fix: bound ipns cache-control to record eol#1166
Open
lidel wants to merge 7 commits into
Open
Conversation
max-age plus stale-while-revalidate could extend past an IPNS record's EOL, letting caches serve an expired record that then fails validation. Cap max-age to the remaining validity and size the stale window to whatever validity is left, so max-age+stale never crosses EOL. An already-expired record is returned with Cache-Control: no-store.
Codecov Report❌ Patch coverage is
@@ Coverage Diff @@
## main #1166 +/- ##
==========================================
+ Coverage 63.39% 63.45% +0.05%
==========================================
Files 268 269 +1
Lines 26967 27002 +35
==========================================
+ Hits 17096 17133 +37
+ Misses 8147 8142 -5
- Partials 1724 1727 +3
... and 6 files with indirect coverage changes 🚀 New features to boost your workflow:
|
gammazero
approved these changes
Jun 3, 2026
1 task
A record's TTL is read from CBOR without a sign check, so a negative value would make min(ttl, remainingValidity) emit a malformed Cache-Control: max-age=-N. Floor ttl at zero before capping.
The raw IPNS record handler set max-age straight from record.TTL(), bypassing the resolver's EOL clamp, so a cache could reuse the record past its signature expiry; a negative TTL also yielded a negative max-age. Cap max-age to the remaining validity and floor it at zero.
calculateBestTTL clamped a record TTL down to the remaining EOL validity but never floored a negative record TTL, so a malformed record with a future EOL leaked a negative duration through namesys.Result.TTL. Floor the record TTL at zero.
GET /routing/v1/ipns/{name} assumed the default 48h record lifetime
for any record whose ValidityType was not EOL (or could not be
parsed), advertising a stale-serving window for a record whose
expiration is unknown. Treat such records as uncacheable instead.
On a cache hit the resolver returned the record's original TTL, so a hit late in the cache window re-advertised the full TTL and could let a downstream cache outlive the record's EOL. Return the cache entry's remaining lifetime instead, which is already bounded by the EOL.
A negative TTL was stored verbatim: the CBOR field went negative and the v1 uint64 protobuf field underflowed to a huge value. NewRecord now floors the TTL at zero, and Validate rejects a record whose CBOR TTL is negative. Tests that need a genuinely negative TTL build records at the wire level via internal/ipnstest, since NewRecord no longer mints them.
1 task
Member
Author
|
Hey @gammazero, heads-up before this lands: it grew quite a bit since your approval (which was on just the initial
I smoke-tested all of it in ipfs/kubo#11349 (bumps kubo to this branch's tip, CI green) and used it there to add user-facing Could you give the post-approval commits a final look? If they look good to you, this is ready to merge and cut a boxo release, which is what lets someguy/delegated-ipfs.dev pick it up and clear the service-worker-gateway 500s (ipfs/service-worker-gateway#1072). Thanks for the earlier review! |
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
3 participants
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.
Problem
GET /routing/v1/ipns/{name}advertised a cache window that outlived the record:max-agewas the record TTL andstale-while-revalidate/stale-if-errorcovered the full remaining validity, so the stale-serving window ran past the record's EOL by up to onemax-age. A cache (the CDN in front ofdelegated-ipfs.dev, a browser, or a Helia HTTP cache) then served a record whose signature had already expired, which validating clients reject, surfacing as sporadic 500s in the service-worker gateway. The other IPNS surfaces had the same gap, and a malformed negative TTL propagated unchecked, even underflowing the v1uint64TTL field.Fix
routing/http/server: capmax-ageto the record's remaining validity and size the stale window to the validity left after it, somax-age+stale never crosses EOL; returnno-storefor a record that is expired or carries an unrecognizedValidityType.gateway: the raw record endpoint (GET /ipns/{name}?format=ipns-record) capsmax-ageto the remaining validity.namesys: the resolver floors a negative record TTL, and a cache hit reports the entry's remaining lifetime instead of the original TTL.ipns:NewRecordfloors a negative TTL (also closing the v1uint64underflow) andValidaterejects a record carrying one.No cache reuses an IPNS response past the record's EOL, and a negative TTL can no longer surface through any IPNS API.