wfe: fix a race in orderForDisplay
#500
Merged
Conversation
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
I have an integration test for an ACME extension ([1]) that instantiate
Pebble. I noticed that I would sometimes get test failures under
`go test -race` like this:
```
==================
WARNING: DATA RACE
Write at 0x00c00026cd80 by goroutine 143:
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).orderForDisplay.func2()
/home/timg/source/pebble/wfe/wfe.go:1876 +0x255
math/rand.(*Rand).Shuffle()
/usr/local/go/src/math/rand/rand.go:265 +0x96
math/rand.Shuffle()
/usr/local/go/src/math/rand/rand.go:470 +0x38
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).orderForDisplay()
/home/timg/source/pebble/wfe/wfe.go:1875 +0x204
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).Order()
/home/timg/source/pebble/wfe/wfe.go:2021 +0x5ed
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).Order-fm()
<autogenerated>:1 +0x69
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).HandleFunc.func1()
/home/timg/source/pebble/wfe/wfe.go:306 +0xa09
github.com/letsencrypt/pebble/v2/wfe.wfeHandlerFunc.ServeHTTP()
/home/timg/source/pebble/wfe/wfe.go:146 +0x56
github.com/letsencrypt/pebble/v2/wfe.(*topHandler).ServeHTTP()
/home/timg/source/pebble/wfe/wfe.go:158 +0x5b
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).HandleFunc.StripPrefix.func2()
/usr/local/go/src/net/http/server.go:2282 +0x471
net/http.HandlerFunc.ServeHTTP()
/usr/local/go/src/net/http/server.go:2220 +0x47
net/http.(*ServeMux).ServeHTTP()
/usr/local/go/src/net/http/server.go:2747 +0x255
net/http.serverHandler.ServeHTTP()
/usr/local/go/src/net/http/server.go:3210 +0x2a1
net/http.(*conn).serve()
/usr/local/go/src/net/http/server.go:2092 +0x12a4
net/http.(*Server).Serve.gowrap3()
/usr/local/go/src/net/http/server.go:3360 +0x4f
Previous read at 0x00c00026cd80 by goroutine 274:
github.com/letsencrypt/pebble/v2/ca.(*CAImpl).CompleteOrder()
/home/timg/source/pebble/ca/ca.go:448 +0x76f
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).FinalizeOrder.func1()
/home/timg/source/pebble/wfe/wfe.go:2215 +0x8d
Goroutine 143 (running) created at:
net/http.(*Server).Serve()
/usr/local/go/src/net/http/server.go:3360 +0x8ec
net/http.(*Server).ServeTLS()
/usr/local/go/src/net/http/server.go:3401 +0x706
net/http.ServeTLS()
/usr/local/go/src/net/http/server.go:2875 +0x269
github.com/tgeoghegan/oidf-box/test.setupPebble.func1()
/home/timg/source/oidf-box/test/harness.go:297 +0x19c
Goroutine 274 (running) created at:
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).FinalizeOrder()
/home/timg/source/pebble/wfe/wfe.go:2214 +0x2fa4
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).FinalizeOrder-fm()
<autogenerated>:1 +0x69
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).HandleFunc.func1()
/home/timg/source/pebble/wfe/wfe.go:306 +0xa09
github.com/letsencrypt/pebble/v2/wfe.wfeHandlerFunc.ServeHTTP()
/home/timg/source/pebble/wfe/wfe.go:146 +0x56
github.com/letsencrypt/pebble/v2/wfe.(*topHandler).ServeHTTP()
/home/timg/source/pebble/wfe/wfe.go:158 +0x5b
github.com/letsencrypt/pebble/v2/wfe.(*WebFrontEndImpl).HandleFunc.StripPrefix.func2()
/usr/local/go/src/net/http/server.go:2282 +0x471
net/http.HandlerFunc.ServeHTTP()
/usr/local/go/src/net/http/server.go:2220 +0x47
net/http.(*ServeMux).ServeHTTP()
/usr/local/go/src/net/http/server.go:2747 +0x255
net/http.serverHandler.ServeHTTP()
/usr/local/go/src/net/http/server.go:3210 +0x2a1
net/http.(*conn).serve()
/usr/local/go/src/net/http/server.go:2092 +0x12a4
net/http.(*Server).Serve.gowrap3()
/usr/local/go/src/net/http/server.go:3360 +0x4f
==================
```
What's happening is that `CAImpl.CompleteOrder` is fetching an `Order`
from the "database" (which is in-memory) and iterating over the
identifiers therein. `WebFrontEndImpl.orderForDisplay` fetches an
`Order` from the database and then wants to construct a shuffled version
of the order to return to the client to ensure nobody makes assumptions
about the layout of these objects.
`orderForDisplay` tries to take a copy of the `acme.Order`, but
evidently that copies *pointers* to the `[]acme.Identifier Identifiers`
and `[]string Authorizations` fields, which means shuffling them in the
copy shuffles them in the DB object and makes the race detector
unhappy.
Explicit `slices.Clone` calls make sure we're not mutating objects from
the database.
[1]: https://github.com/tgeoghegan/oidf-box
tgeoghegan
commented
May 20, 2025
wfe/wfe.go
Outdated
| // below could mutate the order in the database, causing data races. | ||
| result := order.Order | ||
| result.Identifiers = slices.Clone(order.Order.Identifiers) | ||
| result.Authorizations = slices.Clone(order.Order.Authorizations) |
Contributor
Author
There was a problem hiding this comment.
I must admit that I'm not certain if this is the "right", Goish, Gothonic way to fix this, so I'm happy to be shown better solutions.
Contributor
There was a problem hiding this comment.
This definitely works. But it has the downside of not catching any new pointer-valued fields which are added to the Order type in the future.
I think it would be slightly more idiomatic to instead say:
result := acme.Order{
Status: order.Order.Status,
Error: order.Order.Error,
etc
Identifiers: slices.Clone(order.Order.Identifiers),
etc
}This has a similar downside -- if a new field is added to the Order type, it may not get added to this operation. But that will result in the field being missing, rather than subtler bugs, so is much more likely to be caught by unit or integration tests.
This solution is so much more verbose that I'm honestly not sure it counts as "better", but I do think it has worthwhile advantages. Up to you!
Contributor
Author
There was a problem hiding this comment.
Yeah, the thing where unmentioned struct fields get filled with the zero value kinda works against us here. I like yours better because then at least we're not mutating an object after creating it, but I don't know how to do better.
aarongable
previously approved these changes
May 20, 2025
wfe/wfe.go
Outdated
| // below could mutate the order in the database, causing data races. | ||
| result := order.Order | ||
| result.Identifiers = slices.Clone(order.Order.Identifiers) | ||
| result.Authorizations = slices.Clone(order.Order.Authorizations) |
Contributor
There was a problem hiding this comment.
This definitely works. But it has the downside of not catching any new pointer-valued fields which are added to the Order type in the future.
I think it would be slightly more idiomatic to instead say:
result := acme.Order{
Status: order.Order.Status,
Error: order.Order.Error,
etc
Identifiers: slices.Clone(order.Order.Identifiers),
etc
}This has a similar downside -- if a new field is added to the Order type, it may not get added to this operation. But that will result in the field being missing, rather than subtler bugs, so is much more likely to be caught by unit or integration tests.
This solution is so much more verbose that I'm honestly not sure it counts as "better", but I do think it has worthwhile advantages. Up to you!
aarongable
previously approved these changes
May 20, 2025
wfe/wfe.go
Outdated
Comment on lines
1843
to
1844
| // use as the result. We have to make sure to copy the identifiers and | ||
| // authorizations *slices* and not *pointers to the slices* or the mutations |
Contributor
There was a problem hiding this comment.
A more correct way to phrase this would be that we have to copy the contents of the idents and authzs slices, rather than copying the slices themselves, because slices are just pointers with some extra metadata.
Those fields aren't "pointers to slices" (that would be type *[]string).
aarongable
approved these changes
May 21, 2025
kwatson
added a commit
to kwatson/letsencrypt-pebble
that referenced
this pull request
Jun 9, 2025
* 'main' of https://github.com/letsencrypt/pebble: (35 commits) add overriding of ARI response (letsencrypt#501) wfe: fix a race in `orderForDisplay` (letsencrypt#500) Bump golang.org/x/ dependencies (letsencrypt#499) currectly triggers BadSignatureAlgorithmProblem at JWS parse time (letsencrypt#492) use newer validation subdomain for dns-account-01 (fix CI eggsampler/acme error) (letsencrypt#498) Orders don't have a "deactivated" status. (letsencrypt#301) Update golangci-lint (letsencrypt#488) build(deps): bump github.com/go-jose/go-jose/v4 from 4.0.4 to 4.0.5 (letsencrypt#487) Truncate ARI timestamps to millisecond resolution (letsencrypt#485) return logical and compliant ARI windows for expiring certs (letsencrypt#484) Update dependencies (letsencrypt#481) docs: rm mention of subproblems being unimpl'd (letsencrypt#479) Fix(NOISSUE): Fix docker compose file example in README.md (letsencrypt#475) Add support for ACME Profiles (letsencrypt#473) Simplify KU, EKU, and SKID fields of issued certs (letsencrypt#472) Update golangci-lint to 1.60.2 (letsencrypt#474) Update /x/net for compatibility with go1.23 (letsencrypt#470) Reject extra command line args and fix README invocation (letsencrypt#467) Document exposing API and management ports when not using docker-compose.yaml (letsencrypt#465) Implement latest draft-ietf-acme-ari spec (letsencrypt#461) ...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
I have an integration test for an ACME extension (1) that instantiate Pebble. I noticed that I would sometimes get test failures under
go test -racelike this:What's happening is that
CAImpl.CompleteOrderis fetching anOrderfrom the "database" (which is in-memory) and iterating over the identifiers therein.WebFrontEndImpl.orderForDisplayfetches anOrderfrom the database and then wants to construct a shuffled version of the order to return to the client to ensure nobody makes assumptions about the layout of these objects.orderForDisplaytries to take a copy of theacme.Order, but evidently that copies pointers to the[]acme.Identifier Identifiersand[]string Authorizationsfields, which means shuffling them in the "copy" shuffles them in the DB object and makes the race detector unhappy.