Repository navigation
Conversation
Update gvisor.dev/gvisor from 9680d69bf798 (2026-04-13) to 3b87f090dbda1fefedbfbe94183324e9c8e13494 (2026-10-08) on the go branch. This is the newest go branch snapshot whose go.mod still requires Go 1.26; later snapshots require Go 1.27.3. Notable changes in the vendored packages: - tcp: strict RFC 5961 RST sequence validation, DF bit on the handshake, FIN_WAIT1 when a FIN is queued, CUBIC/RACK fixes, and a segment leak fix when TIME-WAIT reuse fails - udp: checksum 0 handling, PacketBuffer leak fix in the forwarder, packet overhead accounted in receive buffer limits - ipv4/ipv6: no echo replies to broadcast/multicast, reassembly length limit, stricter parsing - ICMP echo requests consumed by a default transport handler no longer get the built-in reply (8b0e16201) go mod tidy raises the go directive to 1.26.3 and bumps the indirect golang.org/x/exp and golang.org/x/time dependencies. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Yevhen Vydolob <yvydolob@redhat.com>
The ICMP forwarder's default handler returned true for echo requests
sent to an address the stack owns, intending to leave them to gvisor's
built-in echo reply. Since gvisor 8b0e16201 ("netstack: let ICMP echo
handlers consume requests"), a default handler returning true means it
consumed the packet, and gvisor no longer sends its own reply. Pings
from the VM to the gateway's address therefore went unanswered.
Return false for these packets so gvisor still replies. Older gvisor
versions ignore the return value on this path, so the change is safe
with them too.
Fixes the "should leave pings to the gateway's own address to the
stack" forwarder test.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Yevhen Vydolob <yvydolob@redhat.com>
Contributor
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: evidolob The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:
Approvers can indicate their approval by writing |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps
gvisor.dev/gvisorfrom9680d69bf798(2026-04-13) to3b87f090dbda1fefedbfbe94183324e9c8e13494(2026-10-08) on the upstreamgobranch, and adapts the ICMP forwarder to a gvisor behavior change.Why this commit
3b87f090dbdais the newestgobranch snapshot whosego.modstill requires Go 1.26. Later snapshots require Go 1.27.3 and also include thepkg/stategraph-engine rewrite.go mod tidyraises ourgodirective from 1.26.0 to 1.26.3; the CI matrix (1.26.x, 1.27.x) covers that.Notable upstream changes
ICMP fix (second commit)
gvisor
8b0e16201("netstack: let ICMP echo handlers consume requests") treats a default transport handler returningtrueas having consumed the echo request, and then skips the built-in reply. Our forwarder returnedtruefor pings to the stack's own addresses (e.g. the gateway) to mean "leave it to the stack", so with the bump those pings went unanswered. Theshould leave pings to the gateway's own address to the stacktest caught this. The handler now returnsfalsefor those packets. This also works with the previous gvisor, which ignores the return value on this path.The fix is a separate commit, so the forwarder test fails at the first commit on its own.
Testing
make crossgo vet ./...go test ./pkg/...: all pass, including the forwarder ICMP tests (fails without the second commit)🤖 Generated with Claude Code