Compare commits

..
Author SHA1 Message Date
JackDoan 355b3dcbdb make all lane epochs match 2026-09-25 14:47:13 -05:00
JackDoan 0ee7ace67e avoid race potential when calling installSession for a lane 2026-09-25 14:41:33 -05:00
Wade Simmons 4782c5da9c design: document the multiport lane design
Describe what a lane is, the grouped socket layout, negotiation and port
pairing, how a lane comes up, how a flow stays on one lane and one routine
(including the kernel tun queue feedback loop that forces the lane to come
from the flow hash rather than the routine index), and every level at which
multiport falls back to the old single-port behavior.
2026-09-03 15:12:22 -04:00
Wade Simmons be77501f14 header: show the lane field in the version 1 header diagram
The lane byte was described in prose wedged between the first and second
rows of the table, which both broke up the diagram and left the wire layout
still claiming a single Reserved (uint16). Make it a field: widen the table
to fit Reserved (uint8) and Lane (uint8) with their types, and move the
explanation below the whole diagram.

The prose now covers what the picture can't: lane 0 is what a lanes-unaware
sender and every non-lane packet carry, so the field is compatible in both
directions; the remaining 8 bits stay reserved and zero; the H struct still
stores the pair as one Reserved field, so callers reach the low byte through
H.Lane and EncodeLane; and lane is part of the AEAD's associated data, so it
cannot be altered in flight.
2026-09-03 14:25:01 -04:00
Wade Simmons af439dadbf multiport: SO_REUSEPORT groups per port, routines is per port
Under multiport each port had exactly one socket, so each port had exactly
one core. That hurt worst on the base port, which carries every handshake,
every lighthouse and punch packet, every peer without multiport, and every
tunnel whose lanes are down or firewalled.

Bind multiport.ports consecutive ports and give each one a full group of
`routines` sockets sharing it through SO_REUSEPORT, so the kernel hashes
each arriving 4-tuple onto one of the group. `routines` is now per port and
the total worker count is routines * multiport.ports, computed for you: with
`routines: 8` and `multiport.ports: 4` you get 8 routines per port, 32
total. A routine still owns exactly one socket, which is what lets the read
path own its state without locking.

Sockets are laid out port-major, writers[s*routinesPerPort+r] being the r'th
socket on port listen.port+s, so socket selection is a pure function of
(queue, lane) with no borrowing: laneSock(q,s) = s*R + q%R, and base traffic
takes laneSock(q,0). Sibling routines land on different sockets of the same
port, so a lane's port is served by its whole group.

multiport.ports is a new key and must be > 1; there is no default, since
under these semantics defaulting it would silently multiply the worker
count. The multiport decision moves above tun creation because tun queues
are sized by the total.

Also drop the base-port fallback for unanswered lane probes. A lane now
aims only at its own paired peer port: sharing the base port's destination
buys the lane a source port of its own but costs the peer the receive
spread lanes exist to create, so a lane that cannot reach its port stays
down and its flows ride the base tunnel.
2026-09-03 14:24:57 -04:00
Wade Simmons 53c565eb29 Pick multiport lanes by inside flow hash, not routine index
Lanes were collapsing onto lane 0 and staying there. Nebula writes a
peer's inbound packets to the tun queue matching the socket they arrived
on, and tun_flow_update teaches the kernel to steer that flow's outbound
packets to the same queue, preferring what it learned over the hash. So
while a tunnel's lanes were still down -- which is every tunnel, for its
first moments -- all traffic arrived on socket 0, pinning every flow to
queue 0 on both hosts for as long as it stayed busy. With the lane
following the routine, that meant lane 0 forever.

Two changes break the loop:

Choose the lane from the packet's own 5-tuple hash, so lane spread no
longer depends on tun steering at all. The hash is symmetric, and the
high-addressed side rotates its choice by the low side's port offset, so
a flow's two directions land on partner lanes with exact reverse
4-tuples and each arrives through the conntrack entry the other's probe
opened.

Seed every lane as demanded at lane-set creation, so the first traffic
tick probes them all at once rather than waiting for a flow to hash onto
each. Lanes have to be up before the flows are, not after.

A routine may now send on any lane, so it holds a SendBatch per lane,
built lazily and all borrowing one shared arena -- the slab is the
expensive part, and a batch per (routine, lane) would otherwise cost
hundreds of megabytes. full() counts across the batches so the total
outstanding stays bounded as before.

That also means several routines can write to one socket, which was
already true for base traffic under multiport: the linux batchWriter's
sendmmsg scratch had no lock. Take a mutex there, once per flush.
2026-09-03 14:24:53 -04:00
Wade Simmons 4936b40169 Revert "multiport: log lane up and down at debug"
This reverts commit 06621e622e.
2026-09-02 16:53:10 -04:00
Wade Simmons 06621e622e multiport: log lane up and down at debug
A lane coming up or going down is per-lane, per-tunnel, and repeats on the
keepalive: a node with many tunnels would bury its actual events under it. The
gauges are the place to watch lanes in aggregate, so drop these two to debug for
when you are looking at one tunnel. Probe failures stay at error.
2026-09-02 16:38:53 -04:00
Wade Simmons a304f370f4 multiport: say when lanes are on and what was negotiated
Every multiport log line so far was a reason it turned itself off, so a node
running it looked identical to one that never tried. Add the other half:

- an Info at startup with the lane count and port range
- the negotiated lanes on both "Handshake message received" lines, which
  required allocating the lane set before the log rather than after; it still
  lands before CheckAndComplete/Complete, which was the ordering requirement
- multiport.lanes.up and multiport.lanes.tunnels gauges

The lane field is a slog.Attr so it vanishes on a node without multiport, and
reads zeros when we offered lanes and the peer had none - the case an operator
is actually looking for.

The gauges walk the hostmap instead of keeping a counter at promote/demote: a
tunnel torn down while its lanes are up never demotes them, so a counter would
drift upward forever. Only a multiport node pays for the walk.
2026-09-02 16:38:49 -04:00
Wade Simmons e3efe0c53d multiport: carry a lost install race's counter to the surviving lane window
Two routines can race on a lane's first packet and derive a session each. Only
one gets installed, and the loser was dropped along with the replay-window entry
for the packet it had just accepted, leaving that one counter replayable.

The two sessions hold identical keys, so the winner's window is the same window
in every respect that matters: mark the counter seen on it instead. Closes the
gap without putting a lock on the decrypt path.
2026-09-02 14:48:17 -04:00
Wade Simmons 8c5c740571 multiport: derive lane sessions on demand, install once authenticated
How many lanes a tunnel has is partly the peer's call: it advertises how many it
sends on and we have to be able to receive all of them, up to the header's 256.
Deriving them all when the handshake completes meant a peer advertising a large
count cost us a replay window and two cipher states per lane, per tunnel, for
lanes it may never send on.

Derive each session on the first packet that needs it instead. The TX side asks
through laneSet.session, which installs on the spot — we only ask for lanes we
chose to send on. The RX side can't do that: anyone who can spoof a tunnel's
local index can name any lane, and installing on sight would hand them the same
allocation for free. So laneSession hands back a session without publishing it
and reports that it did; outside.go installs it only once the packet has
decrypted, which is the first moment the lane is known to be real. A spoofer
gets an HKDF per packet and nothing retained.

Only the session table is sized by the peer's advert now. txAddr, demand and
probe are sized by the lanes we will actually send on, so the peer can no longer
size our per-lane tx state either.

Also refuse relayed lane packets before the session lookup rather than after,
so a junk relay packet can't reach the derivation path at all.
2026-09-02 14:48:17 -04:00
Wade Simmons fb20de39b2 multiport: make a lane a session on the tunnel, not a tunnel
Lanes used to be separate HostInfos, each with its own handshake, its own
half-established states, its own lifetime and its own slot bookkeeping. That
bought nothing: a lane is the same tunnel over a different underlay 5-tuple.

Derive lane sessions instead. Noise leaves us with A.eKey == B.dKey, so both
sides HKDF-expand the same two keys with the same per-lane label and land on a
matched pair without exchanging anything. Lanes now cost no handshake, have no
half-established state, and die exactly when their base tunnel does. Which lane
a packet belongs to rides the low byte of the nebula header's Reserved field,
inside the AEAD's associated data.

Receiving on a lane needs no permission, since the session exists the moment the
base handshake completes. Sending on one needs proof the new 5-tuple works, so a
lane stays down until a probe on it is acked and falls back to the base tunnel
the moment it stops being acked. Probing is demand-driven off the connection
manager's per-tunnel traffic tick: a peer we exchange a trickle with never costs
more than its base tunnel, however many lanes are configured. The ack rides the
base session on purpose, so a broken reverse lane can't fail a working one.

Because the data now rides lane counters, the rehandshake, exhaustion and
swap-primary checks take the max counter across the base session and its lanes;
otherwise the base counter would sit near zero while a lane ran its keys past
the nonce ceiling.

Removes OutboundLaneTimer, EnsureLanes, startLaneHandshake, completeLane,
completeLaneResponder, makeLaneTrafficDecision and the lane fields on HostInfo.
Handshake payload field 3 (the per-lane handshake index) is permanently
reserved; peers advertise a TxLanes count instead.
2026-09-02 14:48:17 -04:00
Wade Simmons 8b914e67f4 gofmt
Two double blank lines left behind when the lane commits were rebased onto
master.
2026-09-02 14:48:14 -04:00
Wade Simmons a4c813fd26 handshake: move lane fields to 9 and 10
Fields 6 and 7 are reserved for work in progress, so the multiport lane
adverts move up to 9 and 10. Both are still single-byte tags, so the
encoded payload size does not change.

Also emit the lane fields after CertVersion so MarshalPayload stays in
ascending field-number order, matching what protoc-gen-go would produce
and keeping the hand-built expectations in payload_test.go
straightforward.

Field numbers are the only thing that identifies a lane advert, so a node
on this build and one on the previous numbering will each skip the other's
adverts as unknown fields and fall back to a single vanilla tunnel.
2026-09-02 14:48:11 -04:00
Wade Simmons 11528c6e02 Build multiport lanes lazily, on data-plane demand
Lanes were established for every slot as soon as the base tunnel came up,
so a host with `routines: N` paid N-1 extra Noise handshakes, sessions and
keepalives for every peer it talked to, including peers it exchanges a
trickle with. Only routines that actually carry traffic to a peer need a
lane.

Add a per-slot demand flag to laneState. sendInsideMessage already loads
txLanes[laneSlot] on every direct-path packet; that load missing is now
the signal, and EnsureLanes only starts slots the data plane asked for.
The miss path is unchanged otherwise: the packet rides the base tunnel,
the same fallback used while a lane is down.

noteLaneDemand load-guards its store so repeated misses on a lane-less
slot are plain reads rather than a cache line ping-ponging between the
routines sharing it. The demand check is EnsureLanes' last condition, so
a slot that is pending or inside its backoff keeps the flag for the tick
that can act on it. Consuming the flag also stops a lane whose routine
went quiet from being rebuilt forever after it dies.

The wire format, lane negotiation and port-offset scheme are untouched,
so a lazy node interoperates with an eager one in both directions, and
inbound lane handshakes from a busy peer are still accepted regardless of
our own demand.
2026-09-02 14:48:07 -04:00
JackDoan 192d994273 spread N routines over n lanes
(cherry picked from commit 7043baeb34)
2026-09-02 11:28:32 -04:00
JackDoan 2c9f47c252 port offsets for peer mismatch
(cherry picked from commit 405a78f415)
2026-09-02 11:26:36 -04:00
JackDoan 0ae7ba073c crazy multiport stuff
(cherry picked from commit 0488793a62)
2026-09-02 11:26:30 -04:00
155 changed files with 4212 additions and 7723 deletions

No files matched your search

-142
View File
@@ -1,142 +0,0 @@
#!/usr/bin/env bash
#
# Backport a merged PR to a release branch.
#
# Cherry-picks the merge commit of a merged PR onto release-1.11 and opens a
# backport PR, mirroring the shape of #1842.
#
# Usage: ./backport.sh <pr-number> [target-version] [--continue]
# pr-number The merged PR to backport
# target-version Release version to target (default: 1.11)
# --continue Resume after resolving cherry-pick conflicts: assumes the
# fixes are staged, runs `git cherry-pick --continue`, and
# proceeds to push and open the backport PR.
#
# Requires: gh (authenticated), git, jq.
set -euo pipefail
PR_NUMBER=${1:?usage: backport.sh <pr-number> [target-version] [--continue]}
shift
CONTINUE=0
TARGET_VERSION=""
for arg in "$@"; do
case "$arg" in
--continue) CONTINUE=1 ;;
*) TARGET_VERSION=$arg ;;
esac
done
for cmd in gh git jq; do
command -v "$cmd" >/dev/null || { echo "error: $cmd is required" >&2; exit 1; }
done
# Pull the source PR's metadata. Refuse to backport a PR that never merged.
pr_json=$(gh pr view "$PR_NUMBER" --json title,body,mergedAt,mergeCommit,milestone)
PR_TITLE=$(jq -r '.title' <<<"$pr_json")
PR_BODY=$(jq -r '.body' <<<"$pr_json")
MERGE_SHA=$(jq -r '.mergeCommit.oid // empty' <<<"$pr_json")
PR_MILESTONE=$(jq -r '.milestone.title // empty' <<<"$pr_json")
if [ "$(jq -r '.mergedAt // empty' <<<"$pr_json")" = "" ] || [ -z "$MERGE_SHA" ]; then
echo "error: PR #${PR_NUMBER} is not merged (no merge commit to cherry-pick)" >&2
exit 1
fi
# Default the target to one minor below the PR's milestone: a PR landing in
# v1.12.0 backports to release-1.11. An explicit target-version arg overrides.
if [ -z "$TARGET_VERSION" ]; then
if [[ $PR_MILESTONE =~ ^v?([0-9]+)\.([0-9]+) ]]; then
TARGET_VERSION="${BASH_REMATCH[1]}.$(( BASH_REMATCH[2] - 1 ))"
else
echo "error: PR #${PR_NUMBER} has no v<major>.<minor>.* milestone to infer the target from." >&2
echo "Pass the target version explicitly, e.g. $0 ${PR_NUMBER} 1.11" >&2
exit 1
fi
fi
TARGET_BRANCH="release-${TARGET_VERSION}"
# Slug from the PR title: lower-case, non-alphanumerics to dashes, trimmed.
slug=$(printf '%s' "$PR_TITLE" \
| tr '[:upper:]' '[:lower:]' \
| sed -E 's/[^a-z0-9]+/-/g; s/^-+//; s/-+$//' \
| cut -c1-50 \
| sed -E 's/-+$//')
BRANCH="backport-${TARGET_VERSION//./-}-${slug}"
if [ "$CONTINUE" -eq 1 ]; then
# Resuming: the branch already exists and a cherry-pick is mid-conflict.
current=$(git rev-parse --abbrev-ref HEAD)
if [ "$current" != "$BRANCH" ]; then
echo "error: --continue expects to be on ${BRANCH}, but HEAD is ${current}" >&2
exit 1
fi
echo "Resuming backport of #${PR_NUMBER} on ${BRANCH}"
# Conflicts assumed resolved and staged; core.editor=true keeps the commit message.
git -c core.editor=true cherry-pick --continue
else
# A fresh run switches branches and cherry-picks, so tracked changes must be
# clean. Untracked files are fine, and --continue is exempt (its resolved
# conflicts are meant to be staged).
if [ -n "$(git status --porcelain --untracked-files=no)" ]; then
echo "error: working tree has uncommitted changes to tracked files; commit or stash them first." >&2
exit 1
fi
echo "Backporting #${PR_NUMBER} (${MERGE_SHA}) onto ${TARGET_BRANCH} as ${BRANCH}"
git fetch origin
git checkout -b "$BRANCH" "origin/${TARGET_BRANCH}"
# -m 1 handles a real merge commit; plain cherry-pick handles a squash merge.
if ! { git cherry-pick -x -m 1 "$MERGE_SHA" 2>/dev/null || git cherry-pick -x "$MERGE_SHA"; }; then
echo "error: ${MERGE_SHA} did not cherry-pick cleanly onto ${TARGET_BRANCH}." >&2
echo "Resolve the conflicts, 'git add' them, then re-run:" >&2
echo " $0 ${PR_NUMBER} ${TARGET_VERSION} --continue" >&2
exit 1
fi
fi
title="backport v${TARGET_VERSION}: ${PR_TITLE}"
body=$(printf 'Backport from #%s to release v%s\n\n---\n\n%s' \
"$PR_NUMBER" "$TARGET_VERSION" "$PR_BODY")
# Highest open milestone matching v<version>.* (e.g. v1.11.1), if any.
MILESTONE=$(gh api "repos/{owner}/{repo}/milestones?state=open" \
--jq ".[].title | select(startswith(\"v${TARGET_VERSION}.\"))" \
| sort -V | tail -n1)
milestone_args=()
if [ -n "$MILESTONE" ]; then
milestone_args=(--milestone "$MILESTONE")
else
echo "warning: no open milestone matching v${TARGET_VERSION}.* found; PR will have no milestone" >&2
fi
echo
echo "About to run:"
printf ' git push -u --force-with-lease origin %q\n' "$BRANCH"
printf ' gh pr create --base %q --head %q --title %q --body %q' \
"$TARGET_BRANCH" "$BRANCH" "$title" "$body"
[ -n "$MILESTONE" ] && printf ' --milestone %q' "$MILESTONE"
printf '\n'
echo
read -r -p "Open this pull request? [y/N] " reply
case "$reply" in
[yY] | [yY][eE][sS]) ;;
*) echo "Aborted. The cherry-pick is on local branch ${BRANCH}; push it manually if desired."; exit 0 ;;
esac
git push -u --force-with-lease origin "$BRANCH"
gh pr create \
--base "$TARGET_BRANCH" \
--head "$BRANCH" \
--title "$title" \
--body "$body" \
${milestone_args[@]+"${milestone_args[@]}"}
# The backport now exists, so drop the label that flagged this PR for one.
gh pr edit "$PR_NUMBER" --remove-label needs-backport
# Switch back to the original branch we were on before doing the backport
git checkout -
+1 -1
View File
@@ -98,7 +98,7 @@ jobs:
# WSL2 + Ubuntu so the smoke can run a real linux peer with its own
# netns. iputils-ping is needed for the in-WSL ping check. WSL1 has no
# real kernel and would lack /dev/net/tun, so we have to force WSL2.
- uses: Vampire/setup-wsl@v7
- uses: Vampire/setup-wsl@v3
with:
distribution: Ubuntu-24.04
additional-packages: iputils-ping iproute2
+5 -48
View File
@@ -51,19 +51,15 @@ wsl -d $Distro -- bash -c "rm -rf $WslDir && mkdir -p $WslDir" | Out-Null
$DevName = 'nebula-smoke'
$Ip1 = '192.168.241.1'
$Ip2 = '192.168.241.2'
# Dual stack on purpose: a v4-only overlay never exercises the v6 side of tun.mtu.
$Ip6_1 = 'fd42:4242:241::1'
$Ip6_2 = 'fd42:4242:241::2'
$Mtu = 1300
$Port = 4242
& $NebulaCert ca -name 'smoke-ca' -out-crt "$WorkDir\ca.crt" -out-key "$WorkDir\ca.key"
if ($LASTEXITCODE -ne 0) { throw "nebula-cert ca failed (exit $LASTEXITCODE)" }
& $NebulaCert sign -name 'lighthouse' -networks "$Ip1/24,$Ip6_1/64" -ca-crt "$WorkDir\ca.crt" -ca-key "$WorkDir\ca.key" -out-crt "$WorkDir\lighthouse.crt" -out-key "$WorkDir\lighthouse.key"
& $NebulaCert sign -name 'lighthouse' -networks "$Ip1/24" -ca-crt "$WorkDir\ca.crt" -ca-key "$WorkDir\ca.key" -out-crt "$WorkDir\lighthouse.crt" -out-key "$WorkDir\lighthouse.key"
if ($LASTEXITCODE -ne 0) { throw "nebula-cert sign lighthouse failed (exit $LASTEXITCODE)" }
& $NebulaCert sign -name 'peer' -networks "$Ip2/24,$Ip6_2/64" -ca-crt "$WorkDir\ca.crt" -ca-key "$WorkDir\ca.key" -out-crt "$WorkDir\peer.crt" -out-key "$WorkDir\peer.key"
& $NebulaCert sign -name 'peer' -networks "$Ip2/24" -ca-crt "$WorkDir\ca.crt" -ca-key "$WorkDir\ca.key" -out-crt "$WorkDir\peer.crt" -out-key "$WorkDir\peer.key"
if ($LASTEXITCODE -ne 0) { throw "nebula-cert sign peer failed (exit $LASTEXITCODE)" }
# Windows lighthouse config.
@@ -86,7 +82,7 @@ tun:
drop_local_broadcast: false
drop_multicast: false
tx_queue: 500
mtu: $Mtu
mtu: 1300
network_category: private
logging:
level: info
@@ -130,7 +126,7 @@ tun:
drop_local_broadcast: false
drop_multicast: false
tx_queue: 500
mtu: $Mtu
mtu: 1300
logging:
level: info
format: text
@@ -173,7 +169,7 @@ Write-Host '=== WSL diagnostic ==='
wsl --version 2>&1 | Out-Host
wsl --list --verbose 2>&1 | Out-Host
wsl -d $Distro -u root -- uname -a | Out-Host
wsl -d $Distro -u root -- bash -c "modprobe tun 2>&1 || true; mkdir -p /dev/net; [ -c /dev/net/tun ] || mknod /dev/net/tun c 10 200; chmod 600 /dev/net/tun; { echo 0 > /proc/sys/net/ipv6/conf/all/disable_ipv6; echo 0 > /proc/sys/net/ipv6/conf/default/disable_ipv6; } 2>/dev/null || true; ls -l /dev/net/tun"
wsl -d $Distro -u root -- bash -c "modprobe tun 2>&1 || true; mkdir -p /dev/net; [ -c /dev/net/tun ] || mknod /dev/net/tun c 10 200; chmod 600 /dev/net/tun; ls -l /dev/net/tun"
if ($LASTEXITCODE -ne 0) { throw "failed to prepare /dev/net/tun in WSL (TUN support missing?)" }
# Deliberately no New-NetFirewallRule calls here -- nebula's windows_bypass_wdf
@@ -218,16 +214,6 @@ try {
}
Write-Host "OK: $DevName NetworkCategory=Private"
# v6 silently kept the adapter default of 65535 while v4 was correct.
foreach ($family in @('IPv4', 'IPv6')) {
Wait-Until -TimeoutSec 30 -What "$DevName $family NlMtu=$Mtu" -Predicate {
if ($lhProc.HasExited) { throw "lighthouse exited (code $($lhProc.ExitCode)) before $family mtu was set" }
$rows = @(Get-NetIPInterface -InterfaceAlias $DevName -AddressFamily $family -ErrorAction SilentlyContinue)
$rows.Count -gt 0 -and -not ($rows | Where-Object { $_.NlMtu -ne $Mtu })
}
Write-Host "OK: $DevName $family NlMtu=$Mtu"
}
Wait-Until -TimeoutSec 30 -What "WSL nebula1 with $Ip2" -Predicate {
if ($peerProc.HasExited) { throw "peer exited (code $($peerProc.ExitCode)) before tun was ready" }
$r = wsl -d $Distro -u root -- bash -c "ip -o addr show nebula1 2>/dev/null | grep -q 'inet $Ip2' && echo yes"
@@ -235,13 +221,6 @@ try {
}
Write-Host "OK: WSL nebula1 has $Ip2"
Wait-Until -TimeoutSec 30 -What "WSL nebula1 with $Ip6_2" -Predicate {
if ($peerProc.HasExited) { throw "peer exited (code $($peerProc.ExitCode)) before the v6 address was up" }
$r = wsl -d $Distro -u root -- bash -c "ip -o addr show nebula1 2>/dev/null | grep -q 'inet6 $Ip6_2' && echo yes"
("$r").Trim() -eq 'yes'
}
Write-Host "OK: WSL nebula1 has $Ip6_2"
Wait-Until -TimeoutSec 30 -What "ping from WSL peer to windows lighthouse ($Ip1)" -Predicate {
if ($peerProc.HasExited) { throw "peer exited (code $($peerProc.ExitCode)) before ping succeeded" }
$r = wsl -d $Distro -u root -- bash -c "ping -c1 -W1 $Ip1 >/dev/null 2>&1 && echo OK"
@@ -255,28 +234,6 @@ try {
}
Write-Host "OK: windows lighthouse -> WSL peer"
# Otherwise the v6 networks only prove the interface exists, not that it forwards.
Wait-Until -TimeoutSec 30 -What "v6 ping from WSL peer to windows lighthouse ($Ip6_1)" -Predicate {
if ($peerProc.HasExited) { throw "peer exited (code $($peerProc.ExitCode)) before the v6 ping succeeded" }
$r = wsl -d $Distro -u root -- bash -c "ping -6 -c1 -W1 $Ip6_1 >/dev/null 2>&1 && echo OK"
("$r").Trim() -eq 'OK'
}
Write-Host "OK: WSL peer -> windows lighthouse over v6"
# The NlMtu check above proves the value was written, not that windows honors it. A payload this size only
# crosses the tunnel as fragments cut at NlMtu on the way out of the adapter. Left at the adapter default it
# reaches nebula whole, overflows the udp.MTU (9001) write buffer and is dropped, so the ping never answers.
# Anything smaller would ride out as one oversized datagram and survive on IP fragmentation of the underlay.
$BigPayload = 9000
foreach ($target in @(@{ Family = 'v4'; Ip = $Ip2 }, @{ Family = 'v6'; Ip = $Ip6_2 })) {
Wait-Until -TimeoutSec 30 -What "$($target.Family) ping with a $BigPayload byte payload from windows lighthouse to WSL peer ($($target.Ip))" -Predicate {
if ($peerProc.HasExited) { throw "peer exited (code $($peerProc.ExitCode)) before the large $($target.Family) ping succeeded" }
$null = & ping.exe -n 1 -w 1000 -l $BigPayload $target.Ip
$LASTEXITCODE -eq 0
}
Write-Host "OK: windows lighthouse -> WSL peer, $BigPayload byte $($target.Family) payload"
}
Write-Host ''
Write-Host 'All smoke checks passed.'
}
+1 -21
View File
@@ -7,17 +7,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
## [Unreleased]
### Fixed
- `create-tunnel -address <ip:port>` panicked on a nil RemoteList and, over the ssh debug server, took nebula
down with it. A hostinfo returned by StartHandshake had no RemoteList until the handshake loop's first
pass attached one, and `create-tunnel -address` set the remote before that. The list is now attached when
the handshake starts, so a pending hostinfo never exists without one.
## [1.11.1] - 2026-08-21
See the [v1.11.1](https://github.com/slackhq/nebula/milestone/30?closed=1) milestone for a complete list of changes.
### Changed
- IPv6 packets whose next header is a protocol Nebula does not parse (SCTP, GRE, IP-in-IP, etc.) are now
@@ -26,18 +15,11 @@ See the [v1.11.1](https://github.com/slackhq/nebula/milestone/30?closed=1) miles
their true protocol, so only a `proto: any` rule allows them. If you carry one of these protocols over the
overlay, confirm a `proto: any` rule covers it before upgrading, it may have been passing only through this
bypass. (#1840)
- Drop the dependency on `github.com/cyberdelia/go-metrics-graphite`, which has been unmaintained for over ten
years, by inlining the small amount of code Nebula used. (#1832)
### Fixed
- The ICMPv6 type was read from the wrong byte when classifying IPv6 packets, so the echo identifier used
for conntrack was never picked up. (#1840)
- Enforce outbound message counter limits so a tunnel is rehandshaked before the counter can wrap, preventing
nonce reuse. This is unreachable in practice, but is enforced as a defense-in-depth measure. (#1841)
- Prevent `nebula-cert ca` from running out of memory on 32bit systems when generating encrypted private keys. (#1834)
- Tolerate `ErrDumpInterrupted` when listing tun addresses on Linux, so a transient interrupted netlink dump
no longer aborts startup. (#1835)
## [1.11.0] - 2026-07-23
@@ -902,9 +884,7 @@ created.)
- Initial public release.
[Unreleased]: https://github.com/slackhq/nebula/compare/v1.11.1...HEAD
[1.11.1]: https://github.com/slackhq/nebula/releases/tag/v1.11.1
[1.11.0]: https://github.com/slackhq/nebula/releases/tag/v1.11.0
[Unreleased]: https://github.com/slackhq/nebula/compare/v1.10.3...HEAD
[1.10.3]: https://github.com/slackhq/nebula/releases/tag/v1.10.3
[1.10.2]: https://github.com/slackhq/nebula/releases/tag/v1.10.2
[1.10.1]: https://github.com/slackhq/nebula/releases/tag/v1.10.1
-3
View File
@@ -242,14 +242,11 @@ test-cov-html:
go test -coverprofile=coverage.out
go tool cover -html=coverage.out
# The package builds only compile. The final line links an android binary so a linker-only failure
# cannot pass CI.
build-test-mobile:
GOARCH=amd64 GOOS=ios go build $(shell go list ./... | grep -v '/cmd/\|/examples/')
GOARCH=arm64 GOOS=ios go build $(shell go list ./... | grep -v '/cmd/\|/examples/')
GOARCH=amd64 GOOS=android go build $(shell go list ./... | grep -v '/cmd/\|/examples/')
GOARCH=arm64 GOOS=android go build $(shell go list ./... | grep -v '/cmd/\|/examples/')
GOARCH=arm64 GOOS=android go build -o /dev/null ${NEBULA_CMD_PATH}
bench:
go test -bench=.
+11 -2
View File
@@ -79,7 +79,13 @@ func (b *Bits) clearRange(startPos, count uint64) uint64 {
// handle the potential partial word before pos becomes u64 aligned
word := pos >> 6
bit := pos & 63
take := min(min(uint64(64)-bit, remaining), b.length-pos)
take := uint64(64) - bit
if take > remaining {
take = remaining
}
if take > b.length-pos {
take = b.length - pos
}
var mask uint64
if take == 64 {
mask = math.MaxUint64
@@ -183,7 +189,10 @@ func (b *Bits) Update(l *slog.Logger, i uint64) bool {
func (b *Bits) updateSlow(l *slog.Logger, i uint64) bool {
// If i is a jump, adjust the window, record lost, update current, and return true
if i > b.current {
end := min(i, b.current+b.length)
end := i
if end > b.current+b.length {
end = b.current + b.length
}
count := end - b.current
startPos := (b.current + 1) & b.lengthMask
+1 -1
View File
@@ -120,7 +120,7 @@ func TestCertificate_SignP256_AlwaysNormalized(t *testing.T) {
pub := elliptic.Marshal(elliptic.P256(), priv.PublicKey.X, priv.PublicKey.Y)
rawPriv := priv.D.FillBytes(make([]byte, 32))
for i := range 1000 {
for i := 0; i < 1000; i++ {
if i&1 == 1 {
tbs.Version = Version1
} else {
+4 -4
View File
@@ -151,7 +151,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
var groups []string
if *cf.groups != "" {
for rg := range strings.SplitSeq(*cf.groups, ",") {
for _, rg := range strings.Split(*cf.groups, ",") {
g := strings.TrimSpace(rg)
if g != "" {
groups = append(groups, g)
@@ -171,7 +171,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
}
if *cf.networks != "" {
for rs := range strings.SplitSeq(*cf.networks, ",") {
for _, rs := range strings.Split(*cf.networks, ",") {
rs := strings.Trim(rs, " ")
if rs != "" {
n, err := netip.ParsePrefix(rs)
@@ -193,7 +193,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
}
if *cf.unsafeNetworks != "" {
for rs := range strings.SplitSeq(*cf.unsafeNetworks, ",") {
for _, rs := range strings.Split(*cf.unsafeNetworks, ",") {
rs := strings.Trim(rs, " ")
if rs != "" {
n, err := netip.ParsePrefix(rs)
@@ -221,7 +221,7 @@ func ca(args []string, out io.Writer, errOut io.Writer, pr PasswordReader) error
if !isP11 && *cf.encryption {
passphrase = []byte(os.Getenv("NEBULA_CA_PASSPHRASE"))
if len(passphrase) == 0 {
for range 5 {
for i := 0; i < 5; i++ {
errOut.Write([]byte("Enter passphrase: "))
passphrase, err = pr.ReadPassword()
+1
View File
@@ -1,4 +1,5 @@
//go:build !windows
// +build !windows
package main
+4 -4
View File
@@ -146,7 +146,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
passphrase = []byte(os.Getenv("NEBULA_CA_PASSPHRASE"))
if len(passphrase) == 0 {
// ask for a passphrase until we get one
for range 5 {
for i := 0; i < 5; i++ {
errOut.Write([]byte("Enter passphrase: "))
passphrase, err = pr.ReadPassword()
@@ -203,7 +203,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
}
if *sf.networks != "" {
for rs := range strings.SplitSeq(*sf.networks, ",") {
for _, rs := range strings.Split(*sf.networks, ",") {
rs := strings.Trim(rs, " ")
if rs != "" {
n, err := netip.ParsePrefix(rs)
@@ -228,7 +228,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
}
if *sf.unsafeNetworks != "" {
for rs := range strings.SplitSeq(*sf.unsafeNetworks, ",") {
for _, rs := range strings.Split(*sf.unsafeNetworks, ",") {
rs := strings.Trim(rs, " ")
if rs != "" {
n, err := netip.ParsePrefix(rs)
@@ -247,7 +247,7 @@ func signCert(args []string, out io.Writer, errOut io.Writer, pr PasswordReader)
var groups []string
if *sf.groups != "" {
for rg := range strings.SplitSeq(*sf.groups, ",") {
for _, rg := range strings.Split(*sf.groups, ",") {
g := strings.TrimSpace(rg)
if g != "" {
groups = append(groups, g)
+1
View File
@@ -1,4 +1,5 @@
//go:build !windows
// +build !windows
package main
+1
View File
@@ -1,4 +1,5 @@
//go:build !windows
// +build !windows
package main
+1
View File
@@ -1,4 +1,5 @@
//go:build !linux
// +build !linux
package main
+9 -6
View File
@@ -5,7 +5,6 @@ import (
"errors"
"fmt"
"log/slog"
"maps"
"math"
"os"
"os/signal"
@@ -155,7 +154,9 @@ func (c *C) ReloadConfig() {
defer c.reloadLock.Unlock()
c.oldSettings = make(map[string]any)
maps.Copy(c.oldSettings, c.Settings)
for k, v := range c.Settings {
c.oldSettings[k] = v
}
err := c.Load(c.path)
if err != nil {
@@ -176,7 +177,9 @@ func (c *C) ReloadConfigString(raw string) error {
defer c.reloadLock.Unlock()
c.oldSettings = make(map[string]any)
maps.Copy(c.oldSettings, c.Settings)
for k, v := range c.Settings {
c.oldSettings[k] = v
}
err := c.LoadString(raw)
if err != nil {
@@ -213,7 +216,7 @@ func (c *C) GetStringSlice(k string, d []string) []string {
}
v := make([]string, len(rv))
for i := range v {
for i := 0; i < len(v); i++ {
v[i] = fmt.Sprintf("%v", rv[i])
}
@@ -307,8 +310,8 @@ func (c *C) IsSet(k string) bool {
}
func (c *C) get(k string, v any) any {
parts := strings.SplitSeq(k, ".")
for p := range parts {
parts := strings.Split(k, ".")
for _, p := range parts {
m, ok := v.(map[string]any)
if !ok {
return nil
+32 -5
View File
@@ -174,8 +174,10 @@ func (cm *connectionManager) doTrafficCheck(localIndex uint32, p, nb, out []byte
switch decision {
case deleteTunnel:
// Local delete only, clearing the lighthouse cache if this was the last hostinfo for the vpn addr
cm.intf.closeTunnel(hostinfo)
if cm.hostMap.DeleteHostInfo(hostinfo) {
// Only clearing the lighthouse cache if this is the last hostinfo for this vpn ip in the hostmap
cm.intf.lightHouse.DeleteVpnAddrs(hostinfo.vpnAddrs)
}
case closeTunnel:
cm.intf.sendCloseTunnel(hostinfo)
@@ -195,6 +197,27 @@ func (cm *connectionManager) doTrafficCheck(localIndex uint32, p, nb, out []byte
}
cm.resetRelayTrafficCheck(hostinfo)
cm.maintainLanes(localIndex, decision, hostinfo, now, nb, out)
}
// maintainLanes piggybacks multiport lane probing on the per-tunnel traffic
// tick. This tick is the right place for it precisely because lanes are
// demand-driven: a tunnel only lands here when it has traffic, which is the
// same condition that raises lane demand.
//
// makeTrafficDecision returns a nil hostinfo on some keep-alive paths, so
// re-resolve the index in that case.
func (cm *connectionManager) maintainLanes(localIndex uint32, decision trafficDecision, hostinfo *HostInfo, now time.Time, nb, out []byte) {
if decision == deleteTunnel || decision == closeTunnel {
return
}
if hostinfo == nil {
hostinfo = cm.hostMap.QueryIndex(localIndex)
if hostinfo == nil {
return
}
}
cm.intf.probeLanes(hostinfo, now, nb, out)
}
func (cm *connectionManager) resetRelayTrafficCheck(hostinfo *HostInfo) {
@@ -327,7 +350,10 @@ func (cm *connectionManager) makeTrafficDecision(localIndex uint32, now time.Tim
return closeTunnel, hostinfo, nil
}
if hostinfo.ConnectionState != nil && hostinfo.ConnectionState.messageCounter.Load() >= RejectAfterMessages {
// The highest counter across the base session and its lanes: the lanes carry
// the data, so the base counter alone would sit near zero while a lane runs
// its keys past the nonce ceiling.
if hostinfo.maxMessageCounter() >= RejectAfterMessages {
// Send path can't encrypt a CloseTunnel notify, so just delete locally; the peer recovers via recv_error.
hostinfo.logger(cm.l).Error("Dropping tunnel, message counter is exhausted")
return deleteTunnel, hostinfo, nil
@@ -435,6 +461,7 @@ func (cm *connectionManager) isInactive(hostinfo *HostInfo, now time.Time) (time
return 0, false
}
// Lane traffic is this hostinfo's traffic, so lastUsed already covers it.
inactiveDuration := now.Sub(hostinfo.lastUsed)
if inactiveDuration < cm.getInactivityTimeout() {
// It's not considered inactive
@@ -458,7 +485,7 @@ func (cm *connectionManager) shouldSwapPrimary(current *HostInfo) bool {
return false
}
if current.ConnectionState.messageCounter.Load() >= RehandshakeAfterMessages {
if current.maxMessageCounter() >= RehandshakeAfterMessages {
// This tunnel is being rolled for counter exhaustion, never swap back onto its spent key.
return false
}
@@ -562,7 +589,7 @@ func (cm *connectionManager) tryRehandshake(hostinfo *HostInfo) {
cm.intf.handshakeManager.StartHandshake(hostinfo.vpnAddrs[0], nil)
return
}
if hostinfo.ConnectionState.messageCounter.Load() >= RehandshakeAfterMessages {
if hostinfo.maxMessageCounter() >= RehandshakeAfterMessages {
cm.l.Info("Re-handshaking with remote",
"vpnAddrs", hostinfo.vpnAddrs,
"reason", "message counter rehandshake threshold reached",
-3
View File
@@ -7,7 +7,6 @@ import (
"testing"
"time"
"github.com/rcrowley/go-metrics"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/overlay/overlaytest"
@@ -22,8 +21,6 @@ func newTestLighthouse() *LightHouse {
l: test.NewLogger(),
addrMap: map[netip.Addr]*RemoteList{},
queryChan: make(chan netip.Addr, 10),
// Nothing drains queryChan here, so an 11th query is dropped.
queryDropped: metrics.NilCounter{},
}
lighthouses := []netip.Addr{}
staticList := map[netip.Addr]struct{}{}
+63 -1
View File
@@ -1,9 +1,12 @@
package nebula
import (
"crypto/hkdf"
"crypto/sha256"
"encoding/json"
"fmt"
"log/slog"
"strconv"
"sync"
"sync/atomic"
@@ -26,12 +29,15 @@ const (
// RehandshakeAfterMessages must stay below RejectAfterMessages so tunnels roll before the hard send stop.
const _ = RejectAfterMessages - RehandshakeAfterMessages
// sessionEpoch hands out a receiver-local ordinal to every ConnectionState at creation. The RX
// sessionEpoch hands out a receiver-local ordinal to every tunnel at creation. The RX
// staging sort (overlay/batch) orders packets by (epoch, message counter). A re-handshake never
// rekeys an existing tunnel; it brings up a new hostinfo and ConnectionState with a counter space
// starting near zero, while the old tunnel keeps decrypting until torn down. During that cutover
// one flush batch can hold packets from both tunnels, and the epoch keeps the old tunnel's
// packets sorted first.
//
// It is drawn once per tunnel, by the base session: a tunnel's multiport lane sessions copy it
// rather than draw their own.
var sessionEpoch atomic.Uint64
type ConnectionState struct {
@@ -74,6 +80,53 @@ func newConnectionStateFromResult(r *handshake.Result) (*ConnectionState, error)
return ci, nil
}
// newLaneConnectionState derives multiport lane s's session from the base
// tunnel's material. Each key is an HKDF expansion of the base tunnel's matching
// key, labelled with the lane index, so the pair stays matched with no extra
// negotiation: Noise leaves our send key equal to the peer's receive key, and
// expanding both with the same label preserves that.
//
// The lane gets its own counter and replay window starting from zero. No
// handshake messages were spent on it, so unlike the base session there is
// nothing to seed. The epoch is the tunnel's, not the lane's; see sessionEpoch.
func newLaneConnectionState(m *laneMaterial, lane uint8) (*ConnectionState, error) {
if lane == 0 {
return nil, fmt.Errorf("lane 0 is the base session")
}
eKey, err := deriveLaneKey(m.eKey, lane)
if err != nil {
return nil, err
}
dKey, err := deriveLaneKey(m.dKey, lane)
if err != nil {
return nil, err
}
return &ConnectionState{
myCert: m.myCert,
initiator: m.initiator,
peerCert: m.peerCert,
eKey: noiseutil.NewCipherStateFromKey(eKey, m.cipher),
dKey: noiseutil.NewCipherStateFromKey(dKey, m.cipher),
window: NewBits(ReplayWindow),
epoch: m.epoch,
}, nil
}
// deriveLaneKey expands a base tunnel key into the key for one lane.
func deriveLaneKey(base [32]byte, lane uint8) ([32]byte, error) {
var out [32]byte
// The base key is already unique to this tunnel and direction, so the lane
// index is the only thing that needs to vary; no salt is required.
k, err := hkdf.Key(sha256.New, base[:], nil, laneKeyInfo+" "+strconv.Itoa(int(lane)), len(out))
if err != nil {
return out, err
}
copy(out[:], k)
return out, nil
}
func (cs *ConnectionState) MarshalJSON() ([]byte, error) {
return json.Marshal(m{
"certificate": cs.peerCert,
@@ -118,6 +171,15 @@ func (cs *ConnectionState) Decrypt(l *slog.Logger, messageCounter uint64, packet
return out, nil
}
// noteSeen records a counter that some other session for the same keys already
// accepted, so a packet doesn't become replayable just because the session that
// decrypted it was thrown away. See laneSet.installSession, its only caller.
func (cs *ConnectionState) noteSeen(l *slog.Logger, messageCounter uint64) {
cs.decryptLock.Lock()
cs.window.Update(l, messageCounter)
cs.decryptLock.Unlock()
}
func (cs *ConnectionState) VerifyRelay(l *slog.Logger, messageCounter uint64, packet []byte, nb []byte) error {
cs.decryptLock.Lock()
result := cs.window.Check(l, messageCounter)
+3 -120
View File
@@ -2,8 +2,6 @@ package nebula
import (
"net/netip"
"strconv"
"sync"
"testing"
"time"
@@ -13,7 +11,6 @@ import (
ct "github.com/slackhq/nebula/cert_test"
"github.com/slackhq/nebula/handshake"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/overlay"
"github.com/slackhq/nebula/test"
"github.com/slackhq/nebula/udp"
"github.com/stretchr/testify/assert"
@@ -61,6 +58,7 @@ func runTestHandshake(t *testing.T) (initR, respR *handshake.Result) {
cert.Version2, initCreds, verifier,
func() (uint32, error) { return 1000, nil },
true, header.HandshakeIXPSK0,
nil,
)
require.NoError(t, err)
@@ -68,6 +66,7 @@ func runTestHandshake(t *testing.T) (initR, respR *handshake.Result) {
cert.Version2, respCreds, verifier,
func() (uint32, error) { return 2000, nil },
false, header.HandshakeIXPSK0,
nil,
)
require.NoError(t, err)
@@ -100,7 +99,7 @@ func TestConnectionState_NextMessageCounter(t *testing.T) {
assert.Equal(t, RejectAfterMessages, cs.messageCounter.Load())
// Continued send attempts stay refused and the counter never wraps
for range 10 {
for i := 0; i < 10; i++ {
_, ok = cs.NextMessageCounter()
assert.False(t, ok)
}
@@ -150,122 +149,6 @@ func TestSendNoMetricsCloseTunnelKeepsRebindEpoch(t *testing.T) {
assert.True(t, hostinfo.markOut(1), "a CloseTunnel send must not consume the rebind epoch")
}
// capturingConn is a udp.Conn that records every WriteTo.
type capturingConn struct {
udp.NoopConn
writes [][]byte
addrs []netip.AddrPort
}
func (c *capturingConn) WriteTo(b []byte, addr netip.AddrPort) error {
c.writes = append(c.writes, append([]byte(nil), b...))
c.addrs = append(c.addrs, addr)
return nil
}
// TestSendNoMetricsViaRelay sends through a relay from the mtu sized buffer the cached packet flush uses. Payloads up
// to overlay.MaxMTU go out intact, anything bigger is dropped rather than outgrowing the buffer, which once panicked.
func TestSendNoMetricsViaRelay(t *testing.T) {
tests := []struct {
n int
sent bool
}{
{0, true},
{1300, true},
{overlay.MaxMTU, true},
{overlay.MaxMTU + 1, false}, // too big for the relay's tag
{mtu - 47, false}, // too big for the inner tag, this panicked
{9000, false}, // a full packet at tun.mtu 9000
}
for _, tt := range tests {
t.Run(strconv.Itoa(tt.n), func(t *testing.T) {
// One tunnel to the target, carried end to end, and one to the relay that wraps it
targetInit, targetResp := runTestHandshake(t)
relayInit, relayResp := runTestHandshake(t)
ci, err := newConnectionStateFromResult(targetInit)
require.NoError(t, err)
relayCI, err := newConnectionStateFromResult(relayInit)
require.NoError(t, err)
targetAddr := netip.MustParseAddr("10.0.0.2")
relayAddr := netip.MustParseAddr("10.0.0.3")
relayRemote := netip.MustParseAddrPort("192.0.2.1:4242")
relay := &Relay{Type: TerminalType, State: Established, LocalIndex: 5, RemoteIndex: 6, PeerAddr: targetAddr}
relayHI := &HostInfo{
vpnAddrs: []netip.Addr{relayAddr},
ConnectionState: relayCI,
relayState: RelayState{
relayForByAddr: map[netip.Addr]*Relay{targetAddr: relay},
relayForByIdx: map[uint32]*Relay{relay.LocalIndex: relay},
},
}
relayHI.remote.Store(&relayRemote)
// No remote of its own, so the send has to go via the relay
hostinfo := &HostInfo{
vpnAddrs: []netip.Addr{targetAddr},
ConnectionState: ci,
remoteIndexId: 7,
relayState: RelayState{
relays: []netip.Addr{relayAddr},
relayForByAddr: map[netip.Addr]*Relay{},
relayForByIdx: map[uint32]*Relay{},
},
}
l := test.NewLogger()
hm := newHostMap(l)
hm.Hosts[relayAddr] = relayHI
conn := &capturingConn{}
f := &Interface{
l: l,
hostMap: hm,
messageMetrics: &MessageMetrics{txExhausted: metrics.NewCounter()},
writers: []udp.Conn{conn},
}
f.connectionManager = &connectionManager{intf: f, relayUsed: map[uint32]struct{}{}, relayUsedLock: &sync.RWMutex{}}
payload := make([]byte, tt.n)
for i := range payload {
payload[i] = byte(i)
}
counter := ci.messageCounter.Load()
f.sendNoMetrics(header.Message, 0, ci, hostinfo, netip.AddrPort{}, payload, make([]byte, 12), make([]byte, mtu), 0)
if !tt.sent {
assert.Empty(t, conn.writes)
assert.Equal(t, counter, ci.messageCounter.Load(), "a dropped packet must not spend a counter")
return
}
require.Len(t, conn.writes, 1)
assert.Equal(t, relayRemote, conn.addrs[0])
pkt := conn.writes[0]
// The relay authenticates the whole packet under its own header
outer := &header.H{}
require.NoError(t, outer.Parse(pkt))
assert.Equal(t, header.MessageRelay, outer.Subtype)
assert.Equal(t, relay.RemoteIndex, outer.RemoteIndex)
relayPeer, err := newConnectionStateFromResult(relayResp)
require.NoError(t, err)
require.NoError(t, relayPeer.VerifyRelay(l, outer.MessageCounter, pkt, make([]byte, 12)))
// Inside it, the target decrypts the end to end packet back to the payload
inner := pkt[header.Len : len(pkt)-relayPeer.dKey.Overhead()]
h := &header.H{}
require.NoError(t, h.Parse(inner))
assert.Equal(t, header.Message, h.Type)
assert.Equal(t, hostinfo.remoteIndexId, h.RemoteIndex)
targetPeer, err := newConnectionStateFromResult(targetResp)
require.NoError(t, err)
got, err := targetPeer.Decrypt(l, h.MessageCounter, inner, make([]byte, 12))
require.NoError(t, err)
assert.Equal(t, payload, got)
})
}
}
func TestNewConnectionStateFromResult(t *testing.T) {
initR, respR := runTestHandshake(t)
+42 -2
View File
@@ -67,6 +67,16 @@ type ControlHostInfo struct {
CurrentRemote netip.AddrPort `json:"currentRemote"`
CurrentRelaysToMe []netip.Addr `json:"currentRelaysToMe"`
CurrentRelaysThroughMe []netip.Addr `json:"currentRelaysThroughMe"`
Lanes []ControlLane `json:"lanes,omitempty"`
}
// ControlLane reports one multiport lane of a tunnel. Only lanes we may send on
// are listed; receive-only lanes have no state worth showing.
type ControlLane struct {
Index uint8 `json:"index"`
Up bool `json:"up"`
Remote netip.AddrPort `json:"remote,omitempty"`
MessageCounter uint64 `json:"messageCounter"`
}
// Start actually runs nebula, this is a nonblocking call.
@@ -202,10 +212,15 @@ func (c *Control) RebindUDPServer() {
return
}
// Every socket needs rebinding, not just the base: with multiport each one is bound to its own lane port, and
// even without it the surplus SO_REUSEPORT sockets stay pinned to the interface we came up on otherwise.
//
// A failure here means we are likely still pinned to the interface we came up on, so the rest of this is
// unlikely to help. Say so instead of silently carrying on as if we rebound.
if err := c.f.outside.Rebind(); err != nil {
c.l.Error("Failed to rebind udp socket", "error", err)
for i, w := range c.f.writers {
if err := w.Rebind(); err != nil {
c.l.Error("Failed to rebind udp socket", "error", err, "writer", i)
}
}
// Trigger a lighthouse update, useful for mobile clients that should have an update interval of 0
@@ -385,6 +400,7 @@ func copyHostInfo(h *HostInfo, preferredRanges []netip.Prefix) ControlHostInfo {
CurrentRelaysToMe: h.relayState.CopyRelayIps(),
CurrentRelaysThroughMe: h.relayState.CopyRelayForIps(),
CurrentRemote: h.GetRemote(),
Lanes: copyLanes(h),
}
for i, a := range h.vpnAddrs {
@@ -402,6 +418,30 @@ func copyHostInfo(h *HostInfo, preferredRanges []netip.Prefix) ControlHostInfo {
return chi
}
// copyLanes snapshots the sendable multiport lanes of a tunnel, or nil when it
// has none. txAddr is the lane's gate as well as its destination, so a nil load
// is exactly "this lane is down and its routine is riding the base tunnel".
func copyLanes(h *HostInfo) []ControlLane {
ls := h.lanes
if ls == nil || ls.txLanes < 2 {
return nil
}
lanes := make([]ControlLane, 0, ls.txLanes-1)
for s := 1; s < ls.txLanes; s++ {
l := ControlLane{Index: uint8(s)}
if addr := ls.txAddr[s].Load(); addr != nil {
l.Up = true
l.Remote = *addr
}
if cs := ls.sessions[s].Load(); cs != nil {
l.MessageCounter = cs.messageCounter.Load()
}
lanes = append(lanes, l)
}
return lanes
}
func listHostMapHosts(hl controlHostLister) []ControlHostInfo {
hosts := make([]ControlHostInfo, 0)
pr := hl.GetPreferredRanges()
+1 -1
View File
@@ -247,7 +247,7 @@ func TestControl_ConcurrentStopAndStart(t *testing.T) {
c, _, _ := newReadyControl(t)
var wg sync.WaitGroup
for range 2 {
for i := 0; i < 2; i++ {
wg.Go(func() { c.Stop() })
}
wg.Go(func() { _ = c.Start() })
+1 -1
View File
@@ -105,7 +105,7 @@ func TestControl_GetHostInfoByVpnIp(t *testing.T) {
}
// Make sure we don't have any unexpected fields
assertFields(t, []string{"VpnAddrs", "LocalIndex", "RemoteIndex", "RemoteAddrs", "Cert", "MessageCounter", "CurrentRemote", "CurrentRelaysToMe", "CurrentRelaysThroughMe"}, thi)
assertFields(t, []string{"VpnAddrs", "LocalIndex", "RemoteIndex", "RemoteAddrs", "Cert", "MessageCounter", "CurrentRemote", "CurrentRelaysToMe", "CurrentRelaysThroughMe", "Lanes"}, thi)
assert.Equal(t, &expectedInfo, thi)
test.AssertDeepCopyEqual(t, &expectedInfo, thi)
+526
View File
@@ -0,0 +1,526 @@
# Multiport lanes
Status: experimental. Linux only. Off unless `multiport.ports` is set.
## The problem
A nebula tunnel is one UDP 4-tuple. Everything between the two hosts that makes
a decision per flow makes it once, for the whole tunnel:
- **ECMP / LAG** hashes the 4-tuple and picks one path. A tunnel gets one path's
worth of bandwidth no matter how many exist.
- **NIC RSS** hashes the 4-tuple to one receive queue, so one CPU takes every
interrupt for the tunnel and receive is capped by a single core.
- **Per-flow policers and shapers** see one flow and rate-limit it as one.
- **Cloud per-flow bandwidth caps** are the hard version of that. AWS EC2 meters
each 5-tuple separately and caps a single flow well below what the instance can
do in aggregate -- on the order of 5 Gbps for a single flow within a VPC (more
inside a cluster placement group, or with ENA Express; check the current EC2
network-limits docs for exact numbers) on instances whose aggregate allowance
is many times that. Other providers do the same. Nothing about the path is
saturated when this fires, and no amount of retuning changes it: the hypervisor
is metering *the flow*, so the only way to get more is to be more than one
flow.
Running `routines: N` does not help. Those N sockets share one port through
`SO_REUSEPORT`, and the group is keyed on the exact `(addr, port)` pair, so the
kernel's reuseport hash is the *only* thing spreading the work -- every hash
outside this host still sees a single flow.
Multiport gives one tunnel several underlay 4-tuples, so all of those per-flow
decisions get made several times, independently. A tunnel with N lanes is N
flows to everything counting flows: N ECMP hashes, N receive queues, N of the
cloud provider's per-flow buckets.
There is a second bottleneck, and it is inside this host rather than out on the
network: in FIPS 140 mode the AES-GCM implementation refuses to seal twice under
the same nonce, which forces every encrypt on a session to be serialized behind
one mutex. `routines: N` does not help there either. Because each lane is a
separate session, multiport splits that mutex as well -- see
[The FIPS 140 encrypt lock](#the-fips-140-encrypt-lock).
## What a lane is
A lane is **not** a second tunnel. It is an extra session on the same
`HostInfo`.
Noise leaves both sides with `A.eKey == B.dKey` and `A.dKey == B.eKey`. Expanding
both keys through HKDF-SHA256 with the same per-lane label preserves that
equality, so both sides land on a matched key pair having exchanged nothing:
```
lane s send key = HKDF(base eKey, info: "nebula multiport lane v1 <s>")
lane s recv key = HKDF(base dKey, info: "nebula multiport lane v1 <s>")
```
(`connection_state.go:deriveLaneKey`. The base key is already unique per tunnel
and per direction, so the lane index is the only thing that needs to vary and no
salt is required.)
Consequences worth stating plainly:
- A lane costs **no handshake** and has no half-established state.
- A lane dies exactly when its base tunnel dies. There is no independent
lifetime to reason about, no second teardown path.
- Each lane is a **full** session: its own message counter, its own replay
window, its own cipher states, its own encrypt lock. Flows on different paths
never contend for shared replay state, which is what makes reordering across
lanes harmless, and under FIPS 140 the separate encrypt locks are what let one
tunnel encrypt on more than one core.
- Rolling the base tunnel replaces every lane key, because lanes are derived
from it. `maxMessageCounter` therefore reports the max across the base and all
lane counters, so rehandshake and exhaustion thresholds see the real data
volume rather than the base session's small share.
Which lane a packet belongs to travels in the nebula header, in the low 8 bits
of what used to be `Reserved` (see `header/header.go`). It is part of the AEAD's
associated data, so a lane index cannot be altered in flight -- a packet
decrypts on the lane it claims or not at all.
**Lane 0 is the base tunnel itself**: `HostInfo.ConnectionState`, the base port,
the peer's real remote address. It is not a special case reserved for control
traffic; it carries its share of data flows like any other lane.
## Socket layout
`routines` is **per port**. Each of `multiport.ports` consecutive ports gets its
own full group of `routines` sockets sharing it through `SO_REUSEPORT`, so total
sockets = total routines = total tun queues = `routines * multiport.ports`.
```
routines: 2, multiport.ports: 3, listen.port: 4242
port 4242 port 4243 port 4244
(lane 0/base) (lane 1) (lane 2)
+-------------+ +-------------+ +-------------+
writers[] | 0 | 1 | | 2 | 3 | | 4 | 5 |
+-------------+ +-------------+ +-------------+
routine 0 1 0 1 0 1
```
Sockets are laid out **port-major**: `writers[s*routinesPerPort + r]` is the
r'th socket on port `listen.port+s`. So:
```go
laneSock(q, s) = s*routinesPerPort + q%routinesPerPort // inside.go
egressSock(q) = laneSock(q, 0) // base traffic
```
Two properties fall out of this, and the rest of the design depends on both:
1. **Every routine owns exactly one socket.** `listenIn` blocks in `recvmmsg`
and everything downstream of it -- the batcher, the conntrack cache, the
`txQueue` -- is single-owner and lock-free. More sockets than routines would
need epoll or locks.
2. **Every routine has a sibling socket at the same group position on every
port.** So socket selection is a pure function of `(queue, lane)` with no
borrowing and no shared state, and each port's traffic spreads across its
whole group rather than funnelling into one socket.
The second property is why `routines` is per port rather than a total to divide
up. With one socket per port, each port would be served by a single core -- worst
of all the base port, which carries every handshake, every lighthouse and punch
packet, every peer without multiport, and every tunnel whose lanes are down.
## Negotiation
Multiport capability rides the existing handshake payload as two new protobuf
fields (`handshake/payload.go`): `InitiatorLanes` field 9 and `ResponderLanes`
field 10, each a `LaneDetails{PortCount, BasePort, TxLanes}`.
- `PortCount` / `BasePort` -- the contiguous port range the sender bound, so the
peer knows where to aim its lanes.
- `TxLanes` -- how many lanes the sender may send on, so the peer knows how many
lane sessions it must be prepared to receive on.
Both are `nil` when multiport is off, which keeps the encoded payload
**byte-identical** to a vanilla one. A peer that has never heard of lanes skips
unknown fields as protobuf requires, and gets a plain tunnel.
From the result, `newLaneSet` computes:
```go
sessions = min(max(myLanes, peerTxLanes), 256) // we must be able to RECEIVE all of theirs
txLanes = min(myLanes, peerPortCount, sessions) // we may only SEND on ports they bound
```
Sizing RX by the peer's count and TX by our own is what lets asymmetric hosts
work: a 4-port laptop talking to a 32-port server sends on 4 lanes and receives
on 32.
### Port pairing
Lane `s` targets `peerBasePort + ((s + portOffset) % peerPortCount)`.
`portOffset` is a per-pair FNV hash of the sorted vpn-address pair. Without it,
every small peer would aim its few lanes at a big peer's first few ports and
concentrate that peer's receive work on a couple of sockets.
The rotation has to cancel, though, or the two directions of one flow would take
unrelated 4-tuples and neither side's traffic would arrive through the conntrack
or NAT entry the other's probe opened. So both sides hash the *same* sorted pair
and the higher-addressed side **negates** the result. When the port counts match,
the two rotations cancel exactly: our lane `s`'s 4-tuple is the reverse of the
peer's lane `s`. `laneBias` does the matching rotation on the flow hash for the
same reason.
That pairing only exists when our lane indices map one-to-one onto the peer's
ports, which is exactly the `txLanes == peerPortCount` test `newLaneSet` applies
before setting `laneBias`. If we send on 4 lanes and the peer bound 8 ports, four
of its ports have no lane of ours pointing at them, and no choice of rotation can
make our lane `s` and its lane `s` be each other's reverse. So in that case
`laneBias` stays 0 and each side hashes the flow to a lane on its own. The flow
still works and is still spread; the two directions simply take two unrelated
4-tuples instead of one 4-tuple and its exact reverse, and each direction depends
on its own lane's probe having opened its own conntrack or NAT entry.
## Bringing a lane up
Receiving on a lane needs no permission: the keys are derivable the moment the
base handshake completes. **Sending** on one needs proof the new 4-tuple actually
works, because nothing else would notice a middlebox quietly dropping it. So:
```
lane down --probe (Test/LaneProbe on lane s, from port base+s to peer's lane port)-->
<--ack (Test/LaneProbeAck, on the BASE tunnel)--
lane up
```
- The probe is encrypted with **the lane's own session**, so an ack proves the
whole lane end to end: our source port reached the peer, its reply reached us,
and the keys we derived match the ones it derived.
- The ack rides the **base tunnel** on purpose. A probe proves the peer's lane
works in *its* send direction; answering on our own lane `s` would make the
result depend on a second path that can be broken independently.
- The ack echoes the header's lane, not the payload's, so a peer cannot get us to
vouch for a lane it did not probe.
- A generation byte in the probe is echoed in the ack, so a late ack cannot
promote a lane on the strength of a superseded probe.
`txAddr[s]` is a single `atomic.Pointer[netip.AddrPort]` that is *both* the TX
gate and the destination, so a data-plane routine that loads non-nil has
everything it needs in one atomic read and there is no window where one is set
and the other is not.
Probing is driven by the connection manager's per-tunnel traffic tick
(`maintainLanes` -> `probeLanes`), which only fires for a tunnel with traffic --
the same condition that makes a lane worth having. Timers:
| timer | value |
|---|---|
| probe timeout | 2s (shorter than the 5s tick on purpose) |
| keepalive | 30s |
| retry backoff | 5s, doubling to 60s |
| max failure count | 8 |
Traffic on a lane is not evidence the lane works -- that is the whole reason
lanes need probing -- so a lane that silently breaks is only caught by the
keepalive.
Every lane starts out **demanded**, so the first traffic tick probes all of them
at once. This matters more than it looks: see "flow pinning" below. A lane that
is down also raises demand from the TX path each time a flow wants it, so a
lane that keeps failing is retried only while something still wants it, and an
idle tunnel costs nothing.
A lane aims **only** at its own port. There is no fallback to the peer's base
port when the lane port doesn't answer: a lane sharing the base port's
destination would gain only a source port of its own while costing the peer the
receive spread that is the entire point. A lane that can't reach its port stays
down and its flows ride the base tunnel.
## Flow -> lane -> routine: keeping a flow consistent
This is the part that took the most iterations to get right, so it's worth
spelling out why it is shaped the way it is.
### The kernel's tun queue feedback loop
On Linux, a multiqueue tun device does not simply hash a flow to a queue. It
also *learns*: `tun_flow_update` records "this flow was last seen on queue *q*"
from the packets **we write in**, and `tun_automq_select_queue` prefers what it
learned over the hash for as long as the flow stays busy.
We write an inbound packet to the tun queue with the same index as the UDP
routine that received it (`batchers[rxc.q]`). So the queue a flow's *outbound*
packets arrive on is decided by the socket its *inbound* packets landed on, at
the far end, one RTT ago.
### Why the lane comes from the flow, not the routine
The obvious design -- routine `q` sends on lane `q` -- deadlocks against that
feedback loop:
1. A tunnel comes up. All lanes are down, so all traffic goes out lane 0.
2. The peer receives it all on socket 0 and writes it all to tun queue 0.
3. Both kernels now believe every flow belongs on queue 0.
4. Every flow is read by routine 0, so every flow picks lane 0.
5. Go to 2. The tunnel is pinned to lane 0 for as long as its flows stay busy.
So the lane comes from **the flow's own 5-tuple**, not from the routine index:
```go
s = (laneFlowHash(fwPacket) + laneBias) % txLanes // lanes.go:txLaneForFlow
```
This makes lane spread completely independent of how the kernel steers tun
queues. It also gives the properties you actually want from a flow's point of
view:
- **A flow stays on one lane for its life.** The hash is a pure function of the
5-tuple, so there is no per-packet lane hopping and therefore no reordering
introduced by multiport.
- **One lane's replay window sees one stable set of flows.**
- **Both directions of a flow pick partner lanes.** `laneFlowHash` orders the two
endpoints before hashing, so it returns the same value from either end, and
`laneBias` lines the two sides' choices up. The two directions are exact
reverse 4-tuples, which is what NAT and stateful firewalls need.
`newLaneSet` demanding every lane up front is the other half of this. Lanes have
to be up *before* the flows are: a flow that starts while the lanes are still
down gets its queue pinned by step 2 above and can stay there for its whole
life. Eager demand costs one probe per lane on any tunnel that has traffic, and
nothing at all on one that doesn't.
### The full path
```
outbound inbound (at the peer)
-------- ---------------------
inside flow
| kernel tun hash, or the queue it
| learned from our last write
v
tun queue q -> routine q
|
| s = laneFlowHash(flow) % txLanes lane s arrives on port base+s
v |
lane s session | SO_REUSEPORT hash of the
| | 4-tuple picks one socket
| writers[laneSock(q,s)] v
v routine q' (owner of that socket)
port base+s -------------------------> |
v
tun queue q' (teaches the kernel
flow -> q')
```
Note that RX is entirely socket-agnostic: the lane comes from the header byte,
not from the port the packet arrived on, and roaming is skipped for `lane != 0`
(a lane's source address is a per-lane 4-tuple, not the tunnel's remote -- letting
it roam the hostinfo would point every non-lane packet at a lane port). So a
lane packet may legitimately arrive on any socket, which is what makes the
reuseport spread within a port safe.
### Ordering and locking
Several routines can write to one socket, since the routines whose lane
arithmetic lands on the same index share it. Linux's `batchWriter` serializes
`sendmmsg` with a mutex. Per-flow wire order still holds regardless: a flow is
hashed onto one lane and read by one routine, so nothing else is writing that
flow.
Each routine holds one `txQueue` with one shared arena (~1.16 MB) and builds a
`SendBatch` per lane lazily, on the first packet that picks it, so a routine that
never sends on a lane never pays for one.
### The FIPS 140 encrypt lock
In FIPS 140 mode -- a `boringcrypto` build, or `GODEBUG=fips140=on` -- nebula
uses `noiseutil.CipherAESGCMFIPS140` instead of the plain AES-GCM cipher. That
cipher is the TLS 1.3 GCM (`GCMWithXORCounterNonce`), which **panics** if it is
asked to seal with a counter that is not strictly greater than the last one. That
check is the point: it is the nonce-reuse protection FIPS 140 requires, and
`noiseutil`'s startup self-test refuses to run if the check has gone missing.
The check means encrypts on one session cannot overlap, so
`noiseutil.EncryptLockNeeded` is true and every send path takes that session's
`ConnectionState.writeLock`:
```go
if noiseutil.EncryptLockNeeded {
ci.writeLock.Lock()
}
c := ci.messageCounter.Add(1)
out = header.EncodeLane(scratch, ..., c, lane)
out, encErr := ci.eKey.EncryptDanger(out, out, seg, c, nb)
if noiseutil.EncryptLockNeeded {
ci.writeLock.Unlock()
}
```
The lock has to cover the seal and not just the counter increment. Reserving
counters atomically is easy; the requirement is that the seals *arrive at the
AEAD in counter order*, and only holding the lock across both gives that.
So under FIPS 140 the encrypt cost of a tunnel is pinned to one core. `routines:
N` gives N readers, but every one of them that has a packet for the same peer
queues on the same mutex, and with TSO/USO the lock is taken and released once
per segment -- up to ~45 times for a single superpacket
(`sendInsideEncrypt`). Receive is not affected: `extractFIPSAEAD` deliberately
pulls the inner FIPS AEAD out of the `crypto/tls` wrapper because that inner
implementation is safe to `Open` concurrently, and `decryptLock` is only ever
held around the replay-window check and update.
Lanes split the lock because a lane is a separate `ConnectionState`: its own
`writeLock`, its own message counter, its own nonce sequence. A peer we send to
on `txLanes` lanes has `txLanes` independent encrypt locks, so encrypt for that
one tunnel can run on that many cores at once.
The flow hash is what makes this work rather than merely legal. A flow maps to
one lane, so all of a flow's packets take one lock in counter order -- exactly
what the FIPS AEAD demands -- while different flows to the same peer land on
different locks. Contention is divided, not eliminated: any routine may send on
any lane, so two routines whose flows hash to the same lane still serialize
against each other.
Two caveats. This is only a FIPS 140 benefit -- in a normal build
`EncryptLockNeeded` is false, encrypt is lock-free, and lanes buy path and queue
spread only. And it is not unilateral: `txLanes` is bounded by the peer's port
count, so a peer that binds one port leaves us back on one encrypt lock however
many ports we bound ourselves.
## Falling back to the old behavior
Multiport degrades rather than failing, at every level. This is deliberate:
managed deployments can't be hard-errored on config they don't control.
### Whole-node: back to one port
Any of these turns multiport off, logs why, and leaves a node that binds one
port with `routines` `SO_REUSEPORT` sockets -- bit-for-bit the pre-multiport
configuration:
| condition | log |
|---|---|
| `multiport.enabled: false` | - |
| `multiport.ports` unset, 0, or 1 | `multiport disabled: set multiport.ports > 1 ...` |
| `listen.port + ports - 1 > 65535` | `multiport disabled: would bind ports beyond 65535` |
| platform can't run multiple UDP readers | `multiport disabled: this platform does not support multiple udp readers` |
| that capability couldn't be probed | `multiport disabled: could not probe udp reader support` |
`multiport.ports` is also clamped to 256 (the lane header limit) and to
`maxRoutines / routines`, with a warning, rather than being rejected.
With a dynamic `listen.port: 0`, the first socket binds dynamically and the
range is claimed above it; a partially-occupied range re-rolls with a fresh
dynamic port up to 6 times.
`laneSock` collapses to the identity when multiport is off, so the send path is
unchanged: a routine writes to its own socket and no lane batches are built at
all.
### Per-tunnel: back to a single session
`newLaneSet` returns `nil`, and the tunnel is an ordinary one, when:
- the peer advertised no port count (vanilla peer, or multiport off there), or
- `sessions < 2`, i.e. neither side offers a lane.
`txLanes` can also land at 1 -- we bound ports but the peer bound only one -- in
which case the set exists for RX but we never send on a lane.
### Per-packet: back to the base tunnel
`txLaneForFlow` returns a nil session and the packet rides the base tunnel,
decided per packet with no state to unwind:
- the flow hashed onto lane 0 (its fair share of flows);
- the lane it hashed onto has never come up;
- the lane was **demoted** -- a probe or keepalive went unanswered. Fallback is
immediate, on the very next packet, and the miss re-raises demand so the lane
is re-probed;
- the tunnel is **relayed** -- lanes are direct-only, so `probeLanes` drops them
all when there is no direct path and rebuilds when one returns;
- the peer **roamed** -- a new NAT mapping has no derivable relationship to the
old lane ports, so every lane is torn down and re-probed from a clean backoff.
Standing demand is deliberately kept across a reset, so the lanes that were
actually carrying data come back first.
### Always on the base tunnel
Handshakes, lighthouse traffic, punching, relay carriers, close packets, rejects,
and lane probe *acks* all use the base session and a socket on the base port
(`egressSock`). Base traffic must keep the base source port or a vanilla peer
would see the tunnel's address move and roam-thrash.
`recv_error` is the one deliberate exception: it replies from the socket the
offending packet arrived on, because a lane peer's spoof guard compares our
source address against that lane's remote and would discard a reply from the
base port.
### Wire compatibility
- A vanilla sender emits lane 0 in a field it thinks is reserved, and lane 0 is
the base tunnel, so it reads correctly with no version check.
- We always send the upper 8 reserved bits as zero.
- A lane index above what this tunnel has (a stale lane from a rolled tunnel, or
a peer sending above what it advertised) is dropped **silently** -- a
`recv_error` would tear down a perfectly good base tunnel on the strength of
one odd packet.
- Lane ciphertext wrapped in a relay carrier is refused before the session
lookup, so a junk relay packet can't make us derive a session.
## Security notes
- The lane index is in the AEAD's associated data, so it is authenticated, not
just carried.
- On RX, a lane session derived for an unrecognized lane is **not installed**
until the packet actually decrypts. Anyone who can spoof this tunnel's local
index can name any lane; installing on sight would let them make us hold a
replay window and two cipher states per lane, per tunnel, for lanes they never
send on.
- Two routines racing on a lane's first packet each derive a session. The loser's
is dropped, and its replay-window entry for the packet it just accepted is
handed to the winner (`installSession`) -- the keys are identical, so it is the
same window in every respect that matters.
## Configuration
```yaml
routines: 8 # PER PORT under multiport; total workers = routines * ports
multiport:
enabled: true # default true, but inert without ports
ports: 4 # consecutive ports from listen.port; must be > 1, no default
lanes: 0 # 0 = one per bound port; lower to send on a subset
```
`multiport.ports` has no default on purpose: under these semantics a default
would silently multiply the worker count. Nothing here is reloadable.
Both sides need a port range, and the range
`[listen.port, listen.port+ports-1]` must be open in both directions. Opening
only the base port is the common failure and gives you exactly one working lane
-- the one whose rotation happens to land on the base port.
## Observability
```
nebula-ssh> print-tunnel -vpn-addr <peer>
```
`lanes[]` gives per-lane `up`, `remote` and `messageCounter`, which is the fastest
way to tell "no lanes negotiated" (key absent) from "lanes up but traffic on one"
(counters).
Metrics, registered only when multiport is running so they don't sit at zero on
nodes without it:
- `multiport.lanes.up` -- lanes currently carrying traffic
- `multiport.lanes.tunnels` -- tunnels with any lanes
Both are counted by walking the hostmap rather than kept at promotion/demotion,
because a counter would drift upward forever: a tunnel torn down with its lanes
up never demotes them.
Logs worth grepping: `multiport enabled` and `multiport routines` at startup, the
`lanes` attr on handshake completion (`tx`, `sessions`, `peerBasePort`,
`peerPorts`, `portOffset`), and `Multiport lane up` / `Multiport lane demoted`,
both of which name the `udpAddr` involved.
## Known gaps
- No `readOutsidePackets`-level test for the RX lane drop paths.
- No multiport coverage in the e2e suite.
- `routines * ports > 256` fails at startup from the kernel's tun queue limit
rather than being clamped with a warning.
-131
View File
@@ -1,131 +0,0 @@
# Post-quantum handshake transport compared with QUIC
Status: analysis, 2026-10-14. Compares the fragment transport in
[pq-handshake.md](pq-handshake.md) with how QUIC makes its handshake reliable. TLS is out of
scope; only the transport mechanics are compared.
## Summary
QUIC's handshake reliability is closed-loop: the receiver acknowledges, and the sender
repairs exactly what is missing. Ours is open-loop: repair symbols are sent up front, and a
timer sends fresh symbols. At handshake sizes (K <= 15) a QUIC-style acknowledgement
shrinks to a 16-bit bitmap, and with an MDS code to a single count. So the real comparison
is forward error correction versus acknowledgements, not QUIC versus us.
FEC is the right core. One piece of feedback is worth adding, initiator to responder and for
msg2 only (see [Worth taking from QUIC](#worth-taking-from-quic)).
## What QUIC does
- **Messages as a stream.** Handshake bytes travel as `CRYPTO` frames (offset and length)
and are reassembled in order. Initial and Handshake packets each have their own packet
number space.
- **Immediate acknowledgements.** Receivers must acknowledge Initial and Handshake packets
without delay. ACK frames list ranges of received packet numbers.
- **Loss detection** (RFC 9002). A packet is lost once 3 later packets are acknowledged, or
after 9/8 of the round-trip time. Lost bytes are resent in new packets.
- **Probe timeout (PTO).** If a whole flight vanishes, a probe timer fires. Before any RTT
sample it starts at about 1 s (333 ms initial RTT plus 4x its variance) and doubles each
time. Both endpoints run it.
- **Amplification limit and anti-deadlock.** The server may send at most 3x what it
received until the client's address is validated. The client must keep probing so a
server stuck at that limit cannot deadlock.
- **Retry.** A stateless round trip to validate the address. `NEW_TOKEN` lets later
connections skip it.
- **Congestion control from the first packet.** The initial window is 10 packets of 1200
bytes.
## Side by side
| | QUIC | pq-handshake |
|---|---|---|
| Receiver feedback | ACK ranges, sent immediately | None; the next handshake message is the only acknowledgement |
| Partial loss | +1 to 2 RTT (ACK, detect, resend) | 0 RTT while at most m symbols are lost |
| Total loss | PTO: about 1 s, then doubling | Next PQ tick: 300 ms, then 1 s (alternating), or 100/300/600 ms (sticky) |
| Spurious retransmit | Wasted bytes | Still useful: fresh repair indexes add to what the receiver holds |
| Reassembly | Every byte offset, in order | Any K of up to 256 symbols |
| Who retransmits | Both sides on their own timers | Initiator on a timer; the responder resends msg2 only when a repeat msg1 arrives (rate limited) |
| State from the first packet | Server buffers and acknowledges partial flights | Capped pre-authentication table that never acknowledges |
| Address validation | Retry, 1 RTT | Cookie under load, 1 RTT |
From the [simulation results](pq-handshake.md#simulation-results): phase 2 at 5% random
loss with no FEC completes on the first attempt 39.8% of the time. Under
acknowledgement-based recovery roughly 60% of handshakes would pay at least one extra round
trip. With FEC, 0.4% wait for a retry. Spending bandwidth to save round trips is the stated
design preference, and open-loop recovery delivers it.
## Where the designs already agree
- **1200-byte floor and the 3x amplification budget**, both taken directly from QUIC.
- **Cookie and Retry.** Both cost one round trip and both can be forged by an on-path
attacker. QUIC's Retry integrity tag uses a fixed, published key.
- **On-path injection.** QUIC's Initial keys are derived from the client's connection ID,
which an observer can read, so QUIC accepts the same handshake-blocking injection
weakness documented for our symbols in
[Injected symbols](pq-handshake.md#injected-symbols).
- **Phase 2 msg3 confirmation.** Resending msg3 until any authenticated packet arrives,
with a `Test` from the responder on completion, is QUIC's handshake confirmation and
`HANDSHAKE_DONE`.
## Worth taking from QUIC
### 1. A msg2 progress probe from the initiator
This is the main gap. Today, recovering a partially lost msg2 means the initiator resends a
full msg1 attempt; the responder decodes it, matches it in the answered-msg1 cache, and
resends msg2. In phase 2 that is about 12 fragments (~13 KB) uploaded just to act as an
acknowledgement, and recovery waits for the 300 ms or 1 s tick.
- The initiator already knows the responder's index: msg2 fragments carry it as `FragID`
(`sendFragmentedResponse` passes `hostinfo.localIndexId`).
- A one-packet probe, "msg2: have n of K", sent to that index would recover in about one
round trip. The responder finds it the way phase 2 plans to find msg3.
- Timing comes for free: msg1 send to first msg2 fragment is an RTT sample. Probe at
roughly RTT/8 plus the repair delay after the first msg2 fragment if decoding has not
completed.
- Forging a probe needs the 32-bit responder index, so only an on-path attacker can, and
the existing 3x budget bounds the reply.
- If no msg2 fragment arrived at all, the initiator keeps today's msg1 resend. That is the
same tail-loss case QUIC's PTO covers.
The reverse direction, the responder reporting progress on msg1, should not be added. It
needs per-entry timers on the pre-authentication path, which is exactly the state QUIC
servers carry and the reason QUIC needs Retry.
### 2. Buffer early data in phase 2
QUIC stacks hold a few packets they cannot decrypt yet. The phase 2 plan drops data that
reaches the responder before msg3 decodes. Because repair fragments trail by
`handshakes.pq.repair_delay` (10 ms), losing one msg3 data fragment drops about 10 ms of
the initiator's first data while the repair is in flight. A small queue on the pending
responder entry, replayed once msg3 verifies, avoids this.
### 3. Burst size across remotes
QUIC handshakes on a single path. `handleOutbound` sends the same `now` and `later` sets to
every known remote and to relays (`StartRelays`).
- Phase 1 is fine.
- In phase 2, four remotes means about 48 packets (~54 KB) leaving one uplink at once,
every PQ tick, about 4x QUIC's initial window.
- The Gilbert-Elliott simulation treats each path independently, so it does not capture
this self-inflicted burst on a constrained uplink such as cellular.
- Options: send the full FEC set only to the most likely remote and a reduced set to the
others, or interleave packets across remotes.
## Not worth taking
- **Packet numbers, ACK ranges, an RTT estimator, loss thresholds, congestion control.**
QUIC pays for this machinery because it also serves a data transport. Nebula carries
packets and leaves reliability to the overlay's own transports. Building it only for
handshakes of at most 15 fragments buys nothing a symbol count cannot.
- **Offset-based reassembly.** It gives up the any-K-of-256 property and makes spurious
retransmits wasteful again.
- **An RTT-adaptive retry timer.** Low value. The fixed 100 ms linear schedule already fires
sooner than QUIC's 1 s first PTO, and a spurious resend costs bandwidth, not correctness.
## Conclusion
Keep FEC as the core and add the msg2 progress probe. That gets QUIC's roughly one round
trip tail recovery without pre-authentication state, and cuts phase 2's retry cost on the
uplink by about 10x. Items 2 and 3 are phase 2 concerns; item 1 also helps phase 1.
-348
View File
@@ -1,348 +0,0 @@
# Post-quantum handshakes
Status: draft. Decisions below were made on 2026-10-13; open items are listed at the end.
Phase 1 is implemented (see [Implementation](#implementation-phase-1)).
## Goals
1. **Confidentiality against harvest-now-decrypt-later.** Traffic recorded today must stay
secret after a cryptographically relevant quantum computer exists. This needs a
post-quantum key exchange now.
2. **Post-quantum authentication, later.** Needs certificates whose CA signature and host
keys are post-quantum. This requires reissuing certificates, so it ships second.
3. **Never rely on IP fragmentation.** ML-KEM and ML-DSA do not fit in one packet, so the
handshake must carry messages larger than the path MTU itself.
Non-goals: identity hiding beyond what IX offers today, and signatures at handshake time.
The only signature in the system stays the CA signature inside the certificate.
## Primitives and sizes
| Primitive | Size |
|---|---|
| ML-KEM-1024 encapsulation key | 1568 B |
| ML-KEM-1024 ciphertext | 1568 B |
| ML-DSA-87 public key / signature | 2592 B / 4627 B |
| Today's IX msg1 / msg2 (v2 cert, 151 B) | 252 B / 287 B |
`crypto/mlkem` is available in Go 1.26 (the current `go.mod`); `crypto/mldsa` arrives in
Go 1.27 and is only needed for phase 2.
## Phase 1: ephemeral ML-KEM on today's certificates
A new handshake subtype, `ix_psk2_mlkem1024` (2), IX-shaped (2 messages; the responder
completes on msg1 as today), using the flynn/noise `psk2` modifier. The Noise protocol name
becomes `Noise_IXpsk2_...`, which separates it from the classical subtype.
- **msg1**: the initiator generates a fresh ML-KEM-1024 key pair and places the
encapsulation key in `NebulaHandshakeDetails.KEMKey` (field 9). msg1's payload is mixed
into the Noise transcript, so it cannot be stripped or replaced without msg2 failing.
With the psk modifier Noise also mixes `e` into the cipher key, so msg1's static key and
payload are encrypted, though only under a key derived from the public ephemeral.
- **Responder**: after `ReadMessage(msg1)` it encapsulates to that key, obtaining
`(ct, ss)`, and calls `HandshakeState.SetPresharedKey(psk)` before writing msg2, with
`psk = HKDF-Extract(SHA-256, salt = "nebula ix_psk2_mlkem1024 v1", ikm = ss || ct)`.
The ciphertext is bound in because, unlike the encapsulation key, it is not part of the
Noise transcript.
- **msg2** on the wire is `ct || noise_msg2`. The ciphertext travels outside the Noise
message because the initiator needs `ss` before processing msg2's `psk` token. Any change
to `ct` yields a different `ss` (ML-KEM implicit rejection) and the handshake fails.
Transport keys depend on both the X25519 DH outputs and the ML-KEM shared secret.
Authentication is unchanged and classical. With a minimal v2 certificate msg1 and msg2 are
about 1.84 KB each, so each travels as 2 data and 2 repair fragments of about 936 bytes.
A KEM key on the classical subtype, or a missing or malformed one on this subtype, is a fatal
protocol error; a forged ciphertext only fails that packet.
## Phase 2: pqIX hybrid with certificate v3
Certificate v3 is hybrid: the classical static key and CA signature plus a static
ML-KEM-1024 key and an ML-DSA-87 CA signature. Host authentication is by KEM rather than by
signature, following PQNoise's pqIX, while keeping the classical DH tokens:
- **msg1** (I to R): `e, s`, ephemeral ML-KEM key, initiator cert.
- **msg2** (R to I): `e, s`, ciphertext to I's ephemeral KEM key, ciphertext to I's static
KEM key, responder cert.
- **msg3** (I to R): ciphertext to R's static KEM key, plus first data. R authenticates I
when msg3 decrypts; I authenticates R implicitly when R's first reply decrypts.
msg3 is about 1.6 KB, 2 data and 2 repair fragments, and adds no new message types:
- **Responder.** After sending msg2 the responder keeps a pending Machine, found by its
local index like the initiator's today; this is the multi-message responder support the
manager lacks now. A repeat msg1 maps to it through the answered-msg1 cache and gets msg2
resent with fresh symbols, exactly as in phase 1. Until msg3 verifies, data for that
index is dropped silently: it must not draw a RecvError, which would tear down the
initiator's new tunnel. On completing it sends an empty `Test`/`TestReply` under the
new keys, so the initiator hears back even when there is no traffic.
- **Initiator.** Writing msg3 yields the transport keys, so the hostinfo moves to the main
hostmap and data flows at once, readable only by the real responder. The handshake entry
stays behind in a confirming state: each retry tick resends msg3 with fresh symbols until
any packet from the responder decrypts on the new tunnel, and if that never happens
within the handshake timeout the tunnel is torn down. Losing msg3 costs the data sent
behind it, which the overlay's own transport retransmits.
- msg3's fragments use the responder's index as RemoteIndex and land on the responder's
pending entry, so they never enter the pre-authentication table.
The initiator never caches the responder's certificate, so every handshake has three
messages and carries both full certificates. With ML-DSA-87:
| Message | Size | Data fragments (k) |
|---|---|---|
| msg1 | ~8.2 KB | 8 |
| msg2 | ~9.8 KB | 9 |
| msg3 | ~1.6 KB | 2 |
flynn/noise cannot mix multiple KEM secrets into the chaining key, so phase 2 needs a
KEM-aware Noise symmetric state written in this repository. Both subtypes coexist.
Cryptographic review of the hybrid composition will be done by in-house cryptographers.
## Fragment transport
### Requirements
- No IP fragmentation: every datagram's UDP payload is at most 1232 bytes **after** relay
wrapping (16 B header + 16 B tag), the IPv6 minimum MTU budget also used by DNS Flag Day
2020 and Solana's Turbine.
- Every fragment of a handshake uses the same 5-tuple.
- Bounded pre-authentication state, bounded amplification, selective loss recovery,
tolerance of reordering and duplication, and a strictly validated, fuzzable parser.
### Wire format
Every packet of a fragmented handshake subtype is:
```
Nebula header (16) Version | Type=Handshake | Subtype | Reserved=0 | RemoteIndex | Counter
Fragment header (8) FragID u32 | MsgLen u16 | Index u8 | Flags u8
Cookie (16) only when Flags has 0x01
Symbol SymbolSize bytes
```
- `Counter` is the Noise message number (1, 2, 3). `RemoteIndex` is 0 for msg1, as today.
- `FragID` is the sender's local index (random, non-zero).
- `MsgLen` is the length of the whole message. Both ends derive the layout from it:
`K = ceil(MsgLen / 1160)`, `SymbolSize = ceil(MsgLen / K)`. Every symbol has the same
size. Room for a cookie is always reserved, so a packet is at most
16 + 8 + 16 + 1160 = 1200 bytes with or without one.
- `Index < K` is a data symbol (a slice of the message, the last one zero padded);
`Index >= K` is a repair symbol.
- `Flags` bit 0x01 means a cookie follows the fragment header. Other bits must be zero.
A cookie reply, sent by a responder under load, is a bare 32 byte packet:
```
Nebula header (16) Version | Type=Handshake | Subtype | Reserved=0 | RemoteIndex=FragID | Counter=0
Cookie (16)
```
It names the initiator's FragID, so the initiator finds its attempt the same way it finds
msg2.
### Forward error correction
Messages are protected with a systematic Reed-Solomon code over GF(2^8), implemented in
`handshake/fec`:
- The repair rows form a Cauchy matrix, `coefficient(row, col) = 1 / (row XOR col)` with
`row >= K > col`, so every square submatrix is invertible and **any K distinct symbols
reconstruct the message**. Naive identity-over-Vandermonde constructions are not MDS
(the PAR1 flaw and the 1997 Plank tutorial erratum).
- A repair row depends only on its index, so repair symbols are generated on demand and
their indexes never need to be agreed in advance. Up to 256 symbols exist per message.
- Known-answer vectors in `fec_test.go` pin the encoding and were cross-checked against an
independent implementation; changing them is a wire format change.
- Cost for a 9.8 KB message (k=9): about 13 µs for three repair symbols, about 19 µs to
decode with three data symbols missing.
**The sender chooses m**, the number of repair symbols per attempt. The receiver only needs
any K symbols, so this is policy and can change without a protocol change. The starting
rule is `m = max(2, ceil(K/3))`.
**Retries send fresh symbols.** Attempt 0 sends indexes `0 .. K+m-1`; each retry sends
the next `K+m` unused indexes. The receiver merges all distinct symbols across attempts,
so no received fragment is wasted. Indexes are a deterministic function of the message,
so retransmits stay idempotent: the responder caches msg2 itself and regenerates any
symbol.
**Repair symbols trail the data.** Data symbols go out back to back; repair symbols follow
after a short delay, so a queue tail-drop at the end of the data run does not also take the
repair. In simulation this lifts small messages substantially (k=2 at 10% loss, burst 8:
91.4% to 98.2% first-attempt success). Pacing every fragment was also evaluated; it gains
about 5 more points for k=7..9 on bursty paths, which is not worth the extra machinery,
and the retry with fresh symbols covers the remainder.
### Receiver rules
Implemented in `handshake/fragment` (`Sender`, `Assembly`, and the capped `Table`).
- Fragments are stored individually as they arrive and charged against a memory cap.
Nothing is allocated from a claimed size alone (FragmentSmack, game-engine "fragment
bombs").
- `MsgLen` must be identical for every fragment of a `FragID`; a mismatch drops the whole
entry. msg1 is capped at `MsgLen <= 16384` (K <= 15).
- The decoder keeps at most K distinct symbols, so an entry never exceeds about MsgLen bytes
however many repair symbols arrive.
- When K symbols are present the message is decoded and handed to the handshake `Machine`
as a synthetic `[header | message]`. The Machine never sees fragments.
- A completed entry is forgotten immediately. A late duplicate fragment starts a new entry
and, once enough arrive, reconstructs the same message again; the handshake layer's
duplicate detection recognises it.
- If Noise rejects a decoded message, the entry is discarded and the next attempt
supplies fresh symbols. (Possible refinement: keep a few spare symbols so an injected
bad symbol can be identified by re-encoding instead of waiting for a retry.)
- Pre-authentication msg1 entries live in a table keyed by (source address or relay index,
FragID) with a total memory cap, per-source quotas (/32 for IPv4, /64 for IPv6), a
sliding TTL of about 2 s refreshed by progress, and random eviction under pressure.
msg2 and msg3 reassembly attaches to the existing pending handshake found by RemoteIndex.
### Retransmission and amplification
- Only timers cause transmission. Reactive resends (the responder resending its cached
msg2 when a duplicate msg1 completes) are rate limited to one per handshake per retry
interval, to avoid TFTP's Sorcerer's Apprentice amplification.
- Duplicate msg1 detection compares the decoded message to the cached one **before**
running Noise or certificate verification.
- Before an address is validated, a responder sends at most 3x the bytes it received from
it (QUIC's rule). Normal handshakes sit near 1.1-1.2x.
- In phase 2 the initiator retransmits msg3 until any authenticated packet arrives from
the responder; the responder sends a small confirmation on completion and holds tunnel
traffic until msg3 verifies.
### Denial of service
The responder can no longer complete statelessly on msg1, since msg1 arrives in pieces.
Defenses are the capped table above plus a **cookie under load**, in the style of
WireGuard's mac2, QUIC Retry and the SCTP state cookie:
- Once the msg1 table is at least half full (by bytes or by entries), every msg1 fragment
must carry a valid cookie. One without is answered with a cookie reply and leaves no
state. Fragments that already belong to an entry need one too, so an attacker cannot
hold old entries open.
- A cookie is `HMAC-SHA256(secret, source address and port, relay channel, FragID)`
truncated to 16 bytes. The secret rotates every 2 minutes and the previous one is still
accepted, so a cookie is good for 2 to 4 minutes.
- Replies are never larger than the packet that prompted them (no amplification) and are
rate limited as a whole (a burst of 256, then 1000 per second), so spoofed floods cannot
turn the responder into a reflector.
- The initiator attaches the cookie to its PQ attempt, so every later packet carries it,
and immediately resends fresh symbols to the address the reply came from, at most once
per retry interval. Under load a PQ handshake costs one extra round trip, which in
practice still beats the classical attempt on the next retry tick.
- An attacker on the path can forge a cookie reply; the bogus cookie is simply rejected
and answered with a real one.
A CA-keyed pre-filter MAC was considered and rejected: networks commonly trust many CAs.
#### Injected symbols
The decoder keeps the first K distinct symbols it sees, so one forged symbol in those K
spoils that attempt's decode; Noise rejects the result and the next attempt brings fresh
symbols. Forging one needs the attempt's FragID or index, which are random 32 bit values,
and for msg1 the initiator's exact address and port too, so in practice only an attacker
who sees the traffic can do it. An attacker who can also drop packets gains nothing new.
One who can only inject, and races a forged symbol into every attempt, can stop a PQ
handshake where it could not stop a classical one, since classical msg2s are tried one
packet at a time. Keeping spare symbols to find a consistent subset only raises the number
of forgeries needed per attempt, and authenticating symbols is impossible for msg1, which
travels before any key exists. So the decoder deliberately does neither. A non-sticky
initiator still gets a classical tunnel from its next attempt; a sticky one, by design,
does not downgrade.
## Rollout
- Separate settings for initiating and accepting PQ handshakes, both off by default:
`handshakes.pq.initiate`, `handshakes.pq.accept`. `handshakes.pq.repair_delay` (default
10ms) sets how long repair fragments trail the data fragments.
- `handshakes.pq.require` makes every handshake PQ: it implies `initiate` and `accept`,
never offers a classical attempt, and drops classical msg1s.
- Old nodes silently drop unknown subtypes, so discovery is speculative: an initiator with
`initiate` enabled alternates PQ and classical attempts on successive timer retries,
keeping both pending, and the first to complete wins. A path nothing has been sent over
yet gets the PQ attempt first: an early resend triggered by a lighthouse reply is PQ, and
a relay that comes up pushes the PQ attempt through at once, counting as the PQ turn.
Without that, relays set up during the first attempt handed every relayed handshake to
the classical retry (seen on real hosts: 11 of 11). Once a msg2 fragment or cookie reply
arrives for the PQ attempt the peer evidently speaks PQ, and classical stops getting
turns.
- All `handshakes.pq` settings reload without a restart. Handshakes in flight carry on;
the new settings apply to the next attempt and the next packet.
- `handshakes.pq.sticky` (default off): once a peer has completed a PQ handshake, never fall
back to classical for it again during this process lifetime. This resists active
downgrade after first contact.
- No lighthouse capability advertisement.
## Simulation results
Gilbert-Elliott channel (bad state drops everything), losses expressed as average rate and
mean burst length in packets. First-attempt success for the whole handshake with
`m = max(2, ceil(K/3))` and repair delayed 16 packet slots, compared with no FEC:
| Handshake | Loss | Burst 1 | Burst 2 | Burst 4 | Burst 8 |
|---|---|---|---|---|---|
| phase 1 (2, 2) no FEC | 5% | 81.4% | 85.6% | 87.9% | 89.0% |
| phase 1 (2, 2) FEC | 5% | 99.9% | 99.4% | 99.3% | 98.6% |
| phase 1 (2, 2) FEC | 10% | 99.3% | 97.5% | 97.4% | 96.3% |
| phase 2 (8, 9, 2) no FEC | 5% | 37.8% | 56.0% | 69.3% | 77.1% |
| phase 2 (8, 9, 2) FEC | 5% | 99.6% | 93.7% | 89.9% | 88.4% |
| phase 2 (8, 9, 2) FEC | 10% | 95.3% | 83.5% | 77.9% | 76.7% |
With fresh repair symbols, phase 2 completes within two attempts at least 99.1% of the time
in every scenario above, and within three at least 99.98%.
The simulators and their outputs are in [pq-handshake-sim](pq-handshake-sim/README.md),
including a comparison with classical handshakes and a model of time to a working tunnel
under Nebula's retry schedule.
## Prior art
- **WireGuard / Rosenpass**: fit every message in one packet; cookies and biscuits for
statelessness. Not possible at our sizes.
- **IKEv2**: RFC 7383 fragments only after keys exist, with each fragment authenticated;
RFC 9242/9370 move PQ key exchange into intermediate exchanges for that reason.
- **QUIC / DTLS 1.3**: offset-based reassembly with ACKs; QUIC's 3x amplification limit and
1200-byte floor.
- **Solana Turbine**: Reed-Solomon over 1232-byte UDP packets through untrusted relays;
moved to Merkle-authenticated shreds after unauthenticated repair symbols enabled
poisoning.
- **kcp-go, UDPspeeder, CM256**: RS FEC over UDP; parity must scale up for small groups.
- **gQUIC FEC (removed)**: FEC does not pay for bulk transfer under bursty loss; it does
for short, latency-critical transfers such as handshakes.
- **TFTP, NFS duplicate request cache, memcached UDP, IP reassembly bugs**: retransmit
storms, ID-wrap misassembly, reflection amplification, and allocation from claimed sizes.
## Implementation (phase 1)
- `handshake`: the `ix_psk2_mlkem1024` subtype (`patterns.go`, `kem.go`), the `KEMKey`
payload field, and the psk2 plumbing in `Machine`.
- `handshake/fec`, `handshake/fragment`: the code and the fragment transport.
- `handshake_manager_pq.go` and the manager:
- **Initiator.** Attempts alternate PQ and classical, starting with PQ (sticky peers get
only PQ). The PQ attempt has its own local index registered alongside the classical
one; whichever response completes first wins, and the other index is released
(`settleIndex`). A timed out or deleted pending handshake releases both.
- **Responder.** msg1 fragments go into the capped pre-authentication table. A decoded
msg1 is looked up by hash among recently answered ones before any Noise, KEM or
certificate work: the trailing repair fragments of an attempt rebuild a msg1 its data
fragments already completed, and a retrying initiator offers fresh symbols for the
same msg1. A repeat goes straight to the cached-response resend, which offers fresh
msg2 symbols, is rate limited to one per retry interval and is held to the 3x
amplification budget.
- **Relays** carry fragments unchanged; repair fragments trail via a timer.
- The pre-authentication table is capped at 4 MiB, 1024 entries, 4 per source (IPv4 /32,
IPv6 /64, or relay channel), 16 KiB messages and a 3 s idle timeout. Cookies are
demanded from half full.
- Tests: Noise-level KEM tests, codec and fragment tests and fuzzers, manager unit tests
for the response budget and rate limit, and e2e tests for a clean handshake, loss of
data fragments in both directions, fallback to a peer that does not accept PQ, a lost
msg2 recovered by fresh symbols, sticky mode under an attacker dropping PQ packets,
relays, index cleanup on completion, fallback and timeout, and a responder flooded from
spoofed sources demanding and accepting a cookie. Unit tests cover cookie binding and
rotation, the never-amplify rule, the initiator's rate limited resend, and reload. ML-KEM and HKDF also
pass under `GOEXPERIMENT=boringcrypto` and `GODEBUG=fips140=on`.
## Open items
None for phase 1. Phase 2 needs certificate v3, the KEM-aware Noise symmetric state, and
the msg3 design above.
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
-476
View File
@@ -1,476 +0,0 @@
//go:build e2e_testing
package e2e
import (
"net/netip"
"sync"
"testing"
"time"
"github.com/slackhq/nebula"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/cert_test"
"github.com/slackhq/nebula/e2e/router"
"github.com/slackhq/nebula/handshake/fragment"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/udp"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func pqRequire() m {
return m{"handshakes": m{"pq": m{"require": true, "repair_delay": "0s"}}}
}
func pqConfig(initiate, accept, sticky bool) m {
return m{"handshakes": m{"pq": m{
"initiate": initiate,
"accept": accept,
"sticky": sticky,
"repair_delay": "0s",
}}}
}
// handshakeOf returns the handshake header p carries, looking inside relay
// wrapping, plus the fragment header when the handshake is fragmented.
func handshakeOf(p *udp.Packet) (*header.H, *fragment.Header) {
b := p.Data
h := &header.H{}
if h.Parse(b) != nil {
return nil, nil
}
if h.Type == header.Message && h.Subtype == header.MessageRelay {
// Relays authenticate but do not encrypt the inner packet.
b = b[header.Len:]
if h.Parse(b) != nil {
return nil, nil
}
}
if h.Type != header.Handshake {
return nil, nil
}
if !header.IsFragmented(h.Subtype) {
return h, nil
}
// A relayed packet still has the relay's AEAD tag on the end.
for _, tag := range []int{0, 16} {
if len(b) < header.Len+tag {
break
}
if fh, _, err := fragment.Parse(b[header.Len : len(b)-tag]); err == nil {
return h, &fh
}
}
return h, nil
}
// handshakeTally counts handshake packets by style as the router moves them.
type handshakeTally struct {
sync.Mutex
pq, classical int
maxPQLen int
}
func (ht *handshakeTally) observe(p *udp.Packet) {
h, _ := handshakeOf(p)
if h == nil {
return
}
ht.Lock()
defer ht.Unlock()
if h.Subtype == header.HandshakeIXPSK2MLKEM1024 {
ht.pq++
ht.maxPQLen = max(ht.maxPQLen, len(p.Data))
} else {
ht.classical++
}
}
// isTunnelData reports whether p is ordinary tunnel traffic headed for to,
// directly or as the last hop of a relay.
func isTunnelData(p *udp.Packet, receiver, to *nebula.Control) bool {
h := &header.H{}
if receiver != to || h.Parse(p.Data) != nil || h.Type != header.Message {
return false
}
if h.Subtype == header.MessageRelay {
if h.Parse(p.Data[header.Len:]) != nil {
return false
}
}
return h.Type == header.Message && h.Subtype == header.MessageNone
}
func newPQPair(t *testing.T, mine, theirs m) (myControl, theirControl *nebula.Control, myVpn, theirVpn []netip.Prefix, myUdp, theirUdp netip.AddrPort, r *router.R) {
t.Helper()
ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
myControl, myVpn, myUdp, _ = newSimpleServer(cert.Version2, ca, caKey, "me", "10.128.0.1/24", mine)
theirControl, theirVpn, theirUdp, _ = newSimpleServer(cert.Version2, ca, caKey, "them", "10.128.0.2/24", theirs)
myControl.InjectLightHouseAddr(theirVpn[0].Addr(), theirUdp)
r = router.NewR(t, myControl, theirControl)
myControl.Start()
theirControl.Start()
return
}
func TestPQHandshake(t *testing.T) {
t.Parallel()
// They only accept, which is all a PQ initiator needs.
myControl, theirControl, myVpn, theirVpn, myUdp, theirUdp, r := newPQPair(t, pqConfig(true, true, false), pqConfig(false, true, false))
defer r.RenderFlow()
var tally handshakeTally
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(5*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
assertUdpPacket(t, []byte("Hi from me"), theirControl.GetFromTun(true), myVpn[0].Addr(), theirVpn[0].Addr(), 80, 80)
assert.Zero(t, tally.classical, "no classical handshake should be needed")
assert.GreaterOrEqual(t, tally.pq, 4, "both messages travel as several fragments")
assert.LessOrEqual(t, tally.maxPQLen, fragment.MaxPacketLen)
assertHostInfoPair(t, myUdp, theirUdp, myVpn, theirVpn, myControl, theirControl)
assertTunnel(t, myVpn[0].Addr(), theirVpn[0].Addr(), myControl, theirControl, r)
assert.Empty(t, myControl.ListHostmapIndexes(true), "no pending index may outlive the handshake")
assert.Empty(t, theirControl.ListHostmapIndexes(true))
myControl.Stop()
theirControl.Stop()
}
// TestPQHandshakeLoss drops data fragments of both messages; the repair
// fragments must carry the handshake without a retry.
func TestPQHandshakeLoss(t *testing.T) {
t.Parallel()
myControl, theirControl, myVpn, theirVpn, _, _, r := newPQPair(t, pqConfig(true, true, false), pqConfig(false, true, false))
defer r.RenderFlow()
var tally handshakeTally
dropped := 0
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(5*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
if h, fh := handshakeOf(p); fh != nil {
// Every data fragment of msg1, and the first of msg2.
if (h.MessageCounter == 1 && int(fh.Index) < 2) || (h.MessageCounter == 2 && fh.Index == 0) {
dropped++
return router.Drop
}
}
tally.observe(p)
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
assertUdpPacket(t, []byte("Hi from me"), theirControl.GetFromTun(true), myVpn[0].Addr(), theirVpn[0].Addr(), 80, 80)
assert.Equal(t, 3, dropped)
assert.Zero(t, tally.classical, "the first attempt should succeed from repair fragments")
assertTunnel(t, myVpn[0].Addr(), theirVpn[0].Addr(), myControl, theirControl, r)
myControl.Stop()
theirControl.Stop()
}
// TestPQFallback has a peer that does not accept PQ handshakes. The initiator's
// next attempt is classical and the tunnel comes up anyway.
func TestPQFallback(t *testing.T) {
t.Parallel()
myControl, theirControl, myVpn, theirVpn, myUdp, theirUdp, r := newPQPair(t, pqConfig(true, true, false), nil)
defer r.RenderFlow()
var tally handshakeTally
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(5*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
assertUdpPacket(t, []byte("Hi from me"), theirControl.GetFromTun(true), myVpn[0].Addr(), theirVpn[0].Addr(), 80, 80)
assert.Positive(t, tally.pq, "a PQ attempt comes first")
assert.Positive(t, tally.classical, "then a classical one")
assertHostInfoPair(t, myUdp, theirUdp, myVpn, theirVpn, myControl, theirControl)
assertTunnel(t, myVpn[0].Addr(), theirVpn[0].Addr(), myControl, theirControl, r)
assert.Empty(t, myControl.ListHostmapIndexes(true), "the losing PQ attempt's index must be released")
myControl.Stop()
theirControl.Stop()
}
// TestPQResendFreshSymbols loses the responder's whole first msg2. The
// initiator's next PQ attempt offers fresh msg1 symbols, the responder
// recognises the repeat and resends msg2 with symbols it has not sent before.
func TestPQResendFreshSymbols(t *testing.T) {
t.Parallel()
myControl, theirControl, myVpn, theirVpn, _, _, r := newPQPair(t, pqConfig(true, true, false), pqConfig(false, true, false))
defer r.RenderFlow()
var firstMsg2, freshMsg2, freshMsg1 int
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(10*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
h, fh := handshakeOf(p)
switch {
case h != nil && fh == nil:
// Keep the classical fallback out of the way.
return router.Drop
case fh != nil && h.MessageCounter == 2 && fh.Index < 4:
firstMsg2++
return router.Drop
case fh != nil && h.MessageCounter == 2:
freshMsg2++
case fh != nil && h.MessageCounter == 1 && fh.Index >= 4:
freshMsg1++
}
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
assertUdpPacket(t, []byte("Hi from me"), theirControl.GetFromTun(true), myVpn[0].Addr(), theirVpn[0].Addr(), 80, 80)
assert.Equal(t, 4, firstMsg2, "the first msg2 is 2 data and 2 repair fragments")
assert.Positive(t, freshMsg1, "the retry offers fresh msg1 symbols")
assert.Positive(t, freshMsg2, "the resend offers fresh msg2 symbols")
assertTunnel(t, myVpn[0].Addr(), theirVpn[0].Addr(), myControl, theirControl, r)
myControl.Stop()
theirControl.Stop()
}
// TestPQSticky checks a sticky initiator never offers a classical handshake to a
// peer it has already completed a PQ handshake with.
func TestPQSticky(t *testing.T) {
t.Parallel()
myControl, theirControl, myVpn, theirVpn, _, _, r := newPQPair(t, pqConfig(true, true, true), pqConfig(false, true, false))
defer r.RenderFlow()
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(5*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
theirControl.GetFromTun(true)
t.Log("Rehandshake while an attacker drops every PQ packet")
var tally handshakeTally
myControl.ReHandshake(theirVpn[0].Addr())
r.RouteForAllExitFuncOrTimeout(time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
if h, _ := handshakeOf(p); h != nil && h.Subtype == header.HandshakeIXPSK2MLKEM1024 {
return router.Drop
}
return router.KeepRouting
})
assert.Positive(t, tally.pq, "PQ attempts keep being made")
assert.Zero(t, tally.classical, "sticky must never fall back")
myControl.Stop()
theirControl.Stop()
}
func TestPQRelays(t *testing.T) {
t.Parallel()
ca, _, caKey, _ := cert_test.NewTestCaCert(cert.Version1, cert.Curve_CURVE25519, time.Now(), time.Now().Add(10*time.Minute), nil, nil, []string{})
useRelays := pqConfig(true, true, false)
useRelays["relay"] = m{"use_relays": true}
amRelay := pqConfig(true, true, false)
amRelay["relay"] = m{"am_relay": true}
myControl, myVpnIpNet, _, _ := newSimpleServer(cert.Version1, ca, caKey, "me ", "10.128.0.1/24", useRelays)
relayControl, relayVpnIpNet, relayUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "relay ", "10.128.0.128/24", amRelay)
theirControl, theirVpnIpNet, theirUdpAddr, _ := newSimpleServer(cert.Version1, ca, caKey, "them ", "10.128.0.2/24", useRelays)
myControl.InjectLightHouseAddr(relayVpnIpNet[0].Addr(), relayUdpAddr)
myControl.InjectRelays(theirVpnIpNet[0].Addr(), []netip.Addr{relayVpnIpNet[0].Addr()})
relayControl.InjectLightHouseAddr(theirVpnIpNet[0].Addr(), theirUdpAddr)
r := router.NewR(t, myControl, relayControl, theirControl)
defer r.RenderFlow()
myControl.Start()
relayControl.Start()
theirControl.Start()
var tally handshakeTally
var relayedPQ, relayedClassical int
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpnIpNet[0].Addr(), 80, myVpnIpNet[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(10*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
outer := &header.H{}
if outer.Parse(p.Data) == nil && outer.Type == header.Message && outer.Subtype == header.MessageRelay {
if h, fh := handshakeOf(p); h != nil && fh != nil {
relayedPQ++
assert.LessOrEqual(t, len(p.Data), fragment.MaxPacketLen+32, "a relayed fragment fits the IPv6 minimum MTU")
} else if h != nil {
relayedClassical++
}
}
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
assertUdpPacket(t, []byte("Hi from me"), theirControl.GetFromTun(true), myVpnIpNet[0].Addr(), theirVpnIpNet[0].Addr(), 80, 80)
assert.Positive(t, relayedPQ, "the end to end handshake is PQ through the relay")
assert.Zero(t, relayedClassical, "the relay coming up carries the PQ attempt, not the classical retry")
r.RenderHostmaps("Final hostmaps", myControl, relayControl, theirControl)
myControl.Stop()
relayControl.Stop()
theirControl.Stop()
}
// TestPQTimeoutReleasesIndexes lets a handshake with both a PQ and a classical
// attempt pending time out, and checks neither index is left behind.
func TestPQTimeoutReleasesIndexes(t *testing.T) {
t.Parallel()
mine := pqConfig(true, true, false)
mine["handshakes"].(m)["retries"] = 2
myControl, theirControl, myVpn, theirVpn, _, _, r := newPQPair(t, mine, pqConfig(false, true, false))
defer r.RenderFlow()
var tally handshakeTally
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
r.RouteForAllExitFuncOrTimeout(1500*time.Millisecond, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
return router.Drop
})
assert.Positive(t, tally.pq)
assert.Positive(t, tally.classical)
assert.Nil(t, myControl.GetHostInfoByVpnAddr(theirVpn[0].Addr(), true), "the handshake should have timed out")
assert.Empty(t, myControl.ListHostmapIndexes(true))
myControl.Stop()
theirControl.Stop()
}
// TestPQCookieUnderLoad floods the responder with msg1 fragments from spoofed
// sources until it demands cookies, then checks a real initiator picks up the
// cookie and still completes a PQ handshake.
func TestPQCookieUnderLoad(t *testing.T) {
t.Parallel()
myControl, theirControl, myVpn, theirVpn, _, theirUdp, r := newPQPair(t, pqConfig(true, true, false), pqConfig(false, true, false))
defer r.RenderFlow()
// Half of the 1024 entry table is the cookie threshold. Stopping exactly there
// keeps them from answering the spoofed sources, which the router cannot reach.
t.Log("Flood them with the first fragment of msg1s nobody will finish")
for i := range 512 {
from := netip.AddrPortFrom(netip.AddrFrom4([4]byte{198, 18, byte(i >> 8), byte(i)}), 4242)
s, err := fragment.NewSender(make([]byte, 2000), header.HandshakeIXPSK2MLKEM1024, 0, 1, uint32(i+1))
require.NoError(t, err)
pkts, _, err := s.Attempt(0)
require.NoError(t, err)
theirControl.InjectUDPPacket(&udp.Packet{From: from, To: theirUdp, Data: pkts[0]})
}
var tally handshakeTally
var cookieReplies, msg2, withCookie int
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(5*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
if h, fh := handshakeOf(p); h != nil && h.Subtype == header.HandshakeIXPSK2MLKEM1024 {
switch {
case h.MessageCounter == 0:
cookieReplies++
case h.MessageCounter == 2:
msg2++
case fh != nil && fh.HasCookie():
withCookie++
}
}
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
assertUdpPacket(t, []byte("Hi from me"), theirControl.GetFromTun(true), myVpn[0].Addr(), theirVpn[0].Addr(), 80, 80)
assert.Positive(t, cookieReplies, "a loaded responder asks for a cookie")
assert.Positive(t, withCookie, "the initiator resends with it")
assert.Positive(t, msg2, "and the handshake completes post-quantum")
assert.Zero(t, tally.classical, "the cookie round trip beats the classical fallback")
assertTunnel(t, myVpn[0].Addr(), theirVpn[0].Addr(), myControl, theirControl, r)
myControl.Stop()
theirControl.Stop()
}
// TestPQRequire checks require mode never handshakes classically, in either
// direction.
func TestPQRequire(t *testing.T) {
t.Parallel()
t.Run("initiator", func(t *testing.T) {
// They neither speak PQ nor require it; we never fall back.
myControl, theirControl, myVpn, theirVpn, _, _, r := newPQPair(t, pqRequire(), nil)
defer r.RenderFlow()
var tally handshakeTally
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
r.RouteForAllExitFuncOrTimeout(time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
return router.KeepRouting
})
assert.Positive(t, tally.pq)
assert.Zero(t, tally.classical)
assert.Nil(t, theirControl.GetHostInfoByVpnAddr(myVpn[0].Addr(), false), "no tunnel without PQ")
myControl.Stop()
theirControl.Stop()
})
t.Run("responder", func(t *testing.T) {
// We only speak classical; they refuse it.
myControl, theirControl, myVpn, theirVpn, _, _, r := newPQPair(t, nil, pqRequire())
defer r.RenderFlow()
var answers int
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
r.RouteForAllExitFuncOrTimeout(time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
if h, _ := handshakeOf(p); h != nil && h.MessageCounter == 2 {
answers++
}
return router.KeepRouting
})
assert.Zero(t, answers, "a classical msg1 gets no answer")
assert.Nil(t, theirControl.GetHostInfoByVpnAddr(myVpn[0].Addr(), false))
myControl.Stop()
theirControl.Stop()
})
t.Run("both", func(t *testing.T) {
myControl, theirControl, myVpn, theirVpn, _, _, r := newPQPair(t, pqRequire(), pqRequire())
defer r.RenderFlow()
var tally handshakeTally
myControl.InjectTunPacket(BuildTunUDPPacket(theirVpn[0].Addr(), 80, myVpn[0].Addr(), 80, []byte("Hi from me")))
require.True(t, r.RouteForAllExitFuncOrTimeout(5*time.Second, func(p *udp.Packet, receiver *nebula.Control) router.ExitType {
tally.observe(p)
if isTunnelData(p, receiver, theirControl) {
return router.RouteAndExit
}
return router.KeepRouting
}), "tunnel never came up")
assertUdpPacket(t, []byte("Hi from me"), theirControl.GetFromTun(true), myVpn[0].Addr(), theirVpn[0].Addr(), 80, 80)
assert.Zero(t, tally.classical)
assertTunnel(t, myVpn[0].Addr(), theirVpn[0].Addr(), myControl, theirControl, r)
myControl.Stop()
theirControl.Stop()
})
}
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package router
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package router
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+1
View File
@@ -1,4 +1,5 @@
//go:build e2e_testing
// +build e2e_testing
package e2e
+52 -26
View File
@@ -135,12 +135,12 @@ listen:
# superpacket, so the receive scratch is batch * 64KiB per listening socket (~4MiB per routine at the
# default of 64). Lower this to trade peak per-syscall throughput for memory on constrained hosts.
#batch: 64
# Linux only. Configure socket buffers for the udp side (outside), leave unset to use the system defaults. Values will be doubled by the kernel
# Configure socket buffers for the udp side (outside), leave unset to use the system defaults. Values will be doubled by the kernel
# Default is net.core.rmem_default and net.core.wmem_default (/proc/sys/net/core/rmem_default and /proc/sys/net/core/rmem_default)
# Maximum is limited by memory in the system, SO_RCVBUFFORCE and SO_SNDBUFFORCE is used to avoid having to raise the system wide
# max, net.core.rmem_max and net.core.wmem_max
read_buffer: 3412512
write_buffer: 3412512
#read_buffer: 10485760
#write_buffer: 10485760
# On Windows only
# When true, Nebula installs a WFP (Windows Filtering Platform) PERMIT filter scoped to UDP at the listener port.
@@ -179,9 +179,54 @@ listen:
# Currently, this defaults to 1 which means we have 1 tun queue reader and 1
# UDP queue reader. Setting this above one will set IFF_MULTI_QUEUE on the tun
# device and SO_REUSEPORT on the UDP socket to allow multiple queues.
# With multiport enabled this is the number of routines *per port*, so the total
# is routines * multiport.ports.
# This option is only supported on Linux.
#routines: 1
# EXPERIMENTAL: multiport lanes give each pair of hosts multiple underlay UDP
# flows so overlay traffic is no longer bottlenecked by a single 5-tuple
# (one ECMP path, one NIC RSS queue, one per-flow policer). Instead of every
# socket sharing listen.port, multiport.ports consecutive ports are bound and
# each gets its own group of `routines` sockets sharing it via SO_REUSEPORT, so
# no port (the base port above all, which carries every handshake and every
# vanilla peer) depends on a single core. One extra tunnel ("lane") per port is
# negotiated with capable peers: lane i handshakes from local port listen.port+i
# to the peer's advertised base+((i + pair_offset) mod peer_ports), where
# pair_offset is a per-pair hash that spreads many small peers across a big
# peer's whole port range.
# Each lane is a full Noise session with its own
# keys, nonce counter and replay window, so flows taking different paths
# never fight over shared replay state.
#
# Peers negotiate lanes in the handshake; vanilla peers get a single normal
# tunnel. All control traffic (handshakes, lighthouse, punching, relays) and
# the data fallback stay on the base tunnel/port. Lanes are built lazily: a
# lane is only established once a routine actually has traffic for that peer
# and no lane to carry it, so a peer you exchange a trickle with costs one
# tunnel regardless of how many routines are configured. Established lanes
# are kept alive with their own keepalives, and traffic falls back to the base
# tunnel while a lane is down or not yet up.
#
# Requirements: multiport.ports > 1, Linux, and the port range
# [listen.port, listen.port+multiport.ports-1] reachable through firewalls on
# both sides. A lane whose port is unreachable stays down and its traffic rides
# the base tunnel. With listen.port 0 the base port is dynamic and the next
# ports-1 ports above it are claimed. Degrades to a single port when the
# requirements don't hold. Not reloadable.
#multiport:
#enabled: true
# How many consecutive UDP ports to bind, starting at listen.port. Must be
# greater than 1 for multiport to do anything; there is no default, since
# each port costs a full set of `routines` threads and sockets. Capped at 256
# (the lane header limit). 4-8 is plenty to escape a single ECMP path.
#ports: 0
# How many lanes to run, counting the base tunnel as lane 0. 0 (default)
# means one per bound port. Lowering this sends on a subset of the range,
# which bounds how many extra tunnels each peer pair maintains (useful on a
# big server with many peers); the ports are bound and read either way.
#lanes: 0
punchy:
# Continues to punch inbound/outbound at a regular interval to avoid expiration of firewall nat mappings
# This setting is reloadable.
@@ -262,10 +307,9 @@ tun:
drop_local_broadcast: false
# Toggles forwarding of multicast packets
drop_multicast: false
# Sets the transmit queue length, if you notice lots of transmit drops on the tun it may help to raise this number. Default is 5000
tx_queue: 5000
# Sets the transmit queue length, if you notice lots of transmit drops on the tun it may help to raise this number. Default is 500
tx_queue: 500
# Default MTU for every packet, safe setting is (and the default) 1300 for internet based traffic
# Anything above 8937, here or on a route below, is capped to 8937 with a warning
mtu: 1300
# the use_offloads setting controls if Nebula will attempt to enable GSO and GRO for the tun device. Linux only, not reloadable.
@@ -411,31 +455,13 @@ logging:
#try_interval: 100ms
#retries: 20
# query_buffer is the size of the buffer channel for querying lighthouses, drops are recorded in the lighthouse.query.dropped metric.
#query_buffer: 1024
# query_buffer is the size of the buffer channel for querying lighthouses
#query_buffer: 64
# trigger_buffer is the size of the buffer channel for quickly sending handshakes
# after receiving the response for lighthouse queries
#trigger_buffer: 64
# pq enables handshakes that mix an ephemeral ML-KEM-1024 key exchange into the session keys, protecting
# recorded traffic against a future quantum computer. Authentication is unchanged. These handshakes are too
# large for one packet and travel as fragments with forward error correction.
#pq:
# initiate starts post-quantum handshakes. Attempts alternate between post-quantum and classical until one is
# answered, so peers that do not support them still connect.
#initiate: false
# accept answers post-quantum handshakes from peers.
#accept: false
# require makes every handshake post-quantum: it implies initiate and accept, never offers a classical handshake
# and refuses classical handshakes from peers. Only enable it once every host can speak post-quantum handshakes.
#require: false
# sticky never falls back to a classical handshake with a peer that has completed a post-quantum one since
# nebula started, which defeats an attacker who drops post-quantum handshakes to force a downgrade.
#sticky: false
# repair_delay is how long the forward error correction fragments trail the rest of a handshake message.
#repair_delay: 10ms
# Tunnel manager settings
#tunnels:
# drop_inactive controls whether inactive tunnels are maintained or dropped after the inactive_timeout period has
+1 -1
View File
@@ -22,7 +22,7 @@ func newFixedTicker(t *testing.T, l *slog.Logger, cacheLen int) *ConntrackCacheT
l: l,
cache: make(ConntrackCache, cacheLen),
}
for i := range cacheLen {
for i := 0; i < cacheLen; i++ {
c.cache[Packet{LocalPort: uint16(i) + 1}] = struct{}{}
}
c.cacheTick.Store(1) // cacheV starts at 0, so Get() takes the reset path
+12 -10
View File
@@ -5,33 +5,33 @@ go 1.26.0
require (
dario.cat/mergo v1.0.2
filippo.io/bigmod v0.1.0
github.com/DefinedNet/netlink v1.3.2-dn.1
github.com/anmitsu/go-shlex v0.0.0-20200514113438-38f4b401e2be
github.com/armon/go-radix v1.0.0
github.com/flynn/noise v1.1.0
github.com/gaissmai/bart v0.29.0
github.com/gaissmai/bart v0.28.0
github.com/gogo/protobuf v1.3.2
github.com/google/gopacket v1.1.19
github.com/kardianos/service v1.3.0
github.com/miekg/dns v1.1.73
github.com/miekg/dns v1.1.72
github.com/miekg/pkcs11 v1.1.2
github.com/nbrownus/go-metrics-prometheus v0.0.0-20210712211119-974a6260965f
github.com/prometheus/client_golang v1.24.1
github.com/rcrowley/go-metrics v0.0.0-20201227073835-cf1acfcdf475
github.com/skip2/go-qrcode v0.0.0-20200617195104-da1b6568686e
github.com/stefanberger/go-pkcs11uri v0.0.0-20230803200340-78284954bff6
github.com/stretchr/testify v1.12.1
github.com/stretchr/testify v1.12.0
github.com/vishvananda/netlink v1.3.1
go.uber.org/goleak v1.3.0
go.yaml.in/yaml/v3 v3.0.5
golang.org/x/crypto v0.57.0
golang.org/x/net v0.59.0
golang.org/x/sync v0.23.0
golang.org/x/sys v0.48.0
golang.org/x/term v0.46.0
golang.org/x/crypto v0.54.0
golang.org/x/net v0.57.0
golang.org/x/sync v0.22.0
golang.org/x/sys v0.47.0
golang.org/x/term v0.45.0
golang.zx2c4.com/wintun v0.0.0-20230126152724-0fa3db229ce2
golang.zx2c4.com/wireguard v0.0.0-20230325221338-052af4a8072b
golang.zx2c4.com/wireguard/windows v1.0.1
google.golang.org/protobuf v1.36.12
google.golang.org/protobuf v1.36.11
gopkg.in/yaml.v3 v3.0.1
gvisor.dev/gvisor v0.0.0-20240423190808-9d7a357edefe
)
@@ -45,5 +45,7 @@ require (
github.com/prometheus/common v0.70.1 // indirect
github.com/prometheus/procfs v0.21.1 // indirect
github.com/vishvananda/netns v0.0.5 // indirect
golang.org/x/mod v0.36.0 // indirect
golang.org/x/time v0.5.0 // indirect
golang.org/x/tools v0.45.0 // indirect
)
+26 -20
View File
@@ -3,8 +3,6 @@ dario.cat/mergo v1.0.2 h1:85+piFYR1tMbRrLcDwR18y4UKJ3aH1Tbzi24VRW1TK8=
dario.cat/mergo v1.0.2/go.mod h1:E/hbnu0NxMFBjpMIE34DRGLWqDy0g5FuKDhCb31ngxA=
filippo.io/bigmod v0.1.0 h1:UNzDk7y9ADKST+axd9skUpBQeW7fG2KrTZyOE4uGQy8=
filippo.io/bigmod v0.1.0/go.mod h1:OjOXDNlClLblvXdwgFFOQFJEocLhhtai8vGLy0JCZlI=
github.com/DefinedNet/netlink v1.3.2-dn.1 h1:Md2PsUE1Xuy1Zxs/Bm6356n1kt5bq2HspXlPJ6pEspg=
github.com/DefinedNet/netlink v1.3.2-dn.1/go.mod h1:o4YSn8Jw7zoDcYXDmV77TRd4eS5sSLuE+vu2zAN67C8=
github.com/alecthomas/template v0.0.0-20160405071501-a0175ee3bccc/go.mod h1:LOuyumcjzFXgccqObfd/Ljyb9UuFJ6TxHnclSeseNhc=
github.com/alecthomas/template v0.0.0-20190718012654-fb15b899a751/go.mod h1:LOuyumcjzFXgccqObfd/Ljyb9UuFJ6TxHnclSeseNhc=
github.com/alecthomas/units v0.0.0-20151022065526-2efee857e7cf/go.mod h1:ybxpYRFXyAe+OPACYpWeL0wqObRcbAqCMya13uyzqw0=
@@ -25,8 +23,8 @@ github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSs
github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
github.com/flynn/noise v1.1.0 h1:KjPQoQCEFdZDiP03phOvGi11+SVVhBG2wOWAorLsstg=
github.com/flynn/noise v1.1.0/go.mod h1:xbMo+0i6+IGbYdJhF31t2eR1BIU0CYc12+BNAKwUTag=
github.com/gaissmai/bart v0.29.0 h1:wO6HGE8g9YE0Wm0bCpYxwRzfQ4+fbJKOhL64e5ACGCI=
github.com/gaissmai/bart v0.29.0/go.mod h1:GREWQfTLRWz/c5FTOsIw+KkscuFkIV5t8Rp7Nd1Td5c=
github.com/gaissmai/bart v0.28.0 h1:89yZLo8NmyqD0RYgJ3QO9HhqqGGw+oWhf90cZm69Lko=
github.com/gaissmai/bart v0.28.0/go.mod h1:GREWQfTLRWz/c5FTOsIw+KkscuFkIV5t8Rp7Nd1Td5c=
github.com/go-kit/kit v0.8.0/go.mod h1:xBxKIO96dXMWWy0MnWVtmwkA9/13aqxPnvrjFYMA2as=
github.com/go-kit/kit v0.9.0/go.mod h1:xBxKIO96dXMWWy0MnWVtmwkA9/13aqxPnvrjFYMA2as=
github.com/go-kit/log v0.1.0/go.mod h1:zbhenjAZHb184qTLMA9ZjW7ThYL0H2mk7Q6pNt4vbaY=
@@ -83,8 +81,8 @@ github.com/kr/text v0.1.0/go.mod h1:4Jbv+DJW3UT/LiOwJeYQe1efqtUx/iVham/4vfdArNI=
github.com/kylelemons/godebug v1.1.0 h1:RPNrshWIDI6G2gRW9EHilWtl7Z6Sb1BR0xunSBf0SNc=
github.com/kylelemons/godebug v1.1.0/go.mod h1:9/0rRGxNHcop5bhtWyNeEfOS8JIWk580+fNqagV/RAw=
github.com/matttproud/golang_protobuf_extensions v1.0.1/go.mod h1:D8He9yQNgCq6Z5Ld7szi9bcBfOoFv/3dc6xSMkL2PC0=
github.com/miekg/dns v1.1.73 h1:uhT8nJxmTrPJYClxVxTCX+CVn6qnzSiybRk72Z6DgrE=
github.com/miekg/dns v1.1.73/go.mod h1:RW2Obtfd5NZHvOFe3zYG0W8koWOQtAzyHaLo8vASBuQ=
github.com/miekg/dns v1.1.72 h1:vhmr+TF2A3tuoGNkLDFK9zi36F2LS+hKTRW0Uf8kbzI=
github.com/miekg/dns v1.1.72/go.mod h1:+EuEPhdHOsfk6Wk5TT2CzssZdqkmFhf8r+aVyDEToIs=
github.com/miekg/pkcs11 v1.1.2 h1:/VxmeAX5qU6Q3EwafypogwWbYryHFmF2RpkJmw3m4MQ=
github.com/miekg/pkcs11 v1.1.2/go.mod h1:XsNlhZGX73bx86s2hdc/FuaLm2CPZJemRLMA+WTFxgs=
github.com/modern-go/concurrent v0.0.0-20180228061459-e0a39a4cb421/go.mod h1:6dJC0mAP4ikYIbvyc7fijjWJddQyLn8Ig3JB5CqoB9Q=
@@ -138,8 +136,10 @@ github.com/stretchr/testify v1.2.2/go.mod h1:a8OnRcib4nhh0OaRAV+Yts87kKdq0PP7pXf
github.com/stretchr/testify v1.3.0/go.mod h1:M5WIy9Dh21IEIfnGCwXGc5bZfKNJtfHm1UVUgZn+9EI=
github.com/stretchr/testify v1.4.0/go.mod h1:j7eGeouHqKxXV5pUuKE4zz7dFj8WfuZ+81PSLYec5m4=
github.com/stretchr/testify v1.7.0/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
github.com/stretchr/testify v1.12.1 h1:EuwCh5fleGS7H32xRwO3wRGT7DxrDhLAT6FF8MpWDWE=
github.com/stretchr/testify v1.12.1/go.mod h1:MDEgiDPPsNp5cuIrHPPCyornHKgEVbtFUmoNlxoYthg=
github.com/stretchr/testify v1.12.0 h1:K6Mr6jO9JICuend/5xzTM03ydSV3vdNRYAdPSukj8uI=
github.com/stretchr/testify v1.12.0/go.mod h1:bOYBZb5qJ00vPzWfIqBUZPaxK8jWiXc6d3ErP4Ca9Gw=
github.com/vishvananda/netlink v1.3.1 h1:3AEMt62VKqz90r0tmNhog0r/PpWKmrEShJU0wJW6bV0=
github.com/vishvananda/netlink v1.3.1/go.mod h1:ARtKouGSTGchR8aMwmkzC0qiNPrrWO5JS/XMVl45+b4=
github.com/vishvananda/netns v0.0.5 h1:DfiHV+j8bA32MFM7bfEunvT8IAqQ/NzSJHtcmW5zdEY=
github.com/vishvananda/netns v0.0.5/go.mod h1:SpkAiCQRtJ6TvvxPnOSyH3BMl6unz3xZlaprSwhNNJM=
github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
@@ -155,12 +155,14 @@ golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACk
golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
golang.org/x/crypto v0.0.0-20200622213623-75b288015ac9/go.mod h1:LzIPMQfyMNhhGPhUkYOs5KpL4U8rLKemX1yGLhDgUto=
golang.org/x/crypto v0.0.0-20210322153248-0c34fe9e7dc2/go.mod h1:T9bdIzuCu7OtxOm1hfPfRQxPLYneinmdGuTeoZ9dtd4=
golang.org/x/crypto v0.57.0 h1:3ZVCjf8Ggz7zneR/EHRVx68Ctf+2pmIMP2UFhh9cC6M=
golang.org/x/crypto v0.57.0/go.mod h1:Fdz0i5U6CoizGwLda9DttjSk6qlZo25zYNtR+ycvuZA=
golang.org/x/crypto v0.54.0 h1:YLIA59K4fiNzHzjnZt2tUJQjQtUWfWbeHBqKtk3eScw=
golang.org/x/crypto v0.54.0/go.mod h1:KWL8ny2AZdGR2cWmzeHrp2azQPGogOv+HeQaVEXC2dk=
golang.org/x/lint v0.0.0-20200302205851-738671d3881b/go.mod h1:3xt1FjdF8hUf6vQPIChWIBhFzV8gjjsPE/fR3IyQdNY=
golang.org/x/mod v0.1.1-0.20191105210325-c90efee705ee/go.mod h1:QqPTAvyqsEbceGzBzNggFXnrqF1CaUcvgkdR5Ot7KZg=
golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
golang.org/x/mod v0.3.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
golang.org/x/mod v0.36.0 h1:JJjpVx6myfUsUdAzZuOSTTmRE0PfZeNWzzvKrP7amb4=
golang.org/x/mod v0.36.0/go.mod h1:moc6ELqsWcOw5Ef3xVprK5ul/MvtVvkIXLziUOICjUQ=
golang.org/x/net v0.0.0-20180724234803-3673e40ba225/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
golang.org/x/net v0.0.0-20181114220301-adae6a3d119a/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
golang.org/x/net v0.0.0-20190108225652-1e06a53dbb7e/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4=
@@ -171,8 +173,8 @@ golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLL
golang.org/x/net v0.0.0-20200625001655-4c5254603344/go.mod h1:/O7V0waA8r7cgGh81Ro3o1hOxt32SMVPicZroKQ2sZA=
golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU=
golang.org/x/net v0.0.0-20210226172049-e18ecbb05110/go.mod h1:m0MpNAwzfU5UDzcl9v0D8zg8gWTRqZa9RBIspLL5mdg=
golang.org/x/net v0.59.0 h1:5zfYln+w5XCxwrnMMJPufRgNoXEaGxl0wo5GqPXyues=
golang.org/x/net v0.59.0/go.mod h1:2DA/G1UfVbCpQPeWTmMPGY7Cs2PkBkwu743bVX5PIVg=
golang.org/x/net v0.57.0 h1:K5+3DljvIuDG9/Jv9rvyMywYNFCQ9RSUY6OOTTkT+tE=
golang.org/x/net v0.57.0/go.mod h1:KpXc8iv+r3XplLAG/f7Jsf9RPszJzdR0f58q9vGOuEU=
golang.org/x/oauth2 v0.0.0-20190226205417-e64efc72b421/go.mod h1:gOpvHmFTYa4IltrdGE7lF6nIHvwfUNPOp7c8zoXwtLw=
golang.org/x/sync v0.0.0-20181108010431-42b317875d0f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
golang.org/x/sync v0.0.0-20181221193216-37e7f081c4d4/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
@@ -180,8 +182,8 @@ golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJ
golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
golang.org/x/sync v0.0.0-20201020160332-67f06af15bc9/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
golang.org/x/sync v0.0.0-20201207232520-09787c993a3a/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
golang.org/x/sync v0.23.0 h1:KameEIfc1IkluZyXWLn39Wd4tURc6GbCiISGiZm2bQk=
golang.org/x/sync v0.23.0/go.mod h1:sUUOizhqBxiL6pEWpqNLUiaJn1ShEbZ6BBqskPbjZm0=
golang.org/x/sync v0.22.0 h1:SZjpbeLmrCk4xhRSZFNZW5gFUeCeFgjekvI/+gfScek=
golang.org/x/sync v0.22.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
golang.org/x/sys v0.0.0-20180905080454-ebe1bf3edb33/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
golang.org/x/sys v0.0.0-20181116152217-5ac8a444bdc5/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
@@ -195,11 +197,13 @@ golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7w
golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
golang.org/x/sys v0.0.0-20210124154548-22da62e12c0c/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
golang.org/x/sys v0.0.0-20210603081109-ebe580a85c40/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
golang.org/x/sys v0.48.0 h1:bbX/i/6MgT9BVLM9RT1thmxL04yeTAhbEz4SyadbXoo=
golang.org/x/sys v0.48.0/go.mod h1:hNLxWAXmnKAxqDtdwIYC4bM9oQPEecfsnNMuSxOs3og=
golang.org/x/sys v0.2.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
golang.org/x/sys v0.10.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
golang.org/x/sys v0.47.0 h1:o7XGOvZQCADBQQ4Y7VNq2dRWQR7JmOUW8Kxx4ZsNgWs=
golang.org/x/sys v0.47.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
golang.org/x/term v0.46.0 h1:3+OXuTbaKDgwk8jTi3aSLHRlmWqHEUDUtxnbFigO4YE=
golang.org/x/term v0.46.0/go.mod h1:+K02xbkittuwc0Am4abfA3Fc+XRGXkvBXNO88NCXPoc=
golang.org/x/term v0.45.0 h1:NwWyBmoJCbfTHpxrWoZ9C6/VxOf7ic219I8xZZFdrf0=
golang.org/x/term v0.45.0/go.mod h1:9aqxs0blBcrm/n0L9QW0aRVD+ktan8ssZromtqJC43w=
golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk=
golang.org/x/text v0.3.3/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
@@ -210,6 +214,8 @@ golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtn
golang.org/x/tools v0.0.0-20200130002326-2f3ba24bd6e7/go.mod h1:TB2adYChydJhpapKDTa4BR/hXlZSLoq2Wpct/0txZ28=
golang.org/x/tools v0.0.0-20200619180055-7c47624df98f/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE=
golang.org/x/tools v0.0.0-20210106214847-113979e3529a/go.mod h1:emZCQorbCU4vsT4fOWvOPXz4eW1wZW4PmDk9uLelYpA=
golang.org/x/tools v0.45.0 h1:18qN3FAooORvApf5XjCXgsuayZOEtXf6JK18I3+ONa8=
golang.org/x/tools v0.45.0/go.mod h1:LuUGqqaXcXMEFEruIVJVm5mgDD8vww/z/SR1gQ4uE/0=
golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
@@ -228,8 +234,8 @@ google.golang.org/protobuf v1.20.1-0.20200309200217-e05f789c0967/go.mod h1:A+miE
google.golang.org/protobuf v1.21.0/go.mod h1:47Nbq4nVaFHyn7ilMalzfO3qCViNmqZ2kzikPIcrTAo=
google.golang.org/protobuf v1.23.0/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU=
google.golang.org/protobuf v1.26.0-rc.1/go.mod h1:jlhhOSvTdKEhbULTjvd4ARK9grFBp09yW+WbY/TyQbw=
google.golang.org/protobuf v1.36.12 h1:pJOKDDOyeXErUroCihFAd5LQuwXBSpVnKGrj5o/fwxc=
google.golang.org/protobuf v1.36.12/go.mod h1:HTf+CrKn2C3g5S8VImy6tdcUvCska2kB7j23XfzDpco=
google.golang.org/protobuf v1.36.11 h1:fV6ZwhNocDyBLK0dj+fg8ektcVegBBuEolpbTQyBNVE=
google.golang.org/protobuf v1.36.11/go.mod h1:HTf+CrKn2C3g5S8VImy6tdcUvCska2kB7j23XfzDpco=
gopkg.in/alecthomas/kingpin.v2 v2.2.6/go.mod h1:FMv+mEhP44yOT+4EoQTLFTRgOQ1FBLkstjWtayDeSgw=
gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
gopkg.in/check.v1 v1.0.0-20190902080502-41f04d3bba15/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0=
+5 -18
View File
@@ -31,7 +31,6 @@ package nebula
import (
"bufio"
"context"
"fmt"
"net"
"strconv"
@@ -52,29 +51,17 @@ type graphiteConfigExport struct {
Percentiles []float64 // Percentiles to export from timers and histograms
}
// graphiteTimeout bounds one submission, the connect and the write together. An export is a few KB, a host that can't
// take that in this long isn't going to.
const graphiteTimeout = 30 * time.Second
// graphiteOnce performs a single submission to Graphite, returning a non-nil
// error if it could not connect or write all of it within timeout. It gives up
// as soon as ctx is done.
func graphiteOnce(ctx context.Context, c graphiteConfigExport, timeout time.Duration) error {
// error on failed connections.
func graphiteOnce(c graphiteConfigExport) error {
now := time.Now().Unix()
du := float64(c.DurationUnit)
flushSeconds := float64(c.FlushInterval) / float64(time.Second)
deadline := time.Now().Add(timeout)
d := net.Dialer{Deadline: deadline}
conn, err := d.DialContext(ctx, "tcp", c.Addr.String())
conn, err := net.DialTCP("tcp", nil, c.Addr)
if err != nil {
return err
}
defer conn.Close()
if err := conn.SetDeadline(deadline); err != nil {
return err
}
stop := context.AfterFunc(ctx, func() { _ = conn.Close() })
defer stop()
w := bufio.NewWriter(conn)
c.Registry.Each(func(name string, i any) {
switch metric := i.(type) {
@@ -124,7 +111,7 @@ func graphiteOnce(ctx context.Context, c graphiteConfigExport, timeout time.Dura
fmt.Fprintf(w, "%s.%s.fifteen-minute %.2f %d\n", c.Prefix, name, t.Rate15(), now)
fmt.Fprintf(w, "%s.%s.mean-rate %.2f %d\n", c.Prefix, name, t.RateMean(), now)
}
w.Flush()
})
// A failed write sticks, so this reports the first one
return w.Flush()
return nil
}
-315
View File
@@ -1,315 +0,0 @@
package nebula
import (
"bytes"
"context"
"errors"
"fmt"
"io"
"net"
"strings"
"sync"
"testing"
"time"
"github.com/rcrowley/go-metrics"
"github.com/slackhq/nebula/test"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// graphiteHost accepts connections and hands each one to the test
type graphiteHost struct {
ln net.Listener
conns chan net.Conn
mu sync.Mutex
closed bool
open []net.Conn
}
// newGraphiteHost accepts connections for the test. A non-zero window shrinks each connection's receive buffer, so a
// sender that is not read from blocks after a little whatever the host's TCP tuning. Only for tests that need a stall,
// a small window makes reading very slow on Linux.
func newGraphiteHost(t *testing.T, window int) *graphiteHost {
t.Helper()
ln, err := net.Listen("tcp", "127.0.0.1:0")
require.NoError(t, err)
h := &graphiteHost{ln: ln, conns: make(chan net.Conn, 8)}
done := make(chan struct{})
t.Cleanup(func() {
_ = ln.Close()
h.mu.Lock()
h.closed = true
for _, c := range h.open {
_ = c.Close()
}
h.mu.Unlock()
close(done)
})
go func() {
for {
c, err := ln.Accept()
if err != nil {
return
}
h.mu.Lock()
if h.closed {
h.mu.Unlock()
_ = c.Close()
return
}
h.open = append(h.open, c)
h.mu.Unlock()
if tc, ok := c.(*net.TCPConn); ok && window > 0 {
_ = tc.SetReadBuffer(window)
}
select {
case h.conns <- c:
case <-done:
return
}
}
}()
return h
}
func (h *graphiteHost) addr(t *testing.T) *net.TCPAddr {
a, err := net.ResolveTCPAddr("tcp", h.ln.Addr().String())
require.NoError(t, err)
return a
}
// readExport reads what one send wrote, bounded so a send that never finishes fails the test rather than hanging it
func readExport(t *testing.T, c net.Conn) []byte {
t.Helper()
require.NoError(t, c.SetReadDeadline(time.Now().Add(10*time.Second)))
b, err := io.ReadAll(c)
require.NoError(t, err)
return b
}
func (h *graphiteHost) next(t *testing.T, wait time.Duration) net.Conn {
t.Helper()
select {
case c := <-h.conns:
return c
case <-time.After(wait):
t.Fatal("no connection")
return nil
}
}
// lockedBuffer is a log destination a test can read while the code under test writes to it from another goroutine
type lockedBuffer struct {
mu sync.Mutex
buf bytes.Buffer
}
func (b *lockedBuffer) Write(p []byte) (int, error) {
b.mu.Lock()
defer b.mu.Unlock()
return b.buf.Write(p)
}
func (b *lockedBuffer) String() string {
b.mu.Lock()
defer b.mu.Unlock()
return b.buf.String()
}
func testGraphiteConfig(addr *net.TCPAddr, r metrics.Registry) graphiteConfigExport {
return graphiteConfigExport{Addr: addr, Registry: r, FlushInterval: time.Second, DurationUnit: time.Nanosecond, Prefix: "nebula"}
}
// Enough metrics that one export overflows the socket buffers on any platform, so a send to a host that stops
// reading blocks
func bigGraphiteRegistry(r metrics.Registry, t *testing.T) {
for i := range 200000 {
name := fmt.Sprintf("big.gauge.%d", i)
metrics.GetOrRegisterGauge(name, r).Update(int64(i))
t.Cleanup(func() { r.Unregister(name) })
}
}
func TestGraphiteOnce_sends(t *testing.T) {
r := metrics.NewRegistry()
metrics.GetOrRegisterGauge("g", r).Update(7)
metrics.GetOrRegisterCounter("c", r).Inc(3)
h := newGraphiteHost(t, 0)
require.NoError(t, graphiteOnce(t.Context(), testGraphiteConfig(h.addr(t), r), time.Minute))
b := readExport(t, h.next(t, 5*time.Second))
assert.Contains(t, string(b), "nebula.g.value 7 ")
assert.Contains(t, string(b), "nebula.c.count 3 ")
assert.Contains(t, string(b), "nebula.c.count_ps 3.00 ")
}
// A host that accepts and never reads is given up on, and the failed write is reported
func TestGraphiteOnce_stalledHostTimesOut(t *testing.T) {
r := metrics.NewRegistry()
bigGraphiteRegistry(r, t)
h := newGraphiteHost(t, 4096)
done := make(chan error, 1)
cfg := testGraphiteConfig(h.addr(t), r)
go func() { done <- graphiteOnce(t.Context(), cfg, 300*time.Millisecond) }()
h.next(t, 5*time.Second) // accepted, never read
select {
case err := <-done:
require.Error(t, err)
case <-time.After(10 * time.Second):
t.Fatal("the send did not give up on a host that stopped reading")
}
}
// The timeout covers the whole send, a host that keeps reading but too slowly is still given up on
func TestGraphiteOnce_timeoutCoversTheWholeSend(t *testing.T) {
r := metrics.NewRegistry()
bigGraphiteRegistry(r, t)
h := newGraphiteHost(t, 0)
done := make(chan error, 1)
start := time.Now()
cfg := testGraphiteConfig(h.addr(t), r)
go func() { done <- graphiteOnce(t.Context(), cfg, 500*time.Millisecond) }()
c := h.next(t, 5*time.Second)
stopReading := make(chan struct{})
defer close(stopReading)
// 32KB every 10ms keeps every write well inside the timeout, the whole export takes seconds
go func() {
buf := make([]byte, 32<<10)
for {
select {
case <-stopReading:
return
case <-time.After(10 * time.Millisecond):
}
if _, err := c.Read(buf); err != nil {
return
}
}
}()
select {
case err := <-done:
require.Error(t, err)
assert.Less(t, time.Since(start), 5*time.Second)
case <-time.After(10 * time.Second):
t.Fatal("a host that kept reading slowly held the send past its timeout")
}
}
func TestGraphiteOnce_cancelledCtxDoesNotConnect(t *testing.T) {
h := newGraphiteHost(t, 0)
ctx, cancel := context.WithCancel(t.Context())
cancel()
require.Error(t, graphiteOnce(ctx, testGraphiteConfig(h.addr(t), metrics.NewRegistry()), time.Minute))
select {
case <-h.conns:
t.Fatal("connected on a cancelled ctx")
case <-time.After(200 * time.Millisecond):
}
}
// A stop or reload abandons a stuck send rather than waiting out its timeout
func TestGraphiteOnce_cancelAbandonsSend(t *testing.T) {
r := metrics.NewRegistry()
bigGraphiteRegistry(r, t)
h := newGraphiteHost(t, 4096)
ctx, cancel := context.WithCancel(t.Context())
done := make(chan error, 1)
cfg := testGraphiteConfig(h.addr(t), r)
go func() { done <- graphiteOnce(ctx, cfg, time.Minute) }()
h.next(t, 5*time.Second)
select {
case err := <-done:
t.Fatalf("the send should be stuck on a host that is not reading, it ended with %v", err)
case <-time.After(200 * time.Millisecond):
}
cancel()
select {
case err := <-done:
require.Error(t, err)
case <-time.After(5 * time.Second):
t.Fatal("the send is still stuck after its ctx was cancelled")
}
}
// buildRuntime hands graphite the configured prefix and interval
func TestStatsServer_graphiteExportUsesConfig(t *testing.T) {
const name = "graphite.wiring.test"
metrics.GetOrRegisterCounter(name, nil).Inc(8)
t.Cleanup(func() { metrics.DefaultRegistry.Unregister(name) })
h := newGraphiteHost(t, 0)
s, c := newTestStatsServer(t)
setStatsConfig(c, map[string]any{"type": "graphite", "interval": "4s", "prefix": "pfx", "host": h.ln.Addr().String()})
cfg, err := loadStatsConfig(c)
require.NoError(t, err)
s.runMu.Lock()
fns, _ := s.buildRuntime(t.Context(), cfg)
s.runMu.Unlock()
for _, fn := range fns {
fn()
}
b := readExport(t, h.next(t, 5*time.Second))
assert.Contains(t, string(b), "pfx."+name+".count 8 ")
assert.Contains(t, string(b), "pfx."+name+".count_ps 2.00 ")
}
func TestStatsServer_graphiteFailureIsLogged(t *testing.T) {
logs := &lockedBuffer{}
s, c := newTestStatsServer(t)
s.l = test.NewLoggerWithOutput(logs)
setStatsConfig(c, map[string]any{"type": "graphite", "interval": "1h", "host": "127.0.0.1:" + freeTCPPort(t)})
cfg, err := loadStatsConfig(c)
require.NoError(t, err)
s.runMu.Lock()
fns, _ := s.buildRuntime(t.Context(), cfg)
s.runMu.Unlock()
for _, fn := range fns {
fn()
}
assert.Contains(t, logs.String(), "Graphite export failed")
}
// Stop abandons the running export, so it goes through Start rather than handing buildRuntime a ctx, and a stop is
// not logged as a failure
func TestStatsServer_stopAbandonsGraphiteSend(t *testing.T) {
bigGraphiteRegistry(metrics.DefaultRegistry, t)
h := newGraphiteHost(t, 4096)
logs := &lockedBuffer{}
s, c := newTestStatsServer(t)
s.l = test.NewLoggerWithOutput(logs)
setStatsConfig(c, map[string]any{"type": "graphite", "interval": "50ms", "host": h.ln.Addr().String()})
require.NoError(t, s.reload(c, true))
started := make(chan struct{})
go func() {
s.Start()
close(started)
}()
stuck := h.next(t, 30*time.Second)
s.Stop()
select {
case <-started:
case <-time.After(5 * time.Second):
t.Fatal("Start did not return after Stop")
}
// A closed sender resets the connection once it sees this, one still sending just buffers it. Either way the
// read below doesn't have to drain the export through the small window to find out
_, err := stuck.Write([]byte("x"))
require.NoError(t, err)
require.NoError(t, stuck.SetReadDeadline(time.Now().Add(10*time.Second)))
n, err := io.Copy(io.Discard, stuck)
var ne net.Error
if errors.As(err, &ne) && ne.Timeout() {
t.Fatalf("the send kept the connection after Stop, read %d bytes", n)
}
assert.Less(t, n, int64(1<<20), "the send kept going after Stop")
// The capture loop logs on its own goroutine after Start has returned, give it the chance
assert.Never(t, func() bool { return strings.Contains(logs.String(), "Graphite export failed") }, 500*time.Millisecond, 10*time.Millisecond, "a stop is not a failure")
}
+2 -4
View File
@@ -35,9 +35,7 @@ func NewCredential(
}
// buildHandshakeState creates a noise.HandshakeState from this credential.
// pskPlacement selects the noise psk modifier; 0 means none. A psk for
// placements 2 and up is supplied later with SetPresharedKey.
func (hc *Credential) buildHandshakeState(initiator bool, pattern noise.HandshakePattern, pskPlacement int) (*noise.HandshakeState, error) {
func (hc *Credential) buildHandshakeState(initiator bool, pattern noise.HandshakePattern) (*noise.HandshakeState, error) {
return noise.NewHandshakeState(noise.Config{
CipherSuite: hc.cipherSuite,
Random: rand.Reader,
@@ -45,7 +43,7 @@ func (hc *Credential) buildHandshakeState(initiator bool, pattern noise.Handshak
Initiator: initiator,
StaticKeypair: noise.DHKey{Private: hc.privateKey, Public: hc.Cert.PublicKey()},
PresharedKey: []byte{},
PresharedKeyPlacement: pskPlacement,
PresharedKeyPlacement: 0,
})
}
-216
View File
@@ -1,216 +0,0 @@
// Package fec implements the systematic Reed-Solomon erasure code used to carry
// handshake messages that do not fit in a single UDP packet.
//
// A message is split into K equally sized data symbols (the last one zero
// padded). Symbol indexes 0..K-1 are the data symbols themselves; indexes
// K..MaxSymbols-1 are repair symbols. Any K distinct symbols reconstruct the
// message. Repair symbols are generated on demand, so a sender can emit fresh
// ones on every retransmit and a receiver can merge everything it has seen.
package fec
import (
"errors"
"fmt"
)
// MaxSymbols is the number of distinct symbols, data and repair together, one
// message can be encoded into. It is bounded by the size of GF(2^8).
const MaxSymbols = 256
var (
ErrInvalidLayout = errors.New("invalid fec layout")
ErrIndexRange = errors.New("symbol index out of range")
ErrSymbolSize = errors.New("symbol has the wrong size")
ErrNotEnough = errors.New("not enough symbols to decode")
ErrCorrupt = errors.New("decoded message has non-zero padding")
)
// Layout describes how a message is split into symbols. Both ends derive it
// from the message length alone, so it never has to be sent.
type Layout struct {
MsgLen int // length of the message in bytes
K int // number of data symbols
SymbolSize int // size of every symbol, data and repair
}
// NewLayout returns the layout for a message of msgLen bytes carried in symbols
// of at most maxSymbolSize bytes. Symbols are balanced, so padding is always
// less than K bytes.
func NewLayout(msgLen, maxSymbolSize int) (Layout, error) {
if msgLen <= 0 || maxSymbolSize <= 0 {
return Layout{}, ErrInvalidLayout
}
k := (msgLen + maxSymbolSize - 1) / maxSymbolSize
if k >= MaxSymbols {
return Layout{}, fmt.Errorf("%w: %d data symbols", ErrInvalidLayout, k)
}
return Layout{
MsgLen: msgLen,
K: k,
SymbolSize: (msgLen + k - 1) / k,
}, nil
}
// Encoder produces symbols for one message.
type Encoder struct {
layout Layout
data [][]byte
}
// NewEncoder copies msg into padded data symbols.
func NewEncoder(msg []byte, maxSymbolSize int) (*Encoder, error) {
l, err := NewLayout(len(msg), maxSymbolSize)
if err != nil {
return nil, err
}
buf := make([]byte, l.K*l.SymbolSize)
copy(buf, msg)
data := make([][]byte, l.K)
for i := range data {
data[i] = buf[i*l.SymbolSize : (i+1)*l.SymbolSize]
}
return &Encoder{layout: l, data: data}, nil
}
func (e *Encoder) Layout() Layout {
return e.layout
}
// Symbol appends the symbol with the given index to dst and returns the result.
func (e *Encoder) Symbol(dst []byte, index int) ([]byte, error) {
if index < 0 || index >= MaxSymbols {
return nil, ErrIndexRange
}
if index < e.layout.K {
return append(dst, e.data[index]...), nil
}
start := len(dst)
dst = append(dst, make([]byte, e.layout.SymbolSize)...)
out := dst[start:]
for col, d := range e.data {
mulAdd(out, d, coefficient(index, col))
}
return dst, nil
}
// Decoder collects symbols for one message until it can reconstruct it.
// It stores at most K symbols, so its memory is bounded by the layout no matter
// how many symbols arrive.
type Decoder struct {
layout Layout
have [MaxSymbols]bool
syms map[int][]byte
}
// NewDecoder returns a decoder for a message of msgLen bytes that was encoded
// with the same maxSymbolSize.
func NewDecoder(msgLen, maxSymbolSize int) (*Decoder, error) {
l, err := NewLayout(msgLen, maxSymbolSize)
if err != nil {
return nil, err
}
return &Decoder{layout: l, syms: make(map[int][]byte, l.K)}, nil
}
func (d *Decoder) Layout() Layout {
return d.layout
}
// Stored returns the number of symbols held.
func (d *Decoder) Stored() int {
return len(d.syms)
}
// Has reports whether the symbol with the given index is held.
func (d *Decoder) Has(index int) bool {
return index >= 0 && index < MaxSymbols && d.have[index]
}
// Ready reports whether enough symbols have arrived to decode.
func (d *Decoder) Ready() bool {
return len(d.syms) >= d.layout.K
}
// Add stores a copy of the symbol. Duplicates, and anything beyond the first K
// distinct symbols, are ignored. It returns whether the decoder is ready.
func (d *Decoder) Add(index int, sym []byte) (bool, error) {
if index < 0 || index >= MaxSymbols {
return d.Ready(), ErrIndexRange
}
if len(sym) != d.layout.SymbolSize {
return d.Ready(), ErrSymbolSize
}
if d.have[index] || d.Ready() {
return d.Ready(), nil
}
d.have[index] = true
d.syms[index] = append([]byte(nil), sym...)
return d.Ready(), nil
}
// Decode reconstructs the message from the stored symbols.
func (d *Decoder) Decode() ([]byte, error) {
if !d.Ready() {
return nil, ErrNotEnough
}
l := d.layout
out := make([]byte, l.K*l.SymbolSize)
var missing, repair []int
for i := range l.K {
if s, ok := d.syms[i]; ok {
copy(out[i*l.SymbolSize:], s)
} else {
missing = append(missing, i)
}
}
if len(missing) > 0 {
for i := l.K; i < MaxSymbols && len(repair) < len(missing); i++ {
if d.have[i] {
repair = append(repair, i)
}
}
if len(repair) < len(missing) {
return nil, ErrNotEnough
}
// Each repair symbol r satisfies r = sum_col coefficient(r, col) * data[col].
// Move the known data columns to the left hand side, leaving a square
// Cauchy system in the missing columns.
rhs := make([][]byte, len(repair))
m := make([][]byte, len(repair))
for ri, row := range repair {
rhs[ri] = append([]byte(nil), d.syms[row]...)
m[ri] = make([]byte, len(missing))
mi := 0
for col := range l.K {
if mi < len(missing) && missing[mi] == col {
m[ri][mi] = coefficient(row, col)
mi++
continue
}
mulAdd(rhs[ri], out[col*l.SymbolSize:(col+1)*l.SymbolSize], coefficient(row, col))
}
}
if !invert(m) {
// Unreachable for a Cauchy matrix; kept so a broken invariant fails loudly.
return nil, errors.New("fec: singular decode matrix")
}
for mi, col := range missing {
dst := out[col*l.SymbolSize : (col+1)*l.SymbolSize]
for ri := range repair {
mulAdd(dst, rhs[ri], m[mi][ri])
}
}
}
for _, b := range out[l.MsgLen:] {
if b != 0 {
return nil, ErrCorrupt
}
}
return out[:l.MsgLen], nil
}
-318
View File
@@ -1,318 +0,0 @@
package fec
import (
"crypto/sha256"
"encoding/hex"
"math/rand/v2"
"testing"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// slowMul is carry-less multiplication reduced by gfPoly, used to check the tables.
func slowMul(a, b byte) byte {
var p int
x, y := int(a), int(b)
for y > 0 {
if y&1 != 0 {
p ^= x
}
x <<= 1
if x&0x100 != 0 {
x ^= gfPoly
}
y >>= 1
}
return byte(p)
}
func TestGF(t *testing.T) {
for a := range 256 {
for b := range 256 {
require.Equal(t, slowMul(byte(a), byte(b)), gfMul(byte(a), byte(b)), "%d*%d", a, b)
}
if a != 0 {
require.Equal(t, byte(1), gfMul(byte(a), gfInv(byte(a))), "inv(%d)", a)
}
}
src := make([]byte, 256)
for i := range src {
src[i] = byte(i)
}
for c := range 256 {
dst := make([]byte, 256)
mulAdd(dst, src, byte(c))
for i := range src {
require.Equal(t, gfMul(byte(c), src[i]), dst[i])
}
}
}
func TestLayout(t *testing.T) {
for msgLen := 1; msgLen <= 20000; msgLen++ {
l, err := NewLayout(msgLen, 1176)
require.NoError(t, err)
require.Equal(t, (msgLen+1175)/1176, l.K)
require.LessOrEqual(t, l.SymbolSize, 1176)
require.GreaterOrEqual(t, l.K*l.SymbolSize, msgLen)
require.Less(t, l.K*l.SymbolSize-msgLen, l.K, "padding must be less than one byte per symbol")
}
l, err := NewLayout(65535, 1176)
require.NoError(t, err)
assert.Equal(t, 56, l.K)
_, err = NewLayout(0, 1176)
require.ErrorIs(t, err, ErrInvalidLayout)
_, err = NewLayout(100, 0)
require.ErrorIs(t, err, ErrInvalidLayout)
_, err = NewLayout(256, 1)
require.ErrorIs(t, err, ErrInvalidLayout)
_, err = NewLayout(255, 1)
require.NoError(t, err)
}
func randMsg(r *rand.Rand, n int) []byte {
b := make([]byte, n)
for i := range b {
b[i] = byte(r.Uint32())
}
return b
}
func decodeWith(t *testing.T, enc *Encoder, msgLen, maxSym int, indexes []int) []byte {
t.Helper()
dec, err := NewDecoder(msgLen, maxSym)
require.NoError(t, err)
for _, idx := range indexes {
sym, err := enc.Symbol(nil, idx)
require.NoError(t, err)
_, err = dec.Add(idx, sym)
require.NoError(t, err)
}
require.True(t, dec.Ready())
out, err := dec.Decode()
require.NoError(t, err)
return out
}
// TestExhaustiveSubsets decodes every k-subset of the first k+m symbols for
// small codes, which checks the MDS property directly.
func TestExhaustiveSubsets(t *testing.T) {
r := rand.New(rand.NewPCG(1, 2))
const maxSym = 16
for k := 1; k <= 10; k++ {
for m := 1; m <= 5; m++ {
msgLen := (k-1)*maxSym + 1 + r.IntN(maxSym)
msg := randMsg(r, msgLen)
enc, err := NewEncoder(msg, maxSym)
require.NoError(t, err)
require.Equal(t, k, enc.Layout().K)
n := k + m
for mask := 0; mask < 1<<n; mask++ {
var idx []int
for i := range n {
if mask&(1<<i) != 0 {
idx = append(idx, i)
}
}
if len(idx) != k {
continue
}
require.Equal(t, msg, decodeWith(t, enc, msgLen, maxSym, idx), "k=%d m=%d idx=%v", k, m, idx)
}
}
}
}
// TestRandomHighIndexes draws symbols from the whole index space, the shape
// retransmits produce once fresh repair symbols are in play.
func TestRandomHighIndexes(t *testing.T) {
r := rand.New(rand.NewPCG(3, 4))
for range 2000 {
msgLen := 1 + r.IntN(65535)
msg := randMsg(r, msgLen)
enc, err := NewEncoder(msg, 1176)
require.NoError(t, err)
k := enc.Layout().K
idx := r.Perm(MaxSymbols)[:k]
require.Equal(t, msg, decodeWith(t, enc, msgLen, 1176, idx), "k=%d idx=%v", k, idx)
}
}
func TestDecoderBounds(t *testing.T) {
r := rand.New(rand.NewPCG(5, 6))
msg := randMsg(r, 3000)
enc, err := NewEncoder(msg, 1176)
require.NoError(t, err)
l := enc.Layout()
require.Equal(t, 3, l.K)
require.Equal(t, 1000, l.SymbolSize)
dec, err := NewDecoder(3000, 1176)
require.NoError(t, err)
_, err = dec.Decode()
require.ErrorIs(t, err, ErrNotEnough)
_, err = dec.Add(-1, make([]byte, 1000))
require.ErrorIs(t, err, ErrIndexRange)
_, err = dec.Add(MaxSymbols, make([]byte, 1000))
require.ErrorIs(t, err, ErrIndexRange)
_, err = dec.Add(0, make([]byte, 999))
require.ErrorIs(t, err, ErrSymbolSize)
s5, _ := enc.Symbol(nil, 5)
ready, err := dec.Add(5, s5)
require.NoError(t, err)
require.False(t, ready)
// A duplicate index is ignored, even with different contents.
ready, err = dec.Add(5, make([]byte, 1000))
require.NoError(t, err)
require.False(t, ready)
s0, _ := enc.Symbol(nil, 0)
s9, _ := enc.Symbol(nil, 9)
_, err = dec.Add(0, s0)
require.NoError(t, err)
ready, err = dec.Add(9, s9)
require.NoError(t, err)
require.True(t, ready)
// Symbols beyond K are not stored.
ready, err = dec.Add(1, make([]byte, 1000))
require.NoError(t, err)
require.True(t, ready)
require.Len(t, dec.syms, 3)
out, err := dec.Decode()
require.NoError(t, err)
require.Equal(t, msg, out)
}
func TestCorruptPadding(t *testing.T) {
// 3001 bytes in 3 symbols of 1001 leaves 2 bytes of padding in the last one.
dec, err := NewDecoder(3001, 1176)
require.NoError(t, err)
require.Equal(t, 1001, dec.Layout().SymbolSize)
last := make([]byte, 1001)
last[1000] = 1
for i, s := range [][]byte{make([]byte, 1001), make([]byte, 1001), last} {
_, err = dec.Add(i, s)
require.NoError(t, err)
}
_, err = dec.Decode()
require.ErrorIs(t, err, ErrCorrupt)
}
// TestKnownAnswer pins the encoding. Both ends of a handshake must produce
// identical repair symbols, so any change here is a wire format change.
func TestKnownAnswer(t *testing.T) {
msg := make([]byte, 3000)
for i := range msg {
msg[i] = byte(i*7 + 3)
}
enc, err := NewEncoder(msg, 1176)
require.NoError(t, err)
// Cross-checked against an independent bit-serial GF(2^8) implementation.
want := map[int]string{
3: "52124f58276ec6338d3b3f4eed273bc7cabfe4f0f467928962b046ad1b348f50",
4: "a50928b04738981985a7d4a17ffaf8d7e30014ca64fc53c99800ffb61aec4aa7",
5: "ada4c95a7facbc542aaef584e6f0d081b7b0d0eed69ded03b57d0753a119cfc9",
255: "839066cdf94f963b223fb1c1c1360f698c14f6349b25653ff5b521791a4021b1",
}
for idx, h := range want {
sym, err := enc.Symbol(nil, idx)
require.NoError(t, err)
sum := sha256.Sum256(sym)
assert.Equal(t, h, hex.EncodeToString(sum[:]), "symbol %d", idx)
}
}
func FuzzRoundTrip(f *testing.F) {
f.Add([]byte("hello"), uint16(2), uint64(0))
f.Add(make([]byte, 9000), uint16(1176), uint64(7))
f.Fuzz(func(t *testing.T, msg []byte, maxSym uint16, seed uint64) {
if len(msg) == 0 || maxSym == 0 {
return
}
enc, err := NewEncoder(msg, int(maxSym))
if err != nil {
return
}
r := rand.New(rand.NewPCG(seed, seed))
idx := r.Perm(MaxSymbols)[:enc.Layout().K]
require.Equal(t, msg, decodeWith(t, enc, len(msg), int(maxSym), idx))
})
}
// FuzzDecoder feeds arbitrary symbols to a decoder; it must never panic and
// must either fail or return a message of the declared length.
func FuzzDecoder(f *testing.F) {
f.Add(uint16(3000), []byte{0, 5, 9, 200}, []byte("some symbol bytes"))
f.Fuzz(func(t *testing.T, msgLen uint16, indexes []byte, fill []byte) {
dec, err := NewDecoder(int(msgLen), 1176)
if err != nil {
return
}
size := dec.Layout().SymbolSize
for i, idx := range indexes {
sym := make([]byte, size)
for j := range sym {
if len(fill) > 0 {
sym[j] = fill[(i+j)%len(fill)]
}
}
if _, err := dec.Add(int(idx), sym); err != nil {
t.Fatal(err)
}
}
out, err := dec.Decode()
if err == nil {
require.Len(t, out, int(msgLen))
}
})
}
// BenchmarkRepair and BenchmarkDecode use a v3-sized msg2 (~9.8KB, k=9) with
// three repair symbols, decoding with three data symbols missing.
func BenchmarkRepair(b *testing.B) {
msg := randMsg(rand.New(rand.NewPCG(7, 8)), 9800)
enc, err := NewEncoder(msg, 1176)
require.NoError(b, err)
buf := make([]byte, 0, 1176)
b.SetBytes(int64(len(msg)))
for b.Loop() {
for idx := 9; idx < 12; idx++ {
buf, _ = enc.Symbol(buf[:0], idx)
}
}
}
func BenchmarkDecode(b *testing.B) {
msg := randMsg(rand.New(rand.NewPCG(7, 8)), 9800)
enc, err := NewEncoder(msg, 1176)
require.NoError(b, err)
idx := []int{0, 1, 2, 3, 5, 6, 9, 10, 11}
syms := make([][]byte, len(idx))
for i, x := range idx {
syms[i], _ = enc.Symbol(nil, x)
}
b.SetBytes(int64(len(msg)))
for b.Loop() {
dec, _ := NewDecoder(len(msg), 1176)
for i, x := range idx {
_, _ = dec.Add(x, syms[i])
}
if _, err := dec.Decode(); err != nil {
b.Fatal(err)
}
}
}
-115
View File
@@ -1,115 +0,0 @@
package fec
// Arithmetic over GF(2^8) with the reducing polynomial x^8+x^4+x^3+x^2+1 (0x11d).
// Addition is XOR. Multiplication uses log/exp tables; the exp table is doubled so
// the sum of two logs never needs reducing mod 255.
const gfPoly = 0x11d
var (
gfExp [510]byte
gfLog [256]byte
)
func init() {
x := 1
for i := range 255 {
gfExp[i] = byte(x)
gfExp[i+255] = byte(x)
gfLog[x] = byte(i)
x <<= 1
if x&0x100 != 0 {
x ^= gfPoly
}
}
}
func gfMul(a, b byte) byte {
if a == 0 || b == 0 {
return 0
}
return gfExp[int(gfLog[a])+int(gfLog[b])]
}
// gfInv returns the multiplicative inverse of a. a must be non-zero.
func gfInv(a byte) byte {
return gfExp[255-int(gfLog[a])]
}
// mulAdd computes dst ^= c*src element-wise. dst must be at least as long as src.
func mulAdd(dst, src []byte, c byte) {
switch c {
case 0:
return
case 1:
for i, s := range src {
dst[i] ^= s
}
return
}
var t [256]byte
lc := int(gfLog[c])
for s := 1; s < 256; s++ {
t[s] = gfExp[int(gfLog[s])+lc]
}
for i, s := range src {
dst[i] ^= t[s]
}
}
// coefficient returns the generator matrix entry for repair symbol index row and
// data symbol index col. Repair rows are a Cauchy matrix 1/(x_row + y_col) with
// x_row = row and y_col = col: rows are always >= k and columns < k so the two
// sets are disjoint, which makes every square submatrix invertible and the
// systematic code [I; C] MDS. A row depends only on its index, never on how many
// repair symbols exist, so new repair symbols can be generated at any time.
func coefficient(row, col int) byte {
return gfInv(byte(row) ^ byte(col))
}
// invert inverts the n x n matrix m in place with Gauss-Jordan elimination.
// It returns false if m is singular.
func invert(m [][]byte) bool {
n := len(m)
inv := make([][]byte, n)
for i := range inv {
inv[i] = make([]byte, n)
inv[i][i] = 1
}
for col := range n {
pivot := -1
for r := col; r < n; r++ {
if m[r][col] != 0 {
pivot = r
break
}
}
if pivot < 0 {
return false
}
m[col], m[pivot] = m[pivot], m[col]
inv[col], inv[pivot] = inv[pivot], inv[col]
scale := gfInv(m[col][col])
for c := range n {
m[col][c] = gfMul(m[col][c], scale)
inv[col][c] = gfMul(inv[col][c], scale)
}
for r := 0; r < n; r++ {
if r == col || m[r][col] == 0 {
continue
}
f := m[r][col]
for c := range n {
m[r][c] ^= gfMul(f, m[col][c])
inv[r][c] ^= gfMul(f, inv[col][c])
}
}
}
copy(m, inv)
return true
}
-67
View File
@@ -1,67 +0,0 @@
package fragment
import (
"errors"
"github.com/slackhq/nebula/handshake/fec"
"github.com/slackhq/nebula/header"
)
// ErrMismatch is returned when a fragment disagrees with the message it claims to
// belong to. The caller should discard the whole assembly.
var ErrMismatch = errors.New("fragment does not match its message")
// Assembly collects the fragments of one incoming message. Its memory is bounded
// by the message length: at most K symbols are kept however many arrive.
//
// An Assembly is not safe for concurrent use.
type Assembly struct {
fragID uint32
msgLen uint16
dec *fec.Decoder
received int
}
// NewAssembly starts an assembly for the message described by h. The fragment
// itself still has to be passed to Add.
func NewAssembly(h Header) (*Assembly, error) {
dec, err := fec.NewDecoder(int(h.MsgLen), MaxSymbolSize)
if err != nil {
return nil, err
}
return &Assembly{fragID: h.FragID, msgLen: h.MsgLen, dec: dec}, nil
}
// Add records a fragment parsed by Parse. It returns true once enough symbols
// are held to call Decode.
func (a *Assembly) Add(h Header, sym []byte) (bool, error) {
if h.FragID != a.fragID || h.MsgLen != a.msgLen {
return a.dec.Ready(), ErrMismatch
}
a.received += header.Len + HeaderLen + len(sym)
return a.dec.Add(int(h.Index), sym)
}
// wouldStore returns how many bytes Add would retain for this fragment.
func (a *Assembly) wouldStore(h Header, sym []byte) int {
if h.FragID != a.fragID || h.MsgLen != a.msgLen || a.dec.Ready() || a.dec.Has(int(h.Index)) {
return 0
}
return len(sym)
}
// Decode reconstructs the message. It requires Add to have reported ready.
func (a *Assembly) Decode() ([]byte, error) {
return a.dec.Decode()
}
// Stored returns the number of symbol bytes held.
func (a *Assembly) Stored() int {
return a.dec.Stored() * a.dec.Layout().SymbolSize
}
// Received returns the number of packet bytes received for this message,
// duplicates included. It is the basis for the amplification budget.
func (a *Assembly) Received() int {
return a.received
}
-515
View File
@@ -1,515 +0,0 @@
package fragment
import (
"bytes"
"math/rand/v2"
"sync"
"testing"
"time"
"github.com/slackhq/nebula/header"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
const testSubtype = header.MessageSubType(9)
type testKey struct {
src string
fragID uint32
}
func randMsg(r *rand.Rand, n int) []byte {
b := make([]byte, n)
for i := range b {
b[i] = byte(r.Uint32())
}
return b
}
func testConfig() TableConfig {
return TableConfig{
MaxBytes: 1 << 20,
MaxEntries: 64,
MaxPerSource: 4,
MaxMsgLen: 16384,
TTL: 2 * time.Second,
}
}
// checkInvariants verifies the table's bookkeeping against its contents.
func checkInvariants[K, S comparable](t *testing.T, tb *Table[K, S]) {
t.Helper()
tb.mu.Lock()
defer tb.mu.Unlock()
require.Len(t, tb.order, len(tb.entries))
require.LessOrEqual(t, len(tb.entries), tb.cfg.MaxEntries)
bytes := 0
perSource := map[S]int{}
for i, key := range tb.order {
e, ok := tb.entries[key]
require.True(t, ok)
require.Equal(t, i, e.pos)
bytes += e.a.Stored()
perSource[e.source]++
}
require.Equal(t, bytes, tb.bytes)
require.LessOrEqual(t, tb.bytes, tb.cfg.MaxBytes)
require.Equal(t, perSource, tb.perSource)
for _, n := range perSource {
require.LessOrEqual(t, n, tb.cfg.MaxPerSource)
}
}
func parse(t *testing.T, p []byte) (header.H, Header, []byte) {
t.Helper()
var nh header.H
require.NoError(t, nh.Parse(p))
h, sym, err := Parse(p[header.Len:])
require.NoError(t, err)
return nh, h, sym
}
func TestParse(t *testing.T) {
h := Header{FragID: 0xdeadbeef, MsgLen: 3000, Index: 7}
b := h.Encode(nil)
require.Len(t, b, HeaderLen)
b = append(b, make([]byte, 1000)...)
got, sym, err := Parse(b)
require.NoError(t, err)
assert.Equal(t, h, got)
assert.Len(t, sym, 1000)
_, _, err = Parse(b[:HeaderLen-1])
require.ErrorIs(t, err, ErrShortFragment)
_, _, err = Parse(b[:len(b)-1])
require.ErrorIs(t, err, ErrSymbolSize)
_, _, err = Parse(append(b, 0))
require.ErrorIs(t, err, ErrSymbolSize)
bad := Header{FragID: 0, MsgLen: 3000}.Encode(nil)
_, _, err = Parse(append(bad, make([]byte, 1000)...))
require.ErrorIs(t, err, ErrFragID)
bad = Header{FragID: 1, MsgLen: 0}.Encode(nil)
_, _, err = Parse(bad)
require.ErrorIs(t, err, ErrMsgLen)
bad = Header{FragID: 1, MsgLen: 3000, Flags: 0x02}.Encode(nil)
_, _, err = Parse(append(bad, make([]byte, 1000)...))
require.ErrorIs(t, err, ErrFlags)
t.Run("cookie", func(t *testing.T) {
h := Header{FragID: 7, MsgLen: 3000, Index: 4, Flags: FlagCookie}
for i := range h.Cookie {
h.Cookie[i] = byte(i + 1)
}
b := h.Encode(nil)
require.Len(t, b, HeaderLen+CookieLen)
b = append(b, make([]byte, 1000)...)
got, sym, err := Parse(b)
require.NoError(t, err)
assert.Equal(t, h, got)
assert.True(t, got.HasCookie())
assert.Len(t, sym, 1000)
_, _, err = Parse(b[:HeaderLen+CookieLen-1])
require.ErrorIs(t, err, ErrNoCookie)
})
}
func TestDefaultRepair(t *testing.T) {
for k, want := range map[int]int{1: 2, 2: 2, 6: 2, 7: 3, 9: 3, 10: 4, 14: 5} {
assert.Equal(t, want, DefaultRepair(k), "k=%d", k)
}
}
func TestSender(t *testing.T) {
_, err := NewSender(nil, testSubtype, 0, 1, 1)
require.ErrorIs(t, err, ErrMsgLen)
_, err = NewSender(make([]byte, 65536), testSubtype, 0, 1, 1)
require.ErrorIs(t, err, ErrMsgLen)
_, err = NewSender(make([]byte, 10), testSubtype, 0, 1, 0)
require.ErrorIs(t, err, ErrFragID)
msg := randMsg(rand.New(rand.NewPCG(1, 1)), 9800)
s, err := NewSender(msg, testSubtype, 77, 2, 1234)
require.NoError(t, err)
require.Equal(t, 9, s.Layout().K)
indexes := func(ps [][]byte) []int {
var out []int
for _, p := range ps {
require.LessOrEqual(t, len(p), MaxPacketLen)
nh, h, _ := parse(t, p)
assert.Equal(t, header.Handshake, nh.Type)
assert.Equal(t, testSubtype, nh.Subtype)
assert.Equal(t, uint32(77), nh.RemoteIndex)
assert.Equal(t, uint64(2), nh.MessageCounter)
assert.Equal(t, uint32(1234), h.FragID)
assert.Equal(t, uint16(9800), h.MsgLen)
out = append(out, int(h.Index))
}
return out
}
now, later, err := s.Attempt(3)
require.NoError(t, err)
assert.Equal(t, []int{0, 1, 2, 3, 4, 5, 6, 7, 8}, indexes(now))
assert.Equal(t, []int{9, 10, 11}, indexes(later))
// A retry offers only indexes the receiver has not seen.
now, later, err = s.Attempt(3)
require.NoError(t, err)
assert.Equal(t, []int{12, 13, 14, 15, 16, 17, 18, 19, 20}, indexes(now))
assert.Equal(t, []int{21, 22, 23}, indexes(later))
// A sender rebuilt from the same message resumes where it was told to.
s2, err := NewSender(msg, testSubtype, 77, 2, 1234)
require.NoError(t, err)
s2.SetNext(s.Next())
now2, _, err := s2.Attempt(0)
require.NoError(t, err)
assert.Equal(t, []int{24, 25, 26, 27, 28, 29, 30, 31, 32}, indexes(now2))
s2.SetNext(-1)
assert.Equal(t, 255, s2.Next())
// A cookie rides on every later packet, which still fits MaxPacketLen even
// with a full size symbol.
full, err := NewSender(make([]byte, 2*MaxSymbolSize), testSubtype, 0, 1, 9)
require.NoError(t, err)
full.SetCookie(Cookie{1, 2, 3})
pkts, _, err := full.Attempt(1)
require.NoError(t, err)
for _, p := range pkts {
assert.Len(t, p, MaxPacketLen)
_, h, _ := parse(t, p)
assert.True(t, h.HasCookie())
assert.Equal(t, Cookie{1, 2, 3}, h.Cookie)
}
// Indexes wrap once the space is used up.
for range 19 {
_, _, err = s.Attempt(3)
require.NoError(t, err)
}
now, _, err = s.Attempt(3)
require.NoError(t, err)
assert.Equal(t, []int{252, 253, 254, 255, 0, 1, 2, 3, 4}, indexes(now))
}
// TestLossy runs whole messages through heavy loss, reordering and duplication,
// merging across attempts, and checks the table ends empty.
func TestLossy(t *testing.T) {
r := rand.New(rand.NewPCG(2, 2))
tb := NewTable[testKey, string](testConfig())
now := time.Unix(1000, 0)
for i := range 300 {
msg := randMsg(r, 1+r.IntN(16384))
fragID := uint32(i + 1)
s, err := NewSender(msg, testSubtype, 0, 1, fragID)
require.NoError(t, err)
var got *Result
for attempt := 0; got == nil; attempt++ {
require.Less(t, attempt, 20, "message never completed")
a, b, err := s.Attempt(DefaultRepair(s.Layout().K))
require.NoError(t, err)
var pkts [][]byte
for _, p := range append(a, b...) {
if r.Float64() < 0.3 {
continue
}
pkts = append(pkts, p)
if r.Float64() < 0.2 {
pkts = append(pkts, p)
}
}
r.Shuffle(len(pkts), func(i, j int) { pkts[i], pkts[j] = pkts[j], pkts[i] })
for _, p := range pkts {
_, h, sym := parse(t, p)
res, err := tb.Add(testKey{"a", h.FragID}, "a", now, h, sym)
require.NoError(t, err)
if res != nil {
// The table keeps no memory of completed messages, so a late
// duplicate can complete one again; the handshake layer
// recognises the repeat.
require.True(t, bytes.Equal(msg, res.Msg))
got = res
}
checkInvariants(t, tb)
}
now = now.Add(100 * time.Millisecond)
}
require.Positive(t, got.Received)
// Late fragments of a completed message start a fresh entry; drop it.
tb.Expire(now.Add(time.Hour))
}
assert.Zero(t, tb.Len())
assert.Zero(t, tb.Bytes())
}
// TestConcurrent spreads each message's fragments across goroutines, as
// multiple UDP reader routines would.
func TestConcurrent(t *testing.T) {
tb := NewTable[testKey, string](testConfig())
now := time.Unix(1000, 0)
r := rand.New(rand.NewPCG(3, 3))
const msgs = 50
want := make(map[uint32][]byte, msgs)
var pkts [][]byte
for id := uint32(1); id <= msgs; id++ {
msg := randMsg(r, 1+r.IntN(16384))
want[id] = msg
s, err := NewSender(msg, testSubtype, 0, 1, id)
require.NoError(t, err)
a, b, err := s.Attempt(DefaultRepair(s.Layout().K))
require.NoError(t, err)
pkts = append(pkts, a...)
pkts = append(pkts, b...)
}
r.Shuffle(len(pkts), func(i, j int) { pkts[i], pkts[j] = pkts[j], pkts[i] })
var mu sync.Mutex
done := map[uint32]bool{}
var wg sync.WaitGroup
for w := range 8 {
wg.Go(func() {
for i := w; i < len(pkts); i += 8 {
h, sym, err := Parse(pkts[i][16:])
if err != nil {
t.Error(err)
return
}
src := string(rune('a' + h.FragID))
res, err := tb.Add(testKey{src, h.FragID}, src, now, h, sym)
if err != nil {
t.Error(err)
return
}
if res != nil {
if !bytes.Equal(want[h.FragID], res.Msg) {
t.Errorf("message %d decoded wrong", h.FragID)
}
mu.Lock()
done[h.FragID] = true
mu.Unlock()
}
}
})
}
wg.Wait()
assert.Len(t, done, msgs)
checkInvariants(t, tb)
}
func firstFragment(t *testing.T, msgLen int, fragID uint32) (Header, []byte) {
t.Helper()
s, err := NewSender(make([]byte, msgLen), testSubtype, 0, 1, fragID)
require.NoError(t, err)
now, _, err := s.Attempt(0)
require.NoError(t, err)
_, h, sym := parse(t, now[0])
return h, sym
}
func TestTableLimits(t *testing.T) {
now := time.Unix(1000, 0)
t.Run("max message length", func(t *testing.T) {
tb := NewTable[testKey, string](testConfig())
h, sym := firstFragment(t, 16385, 1)
_, err := tb.Add(testKey{"a", 1}, "a", now, h, sym)
require.ErrorIs(t, err, ErrTooLarge)
assert.Zero(t, tb.Len())
})
t.Run("per source quota", func(t *testing.T) {
cfg := testConfig()
cfg.MaxPerSource = 2
tb := NewTable[testKey, string](cfg)
for id := uint32(1); id <= 3; id++ {
h, sym := firstFragment(t, 5000, id)
_, err := tb.Add(testKey{"a", id}, "a", now, h, sym)
if id <= 2 {
require.NoError(t, err)
} else {
require.ErrorIs(t, err, ErrSourceQuota)
}
}
h, sym := firstFragment(t, 5000, 4)
_, err := tb.Add(testKey{"b", 4}, "b", now, h, sym)
require.NoError(t, err)
assert.Equal(t, 3, tb.Len())
checkInvariants(t, tb)
})
t.Run("max entries evicts", func(t *testing.T) {
cfg := testConfig()
cfg.MaxEntries = 4
tb := NewTable[testKey, string](cfg)
for id := uint32(1); id <= 10; id++ {
h, sym := firstFragment(t, 5000, id)
src := string(rune('a' + id))
_, err := tb.Add(testKey{src, id}, src, now, h, sym)
require.NoError(t, err)
checkInvariants(t, tb)
}
assert.Equal(t, 4, tb.Len())
})
t.Run("max bytes evicts others", func(t *testing.T) {
cfg := testConfig()
cfg.MaxBytes = 3 * 1000
tb := NewTable[testKey, string](cfg)
for id := uint32(1); id <= 10; id++ {
h, sym := firstFragment(t, 5000, id) // 5 symbols of 1000
src := string(rune('a' + id))
_, err := tb.Add(testKey{src, id}, src, now, h, sym)
require.NoError(t, err)
checkInvariants(t, tb)
}
assert.Equal(t, 3000, tb.Bytes())
})
t.Run("no room removes new entry", func(t *testing.T) {
cfg := testConfig()
cfg.MaxBytes = 500
tb := NewTable[testKey, string](cfg)
h, sym := firstFragment(t, 5000, 1)
_, err := tb.Add(testKey{"a", 1}, "a", now, h, sym)
require.ErrorIs(t, err, ErrNoRoom)
assert.Zero(t, tb.Len())
checkInvariants(t, tb)
})
t.Run("duplicates are not charged", func(t *testing.T) {
tb := NewTable[testKey, string](testConfig())
h, sym := firstFragment(t, 5000, 1)
for range 5 {
_, err := tb.Add(testKey{"a", 1}, "a", now, h, sym)
require.NoError(t, err)
}
assert.Equal(t, 1000, tb.Bytes())
})
t.Run("mismatch drops the entry", func(t *testing.T) {
tb := NewTable[testKey, string](testConfig())
h, sym := firstFragment(t, 5000, 1)
_, err := tb.Add(testKey{"a", 1}, "a", now, h, sym)
require.NoError(t, err)
h2, sym2 := firstFragment(t, 6000, 1)
_, err = tb.Add(testKey{"a", 1}, "a", now, h2, sym2)
require.ErrorIs(t, err, ErrMismatch)
assert.Zero(t, tb.Len())
assert.Zero(t, tb.Bytes())
checkInvariants(t, tb)
})
t.Run("load", func(t *testing.T) {
cfg := testConfig()
cfg.MaxEntries = 4
cfg.MaxBytes = 10000
tb := NewTable[testKey, string](cfg)
assert.Zero(t, tb.Load())
h, sym := firstFragment(t, 5000, 1)
_, err := tb.Add(testKey{"a", 1}, "a", now, h, sym)
require.NoError(t, err)
assert.InDelta(t, 0.25, tb.Load(), 0.001, "one of four entries")
h, sym = firstFragment(t, 5000, 2)
_, err = tb.Add(testKey{"b", 2}, "b", now, h, sym)
require.NoError(t, err)
assert.InDelta(t, 0.5, tb.Load(), 0.001)
})
t.Run("ttl", func(t *testing.T) {
tb := NewTable[testKey, string](testConfig())
s, err := NewSender(make([]byte, 5000), testSubtype, 0, 1, 1)
require.NoError(t, err)
pkts, _, err := s.Attempt(0)
require.NoError(t, err)
_, h, sym := parse(t, pkts[0])
_, err = tb.Add(testKey{"a", 1}, "a", now, h, sym)
require.NoError(t, err)
// Activity within the TTL keeps the entry and slides its deadline.
_, h, sym = parse(t, pkts[1])
_, err = tb.Add(testKey{"a", 1}, "a", now.Add(1500*time.Millisecond), h, sym)
require.NoError(t, err)
assert.Equal(t, 2000, tb.Bytes())
tb.Expire(now.Add(3 * time.Second))
assert.Equal(t, 1, tb.Len())
// A stale entry is restarted rather than merged.
_, h, sym = parse(t, pkts[2])
_, err = tb.Add(testKey{"a", 1}, "a", now.Add(10*time.Second), h, sym)
require.NoError(t, err)
assert.Equal(t, 1000, tb.Bytes())
tb.Expire(now.Add(time.Minute))
assert.Zero(t, tb.Len())
checkInvariants(t, tb)
})
}
func FuzzParse(f *testing.F) {
f.Add(append(Header{FragID: 1, MsgLen: 3000}.Encode(nil), make([]byte, 1000)...))
f.Add([]byte{0, 0, 0, 1, 0, 1, 0, 0, 7})
f.Fuzz(func(t *testing.T, b []byte) {
h, sym, err := Parse(b)
if err != nil {
return
}
require.Equal(t, b, append(h.Encode(nil), sym...))
l, err := h.Layout()
require.NoError(t, err)
require.LessOrEqual(t, l.SymbolSize, MaxSymbolSize)
})
}
// FuzzTable drives a small table with arbitrary fragments and checks its
// bookkeeping after every step.
func FuzzTable(f *testing.F) {
f.Add([]byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12})
f.Fuzz(func(t *testing.T, ops []byte) {
tb := NewTable[testKey, string](TableConfig{
MaxBytes: 4000,
MaxEntries: 3,
MaxPerSource: 2,
MaxMsgLen: 6000,
TTL: time.Second,
})
now := time.Unix(0, 0)
for len(ops) >= 4 {
src := string(rune('a' + ops[0]%4))
fragID := uint32(ops[1]%5) + 1
msgLen := uint16(ops[2])*32 + 1
index := ops[3]
ops = ops[4:]
h := Header{FragID: fragID, MsgLen: msgLen, Index: index}
l, err := h.Layout()
require.NoError(t, err)
sym := make([]byte, l.SymbolSize)
if index%7 == 0 {
now = now.Add(600 * time.Millisecond)
tb.Expire(now)
}
res, err := tb.Add(testKey{src, fragID}, src, now, h, sym)
if res != nil {
require.NoError(t, err)
require.Len(t, res.Msg, int(msgLen))
}
checkInvariants(t, tb)
}
})
}
-118
View File
@@ -1,118 +0,0 @@
package fragment
import (
"errors"
"math"
"github.com/slackhq/nebula/handshake/fec"
"github.com/slackhq/nebula/header"
)
// DefaultRepair is the starting number of repair symbols per attempt for a
// message of k data symbols. Small messages get proportionally more repair.
func DefaultRepair(k int) int {
return max(2, (k+2)/3)
}
// Sender produces the fragment packets of one outgoing handshake message.
//
// The first attempt sends the K data symbols followed by repair symbols. Every
// later attempt sends symbols with indexes the receiver has not been offered yet,
// so whatever it already holds still counts toward K. Symbols are a deterministic
// function of the message, which keeps retransmits idempotent. Once all
// fec.MaxSymbols indexes are used the sequence wraps and repeats.
//
// A Sender is not safe for concurrent use.
type Sender struct {
enc *fec.Encoder
subtype header.MessageSubType
remoteIndex uint32
counter uint64
fragID uint32
next int
cookie *Cookie
}
// NewSender prepares msg for sending. subtype, remoteIndex and counter fill the
// Nebula header of every fragment; fragID is the sender's local index.
func NewSender(msg []byte, subtype header.MessageSubType, remoteIndex uint32, counter uint64, fragID uint32) (*Sender, error) {
if len(msg) == 0 || len(msg) > math.MaxUint16 {
return nil, ErrMsgLen
}
if fragID == 0 {
return nil, ErrFragID
}
enc, err := fec.NewEncoder(msg, MaxSymbolSize)
if err != nil {
return nil, err
}
return &Sender{
enc: enc,
subtype: subtype,
remoteIndex: remoteIndex,
counter: counter,
fragID: fragID,
}, nil
}
func (s *Sender) Layout() fec.Layout {
return s.enc.Layout()
}
// SetNext sets the index the next attempt starts from, so a sender rebuilt from
// a cached message can carry on offering fresh symbols.
func (s *Sender) SetNext(index int) {
s.next = ((index % fec.MaxSymbols) + fec.MaxSymbols) % fec.MaxSymbols
}
// SetCookie attaches a cookie from a responder under load to every later packet.
func (s *Sender) SetCookie(c Cookie) {
s.cookie = &c
}
// Next returns the index the next attempt starts from.
func (s *Sender) Next() int {
return s.next
}
// Attempt returns the packets for one transmission attempt carrying K+repair
// symbols. now should be sent immediately and later after the repair delay, so
// a tail drop at the end of the first burst does not also take the repair.
func (s *Sender) Attempt(repair int) (now, later [][]byte, err error) {
if repair < 0 {
return nil, nil, errors.New("negative repair count")
}
k := s.enc.Layout().K
now = make([][]byte, 0, k)
later = make([][]byte, 0, repair)
for i := range k + repair {
p, err := s.packet(s.next)
if err != nil {
return nil, nil, err
}
s.next = (s.next + 1) % fec.MaxSymbols
if i < k {
now = append(now, p)
} else {
later = append(later, p)
}
}
return now, later, nil
}
func (s *Sender) packet(index int) ([]byte, error) {
l := s.enc.Layout()
p := make([]byte, header.Len, header.Len+HeaderLen+CookieLen+l.SymbolSize)
header.Encode(p, header.Version, header.Handshake, s.subtype, s.remoteIndex, s.counter)
h := Header{
FragID: s.fragID,
MsgLen: uint16(l.MsgLen),
Index: uint8(index),
}
if s.cookie != nil {
h.Flags |= FlagCookie
h.Cookie = *s.cookie
}
p = h.Encode(p)
return s.enc.Symbol(p, index)
}
-220
View File
@@ -1,220 +0,0 @@
package fragment
import (
"errors"
"math/rand/v2"
"sync"
"time"
)
var (
ErrTooLarge = errors.New("fragmented message exceeds the size limit")
ErrSourceQuota = errors.New("source has too many messages in progress")
ErrNoRoom = errors.New("no room for fragment")
)
// TableConfig bounds a Table.
type TableConfig struct {
MaxBytes int // symbol bytes held across all entries
MaxEntries int // messages in progress across all sources
MaxPerSource int // messages in progress per source
MaxMsgLen int // largest message accepted
TTL time.Duration // an entry idle this long is dropped
}
// Table reassembles messages from senders that have not authenticated yet, the
// responder's view of msg1. Every resource is capped. When a cap is hit a
// random entry is evicted; random eviction is harder to aim than LRU, which a
// flood can use to flush specific entries.
//
// K identifies one message, typically the source plus FragID. S identifies the
// source for quotas, typically the address masked to /32 or /64.
//
// A Table is safe for concurrent use.
type Table[K, S comparable] struct {
mu sync.Mutex
cfg TableConfig
entries map[K]*tableEntry[K, S]
order []K
perSource map[S]int
bytes int
}
type tableEntry[K, S comparable] struct {
a *Assembly
source S
lastSeen time.Time
pos int
}
func NewTable[K, S comparable](cfg TableConfig) *Table[K, S] {
return &Table[K, S]{
cfg: cfg,
entries: make(map[K]*tableEntry[K, S]),
perSource: make(map[S]int),
}
}
// Result describes a completed message.
type Result struct {
Msg []byte // the decoded message
Received int // packet bytes received for it, for the amplification budget
}
// Add records a fragment parsed by Parse. When the message is complete its entry
// is removed and the result returned; otherwise the result is nil. A decode
// error means the message is lost and must be retransmitted.
func (t *Table[K, S]) Add(key K, source S, now time.Time, h Header, sym []byte) (*Result, error) {
if int(h.MsgLen) > t.cfg.MaxMsgLen {
return nil, ErrTooLarge
}
t.mu.Lock()
e, ok := t.entries[key]
if ok && now.Sub(e.lastSeen) > t.cfg.TTL {
t.remove(key, e)
ok = false
}
if !ok {
var err error
e, err = t.insert(key, source, h)
if err != nil {
t.mu.Unlock()
return nil, err
}
}
e.lastSeen = now
if !t.makeRoom(key, e.a.wouldStore(h, sym)) {
if e.a.Stored() == 0 {
t.remove(key, e)
}
t.mu.Unlock()
return nil, ErrNoRoom
}
before := e.a.Stored()
ready, err := e.a.Add(h, sym)
t.bytes += e.a.Stored() - before
if err != nil || !ready {
if errors.Is(err, ErrMismatch) || (err != nil && e.a.Stored() == 0) {
t.remove(key, e)
}
t.mu.Unlock()
return nil, err
}
t.remove(key, e)
t.mu.Unlock()
msg, err := e.a.Decode()
if err != nil {
return nil, err
}
return &Result{Msg: msg, Received: e.a.Received()}, nil
}
// Expire drops every entry idle for longer than the TTL.
func (t *Table[K, S]) Expire(now time.Time) {
t.mu.Lock()
defer t.mu.Unlock()
for i := 0; i < len(t.order); {
key := t.order[i]
e := t.entries[key]
if now.Sub(e.lastSeen) > t.cfg.TTL {
// remove swaps the last entry into position i, so look at i again.
t.remove(key, e)
continue
}
i++
}
}
// Len returns the number of messages in progress.
func (t *Table[K, S]) Len() int {
t.mu.Lock()
defer t.mu.Unlock()
return len(t.entries)
}
// Load returns how full the table is, from 0 to 1, by whichever of its byte and
// entry limits is closer.
func (t *Table[K, S]) Load() float64 {
t.mu.Lock()
defer t.mu.Unlock()
return max(float64(t.bytes)/float64(t.cfg.MaxBytes), float64(len(t.entries))/float64(t.cfg.MaxEntries))
}
// Bytes returns the number of symbol bytes held.
func (t *Table[K, S]) Bytes() int {
t.mu.Lock()
defer t.mu.Unlock()
return t.bytes
}
func (t *Table[K, S]) insert(key K, source S, h Header) (*tableEntry[K, S], error) {
if t.perSource[source] >= t.cfg.MaxPerSource {
return nil, ErrSourceQuota
}
a, err := NewAssembly(h)
if err != nil {
return nil, err
}
for len(t.entries) >= t.cfg.MaxEntries {
if !t.evictRandom(nil) {
return nil, ErrNoRoom
}
}
e := &tableEntry[K, S]{a: a, source: source, pos: len(t.order)}
t.entries[key] = e
t.order = append(t.order, key)
t.perSource[source]++
return e, nil
}
// makeRoom evicts random entries other than keep until n more bytes fit.
func (t *Table[K, S]) makeRoom(keep K, n int) bool {
for t.bytes+n > t.cfg.MaxBytes {
if !t.evictRandom(&keep) {
return false
}
}
return true
}
func (t *Table[K, S]) evictRandom(keep *K) bool {
candidates := len(t.order)
if keep != nil {
if _, ok := t.entries[*keep]; ok {
candidates--
}
}
if candidates <= 0 {
return false
}
for {
key := t.order[rand.IntN(len(t.order))]
if keep != nil && key == *keep {
continue
}
t.remove(key, t.entries[key])
return true
}
}
func (t *Table[K, S]) remove(key K, e *tableEntry[K, S]) {
last := len(t.order) - 1
moved := t.order[last]
t.order[e.pos] = moved
t.entries[moved].pos = e.pos
t.order = t.order[:last]
delete(t.entries, key)
t.bytes -= e.a.Stored()
if t.perSource[e.source] <= 1 {
delete(t.perSource, e.source)
} else {
t.perSource[e.source]--
}
}
-121
View File
@@ -1,121 +0,0 @@
// Package fragment carries handshake messages that do not fit in one UDP
// packet. A message is Reed-Solomon coded into equally sized symbols (see
// handshake/fec) and each symbol travels in its own packet:
//
// Nebula header (16) | Fragment header (8) | [cookie (16)] | symbol
//
// The fragment header is FragID u32 | MsgLen u16 | Index u8 | Flags u8, all big
// endian. Both ends derive the symbol layout from MsgLen, so any MsgLen-derived
// value never has to be sent. FlagCookie marks a cookie a responder under load
// asked for; room for it is always reserved, so a fragment never outgrows
// MaxPacketLen.
package fragment
import (
"encoding/binary"
"errors"
"github.com/slackhq/nebula/handshake/fec"
"github.com/slackhq/nebula/header"
)
const (
// HeaderLen is the size of the fragment header that follows the Nebula header.
HeaderLen = 8
// CookieLen is the size of the cookie a fragment carries when FlagCookie is set.
CookieLen = 16
// FlagCookie marks a fragment that carries a cookie after its header.
FlagCookie uint8 = 0x01
// MaxPacketLen bounds every fragment packet, Nebula header included. A relayed
// fragment gains a 16 byte header and a 16 byte tag, which lands exactly on
// the 1232 byte UDP payload available on an IPv6 minimum MTU path.
MaxPacketLen = 1200
// MaxSymbolSize is the largest symbol that fits in MaxPacketLen alongside a cookie.
MaxSymbolSize = MaxPacketLen - header.Len - HeaderLen - CookieLen
)
var (
ErrShortFragment = errors.New("fragment is too short")
ErrFragID = errors.New("fragment has a zero FragID")
ErrMsgLen = errors.New("fragment has a zero MsgLen")
ErrFlags = errors.New("fragment has reserved flags set")
ErrNoCookie = errors.New("fragment is too short for its cookie")
ErrSymbolSize = errors.New("fragment symbol size does not match its MsgLen")
)
// Cookie proves a sender received a responder's reply at its address.
type Cookie [CookieLen]byte
// Header is the fragment header.
type Header struct {
FragID uint32 // the sender's local index for this handshake, never zero
MsgLen uint16 // length of the whole message
Index uint8 // symbol index, data below K and repair from K
Flags uint8 // FlagCookie, other bits reserved and zero
Cookie Cookie // valid when Flags has FlagCookie
}
func (h Header) HasCookie() bool {
return h.Flags&FlagCookie != 0
}
// Encode appends the header, and its cookie if it has one, to b.
func (h Header) Encode(b []byte) []byte {
b = binary.BigEndian.AppendUint32(b, h.FragID)
b = binary.BigEndian.AppendUint16(b, h.MsgLen)
b = append(b, h.Index, h.Flags)
if h.HasCookie() {
b = append(b, h.Cookie[:]...)
}
return b
}
// Layout returns the symbol layout the header's MsgLen implies.
func (h Header) Layout() (fec.Layout, error) {
return fec.NewLayout(int(h.MsgLen), MaxSymbolSize)
}
// Parse validates the bytes following the Nebula header and splits them into
// the fragment header and the symbol. The symbol aliases b.
func Parse(b []byte) (Header, []byte, error) {
if len(b) < HeaderLen {
return Header{}, nil, ErrShortFragment
}
h := Header{
FragID: binary.BigEndian.Uint32(b[0:4]),
MsgLen: binary.BigEndian.Uint16(b[4:6]),
Index: b[6],
Flags: b[7],
}
if h.FragID == 0 {
return Header{}, nil, ErrFragID
}
if h.MsgLen == 0 {
return Header{}, nil, ErrMsgLen
}
if h.Flags&^FlagCookie != 0 {
return Header{}, nil, ErrFlags
}
b = b[HeaderLen:]
if h.HasCookie() {
if len(b) < CookieLen {
return Header{}, nil, ErrNoCookie
}
copy(h.Cookie[:], b)
b = b[CookieLen:]
}
l, err := h.Layout()
if err != nil {
return Header{}, nil, err
}
sym := b
if len(sym) != l.SymbolSize {
return Header{}, nil, ErrSymbolSize
}
return h, sym, nil
}
+14 -3
View File
@@ -24,9 +24,20 @@ message NebulaHandshakeDetails {
uint64 Cookie = 4 [deprecated = true];
uint64 Time = 5;
uint32 CertVersion = 8;
// KEMKey is the initiator's ephemeral ML-KEM-1024 encapsulation key. It is
// only present in msg1 of the ix_psk2_mlkem1024 subtype.
bytes KEMKey = 9;
// reserved for WIP multiport
reserved 6, 7;
// Multiport lane negotiation. Absent on hosts without multiport enabled;
// vanilla nebula treats 9 and 10 as unknown fields and skips them.
LaneDetails InitiatorLanes = 9;
LaneDetails ResponderLanes = 10;
}
// LaneDetails advertises a host's multiport lane capability. On a base
// handshake LaneIndex is 0 and PortCount/BasePort describe the sender's
// consecutively bound UDP ports. On a lane handshake the initiator sets
// LaneIndex to its (nonzero) lane number.
message LaneDetails {
uint32 PortCount = 1;
uint32 BasePort = 2;
uint32 LaneIndex = 3;
}
+1
View File
@@ -71,6 +71,7 @@ func newTestMachine(
cs.version, cs.getCredential,
verifier, func() (uint32, error) { return localIndex, nil },
initiator, header.HandshakeIXPSK0,
nil,
)
require.NoError(t, err)
return m
-68
View File
@@ -1,68 +0,0 @@
package handshake
import (
"crypto/hkdf"
"crypto/mlkem"
"crypto/sha256"
"errors"
"fmt"
)
// kemPSKSalt separates the psk derivation from any other use of the secret.
const kemPSKSalt = "nebula ix_psk2_mlkem1024 v1"
var ErrInvalidKEMKey = errors.New("peer sent an invalid ML-KEM encapsulation key")
// kemPSK derives the noise psk from an ML-KEM shared secret. The ciphertext is
// bound in because, unlike the encapsulation key, it travels outside the noise
// transcript.
func kemPSK(sharedKey, ciphertext []byte) ([]byte, error) {
secret := make([]byte, 0, len(sharedKey)+len(ciphertext))
secret = append(secret, sharedKey...)
secret = append(secret, ciphertext...)
return hkdf.Extract(sha256.New, secret, []byte(kemPSKSalt))
}
// kemEncapsulate is the responder half: encapsulate to the initiator's key,
// install the psk, and remember the ciphertext to prefix the response with.
func (m *Machine) kemEncapsulate(encapsulationKey []byte) error {
ek, err := mlkem.NewEncapsulationKey1024(encapsulationKey)
if err != nil {
return fmt.Errorf("%w: %w", ErrInvalidKEMKey, err)
}
sharedKey, ct := ek.Encapsulate()
psk, err := kemPSK(sharedKey, ct)
if err != nil {
return err
}
if err := m.hs.SetPresharedKey(psk); err != nil {
return err
}
m.kemCiphertext = ct
return nil
}
// kemDecapsulate is the initiator half: recover the shared secret from the
// ciphertext prefixing the response and install the psk. It returns the rest of
// the message, which is the noise message.
func (m *Machine) kemDecapsulate(msg []byte) ([]byte, error) {
if len(msg) < mlkem.CiphertextSize1024 {
return nil, ErrPacketTooShort
}
if m.kemKey == nil {
return nil, errors.New("no ML-KEM decapsulation key for this handshake")
}
ct := msg[:mlkem.CiphertextSize1024]
sharedKey, err := m.kemKey.Decapsulate(ct)
if err != nil {
return nil, err
}
psk, err := kemPSK(sharedKey, ct)
if err != nil {
return nil, err
}
if err := m.hs.SetPresharedKey(psk); err != nil {
return nil, err
}
return msg[mlkem.CiphertextSize1024:], nil
}
-208
View File
@@ -1,208 +0,0 @@
package handshake
import (
"bytes"
"crypto/mlkem"
"net/netip"
"testing"
"time"
"github.com/flynn/noise"
"github.com/slackhq/nebula/cert"
ct "github.com/slackhq/nebula/cert_test"
"github.com/slackhq/nebula/header"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func newKEMMachine(t *testing.T, cs *testCertState, v CertVerifier, initiator bool, index uint32) *Machine {
t.Helper()
m, err := NewMachine(
cs.version, cs.getCredential, v,
func() (uint32, error) { return index, nil },
initiator, header.HandshakeIXPSK2MLKEM1024,
)
require.NoError(t, err)
return m
}
type kemFixture struct {
initCS, respCS *testCertState
v CertVerifier
}
func newKEMFixture(t *testing.T, cipher noise.CipherFunc) kemFixture {
ca, _, caKey, _ := ct.NewTestCaCert(
cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
)
return kemFixture{
initCS: newTestCertStateWithCipher(t, ca, caKey, "initiator", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")}, cipher),
respCS: newTestCertStateWithCipher(t, ca, caKey, "responder", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")}, cipher),
v: testVerifier(ct.NewTestCAPool(ca)),
}
}
func requireKeysMatch(t *testing.T, a, b *Result) {
t.Helper()
c, err := a.EKey.Encrypt(nil, nil, []byte("hello"))
require.NoError(t, err)
p, err := b.DKey.Decrypt(nil, nil, c)
require.NoError(t, err)
assert.Equal(t, []byte("hello"), p)
c, err = b.EKey.Encrypt(nil, nil, []byte("world"))
require.NoError(t, err)
p, err = a.DKey.Decrypt(nil, nil, c)
require.NoError(t, err)
assert.Equal(t, []byte("world"), p)
}
func TestKEMHandshake(t *testing.T) {
for _, cipher := range []noise.CipherFunc{noise.CipherChaChaPoly, noise.CipherAESGCM} {
t.Run(cipher.CipherName(), func(t *testing.T) {
f := newKEMFixture(t, cipher)
initM := newKEMMachine(t, f.initCS, f.v, true, 1000)
respM := newKEMMachine(t, f.respCS, f.v, false, 2000)
msg1, err := initM.Initiate(nil)
require.NoError(t, err)
assert.Equal(t, header.HandshakeIXPSK2MLKEM1024, header.MessageSubType(msg1[1]))
assert.Greater(t, len(msg1), header.Len+mlkem.EncapsulationKeySize1024)
msg2, respR, err := respM.ProcessPacket(nil, msg1)
require.NoError(t, err)
require.NotNil(t, respR)
assert.Greater(t, len(msg2), header.Len+mlkem.CiphertextSize1024)
_, initR, err := initM.ProcessPacket(nil, msg2)
require.NoError(t, err)
require.NotNil(t, initR)
assert.Equal(t, "responder", initR.RemoteCert.Certificate.Name())
assert.Equal(t, "initiator", respR.RemoteCert.Certificate.Name())
assert.Equal(t, uint32(2000), initR.RemoteIndex)
assert.Equal(t, uint32(1000), respR.RemoteIndex)
requireKeysMatch(t, initR, respR)
})
}
}
// TestKEMSecretMatters checks the ML-KEM secret really reaches the transport
// keys: an initiator holding a different decapsulation key cannot complete even
// though every DH input is genuine.
func TestKEMSecretMatters(t *testing.T) {
f := newKEMFixture(t, noise.CipherChaChaPoly)
initM := newKEMMachine(t, f.initCS, f.v, true, 1000)
respM := newKEMMachine(t, f.respCS, f.v, false, 2000)
msg1, err := initM.Initiate(nil)
require.NoError(t, err)
msg2, _, err := respM.ProcessPacket(nil, msg1)
require.NoError(t, err)
other, err := mlkem.GenerateKey1024()
require.NoError(t, err)
initM.kemKey = other
_, res, err := initM.ProcessPacket(nil, msg2)
require.Error(t, err)
assert.Nil(t, res)
}
func TestKEMTamperedCiphertext(t *testing.T) {
f := newKEMFixture(t, noise.CipherChaChaPoly)
initM := newKEMMachine(t, f.initCS, f.v, true, 1000)
respM := newKEMMachine(t, f.respCS, f.v, false, 2000)
msg1, err := initM.Initiate(nil)
require.NoError(t, err)
msg2, respR, err := respM.ProcessPacket(nil, msg1)
require.NoError(t, err)
t.Run("truncated", func(t *testing.T) {
_, res, err := initM.ProcessPacket(nil, msg2[:header.Len+mlkem.CiphertextSize1024-1])
require.ErrorIs(t, err, ErrPacketTooShort)
assert.Nil(t, res)
assert.False(t, initM.Failed())
})
t.Run("flipped ciphertext bit", func(t *testing.T) {
bad := bytes.Clone(msg2)
bad[header.Len+10] ^= 1
_, res, err := initM.ProcessPacket(nil, bad)
require.Error(t, err)
assert.Nil(t, res)
assert.False(t, initM.Failed(), "a forged ciphertext must not kill the handshake")
})
t.Run("genuine response still completes", func(t *testing.T) {
_, initR, err := initM.ProcessPacket(nil, msg2)
require.NoError(t, err)
require.NotNil(t, initR)
requireKeysMatch(t, initR, respR)
})
}
// TestKEMBadMsg1 drives processPayload by hand. With the psk modifier msg1's
// payload is encrypted under a key derived from the public ephemeral, so a
// forged payload would need re-encrypting; the parsing rules are what matter.
func TestKEMBadMsg1(t *testing.T) {
f := newKEMFixture(t, noise.CipherChaChaPoly)
msg1Flags := subtypeInfos[header.HandshakeIXPSK2MLKEM1024].msgs[0]
cert := f.initCS.creds[cert.Version2]
payload := func(kemKey []byte) []byte {
return MarshalPayload(nil, Payload{
Cert: cert.Bytes,
CertVersion: uint32(cert.Cert.Version()),
InitiatorIndex: 1000,
Time: 1,
KEMKey: kemKey,
})
}
dk, err := mlkem.GenerateKey1024()
require.NoError(t, err)
t.Run("missing KEM key", func(t *testing.T) {
m := newKEMMachine(t, f.respCS, f.v, false, 2000)
err := m.processPayload(payload(nil), msg1Flags)
require.ErrorIs(t, err, ErrUnexpectedContent)
assert.True(t, m.Failed())
})
t.Run("invalid KEM key", func(t *testing.T) {
m := newKEMMachine(t, f.respCS, f.v, false, 2000)
// Coefficients of 0xfff exceed the modulus, which ML-KEM must reject.
bad := bytes.Repeat([]byte{0xff}, mlkem.EncapsulationKeySize1024)
err := m.kemEncapsulate(bad)
require.ErrorIs(t, err, ErrInvalidKEMKey)
err = m.kemEncapsulate(bad[:100])
require.ErrorIs(t, err, ErrInvalidKEMKey)
})
t.Run("valid KEM key", func(t *testing.T) {
m := newKEMMachine(t, f.respCS, f.v, false, 2000)
require.NoError(t, m.kemEncapsulate(dk.EncapsulationKey().Bytes()))
assert.Len(t, m.kemCiphertext, mlkem.CiphertextSize1024)
})
t.Run("KEM key on the classical subtype", func(t *testing.T) {
m := newTestMachine(t, f.respCS, f.v, false, 2000)
classical := subtypeInfos[header.HandshakeIXPSK0].msgs[0]
err := m.processPayload(payload(dk.EncapsulationKey().Bytes()), classical)
require.ErrorIs(t, err, ErrUnexpectedContent)
assert.True(t, m.Failed())
})
}
// TestKEMDistinctFromClassical makes sure the two subtypes cannot be mixed: a
// classical responder cannot answer a KEM initiator.
func TestKEMDistinctFromClassical(t *testing.T) {
f := newKEMFixture(t, noise.CipherChaChaPoly)
initM := newKEMMachine(t, f.initCS, f.v, true, 1000)
respM := newTestMachine(t, f.respCS, f.v, false, 2000)
msg1, err := initM.Initiate(nil)
require.NoError(t, err)
_, _, err = respM.ProcessPacket(nil, msg1)
require.ErrorIs(t, err, ErrSubtypeMismatch)
}
+39 -55
View File
@@ -2,8 +2,6 @@ package handshake
import (
"bytes"
"crypto/mlkem"
"errors"
"fmt"
"slices"
"time"
@@ -41,6 +39,14 @@ type Result struct {
HandshakeTime uint64
MessageIndex uint64 // number of messages exchanged during the handshake
Initiator bool
// Multiport lane negotiation, from the peer's LaneDetails. All zero when
// the peer did not advertise (vanilla peer or multiport disabled).
// PeerTxLanes is how many lanes the peer may send on, which is how many lane
// sessions we need in order to receive everything it sends.
PeerPortCount uint32
PeerBasePort uint32
PeerTxLanes uint32
}
// Machine drives a Noise handshake through N messages. It handles Noise
@@ -63,21 +69,21 @@ type Machine struct {
verifier CertVerifier
result *Result
msgs []msgFlags
lanes *LaneDetails // our multiport advert; nil emits a vanilla payload
myVersion cert.Version
subtype header.MessageSubType
indexAllocated bool
remoteCertSet bool
payloadSet bool
failed bool
kemKey *mlkem.DecapsulationKey1024 // initiator's ephemeral ML-KEM key, until completion
kemCiphertext []byte // responder's ciphertext to prefix its response with
}
// NewMachine creates a handshake state machine. The subtype determines both
// the noise pattern and the per-message content layout. The credential for
// `version` is fetched via getCred and used to seed the noise.HandshakeState.
// IndexAllocator is called lazily when the first outgoing payload is built.
// lanes, when non-nil, is emitted as this side's multiport advert on every
// payload-bearing message; nil produces byte-identical vanilla payloads.
func NewMachine(
version cert.Version,
getCred GetCredentialFunc,
@@ -85,6 +91,7 @@ func NewMachine(
allocIndex IndexAllocator,
initiator bool,
subtype header.MessageSubType,
lanes *LaneDetails,
) (*Machine, error) {
info, err := subtypeInfoFor(subtype)
if err != nil {
@@ -96,7 +103,7 @@ func NewMachine(
return nil, fmt.Errorf("%w: %v", ErrNoCredential, version)
}
hs, err := cred.buildHandshakeState(initiator, info.pattern, info.pskPlacement)
hs, err := cred.buildHandshakeState(initiator, info.pattern)
if err != nil {
return nil, fmt.Errorf("build noise state: %w", err)
}
@@ -108,6 +115,7 @@ func NewMachine(
getCred: getCred,
allocIndex: allocIndex,
verifier: verifier,
lanes: lanes,
myVersion: version,
result: &Result{
Initiator: initiator,
@@ -142,8 +150,7 @@ func (m *Machine) requireComplete() error {
return nil
}
// myMsgFlags returns the flags for the current outgoing message, which are
// also the flags of the next incoming message before it is read.
// myMsgFlags returns the flags for the current outgoing message.
func (m *Machine) myMsgFlags() msgFlags {
idx := m.hs.MessageIndex()
if idx < len(m.msgs) {
@@ -231,18 +238,7 @@ func (m *Machine) ProcessPacket(out, packet []byte) ([]byte, *Result, error) {
// noise returns (cs1, cs2) where cs1 is the initiator->responder cipher.
// For 3-message patterns where a responder finishes by reading the final
// message, this ordering would be wrong; revisit when XX/pqIX lands.
noiseMsg := packet[header.Len:]
if m.myMsgFlags().kemCiphertext {
var err error
// Installing a psk derived from a forged ciphertext is harmless: ReadMessage
// then fails without advancing, and the next packet installs its own.
noiseMsg, err = m.kemDecapsulate(noiseMsg)
if err != nil {
return nil, nil, err
}
}
msg, eKey, dKey, err := m.hs.ReadMessage(nil, noiseMsg)
msg, eKey, dKey, err := m.hs.ReadMessage(nil, packet[header.Len:])
if err != nil {
// Noise ReadMessage failed. The noise library checkpoints and rolls back
// on failure, so the Machine is still alive. The caller can retry with
@@ -293,8 +289,6 @@ func (m *Machine) ProcessPacket(out, packet []byte) ([]byte, *Result, error) {
}
func (m *Machine) completed(eKey, dKey *noise.CipherState) *Result {
m.kemKey = nil
m.kemCiphertext = nil
m.result.EKey = eKey
m.result.DKey = dKey
m.result.MessageIndex = uint64(m.hs.MessageIndex())
@@ -303,7 +297,7 @@ func (m *Machine) completed(eKey, dKey *noise.CipherState) *Result {
func (m *Machine) processPayload(msg []byte, flags msgFlags) error {
if len(msg) == 0 {
if flags.expectsPayload || flags.expectsCert || flags.kemKey {
if flags.expectsPayload || flags.expectsCert {
m.failed = true
return ErrMissingContent
}
@@ -317,7 +311,8 @@ func (m *Machine) processPayload(msg []byte, flags msgFlags) error {
}
// Assert the payload contains exactly what we expect
hasPayloadData := payload.InitiatorIndex != 0 || payload.ResponderIndex != 0 || payload.Time != 0
hasPayloadData := payload.InitiatorIndex != 0 || payload.ResponderIndex != 0 || payload.Time != 0 ||
payload.InitiatorLanes != nil || payload.ResponderLanes != nil
if hasPayloadData != flags.expectsPayload {
m.failed = true
return ErrUnexpectedContent
@@ -329,12 +324,6 @@ func (m *Machine) processPayload(msg []byte, flags msgFlags) error {
return ErrUnexpectedContent
}
hasKEMKey := len(payload.KEMKey) > 0
if hasKEMKey != flags.kemKey {
m.failed = true
return ErrUnexpectedContent
}
// Process payload
if flags.expectsPayload {
var remoteIndex uint32
@@ -352,6 +341,23 @@ func (m *Machine) processPayload(msg []byte, flags msgFlags) error {
m.result.RemoteIndex = remoteIndex
m.result.HandshakeTime = payload.Time
m.payloadSet = true
// Multiport advert from the peer's side of the exchange. Out-of-range
// values mean a peer we can't pair lanes with; ignore the advert
// rather than failing the handshake — the tunnel itself is fine, it
// just won't get lanes. Semantic policing (port-count caps, lane
// clamping) belongs to the handshake manager.
var peerLanes *LaneDetails
if m.result.Initiator {
peerLanes = payload.ResponderLanes
} else {
peerLanes = payload.InitiatorLanes
}
if peerLanes != nil && peerLanes.BasePort <= 0xffff && peerLanes.PortCount <= 0xffff && peerLanes.TxLanes <= 0xffff {
m.result.PeerPortCount = peerLanes.PortCount
m.result.PeerBasePort = peerLanes.BasePort
m.result.PeerTxLanes = peerLanes.TxLanes
}
}
// Process certificate
@@ -361,13 +367,6 @@ func (m *Machine) processPayload(msg []byte, flags msgFlags) error {
}
}
if flags.kemKey {
if err := m.kemEncapsulate(payload.KEMKey); err != nil {
m.failed = true
return err
}
}
return nil
}
@@ -412,21 +411,11 @@ func (m *Machine) validateCert(payload Payload) error {
}
func (m *Machine) marshalOutgoing(flags msgFlags) ([]byte, error) {
if !flags.expectsPayload && !flags.expectsCert && !flags.kemKey {
if !flags.expectsPayload && !flags.expectsCert {
return nil, nil
}
var p Payload
if flags.kemKey {
if m.kemKey == nil {
dk, err := mlkem.GenerateKey1024()
if err != nil {
return nil, fmt.Errorf("generate ML-KEM key: %w", err)
}
m.kemKey = dk
}
p.KEMKey = m.kemKey.EncapsulationKey().Bytes()
}
if flags.expectsPayload {
if !m.indexAllocated {
index, err := m.allocIndex()
@@ -439,9 +428,11 @@ func (m *Machine) marshalOutgoing(flags msgFlags) ([]byte, error) {
if m.result.Initiator {
p.InitiatorIndex = m.result.LocalIndex
p.InitiatorLanes = m.lanes
} else {
p.ResponderIndex = m.result.LocalIndex
p.InitiatorIndex = m.result.RemoteIndex
p.ResponderLanes = m.lanes
}
p.Time = uint64(time.Now().UnixNano())
}
@@ -478,13 +469,6 @@ func (m *Machine) buildResponse(out []byte) ([]byte, *noise.CipherState, *noise.
uint64(m.hs.MessageIndex()+1),
)
if flags.kemCiphertext {
if m.kemCiphertext == nil {
return nil, nil, nil, errors.New("no ML-KEM ciphertext to send")
}
out = append(out, m.kemCiphertext...)
}
// noise.WriteMessage appends the encrypted handshake message to out,
// reusing capacity when present.
//
+111
View File
@@ -0,0 +1,111 @@
package handshake
import (
"net/netip"
"testing"
"time"
"github.com/slackhq/nebula/cert"
ct "github.com/slackhq/nebula/cert_test"
"github.com/slackhq/nebula/header"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// newTestLaneMachine is newTestMachine with a lane advert attached.
func newTestLaneMachine(
t *testing.T,
cs *testCertState,
verifier CertVerifier,
initiator bool,
localIndex uint32,
lanes *LaneDetails,
) *Machine {
t.Helper()
m, err := NewMachine(
cs.version, cs.getCredential,
verifier, func() (uint32, error) { return localIndex, nil },
initiator, header.HandshakeIXPSK0,
lanes,
)
require.NoError(t, err)
return m
}
func doFullLaneHandshake(t *testing.T, initLanes, respLanes *LaneDetails) (initR, respR *Result) {
t.Helper()
ca, _, caKey, _ := ct.NewTestCaCert(
cert.Version2, cert.Curve_CURVE25519, time.Time{}, time.Time{}, nil, nil, nil,
)
caPool := ct.NewTestCAPool(ca)
v := testVerifier(caPool)
initCS := newTestCertState(t, ca, caKey, "initiator", []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")})
respCS := newTestCertState(t, ca, caKey, "responder", []netip.Prefix{netip.MustParsePrefix("10.0.0.2/24")})
initM := newTestLaneMachine(t, initCS, v, true, 1000, initLanes)
respM := newTestLaneMachine(t, respCS, v, false, 2000, respLanes)
msg1, err := initM.Initiate(nil)
require.NoError(t, err)
resp, respR, err := respM.ProcessPacket(nil, msg1)
require.NoError(t, err)
require.NotNil(t, respR)
_, initR, err = initM.ProcessPacket(nil, resp)
require.NoError(t, err)
require.NotNil(t, initR)
return initR, respR
}
func TestMachineLaneAdvertBothSides(t *testing.T) {
initR, respR := doFullLaneHandshake(t,
&LaneDetails{PortCount: 8, BasePort: 4242, TxLanes: 8},
&LaneDetails{PortCount: 4, BasePort: 5353, TxLanes: 3},
)
// Each side's Result carries the peer's advert.
assert.Equal(t, uint32(4), initR.PeerPortCount)
assert.Equal(t, uint32(5353), initR.PeerBasePort)
assert.Equal(t, uint32(3), initR.PeerTxLanes)
assert.Equal(t, uint32(8), respR.PeerPortCount)
assert.Equal(t, uint32(4242), respR.PeerBasePort)
assert.Equal(t, uint32(8), respR.PeerTxLanes)
}
func TestMachineLaneAdvertAsymmetric(t *testing.T) {
// Vanilla initiator, multiport responder and vice versa: the nil side
// yields all-zero peer fields on the other end.
initR, respR := doFullLaneHandshake(t, nil, &LaneDetails{PortCount: 4, BasePort: 5353})
assert.Equal(t, uint32(4), initR.PeerPortCount)
assert.Equal(t, uint32(0), respR.PeerPortCount)
assert.Equal(t, uint32(0), respR.PeerBasePort)
initR, respR = doFullLaneHandshake(t, &LaneDetails{PortCount: 8, BasePort: 4242}, nil)
assert.Equal(t, uint32(0), initR.PeerPortCount)
assert.Equal(t, uint32(8), respR.PeerPortCount)
}
func TestMachineLaneAdvertOutOfRangeIgnored(t *testing.T) {
// A BasePort that can't be a real UDP port is ignored, not fatal.
initR, respR := doFullLaneHandshake(t,
&LaneDetails{PortCount: 8, BasePort: 70000, TxLanes: 8},
&LaneDetails{PortCount: 4, BasePort: 5353, TxLanes: 4},
)
assert.Equal(t, uint32(0), respR.PeerPortCount)
assert.Equal(t, uint32(0), respR.PeerBasePort)
assert.Equal(t, uint32(0), respR.PeerTxLanes)
// The sane side still negotiates.
assert.Equal(t, uint32(4), initR.PeerPortCount)
// An out-of-range TxLanes drops the whole advert the same way: a lane index
// that does not fit the header is as unusable as an impossible port.
initR, respR = doFullLaneHandshake(t,
&LaneDetails{PortCount: 8, BasePort: 4242, TxLanes: 0x10000},
&LaneDetails{PortCount: 4, BasePort: 5353, TxLanes: 4},
)
assert.Equal(t, uint32(0), respR.PeerPortCount)
assert.Equal(t, uint32(4), initR.PeerPortCount)
}
+2
View File
@@ -444,6 +444,7 @@ func TestMachineThreeMessagePattern(t *testing.T) {
initCS.getCredential, v,
func() (uint32, error) { return 1000, nil },
true, header.HandshakeXXPSK0,
nil,
)
require.NoError(t, err)
@@ -452,6 +453,7 @@ func TestMachineThreeMessagePattern(t *testing.T) {
respCS.getCredential, v,
func() (uint32, error) { return 2000, nil },
false, header.HandshakeXXPSK0,
nil,
)
require.NoError(t, err)
+2 -21
View File
@@ -11,16 +11,13 @@ import (
type msgFlags struct {
expectsPayload bool // message carries indexes and time
expectsCert bool // message carries the certificate
kemKey bool // payload carries an ML-KEM encapsulation key
kemCiphertext bool // message is prefixed with the ML-KEM ciphertext for the peer's key
}
// subtypeInfo bundles the noise pattern with the per-message flags for a
// given handshake subtype.
type subtypeInfo struct {
pattern noise.HandshakePattern
pskPlacement int // noise psk modifier position, 0 means no psk token
msgs []msgFlags
pattern noise.HandshakePattern
msgs []msgFlags
}
// subtypeInfos defines the noise pattern and message content layout for each
@@ -35,22 +32,6 @@ var subtypeInfos = map[header.MessageSubType]subtypeInfo{
},
},
// IX with an ephemeral ML-KEM-1024 exchange:
// msg1 (I->R): payload + cert + encapsulation key
// msg2 (R->I): ciphertext || noise(payload + cert)
// The responder encapsulates after reading msg1 and mixes the shared secret in
// through the psk2 token at the end of msg2. The ciphertext travels outside the
// noise message because the initiator needs the secret before it can process
// that token.
header.HandshakeIXPSK2MLKEM1024: {
pattern: noise.HandshakeIX,
pskPlacement: 2,
msgs: []msgFlags{
{expectsPayload: true, expectsCert: true, kemKey: true},
{expectsPayload: true, expectsCert: true, kemCiphertext: true},
},
},
// XX: 3 messages
// msg1 (I->R): payload only
// msg2 (R->I): payload + cert
+111 -12
View File
@@ -20,17 +20,40 @@ type Payload struct {
ResponderIndex uint32
Time uint64
CertVersion uint32
KEMKey []byte // ML-KEM encapsulation key, only in subtypes that carry one
// Multiport lane negotiation; nil when the sender has multiport disabled
// (which keeps the encoded payload byte-identical to a vanilla one).
InitiatorLanes *LaneDetails
ResponderLanes *LaneDetails
}
// LaneDetails advertises multiport lane capability: the contiguous UDP port
// range the sender bound, and how many lanes it may send on. The receiver needs
// PortCount/BasePort to aim its own lanes and TxLanes to know how many lane
// sessions to derive for receiving.
type LaneDetails struct {
PortCount uint32
BasePort uint32
TxLanes uint32
}
// Proto field numbers for NebulaHandshakeDetails
const (
fieldCert = 1 // bytes
fieldInitiatorIndex = 2 // uint32
fieldResponderIndex = 3 // uint32
fieldTime = 5 // uint64
fieldCertVersion = 8 // uint32
fieldKEMKey = 9 // bytes
fieldCert = 1 // bytes
fieldInitiatorIndex = 2 // uint32
fieldResponderIndex = 3 // uint32
fieldTime = 5 // uint64
fieldCertVersion = 8 // uint32
fieldInitiatorLanes = 9 // LaneDetails
fieldResponderLanes = 10 // LaneDetails
)
// Proto field numbers for LaneDetails.
// Field 3 was a per-lane handshake index and is permanently reserved.
const (
fieldLanePortCount = 1 // uint32
fieldLaneBasePort = 2 // uint32
fieldLaneTxLanes = 4 // uint32
)
// MarshalPayload encodes a handshake payload in protobuf wire format compatible
@@ -59,9 +82,15 @@ func MarshalPayload(out []byte, p Payload) []byte {
details = protowire.AppendTag(details, fieldCertVersion, protowire.VarintType)
details = protowire.AppendVarint(details, uint64(p.CertVersion))
}
if len(p.KEMKey) > 0 {
details = protowire.AppendTag(details, fieldKEMKey, protowire.BytesType)
details = protowire.AppendBytes(details, p.KEMKey)
// Emitted last to keep the encoding in ascending field-number order, which
// is what protoc-gen-go would produce for the same message.
if p.InitiatorLanes != nil {
details = protowire.AppendTag(details, fieldInitiatorLanes, protowire.BytesType)
details = protowire.AppendBytes(details, p.InitiatorLanes.marshal(nil))
}
if p.ResponderLanes != nil {
details = protowire.AppendTag(details, fieldResponderLanes, protowire.BytesType)
details = protowire.AppendBytes(details, p.ResponderLanes.marshal(nil))
}
out = protowire.AppendTag(out, 1, protowire.BytesType)
@@ -70,6 +99,20 @@ func MarshalPayload(out []byte, p Payload) []byte {
return out
}
// marshal appends the LaneDetails submessage fields to out. All fields are
// emitted unconditionally: a LaneDetails is only present at all when multiport
// is negotiating, and explicit zeros keep the parser's presence semantics
// trivial.
func (d *LaneDetails) marshal(out []byte) []byte {
out = protowire.AppendTag(out, fieldLanePortCount, protowire.VarintType)
out = protowire.AppendVarint(out, uint64(d.PortCount))
out = protowire.AppendTag(out, fieldLaneBasePort, protowire.VarintType)
out = protowire.AppendVarint(out, uint64(d.BasePort))
out = protowire.AppendTag(out, fieldLaneTxLanes, protowire.VarintType)
out = protowire.AppendVarint(out, uint64(d.TxLanes))
return out
}
// UnmarshalPayload decodes a protobuf-encoded NebulaHandshake message.
func UnmarshalPayload(b []byte) (Payload, error) {
var p Payload
@@ -167,7 +210,7 @@ func unmarshalPayloadDetails(p *Payload, b []byte) error {
}
p.CertVersion = uint32(v)
b = b[n:]
case fieldKEMKey:
case fieldInitiatorLanes:
if typ != protowire.BytesType {
return errInvalidHandshakeDetails
}
@@ -175,7 +218,63 @@ func unmarshalPayloadDetails(p *Payload, b []byte) error {
if n < 0 {
return errInvalidHandshakeDetails
}
p.KEMKey = append([]byte(nil), v...)
p.InitiatorLanes = new(LaneDetails)
if err := unmarshalLaneDetails(p.InitiatorLanes, v); err != nil {
return err
}
b = b[n:]
case fieldResponderLanes:
if typ != protowire.BytesType {
return errInvalidHandshakeDetails
}
v, n := protowire.ConsumeBytes(b)
if n < 0 {
return errInvalidHandshakeDetails
}
p.ResponderLanes = new(LaneDetails)
if err := unmarshalLaneDetails(p.ResponderLanes, v); err != nil {
return err
}
b = b[n:]
default:
n := protowire.ConsumeFieldValue(num, typ, b)
if n < 0 {
return errInvalidHandshakeDetails
}
b = b[n:]
}
}
return nil
}
func unmarshalLaneDetails(d *LaneDetails, b []byte) error {
for len(b) > 0 {
num, typ, n := protowire.ConsumeTag(b)
if n < 0 {
return errInvalidHandshakeDetails
}
b = b[n:]
// Same contract as the details parser: known fields hard-fail on a
// wire-type mismatch, unknown fields are skipped, repeated singular
// fields follow proto3 last-wins.
switch num {
case fieldLanePortCount, fieldLaneBasePort, fieldLaneTxLanes:
if typ != protowire.VarintType {
return errInvalidHandshakeDetails
}
v, n := protowire.ConsumeVarint(b)
if n < 0 || v > math.MaxUint32 {
return errInvalidHandshakeDetails
}
switch num {
case fieldLanePortCount:
d.PortCount = uint32(v)
case fieldLaneBasePort:
d.BasePort = uint32(v)
case fieldLaneTxLanes:
d.TxLanes = uint32(v)
}
b = b[n:]
default:
n := protowire.ConsumeFieldValue(num, typ, b)
+137 -11
View File
@@ -117,23 +117,134 @@ func TestPayloadUnknownFields(t *testing.T) {
assert.Equal(t, uint32(88), got.ResponderIndex)
})
t.Run("reserved fields 6 and 7 are skipped", func(t *testing.T) {
// Fields 6 and 7 are reserved in the proto definition
t.Run("unknown field inside LaneDetails is skipped", func(t *testing.T) {
var lane []byte
lane = protowire.AppendTag(lane, fieldLanePortCount, protowire.VarintType)
lane = protowire.AppendVarint(lane, 4)
lane = protowire.AppendTag(lane, 50, protowire.VarintType) // unknown subfield
lane = protowire.AppendVarint(lane, 9999)
lane = protowire.AppendTag(lane, fieldLaneBasePort, protowire.VarintType)
lane = protowire.AppendVarint(lane, 4242)
var details []byte
details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
details = protowire.AppendVarint(details, 100)
details = protowire.AppendTag(details, 6, protowire.VarintType)
details = protowire.AppendVarint(details, 1)
details = protowire.AppendTag(details, 7, protowire.VarintType)
details = protowire.AppendVarint(details, 2)
details = protowire.AppendTag(details, fieldInitiatorLanes, protowire.BytesType)
details = protowire.AppendBytes(details, lane)
var data []byte
data = protowire.AppendTag(data, 1, protowire.BytesType)
data = protowire.AppendBytes(data, details)
got, err := UnmarshalPayload(wrapDetails(details))
require.NoError(t, err)
assert.Equal(t, uint32(100), got.InitiatorIndex)
require.NotNil(t, got.InitiatorLanes)
assert.Equal(t, uint32(4), got.InitiatorLanes.PortCount)
assert.Equal(t, uint32(4242), got.InitiatorLanes.BasePort)
})
}
func TestPayloadLaneDetails(t *testing.T) {
t.Run("round trip both sides", func(t *testing.T) {
data := MarshalPayload(nil, Payload{
InitiatorIndex: 12345,
Time: 999,
InitiatorLanes: &LaneDetails{PortCount: 8, BasePort: 4242, TxLanes: 3},
ResponderLanes: &LaneDetails{PortCount: 4, BasePort: 5353},
})
got, err := UnmarshalPayload(data)
require.NoError(t, err)
assert.Equal(t, uint32(100), got.InitiatorIndex)
require.NotNil(t, got.InitiatorLanes)
assert.Equal(t, LaneDetails{PortCount: 8, BasePort: 4242, TxLanes: 3}, *got.InitiatorLanes)
require.NotNil(t, got.ResponderLanes)
assert.Equal(t, LaneDetails{PortCount: 4, BasePort: 5353}, *got.ResponderLanes)
})
t.Run("zero-valued LaneDetails survives the round trip", func(t *testing.T) {
// Presence is what negotiation keys on; an all-zero advert must not
// decay to nil.
data := MarshalPayload(nil, Payload{
InitiatorIndex: 1,
InitiatorLanes: &LaneDetails{},
})
got, err := UnmarshalPayload(data)
require.NoError(t, err)
require.NotNil(t, got.InitiatorLanes)
assert.Equal(t, LaneDetails{}, *got.InitiatorLanes)
assert.Nil(t, got.ResponderLanes)
})
t.Run("nil lanes marshal byte-identical to a vanilla payload", func(t *testing.T) {
p := Payload{
Cert: []byte("cert"),
CertVersion: 2,
InitiatorIndex: 100,
Time: 999,
}
// The vanilla encoding of the same fields, built by hand in field order.
var details []byte
details = protowire.AppendTag(details, fieldCert, protowire.BytesType)
details = protowire.AppendBytes(details, p.Cert)
details = protowire.AppendTag(details, fieldInitiatorIndex, protowire.VarintType)
details = protowire.AppendVarint(details, uint64(p.InitiatorIndex))
details = protowire.AppendTag(details, fieldTime, protowire.VarintType)
details = protowire.AppendVarint(details, p.Time)
details = protowire.AppendTag(details, fieldCertVersion, protowire.VarintType)
details = protowire.AppendVarint(details, uint64(p.CertVersion))
assert.Equal(t, wrapDetails(details), MarshalPayload(nil, p))
})
t.Run("lane field with wrong wire type rejected", func(t *testing.T) {
for _, field := range []protowire.Number{fieldInitiatorLanes, fieldResponderLanes} {
var details []byte
details = protowire.AppendTag(details, field, protowire.VarintType)
details = protowire.AppendVarint(details, 1)
_, err := UnmarshalPayload(wrapDetails(details))
assert.Error(t, err)
}
})
t.Run("lane subfield with wrong wire type rejected", func(t *testing.T) {
var lane []byte
lane = protowire.AppendTag(lane, fieldLanePortCount, protowire.BytesType)
lane = protowire.AppendBytes(lane, []byte{1, 2, 3})
var details []byte
details = protowire.AppendTag(details, fieldInitiatorLanes, protowire.BytesType)
details = protowire.AppendBytes(details, lane)
_, err := UnmarshalPayload(wrapDetails(details))
assert.Error(t, err)
})
t.Run("truncated LaneDetails submessage rejected", func(t *testing.T) {
var details []byte
details = protowire.AppendTag(details, fieldInitiatorLanes, protowire.BytesType)
details = append(details, 0x0a, 0x01, 0x02) // length 10, only 2 bytes
_, err := UnmarshalPayload(wrapDetails(details))
assert.Error(t, err)
})
t.Run("truncated varint inside LaneDetails rejected", func(t *testing.T) {
var lane []byte
lane = protowire.AppendTag(lane, fieldLaneBasePort, protowire.VarintType)
lane = append(lane, 0x80) // incomplete varint
var details []byte
details = protowire.AppendTag(details, fieldResponderLanes, protowire.BytesType)
details = protowire.AppendBytes(details, lane)
_, err := UnmarshalPayload(wrapDetails(details))
assert.Error(t, err)
})
t.Run("lane subfield varint overflow rejected", func(t *testing.T) {
var lane []byte
lane = protowire.AppendTag(lane, fieldLaneTxLanes, protowire.VarintType)
lane = protowire.AppendVarint(lane, math.MaxUint32+1)
var details []byte
details = protowire.AppendTag(details, fieldInitiatorLanes, protowire.BytesType)
details = protowire.AppendBytes(details, lane)
_, err := UnmarshalPayload(wrapDetails(details))
assert.Error(t, err)
})
}
@@ -328,6 +439,12 @@ func FuzzPayload(f *testing.F) {
Time: 3,
CertVersion: 2,
}))
f.Add(MarshalPayload(nil, Payload{
InitiatorIndex: 1,
Time: 3,
InitiatorLanes: &LaneDetails{PortCount: 8, BasePort: 4242, TxLanes: 2},
ResponderLanes: &LaneDetails{PortCount: 4, BasePort: 5353},
}))
f.Add([]byte{})
f.Add([]byte{0xff})
@@ -357,5 +474,14 @@ func payloadsEqual(a, b Payload) bool {
a.InitiatorIndex == b.InitiatorIndex &&
a.ResponderIndex == b.ResponderIndex &&
a.Time == b.Time &&
a.CertVersion == b.CertVersion
a.CertVersion == b.CertVersion &&
laneDetailsEqual(a.InitiatorLanes, b.InitiatorLanes) &&
laneDetailsEqual(a.ResponderLanes, b.ResponderLanes)
}
func laneDetailsEqual(a, b *LaneDetails) bool {
if a == nil || b == nil {
return a == b
}
return *a == *b
}
+118 -164
View File
@@ -10,15 +10,12 @@ import (
"net/netip"
"slices"
"sync"
"sync/atomic"
"time"
"github.com/rcrowley/go-metrics"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/handshake"
"github.com/slackhq/nebula/handshake/fragment"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/ratelimit"
"github.com/slackhq/nebula/udp"
)
@@ -52,14 +49,20 @@ type HandshakeConfig struct {
tryInterval time.Duration
retries int64
triggerBuffer int
pq pqHandshakeConfig
// Multiport lane parameters; laneCount == 0 means multiport is disabled.
// laneCount includes implicit lane 0 (the base tunnel), so lanes
// 1..laneCount-1 may carry traffic. lanePortCount/laneBasePort describe our
// own bound port range and are advertised in every handshake payload.
laneCount int
lanePortCount uint16
laneBasePort uint16
messageMetrics *MessageMetrics
}
type HandshakeManager struct {
// Mutex for interacting with the vpnIps and indexes maps.
// Lock order is hostmap, then this lock, then the lighthouse, then a RemoteList.
// Mutex for interacting with the vpnIps and indexes maps
sync.RWMutex
vpnIps map[netip.Addr]*HandshakeHostInfo
@@ -78,18 +81,6 @@ type HandshakeManager struct {
// can be used to trigger outbound handshake for the given vpnIp
trigger chan netip.Addr
// pq holds the post-quantum settings, which change on reload.
pq atomic.Pointer[pqHandshakeConfig]
// pqTable reassembles post-quantum msg1s from senders that have not authenticated yet.
pqTable *fragment.Table[pqFragKey, pqFragSource]
// pqPeers holds vpn addresses that completed a post-quantum handshake, for sticky mode.
pqPeers sync.Map
// pqAnswered recognises repeats of post-quantum msg1s already answered.
pqAnswered *pqAnswered
// pqCookies and pqCookieLimiter serve cookies while pqTable is under load.
pqCookies *pqCookies
pqCookieLimiter *ratelimit.Limiter
}
type HandshakeHostInfo struct {
@@ -105,8 +96,6 @@ type HandshakeHostInfo struct {
hostinfo *HostInfo
machine *handshake.Machine // The handshake state machine, set during stage 0 (initiator) or beginHandshake (responder multi-message)
pq *pqAttempt // The initiator's post-quantum attempt, pending alongside the classical one
lastPQ bool // Whether the latest attempt was post-quantum
}
func (hh *HandshakeHostInfo) cachePacket(l *slog.Logger, t header.MessageType, st header.MessageSubType, packet []byte, f packetCallback, m *cachedPacketMetrics) {
@@ -135,7 +124,7 @@ func (hh *HandshakeHostInfo) cachePacket(l *slog.Logger, t header.MessageType, s
}
func NewHandshakeManager(l *slog.Logger, mainHostMap *HostMap, lightHouse *LightHouse, outside udp.Conn, config HandshakeConfig) *HandshakeManager {
hm := &HandshakeManager{
return &HandshakeManager{
vpnIps: map[netip.Addr]*HandshakeHostInfo{},
indexes: map[uint32]*HandshakeHostInfo{},
mainHostMap: mainHostMap,
@@ -148,16 +137,7 @@ func NewHandshakeManager(l *slog.Logger, mainHostMap *HostMap, lightHouse *Light
metricInitiated: metrics.GetOrRegisterCounter("handshake_manager.initiated", nil),
metricTimedOut: metrics.GetOrRegisterCounter("handshake_manager.timed_out", nil),
l: l,
pqTable: fragment.NewTable[pqFragKey, pqFragSource](pqTableConfig),
pqAnswered: newPQAnswered(hsTimeout(config.retries, config.tryInterval)),
}
pq := config.pq
hm.pq.Store(&pq)
hm.pqCookies = newPQCookies(time.Now())
// Cookie replies are tiny, but a flood of spoofed fragments must not turn
// into a flood of replies.
hm.pqCookieLimiter, _ = ratelimit.New(256, time.Millisecond)
return hm
}
func (hm *HandshakeManager) Run(ctx context.Context) {
@@ -172,8 +152,6 @@ func (hm *HandshakeManager) Run(ctx context.Context) {
hm.handleOutbound(vpnIP, true)
case now := <-clockSource.C:
hm.NextOutboundHandshakeTimerTick(now)
hm.pqTable.Expire(now)
hm.pqAnswered.expire(now)
}
}
}
@@ -183,7 +161,7 @@ func (hm *HandshakeManager) HandleIncoming(via ViaSender, packet []byte, h *head
// don't yet support) are dropped here rather than silently routed through
// the IX path. Add a case when introducing a new pattern.
switch h.Subtype {
case header.HandshakeIXPSK0, pqSubtype:
case header.HandshakeIXPSK0:
// supported
default:
hm.l.Debug("dropping handshake with unsupported subtype",
@@ -199,11 +177,6 @@ func (hm *HandshakeManager) HandleIncoming(via ViaSender, packet []byte, h *head
}
}
if header.IsFragmented(h.Subtype) {
hm.handleFragment(via, packet, h)
return
}
// First message of a new handshake. The wire format requires RemoteIndex
// to be zero here (the initiator has no responder index to fill in yet),
// and generateIndex never allocates 0, so any non-zero RemoteIndex on a
@@ -215,11 +188,7 @@ func (hm *HandshakeManager) HandleIncoming(via ViaSender, packet []byte, h *head
"from", via, "remoteIndex", h.RemoteIndex)
return
}
if hm.pqConfig().require {
hm.l.Debug("dropping classical handshake, post-quantum handshakes are required", "from", via)
return
}
hm.beginHandshake(via, packet, h, 0)
hm.beginHandshake(via, packet, h)
return
}
@@ -268,9 +237,6 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
"style": header.SubTypeName(header.Handshake, hh.machine.Subtype()),
})
}
if hh.pq != nil {
fields = append(fields, "pqInitiatorIndex", hh.pq.index)
}
hh.hostinfo.logger(hm.l).Info("Handshake timed out", fields...)
hm.metricTimedOut.Inc(1)
hm.DeleteHostInfo(hostinfo)
@@ -280,45 +246,33 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
// Increment the counter to increase our delay, linear backoff
hh.counter++
// Pick this attempt's style and make sure its first message exists. A PQ
// attempt sends msg1 as fragments: now goes out immediately, later after the
// repair delay, and every attempt offers fresh repair symbols.
var now, later [][]byte
var machine *handshake.Machine
hh.lastPQ = hm.usePQ(hh, vpnIp, lighthouseTriggered)
if hh.lastPQ {
if hh.pq == nil && !hm.buildPQAttempt(hh) {
// Check if we have a handshake packet to transmit yet
if !hh.ready {
if !hm.buildStage0Packet(hh) {
hm.OutboundHandshakeTimer.Add(vpnIp, hm.config.tryInterval*time.Duration(hh.counter))
return
}
var err error
now, later, err = hh.pq.sender.Attempt(fragment.DefaultRepair(hh.pq.sender.Layout().K))
if err != nil {
hostinfo.logger(hm.l).Error("Failed to fragment handshake", "error", err)
hm.OutboundHandshakeTimer.Add(vpnIp, hm.config.tryInterval*time.Duration(hh.counter))
return
}
machine = hh.pq.machine
} else {
if !hh.ready {
if !hm.buildStage0Packet(hh) {
hm.OutboundHandshakeTimer.Add(vpnIp, hm.config.tryInterval*time.Duration(hh.counter))
return
}
}
// TODO: this hardcodes "always retransmit stage 0", which is correct for
// IX (the initiator only ever sends one packet, msg1) but wrong the
// moment a 3+ message pattern lands. The retry loop should resend the
// most recent outgoing message, not always stage 0. That implies
// HandshakeHostInfo tracking a single "currentOutbound" packet (bytes +
// header metadata) that gets replaced as the handshake progresses,
// instead of indexing into HandshakePacket.
now = [][]byte{hostinfo.HandshakePacket[handshakePacketStage0]}
machine = hh.machine
}
// TODO: this hardcodes "always retransmit stage 0", which is correct for
// IX (the initiator only ever sends one packet, msg1) but wrong the
// moment a 3+ message pattern lands. The retry loop should resend the
// most recent outgoing message, not always stage 0. That implies
// HandshakeHostInfo tracking a single "currentOutbound" packet (bytes +
// header metadata) that gets replaced as the handshake progresses,
// instead of indexing into HandshakePacket.
stage0 := hostinfo.HandshakePacket[handshakePacketStage0]
hsFields := m{
"stage": uint64(machine.MessageIndex()),
"style": header.SubTypeName(header.Handshake, machine.Subtype()),
"stage": uint64(hh.machine.MessageIndex()),
"style": header.SubTypeName(header.Handshake, hh.machine.Subtype()),
}
// Get a remotes object if we don't already have one.
// This is mainly to protect us as this should never be the case
// NB ^ This comment doesn't jive. It's how the thing gets initialized.
// It's the common path. Should it update every time, in case a future LH query/queries give us more info?
if hostinfo.remotes == nil {
hostinfo.remotes = hm.lightHouse.QueryCache([]netip.Addr{vpnIp})
}
remotes := hostinfo.remotes.CopyAddrs(hm.mainHostMap.GetPreferredRanges())
@@ -346,10 +300,8 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
// Send the handshake to all known ips, stage 2 takes care of assigning the hostinfo.remote based on the first to reply
var sentTo []netip.AddrPort
hostinfo.remotes.ForEach(hm.mainHostMap.GetPreferredRanges(), func(addr netip.AddrPort, _ bool) {
hm.messageMetrics.Tx(header.Handshake, machine.Subtype(), int64(len(now)+len(later)))
err := hm.sendFragmented(now, later, func(p []byte) error {
return hm.outside.WriteTo(p, addr)
})
hm.messageMetrics.Tx(header.Handshake, hh.machine.Subtype(), 1)
err := hm.outside.WriteTo(stage0, addr)
if err != nil {
// These repeat every attempt, so match the success log below and only shout when the remotes changed
level := slog.LevelDebug
@@ -385,7 +337,7 @@ func (hm *HandshakeManager) handleOutbound(vpnIp netip.Addr, lighthouseTriggered
)
}
hm.f.relayManager.StartRelays(hm.f, vpnIp, hh, now, later)
hm.f.relayManager.StartRelays(hm.f, vpnIp, hh, stage0)
// If a lighthouse triggered this attempt then we are still in the timer wheel and do not need to re-add
if !lighthouseTriggered {
@@ -427,9 +379,6 @@ func (hm *HandshakeManager) StartHandshake(vpnAddr netip.Addr, cacheCb func(*Han
hostinfo := &HostInfo{
vpnAddrs: []netip.Addr{vpnAddr},
HandshakePacket: make(map[uint8][]byte, 0),
// Attached here so a pending hostinfo never exists without one: SetRemote dereferences
// it, and create-tunnel -address can reach SetRemote before the handshake loop runs.
remotes: hm.lightHouse.QueryCache([]netip.Addr{vpnAddr}),
relayState: RelayState{
relays: nil,
relayForByAddr: map[netip.Addr]*Relay{},
@@ -594,11 +543,10 @@ func (hm *HandshakeManager) DeleteHostInfo(hostinfo *HostInfo) {
func (hm *HandshakeManager) unlockedDeleteHostInfo(hostinfo *HostInfo) {
for _, addr := range hostinfo.vpnAddrs {
// Only delete the pending entry if it is actually ours: an
// unconditional delete could evict a concurrently pending handshake for
// the same address.
if cur, ok := hm.vpnIps[addr]; ok && cur.hostinfo == hostinfo {
// A PQ attempt holds its own index alongside the hostinfo's.
if idx := cur.pqIndex(); idx != 0 && hm.indexes[idx] == cur {
delete(hm.indexes, idx)
}
delete(hm.vpnIps, addr)
}
}
@@ -709,8 +657,9 @@ func hsTimeout(tries int64, interval time.Duration) time.Duration {
return time.Duration(tries / 2 * ((2 * int64(interval)) + (tries-1)*int64(interval)))
}
// initiatingVersion picks the certificate version to initiate a handshake with.
func (hm *HandshakeManager) initiatingVersion(hh *HandshakeHostInfo, cs *CertState) cert.Version {
// buildStage0Packet creates the initial handshake packet for the initiator.
func (hm *HandshakeManager) buildStage0Packet(hh *HandshakeHostInfo) bool {
cs := hm.f.pki.getCertState()
v := cs.DefaultVersion()
if hh.initiatingVersionOverride != cert.VersionPre1 {
v = hh.initiatingVersionOverride
@@ -722,13 +671,6 @@ func (hm *HandshakeManager) initiatingVersion(hh *HandshakeHostInfo, cs *CertSta
}
}
}
return v
}
// buildStage0Packet creates the initial handshake packet for the initiator.
func (hm *HandshakeManager) buildStage0Packet(hh *HandshakeHostInfo) bool {
cs := hm.f.pki.getCertState()
v := hm.initiatingVersion(hh, cs)
cred := cs.GetCredential(v)
if cred == nil {
@@ -741,6 +683,7 @@ func (hm *HandshakeManager) buildStage0Packet(hh *HandshakeHostInfo) bool {
v, cs.GetCredential,
hm.certVerifier(), func() (uint32, error) { return hm.allocateIndex(hh) },
true, header.HandshakeIXPSK0,
hm.laneAdvert(),
)
if err != nil {
hm.f.l.Error("Failed to create handshake machine",
@@ -764,13 +707,39 @@ func (hm *HandshakeManager) buildStage0Packet(hh *HandshakeHostInfo) bool {
return true
}
// laneAdvert returns our multiport advert for a handshake payload, or nil
// when multiport is disabled (which keeps the payload byte-identical to
// vanilla).
func (hm *HandshakeManager) laneAdvert() *handshake.LaneDetails {
if hm.config.laneCount == 0 {
return nil
}
return &handshake.LaneDetails{
PortCount: uint32(hm.config.lanePortCount),
BasePort: uint32(hm.config.laneBasePort),
TxLanes: uint32(hm.config.laneCount),
}
}
// maybeAllocLanes sets up the multiport lanes for a just-completed tunnel. Must
// run before the hostinfo becomes visible in the hostmap: the data plane reads
// hostinfo.lanes without synchronizing on it. The sessions themselves are derived
// later, on the first packet that needs each one.
func (hm *HandshakeManager) maybeAllocLanes(hostinfo *HostInfo, result *handshake.Result) {
if hm.config.laneCount == 0 || result.PeerPortCount == 0 {
return
}
if len(hm.f.myVpnAddrs) == 0 || len(hostinfo.vpnAddrs) == 0 {
return
}
hostinfo.lanes = newLaneSet(result, hm.config.laneCount, hostinfo.ConnectionState.epoch, hm.f.myVpnAddrs[0], hostinfo.vpnAddrs[0])
}
// beginHandshake handles an incoming handshake packet that doesn't match any
// existing pending handshake. It creates a new responder Machine and processes
// the first message.
//
// received is the number of bytes a fragmented msg1 arrived in, which bounds the
// response sent back before the address is validated; 0 for unfragmented ones.
func (hm *HandshakeManager) beginHandshake(via ViaSender, packet []byte, h *header.H, received int) {
func (hm *HandshakeManager) beginHandshake(via ViaSender, packet []byte, h *header.H) {
f := hm.f
cs := f.pki.getCertState()
@@ -784,7 +753,8 @@ func (hm *HandshakeManager) beginHandshake(via ViaSender, packet []byte, h *head
machine, err := handshake.NewMachine(
v, cs.GetCredential,
hm.certVerifier(), func() (uint32, error) { return generateIndex(f.l) },
false, h.Subtype,
false, header.HandshakeIXPSK0,
hm.laneAdvert(),
)
if err != nil {
f.l.Error("Failed to create handshake machine", "from", via, "error", err)
@@ -841,6 +811,10 @@ func (hm *HandshakeManager) beginHandshake(via ViaSender, packet []byte, h *head
},
}
// Lanes are allocated before the log line so it can report what was actually
// negotiated, and must in any case be in place before CheckAndComplete below.
hm.maybeAllocLanes(hostinfo, result)
msg := "Handshake message received"
if !anyVpnAddrsInCommon {
msg = "Handshake message received, but no vpnNetworks in common."
@@ -855,6 +829,7 @@ func (hm *HandshakeManager) beginHandshake(via ViaSender, packet []byte, h *head
"initiatorIndex", result.RemoteIndex,
"responderIndex", result.LocalIndex,
"handshake", m{"stage": uint64(machine.MessageIndex()), "style": header.SubTypeName(header.Handshake, machine.Subtype())},
laneLogAttr(hm.config.laneCount, hostinfo.lanes),
)
// packet aliases the listener's incoming buffer, so this copy must stay.
@@ -874,15 +849,11 @@ func (hm *HandshakeManager) beginHandshake(via ViaSender, packet []byte, h *head
existing, err := hm.CheckAndComplete(hostinfo, handshakePacketStage0, f)
if err != nil {
hm.handleCheckAndCompleteError(err, existing, hostinfo, via, machine.Subtype(), received)
hm.handleCheckAndCompleteError(err, existing, hostinfo, via)
return
}
hm.sendHandshakeResponse(via, response, hostinfo, false, received)
if header.IsFragmented(machine.Subtype()) {
hm.pqAnswered.add(packet[header.Len:], hostinfo, time.Now())
}
hm.notePQPeer(machine.Subtype(), vpnAddrs)
hm.sendHandshakeResponse(via, response, hostinfo, false)
hostinfo.remotes.RefreshFromHandshake(vpnAddrs)
// Don't wait for UpdateWorker
@@ -893,32 +864,23 @@ func (hm *HandshakeManager) beginHandshake(via ViaSender, packet []byte, h *head
// continueHandshake feeds an incoming packet to an existing pending handshake Machine.
func (hm *HandshakeManager) continueHandshake(via ViaSender, hh *HandshakeHostInfo, packet []byte) {
f := hm.f
hh.Lock()
defer hh.Unlock()
if !hm.stillPending(hh, hh.hostinfo.localIndexId) {
// Re-verify hh is still tracked. Between queryIndex returning and us taking
// hh.Lock, handleOutbound may have timed out and deleted it. Once we hold
// hh.Lock no other deleter can race our index: handleOutbound also takes
// hh.Lock first, and handleRecvError targets a main-hostmap entry with a
// different localIndexId.
hm.RLock()
cur, ok := hm.indexes[hh.hostinfo.localIndexId]
hm.RUnlock()
if !ok || cur != hh {
return
}
hm.processResponse(via, hh, hh.machine, packet)
}
// stillPending re-verifies hh is still tracked under index. Between queryIndex
// returning and the caller taking hh.Lock, handleOutbound may have timed out and
// deleted it. Once hh.Lock is held no other deleter can race the index:
// handleOutbound also takes hh.Lock first, and handleRecvError targets a
// main-hostmap entry with a different localIndexId. The caller must hold hh.Lock.
func (hm *HandshakeManager) stillPending(hh *HandshakeHostInfo, index uint32) bool {
hm.RLock()
defer hm.RUnlock()
cur, ok := hm.indexes[index]
return ok && cur == hh
}
// processResponse runs a responder's message through one of the initiator's
// pending attempts and completes the tunnel when the handshake finishes. The
// caller must hold hh.Lock.
func (hm *HandshakeManager) processResponse(via ViaSender, hh *HandshakeHostInfo, machine *handshake.Machine, packet []byte) {
f := hm.f
hostinfo := hh.hostinfo
if !via.IsRelayed {
if !f.lightHouse.GetRemoteAllowList().AllowAll(hostinfo.vpnAddrs, via.UdpAddr.Addr()) {
@@ -928,6 +890,7 @@ func (hm *HandshakeManager) processResponse(via ViaSender, hh *HandshakeHostInfo
}
}
machine := hh.machine
if machine == nil {
f.l.Error("No handshake machine available for continuation",
"vpnAddrs", hostinfo.vpnAddrs, "from", via)
@@ -950,7 +913,7 @@ func (hm *HandshakeManager) processResponse(via ViaSender, hh *HandshakeHostInfo
}
if response != nil {
hm.sendHandshakeResponse(via, response, hostinfo, false, 0)
hm.sendHandshakeResponse(via, response, hostinfo, false)
}
if result == nil {
@@ -1030,18 +993,25 @@ func (hm *HandshakeManager) processResponse(via ViaSender, hh *HandshakeHostInfo
hm.DeleteHostInfo(hostinfo)
hm.StartHandshake(hostinfo.vpnAddrs[0], func(newHH *HandshakeHostInfo) {
newHH.hostinfo.remotes = hostinfo.remotes
newHH.hostinfo.remotes.BlockRemote(via)
newHH.packetStore = hh.packetStore
hh.packetStore = []*cachedPacket{}
hostinfo.vpnAddrs = vpnAddrs
f.sendCloseTunnel(hostinfo)
})
// sendCloseTunnel may take the hostmap lock, so it must not run under the handshake
// manager lock that StartHandshake holds around the callback.
hostinfo.vpnAddrs = vpnAddrs
f.sendCloseTunnel(hostinfo)
return
}
duration := time.Since(hh.startTime).Nanoseconds()
hostinfo.vpnAddrs = vpnAddrs
hostinfo.buildNetworks(f.myVpnNetworksTable, remoteCert.Certificate)
// Lanes are allocated before the log line so it can report what was actually
// negotiated, and must in any case be in place before Complete below.
hm.maybeAllocLanes(hostinfo, result)
msg := "Handshake message received"
if !anyVpnAddrsInCommon {
msg = "Handshake message received, but no vpnNetworks in common."
@@ -1058,15 +1028,10 @@ func (hm *HandshakeManager) processResponse(via ViaSender, hh *HandshakeHostInfo
"handshake", m{"stage": uint64(machine.MessageIndex()), "style": header.SubTypeName(header.Handshake, machine.Subtype())},
"durationNs", duration,
"sentCachedPackets", len(hh.packetStore),
laneLogAttr(hm.config.laneCount, hostinfo.lanes),
)
hostinfo.vpnAddrs = vpnAddrs
hostinfo.buildNetworks(f.myVpnNetworksTable, remoteCert.Certificate)
// The classical and PQ attempts each hold an index; keep the one that won.
hm.settleIndex(hh, result.LocalIndex)
hm.Complete(hostinfo, f)
hm.notePQPeer(machine.Subtype(), vpnAddrs)
if len(hh.packetStore) > 0 {
if f.l.Enabled(context.Background(), slog.LevelDebug) {
@@ -1145,23 +1110,20 @@ func (hm *HandshakeManager) validatePeerCert(via ViaSender, remoteCert *cert.Cac
// cached is true when msg is a stored response being retransmitted because
// the peer's stage-1 retransmit landed (the ErrAlreadySeen path); false on a
// fresh response.
//
// received is the number of bytes a fragmented msg1 arrived in; it bounds what a
// fragmented response may send back. It is ignored for unfragmented subtypes.
func (hm *HandshakeManager) sendHandshakeResponse(via ViaSender, msg []byte, hostinfo *HostInfo, cached bool, received int) {
func (hm *HandshakeManager) sendHandshakeResponse(via ViaSender, msg []byte, hostinfo *HostInfo, cached bool) {
if msg == nil {
return
}
f := hm.f
subtype := header.MessageSubType(msg[1])
f.messageMetrics.Tx(header.Handshake, header.MessageSubType(msg[1]), 1)
// Common log fields. peerCert may be nil during intermediate
// multi-message flows (handshake hasn't completed yet); skip the cert
// block if so.
logFields := []any{
"vpnAddrs", hostinfo.vpnAddrs,
"handshake", m{"stage": uint64(2), "style": header.SubTypeName(header.Handshake, subtype)},
"handshake", m{"stage": uint64(2), "style": header.SubTypeName(header.Handshake, header.HandshakeIXPSK0)},
"cached", cached,
"initiatorIndex", hostinfo.remoteIndexId,
"responderIndex", hostinfo.localIndexId,
@@ -1175,20 +1137,12 @@ func (hm *HandshakeManager) sendHandshakeResponse(via ViaSender, msg []byte, hos
)
}
if header.IsFragmented(subtype) {
fields := append(logFields, "from", via)
if err := hm.sendFragmentedResponse(via, msg, hostinfo, cached, received); err != nil {
f.l.Debug("Did not send handshake message", append(fields, "error", err)...)
} else {
f.l.Info("Handshake message sent", fields...)
}
return
}
f.messageMetrics.Tx(header.Handshake, subtype, 1)
if !via.IsRelayed {
fields := append(logFields, "from", via)
err := f.outside.WriteTo(msg, via.UdpAddr)
// Reply from the socket the handshake arrived on so the initiator sees
// the source port it targeted. Identical to f.outside under vanilla
// config (all writers share one port); required for multiport lanes.
err := f.writers[via.SockIdx].WriteTo(msg, via.UdpAddr)
if err != nil {
f.l.Error("Failed to send handshake message", append(fields, "error", err)...)
} else {
@@ -1212,10 +1166,10 @@ func (hm *HandshakeManager) sendHandshakeResponse(via ViaSender, msg []byte, hos
// This only fires from the responder-side beginHandshake path, after the
// peer cert has been validated and ConnectionState populated, so peerCert
// is always non-nil for the cases that log it.
func (hm *HandshakeManager) handleCheckAndCompleteError(err error, existing, hostinfo *HostInfo, via ViaSender, subtype header.MessageSubType, received int) {
func (hm *HandshakeManager) handleCheckAndCompleteError(err error, existing, hostinfo *HostInfo, via ViaSender) {
f := hm.f
peerCert := hostinfo.ConnectionState.peerCert
hsFields := m{"stage": uint64(1), "style": header.SubTypeName(header.Handshake, subtype)}
hsFields := m{"stage": uint64(1), "style": header.SubTypeName(header.Handshake, header.HandshakeIXPSK0)}
switch err {
case ErrAlreadySeen:
@@ -1226,7 +1180,7 @@ func (hm *HandshakeManager) handleCheckAndCompleteError(err error, existing, hos
// ephemeral keys; a freshly-built one would have different keys and break
// the tunnel even though both sides "completed" the handshake.
if msg := existing.HandshakePacket[handshakePacketStage2]; msg != nil {
hm.sendHandshakeResponse(via, msg, existing, true, received)
hm.sendHandshakeResponse(via, msg, existing, true)
}
case ErrExistingHostInfo:
-699
View File
@@ -1,699 +0,0 @@
package nebula
import (
"context"
"crypto/hmac"
"crypto/rand"
"crypto/sha256"
"encoding/binary"
"errors"
"log/slog"
"net/netip"
"sync"
"time"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/handshake"
"github.com/slackhq/nebula/handshake/fragment"
"github.com/slackhq/nebula/header"
)
const (
// pqSubtype is the post-quantum handshake this node speaks.
pqSubtype = header.HandshakeIXPSK2MLKEM1024
// DefaultPQRepairDelay is how long repair fragments trail the data fragments.
DefaultPQRepairDelay = 10 * time.Millisecond
// pqAmplification bounds what a responder sends in reply to an unvalidated
// address, as a multiple of what it received from it.
pqAmplification = 3
)
// pqTableConfig bounds the responder's reassembly of msg1 fragments from senders
// that have not authenticated yet.
var pqTableConfig = fragment.TableConfig{
MaxBytes: 4 << 20,
MaxEntries: 1024,
MaxPerSource: 4,
MaxMsgLen: 16384,
TTL: 3 * time.Second,
}
type pqHandshakeConfig struct {
initiate bool // start PQ handshakes
accept bool // answer PQ handshakes
require bool // only PQ handshakes, in either direction
sticky bool // never fall back to classical for a peer that completed PQ
repairDelay time.Duration // delay between data and repair fragments
}
func pqHandshakeConfigFrom(c *config.C) pqHandshakeConfig {
cfg := pqHandshakeConfig{
initiate: c.GetBool("handshakes.pq.initiate", false),
accept: c.GetBool("handshakes.pq.accept", false),
require: c.GetBool("handshakes.pq.require", false),
sticky: c.GetBool("handshakes.pq.sticky", false),
repairDelay: c.GetDuration("handshakes.pq.repair_delay", DefaultPQRepairDelay),
}
if cfg.require {
cfg.initiate = true
cfg.accept = true
}
return cfg
}
// pqConfig returns the post-quantum settings in effect.
func (hm *HandshakeManager) pqConfig() *pqHandshakeConfig {
return hm.pq.Load()
}
// reloadPQ applies changed post-quantum settings. Handshakes already in flight
// keep going; the new settings apply to the next attempt and packet.
func (hm *HandshakeManager) reloadPQ(c *config.C) {
if !c.HasChanged("handshakes.pq") {
return
}
old := hm.pqConfig()
cfg := pqHandshakeConfigFrom(c)
hm.pq.Store(&cfg)
hm.l.Info("Post-quantum handshake settings changed",
"initiate", cfg.initiate, "oldInitiate", old.initiate,
"accept", cfg.accept, "oldAccept", old.accept,
"require", cfg.require, "oldRequire", old.require,
"sticky", cfg.sticky, "oldSticky", old.sticky,
"repairDelay", cfg.repairDelay, "oldRepairDelay", old.repairDelay,
)
}
// pqAttempt is the initiator's post-quantum attempt for a pending handshake. It
// lives alongside the classical attempt in HandshakeHostInfo and has its own
// local index, so whichever answer arrives first wins.
type pqAttempt struct {
machine *handshake.Machine
index uint32
sender *fragment.Sender
msg2 *fragment.Assembly
cookieResendAt time.Time // last resend prompted by a cookie reply
heard bool // a msg2 fragment or cookie reply arrived, so the peer speaks PQ
}
// pqFragKey identifies one msg1 being reassembled.
type pqFragKey struct {
addr netip.AddrPort
relay uint32
fragID uint32
}
// pqFragSource groups msg1 reassemblies for the per-source quota: an IPv4 /32,
// an IPv6 /64, or one relay channel.
type pqFragSource struct {
prefix netip.Prefix
relay uint32
}
func pqFragKeyFor(via ViaSender, fragID uint32) (pqFragKey, pqFragSource) {
if via.IsRelayed {
return pqFragKey{relay: via.relay.LocalIndex, fragID: fragID},
pqFragSource{relay: via.relay.LocalIndex}
}
addr := via.UdpAddr.Addr().Unmap()
bits := 64
if addr.Is4() {
bits = 32
}
prefix, _ := addr.Prefix(bits)
return pqFragKey{addr: via.UdpAddr, fragID: fragID}, pqFragSource{prefix: prefix}
}
// reassembled rebuilds the packet the handshake Machine expects from the header
// of a fragment and the decoded message.
func reassembled(h *header.H, msg []byte) []byte {
p := make([]byte, header.Len, header.Len+len(msg))
header.Encode(p, header.Version, header.Handshake, h.Subtype, h.RemoteIndex, h.MessageCounter)
return append(p, msg...)
}
// usePQ decides whether this outbound attempt is post-quantum. A peer that does
// not speak PQ silently drops it, so timer retries alternate between PQ and
// classical, starting with PQ. An attempt triggered early by a lighthouse reply
// is sending to addresses nothing has been sent to yet, and a new path gets the
// PQ attempt first. With require, or for a peer that completed a PQ handshake
// when sticky is on, every attempt is PQ. Once anything has come back for the PQ
// attempt the peer evidently speaks PQ, and a classical attempt would only race
// it, so classical stops getting turns.
func (hm *HandshakeManager) usePQ(hh *HandshakeHostInfo, vpnIp netip.Addr, triggered bool) bool {
cfg := hm.pqConfig()
switch {
case !cfg.initiate:
return false
case cfg.require:
return true
case hh.pq != nil && hh.pq.heard:
return true
case cfg.sticky:
if _, ok := hm.pqPeers.Load(vpnIp); ok {
return true
}
}
return hh.counter <= 1 || triggered || !hh.lastPQ
}
// relayEstablished pushes the pending handshake with target through a relay that
// just came up, rather than leaving it for the next retry. Nothing has gone over
// the relay yet, so like any new path it gets the PQ attempt first, which then
// counts as the PQ turn: the next retry is classical, as if it had been sent on
// schedule. Without this, relays coming up while the first attempt is in flight
// handed every relayed handshake to the classical retry.
func (hm *HandshakeManager) relayEstablished(target netip.Addr, relayHI *HostInfo, relay *Relay) {
hh := hm.queryVpnIp(target)
if hh == nil {
return
}
hh.Lock()
defer hh.Unlock()
if hm.queryVpnIp(target) != hh {
return
}
var now, later [][]byte
switch {
case hh.pq != nil:
var err error
now, later, err = hh.pq.sender.Attempt(fragment.DefaultRepair(hh.pq.sender.Layout().K))
if err != nil {
return
}
hh.lastPQ = true
case hh.ready:
now = [][]byte{hh.hostinfo.HandshakePacket[handshakePacketStage0]}
default:
return
}
f := hm.f
hh.hostinfo.logger(hm.l).Info("Send handshake via new relay", "relay", relayHI.vpnAddrs[0])
_ = hm.sendFragmented(now, later, func(p []byte) error {
f.SendVia(relayHI, relay, p, make([]byte, 12), make([]byte, mtu), false, 0)
return nil
})
}
// buildPQAttempt creates the initiator's PQ Machine and msg1 sender.
func (hm *HandshakeManager) buildPQAttempt(hh *HandshakeHostInfo) bool {
cs := hm.f.pki.getCertState()
v := hm.initiatingVersion(hh, cs)
if cs.GetCredential(v) == nil {
hm.f.l.Error("Unable to handshake with host because no certificate is available",
"vpnAddrs", hh.hostinfo.vpnAddrs, "certVersion", v)
return false
}
pq := &pqAttempt{}
machine, err := handshake.NewMachine(
v, cs.GetCredential,
hm.certVerifier(), func() (uint32, error) { return hm.allocatePQIndex(hh, pq) },
true, pqSubtype,
)
if err != nil {
hm.f.l.Error("Failed to create handshake machine",
"vpnAddrs", hh.hostinfo.vpnAddrs, "error", err)
return false
}
msg, err := machine.Initiate(nil)
if err != nil {
hm.f.l.Error("Failed to initiate handshake",
"vpnAddrs", hh.hostinfo.vpnAddrs, "error", err)
hm.releasePQIndex(hh, pq)
return false
}
sender, err := fragment.NewSender(msg[header.Len:], pqSubtype, 0, 1, pq.index)
if err != nil {
hm.f.l.Error("Failed to fragment handshake",
"vpnAddrs", hh.hostinfo.vpnAddrs, "error", err)
hm.releasePQIndex(hh, pq)
return false
}
pq.machine = machine
pq.sender = sender
hh.pq = pq
return true
}
// allocatePQIndex registers a local index for the PQ attempt. Unlike
// allocateIndex it leaves hostinfo.localIndexId alone; that is settled when one
// of the attempts completes.
func (hm *HandshakeManager) allocatePQIndex(hh *HandshakeHostInfo, pq *pqAttempt) (uint32, error) {
hm.mainHostMap.RLock()
defer hm.mainHostMap.RUnlock()
hm.Lock()
defer hm.Unlock()
for range 32 {
index, err := generateIndex(hm.l)
if err != nil {
return 0, err
}
_, inPending := hm.indexes[index]
_, inMain := hm.mainHostMap.Indexes[index]
if !inMain && !inPending {
pq.index = index
hm.indexes[index] = hh
return index, nil
}
}
return 0, errors.New("failed to generate unique localIndexId")
}
func (hm *HandshakeManager) releasePQIndex(hh *HandshakeHostInfo, pq *pqAttempt) {
hm.Lock()
defer hm.Unlock()
if pq.index != 0 && hm.indexes[pq.index] == hh {
delete(hm.indexes, pq.index)
}
}
// settleIndex makes winner the hostinfo's local index and releases the index of
// the attempt that lost.
func (hm *HandshakeManager) settleIndex(hh *HandshakeHostInfo, winner uint32) {
hm.Lock()
defer hm.Unlock()
for _, idx := range []uint32{hh.hostinfo.localIndexId, hh.pqIndex()} {
if idx != 0 && idx != winner && hm.indexes[idx] == hh {
delete(hm.indexes, idx)
}
}
hh.hostinfo.localIndexId = winner
}
func (hh *HandshakeHostInfo) pqIndex() uint32 {
if hh.pq == nil {
return 0
}
return hh.pq.index
}
// handleFragment routes one fragment of a PQ handshake message.
func (hm *HandshakeManager) handleFragment(via ViaSender, packet []byte, h *header.H) {
if h.MessageCounter == pqCookieCounter {
hm.handleCookie(via, packet, h)
return
}
fh, sym, err := fragment.Parse(packet[header.Len:])
if err != nil {
hm.l.Debug("dropping malformed handshake fragment", "from", via, "error", err)
return
}
switch h.MessageCounter {
case 1:
if !hm.pqConfig().accept {
hm.l.Debug("dropping post-quantum handshake, not accepting them", "from", via)
return
}
if h.RemoteIndex != 0 {
hm.l.Debug("dropping stage-1 handshake with non-zero RemoteIndex",
"from", via, "remoteIndex", h.RemoteIndex)
return
}
now := time.Now()
key, src := pqFragKeyFor(via, fh.FragID)
if hm.pqTable.Load() >= pqCookieLoad && !(fh.HasCookie() && hm.pqCookies.valid(key, fh.Cookie, now)) {
hm.sendCookie(via, fh, key, len(packet), now)
return
}
res, err := hm.pqTable.Add(key, src, now, fh, sym)
if err != nil {
hm.l.Debug("dropping handshake fragment", "from", via, "error", err)
return
}
if res == nil {
return
}
if existing := hm.pqAnswered.lookup(res.Msg, time.Now()); existing != nil {
// Already answered: resend, rate limited, with fresh symbols.
if msg := existing.HandshakePacket[handshakePacketStage2]; msg != nil {
hm.sendHandshakeResponse(via, msg, existing, true, res.Received)
}
return
}
hm.beginHandshake(via, reassembled(h, res.Msg), h, res.Received)
case 2:
if hh := hm.queryIndex(h.RemoteIndex); hh != nil {
hm.continuePQ(via, hh, h, fh, sym)
}
}
}
// continuePQ collects msg2 fragments for the initiator's PQ attempt.
func (hm *HandshakeManager) continuePQ(via ViaSender, hh *HandshakeHostInfo, h *header.H, fh fragment.Header, sym []byte) {
hh.Lock()
defer hh.Unlock()
pq := hh.pq
if pq == nil || pq.index != h.RemoteIndex || !hm.stillPending(hh, pq.index) {
return
}
pq.heard = true
if pq.msg2 == nil {
a, err := fragment.NewAssembly(fh)
if err != nil {
return
}
pq.msg2 = a
}
ready, err := pq.msg2.Add(fh, sym)
if errors.Is(err, fragment.ErrMismatch) {
// The responder's FragID and MsgLen never change, so this is a stray or a
// forgery. Start over from this fragment rather than wedge.
pq.msg2 = nil
return
}
if err != nil || !ready {
return
}
msg, err := pq.msg2.Decode()
pq.msg2 = nil
if err != nil {
hm.l.Debug("failed to decode handshake fragments", "from", via, "error", err)
return
}
hm.processResponse(via, hh, pq.machine, reassembled(h, msg))
}
// sendFragmented sends packets to a peer: now immediately and later after the
// repair delay. It returns the first error from the immediate sends; errors from
// delayed sends are only logged.
func (hm *HandshakeManager) sendFragmented(now, later [][]byte, send func([]byte) error) error {
var first error
for _, p := range now {
if err := send(p); err != nil && first == nil {
first = err
}
}
sendLater := func() {
for _, p := range later {
if err := send(p); err != nil {
hm.l.Debug("Failed to send handshake repair fragment", "error", err)
}
}
}
delay := hm.pqConfig().repairDelay
switch {
case len(later) == 0:
case delay <= 0:
sendLater()
default:
time.AfterFunc(delay, sendLater)
}
return first
}
// sendFragmentedResponse sends the responder's msg2 as fragments. A resend of a
// cached response (cached) offers fresh symbols and is rate limited to one per
// retry interval. received is how many bytes the triggering msg1 took, which
// bounds what may be sent back before the address is validated.
func (hm *HandshakeManager) sendFragmentedResponse(via ViaSender, msg []byte, hostinfo *HostInfo, cached bool, received int) error {
now := time.Now()
if cached {
last := hostinfo.handshakeResendAt.Load()
if now.UnixNano()-last < int64(hm.config.tryInterval) ||
!hostinfo.handshakeResendAt.CompareAndSwap(last, now.UnixNano()) {
return errors.New("handshake response resend rate limited")
}
} else {
hostinfo.handshakeResendAt.Store(now.UnixNano())
}
var h header.H
if err := h.Parse(msg); err != nil {
return err
}
s, err := fragment.NewSender(msg[header.Len:], h.Subtype, h.RemoteIndex, h.MessageCounter, hostinfo.localIndexId)
if err != nil {
return err
}
l := s.Layout()
repair := fragment.DefaultRepair(l.K)
if received > 0 {
pktLen := header.Len + fragment.HeaderLen + l.SymbolSize
budget := pqAmplification * received / pktLen
if budget < l.K {
return errors.New("handshake response exceeds the amplification budget")
}
repair = min(repair, budget-l.K)
}
n := uint32(l.K + repair)
s.SetNext(int(hostinfo.handshakeNextSymbol.Add(n) - n))
pkts, later, err := s.Attempt(repair)
if err != nil {
return err
}
if via.IsRelayed {
if via.relay == nil {
return errors.New("both addr and via.relay are nil")
}
// We received a valid handshake on this relay, so make sure the relay
// state reflects that, in case it had been marked Disestablished.
hostinfo.relayState.InsertRelayTo(via.relayHI.vpnAddrs[0])
via.relayHI.relayState.UpdateRelayForByIdxState(via.relay.LocalIndex, Established)
}
hm.f.messageMetrics.Tx(header.Handshake, h.Subtype, int64(len(pkts)+len(later)))
return hm.sendFragmented(pkts, later, hm.sendTo(via))
}
// notePQPeer remembers peers that completed a PQ handshake, for sticky mode.
func (hm *HandshakeManager) notePQPeer(subtype header.MessageSubType, vpnAddrs []netip.Addr) {
if !hm.pqConfig().sticky || subtype != pqSubtype {
return
}
for _, a := range vpnAddrs {
hm.pqPeers.Store(a, struct{}{})
}
if hm.l.Enabled(context.Background(), slog.LevelDebug) {
hm.l.Debug("Remembering post-quantum peer", "vpnAddrs", vpnAddrs)
}
}
// pqAnswered remembers the msg1s a responder recently answered, keyed by hash,
// so a repeat is recognised before any noise, KEM or certificate work. Repeats
// are routine: the trailing repair fragments of an attempt rebuild a msg1 that
// its data fragments already completed, and a retrying initiator offers fresh
// symbols for the same msg1. Entries only exist for completed handshakes, which
// need a valid certificate, and live as long as an initiator keeps retrying.
type pqAnswered struct {
sync.Mutex
ttl time.Duration
entries map[[sha256.Size]byte]pqAnsweredEntry
}
type pqAnsweredEntry struct {
hostinfo *HostInfo
at time.Time
}
const pqAnsweredMax = 4096
func newPQAnswered(ttl time.Duration) *pqAnswered {
return &pqAnswered{ttl: ttl, entries: map[[sha256.Size]byte]pqAnsweredEntry{}}
}
func (a *pqAnswered) add(msg1 []byte, hostinfo *HostInfo, now time.Time) {
a.Lock()
defer a.Unlock()
if len(a.entries) >= pqAnsweredMax {
for k := range a.entries {
delete(a.entries, k)
break
}
}
a.entries[sha256.Sum256(msg1)] = pqAnsweredEntry{hostinfo: hostinfo, at: now}
}
func (a *pqAnswered) lookup(msg1 []byte, now time.Time) *HostInfo {
a.Lock()
defer a.Unlock()
e, ok := a.entries[sha256.Sum256(msg1)]
if !ok || now.Sub(e.at) > a.ttl {
return nil
}
return e.hostinfo
}
func (a *pqAnswered) expire(now time.Time) {
a.Lock()
defer a.Unlock()
for k, e := range a.entries {
if now.Sub(e.at) > a.ttl {
delete(a.entries, k)
}
}
}
// A responder whose msg1 table is under load stops keeping state for senders
// that have not shown they receive at their address. It answers a fragment
// without a valid cookie with a cookie reply and keeps nothing; the initiator
// resends with the cookie attached. This is WireGuard's under-load cookie, QUIC's
// Retry and SCTP's state cookie: one extra round trip, only while under attack.
const (
// pqCookieLoad is the table load above which msg1 fragments need a cookie.
pqCookieLoad = 0.5
// pqCookieCounter marks a cookie reply in the header's message counter.
// Handshake messages count from 1.
pqCookieCounter = 0
// pqCookieRotate is how often the cookie secret changes. A cookie stays
// valid for between one and two rotations.
pqCookieRotate = 2 * time.Minute
pqCookieReplyLen = header.Len + fragment.CookieLen
)
// pqCookies mints and checks cookies: an HMAC over the sender's address (or relay
// channel) and FragID under a rotating secret, with the previous secret still
// accepted.
type pqCookies struct {
sync.Mutex
secrets [2][32]byte
rotatedAt time.Time
}
func newPQCookies(now time.Time) *pqCookies {
c := &pqCookies{rotatedAt: now}
_, _ = rand.Read(c.secrets[0][:])
_, _ = rand.Read(c.secrets[1][:])
return c
}
func (c *pqCookies) rotate(now time.Time) {
if now.Sub(c.rotatedAt) < pqCookieRotate {
return
}
c.secrets[1] = c.secrets[0]
_, _ = rand.Read(c.secrets[0][:])
c.rotatedAt = now
}
func cookieMAC(secret []byte, key pqFragKey) fragment.Cookie {
mac := hmac.New(sha256.New, secret)
addr := key.addr.Addr().As16()
mac.Write(addr[:])
var b [10]byte
binary.BigEndian.PutUint16(b[0:], key.addr.Port())
binary.BigEndian.PutUint32(b[2:], key.relay)
binary.BigEndian.PutUint32(b[6:], key.fragID)
mac.Write(b[:])
var c fragment.Cookie
copy(c[:], mac.Sum(nil))
return c
}
func (c *pqCookies) make(key pqFragKey, now time.Time) fragment.Cookie {
c.Lock()
defer c.Unlock()
c.rotate(now)
return cookieMAC(c.secrets[0][:], key)
}
func (c *pqCookies) valid(key pqFragKey, cookie fragment.Cookie, now time.Time) bool {
c.Lock()
defer c.Unlock()
c.rotate(now)
for _, s := range c.secrets {
want := cookieMAC(s[:], key)
if hmac.Equal(want[:], cookie[:]) {
return true
}
}
return false
}
// sendCookie answers a msg1 fragment that arrived without a valid cookie while
// under load. The reply is never larger than what triggered it, and replies are
// rate limited as a whole.
func (hm *HandshakeManager) sendCookie(via ViaSender, fh fragment.Header, key pqFragKey, received int, now time.Time) {
if received < pqCookieReplyLen {
return
}
if _, ok := hm.pqCookieLimiter.Allow(); !ok {
return
}
c := hm.pqCookies.make(key, now)
reply := make([]byte, header.Len, pqCookieReplyLen)
header.Encode(reply, header.Version, header.Handshake, pqSubtype, fh.FragID, pqCookieCounter)
reply = append(reply, c[:]...)
hm.f.messageMetrics.Tx(header.Handshake, pqSubtype, 1)
if err := hm.sendTo(via)(reply); err != nil {
hm.l.Debug("Failed to send handshake cookie", "to", via, "error", err)
}
}
// handleCookie attaches a responder's cookie to the PQ attempt it names and
// resends right away, at most once per retry interval, rather than waiting for
// the next attempt.
func (hm *HandshakeManager) handleCookie(via ViaSender, packet []byte, h *header.H) {
if len(packet) != pqCookieReplyLen {
return
}
hh := hm.queryIndex(h.RemoteIndex)
if hh == nil {
return
}
hh.Lock()
defer hh.Unlock()
pq := hh.pq
if pq == nil || pq.index != h.RemoteIndex || !hm.stillPending(hh, pq.index) {
return
}
var c fragment.Cookie
copy(c[:], packet[header.Len:])
pq.sender.SetCookie(c)
pq.heard = true
now := time.Now()
if now.Sub(pq.cookieResendAt) < hm.config.tryInterval {
return
}
pq.cookieResendAt = now
pkts, later, err := pq.sender.Attempt(fragment.DefaultRepair(pq.sender.Layout().K))
if err != nil {
return
}
hm.messageMetrics.Tx(header.Handshake, pqSubtype, int64(len(pkts)+len(later)))
if err := hm.sendFragmented(pkts, later, hm.sendTo(via)); err != nil {
hm.l.Debug("Failed to resend handshake with cookie", "to", via, "error", err)
}
}
// sendTo returns a function sending a packet back the way via came: directly, or
// through the relay it arrived on.
func (hm *HandshakeManager) sendTo(via ViaSender) func([]byte) error {
f := hm.f
if !via.IsRelayed {
return func(p []byte) error {
return f.outside.WriteTo(p, via.UdpAddr)
}
}
return func(p []byte) error {
if via.relay == nil {
return errors.New("both addr and via.relay are nil")
}
f.SendVia(via.relayHI, via.relay, p, make([]byte, 12), make([]byte, mtu), false, 0)
return nil
}
}
-398
View File
@@ -1,398 +0,0 @@
package nebula
import (
"math/rand/v2"
"net/netip"
"sync"
"testing"
"time"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/handshake/fragment"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/test"
"github.com/slackhq/nebula/udp"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
// recordingConn captures what is written to it.
type recordingConn struct {
udp.NoopConn
sync.Mutex
sent [][]byte
}
func (c *recordingConn) WriteTo(b []byte, _ netip.AddrPort) error {
c.Lock()
defer c.Unlock()
c.sent = append(c.sent, append([]byte(nil), b...))
return nil
}
func (c *recordingConn) take() [][]byte {
c.Lock()
defer c.Unlock()
s := c.sent
c.sent = nil
return s
}
func newPQTestManager(t *testing.T) (*HandshakeManager, *recordingConn) {
l := test.NewLogger()
conn := &recordingConn{}
cfg := defaultHandshakeConfig
cfg.pq = pqHandshakeConfig{initiate: true, accept: true}
hm := NewHandshakeManager(l, newHostMap(l), newTestLighthouse(), conn, cfg)
hm.f = &Interface{
handshakeManager: hm,
outside: conn,
messageMetrics: newMessageMetricsOnlyRecvError(),
l: l,
}
return hm, conn
}
// fragmentIndexes checks every packet is a fragment of the expected response and
// returns their symbol indexes.
func fragmentIndexes(t *testing.T, pkts [][]byte, fragID, remoteIndex uint32) []int {
t.Helper()
var out []int
for _, p := range pkts {
require.LessOrEqual(t, len(p), fragment.MaxPacketLen)
var h header.H
require.NoError(t, h.Parse(p))
assert.Equal(t, pqSubtype, h.Subtype)
assert.Equal(t, remoteIndex, h.RemoteIndex)
assert.Equal(t, uint64(2), h.MessageCounter)
fh, _, err := fragment.Parse(p[header.Len:])
require.NoError(t, err)
assert.Equal(t, fragID, fh.FragID)
out = append(out, int(fh.Index))
}
return out
}
func TestSendFragmentedResponse(t *testing.T) {
hm, conn := newPQTestManager(t)
via := ViaSender{UdpAddr: netip.MustParseAddrPort("192.0.2.1:4242")}
hostinfo := &HostInfo{localIndexId: 77}
// A 2000 byte message is 2 symbols of 1000, so packets of 1024 bytes.
msg := make([]byte, header.Len, header.Len+2000)
header.Encode(msg, header.Version, header.Handshake, pqSubtype, 5, 2)
r := rand.New(rand.NewPCG(1, 1))
for range 2000 {
msg = append(msg, byte(r.Uint32()))
}
const pktLen = header.Len + fragment.HeaderLen + 1000
t.Run("first send", func(t *testing.T) {
require.NoError(t, hm.sendFragmentedResponse(via, msg, hostinfo, false, 0))
assert.Equal(t, []int{0, 1, 2, 3}, fragmentIndexes(t, conn.take(), 77, 5))
})
t.Run("resends are rate limited", func(t *testing.T) {
require.Error(t, hm.sendFragmentedResponse(via, msg, hostinfo, true, 10*pktLen))
assert.Empty(t, conn.take())
})
allowResend := func() {
hostinfo.handshakeResendAt.Store(time.Now().Add(-time.Second).UnixNano())
}
t.Run("resend offers fresh symbols", func(t *testing.T) {
allowResend()
require.NoError(t, hm.sendFragmentedResponse(via, msg, hostinfo, true, 10*pktLen))
assert.Equal(t, []int{4, 5, 6, 7}, fragmentIndexes(t, conn.take(), 77, 5))
})
t.Run("amplification budget trims repair", func(t *testing.T) {
allowResend()
// 3x of a little over one packet allows 3 packets: both data and 1 repair.
require.NoError(t, hm.sendFragmentedResponse(via, msg, hostinfo, true, pktLen+10))
assert.Equal(t, []int{8, 9, 10}, fragmentIndexes(t, conn.take(), 77, 5))
})
t.Run("amplification budget below the data refuses", func(t *testing.T) {
allowResend()
require.Error(t, hm.sendFragmentedResponse(via, msg, hostinfo, true, pktLen/2))
assert.Empty(t, conn.take())
})
}
func TestPQAnswered(t *testing.T) {
now := time.Unix(1000, 0)
a := newPQAnswered(5 * time.Second)
hi := &HostInfo{}
assert.Nil(t, a.lookup([]byte("msg1"), now))
a.add([]byte("msg1"), hi, now)
assert.Same(t, hi, a.lookup([]byte("msg1"), now.Add(time.Second)))
assert.Nil(t, a.lookup([]byte("other"), now))
assert.Nil(t, a.lookup([]byte("msg1"), now.Add(6*time.Second)), "stale entries are not answered")
a.expire(now.Add(6 * time.Second))
assert.Empty(t, a.entries)
for i := range pqAnsweredMax + 10 {
a.add([]byte{byte(i), byte(i >> 8)}, hi, now)
}
assert.Len(t, a.entries, pqAnsweredMax)
}
func TestPQFragKeyFor(t *testing.T) {
k1, s1 := pqFragKeyFor(ViaSender{UdpAddr: netip.MustParseAddrPort("192.0.2.1:4242")}, 9)
k2, s2 := pqFragKeyFor(ViaSender{UdpAddr: netip.MustParseAddrPort("192.0.2.1:4243")}, 9)
assert.NotEqual(t, k1, k2, "different ports are different messages")
assert.Equal(t, s1, s2, "but share an IPv4 /32 quota")
_, s3 := pqFragKeyFor(ViaSender{UdpAddr: netip.MustParseAddrPort("[2001:db8::1]:1")}, 9)
_, s4 := pqFragKeyFor(ViaSender{UdpAddr: netip.MustParseAddrPort("[2001:db8::ffff]:1")}, 9)
assert.Equal(t, s3, s4, "an IPv6 /64 shares a quota")
_, s5 := pqFragKeyFor(ViaSender{UdpAddr: netip.MustParseAddrPort("[::ffff:192.0.2.1]:1")}, 9)
assert.Equal(t, s1, s5, "v4-mapped addresses count as IPv4")
}
func TestPQCookies(t *testing.T) {
now := time.Unix(1000, 0)
c := newPQCookies(now)
key := pqFragKey{addr: netip.MustParseAddrPort("192.0.2.1:4242"), fragID: 9}
cookie := c.make(key, now)
assert.True(t, c.valid(key, cookie, now))
for _, other := range []pqFragKey{
{addr: netip.MustParseAddrPort("192.0.2.2:4242"), fragID: 9},
{addr: netip.MustParseAddrPort("192.0.2.1:4243"), fragID: 9},
{addr: netip.MustParseAddrPort("192.0.2.1:4242"), fragID: 10},
{addr: netip.MustParseAddrPort("192.0.2.1:4242"), relay: 1, fragID: 9},
} {
assert.False(t, c.valid(other, cookie, now), "%+v", other)
}
now = now.Add(pqCookieRotate)
assert.True(t, c.valid(key, cookie, now), "the previous secret is still accepted")
assert.NotEqual(t, cookie, c.make(key, now), "new cookies come from the new secret")
now = now.Add(pqCookieRotate)
assert.False(t, c.valid(key, cookie, now), "two rotations retire a cookie")
}
// pqMsg1Fragment builds the first fragment of a msg1, optionally with a cookie.
func pqMsg1Fragment(t *testing.T, fragID uint32, cookie *fragment.Cookie) ([]byte, *header.H) {
t.Helper()
s, err := fragment.NewSender(make([]byte, 2000), pqSubtype, 0, 1, fragID)
require.NoError(t, err)
if cookie != nil {
s.SetCookie(*cookie)
}
pkts, _, err := s.Attempt(0)
require.NoError(t, err)
var h header.H
require.NoError(t, h.Parse(pkts[0]))
return pkts[0], &h
}
// fillPQTable puts the responder's msg1 table at the cookie threshold.
func fillPQTable(t *testing.T, hm *HandshakeManager) {
t.Helper()
now := time.Now()
for i := 0; hm.pqTable.Load() < pqCookieLoad; i++ {
p, _ := pqMsg1Fragment(t, uint32(i+1), nil)
fh, sym, err := fragment.Parse(p[header.Len:])
require.NoError(t, err)
addr := netip.AddrPortFrom(netip.AddrFrom4([4]byte{198, 18, byte(i >> 8), byte(i)}), 4242)
key, src := pqFragKeyFor(ViaSender{UdpAddr: addr}, fh.FragID)
_, err = hm.pqTable.Add(key, src, now, fh, sym)
require.NoError(t, err)
}
}
func TestPQCookieUnderLoad(t *testing.T) {
hm, conn := newPQTestManager(t)
via := ViaSender{UdpAddr: netip.MustParseAddrPort("192.0.2.1:4242")}
t.Run("no cookie needed below the threshold", func(t *testing.T) {
p, h := pqMsg1Fragment(t, 5000, nil)
before := hm.pqTable.Len()
hm.handleFragment(via, p, h)
assert.Empty(t, conn.take())
assert.Equal(t, before+1, hm.pqTable.Len())
})
fillPQTable(t, hm)
var cookie fragment.Cookie
t.Run("a fragment without a cookie gets one and leaves no state", func(t *testing.T) {
p, h := pqMsg1Fragment(t, 7, nil)
before := hm.pqTable.Len()
hm.handleFragment(via, p, h)
assert.Equal(t, before, hm.pqTable.Len())
sent := conn.take()
require.Len(t, sent, 1)
reply := sent[0]
require.Len(t, reply, pqCookieReplyLen)
var rh header.H
require.NoError(t, rh.Parse(reply))
assert.Equal(t, header.Handshake, rh.Type)
assert.Equal(t, pqSubtype, rh.Subtype)
assert.Equal(t, uint32(7), rh.RemoteIndex, "the reply names the initiator's FragID")
assert.Equal(t, uint64(pqCookieCounter), rh.MessageCounter)
copy(cookie[:], reply[header.Len:])
})
t.Run("a forged cookie gets a fresh one", func(t *testing.T) {
bad := cookie
bad[0] ^= 1
p, h := pqMsg1Fragment(t, 7, &bad)
before := hm.pqTable.Len()
hm.handleFragment(via, p, h)
assert.Equal(t, before, hm.pqTable.Len())
assert.Len(t, conn.take(), 1)
})
t.Run("the cookie gets the fragment in", func(t *testing.T) {
p, h := pqMsg1Fragment(t, 7, &cookie)
before := hm.pqTable.Len()
hm.handleFragment(via, p, h)
assert.Equal(t, before+1, hm.pqTable.Len())
assert.Empty(t, conn.take())
})
t.Run("a cookie is bound to its address", func(t *testing.T) {
p, h := pqMsg1Fragment(t, 7, &cookie)
hm.handleFragment(ViaSender{UdpAddr: netip.MustParseAddrPort("192.0.2.99:4242")}, p, h)
assert.Len(t, conn.take(), 1, "another address gets its own cookie instead")
})
t.Run("a reply never outgrows what prompted it", func(t *testing.T) {
p, h := pqMsg1Fragment(t, 8, nil)
hm.sendCookie(via, fragment.Header{FragID: 8}, pqFragKey{}, pqCookieReplyLen-1, time.Now())
assert.Empty(t, conn.take())
hm.handleFragment(via, p, h)
assert.Len(t, conn.take(), 1)
})
}
func TestPQHandleCookie(t *testing.T) {
hm, conn := newPQTestManager(t)
via := ViaSender{UdpAddr: netip.MustParseAddrPort("192.0.2.1:4242")}
s, err := fragment.NewSender(make([]byte, 2000), pqSubtype, 0, 1, 99)
require.NoError(t, err)
_, _, err = s.Attempt(2)
require.NoError(t, err)
hh := &HandshakeHostInfo{hostinfo: &HostInfo{}, pq: &pqAttempt{index: 99, sender: s}}
hm.indexes[99] = hh
reply := make([]byte, header.Len, pqCookieReplyLen)
header.Encode(reply, header.Version, header.Handshake, pqSubtype, 99, pqCookieCounter)
reply = append(reply, bytes16(0xab)...)
var h header.H
require.NoError(t, h.Parse(reply))
hm.handleFragment(via, reply, &h)
assert.True(t, hh.pq.heard, "a cookie reply shows the peer speaks PQ")
sent := conn.take()
require.Len(t, sent, 4, "an immediate resend of fresh symbols")
for i, p := range sent {
fh, _, err := fragment.Parse(p[header.Len:])
require.NoError(t, err)
assert.True(t, fh.HasCookie())
assert.Equal(t, fragment.Cookie(bytes16(0xab)), fh.Cookie)
assert.Equal(t, uint8(4+i), fh.Index)
}
hm.handleFragment(via, reply, &h)
assert.Empty(t, conn.take(), "immediate resends are rate limited")
t.Run("replies for other indexes are ignored", func(t *testing.T) {
other := append([]byte(nil), reply...)
header.Encode(other, header.Version, header.Handshake, pqSubtype, 98, pqCookieCounter)
var oh header.H
require.NoError(t, oh.Parse(other))
hh.pq.cookieResendAt = time.Time{}
hm.handleFragment(via, other, &oh)
assert.Empty(t, conn.take())
})
}
func bytes16(b byte) []byte {
out := make([]byte, 16)
for i := range out {
out[i] = b
}
return out
}
func TestReloadPQ(t *testing.T) {
l := test.NewLogger()
c := config.NewC(l)
require.NoError(t, c.LoadString(`handshakes: {pq: {initiate: false}}`))
cfg := defaultHandshakeConfig
cfg.pq = pqHandshakeConfigFrom(c)
hm := NewHandshakeManager(l, newHostMap(l), newTestLighthouse(), &udp.NoopConn{}, cfg)
c.RegisterReloadCallback(hm.reloadPQ)
assert.Equal(t, pqHandshakeConfig{repairDelay: DefaultPQRepairDelay}, *hm.pqConfig())
require.NoError(t, c.ReloadConfigString(`handshakes: {pq: {initiate: true, accept: true, sticky: true, repair_delay: 25ms}}`))
assert.Equal(t, pqHandshakeConfig{initiate: true, accept: true, sticky: true, repairDelay: 25 * time.Millisecond}, *hm.pqConfig())
before := hm.pqConfig()
require.NoError(t, c.ReloadConfigString(`handshakes: {try_interval: 1s, pq: {initiate: true, accept: true, sticky: true, repair_delay: 25ms}}`))
assert.Same(t, before, hm.pqConfig(), "an unrelated change leaves the settings alone")
require.NoError(t, c.ReloadConfigString(`handshakes: {}`))
assert.Equal(t, pqHandshakeConfig{repairDelay: DefaultPQRepairDelay}, *hm.pqConfig(), "removing the section restores the defaults")
require.NoError(t, c.ReloadConfigString(`handshakes: {pq: {require: true}}`))
assert.Equal(t, pqHandshakeConfig{initiate: true, accept: true, require: true, repairDelay: DefaultPQRepairDelay}, *hm.pqConfig(), "require implies initiate and accept")
}
func TestUsePQ(t *testing.T) {
hm, _ := newPQTestManager(t)
peer := netip.MustParseAddr("10.1.0.1")
set := func(cfg pqHandshakeConfig) { hm.pq.Store(&cfg) }
// attempts plays a run of attempts, given whether each was triggered early,
// and returns which were PQ.
attempts := func(triggered ...bool) []bool {
hh := &HandshakeHostInfo{}
var out []bool
for _, tr := range triggered {
hh.counter++
hh.lastPQ = hm.usePQ(hh, peer, tr)
out = append(out, hh.lastPQ)
}
return out
}
set(pqHandshakeConfig{})
assert.Equal(t, []bool{false, false, false}, attempts(false, false, false), "off")
set(pqHandshakeConfig{initiate: true})
assert.Equal(t, []bool{true, false, true, false}, attempts(false, false, false, false), "timer retries alternate, PQ first")
assert.Equal(t, []bool{true, true, false, true, false}, attempts(false, true, false, true, false), "an early trigger is a new path, which gets PQ first")
assert.Equal(t, []bool{true, true, false}, attempts(true, true, false), "a triggered first attempt is PQ")
t.Run("hearing back stops the classical turns", func(t *testing.T) {
hh := &HandshakeHostInfo{pq: &pqAttempt{}}
hh.counter, hh.lastPQ = 1, true
hh.counter++
assert.False(t, hm.usePQ(hh, peer, false), "unheard, the classical attempt gets its turn")
hh.pq.heard = true
assert.True(t, hm.usePQ(hh, peer, false))
})
set(pqHandshakeConfig{initiate: true, accept: true, require: true})
assert.Equal(t, []bool{true, true, true, true}, attempts(false, false, true, false), "require never offers classical")
set(pqHandshakeConfig{initiate: true, sticky: true})
assert.Equal(t, []bool{true, false}, attempts(false, false), "sticky alternates until the peer has done PQ")
hm.notePQPeer(pqSubtype, []netip.Addr{peer})
assert.Equal(t, []bool{true, true, true}, attempts(false, false, false), "then only PQ")
}
+2 -68
View File
@@ -11,7 +11,6 @@ import (
"github.com/slackhq/nebula/test"
"github.com/slackhq/nebula/udp"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
)
func Test_NewHandshakeManagerVpnIp(t *testing.T) {
@@ -43,6 +42,8 @@ func Test_NewHandshakeManagerVpnIp(t *testing.T) {
i2 := blah.StartHandshake(ip, nil)
assert.Same(t, i, i2)
i.remotes = NewRemoteList([]netip.Addr{}, nil)
// Adding something to pending should not affect the main hostmap
assert.Empty(t, mainHM.Hosts)
@@ -234,70 +235,3 @@ func TestHandleIncomingDispatch(t *testing.T) {
assert.Empty(t, hm.indexes, "orphan stage-2 must not create state")
})
}
// Test_PendingHandshakeHasRemotes covers create-tunnel -address: a pending hostinfo must already
// carry its RemoteList when StartHandshake runs its callback, so SetRemote there, before the
// handshake loop has run, is safe and the address becomes one the handshake is sent to.
func Test_PendingHandshakeHasRemotes(t *testing.T) {
l := test.NewLogger()
mainHM := newHostMap(l)
preferredRanges := []netip.Prefix{}
mainHM.preferredRanges.Store(&preferredRanges)
hm := NewHandshakeManager(l, mainHM, newTestLighthouse(), &udp.NoopConn{}, defaultHandshakeConfig)
vpnIp := netip.MustParseAddr("172.1.1.2")
remote := netip.MustParseAddrPort("192.168.1.9:4242")
hi := hm.StartHandshake(vpnIp, func(hh *HandshakeHostInfo) {
require.NotNil(t, hh.hostinfo.remotes, "a pending handshake must never be without its RemoteList")
hh.hostinfo.SetRemote(remote)
})
assert.Contains(t, hi.remotes.CopyAddrs(preferredRanges), remote, "the handshake must be sent to the supplied remote")
assert.Equal(t, remote, hi.GetRemote())
// A RemoteList keeps one learned address per host, so a second remote replaces the
// first — the same behaviour change-remote has on an established tunnel.
other := netip.MustParseAddrPort("192.168.1.10:4242")
hi.SetRemote(other)
assert.Equal(t, []netip.AddrPort{other}, hi.remotes.CopyAddrs(preferredRanges))
assert.Equal(t, other, hi.GetRemote())
}
// newRemotesRaceFixture builds a handshake manager with an established tunnel to vpnAddr whose
// RemoteList is the lighthouse's cached list, the state every teardown-during-handshake race starts from.
func newRemotesRaceFixture(t *testing.T, vpnAddr netip.Addr) (*HandshakeManager, *Interface, *HostInfo) {
t.Helper()
l := test.NewLogger()
mainHM := newHostMap(l)
preferredRanges := []netip.Prefix{}
mainHM.preferredRanges.Store(&preferredRanges)
lh := newTestLighthouse()
hm := NewHandshakeManager(l, mainHM, lh, &udp.NoopConn{}, defaultHandshakeConfig)
f := &Interface{hostMap: mainHM, lightHouse: lh, handshakeManager: hm, l: l}
hm.f = f
established := &HostInfo{
vpnAddrs: []netip.Addr{vpnAddr},
localIndexId: 1,
remoteIndexId: 101,
remotes: lh.QueryCache([]netip.Addr{vpnAddr}),
HandshakePacket: map[uint8][]byte{},
}
mainHM.Lock()
mainHM.unlockedAddHostInfo(established, f)
mainHM.Unlock()
return hm, f, established
}
// With no handshake pending, tearing down the last tunnel still evicts the list.
func Test_CloseTunnelEvictsWithoutPendingHandshake(t *testing.T) {
vpnAddr := netip.MustParseAddr("172.1.1.2")
_, f, established := newRemotesRaceFixture(t, vpnAddr)
f.closeTunnel(established)
assert.NotContains(t, f.hostMap.Hosts, vpnAddr)
assert.NotContains(t, f.lightHouse.addrMap, vpnAddr)
}
+53 -34
View File
@@ -8,27 +8,34 @@ import (
)
//Version 1 header:
// 0 31
// |-----------------------------------------------------------------------|
// | Version (uint4) | Type (uint4) | Subtype (uint8) | Reserved (uint16) | 32
// |-----------------------------------------------------------------------|
// | Remote index (uint32) | 64
// |-----------------------------------------------------------------------|
// | Message counter | 96
// | (uint64) | 128
// |-----------------------------------------------------------------------|
// | payload... |
// 0 31
// |------------------------------------------------------------------------------------|
// | Version (uint4) | Type (uint4) | Subtype (uint8) | Reserved (uint8) | Lane (uint8) | 32
// |------------------------------------------------------------------------------------|
// | Remote index (uint32) | 64
// |------------------------------------------------------------------------------------|
// | Message counter | 96
// | (uint64) | 128
// |------------------------------------------------------------------------------------|
// | payload... |
//
// Lane is the multiport lane index, carved out of the low 8 bits of what was a
// single Reserved (uint16) before multiport. Lane 0 is the base tunnel, which is
// what every sender that does not know about lanes emits and what every non-lane
// packet — handshakes, lighthouse, relays, close — carries, so the field is
// compatible in both directions with a peer that has never heard of it. The
// remaining 8 bits stay reserved and are always sent as zero. The H struct still
// holds the pair as one Reserved field; use H.Lane and EncodeLane to reach the
// low byte.
//
// Lane is part of the AEAD's associated data, so a lane index cannot be altered
// in flight: a packet decrypts on the lane it claims or not at all.
type m = map[string]any
const (
Version uint8 = 1
Len = 16
// MaxOverhead is the most a payload grows on its way to the underlay.
// A relayed packet has two headers and two AEAD tags, one for the inner tunnel and another for the relay tunnel.
// Every supported cipher (today) has a 16 byte tag.
MaxOverhead = 2 * (Len + 16)
)
type MessageType uint8
@@ -62,29 +69,30 @@ const (
const (
TestRequest MessageSubType = 0
TestReply MessageSubType = 1
// LaneProbe is sent on a multiport lane to prove the lane's 5-tuple is
// usable; LaneProbeAck answers it on the base tunnel.
LaneProbe MessageSubType = 2
LaneProbeAck MessageSubType = 3
)
// MaxLane is the largest lane index the header can carry.
const MaxLane = 0xff
// laneMask covers the bits of Reserved that hold the lane index.
const laneMask uint16 = 0x00ff
const (
HandshakeIXPSK0 MessageSubType = 0
HandshakeXXPSK0 MessageSubType = 1
// HandshakeIXPSK2MLKEM1024 is IX with an ephemeral ML-KEM-1024 exchange mixed
// in through the psk2 slot. Its messages exceed the MTU and travel as
// fragments (see handshake/fragment).
HandshakeIXPSK2MLKEM1024 MessageSubType = 2
)
// IsFragmented reports whether handshake messages of this subtype travel as
// fragments.
func IsFragmented(st MessageSubType) bool {
return st == HandshakeIXPSK2MLKEM1024
}
var ErrHeaderTooShort = errors.New("header is too short")
var subTypeTestMap = map[MessageSubType]string{
TestRequest: "testRequest",
TestReply: "testReply",
TestRequest: "testRequest",
TestReply: "testReply",
LaneProbe: "laneProbe",
LaneProbeAck: "laneProbeAck",
}
var subTypeNoneMap = map[MessageSubType]string{0: "none"}
@@ -99,8 +107,7 @@ var subTypeMap = map[MessageType]*map[MessageSubType]string{
Test: &subTypeTestMap,
CloseTunnel: &subTypeNoneMap,
Handshake: {
HandshakeIXPSK0: "ix_psk0",
HandshakeIXPSK2MLKEM1024: "ix_psk2_mlkem1024",
HandshakeIXPSK0: "ix_psk0",
},
Control: &subTypeNoneMap,
}
@@ -117,10 +124,16 @@ type H struct {
// Encode uses the provided byte array to encode the provided header values into.
// Byte array must be capped higher than HeaderLen or this will panic
func Encode(b []byte, v uint8, t MessageType, st MessageSubType, ri uint32, c uint64) []byte {
return EncodeLane(b, v, t, st, ri, c, 0)
}
// EncodeLane is Encode with an explicit multiport lane index, which is carried
// in the low 8 bits of Reserved.
func EncodeLane(b []byte, v uint8, t MessageType, st MessageSubType, ri uint32, c uint64, lane uint8) []byte {
b = b[:Len]
b[0] = v<<4 | byte(t&0x0f)
b[1] = byte(st)
binary.BigEndian.PutUint16(b[2:4], 0)
binary.BigEndian.PutUint16(b[2:4], uint16(lane))
binary.BigEndian.PutUint32(b[4:8], ri)
binary.BigEndian.PutUint64(b[8:16], c)
return b
@@ -153,7 +166,13 @@ func (h *H) Encode(b []byte) ([]byte, error) {
return nil, errors.New("nil header")
}
return Encode(b, h.Version, h.Type, h.Subtype, h.RemoteIndex, h.MessageCounter), nil
return EncodeLane(b, h.Version, h.Type, h.Subtype, h.RemoteIndex, h.MessageCounter, h.Lane()), nil
}
// Lane returns the multiport lane index carried in Reserved. Lane 0 is the base
// tunnel, which is what any sender that does not know about lanes will report.
func (h *H) Lane() uint8 {
return uint8(h.Reserved & laneMask)
}
// Parse is a helper function to parses given bytes into new Header struct
@@ -211,9 +230,9 @@ func IsValidSubType(t MessageType, s MessageSubType) bool {
case Message:
return s == MessageNone || s == MessageRelay
case Handshake:
return s == HandshakeIXPSK0 || s == HandshakeIXPSK2MLKEM1024
return s == HandshakeIXPSK0
case Test:
return s == TestReply || s == TestRequest
return s == TestReply || s == TestRequest || s == LaneProbe || s == LaneProbeAck
case Control, CloseTunnel, RecvError, LightHouse:
return s == 0
default:
+26 -4
View File
@@ -52,6 +52,28 @@ func TestParse(t *testing.T) {
}
}
func TestEncodeLane(t *testing.T) {
b := EncodeLane(make([]byte, Len), Version, Message, MessageNone, 10, 9, 3)
assert.Equal(t, []byte{0x11, 0x0, 0x0, 0x3}, b[:4])
h := &H{}
require.NoError(t, h.Parse(b))
assert.Equal(t, uint16(3), h.Reserved)
assert.Equal(t, uint8(3), h.Lane())
// Encode is EncodeLane on the base tunnel, and the H method round trips the lane.
assert.Equal(t,
EncodeLane(make([]byte, Len), Version, Message, MessageNone, 10, 9, 0),
Encode(make([]byte, Len), Version, Message, MessageNone, 10, 9))
rt, err := h.Encode(make([]byte, Len))
require.NoError(t, err)
assert.Equal(t, b, rt)
// Only the low 8 bits of Reserved are the lane.
assert.Equal(t, uint8(0x2a), (&H{Reserved: 0xff2a}).Lane())
}
func TestTypeName(t *testing.T) {
assert.Equal(t, "test", TypeName(Test))
assert.Equal(t, "test", (&H{Type: Test}).TypeName())
@@ -96,8 +118,7 @@ func TestTypeMap(t *testing.T) {
Test: &subTypeTestMap,
CloseTunnel: &subTypeNoneMap,
Handshake: {
HandshakeIXPSK0: "ix_psk0",
HandshakeIXPSK2MLKEM1024: "ix_psk2_mlkem1024",
HandshakeIXPSK0: "ix_psk0",
},
Control: &subTypeNoneMap,
}, subTypeMap)
@@ -123,13 +144,14 @@ func TestIsValidSubType(t *testing.T) {
assert.False(t, IsValidSubType(Message, 2))
assert.True(t, IsValidSubType(Handshake, HandshakeIXPSK0))
assert.True(t, IsValidSubType(Handshake, HandshakeIXPSK2MLKEM1024))
// HandshakeXXPSK0 is defined but not a wire-valid subtype.
assert.False(t, IsValidSubType(Handshake, HandshakeXXPSK0))
assert.True(t, IsValidSubType(Test, TestRequest))
assert.True(t, IsValidSubType(Test, TestReply))
assert.False(t, IsValidSubType(Test, 2))
assert.True(t, IsValidSubType(Test, LaneProbe))
assert.True(t, IsValidSubType(Test, LaneProbeAck))
assert.False(t, IsValidSubType(Test, 4))
// These types only ever carry subtype 0.
for _, mt := range []MessageType{Control, CloseTunnel, RecvError, LightHouse} {
+71 -5
View File
@@ -6,6 +6,7 @@ import (
"errors"
"fmt"
"log/slog"
"net"
"net/netip"
"slices"
"sync"
@@ -17,6 +18,7 @@ import (
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/logging"
)
const defaultPromoteEvery = 1000 // Count of packets sent before we try moving a tunnel to a preferred underlay ip address
@@ -260,11 +262,6 @@ type HostInfo struct {
// We need these to avoid replayed handshake packets creating new hostinfos which causes churn
HandshakePacket map[uint8][]byte
// A responder resending a fragmented handshake response offers symbols it has
// not sent before, at most once per retry interval.
handshakeResendAt atomic.Int64 // unix nanos of the last send
handshakeNextSymbol atomic.Uint32 // next fragment symbol index to offer
// nextLHQuery is the earliest we can ask the lighthouse for new information.
// This is used to limit lighthouse re-queries in chatty clients
nextLHQuery atomic.Int64
@@ -281,6 +278,11 @@ type HostInfo struct {
// This value will be behind against actual tunnel utilization in the hot path.
// This should only be used by the ConnectionManagers ticker routine.
lastUsed time.Time
// lanes holds this tunnel's multiport lane sessions. Allocated when the
// handshake completes if both sides advertised multiport, nil otherwise.
// Immutable once the hostinfo is published to the data plane.
lanes *laneSet
}
type ViaSender struct {
@@ -288,6 +290,11 @@ type ViaSender struct {
relayHI *HostInfo // relayHI is the host info object of the relay
relay *Relay // relay contains the rest of the relay information, including the PeerIP of the host trying to communicate with us.
IsRelayed bool // IsRelayed is true if the packet was sent through a relay
// SockIdx is the local socket (Interface.writers index) the packet
// arrived on. Replies that must originate from the same 4-tuple egress
// f.writers[SockIdx].
SockIdx int
}
func (v ViaSender) String() string {
@@ -477,6 +484,9 @@ func (hm *HostMap) unlockedMakePrimary(hostinfo *HostInfo) bool {
// any tunnel to the peer), which the caller uses to decide whether to clear learned lighthouse
// state and disestablish relays.
func (hm *HostMap) unlockedDeleteHostInfo(hostinfo *HostInfo) bool {
// Lane sessions hang off this hostinfo, so deleting it takes them with it
// and there is nothing extra to unwind here.
// Remove this hostinfo from each of its address lists. The lists are independent, so a
// sibling is never promoted to an address it does not own and no other list is touched.
final := true
@@ -933,3 +943,59 @@ func (i *HostInfo) logger(l *slog.Logger) *slog.Logger {
return li
}
// Utility functions
func localAddrs(l *slog.Logger, allowList *LocalAllowList) []netip.Addr {
//FIXME: This function is pretty garbage
var finalAddrs []netip.Addr
ifaces, _ := net.Interfaces()
for _, i := range ifaces {
allow := allowList.AllowName(i.Name)
if l.Enabled(context.Background(), logging.LevelTrace) {
l.Log(context.Background(), logging.LevelTrace, "localAllowList.AllowName",
"interfaceName", i.Name,
"allow", allow,
)
}
if !allow {
continue
}
addrs, _ := i.Addrs()
for _, rawAddr := range addrs {
var addr netip.Addr
switch v := rawAddr.(type) {
case *net.IPNet:
//continue
addr, _ = netip.AddrFromSlice(v.IP)
case *net.IPAddr:
addr, _ = netip.AddrFromSlice(v.IP)
}
if !addr.IsValid() {
if l.Enabled(context.Background(), slog.LevelDebug) {
l.Debug("addr was invalid", "localAddr", rawAddr)
}
continue
}
addr = addr.Unmap()
if addr.IsLoopback() == false && addr.IsLinkLocalUnicast() == false {
isAllowed := allowList.Allow(addr)
if l.Enabled(context.Background(), logging.LevelTrace) {
l.Log(context.Background(), logging.LevelTrace, "localAllowList.Allow",
"localAddr", addr,
"allowed", isAllowed,
)
}
if !isAllowed {
continue
}
finalAddrs = append(finalAddrs, addr)
}
}
}
return finalAddrs
}
+84 -44
View File
@@ -10,14 +10,12 @@ import (
"github.com/slackhq/nebula/firewall"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/iputil"
"github.com/slackhq/nebula/logging"
"github.com/slackhq/nebula/noiseutil"
"github.com/slackhq/nebula/overlay/batch"
"github.com/slackhq/nebula/overlay/tio"
"github.com/slackhq/nebula/routing"
)
func (f *Interface) consumeInsidePacket(pkt tio.Packet, fwPacket *firewall.ParsedPacket, nb []byte, sendBatch *batch.SendBatch, rejectBuf []byte, q int, localCache firewall.ConntrackCache) {
func (f *Interface) consumeInsidePacket(pkt tio.Packet, fwPacket *firewall.ParsedPacket, nb []byte, tx *txQueue, rejectBuf []byte, q int, localCache firewall.ConntrackCache) {
// borrowed: pkt.Bytes is owned by the originating tio.Queue and is
// only valid until the next Read on that queue. Every consumer below
// (parse, self-forward, handshake cache, sendInsideMessage) reads it
@@ -112,7 +110,7 @@ func (f *Interface) consumeInsidePacket(pkt tio.Packet, fwPacket *firewall.Parse
dropReason := f.firewall.Drop(fwPacket.Packet, false, hostinfo, f.pki.GetCAPool(), localCache)
if dropReason == nil {
f.sendInsideMessage(hostinfo, pkt, nb, sendBatch)
f.sendInsideMessage(hostinfo, pkt, &fwPacket.Packet, nb, tx)
} else {
f.rejectInside(packet, rejectBuf, q)
if f.l.Enabled(context.Background(), slog.LevelDebug) {
@@ -124,13 +122,13 @@ func (f *Interface) consumeInsidePacket(pkt tio.Packet, fwPacket *firewall.Parse
}
}
func (f *Interface) sendInsideEncrypt(hostinfo *HostInfo, ci *ConnectionState, seg, scratch, nb []byte) []byte {
func (f *Interface) sendInsideEncrypt(hostinfo *HostInfo, ci *ConnectionState, lane uint8, seg, scratch, nb []byte) []byte {
if noiseutil.EncryptLockNeeded {
ci.writeLock.Lock()
}
c := ci.messageCounter.Add(1)
out := header.Encode(scratch, header.Version, header.Message, 0, hostinfo.remoteIndexId, c)
out := header.EncodeLane(scratch, header.Version, header.Message, 0, hostinfo.remoteIndexId, c, lane)
out, encErr := ci.eKey.EncryptDanger(out, out, seg, c, nb)
if noiseutil.EncryptLockNeeded {
@@ -155,12 +153,19 @@ func (f *Interface) sendInsideEncrypt(hostinfo *HostInfo, ci *ConnectionState, s
// kernel-supplied superpacket bytes never get written into a separate
// scratch arena: SegmentSuperpacket builds each segment's plaintext in
// segScratch[:segLen] in turn, and we encrypt directly into a fresh SendBatch slot.
func (f *Interface) sendInsideMessage(hostinfo *HostInfo, pkt tio.Packet, nb []byte, sendBatch *batch.SendBatch) {
//
// When this flow has a usable multiport lane to this peer, the direct path swaps
// to that lane's session and socket below. Relay and base traffic stays on
// tx.base (socket 0).
func (f *Interface) sendInsideMessage(hostinfo *HostInfo, pkt tio.Packet, fwPacket *firewall.Packet, nb []byte, tx *txQueue) {
ci := hostinfo.ConnectionState
if ci.eKey == nil {
return
}
// Base and relay traffic stays on socket 0; the direct path may swap to tx.lane below.
sendBatch := tx.base
// One traffic-out mark covers every segment of the superpacket; doing it
// per segment in sendInsideEncrypt paid an atomic store up to ~45 extra
// times per TSO packet, inside writeLock under boring crypto.
@@ -199,10 +204,10 @@ func (f *Interface) sendInsideMessage(hostinfo *HostInfo, pkt tio.Packet, nb []b
}
err = tio.SegmentSuperpacket(pkt, func(seg []byte) error {
//relay header + header + plaintext + AEAD tag + relay tag
scratch := sendBatch.Reserve(len(seg) + header.MaxOverhead)
//relay header + header + plaintext + AEAD tag (16 bytes for both AES-GCM and ChaCha20-Poly1305) + relay tag
scratch := sendBatch.Reserve(header.Len + header.Len + len(seg) + 16 + 16)
innerPacket := f.sendInsideEncrypt(hostinfo, ci, seg, scratch[header.Len:], nb)
innerPacket := f.sendInsideEncrypt(hostinfo, ci, 0, seg, scratch[header.Len:], nb)
if innerPacket == nil {
return nil
}
@@ -223,11 +228,28 @@ func (f *Interface) sendInsideMessage(hostinfo *HostInfo, pkt tio.Packet, nb []b
return
}
// Direct path: prefer this flow's multiport lane once it is proven usable.
// txLaneForFlow hands back the lane's session and destination together, so
// there is no window where one is set and the other is not, and a demotion
// drops us back onto the base tunnel on the very next packet.
//
// A miss is also how a lane gets re-probed after a demotion: txLane raises
// demand, which the connection manager's next tick on this tunnel picks up.
// Until the lane is up the traffic rides the base tunnel, the same fallback
// a demoted lane uses.
lane := uint8(0)
if s, lci, laneRemote := hostinfo.lanes.txLaneForFlow(fwPacket); lci != nil {
lane = uint8(s)
ci = lci
remote = laneRemote
sendBatch = tx.laneBatch(f, s)
}
err := tio.SegmentSuperpacket(pkt, func(seg []byte) error {
// header + plaintext + AEAD tag (16 bytes for both AES-GCM and ChaCha20-Poly1305)
scratch := sendBatch.Reserve(header.Len + len(seg) + 16)
out := f.sendInsideEncrypt(hostinfo, ci, seg, scratch, nb)
out := f.sendInsideEncrypt(hostinfo, ci, lane, seg, scratch, nb)
if out == nil {
return nil
}
@@ -516,39 +538,61 @@ func (f *Interface) SendVia(via *HostInfo, relay *Relay, ad, nb, out []byte, noc
return
}
err = f.writers[q].WriteTo(toSend, via.GetRemote())
err = f.writers[f.egressSock(q)].WriteTo(toSend, via.GetRemote())
if err != nil {
if l, ok := logging.RateLimited(f.l, f.udpRelayTxErrLimiter); ok {
via.logger(l).Info("Failed to WriteTo in sendVia", "error", err)
}
via.logger(f.l).Info("Failed to WriteTo in sendVia", "error", err)
}
}
// egressSock picks the socket a tunnel packet leaves from.
//
// Everything that is not lane data plane leaves from the base port: handshakes, keepalives, close packets, rejects and
// relay carriers all belong to the base tunnel's 4-tuple, which is the only one a peer's spoof/roam checks and a
// vanilla peer's expectations know about. Lane data goes through laneSock instead and never comes here.
//
// Which socket on the base port doesn't matter — they share an address, so they produce identical packets — so keep to
// this routine's own share of the group and leave the rest of it uncontended. Without multiport that is q itself, since
// every socket is on the base port.
func (f *Interface) egressSock(q int) int {
return f.laneSock(q, 0)
}
// laneSock returns the index in writers of a socket bound to lane s's port, for a
// routine that reads queue q.
//
// Under multiport the sockets are laid out port-major — writers[s*routinesPerPort
// + r] is the r'th socket on port listen.port+s — so every routine has a sibling
// socket on every port and the arithmetic is a lane index away. Routines pick the
// sibling matching their own position in their group, which spreads the writers
// for one port over that port's whole group rather than funnelling them onto its
// first socket. It is a pure function of (q, s), so a flow always leaves from the
// same socket and cannot reorder itself across two of them.
//
// Without multiport there is one port and every socket is on it, so any lane
// resolves to q's own socket.
func (f *Interface) laneSock(q, s int) int {
if !f.multiport {
return q
}
return s*f.routinesPerPort + q%f.routinesPerPort
}
func (f *Interface) sendNoMetrics(t header.MessageType, st header.MessageSubType, ci *ConnectionState, hostinfo *HostInfo, remote netip.AddrPort, p, nb, out []byte, q int) {
if ci.eKey == nil {
return
}
q = f.egressSock(q)
useRelay := !remote.IsValid() && !hostinfo.GetRemote().IsValid()
// The encrypt appends to out in place and a relayed packet is then wrapped in place,
// so out has to fit the whole packet up front or the encrypt reallocates out from under SendVia.
// We cap the MTU of tun devices and routes we create ourselves, so only a host that set a bigger MTU than we allow gets here.
need := header.Len + len(p) + ci.eKey.Overhead()
if useRelay {
// The relay tunnel runs the same cipher, and adds its own header and tag
need += header.Len + ci.eKey.Overhead()
}
if cap(out) < need {
if l, ok := logging.RateLimited(f.l, f.txOversizeLimiter); ok {
hostinfo.logger(l).Error("Dropping outbound packet, out buffer not large enough", "outCap", cap(out), "payloadLen", len(p), "relayed", useRelay)
}
return
}
fullOut := out
if useRelay {
if len(out) < header.Len {
// out always has a capacity of mtu, but not always a length greater than the header.Len.
// Grow it to make sure the next operation works.
out = out[:header.Len]
}
// Save a header's worth of data at the front of the 'out' buffer.
out = out[header.Len:header.Len]
out = out[header.Len:]
}
if noiseutil.EncryptLockNeeded {
@@ -595,22 +639,18 @@ func (f *Interface) sendNoMetrics(t header.MessageType, st header.MessageSubType
if remote.IsValid() {
err = f.writers[q].WriteTo(out, remote)
if err != nil {
if l, ok := logging.RateLimited(f.l, f.udpTxErrLimiter); ok {
hostinfo.logger(l).Error("Failed to write outgoing packet",
"error", err,
"udpAddr", remote,
)
}
hostinfo.logger(f.l).Error("Failed to write outgoing packet",
"error", err,
"udpAddr", remote,
)
}
} else if hr := hostinfo.GetRemote(); hr.IsValid() {
err = f.writers[q].WriteTo(out, hr)
if err != nil {
if l, ok := logging.RateLimited(f.l, f.udpTxErrLimiter); ok {
hostinfo.logger(l).Error("Failed to write outgoing packet",
"error", err,
"udpAddr", hr,
)
}
hostinfo.logger(f.l).Error("Failed to write outgoing packet",
"error", err,
"udpAddr", hr,
)
}
} else {
// Try to send via a relay
+175 -36
View File
@@ -24,7 +24,6 @@ import (
"github.com/slackhq/nebula/overlay"
"github.com/slackhq/nebula/overlay/batch"
"github.com/slackhq/nebula/overlay/tio"
"github.com/slackhq/nebula/ratelimit"
"github.com/slackhq/nebula/udp"
)
@@ -44,10 +43,22 @@ type InterfaceConfig struct {
DropLocalBroadcast bool
DropMulticast bool
routines int
MessageMetrics *MessageMetrics
version string
relayManager *relayManager
punchy *Punchy
// Multiport means the sockets are spread over a range of ports
// (listen.port+slot) rather than all sharing listen.port, and that lane
// tunnels are negotiated with capable peers.
Multiport bool
// RoutinesPerPort is how many sockets share each port under multiport, and so
// the stride between port slots in writers: writers[s*RoutinesPerPort+r] is
// the r'th socket bound to listen.port+s. It is `routines` as configured,
// while routines above is that times the number of ports.
RoutinesPerPort int
// LaneCount is the number of lanes counting the base tunnel as lane 0
// (multiport.lanes, clamped to the number of ports bound).
LaneCount int
MessageMetrics *MessageMetrics
version string
relayManager *relayManager
punchy *Punchy
tryPromoteEvery uint32
reQueryEvery uint32
@@ -88,6 +99,9 @@ type Interface struct {
dropLocalBroadcast bool
dropMulticast bool
routines int
multiport bool
routinesPerPort int
laneCount int
disconnectInvalid atomic.Bool
closed atomic.Bool
// cpuAffinity, when non-empty, names the CPUs each TUN reader goroutine
@@ -125,10 +139,6 @@ type Interface struct {
batchers []*batch.MultiCoalescer
wg sync.WaitGroup
udpTxErrLimiter *ratelimit.Limiter // udp direct transmit error log limiter
udpRelayTxErrLimiter *ratelimit.Limiter // udp relay transmit error log limiter
txOversizeLimiter *ratelimit.Limiter // oversized outbound packet drop log limiter
// fatalErr holds the first unexpected reader error that caused shutdown.
// nil means "no fatal error" (yet)
fatalErr atomic.Pointer[error]
@@ -203,19 +213,6 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
return nil, errors.New("no connection manager")
}
udpTxErrLimiter, err := ratelimit.New(1, time.Second)
if err != nil {
return nil, err
}
udpRelayTxErrLimiter, err := ratelimit.New(1, time.Second)
if err != nil {
return nil, err
}
txOversizeLimiter, err := ratelimit.New(1, time.Second)
if err != nil {
return nil, err
}
if c.routines <= 1 {
c.PinThreads = false //pinning is not useful unless there's more than one tun reader
}
@@ -235,6 +232,9 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
dropLocalBroadcast: c.DropLocalBroadcast,
dropMulticast: c.DropMulticast,
routines: c.routines,
multiport: c.Multiport,
routinesPerPort: max(c.RoutinesPerPort, 1),
laneCount: c.LaneCount,
version: c.version,
writers: make([]udp.Conn, c.routines),
batchers: make([]*batch.MultiCoalescer, c.routines),
@@ -248,9 +248,6 @@ func NewInterface(ctx context.Context, c *InterfaceConfig) (*Interface, error) {
conntrackCacheTimeout: c.ConntrackCacheTimeout,
cpuAffinity: c.CpuAffinity,
pinThreads: c.PinThreads,
udpTxErrLimiter: udpTxErrLimiter,
udpRelayTxErrLimiter: udpRelayTxErrLimiter,
txOversizeLimiter: txOversizeLimiter,
metricHandshakes: metrics.GetOrRegisterHistogram("handshakes", nil, metrics.NewExpDecaySample(1028, 0.015)),
metricTxDropped: metrics.GetOrRegisterCounter("udp.tx.dropped", nil),
@@ -297,7 +294,10 @@ func (f *Interface) activate() error {
"fips140Enforced", fips140.Enforced(),
)
if f.routines > 1 && !f.outside.SupportsMultipleReaders() {
// Under multiport each socket has exactly one reader on its own port, so
// the shared-port multi-reader capability is irrelevant (and main.go
// already hard-errored on unsupported platforms).
if f.routines > 1 && !f.multiport && !f.outside.SupportsMultipleReaders() {
f.routines = 1
f.l.Warn("multiple udp readers are not supported on this platform, falling back to a single routine")
}
@@ -310,6 +310,11 @@ func (f *Interface) activate() error {
return err
}
if len(queues) < f.routines {
if f.multiport {
// The lane sockets are already bound one-per-routine; shrinking
// the routine count would leave bound ports with no reader.
return fmt.Errorf("multiport requires %d tun queues, device provided %d", f.routines, len(queues))
}
// TODO: this clamp is only safe because it is unreachable when the
// udp side has multiple readers (linux Queues opens exactly n or
// errors; every other platform already clamped routines to 1 above).
@@ -410,7 +415,7 @@ func (f *Interface) listenOut(i int) {
rxc := newRxContext(f, i)
listener := func(fromUdpAddr netip.AddrPort, payload []byte) {
f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr}, payload, rxc)
f.readOutsidePackets(ViaSender{UdpAddr: fromUdpAddr, SockIdx: i}, payload, rxc)
}
flusher := func() {
@@ -452,6 +457,114 @@ func (f *Interface) pinThisThread(i int) {
}
}
// txQueue is the per-routine TX state owned by one listenIn goroutine.
//
// base carries base-session data, relay carriers, and everything on a tunnel
// without lanes. It goes out a socket on the base port — egressSock's pick — since
// base traffic must keep the base source port or a vanilla peer would see it move
// and roam-thrash. lane[s] goes out a socket on listen.port+s and carries traffic
// encrypted with lane s's session; lane[0] is base, and the rest are built on the
// first packet that picks them, since a routine that never sends on a lane should
// not hold a batch for it. Both come from laneSock, so a routine writes to its own
// share of each port's socket group.
//
// Which lane a packet rides comes from its own flow hash, not from this
// routine's index. That is deliberate. Which routine reads a flow is the
// kernel's decision: it hashes the flow to a tun queue, but it also *learns*
// the queue we write that flow's inbound packets to, and prefers what it
// learned. So if the lane followed the routine, a peer whose lanes were still
// down — every peer, for the first moments of a tunnel — would write all of its
// inbound traffic to queue 0, teaching both kernels to steer every flow to
// queue 0, and every tunnel would collapse onto lane 0 and stay there for as
// long as its flows kept busy. Hashing here makes lane spread independent of
// tun steering entirely.
//
// Every batch borrows arena, so a routine holding a batch per lane still costs
// one slab. The arena is reset by flush once every batch over it is drained.
//
// Several routines can still write to one socket — the sockets on a port are
// shared by the routines whose lane arithmetic lands on them — which the underlay
// serializes (see batchWriter). Per-flow wire order still holds: a flow is hashed
// onto one lane and read by one routine, so nothing else is writing it.
type txQueue struct {
// q is the queue this state belongs to, which laneSock needs to resolve a lane
// to one of its port's sockets.
q int
base *batch.SendBatch
lane []*batch.SendBatch
arena *batch.Arena
// live is every batch built so far, in build order, so base is first: see
// flush. Kept as its own slice because lane is mostly nil holes and both
// full and flush walk this per read batch.
live []txBatch
}
// txBatch is a live batch and the index in writers of the socket it flushes to.
type txBatch struct {
sb *batch.SendBatch
sock int
}
func (f *Interface) newTxQueue(q int) *txQueue {
baseSock := f.egressSock(q)
arena := batch.NewArena(batch.SendBatchCap * (udp.MTU + 32))
base := batch.NewSendBatchSharedArena(f.writers[baseSock], batch.SendBatchCap, arena)
tx := &txQueue{
q: q,
base: base,
arena: arena,
live: []txBatch{{sb: base, sock: baseSock}},
}
if f.multiport && f.laneCount > 1 {
tx.lane = make([]*batch.SendBatch, f.laneCount)
tx.lane[0] = base
}
return tx
}
// laneBatch returns the batch for lane s, building it the first time this
// routine sends on that lane. Lanes this queue doesn't cover fall back to base,
// which is also lane 0's batch.
func (tx *txQueue) laneBatch(f *Interface, s int) *batch.SendBatch {
if s <= 0 || s >= len(tx.lane) {
return tx.base
}
sb := tx.lane[s]
if sb == nil {
sock := f.laneSock(tx.q, s)
sb = batch.NewSendBatchSharedArena(f.writers[sock], batch.SendBatchCap, tx.arena)
tx.lane[s] = sb
tx.live = append(tx.live, txBatch{sb: sb, sock: sock})
}
return sb
}
// full reports a full sendmmsg worth of work queued across every lane, rather
// than on any one of them: the arena is shared, so it is the total that bounds
// how much is outstanding.
func (tx *txQueue) full() bool {
n := 0
for _, b := range tx.live {
n += b.sb.Len()
}
return n >= batch.SendBatchCap
}
// flush drains base before the lanes so that when a flow moves from the base
// session onto a freshly promoted lane mid-window, its packets still leave this
// host in encryption order. Resetting the shared arena is this queue's job,
// since no single batch's Flush can know the others are done with it.
func (tx *txQueue) flush(f *Interface) {
for _, b := range tx.live {
if b.sb.Len() > 0 {
f.flushSendBatch(b.sb, b.sock)
}
}
tx.arena.Reset()
}
func (f *Interface) listenIn(queue tio.Queue, i int) {
// Pinning this thread (and goroutine) to a single CPU keeps every sendmmsg from this goroutine going through the
// same TX ring on the nic, so the wire sees per-flow order. Skip entirely when tun.pin_threads is false.
@@ -460,8 +573,7 @@ func (f *Interface) listenIn(queue tio.Queue, i int) {
}
rejectBuf := make([]byte, mtu)
arenaSize := batch.SendBatchCap * (udp.MTU + 32)
sb := batch.NewSendBatch(f.writers[i], batch.SendBatchCap, arenaSize)
tx := f.newTxQueue(i)
fwPacket := &firewall.ParsedPacket{}
nb := make([]byte, 12, 12)
@@ -479,15 +591,15 @@ func (f *Interface) listenIn(queue tio.Queue, i int) {
}
for _, pkt := range pkts {
f.consumeInsidePacket(pkt, fwPacket, nb, sb, rejectBuf, i, conntrackCache.Get())
f.consumeInsidePacket(pkt, fwPacket, nb, tx, rejectBuf, i, conntrackCache.Get())
// Flush incrementally once a full sendmmsg batch has
// accumulated so the first packets of a deep read drain
// hit the wire while the rest are still being encrypted.
if sb.Len() >= batch.SendBatchCap {
f.flushSendBatch(sb, i)
if tx.full() {
tx.flush(f)
}
}
f.flushSendBatch(sb, i)
tx.flush(f)
}
f.l.Debug("overlay reader is done", "reader", i)
@@ -645,19 +757,33 @@ func (f *Interface) reloadMisc(c *config.C) {
}
}
// statsEmitter updates the interface's gauges, the stats server runs it on its capture loop.
func (f *Interface) statsEmitter() func() {
func (f *Interface) emitStats(ctx context.Context, i time.Duration) {
ticker := time.NewTicker(i)
defer ticker.Stop()
udpStats := udp.NewUDPStatsEmitter(f.writers)
certExpirationGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
certInitiatingVersion := metrics.GetOrRegisterGauge("certificate.initiating_version", nil)
certMaxVersion := metrics.GetOrRegisterGauge("certificate.max_version", nil)
return func() {
// Registered only when we run multiport, so these don't sit at zero on a node
// that was never going to have a lane and read as a broken feature.
var lanesUpGauge, laneTunnelsGauge metrics.Gauge
if f.multiport && f.laneCount > 1 {
lanesUpGauge = metrics.GetOrRegisterGauge("multiport.lanes.up", nil)
laneTunnelsGauge = metrics.GetOrRegisterGauge("multiport.lanes.tunnels", nil)
}
emit := func() {
f.firewall.EmitStats()
f.handshakeManager.EmitStats()
udpStats()
if lanesUpGauge != nil {
f.emitLaneStats(lanesUpGauge, laneTunnelsGauge)
}
certState := f.pki.getCertState()
defaultCrt := certState.GetDefaultCertificate()
certExpirationGauge.Update(int64(defaultCrt.NotAfter().Sub(time.Now()) / time.Second))
@@ -670,6 +796,19 @@ func (f *Interface) statsEmitter() func() {
certMaxVersion.Update(int64(certState.v1Cert.Version()))
}
}
// Prime gauges so a Prometheus scrape that lands before the first tick
// sees real values instead of the zero defaults (issue #907).
emit()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
emit()
}
}
}
func (f *Interface) GetHostInfo(vpnIp netip.Addr) *HostInfo {
+14 -5
View File
@@ -3,6 +3,7 @@
package nebula
import (
"context"
"net/netip"
"testing"
"time"
@@ -17,8 +18,11 @@ import (
"github.com/stretchr/testify/require"
)
// Priming before the first tick (issue #907) is Start's job, see TestStatsServer_Start_primes.
func Test_statsEmitter_setsGauges(t *testing.T) {
// Test_emitStats_primesGauges covers issue #907: a Prometheus scrape that
// landed before the first ticker fire used to read 0 for the cert gauges.
// emitStats now primes the gauges before entering the ticker loop. We assert
// the gauge is zero before the first call and non-zero after.
func Test_emitStats_primesGauges(t *testing.T) {
defer metrics.DefaultRegistry.UnregisterAll()
l := test.NewLogger()
@@ -52,12 +56,17 @@ func Test_statsEmitter_setsGauges(t *testing.T) {
ifce.pki.cs.Store(cs)
ttlGauge := metrics.GetOrRegisterGauge("certificate.ttl_seconds", nil)
require.Zero(t, ttlGauge.Value(), "gauge should be zero before the emitter runs")
require.Zero(t, ttlGauge.Value(), "gauge should be zero before emitStats runs")
ifce.statsEmitter()()
// Pre-cancel the context so emitStats returns after priming the gauges
// without ever reading from ticker.C. The one hour interval is just a
// belt-and-suspenders, the test does not expect the ticker to fire.
ctx, cancel := context.WithCancel(context.Background())
cancel()
ifce.emitStats(ctx, time.Hour)
ttl := ttlGauge.Value()
assert.Positive(t, ttl, "ttl gauge should be set by one run of the emitter")
assert.Positive(t, ttl, "ttl gauge should be primed by emitStats before its first tick")
assert.LessOrEqual(t, ttl, int64(3600))
assert.Equal(t, int64(cert.Version1), metrics.GetOrRegisterGauge("certificate.initiating_version", nil).Value())
assert.Equal(t, int64(cert.Version1), metrics.GetOrRegisterGauge("certificate.max_version", nil).Value())
+13 -3
View File
@@ -45,7 +45,8 @@ func setTransportChecksum4(packet []byte) {
if ihl < ipv4.HeaderLen || end < ihl || end > len(packet) {
return
}
// do not attempt to set checksums for fragments
// The checksum covers the whole datagram, which a fragment (MF set or a
// non-zero offset) does not carry.
if binary.BigEndian.Uint16(packet[6:8])&0x3fff != 0 {
return
}
@@ -67,7 +68,10 @@ func setTransportChecksum6(packet []byte) {
return
}
// we must find the transport header in order to write a checksum
// The checksum covers the whole datagram, which a fragment does not carry.
// An unknown extension header hides where the transport header starts. A
// chain longer than the walk's budget ends it early, at an offset that was
// never checked against the packet.
proto, offset, _, anyFragment, err := IPv6FindUpperProtocol(packet[:end])
if err != nil || anyFragment || offset >= end {
return
@@ -105,7 +109,8 @@ func transportExtent(transport []byte, proto uint8) ([]byte, bool) {
// writeTransportChecksum stores the checksum of transport, taken over the
// pseudo-header sum csum, in the header's checksum field. A UDP checksum that
// computes to zero goes on the wire as 0xffff: zero means no checksum was
// computed (RFC 768), and over IPv6 the checksum is mandatory (RFC 8200 section 8.1).
// computed (RFC 768), and over IPv6 the checksum is mandatory (RFC 8200
// section 8.1).
func writeTransportChecksum(transport []byte, proto uint8, csum uint32) {
var at, minLen int
switch proto {
@@ -128,6 +133,11 @@ func writeTransportChecksum(transport []byte, proto uint8, csum uint32) {
binary.BigEndian.PutUint16(transport[at:], sum)
}
// fold reduces a pseudo-header sum to the 16 bit seed Checksum takes. Carrying
// the high half back into the low half is what keeps the reduction lossless, so
// the seed sums exactly as the wider value would; 0xffff is its fixed point.
// Every term of that sum comes from a 16 bit field, so it stays far below the
// width at which the accumulator would wrap.
func fold(csum uint32) uint16 {
for csum > 0xffff {
csum = (csum >> 16) + (csum & 0xffff)
+16 -13
View File
@@ -345,11 +345,14 @@ func ipv6CreateRejectTCPPacket(packet []byte, out []byte, offset int) []byte {
return out
}
// IPv6FindUpperProtocol walks the ipv6 extension header chain and returns the upper layer protocol, the offset it begins at, and whether the packet is a non-first fragment.
// Only the RFC 8200 and IANA extension headers below are walked.
// Everything else, including Mobility (135), HIP (139), Shim6 (140), experimental 253/254, and real upper layer protocols like SCTP or GRE, is terminal.
// For a non-first fragment the returned protocol is the fragmented protocol and offset points at the fragment header, there is no transport header to locate.
// Returns ErrIPv6CouldNotFindPayload if packet is smaller than an ipv6 header, or the chain is truncated, or if more than 8 extension headers are present.
// IPv6FindUpperProtocol walks the ipv6 extension header chain and returns the upper layer protocol, the
// offset it begins at, and whether the packet is a non-first fragment. Only the RFC 8200 and IANA extension
// headers below are walked. Everything else, including Mobility (135), HIP (139), Shim6 (140), experimental
// 253/254, and real upper layer protocols like SCTP or GRE, is terminal. Walking those as extension headers
// is a firewall bypass, so they fail closed. For a non-first fragment the returned protocol is the fragmented
// protocol and offset points at the fragment header, there is no transport header to locate. Returns
// ErrIPv6CouldNotFindPayload if packet is smaller than an ipv6 header or the chain is truncated before a
// terminal protocol is reached.
func IPv6FindUpperProtocol(packet []byte) (nextHeader uint8, offset int, isFragment bool, anyFragment bool, err error) {
const maxIPv6ExtHeaders = 8
if len(packet) < ipv6.HeaderLen {
@@ -362,14 +365,14 @@ func IPv6FindUpperProtocol(packet []byte) (nextHeader uint8, offset int, isFragm
switch nextHeader {
case 0, 43, 60: // Hop-by-Hop, Routing, Destination
if len(packet) < offset+2 {
return 0, 0, false, false, ErrIPv6CouldNotFindPayload
return nextHeader, offset, isFragment, anyFragment, ErrIPv6CouldNotFindPayload
}
nextHeader = packet[offset]
offset += (int(packet[offset+1]) + 1) << 3
case 44: // Fragment
if len(packet) < offset+8 {
return 0, 0, false, false, ErrIPv6CouldNotFindPayload
return nextHeader, offset, isFragment, anyFragment, ErrIPv6CouldNotFindPayload
}
anyFragment = true
// Non-first fragments carry no transport header, report the fragmented protocol and stop
@@ -381,21 +384,21 @@ func IPv6FindUpperProtocol(packet []byte) (nextHeader uint8, offset int, isFragm
case 51: // AH
if len(packet) < offset+2 {
return 0, 0, false, false, ErrIPv6CouldNotFindPayload
return nextHeader, offset, isFragment, anyFragment, ErrIPv6CouldNotFindPayload
}
nextHeader = packet[offset]
offset += (int(packet[offset+1]) + 2) << 2
default:
// A prior extension header can declare a length that advances offset past the packet. The terminal
// protocol's header isn't actually here, so treat the chain as truncated rather than classifying it.
if offset > len(packet) {
// A prior extension header can declare a length that advances offset past the packet, reject this.
return 0, 0, false, false, ErrIPv6CouldNotFindPayload
return nextHeader, offset, isFragment, anyFragment, ErrIPv6CouldNotFindPayload
}
return nextHeader, offset, isFragment, anyFragment, nil //success!
return nextHeader, offset, isFragment, anyFragment, nil
}
}
// Fail closed
return 0, 0, false, false, ErrIPv6CouldNotFindPayload
return nextHeader, offset, isFragment, anyFragment, nil
}
func CreateICMPEchoResponse(packet, out []byte) []byte {
+5 -17
View File
@@ -529,17 +529,6 @@ func Test_IPv6FindUpperProtocol(t *testing.T) {
firstFragToUDP := []byte{17, 0, 0, 1, 0, 0, 0, 1} // frag offset 0, M=1, next = UDP
nonFirstFrag := []byte{17, 0, 0, 9, 0, 0, 0, 1} // frag offset non-zero, next = UDP
transport := []byte{0, 80, 1, 187, 0, 0, 0, 0} // stand-in bytes, IPv6FindUpperProtocol never reads ports
extToHbH := []byte{0, 0, 0, 0, 0, 0, 0, 0} // len 0 -> 8 bytes, next = Hop-by-Hop
// 8 walked headers spend the budget before the 9th slot's terminal is ever classified. This is the
// report 4040546 shape: without failing closed the walk hands back offset 104, and a caller reads ports
// from the attacker's bytes there.
var budgetChain []byte
for range 7 {
budgetChain = append(budgetChain, extToHbH...)
}
budgetChain = append(budgetChain, extToTCP...)
budgetChain = append(budgetChain, transport...)
tests := []struct {
name string
@@ -560,10 +549,9 @@ func Test_IPv6FindUpperProtocol(t *testing.T) {
{"first fragment walks to transport", 44, append(firstFragToUDP, transport...), 17, ipv6.HeaderLen + 8, false, true, nil},
{"non-first fragment stops", 44, append(nonFirstFrag, transport...), 17, ipv6.HeaderLen, true, true, nil},
{"unknown protocol is terminal", 132, transport, 132, ipv6.HeaderLen, false, false, nil}, // SCTP
{"truncated extension header", 0, nil, 0, 0, false, false, ErrIPv6CouldNotFindPayload},
{"truncated extension header", 0, nil, 0, ipv6.HeaderLen, false, false, ErrIPv6CouldNotFindPayload},
// Destination Options with a declared length (255+1)*8 = 2048 that runs past the 48 byte buffer, next = SCTP
{"extension length past buffer", 0, []byte{132, 255, 0, 0, 0, 0, 0, 0}, 0, 0, false, false, ErrIPv6CouldNotFindPayload},
{"budget exhausted fails closed", 0, budgetChain, 0, 0, false, false, ErrIPv6CouldNotFindPayload},
{"extension length past buffer", 60, []byte{132, 255, 0, 0, 0, 0, 0, 0}, 132, ipv6.HeaderLen + 2048, false, false, ErrIPv6CouldNotFindPayload},
}
for _, tt := range tests {
@@ -571,10 +559,10 @@ func Test_IPv6FindUpperProtocol(t *testing.T) {
packet := makeIPv6Packet(src, dst, tt.nextHeader, tt.payload)
proto, offset, isFragment, anyFragment, err := IPv6FindUpperProtocol(packet)
if tt.wantErr != nil {
require.ErrorIs(t, err, tt.wantErr)
} else {
require.NoError(t, err)
assert.ErrorIs(t, err, tt.wantErr)
return
}
require.NoError(t, err)
assert.Equal(t, tt.wantProto, proto)
assert.Equal(t, tt.wantOffset, offset)
assert.Equal(t, tt.wantFragment, isFragment)
+721
View File
@@ -0,0 +1,721 @@
package nebula
import (
"context"
"hash/fnv"
"log/slog"
"net/netip"
"sync"
"sync/atomic"
"time"
"github.com/flynn/noise"
"github.com/rcrowley/go-metrics"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/firewall"
"github.com/slackhq/nebula/handshake"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/noiseutil"
)
// Multiport lanes give one tunnel several underlay 5-tuples, so its traffic
// spreads over ECMP paths, NIC receive queues and per-flow policers instead of
// funnelling through a single flow. Each inside flow picks a lane by hashing its
// own 5-tuple, so the spread doesn't depend on how a kernel steers tun queues
// (see txQueue).
//
// A lane is not a second tunnel: it is an extra session on the same HostInfo.
// Noise leaves us with A.eKey == B.dKey, so both sides expand the same two keys
// with the same per-lane label and land on a matched pair without exchanging
// anything. A lane therefore costs no handshake, has no half-established state,
// and dies exactly when its base tunnel does. Which lane a packet belongs to
// travels in the nebula header, inside the AEAD's associated data.
//
// Lane 0 is the base tunnel itself: HostInfo.ConnectionState, the base port, and
// the peer's real remote address, and it carries its share of flows like any
// other. Lane s > 0 egresses a socket on listen.port+s (see laneSock) toward the
// peer's advertised port range. Receiving on a lane
// needs no permission — the keys are derivable the moment the base handshake
// completes — but sending on one needs proof the new 5-tuple actually works,
// since nothing else would notice a middlebox quietly dropping it. So a lane
// stays down until a probe on it is acked, and falls back to the base tunnel the
// moment it stops being acked.
//
// Both directions are pay-per-use. A lane session is derived on the first packet
// that needs it, because how many lanes exist is partly the peer's call: it
// advertises how many it sends on, and we have to be able to receive all of
// them. Deriving them all up front would let a peer advertising the maximum cost
// us a replay window and two cipher states per lane, per tunnel, for lanes it
// may never send on.
const (
// laneKeyInfo is the HKDF label prefix for lane key expansion. Changing it
// means older builds derive different keys and drop our lane traffic; the
// base tunnel would keep working, so the failure would be a silent loss of
// lanes rather than of connectivity.
laneKeyInfo = "nebula multiport lane v1"
// laneRetryBase and laneRetryMax bound the backoff between probes of a lane
// that will not come up, so a peer whose lane ports are firewalled costs one
// packet a minute rather than one per traffic tick.
laneRetryBase = 5 * time.Second
laneRetryMax = 60 * time.Second
// laneMaxFails caps the failure counter; the backoff saturates well before.
laneMaxFails = 8
// laneProbeTimeout is how long a probe may go unacked before it counts as a
// failure. It is shorter than the connection manager's check interval on
// purpose: an outstanding probe is judged on the next tick either way, and a
// longer timeout would only delay that by a whole tick.
laneProbeTimeout = 2 * time.Second
// laneKeepalive is how often a lane that is up re-proves its path. Traffic
// on a lane is not evidence the lane works — that is the whole reason lanes
// need probing — so a lane that silently breaks is only caught here.
laneKeepalive = 30 * time.Second
)
// laneSet holds a peer's lane sessions and the state deciding which lanes may
// carry traffic. It is built when the base handshake completes and never
// resized, so the slices and their lengths are immutable; mu guards the fields
// under it, and sessions/txAddr/demand are atomics read by the data plane
// without it.
type laneSet struct {
// sessions[s] holds lane s's session once something has needed it, and nil
// until then. sessions[0] is never populated: lane 0 is the base tunnel's own
// ConnectionState. The length is immutable, so the data plane bounds-checks
// and loads with no locking.
sessions []atomic.Pointer[ConnectionState]
// material is what a lane session is derived from, kept because sessions are
// derived lazily and the handshake result is long gone by then.
material laneMaterial
// txAddr[s] holds lane s's remote address while the lane is proven usable
// and nil otherwise. This single atomic is both the TX gate and the
// destination, so a routine that loads non-nil has everything it needs.
// Sized txLanes: lanes above that never send.
txAddr []atomic.Pointer[netip.AddrPort]
// demand[s] is raised at creation, and again by the TX path whenever a flow
// hashes onto lane s while it is down. Probing is demand-driven, so a peer we
// never send to costs nothing beyond its base tunnel no matter how many lanes
// are configured, and a lane that keeps failing is only retried while
// something still wants it. Sized txLanes.
demand []atomic.Bool
// txLanes is how many lanes we may send on — our lane count clamped to the
// ports the peer bound — and so the modulus a flow's hash is reduced by.
// Lanes from txLanes up can only receive, which is how a peer with more
// routines than us still spreads its own traffic.
// Immutable, and the length of every TX-side slice here.
txLanes int
mu sync.Mutex
// peerPortCount and peerBasePort are the peer's advertised port range and
// portOffset is this pair's rotation within it: lane s targets
// peerBasePort + ((s + portOffset) % peerPortCount).
peerPortCount uint16
peerBasePort uint16
portOffset uint16
// laneBias rotates the flow hash before it picks a lane, so the two sides of
// a flow land on lanes that are each other's partner rather than at
// independent points in the range. See newLaneSet.
laneBias uint16
// peerAddr is the address the current lane targets were built from. The
// peer's lane ports have no derivable relationship to a new NAT mapping, so
// a roam invalidates every lane rather than moving it.
peerAddr netip.Addr
// probe[s] is lane s's probe and backoff state. Sized txLanes.
probe []laneProbeState
}
// laneMaterial is everything a lane session is derived from. The two base keys
// are the same secret the base tunnel's own cipher states already hold — the
// noiseutil.CipherState interface just doesn't hand them back, so a lane set
// keeps its own copy rather than a reference to the session.
type laneMaterial struct {
eKey, dKey [32]byte
cipher noise.CipherFunc
myCert cert.Certificate
peerCert *cert.CachedCertificate
initiator bool
epoch uint64
}
type laneProbeState struct {
// gen is the generation of the last probe sent, echoed in the ack so a late
// ack cannot promote a lane on the strength of a superseded probe.
gen uint8
// fails is the consecutive failure count driving retryAt.
fails uint8
// sentAt is when the outstanding probe went out, zero when none is pending.
sentAt time.Time
// target is where the outstanding probe went, promoted to txAddr on ack. It
// outlives the probe, which is what lets a demotion log the address that
// stopped answering.
target netip.AddrPort
// lastAck is when the lane was last confirmed usable, driving the keepalive.
lastAck time.Time
// retryAt is the earliest we may probe this lane again.
retryAt time.Time
}
// newLaneSet sets up the lanes for a freshly completed base handshake. It
// returns nil when the pair has no lane beyond the base tunnel, which is the
// normal answer for a peer running without multiport. No session is derived
// here; each is derived on the first packet that needs it.
//
// epoch is the base session's, which every lane session will share.
func newLaneSet(r *handshake.Result, myLanes int, epoch uint64, myAddr, peerAddr netip.Addr) *laneSet {
// PeerPortCount and PeerBasePort are already bounded to uint16 by the
// handshake payload parser. A zero port count is a peer that did not
// advertise multiport at all, so there is no lane to be had in either
// direction.
peerPorts := uint16(r.PeerPortCount)
if peerPorts == 0 {
return nil
}
// Sessions have to cover both directions: we send on our lanes and receive
// on the peer's, and one derived session serves both ends of a lane index.
// Only the session table is sized by the peer's advertised count; the TX-side
// state is sized by what we will actually send on.
n := min(max(myLanes, int(r.PeerTxLanes)), header.MaxLane+1)
if n < 2 {
return nil
}
txLanes := min(myLanes, int(peerPorts), n)
offset := lanePortOffset(myAddr, peerAddr, peerPorts)
// A flow picks its lane from a hash both sides compute identically, so with
// the ranges lined up the two directions of a flow would pick the same lane
// index — and lane s targets the peer's lane (s + portOffset), not lane s. The
// high-addressed side rotates its choice by the low side's offset, which is
// its own negated, so the two directions land on partner lanes and their
// 4-tuples are exact reverses: each side's traffic then arrives through the
// conntrack entry the other's probe opened. With mismatched ranges there are
// no partner lanes to find, so don't pretend: hash straight.
bias := uint16(0)
if txLanes == int(peerPorts) && peerAddr.Less(myAddr) {
bias = (peerPorts - offset) % peerPorts
}
ls := &laneSet{
sessions: make([]atomic.Pointer[ConnectionState], n),
material: laneMaterial{
eKey: r.EKey.UnsafeKey(),
dKey: r.DKey.UnsafeKey(),
cipher: r.Cipher,
myCert: r.MyCert,
peerCert: r.RemoteCert,
initiator: r.Initiator,
epoch: epoch,
},
txAddr: make([]atomic.Pointer[netip.AddrPort], txLanes),
demand: make([]atomic.Bool, txLanes),
probe: make([]laneProbeState, txLanes),
txLanes: txLanes,
peerPortCount: peerPorts,
peerBasePort: uint16(r.PeerBasePort),
portOffset: offset,
laneBias: bias,
}
// Every lane starts out demanded, so the first traffic tick on this tunnel
// probes all of them at once instead of waiting for a flow to hash onto each.
// Lanes have to be up *before* the flows are, not after: the peer writes a
// flow's inbound packets to the tun queue matching the socket they arrived on,
// and the kernel remembers that for as long as the flow stays busy. A flow that
// starts while the lanes are still down therefore gets pinned to queue 0 on
// both hosts for its whole life. This costs one probe per lane on any tunnel
// with traffic; an idle tunnel is never ticked, so it still costs nothing.
for s := 1; s < txLanes; s++ {
ls.demand[s].Store(true)
}
return ls
}
// laneLogAttr summarizes what a tunnel negotiated, for the handshake log lines.
// It is an empty attr, which slog drops, on a node not running multiport. A peer
// that negotiated no lanes still logs, with zeros: "we offered and got nothing"
// is exactly what you want to see when you expected lanes and have none.
func laneLogAttr(myLanes int, ls *laneSet) slog.Attr {
if myLanes == 0 {
return slog.Attr{}
}
if ls == nil {
return slog.Any("lanes", m{"tx": 0, "sessions": 0})
}
// Every field read here is immutable once the set is built.
return slog.Any("lanes", m{
"tx": ls.txLanes,
"sessions": len(ls.sessions),
"peerBasePort": ls.peerBasePort,
"peerPorts": ls.peerPortCount,
"portOffset": ls.portOffset,
})
}
// emitLaneStats reports how many lanes are carrying traffic and how many tunnels
// have any. Both are counted by walking the hostmap, because a counter kept at
// promotion and demotion would drift upward forever: a tunnel torn down while its
// lanes are up never demotes them. Only a node running multiport pays for the
// walk.
func (f *Interface) emitLaneStats(up, tunnels metrics.Gauge) {
var nUp, nTunnels int64
f.hostMap.ForEachIndex(func(hostinfo *HostInfo) {
ls := hostinfo.lanes
if ls == nil {
return
}
nTunnels++
for s := 1; s < ls.txLanes; s++ {
if ls.txAddr[s].Load() != nil {
nUp++
}
}
})
up.Update(nUp)
tunnels.Update(nTunnels)
}
// lanePortOffset returns the rotation applied to this pair's lane target ports,
// in [0, peerPortCount). Without it every low-routine peer would aim its few
// lanes at a big peer's first few ports, concentrating the big peer's receive
// work on a couple of sockets; the hash spreads pairs across the whole range.
//
// Both sides hash the same sorted vpn-address pair and the higher address
// negates the result, so when port counts match the two sides' rotations
// cancel: our lane s's 4-tuple stays the reverse of the peer's lane s, and each
// side's probe opens the conntrack entry the other's arrives through. (The one
// lane a nonzero rotation lands on the peer's base port has no partner lane, so
// behind a port-restricted NAT it is the one lane that may never come up.)
func lanePortOffset(myAddr, peerAddr netip.Addr, peerPortCount uint16) uint16 {
if peerPortCount == 0 {
return 0
}
lo, hi := myAddr, peerAddr
if hi.Less(lo) {
lo, hi = hi, lo
}
h := fnv.New32a()
b := lo.As16()
h.Write(b[:])
b = hi.As16()
h.Write(b[:])
o := uint16(h.Sum32() % uint32(peerPortCount))
if myAddr == hi {
o = (peerPortCount - o) % peerPortCount
}
return o
}
// laneSession returns the session to decrypt a lane s packet with, deriving one
// if this is the first packet to claim that lane. A nil session with no error
// means this tunnel has no lane s at all.
//
// cached reports whether the session was already in the table. A fresh one is
// deliberately left out of it: anyone who can spoof this tunnel's local index
// can name any lane, and installing on sight would let them make us hold a
// replay window and two cipher states per lane without authenticating anything.
// The caller must install with installSession once the packet decrypts, which is
// the first moment the lane is known to be real, and must use the session
// installSession hands back from then on.
func (i *HostInfo) laneSession(s uint8) (ci *ConnectionState, cached bool, err error) {
ls := i.lanes
if ls == nil || s == 0 || int(s) >= len(ls.sessions) {
return nil, false, nil
}
if cs := ls.sessions[s].Load(); cs != nil {
return cs, true, nil
}
cs, err := newLaneConnectionState(&ls.material, s)
if err != nil {
return nil, false, err
}
return cs, false, nil
}
// installSession publishes a session derived by laneSession, so the next packet
// on the lane doesn't have to derive it again, and returns the session that is
// now installed. cs must have already decrypted the packet at messageCounter.
//
// Two routines can race on a lane's first packet and derive a session each, and
// only one of them can be installed. Every session derived for a lane holds the
// same key, so a second one that ever encrypted would restart the message
// counter and repeat nonces under it. Returning the winner solves this.
func (ls *laneSet) installSession(l *slog.Logger, s uint8, cs *ConnectionState, messageCounter uint64) *ConnectionState {
if ls.sessions[s].CompareAndSwap(nil, cs) {
return cs
}
winner := ls.sessions[s].Load()
winner.noteSeen(l, messageCounter)
return winner
}
// session returns lane s's session for our own use, deriving and installing it
// if it doesn't exist yet. Unlike the RX path this needs no proof the lane is
// real: we only ask for lanes we chose to send on. s must be a lane this set
// covers.
func (ls *laneSet) session(s int) (*ConnectionState, error) {
if cs := ls.sessions[s].Load(); cs != nil {
return cs, nil
}
cs, err := newLaneConnectionState(&ls.material, uint8(s))
if err != nil {
return nil, err
}
if !ls.sessions[s].CompareAndSwap(nil, cs) {
return ls.sessions[s].Load(), nil
}
return cs, nil
}
// maxMessageCounter returns the highest counter across the base session and
// every lane session. Data rides the lanes, so the base counter alone would
// never reach the rehandshake or exhaustion thresholds and the lane keys would
// be used past their data-volume margin. Rolling the base tunnel replaces the
// lane keys with it, since lanes are derived from it.
func (i *HostInfo) maxMessageCounter() uint64 {
if i.ConnectionState == nil {
return 0
}
c := i.ConnectionState.messageCounter.Load()
if ls := i.lanes; ls != nil {
for s := range ls.sessions {
cs := ls.sessions[s].Load()
if cs == nil {
// Never derived, so it has never sent anything either.
continue
}
if lc := cs.messageCounter.Load(); lc > c {
c = lc
}
}
}
return c
}
// txLane returns the session and destination for lane s, or a nil session when
// the lane is down and the caller must use the base tunnel. A miss raises
// demand, which is what gets a down lane probed again, so we pay for a lane
// exactly where real traffic wanted one. Callers on the data plane come through
// txLaneForFlow.
func (ls *laneSet) txLane(s int) (*ConnectionState, netip.AddrPort) {
if ls == nil || s <= 0 || s >= ls.txLanes {
return nil, netip.AddrPort{}
}
if addr := ls.txAddr[s].Load(); addr != nil {
// The lane is only up because a probe was acked on it, and that probe
// derived the session, so this load cannot miss. Fall back rather than
// derive here anyway: this is the hot path and a nil is not worth an HKDF.
if cs := ls.sessions[s].Load(); cs != nil {
return cs, *addr
}
return nil, netip.AddrPort{}
}
// Load-guarded so the common case of a lane that will not come up is a
// plain read and cannot ping-pong the cache line these flags share.
if !ls.demand[s].Load() {
ls.demand[s].Store(true)
}
return nil, netip.AddrPort{}
}
// txLaneForFlow picks the lane a flow rides and returns it with its session and
// destination, or lane 0 and a nil session when the flow belongs on the base
// tunnel — either because the hash landed on lane 0 or because the lane it
// landed on is down.
//
// The lane comes from the flow rather than from the sending routine so that lane
// use doesn't depend on how the kernel steers tun queues; see txQueue. It is a
// pure function of the 5-tuple, so a flow stays on one lane for its life: no
// per-packet reordering, and one lane's replay window sees one set of flows.
func (ls *laneSet) txLaneForFlow(p *firewall.Packet) (int, *ConnectionState, netip.AddrPort) {
if ls == nil || ls.txLanes < 2 {
return 0, nil, netip.AddrPort{}
}
s := int((laneFlowHash(p) + uint32(ls.laneBias)) % uint32(ls.txLanes))
if s == 0 {
// Lane 0 is the base tunnel, and a full share of flows belongs on it.
return 0, nil, netip.AddrPort{}
}
cs, addr := ls.txLane(s)
return s, cs, addr
}
// laneFlowHash hashes a 5-tuple to the same value from either end of the flow,
// which is what lets both peers pick partner lanes for it (see laneBias). FNV-1a
// by hand rather than through hash/fnv: this runs per packet, and the interface
// there would escape the addresses to the heap.
func laneFlowHash(p *firewall.Packet) uint32 {
// Order the two endpoints so the direction of travel cannot change the hash.
aAddr, aPort := p.LocalAddr, p.LocalPort
bAddr, bPort := p.RemoteAddr, p.RemotePort
if bAddr.Less(aAddr) || (aAddr == bAddr && bPort < aPort) {
aAddr, aPort, bAddr, bPort = bAddr, bPort, aAddr, aPort
}
const prime = 16777619
h := uint32(2166136261)
x, y := aAddr.As16(), bAddr.As16()
for i := range x {
h = (h ^ uint32(x[i])) * prime
h = (h ^ uint32(y[i])) * prime
}
for _, b := range [5]byte{byte(aPort >> 8), byte(aPort), byte(bPort >> 8), byte(bPort), p.Protocol} {
h = (h ^ uint32(b)) * prime
}
return h
}
// laneTargetPortLocked returns the peer port lane s aims at. Only meaningful
// when peerPortCount is nonzero, which txLanes > 0 guarantees.
func (ls *laneSet) laneTargetPortLocked(s int) uint16 {
return ls.peerBasePort + uint16((s+int(ls.portOffset))%int(ls.peerPortCount))
}
// laneTargetLocked returns where lane s's probes go: the peer port this lane is
// paired with, on the peer's current direct address.
//
// A lane only ever aims at its own port. There is no fallback to the peer's base
// port when the lane port doesn't answer — a lane sharing the base port's
// destination gains only a source port of its own, while costing the peer the
// receive spread that is the whole point, so a lane that can't reach its port
// stays down and its traffic rides the base tunnel.
func (ls *laneSet) laneTargetLocked(s int, addr netip.Addr) netip.AddrPort {
return netip.AddrPortFrom(addr, ls.laneTargetPortLocked(s))
}
// laneRetryDelay is the backoff after fails consecutive probe failures.
func laneRetryDelay(fails uint8) time.Duration {
d := laneRetryBase << min(fails, 4)
if d > laneRetryMax {
d = laneRetryMax
}
return d
}
// noteAck records an acked probe for lane s, promoting the lane if it was down.
// gen must match the outstanding probe. Reports the target the lane came up on,
// and whether this ack is what promoted it.
func (ls *laneSet) noteAck(s int, gen uint8, now time.Time) (netip.AddrPort, bool) {
if s <= 0 || s >= len(ls.probe) {
return netip.AddrPort{}, false
}
ls.mu.Lock()
defer ls.mu.Unlock()
p := &ls.probe[s]
if p.sentAt.IsZero() || p.gen != gen {
// No probe outstanding, or an ack for a probe we have already given up
// on. Either way it says nothing about the lane's current path.
return netip.AddrPort{}, false
}
p.sentAt = time.Time{}
p.lastAck = now
p.fails = 0
p.retryAt = time.Time{}
target := p.target
if ls.txAddr[s].Load() != nil {
// Keepalive for a lane already up.
return target, false
}
ls.txAddr[s].Store(&target)
return target, true
}
// probeLanes runs one lane maintenance pass for a peer: it demotes lanes whose
// probe went unanswered, re-proves lanes that have been up a while without one,
// and probes down lanes the data plane asked for. Driven by the connection
// manager's per-tunnel traffic tick, which only fires for a live tunnel — the
// same condition that produces lane demand in the first place.
func (f *Interface) probeLanes(hostinfo *HostInfo, now time.Time, nb, out []byte) {
ls := hostinfo.lanes
if ls == nil || ls.txLanes < 2 {
return
}
remote := hostinfo.GetRemote()
ls.mu.Lock()
defer ls.mu.Unlock()
if !remote.IsValid() {
// Relayed, or otherwise without a direct path. Lanes are direct-only,
// so drop them all; a later tick rebuilds if a direct path returns.
ls.resetLocked()
return
}
if ls.peerAddr != remote.Addr() {
if ls.peerAddr.IsValid() {
// A roam is a new path, not a failure: forget the lanes built on
// the old one and let demand re-probe from a clean backoff. On the
// first pass there is nothing built yet, so just record the address.
ls.resetLocked()
}
ls.peerAddr = remote.Addr()
}
for s := 1; s < ls.txLanes; s++ {
p := &ls.probe[s]
up := ls.txAddr[s].Load() != nil
if !p.sentAt.IsZero() {
if now.Sub(p.sentAt) < laneProbeTimeout {
continue
}
// An aged-out probe is a failure whether it was bringing the lane
// up or keeping it up.
p.sentAt = time.Time{}
p.fails = min(p.fails+1, laneMaxFails)
p.retryAt = now.Add(laneRetryDelay(p.fails))
if up {
ls.txAddr[s].Store(nil)
hostinfo.logger(f.l).Info("Multiport lane demoted, probe unanswered", "lane", s, "udpAddr", p.target)
}
continue
}
if up {
if now.Sub(p.lastAck) < laneKeepalive {
continue
}
} else if now.Before(p.retryAt) || !ls.demand[s].Swap(false) {
continue
}
p.gen++
p.target = ls.laneTargetLocked(s, remote.Addr())
if f.sendLaneProbe(hostinfo, s, p.gen, p.target, nb, out) {
p.sentAt = now
} else {
p.fails = min(p.fails+1, laneMaxFails)
p.retryAt = now.Add(laneRetryDelay(p.fails))
}
}
}
// resetLocked takes every lane down and clears its probe state, without
// counting it as a failure.
//
// Demand is deliberately left standing: it records that a routine has real
// traffic for this peer, which a roam or a relay detour does not change. Keeping
// it re-probes the lanes that were actually carrying data as soon as a path
// exists again, while a lane whose routine has gone quiet stays down.
func (ls *laneSet) resetLocked() {
for s := 1; s < len(ls.probe); s++ {
ls.txAddr[s].Store(nil)
ls.probe[s] = laneProbeState{}
}
}
// sendLaneProbe sends a probe on lane s to addr from a socket on lane s's port.
// The probe is an ordinary Test packet encrypted with the lane's session, so an
// ack proves the whole lane: our source port reached the peer, its reply reached
// us, and the keys we derived for this lane match the ones it derived. Reports
// whether the probe made it onto the wire.
//
// It goes out the first socket on the port, since the connection manager runs
// this and has no queue of its own; every socket on the port has the same address,
// so the probe proves the path for whichever one the data plane picks.
func (f *Interface) sendLaneProbe(hostinfo *HostInfo, s int, gen uint8, addr netip.AddrPort, nb, out []byte) bool {
// The first probe on a lane is what derives its session.
ci, err := hostinfo.lanes.session(s)
if err != nil {
hostinfo.logger(f.l).Error("Failed to derive multiport lane session", "error", err, "lane", s)
return false
}
if ci == nil || ci.eKey == nil {
return false
}
if noiseutil.EncryptLockNeeded {
ci.writeLock.Lock()
}
c, ok := ci.NextMessageCounter()
if !ok {
if noiseutil.EncryptLockNeeded {
ci.writeLock.Unlock()
}
f.dropExhausted(hostinfo, c, "Dropping multiport lane probe, lane message counter is exhausted")
return false
}
b := header.EncodeLane(out[:0], header.Version, header.Test, header.LaneProbe, hostinfo.remoteIndexId, c, uint8(s))
b, err = ci.eKey.EncryptDanger(b, b, []byte{uint8(s), gen}, c, nb)
if noiseutil.EncryptLockNeeded {
ci.writeLock.Unlock()
}
if err != nil {
hostinfo.logger(f.l).Error("Failed to encrypt multiport lane probe", "error", err, "lane", s)
return false
}
f.messageMetrics.Tx(header.Test, header.LaneProbe, 1)
if err := f.writers[f.laneSock(0, s)].WriteTo(b, addr); err != nil {
hostinfo.logger(f.l).Error("Failed to send multiport lane probe", "error", err, "lane", s, "udpAddr", addr)
return false
}
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("Multiport lane probe sent", "lane", s, "gen", gen, "udpAddr", addr)
}
return true
}
// handleLaneProbe answers a peer's lane probe. The ack rides the base tunnel on
// purpose: a probe proves the peer's lane s works in its send direction, and
// answering on our own lane s would make the result depend on a second path
// that may be broken independently.
func (f *Interface) handleLaneProbe(hostinfo *HostInfo, lane uint8, payload []byte, rxc *rxContext) {
if lane == 0 || len(payload) < 2 {
return
}
// Echo the header's lane rather than the payload's, so a peer cannot get us
// to vouch for a lane it did not actually probe.
f.send(header.Test, header.LaneProbeAck, hostinfo.ConnectionState, hostinfo,
[]byte{lane, payload[1]}, rxc.nb, rxc.scratch[:0])
}
// handleLaneProbeAck promotes the lane a peer just acked.
func (f *Interface) handleLaneProbeAck(hostinfo *HostInfo, payload []byte) {
ls := hostinfo.lanes
if ls == nil || len(payload) < 2 {
return
}
// The target is worth logging next to the demotion that names the same
// address, so a flapping lane can be read off the logs.
if target, promoted := ls.noteAck(int(payload[0]), payload[1], time.Now()); promoted {
hostinfo.logger(f.l).Info("Multiport lane up", "lane", payload[0], "udpAddr", target)
}
}
+1061
View File
File diff suppressed because it is too large. Load diff
+7 -71
View File
@@ -15,12 +15,10 @@ import (
"time"
"github.com/gaissmai/bart"
"github.com/rcrowley/go-metrics"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/logging"
"github.com/slackhq/nebula/nicfinder"
"github.com/slackhq/nebula/udp"
"github.com/slackhq/nebula/util"
)
@@ -36,18 +34,12 @@ type LightHouse struct {
myVpnNetworks []netip.Prefix
myVpnNetworksTable *bart.Lite
// myVpnAddrsTable contains our overlay host addrs, as opposed to the overlay networks
myVpnAddrsTable *bart.Lite
punchy *Punchy
punchy *Punchy
// localAddrsFn enumerates the underlay addresses we advertise. It is a field so tests can supply simulated
// addresses rather than whatever this machine's NICs happen to be. Set it before Start.
localAddrsFn func(*LocalAllowList) []netip.Addr
// lastLocalAddrsErr is the previous localAddrsFn failure. Enumeration runs on every update, so an
// unchanged failure is demoted to Debug rather than warning every lighthouse.interval forever.
lastLocalAddrsErr atomic.Pointer[string]
// Local cache of answers from light houses
// map of vpn addr to answers
addrMap map[netip.Addr]*RemoteList
@@ -82,8 +74,6 @@ type LightHouse struct {
updateTrigger chan struct{}
queryChan chan netip.Addr
// queryDropped counts queries QueryServer dropped because queryChan was full.
queryDropped metrics.Counter
calculatedRemotes atomic.Pointer[bart.Table[[]*calculatedRemote]] // Maps VpnAddr to []*calculatedRemote
@@ -114,19 +104,15 @@ func NewLightHouseFromConfig(ctx context.Context, l *slog.Logger, c *config.C, c
amLighthouse: amLighthouse,
myVpnNetworks: cs.myVpnNetworks,
myVpnNetworksTable: cs.myVpnNetworksTable,
myVpnAddrsTable: cs.myVpnAddrsTable,
addrMap: make(map[netip.Addr]*RemoteList),
nebulaPort: nebulaPort,
punchy: p,
updateTrigger: make(chan struct{}, 1),
queryChan: make(chan netip.Addr, c.GetUint32("handshakes.query_buffer", 1024)),
queryDropped: metrics.GetOrRegisterCounter("lighthouse.query.dropped", nil),
queryChan: make(chan netip.Addr, c.GetUint32("handshakes.query_buffer", 64)),
l: l,
}
h.localAddrsFn = func(al *LocalAllowList) []netip.Addr {
addrs, err := nicfinder.Find(ctx, h.l, al)
h.logLocalAddrsErr(err)
return addrs
return localAddrs(h.l, al)
}
lighthouses := make([]netip.Addr, 0)
@@ -359,13 +345,7 @@ func (lh *LightHouse) reload(c *config.C, initial bool) error {
}
}
// am_relay decides whether relay.relays applies to us at all
if initial || c.HasChanged("relay.relays") || c.HasChanged("relay.am_relay") {
var before []netip.Addr
if !initial {
before = lh.GetRelaysForMe()
}
if initial || c.HasChanged("relay.relays") {
switch c.GetBool("relay.am_relay", false) {
case true:
// Relays aren't allowed to specify other relays
@@ -390,11 +370,6 @@ func (lh *LightHouse) reload(c *config.C, initial bool) error {
}
lh.relaysForMe.Store(&relaysForMe)
}
// Tell the lighthouses now rather than at the next interval, but only when what we advertise changed
if !initial && !slices.Equal(before, lh.GetRelaysForMe()) {
lh.TriggerUpdate()
}
}
return nil
@@ -527,19 +502,14 @@ func (lh *LightHouse) Query(vpnAddr netip.Addr) *RemoteList {
return nil
}
// QueryServer is asynchronous and non-blocking, no reply should be expected.
// If the query channel is full, the query is dropped.
// QueryServer is asynchronous so no reply should be expected
func (lh *LightHouse) QueryServer(vpnAddr netip.Addr) {
// Don't put lighthouse addrs in the query channel because we can't query lighthouses about lighthouses
if lh.amLighthouse || lh.IsLighthouseAddr(vpnAddr) {
return
}
select {
case lh.queryChan <- vpnAddr:
default:
lh.queryDropped.Inc(1)
}
lh.queryChan <- vpnAddr
}
func (lh *LightHouse) QueryCache(vpnAddrs []netip.Addr) *RemoteList {
@@ -943,26 +913,6 @@ func (lh *LightHouse) TriggerUpdate() {
}
}
// logLocalAddrsErr reports a localAddrs failure at Warn the first time it is seen and at Debug while
// it persists unchanged, so a permanent failure does not warn on every update forever.
func (lh *LightHouse) logLocalAddrsErr(err error) {
if err == nil {
lh.lastLocalAddrsErr.Store(nil)
return
}
msg := err.Error()
prev := lh.lastLocalAddrsErr.Swap(&msg)
if prev != nil && *prev == msg {
if lh.l.Enabled(context.Background(), slog.LevelDebug) {
lh.l.Debug("Failed to collect local addresses to advertise", "error", err)
}
return
}
lh.l.Warn("Failed to collect local addresses to advertise", "error", err)
}
func (lh *LightHouse) SendUpdate() {
var v4 []*V4AddrPort
var v6 []*V6AddrPort
@@ -1208,17 +1158,6 @@ func (lhh *LightHouseHandler) handleHostQuery(n *NebulaMeta, fromVpnAddrs []neti
return
}
// Don't respond to requests for us.
if lhh.lh.myVpnAddrsTable.Contains(queryVpnAddr) {
if lhh.l.Enabled(context.Background(), slog.LevelDebug) {
lhh.l.Debug("Ignoring HostQuery for one of my own addresses",
"fromVpnAddrs", fromVpnAddrs,
"queryVpnAddr", queryVpnAddr,
)
}
return
}
found, ln, err := lhh.lh.queryAndPrepMessage(queryVpnAddr, func(c *cache) (int, error) {
n = lhh.resetMeta()
n.Type = NebulaMeta_HostQueryReply
@@ -1255,13 +1194,10 @@ func (lhh *LightHouseHandler) handleHostQuery(n *NebulaMeta, fromVpnAddrs []neti
// sendHostPunchNotification signals the other side to punch some zero byte udp packets
func (lhh *LightHouseHandler) sendHostPunchNotification(n *NebulaMeta, fromVpnAddrs []netip.Addr, punchNotifDest netip.Addr, w EncWriter) {
whereToPunch := fromVpnAddrs[0]
// Resolve the target before taking the querier's RemoteList lock below. GetHostInfo takes the
// hostmap lock, and the hostmap is taken before the lighthouse, which is taken before a
// RemoteList, so looking it up inside the callback would deadlock.
targetHI := lhh.lh.ifce.GetHostInfo(punchNotifDest)
found, ln, err := lhh.lh.queryAndPrepMessage(whereToPunch, func(c *cache) (int, error) {
n = lhh.resetMeta()
n.Type = NebulaMeta_HostPunchNotification
targetHI := lhh.lh.ifce.GetHostInfo(punchNotifDest)
var useVersion cert.Version
if targetHI == nil {
useVersion = lhh.lh.ifce.GetCertState().initiatingVersion
+53 -312
View File
@@ -1,21 +1,16 @@
package nebula
import (
"bytes"
"encoding/binary"
"errors"
"fmt"
"net/netip"
"sync"
"testing"
"time"
"github.com/gaissmai/bart"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/test"
"github.com/slackhq/nebula/udp"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"go.yaml.in/yaml/v3"
@@ -32,27 +27,15 @@ func TestOldIPv4Only(t *testing.T) {
assert.Equal(t, binary.BigEndian.Uint32(bp[:]), m.GetAddr())
}
func testCertState(networks ...netip.Prefix) *CertState {
cs := &CertState{
myVpnNetworks: networks,
myVpnNetworksTable: new(bart.Lite),
myVpnAddrs: make([]netip.Addr, 0, len(networks)),
myVpnAddrsTable: new(bart.Lite),
}
for _, n := range networks {
cs.myVpnNetworksTable.Insert(n)
cs.myVpnAddrs = append(cs.myVpnAddrs, n.Addr())
cs.myVpnAddrsTable.Insert(netip.PrefixFrom(n.Addr(), n.Addr().BitLen()))
}
return cs
}
func Test_lhStaticMapping(t *testing.T) {
l := test.NewLogger()
myVpnNet := netip.MustParsePrefix("10.128.0.1/16")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
lh1 := "10.128.0.2"
c := config.NewC(l)
@@ -72,7 +55,12 @@ func Test_lhStaticMapping(t *testing.T) {
func TestReloadLighthouseInterval(t *testing.T) {
l := test.NewLogger()
myVpnNet := netip.MustParsePrefix("10.128.0.1/16")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
lh1 := "10.128.0.2"
c := config.NewC(l)
@@ -102,7 +90,12 @@ func TestReloadLighthouseInterval(t *testing.T) {
func BenchmarkLighthouseHandleRequest(b *testing.B) {
l := test.NewLogger()
myVpnNet := netip.MustParsePrefix("10.128.0.1/0")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
c := config.NewC(l)
lh, err := NewLightHouseFromConfig(b.Context(), l, c, cs, nil, nil)
@@ -202,7 +195,12 @@ func TestLighthouse_Memory(t *testing.T) {
c.Settings["listen"] = map[string]any{"port": 4242}
myVpnNet := netip.MustParsePrefix("10.128.0.1/24")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
lh.ifce = &mockEncWriter{}
require.NoError(t, err)
@@ -282,7 +280,12 @@ func TestLighthouse_reload(t *testing.T) {
c.Settings["listen"] = map[string]any{"port": 4242}
myVpnNet := netip.MustParsePrefix("10.128.0.1/24")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
require.NoError(t, err)
@@ -312,7 +315,12 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
}
myVpnNet := netip.MustParsePrefix("10.128.0.1/24")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
require.NoError(t, err)
@@ -421,9 +429,7 @@ func TestLighthouse_reloadStaticHostMap(t *testing.T) {
assert.Equal(t, []netip.AddrPort{netip.MustParseAddrPort("3.3.3.3:4242")}, rl.CopyAddrs([]netip.Prefix{}))
}
// sendLHHostRequest delivers a HostQuery to lhh and hands back the writer that
// captured what it emitted. Pass a nil filter to see every message.
func sendLHHostRequest(fromAddr netip.AddrPort, myVpnIp, queryVpnIp netip.Addr, lhh *LightHouseHandler, filter *NebulaMeta_MessageType) *testEncWriter {
func newLHHostRequest(fromAddr netip.AddrPort, myVpnIp, queryVpnIp netip.Addr, lhh *LightHouseHandler) testLhReply {
req := &NebulaMeta{
Type: NebulaMeta_HostQuery,
Details: &NebulaMetaDetails{},
@@ -441,59 +447,12 @@ func sendLHHostRequest(fromAddr netip.AddrPort, myVpnIp, queryVpnIp netip.Addr,
panic(err)
}
w := &testEncWriter{metaFilter: filter}
lhh.HandleRequest(fromAddr, []netip.Addr{myVpnIp}, b, w)
return w
}
func newLHHostRequest(fromAddr netip.AddrPort, myVpnIp, queryVpnIp netip.Addr, lhh *LightHouseHandler) testLhReply {
filter := NebulaMeta_HostQueryReply
return sendLHHostRequest(fromAddr, myVpnIp, queryVpnIp, lhh, &filter).lastReply
}
func TestLighthouse_IgnoresHostQueryForItself(t *testing.T) {
// Validate that we don't answer host queries for our own address.
l := test.NewLogger()
myVpnNet := netip.MustParsePrefix("10.128.0.1/24")
myVpnIp := myVpnNet.Addr()
c := config.NewC(l)
c.Settings["lighthouse"] = map[string]any{"am_lighthouse": true}
c.Settings["listen"] = map[string]any{"port": 4242}
// Add a static_host_map entry for ourselves, so our address
// is in the addrMap.
c.Settings["static_host_map"] = map[string]any{
myVpnIp.String(): []any{"192.168.100.1:4242"},
w := &testEncWriter{
metaFilter: &filter,
}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, testCertState(myVpnNet), nil, nil)
require.NoError(t, err)
lh.ifce = &mockEncWriter{}
lhh := lh.NewRequestHandler()
peerVpnIp := netip.MustParseAddr("10.128.0.2")
peerUdpAddr := netip.MustParseAddrPort("10.0.0.2:4242")
otherVpnIp := netip.MustParseAddr("10.128.0.3")
otherUdpAddr := netip.MustParseAddrPort("10.0.0.3:4242")
newLHHostUpdate(peerUdpAddr, peerVpnIp, []netip.AddrPort{peerUdpAddr}, lhh)
newLHHostUpdate(otherUdpAddr, otherVpnIp, []netip.AddrPort{otherUdpAddr}, lhh)
// Control: a query about a real peer is still answered, and still ends with
// the punch notification aimed at the host that was asked about.
w := sendLHHostRequest(peerUdpAddr, peerVpnIp, otherVpnIp, lhh, nil)
require.NotNil(t, w.lastReply.msg)
assert.Equal(t, NebulaMeta_HostPunchNotification, w.lastReply.msg.Type)
assert.Equal(t, otherVpnIp, w.lastReply.vpnIp)
// Now validate that we don't send to ourselves.
found, _, err := lh.queryAndPrepMessage(myVpnIp, func(*cache) (int, error) { return 0, nil })
require.NoError(t, err)
require.True(t, found, "the lighthouse should hold a cache entry for its own address")
w = sendLHHostRequest(peerUdpAddr, peerVpnIp, myVpnIp, lhh, nil)
assert.Nil(t, w.lastReply.msg, "a query about our own address must produce no reply and no punch notification")
lhh.HandleRequest(fromAddr, []netip.Addr{myVpnIp}, b, w)
return w.lastReply
}
func newLHHostUpdate(fromAddr netip.AddrPort, vpnIp netip.Addr, addrs []netip.AddrPort, lhh *LightHouseHandler) {
@@ -683,7 +642,12 @@ func TestLighthouse_Dont_Delete_Static_Hosts(t *testing.T) {
}
myVpnNet := netip.MustParsePrefix("10.128.0.1/24")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
require.NoError(t, err)
lh.ifce = &mockEncWriter{}
@@ -744,7 +708,12 @@ func TestLighthouse_DeletesWork(t *testing.T) {
}
myVpnNet := netip.MustParsePrefix("10.128.0.1/24")
cs := testCertState(myVpnNet)
nt := new(bart.Lite)
nt.Insert(myVpnNet)
cs := &CertState{
myVpnNetworks: []netip.Prefix{myVpnNet},
myVpnNetworksTable: nt,
}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
require.NoError(t, err)
lh.ifce = &mockEncWriter{}
@@ -769,231 +738,3 @@ func TestLighthouse_DeletesWork(t *testing.T) {
out = lh.Query(testHost)
assert.Nil(t, out)
}
func TestLightHouse_logLocalAddrsErr(t *testing.T) {
out := &bytes.Buffer{}
lh := &LightHouse{l: test.NewLoggerWithOutput(out)}
// The first sighting of a failure warns.
lh.logLocalAddrsErr(errors.New("permission denied"))
assert.Contains(t, out.String(), "level=WARN")
assert.Contains(t, out.String(), "permission denied")
// Repeating unchanged does not warn again, which is what keeps a permanent failure from warning
// on every lighthouse.interval for the life of the process.
out.Reset()
lh.logLocalAddrsErr(errors.New("permission denied"))
assert.NotContains(t, out.String(), "level=WARN")
// A different failure is a new event and warns.
out.Reset()
lh.logLocalAddrsErr(errors.New("something else"))
assert.Contains(t, out.String(), "level=WARN")
assert.Contains(t, out.String(), "something else")
// Recovering resets, so the same failure returning later warns again.
out.Reset()
lh.logLocalAddrsErr(nil)
assert.Empty(t, out.String())
lh.logLocalAddrsErr(errors.New("something else"))
assert.Contains(t, out.String(), "level=WARN")
}
// heldEncWriter holds the query worker in its first GetHostInfo until release closes.
type heldEncWriter struct {
mockEncWriter
once sync.Once
held chan struct{}
release chan struct{}
}
func (w *heldEncWriter) GetHostInfo(netip.Addr) *HostInfo {
w.once.Do(func() { close(w.held) })
<-w.release
return nil
}
// QueryServer never waits on the query worker, since the packet readers call it and nothing drains the channel at
// shutdown; a query that finds the channel full is dropped and counted. The worker is held on its first query so the
// channel fills.
func TestLighthouse_QueryServerDoesNotBlock(t *testing.T) {
l := test.NewLogger()
cs := testCertState(netip.MustParsePrefix("10.128.0.1/16"))
lh1 := "10.128.0.2"
c := config.NewC(l)
c.Settings["lighthouse"] = map[string]any{"hosts": []any{lh1}}
c.Settings["static_host_map"] = map[string]any{lh1: []any{"1.1.1.1:4242"}}
c.Settings["handshakes"] = map[string]any{"query_buffer": 1}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, cs, nil, nil)
require.NoError(t, err)
w := &heldEncWriter{held: make(chan struct{}), release: make(chan struct{})}
lh.ifce = w
t.Cleanup(func() { close(w.release) })
lh.QueryServer(netip.MustParseAddr("10.128.0.10"))
select {
case <-w.held:
case <-time.After(5 * time.Second):
t.Fatal("the query worker never took the first query")
}
dropped := lh.queryDropped.Count()
done := make(chan struct{})
go func() {
lh.QueryServer(netip.MustParseAddr("10.128.0.11"))
lh.QueryServer(netip.MustParseAddr("10.128.0.12"))
close(done)
}()
select {
case <-done:
case <-time.After(5 * time.Second):
t.Fatal("QueryServer blocked on a full query channel")
}
assert.EqualValues(t, 1, lh.queryDropped.Count()-dropped)
}
// hostmapEncWriter answers GetHostInfo from a real hostmap, the way Interface does.
type hostmapEncWriter struct {
mockEncWriter
hm *HostMap
}
func (w *hostmapEncWriter) GetHostInfo(a netip.Addr) *HostInfo { return w.hm.QueryVpnAddr(a) }
func (w *hostmapEncWriter) GetCertState() *CertState {
return &CertState{initiatingVersion: cert.Version1}
}
// A lighthouse runs four paths that each hold one lock while waiting for the next: CheckAndComplete
// holds the hostmap and takes the handshake manager, StartHandshake holds the handshake manager and
// takes the lighthouse, a host update holds the lighthouse and takes the sender's RemoteList, and a
// punch notification holds the querier's RemoteList while it resolves the target. If that last step
// reaches the hostmap under the RemoteList lock, the four deadlock. Drive them all at once.
func TestLighthouse_PunchNotificationDoesNotCloseLockCycle(t *testing.T) {
l := test.NewLogger()
myVpnNet := netip.MustParsePrefix("10.128.0.1/24")
c := config.NewC(l)
c.Settings["lighthouse"] = map[string]any{"am_lighthouse": true}
c.Settings["listen"] = map[string]any{"port": 4242}
lh, err := NewLightHouseFromConfig(t.Context(), l, c, testCertState(myVpnNet), nil, nil)
require.NoError(t, err)
mainHM := newHostMap(l)
pr := []netip.Prefix{}
mainHM.preferredRanges.Store(&pr)
lh.ifce = &hostmapEncWriter{hm: mainHM}
hm := NewHandshakeManager(l, mainHM, lh, &udp.NoopConn{}, defaultHandshakeConfig)
f := &Interface{hostMap: mainHM, lightHouse: lh, handshakeManager: hm, l: l}
hm.f = f
peer := netip.MustParseAddr("10.128.0.2")
peerUdp := netip.MustParseAddrPort("10.0.0.2:4242")
target := netip.MustParseAddr("10.128.0.3")
targetUdp := netip.MustParseAddrPort("10.0.0.3:4242")
outbound := netip.MustParseAddr("10.128.0.4")
inbound := netip.MustParseAddr("10.128.0.5")
seed := lh.NewRequestHandler()
newLHHostUpdate(peerUdp, peer, []netip.AddrPort{peerUdp}, seed)
newLHHostUpdate(targetUdp, target, []netip.AddrPort{targetUdp}, seed)
stop := make(chan struct{})
var wg sync.WaitGroup
loop := func(body func()) {
defer wg.Done()
for {
select {
case <-stop:
return
default:
body()
}
}
}
wg.Add(4)
// Outbound handshakes keep starting and being dropped.
go loop(func() {
hm.DeleteHostInfo(hm.StartHandshake(outbound, nil))
})
// Inbound handshakes keep completing and being torn down.
var idx uint32
go loop(func() {
idx++
hi := &HostInfo{
vpnAddrs: []netip.Addr{inbound},
localIndexId: idx,
remoteIndexId: idx,
remotes: lh.QueryCache([]netip.Addr{inbound}),
HandshakePacket: map[uint8][]byte{handshakePacketStage0: {byte(idx)}},
}
if _, err := hm.CheckAndComplete(hi, handshakePacketStage0, f); err == nil {
mainHM.DeleteHostInfo(hi)
}
})
// The peer keeps sending host updates.
updates := lh.NewRequestHandler()
go loop(func() {
newLHHostUpdate(peerUdp, peer, []netip.AddrPort{peerUdp}, updates)
})
// The same peer keeps querying for the target, which ends in a punch notification.
queries := lh.NewRequestHandler()
go loop(func() {
sendLHHostRequest(peerUdp, peer, target, queries, nil)
})
time.Sleep(time.Second)
close(stop)
done := make(chan struct{})
go func() { wg.Wait(); close(done) }()
select {
case <-done:
case <-time.After(5 * time.Second):
t.Fatal("deadlock: the lighthouse paths never finished after stop was signalled")
}
}
func TestLighthouse_reloadRelaysTriggersUpdate(t *testing.T) {
l := test.NewLogger()
c := config.NewC(l)
base := "lighthouse:\n am_lighthouse: true\nlisten:\n port: 4242\n"
require.NoError(t, c.LoadString(base+"relay:\n relays: [10.128.0.5]\n"))
lh, err := NewLightHouseFromConfig(t.Context(), l, c, testCertState(netip.MustParsePrefix("10.128.0.1/24")), nil, nil)
require.NoError(t, err)
assert.Empty(t, lh.updateTrigger)
require.NoError(t, c.ReloadConfigString(base+"relay:\n relays: [10.128.0.6]\n"))
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.6")}, lh.GetRelaysForMe())
require.Len(t, lh.updateTrigger, 1)
<-lh.updateTrigger
// The config changed but what we advertise did not, a bad entry is skipped
require.NoError(t, c.ReloadConfigString(base+"relay:\n relays: [10.128.0.6, nope]\n"))
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.6")}, lh.GetRelaysForMe())
assert.Empty(t, lh.updateTrigger)
}
func TestLighthouse_reloadAmRelayAlone(t *testing.T) {
l := test.NewLogger()
c := config.NewC(l)
base := "lighthouse:\n am_lighthouse: true\nlisten:\n port: 4242\nrelay:\n relays: [10.128.0.5]\n"
require.NoError(t, c.LoadString(base+" am_relay: false\n"))
lh, err := NewLightHouseFromConfig(t.Context(), l, c, testCertState(netip.MustParsePrefix("10.128.0.1/24")), nil, nil)
require.NoError(t, err)
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.5")}, lh.GetRelaysForMe())
// Relays aren't allowed to have relays, flipping am_relay alone must drop them and tell the lighthouses
require.NoError(t, c.ReloadConfigString(base+" am_relay: true\n"))
assert.Empty(t, lh.GetRelaysForMe())
require.Len(t, lh.updateTrigger, 1)
<-lh.updateTrigger
require.NoError(t, c.ReloadConfigString(base+" am_relay: false\n"))
assert.Equal(t, []netip.Addr{netip.MustParseAddr("10.128.0.5")}, lh.GetRelaysForMe())
require.Len(t, lh.updateTrigger, 1)
}
-31
View File
@@ -1,31 +0,0 @@
package logging
import "log/slog"
// RateLimiter is what RateLimited needs from a rate limiter, such as a *ratelimit.Limiter.
// Allow reports whether this call may proceed, and how many calls were denied since the last one that did.
type RateLimiter interface {
Allow() (denied uint64, ok bool)
}
// RateLimited checks lim before anything is built.
// When allowed it returns l, with a "suppressed" count attached if lim denied calls since its last allowed one.
// A nil lim allows everything.
//
// if l, ok := logging.RateLimited(f.l, f.udpTxErrLimiter); ok {
// hostinfo.logger(l).Error("Failed to write outgoing packet", "error", err)
// }
func RateLimited(l *slog.Logger, lim RateLimiter) (*slog.Logger, bool) {
if lim == nil {
return l, true
}
denied, ok := lim.Allow()
if !ok {
return nil, false
}
if denied > 0 {
return l.With("suppressed", denied), true
}
return l, true
}
-106
View File
@@ -1,106 +0,0 @@
package logging
import (
"bytes"
"errors"
"io"
"log/slog"
"net/netip"
"strings"
"testing"
"time"
"github.com/slackhq/nebula/ratelimit"
)
type fakeLimiter struct {
denied uint64
ok bool
}
func (f fakeLimiter) Allow() (uint64, bool) { return f.denied, f.ok }
func TestRateLimited(t *testing.T) {
var buf bytes.Buffer
base := NewLogger(&buf)
if l, ok := RateLimited(base, fakeLimiter{ok: false}); ok || l != nil {
t.Fatalf("denied call: ok=%v logger=%v", ok, l)
}
if l, ok := RateLimited(base, fakeLimiter{ok: true}); !ok || l != base {
t.Fatal("nothing denied should hand back the same logger")
}
if l, ok := RateLimited(base, nil); !ok || l != base {
t.Fatal("a nil limiter should allow with the same logger")
}
l, ok := RateLimited(base, fakeLimiter{denied: 7, ok: true})
if !ok {
t.Fatal("allowed call denied")
}
l.Info("hello", "k", "v")
if !strings.Contains(buf.String(), "suppressed=7") {
t.Fatalf("want suppressed=7, got %q", buf.String())
}
}
// The benchmarks compare a log line through slog directly against the same line behind RateLimited: logged with
// nothing denied, denied, and logged after denials.
var (
benchErr = errors.New("sendto: no buffer space available")
benchAddr = netip.MustParseAddrPort("192.0.2.1:4242")
)
func benchLogger() *slog.Logger {
return NewLogger(io.Discard).With("localIndex", 1234, "remoteIndex", 5678)
}
func BenchmarkRateLimitedBaseline(b *testing.B) {
lg := benchLogger()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
lg.Error("Failed to write outgoing packet", "error", benchErr, "udpAddr", benchAddr)
}
}
func BenchmarkRateLimitedAllowed(b *testing.B) {
lg, lim := benchLogger(), mustLimiter(b, 1, 0)
b.ReportAllocs()
for i := 0; i < b.N; i++ {
if l, ok := RateLimited(lg, lim); ok {
l.Error("Failed to write outgoing packet", "error", benchErr, "udpAddr", benchAddr)
}
}
}
func BenchmarkRateLimitedDenied(b *testing.B) {
lg, lim := benchLogger(), mustLimiter(b, 1, time.Hour)
lim.Allow()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
if l, ok := RateLimited(lg, lim); ok {
l.Error("Failed to write outgoing packet", "error", benchErr, "udpAddr", benchAddr)
}
}
}
func BenchmarkRateLimitedAfterDenied(b *testing.B) {
lg, lim := benchLogger(), fakeLimiter{denied: 5, ok: true}
b.ReportAllocs()
for i := 0; i < b.N; i++ {
if l, ok := RateLimited(lg, lim); ok {
l.Error("Failed to write outgoing packet", "error", benchErr, "udpAddr", benchAddr)
}
}
}
func mustLimiter(b *testing.B, burst int, every time.Duration) *ratelimit.Limiter {
l, err := ratelimit.New(burst, every)
if err != nil {
b.Fatal(err)
}
return l
}
+193 -59
View File
@@ -4,6 +4,7 @@ import (
"context"
"fmt"
"log/slog"
"math"
"net"
"net/netip"
"os"
@@ -14,6 +15,7 @@ import (
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/cpupick"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/noiseutil"
"github.com/slackhq/nebula/overlay"
"github.com/slackhq/nebula/sshd"
@@ -32,12 +34,9 @@ const _ = noiseutil.RejectHeadroom - 4*maxRoutines
func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, deviceFactory overlay.DeviceFactory) (retcon *Control, reterr error) {
ctx, cancel := context.WithCancel(context.Background())
// The goroutines started below stop only when this context does, and only a caller holding the
// Control can arrange that. Cancel whenever we are not handing one back, which covers an error
// and a config test alike: a config test used to leave the lighthouse query worker, and a
// hostname resolver per dns named static host, running for the life of the process.
// Automatically cancel the context if Main returns an error, to signal all created goroutines to quit.
defer func() {
if retcon == nil {
if reterr != nil {
cancel()
}
}()
@@ -113,6 +112,105 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
l.Info("Using multiple routines", "routines", routines)
}
port := c.GetInt("listen.port", 0)
batchSize := c.GetInt("listen.batch", 64)
if batchSize < 1 {
oldBatch := batchSize
batchSize = 1
l.Warn("listen.batch size is invalid", "provided", oldBatch, "overridden to", batchSize)
}
offloads := c.GetBool("listen.udp_offloads", false)
var listenHost netip.Addr
if !configTest {
rawListenHost := c.GetString("listen.host", "0.0.0.0")
if rawListenHost == "[::]" {
// Old guidance was to provide the literal `[::]` in `listen.host` but that won't resolve.
listenHost = netip.IPv6Unspecified()
} else {
ips, err := net.DefaultResolver.LookupNetIP(context.Background(), "ip", rawListenHost)
if err != nil {
return nil, util.ContextualizeIfNeeded("Failed to resolve listen.host", err)
}
if len(ips) == 0 {
return nil, util.ContextualizeIfNeeded("Failed to resolve listen.host", err)
}
listenHost = ips[0].Unmap()
}
}
// Multiport lanes: bind a range of consecutive UDP ports (listen.port+p)
// instead of SO_REUSEPORT-sharing one, and derive one extra session per lane
// with capable peers, so a tunnel's inside flows spread over several underlay
// 5-tuples instead of one. Defaults on, degrading gracefully when
// preconditions aren't met — managed deployments (dnclient) can't be
// hard-errored on config they don't control.
//
// `routines` is per port here rather than a total to divide up: every port
// gets its own full set, and a port's routines share it through SO_REUSEPORT
// so the kernel hashes each arriving 4-tuple onto one of them. That is what
// keeps a port from being served by a single core — the base port above all,
// since every handshake, every peer without multiport, and every tunnel whose
// lanes are down or firewalled arrives there. A routine still owns exactly one
// socket, which is what lets the read path own its state without locking, so
// the worker count is routines * ports and everything sized by routines from
// here down means that product.
routinesPerPort := routines
multiportPorts := 1
multiport := c.GetBool("multiport.enabled", true)
if multiport {
multiportPorts = c.GetInt("multiport.ports", 0)
if multiportPorts > header.MaxLane+1 {
// A lane index rides in one byte of the nebula header, so a port we
// could never address a lane on is a port we would never send from.
l.Warn("multiport.ports clamped to the lane header limit", "ports", multiportPorts, "limit", header.MaxLane+1)
multiportPorts = header.MaxLane + 1
}
if multiportPorts*routinesPerPort > maxRoutines {
clamped := max(maxRoutines/routinesPerPort, 1)
l.Warn("multiport.ports clamped to the routine limit",
"ports", multiportPorts, "clampedTo", clamped, "routinesPerPort", routinesPerPort, "limit", maxRoutines)
multiportPorts = clamped
}
if multiportPorts < 2 {
// A single port is a plain SO_REUSEPORT listener, which is what a node
// without multiport already runs. Say so rather than claiming lanes.
l.Info("multiport disabled: set multiport.ports > 1 to bind a lane port range")
multiport = false
}
}
if multiport && port != 0 && port+multiportPorts-1 > math.MaxUint16 {
l.Warn("multiport disabled: would bind ports beyond 65535", "listen.port", port, "ports", multiportPorts)
multiport = false
}
if multiport && !configTest {
// Every socket needs its own reader; a platform that can't run multiple
// readers would silently strand all but one of them as blackholes. Probe
// capability before sizing tun queues and routines to the port range.
probe, err := udp.NewListener(l, udp.Settings{
Listen: netip.AddrPortFrom(listenHost, 0),
Batch: 1,
})
if err != nil {
// We could not confirm support, so don't gamble a bound port range on it. The real bind below reports
// the underlying error if it is not transient.
l.Warn("multiport disabled: could not probe udp reader support", "error", err)
multiport = false
} else {
if !probe.SupportsMultipleReaders() {
l.Warn("multiport disabled: this platform does not support multiple udp readers")
multiport = false
}
_ = probe.Close()
}
}
if multiport {
routines = routinesPerPort * multiportPorts
l.Info("multiport routines", "ports", multiportPorts, "routinesPerPort", routinesPerPort, "routines", routines)
}
// EXPERIMENTAL
// Intentionally not documented yet while we do more testing and determine
// a good default value.
@@ -147,7 +245,6 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
// set up our UDP listener
udpConns := make([]udp.Conn, routines)
port := c.GetInt("listen.port", 0)
// Callers get no handle to these until the Control is returned, release them on any error.
defer func() {
@@ -161,54 +258,72 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
}()
if !configTest {
rawListenHost := c.GetString("listen.host", "0.0.0.0")
var listenHost netip.Addr
if rawListenHost == "[::]" {
// Old guidance was to provide the literal `[::]` in `listen.host` but that won't resolve.
listenHost = netip.IPv6Unspecified()
} else {
ips, err := net.DefaultResolver.LookupNetIP(context.Background(), "ip", rawListenHost)
if err != nil {
return nil, util.ContextualizeIfNeeded("Failed to resolve listen.host", err)
}
if len(ips) == 0 {
return nil, util.ContextualizeIfNeeded("Failed to resolve listen.host", err)
}
listenHost = ips[0].Unmap()
}
for i := 0; i < routines; i++ {
listen := netip.AddrPortFrom(listenHost, uint16(port))
l.Info("listening", "addr", listen)
batchSize := c.GetInt("listen.batch", 64)
if batchSize < 1 {
oldBatch := batchSize
batchSize = 1
l.Warn("listen.batch size is invalid", "provided", oldBatch, "overridden to", batchSize)
}
udpSettings := udp.Settings{
Listen: listen,
Multi: routines > 1,
Batch: batchSize,
Offloads: c.GetBool("listen.udp_offloads", false),
}
udpServer, err := udp.NewListener(l, udpSettings)
if err != nil {
return nil, util.NewContextualError("Failed to open udp listener", m{"queue": i}, err)
}
udpServer.ReloadConfig(c)
udpConns[i] = udpServer
// If port is dynamic, discover it before the next pass through the for loop
// This way all routines will use the same port correctly
if port == 0 {
uPort, err := udpServer.LocalAddr()
if err != nil {
return nil, util.NewContextualError("Failed to get listening port", nil, err)
// With a dynamic listen.port, multiport binds the first socket dynamically
// and then claims the next ports-1 above it; if that range turns out to be
// partially occupied, re-roll with a fresh dynamic port.
dynamic := port == 0
var bindErr error
for attempt := 0; attempt < 6; attempt++ {
bindErr = nil
for i := 0; i < routines; i++ {
// Routines are laid out port-major: routine i serves port slot
// i/routinesPerPort, so a port's routines are a contiguous run of
// indices and cpu pinning, which walks the index, spreads each
// port's group across the cores rather than stacking it on one.
slot := 0
if multiport {
slot = i / routinesPerPort
}
port = int(uPort.Port())
udpServer, err := udp.NewListener(l, udp.Settings{
Listen: netip.AddrPortFrom(listenHost, uint16(port+slot)),
// Under multiport the destination port narrows a packet to one
// port's routines and SO_REUSEPORT picks among them by 4-tuple
// hash, so both halves of the steering are load spreading: no
// port depends on a single core, and a lane still has a source
// port of its own to send from.
Multi: routines > 1,
Batch: batchSize,
Offloads: offloads,
})
if err != nil {
bindErr = util.NewContextualError("Failed to open udp listener", m{"queue": i}, err)
break
}
udpServer.ReloadConfig(c)
udpConns[i] = udpServer
// If port is dynamic, discover it before the next pass through the for loop
// This way all routines will use the same port correctly
if port == 0 {
uPort, err := udpServer.LocalAddr()
if err != nil {
return nil, util.NewContextualError("Failed to get listening port", nil, err)
}
port = int(uPort.Port())
if multiport && port+multiportPorts-1 > math.MaxUint16 {
bindErr = util.NewContextualError("multiport dynamic port too close to 65535", m{"port": port}, nil)
break
}
}
l.Info("listening", "addr", netip.AddrPortFrom(listenHost, uint16(port+slot)), "socket", i)
}
if bindErr == nil {
break
}
if !(multiport && dynamic) {
return nil, bindErr
}
for i := range udpConns {
if udpConns[i] != nil {
_ = udpConns[i].Close()
udpConns[i] = nil
}
}
port = 0
l.Debug("multiport dynamic port range collided, retrying", "attempt", attempt+1)
}
if bindErr != nil {
return nil, bindErr
}
}
@@ -232,11 +347,29 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
retries: int64(c.GetInt("handshakes.retries", DefaultHandshakeRetries)),
triggerBuffer: c.GetInt("handshakes.trigger_buffer", DefaultHandshakeTriggerBuffer),
messageMetrics: messageMetrics,
pq: pqHandshakeConfigFrom(c),
}
if multiport {
// A lane needs a port to send from, so we can send on no more lanes than we
// bound ports. multiport.lanes below that sends on a subset of the range,
// which is only interesting for narrowing an experiment: the ports are bound
// and read either way.
lanes := c.GetInt("multiport.lanes", 0)
if lanes <= 0 || lanes > multiportPorts {
lanes = multiportPorts
}
handshakeConfig.laneCount = lanes
handshakeConfig.lanePortCount = uint16(multiportPorts)
handshakeConfig.laneBasePort = uint16(port)
// Every other multiport log line is a reason it turned itself off, so say
// plainly when it is on and with what. A peer only gets lanes if it also
// advertises a port range, so this is our half of the negotiation.
l.Info("multiport enabled", "lanes", lanes, "basePort", port, "ports", multiportPorts,
"routinesPerPort", routinesPerPort)
}
handshakeManager := NewHandshakeManager(l, hostMap, lightHouse, udpConns[0], handshakeConfig)
c.RegisterReloadCallback(handshakeManager.reloadPQ)
lightHouse.handshakeTrigger = handshakeManager.trigger
ds, err := newDnsServerFromConfig(ctx, l, pki, hostMap, c)
@@ -291,6 +424,9 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
DropLocalBroadcast: c.GetBool("tun.drop_local_broadcast", false),
DropMulticast: c.GetBool("tun.drop_multicast", false),
routines: routines,
Multiport: multiport,
RoutinesPerPort: routinesPerPort,
LaneCount: handshakeConfig.laneCount,
MessageMetrics: messageMetrics,
version: buildVersion,
relayManager: NewRelayManager(ctx, l, hostMap, c),
@@ -302,7 +438,6 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
}
var ifce *Interface
var emitStats func()
if !configTest {
ifce, err = NewInterface(ctx, ifConfig)
if err != nil {
@@ -321,12 +456,9 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
go handshakeManager.Run(ctx)
punchy.Start(ctx, ifce, hostMap, lightHouse)
// On the stats server's loop, so the gauges follow the interval it accepted
emitStats = ifce.statsEmitter()
}
stats, err := newStatsServerFromConfig(ctx, l, c, buildVersion, configTest, emitStats)
stats, err := newStatsServerFromConfig(ctx, l, c, buildVersion, configTest)
if err != nil {
return nil, util.ContextualizeIfNeeded("Failed to start stats emitter", err)
}
@@ -335,6 +467,8 @@ func Main(c *config.C, configTest bool, buildVersion string, l *slog.Logger, dev
return nil, nil
}
go ifce.emitStats(ctx, c.GetDuration("stats.interval", time.Second*10))
attachCommands(l, c, ssh, ifce)
networkChanges := udp.NewNetworkChangeMonitor(ctx, l, c)
-74
View File
@@ -1,21 +1,12 @@
package nebula
import (
"fmt"
"net/netip"
"os"
"path/filepath"
"testing"
"time"
"github.com/slackhq/nebula/cert"
cert_test "github.com/slackhq/nebula/cert_test"
"github.com/slackhq/nebula/config"
"github.com/slackhq/nebula/test"
"github.com/slackhq/nebula/util"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"go.uber.org/goleak"
)
func TestParseCpuAffinity(t *testing.T) {
@@ -58,68 +49,3 @@ func TestParseCpuAffinity(t *testing.T) {
assert.Nil(t, parseCpuAffinity(newConfig([]any{1 << 20}), l, 1))
}
}
// TestMain_ConfigTestReleasesItsGoroutines pins the rule that Main only leaves goroutines running
// when it hands back a Control to stop them with.
//
// A config test gets no Control, so anything it started had nothing to stop it: the lighthouse
// query worker, and a hostname resolver per dns named static host, ran for the life of the
// process. That matters to every embedder that validates a config in process rather than by
// exec'ing, dnclient and the apple clients included, because they do it on each config load and
// the leak accumulates.
func TestMain_ConfigTestReleasesItsGoroutines(t *testing.T) {
defer goleak.VerifyNone(t, goleak.IgnoreCurrent())
l := test.NewLogger()
dir := t.TempDir()
before := time.Now().Add(-time.Hour)
after := time.Now().Add(time.Hour)
ca, _, caKey, caPEM := cert_test.NewTestCaCert(cert.Version2, cert.Curve_CURVE25519, before, after, nil, nil, nil)
networks := []netip.Prefix{netip.MustParsePrefix("10.0.0.1/24")}
_, _, keyPEM, certPEM := cert_test.NewTestCert(
cert.Version2, cert.Curve_CURVE25519, ca, caKey, "config-test", before, after, networks, nil, nil)
caPath := filepath.Join(dir, "ca.pem")
certPath := filepath.Join(dir, "cert.pem")
keyPath := filepath.Join(dir, "key.pem")
require.NoError(t, os.WriteFile(caPath, caPEM, 0o600))
require.NoError(t, os.WriteFile(certPath, certPEM, 0o600))
require.NoError(t, os.WriteFile(keyPath, keyPEM, 0o600))
// A static host by address, not by name: the query worker is the goroutine under test and a
// hostname would drag a real dns lookup into a unit test.
configBody := fmt.Sprintf(`
pki:
ca: %s
cert: %s
key: %s
static_host_map:
"10.0.0.2": ["192.0.2.1:4242"]
lighthouse:
hosts:
- "10.0.0.2"
listen:
host: 127.0.0.1
port: 0
tun:
disabled: true
firewall:
outbound:
- port: any
proto: any
host: any
inbound:
- port: any
proto: any
host: any
`, caPath, certPath, keyPath)
require.NoError(t, os.WriteFile(filepath.Join(dir, "config.yml"), []byte(configBody), 0o600))
c := config.NewC(l)
require.NoError(t, c.Load(dir))
ctrl, err := Main(c, true, "config-test", l, nil)
require.NoError(t, err)
require.Nil(t, ctrl, "a config test hands back nothing to stop, so it must stop itself")
}
-39
View File
@@ -1,39 +0,0 @@
//go:build !android
package nicfinder
import (
"context"
"errors"
"fmt"
"log/slog"
"net"
"net/netip"
)
func localAddrs(ctx context.Context, l *slog.Logger, filter Filter) ([]netip.Addr, error) {
ifaces, err := net.Interfaces()
if err != nil {
return nil, fmt.Errorf("failed to enumerate local interfaces: %w", err)
}
var out []netip.Addr
var errs []error
for _, i := range ifaces {
if !allowName(ctx, l, filter, i.Name) {
continue
}
addrs, err := i.Addrs()
if err != nil {
errs = append(errs, fmt.Errorf("failed to get addresses for %s: %w", i.Name, err))
continue
}
for _, rawAddr := range addrs {
if addr, ok := allowedAddr(ctx, l, filter, rawAddr); ok {
out = append(out, addr)
}
}
}
return out, errors.Join(errs...)
}
-104
View File
@@ -1,104 +0,0 @@
//go:build android
package nicfinder
import (
"context"
"errors"
"fmt"
"log/slog"
"net/netip"
"os"
"github.com/DefinedNet/netlink"
"github.com/DefinedNet/netlink/nl"
"golang.org/x/sys/unix"
)
// Android denies bind() on netlink route sockets and RTM_GETLINK, so net.Interfaces fails with EACCES.
// Enumerate with an RTM_GETADDR dump on an unbound socket instead.
// localInterfaces returns every interface that has an address, with its addresses. It lists
// addresses over a netlink socket that is never bound and names each index with SIOCGIFNAME, the
// same operations Bionic's getifaddrs uses for app UIDs, which the sandbox permits.
func localAddrs(ctx context.Context, l *slog.Logger, filter Filter) ([]netip.Addr, error) {
var errs []error
addrs, err := netlinkAddrList()
if err != nil {
if !errors.Is(err, netlink.ErrDumpInterrupted) {
return nil, fmt.Errorf("failed to enumerate local interfaces: %w", err)
}
// The address table kept changing under the dump and netlink gave up retrying. What it
// returned is still worth advertising.
errs = append(errs, fmt.Errorf("local addresses may be incomplete: %w", err))
}
// Any socket will do for SIOCGIFNAME.
sfd, err := unix.Socket(unix.AF_INET, unix.SOCK_DGRAM|unix.SOCK_CLOEXEC, 0)
if err != nil {
return nil, os.NewSyscallError("socket", err)
}
defer unix.Close(sfd)
var out []netip.Addr
nameAllowed := make(map[int]bool) // by interface index, once named
for _, a := range addrs {
allow, named := nameAllowed[a.LinkIndex]
if (named && !allow) || a.IPNet == nil {
continue
}
addr, ok := allowedAddr(ctx, l, filter, a.IPNet)
if !ok {
continue
}
if !named {
name, err := interfaceNameByIndex(sfd, a.LinkIndex)
// ENODEV means the interface went away after the dump. Any other failure is reported
// and the interface skipped, so the rest are still returned.
if err != nil && !errors.Is(err, unix.ENODEV) {
errs = append(errs, fmt.Errorf("failed to resolve name of interface %d: %w", a.LinkIndex, err))
}
allow = err == nil && allowName(ctx, l, filter, name)
nameAllowed[a.LinkIndex] = allow
}
if allow {
out = append(out, addr)
}
}
return out, errors.Join(errs...)
}
// netlinkAddrList is netlink.AddrList over a socket that is never bound. The kernel autobinds it
// on the first send, which the sandbox permits.
func netlinkAddrList() ([]netlink.Addr, error) {
fd, err := unix.Socket(unix.AF_NETLINK, unix.SOCK_RAW|unix.SOCK_CLOEXEC, unix.NETLINK_ROUTE)
if err != nil {
return nil, os.NewSyscallError("socket", err)
}
s, err := nl.NewNetlinkSocketFromFd(fd)
if err != nil {
unix.Close(fd)
return nil, err
}
h := netlink.NewHandleFromSockets(
map[int]*nl.SocketHandle{unix.NETLINK_ROUTE: {Socket: s}},
netlink.HandleOptions{RetryInterrupted: true},
)
defer h.Close()
if err := h.SetSocketTimeout(netlink.GetSocketTimeout()); err != nil {
return nil, err
}
return h.AddrList(nil, netlink.FAMILY_ALL)
}
// interfaceNameByIndex resolves an interface index with SIOCGIFNAME.
func interfaceNameByIndex(fd int, index int) (string, error) {
var ifr unix.Ifreq
ifr.SetUint32(uint32(index)) // ifr_ifindex is in the ifr_ifru union
if err := unix.IoctlIfreq(fd, unix.SIOCGIFNAME, &ifr); err != nil {
return "", err
}
return ifr.Name(), nil
}
-59
View File
@@ -1,59 +0,0 @@
package nicfinder
import (
"context"
"log/slog"
"net"
"net/netip"
"github.com/slackhq/nebula/logging"
)
// Filter decides which interfaces, by name, and which of their addresses Find returns.
type Filter interface {
AllowName(string) bool
Allow(netip.Addr) bool
}
// Find returns the addresses of this host's network interfaces that allowList accepts, with
// loopback and link-local addresses left out. When only some interfaces could be read it returns
// their addresses alongside an error for the rest. allowList must not be nil; it is called as
// given, so a nil pointer of a type whose methods accept a nil receiver is fine.
func Find(ctx context.Context, l *slog.Logger, allowList Filter) ([]netip.Addr, error) {
return localAddrs(ctx, l, allowList)
}
// allowName asks filter about an interface name and traces the answer.
func allowName(ctx context.Context, l *slog.Logger, filter Filter, name string) bool {
allow := filter.AllowName(name)
if l.Enabled(ctx, logging.LevelTrace) {
l.Log(ctx, logging.LevelTrace, "localAllowList.AllowName", "interfaceName", name, "allow", allow)
}
return allow
}
// allowedAddr converts one address a platform reports and says whether to advertise it: it must
// be a *net.IPNet or *net.IPAddr carrying an IP, not be loopback or link-local, and be accepted by
// filter, whose answer is traced. IPv4 comes back unmapped.
func allowedAddr(ctx context.Context, l *slog.Logger, filter Filter, rawAddr net.Addr) (netip.Addr, bool) {
var ip net.IP
switch v := rawAddr.(type) {
case *net.IPNet:
ip = v.IP
case *net.IPAddr:
ip = v.IP
}
addr, ok := netip.AddrFromSlice(ip)
if !ok {
return netip.Addr{}, false
}
addr = addr.Unmap()
if addr.IsLoopback() || addr.IsLinkLocalUnicast() {
return netip.Addr{}, false
}
isAllowed := filter.Allow(addr)
if l.Enabled(ctx, logging.LevelTrace) {
l.Log(ctx, logging.LevelTrace, "localAllowList.Allow", "localAddr", addr, "allowed", isAllowed)
}
return addr, isAllowed
}
-117
View File
@@ -1,117 +0,0 @@
package nicfinder
import (
"context"
"net"
"net/netip"
"regexp"
"testing"
"github.com/slackhq/nebula/test"
"github.com/stretchr/testify/assert"
)
type allowAll struct{}
func (allowAll) Allow(netip.Addr) bool { return true }
func (allowAll) AllowName(string) bool { return true }
type testAllowList struct {
bannedName *regexp.Regexp
bannedAddr netip.Addr
}
func (l *testAllowList) Allow(addr netip.Addr) bool {
return addr != l.bannedAddr
}
func (l *testAllowList) AllowName(n string) bool {
return !l.bannedName.MatchString(n)
}
func newTestAllowList(bannedName string) *testAllowList {
return &testAllowList{
bannedName: regexp.MustCompile(bannedName),
}
}
func TestAllowedAddr(t *testing.T) {
ctx, l := context.Background(), test.NewLogger()
allowed := func(filter Filter, addrs []net.Addr) []netip.Addr {
var out []netip.Addr
for _, rawAddr := range addrs {
if addr, ok := allowedAddr(ctx, l, filter, rawAddr); ok {
out = append(out, addr)
}
}
return out
}
addrs := []net.Addr{
&net.IPNet{IP: net.ParseIP("127.0.0.1"), Mask: net.CIDRMask(8, 32)},
&net.IPNet{IP: net.ParseIP("::1"), Mask: net.CIDRMask(128, 128)},
&net.IPNet{IP: net.ParseIP("10.0.0.5"), Mask: net.CIDRMask(24, 32)},
&net.IPNet{IP: net.ParseIP("169.254.10.10"), Mask: net.CIDRMask(16, 32)},
&net.IPNet{IP: net.ParseIP("fe80::1"), Mask: net.CIDRMask(64, 128)},
&net.IPAddr{IP: net.ParseIP("fd00::5")},
&net.IPNet{},
otherAddr{},
nil,
}
// Loopback, link local and anything without an IP are dropped, and IPv4 comes back unmapped
// even though net.ParseIP hands over a 16 byte slice.
assert.Equal(t, []netip.Addr{
netip.MustParseAddr("10.0.0.5"),
netip.MustParseAddr("fd00::5"),
}, allowed(allowAll{}, addrs))
// The filter sees the unmapped form and can drop a single address.
al := newTestAllowList("none")
al.bannedAddr = netip.MustParseAddr("10.0.0.5")
assert.Equal(t, []netip.Addr{netip.MustParseAddr("fd00::5")}, allowed(al, addrs))
}
// otherAddr is a net.Addr that is neither *net.IPNet nor *net.IPAddr. netlink.Addr is one such
// type: it satisfies net.Addr through an embedded *net.IPNet, and passing it directly once
// silently produced no addresses.
type otherAddr struct{}
func (otherAddr) Network() string { return "other" }
func (otherAddr) String() string { return "other" }
// hostFilter counts AllowName calls and can reject every name or every address.
type hostFilter struct {
rejectNames bool
rejectAddrs bool
asked map[string]int
}
func (f *hostFilter) Allow(netip.Addr) bool { return !f.rejectAddrs }
func (f *hostFilter) AllowName(name string) bool {
f.asked[name]++
return !f.rejectNames
}
// Runs this platform's localAddrs against the host and checks it honors the whole filter.
func TestLocalAddrsAppliesFilter(t *testing.T) {
ctx, l := context.Background(), test.NewLogger()
f := &hostFilter{asked: map[string]int{}}
all, err := localAddrs(ctx, l, f)
if len(f.asked) == 0 {
t.Skipf("no interfaces to enumerate here: %v", err)
}
for name, n := range f.asked {
assert.Equal(t, 1, n, "AllowName asked %d times for %s", n, name)
}
// Loopback and link-local never come back, whatever the filter says.
for _, addr := range all {
assert.False(t, addr.IsLoopback() || addr.IsLinkLocalUnicast(), "%s", addr)
}
none, _ := localAddrs(ctx, l, &hostFilter{rejectNames: true, asked: map[string]int{}})
assert.Empty(t, none, "addresses got past a filter that rejects every interface name")
none, _ = localAddrs(ctx, l, &hostFilter{rejectAddrs: true, asked: map[string]int{}})
assert.Empty(t, none, "addresses got past a filter that rejects every address")
}
+1
View File
@@ -1,4 +1,5 @@
//go:build boringcrypto
// +build boringcrypto
package noiseutil
+1
View File
@@ -1,4 +1,5 @@
//go:build boringcrypto
// +build boringcrypto
package noiseutil
+27
View File
@@ -1,6 +1,7 @@
package noiseutil
import (
"crypto/cipher"
"errors"
"fmt"
"math"
@@ -52,3 +53,29 @@ func NewCipherState(s *noise.CipherState, cipherFunc noise.CipherFunc) CipherSta
panic(fmt.Sprintf("noiseutil: unsupported cipher %q", cipherFunc.CipherName()))
}
}
// NewCipherStateFromKey builds a data-plane CipherState directly from raw key
// bytes, bypassing the noise handshake. Multiport lanes use this to key a
// derived session off the base tunnel's negotiated keys; the caller owns the
// guarantee that key is used with exactly one CipherState so nonces never
// repeat.
func NewCipherStateFromKey(key [32]byte, cipherFunc noise.CipherFunc) CipherState {
c := cipherFunc.Cipher(key)
if cs, ok := c.(CipherState); ok {
return cs
}
aead, ok := c.(cipher.AEAD)
if !ok {
panic(fmt.Sprintf("noiseutil: cipher %q does not expose an AEAD", cipherFunc.CipherName()))
}
switch cipherFunc.CipherName() {
case noise.CipherAESGCM.CipherName():
return &CipherStateAESGCM{c: aead}
case noise.CipherChaChaPoly.CipherName():
return &CipherStateChaChaPoly{c: aead}
default:
panic(fmt.Sprintf("noiseutil: unsupported cipher %q", cipherFunc.CipherName()))
}
}
+74 -21
View File
@@ -102,12 +102,49 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
// recvError if necessary
if hostinfo == nil || hostinfo.ConnectionState == nil {
if !via.IsRelayed {
f.maybeSendRecvError(via.UdpAddr, h.RemoteIndex)
f.maybeSendRecvError(via.UdpAddr, h.RemoteIndex, via.SockIdx)
}
return
}
if len(packet) < header.Len+hostinfo.ConnectionState.dKey.Overhead() {
// Which session decrypts this packet is the lane index in the header. Lane 0
// is the base tunnel; a higher lane is one of the sessions derived from it.
ci := hostinfo.ConnectionState
lane := h.Lane()
laneCached := true
if lane != 0 {
if isMessageRelay {
// A relay carrier is always the base tunnel, so lane ciphertext can
// never legitimately arrive wrapped in one. Checked before the lookup
// below so a junk relay packet can't make us derive a session.
f.messageMetrics.RxInvalid(1)
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("Refusing relayed multiport lane packet", "from", via, "header", h)
}
return
}
var err error
ci, laneCached, err = hostinfo.laneSession(lane)
if err != nil {
f.messageMetrics.RxInvalid(1)
hostinfo.logger(f.l).Error("Failed to derive multiport lane session", "error", err, "lane", lane)
return
}
if ci == nil {
// A lane this tunnel doesn't have: a stale lane from a tunnel that has
// since rolled, or a peer sending above what it advertised. Dropping
// silently is right for both — a recv_error would tear down a
// perfectly good base tunnel on the strength of one odd packet.
f.messageMetrics.RxInvalid(1)
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("Unknown multiport lane", "from", via, "header", h)
}
return
}
}
if len(packet) < header.Len+ci.dKey.Overhead() {
f.messageMetrics.RxInvalid(1)
if f.l.Enabled(context.Background(), slog.LevelDebug) {
f.l.Debug("packet too small", "from", via, "length", len(packet))
@@ -118,7 +155,7 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
// All remaining packets are encrypted
if isMessageRelay {
// Relay packets are special, this branch should always early-return
err = hostinfo.ConnectionState.VerifyRelay(f.l, h.MessageCounter, packet, rxc.nb)
err = ci.VerifyRelay(f.l, h.MessageCounter, packet, rxc.nb)
if err != nil {
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("Failed to verify relay packet", "error", err, "from", via, "header", h)
@@ -129,7 +166,7 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
return
}
out, err := hostinfo.ConnectionState.Decrypt(f.l, h.MessageCounter, packet, rxc.nb)
out, err := ci.Decrypt(f.l, h.MessageCounter, packet, rxc.nb)
if err != nil {
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("Failed to decrypt packet", "error", err, "from", via, "header", h)
@@ -137,8 +174,17 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
return
}
// Roam before we respond
f.handleHostRoaming(hostinfo, via)
if !laneCached {
// The packet decrypted, so the peer really is using this lane and the session we derived for it is valid
ci = hostinfo.lanes.installSession(f.l, lane, ci, h.MessageCounter)
}
// Roam before we respond, but only on the base tunnel: a lane's source
// address is a per-lane 4-tuple, not the tunnel's remote, and letting it
// roam the hostinfo would point every non-lane packet at a lane port.
if lane == 0 {
f.handleHostRoaming(hostinfo, via)
}
f.connectionManager.In(hostinfo)
switch h.Type {
@@ -160,7 +206,9 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
case header.TestReply:
// No-op, useful for the Roaming and connectionManager side-effects above
case header.TestRequest:
if header.MaxOverhead+len(out) > len(rxc.scratch) {
const maxCipherOverhead = 16 //todo we use this too often, needs a real importable const
const maxOverhead = header.Len + header.Len + maxCipherOverhead + maxCipherOverhead
if maxOverhead+len(out) > len(rxc.scratch) {
// A reply that cannot fit in scratch is dropped no matter the log level.
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("dropping oversized test request", "payloadLen", len(out), "from", via)
@@ -168,6 +216,10 @@ func (f *Interface) readOutsidePackets(via ViaSender, packet []byte, rxc *rxCont
return
}
f.send(header.Test, header.TestReply, hostinfo.ConnectionState, hostinfo, out, rxc.nb, rxc.scratch[:0])
case header.LaneProbe:
f.handleLaneProbe(hostinfo, lane, out, rxc)
case header.LaneProbeAck:
f.handleLaneProbeAck(hostinfo, out)
default:
hostinfo.logger(f.l).Error("IsValidSubType was true, but unexpected test subtype seen", "from", via, "header", h)
return
@@ -212,6 +264,7 @@ func (f *Interface) handleOutsideRelayPacket(hostinfo *HostInfo, via ViaSender,
relayHI: hostinfo,
relay: relay,
IsRelayed: true,
SockIdx: via.SockIdx,
}
f.readOutsidePackets(via, signedPayload, rxc)
case ForwardingType:
@@ -261,16 +314,11 @@ func (f *Interface) handleOutsideRelayPacket(hostinfo *HostInfo, via ViaSender,
// closeTunnel closes a tunnel locally, it does not send a closeTunnel packet to the remote
func (f *Interface) closeTunnel(hostInfo *HostInfo) {
// Hold the hostmap lock across both the delete and the lighthouse eviction,
// to avoid a race with handshake completion.
f.hostMap.Lock()
defer f.hostMap.Unlock()
if !f.hostMap.unlockedDeleteHostInfo(hostInfo) {
return
final := f.hostMap.DeleteHostInfo(hostInfo)
if final {
// We no longer have any tunnels with this vpn addr, clear learned lighthouse state to lower memory usage
f.lightHouse.DeleteVpnAddrs(hostInfo.vpnAddrs)
}
// We no longer have any tunnels with this vpn addr, clear learned lighthouse state to lower memory usage
f.lightHouse.DeleteVpnAddrs(hostInfo.vpnAddrs)
}
// sendCloseTunnel is a helper function to send a proper close tunnel packet to a remote
@@ -357,7 +405,7 @@ func parseV6(data []byte, incoming bool, fp *firewall.ParsedPacket) error {
// and cannot drift into misreading an unknown protocol (SCTP, GRE, etc.) as a forged transport.
proto, offset, isFragment, anyFragment, err := iputil.IPv6FindUpperProtocol(data)
if err != nil {
return err
return ErrIPv6PacketTooShort
}
fp.Protocol = proto
@@ -482,6 +530,8 @@ func (f *Interface) handleOutsideMessagePacket(hostinfo *HostInfo, messageCounte
dropReason := f.firewall.Drop(rxc.fwPacket.Packet, true, hostinfo, f.pki.GetCAPool(), rxc.ctCache.Get())
if dropReason != nil {
// The reject rides the base tunnel: it is a control response, not lane
// data, and the lane it arrived on says nothing about where it belongs.
f.rejectOutside(out, hostinfo.ConnectionState, hostinfo, rxc.nb, rxc.scratch, rxc.q)
if f.l.Enabled(context.Background(), slog.LevelDebug) {
hostinfo.logger(f.l).Debug("dropping inbound packet", "fwPacket", rxc.fwPacket, "reason", dropReason)
@@ -495,17 +545,20 @@ func (f *Interface) handleOutsideMessagePacket(hostinfo *HostInfo, messageCounte
}
}
func (f *Interface) maybeSendRecvError(endpoint netip.AddrPort, index uint32) {
func (f *Interface) maybeSendRecvError(endpoint netip.AddrPort, index uint32, q int) {
if f.sendRecvErrorConfig.ShouldRecvError(endpoint) {
f.sendRecvError(endpoint, index)
f.sendRecvError(endpoint, index, q)
}
}
func (f *Interface) sendRecvError(endpoint netip.AddrPort, index uint32) {
// sendRecvError replies from the socket the offending packet arrived on (q).
// A lane peer's spoof guard compares our source addr against the lane's
// remote, so a reply from the base port would be discarded.
func (f *Interface) sendRecvError(endpoint netip.AddrPort, index uint32, q int) {
f.messageMetrics.Tx(header.RecvError, 0, 1)
b := header.Encode(make([]byte, header.Len), header.Version, header.RecvError, 0, index, 0)
_ = f.outside.WriteTo(b, endpoint)
_ = f.writers[q].WriteTo(b, endpoint)
if f.l.Enabled(context.Background(), slog.LevelDebug) {
f.l.Debug("Recv error sent",
"index", index,
+7 -7
View File
@@ -117,7 +117,7 @@ func Test_newPacket_v6(t *testing.T) {
require.NoError(t, err)
err = newPacket(buffer.Bytes(), true, p)
require.ErrorIs(t, err, iputil.ErrIPv6CouldNotFindPayload)
require.ErrorIs(t, err, ErrIPv6PacketTooShort)
// A v6 packet with a hop-by-hop extension
// ICMPv6 Payload (Echo Request)
@@ -151,12 +151,12 @@ func Test_newPacket_v6(t *testing.T) {
// A full IPv6 header and 1 byte in the first extension, but missing
// the length byte.
err = newPacket(buffer.Bytes()[:41], true, p)
require.ErrorIs(t, err, iputil.ErrIPv6CouldNotFindPayload)
require.ErrorIs(t, err, ErrIPv6PacketTooShort)
// A full IPv6 header plus 1 full extension, but only 1 byte of the
// next layer, missing length byte
err = newPacket(buffer.Bytes()[:49], true, p)
require.ErrorIs(t, err, iputil.ErrIPv6CouldNotFindPayload)
require.ErrorIs(t, err, ErrIPv6PacketTooShort)
err = nil
// A good ICMP packet
@@ -354,12 +354,12 @@ func Test_newPacket_v6(t *testing.T) {
// Ensure buffer bounds checking during processing, a truncated AH header can't reach the payload
err = newPacket(b[:41], true, p)
require.ErrorIs(t, err, iputil.ErrIPv6CouldNotFindPayload)
require.ErrorIs(t, err, ErrIPv6PacketTooShort)
// Invalid AH header
b = buffer.Bytes()
err = newPacket(b, true, p)
require.ErrorIs(t, err, iputil.ErrIPv6CouldNotFindPayload)
require.ErrorIs(t, err, ErrIPv6PacketTooShort)
}
func Test_newPacket_ipv6Fragment(t *testing.T) {
@@ -466,7 +466,7 @@ func Test_newPacket_ipv6Fragment(t *testing.T) {
// Too short of a fragment packet
err = newPacket(secondFrag[:len(secondFrag)-10], false, p)
require.ErrorIs(t, err, iputil.ErrIPv6CouldNotFindPayload)
require.ErrorIs(t, err, ErrIPv6PacketTooShort)
}
func BenchmarkParseV6(b *testing.B) {
@@ -707,7 +707,7 @@ func Test_newPacket_v6ExtHeaderPastBuffer(t *testing.T) {
pkt[40] = byte(layers.IPProtocolSCTP) // Dest Options next header = SCTP
pkt[41] = 255 // declared length (255+1)*8 = 2048, past the 48 byte buffer
require.ErrorIs(t, newPacket(pkt, true, p), iputil.ErrIPv6CouldNotFindPayload)
require.ErrorIs(t, newPacket(pkt, true, p), ErrIPv6PacketTooShort)
}
// Test_newPacket_v6ExtHeaderConfusion is a regression test for parseV6 walking any unrecognized
+2 -2
View File
@@ -127,7 +127,7 @@ func TestPseudoSumIPv6MatchesReference(t *testing.T) {
func TestIPv4HdrChecksumMatchesReference(t *testing.T) {
rng := rand.New(rand.NewSource(0x1791))
for _, hdrLen := range []int{20, 24, 40, 60} {
for trial := range 200 {
for trial := 0; trial < 200; trial++ {
hdr := make([]byte, hdrLen)
rng.Read(hdr)
hdr[0] = 0x40 | byte(hdrLen/4)
@@ -156,7 +156,7 @@ func TestIPv4HdrChecksumMatchesReference(t *testing.T) {
// the way a receiver does (pseudo-header + L4 must sum to all-ones).
func TestChecksumSeedReceiverAcceptance(t *testing.T) {
rng := rand.New(rand.NewSource(0x1826))
for trial := range 200 {
for trial := 0; trial < 200; trial++ {
src := [4]byte{byte(rng.Intn(256)), byte(rng.Intn(256)), byte(rng.Intn(256)), byte(rng.Intn(256))}
dst := [4]byte{byte(rng.Intn(256)), byte(rng.Intn(256)), byte(rng.Intn(256)), byte(rng.Intn(256))}
payLen := rng.Intn(1500)
+1 -1
View File
@@ -540,7 +540,7 @@ func TestCoalescerCapBySegments(t *testing.T) {
c := newTestTCPCoalescer(t, w)
pay := make([]byte, 512)
seq := uint32(1000)
for range tcpCoalesceMaxSegs + 5 {
for i := 0; i < tcpCoalesceMaxSegs+5; i++ {
if err := c.Commit(buildTCPv4(seq, tcpAck, pay)); err != nil {
t.Fatal(err)
}
+28 -9
View File
@@ -13,19 +13,36 @@ type batchWriter interface {
// One SendBatch is owned by each listenIn goroutine; no locking is needed.
// Slots are backed by an Arena (see its docs)
type SendBatch struct {
out batchWriter
bufs [][]byte
dsts []netip.AddrPort
arena *Arena
out batchWriter
bufs [][]byte
dsts []netip.AddrPort
// arena backs the slots. borrowed means someone else owns it and Flush must
// leave it alone.
arena *Arena
borrowed bool
}
// NewSendBatch makes a SendBatch with batchCap slots and an arenaSize byte buffer for slices to back those slots
func NewSendBatch(out batchWriter, batchCap, arenaSize int) *SendBatch {
return newSendBatch(out, batchCap, NewArena(arenaSize), false)
}
// NewSendBatchSharedArena makes a SendBatch that borrows arena rather than
// allocating one. Several batches over different sockets can then split a single
// slab, which is what multiport lanes need: the slab is the expensive part of a
// SendBatch and one routine may hold a batch per lane. Flush leaves the arena
// alone, so the owner must Reset it once every batch sharing it is drained.
func NewSendBatchSharedArena(out batchWriter, batchCap int, arena *Arena) *SendBatch {
return newSendBatch(out, batchCap, arena, true)
}
func newSendBatch(out batchWriter, batchCap int, arena *Arena, borrowed bool) *SendBatch {
return &SendBatch{
out: out,
bufs: make([][]byte, 0, batchCap),
dsts: make([]netip.AddrPort, 0, batchCap),
arena: NewArena(arenaSize),
out: out,
bufs: make([][]byte, 0, batchCap),
dsts: make([]netip.AddrPort, 0, batchCap),
arena: arena,
borrowed: borrowed,
}
}
@@ -54,6 +71,8 @@ func (b *SendBatch) Flush() (int, error) {
clear(b.bufs)
b.bufs = b.bufs[:0]
b.dsts = b.dsts[:0]
b.arena.Reset()
if !b.borrowed {
b.arena.Reset()
}
return written, err
}
+2 -2
View File
@@ -28,7 +28,7 @@ func TestSendBatchReserveCommitFlush(t *testing.T) {
b := NewSendBatch(fw, 4, 32)
ap := netip.MustParseAddrPort("10.0.0.1:4242")
for i := range 4 {
for i := 0; i < 4; i++ {
slot := b.Reserve(32)
if cap(slot) != 32 {
t.Fatalf("slot %d: cap=%d want 32", i, cap(slot))
@@ -72,7 +72,7 @@ func TestSendBatchSlotsDoNotOverlap(t *testing.T) {
b := NewSendBatch(fw, 3, 8)
ap := netip.MustParseAddrPort("10.0.0.1:80")
for i := range 3 {
for i := 0; i < 3; i++ {
s := b.Reserve(8)
pkt := append(s[:0], byte(0xA0+i), byte(0xB0+i))
b.Commit(pkt, ap)
+3 -3
View File
@@ -129,7 +129,7 @@ func TestUDPCoalescerCoalescesEqualSized(t *testing.T) {
w := &fakeTunWriter{gsoEnabled: true}
c := newTestUDPCoalescer(t, w)
pay := make([]byte, 1200)
for range 3 {
for i := 0; i < 3; i++ {
if err := c.Commit(buildUDPv4(1000, 53, pay)); err != nil {
t.Fatal(err)
}
@@ -259,7 +259,7 @@ func TestUDPCoalescerCapsAtMaxSegs(t *testing.T) {
w := &fakeTunWriter{gsoEnabled: true}
c := newTestUDPCoalescer(t, w)
pay := make([]byte, 100)
for range udpCoalesceMaxSegs + 5 {
for i := 0; i < udpCoalesceMaxSegs+5; i++ {
if err := c.Commit(buildUDPv4(1000, 53, pay)); err != nil {
t.Fatal(err)
}
@@ -317,7 +317,7 @@ func TestUDPCoalescerIPv6Coalesces(t *testing.T) {
w := &fakeTunWriter{gsoEnabled: true}
c := newTestUDPCoalescer(t, w)
pay := make([]byte, 1200)
for range 3 {
for i := 0; i < 3; i++ {
if err := c.Commit(buildUDPv6(1000, 53, pay)); err != nil {
t.Fatal(err)
}
+1 -1
View File
@@ -151,7 +151,7 @@ func TestChecksumTailPaths(t *testing.T) {
offsets := []int{0, 1, 3, 7, 15} // mix of aligned and odd starts
for k := 0; k <= maxK; k++ {
for tail := range 64 {
for tail := 0; tail < 64; tail++ {
length := 64*k + tail
for _, seed := range seeds {
for _, off := range offsets {
-57
View File
@@ -1,57 +0,0 @@
//go:build !e2e_testing
package overlay
import (
"fmt"
"log/slog"
"golang.org/x/sys/windows"
"golang.org/x/sys/windows/registry"
)
var tcpipInterfaceKeys = []string{
`SYSTEM\CurrentControlSet\Services\Tcpip\Parameters\Interfaces`,
`SYSTEM\CurrentControlSet\Services\Tcpip6\Parameters\Interfaces`,
}
// disableDNSRegistration prevents publishing overlay addresses in the host's
// DNS records, where clients outside the overlay cannot reach them.
// Failures are non-fatal so registry access cannot prevent tunnel creation.
func disableDNSRegistration(l *slog.Logger, guid windows.GUID) {
id := guid.String()
for _, base := range tcpipInterfaceKeys {
path := base + `\` + id
if err := setDNSRegistrationValues(path); err != nil {
l.Warn("Failed to disable DNS registration on the tun adapter", "error", err, "key", path)
continue
}
l.Debug("Disabled DNS registration on the tun adapter", "key", path)
}
}
func setDNSRegistrationValues(path string) error {
// The stack may not have created the adapter's interface key yet.
k, _, err := registry.CreateKey(registry.LOCAL_MACHINE, path, registry.SET_VALUE)
if err != nil {
return fmt.Errorf("open HKLM\\%s: %w", path, err)
}
defer k.Close()
// Neither RegistrationEnabled nor DisableDynamicUpdate alone reliably
// prevents registration, so all three settings are required.
values := []struct {
name string
data uint32
}{
{"RegistrationEnabled", 0},
{"DisableDynamicUpdate", 1},
{"MaxNumberOfAddressesToRegister", 0},
}
for _, v := range values {
if err := k.SetDWordValue(v.name, v.data); err != nil {
return fmt.Errorf("set %s: %w", v.name, err)
}
}
return nil
}
+1
View File
@@ -1,4 +1,5 @@
//go:build !e2e_testing
// +build !e2e_testing
package overlay
+1
View File
@@ -1,4 +1,5 @@
//go:build !e2e_testing
// +build !e2e_testing
package overlay
Loaded 100 of 155 files, more files were not shown because too many files have changed in this diff. Show more