fuzz: use ImmediateBackgroundTaskRunner to silence DEBUG_LOCKORDER #35284

pull Crypt-iQ wants to merge 1 commits into bitcoin:master from Crypt-iQ:05122026/imm_bg_std_thread changing 1 files +13 −0

Crypt-iQ commented at 4:04 PM on May 13, 2026: contributor
DEBUG_LOCKORDER was reporting a false positive deadlock with the cmpctblock fuzz harness when using ImmediateTaskRunner. Since it is single-threaded, ImmediateTaskRunner callbacks added LockOrders that could never happen outside of a fuzz test.

First a block would get connected:
- LOCK(mempool.cs)
- BlockConnected (fuzz test runs in same thread)
- LOCK(m_tx_download_mutex)
Then a later iteration of the LIMITED_WHILE would send a TX:
- LOCK(m_tx_download_mutex)
- LOCK(mempool.cs)
causing a false positive deadlock. Normally, the BlockConnected callback would run in a different thread and no deadlock is reported.

Fix this by launching a thread that runs the callback and is immediately joined. I compared this PR to another branch (https://github.com/Crypt-iQ/bitcoin/commit/0028847c6b8b937031228e9a1fce83982976d86e) that uses a ThreadPool and found this PR to be more stable and ~4% quicker.

fuzz: use ImmediateBackgroundTaskRunner to silence DEBUG_LOCKORDER

DEBUG_LOCKORDER was reporting a false positive deadlock with the
cmpctblock fuzz harness when using ImmediateTaskRunner. Since it is
single-threaded, ImmediateTaskRunner callbacks added LockOrders that
could never happen outside of a fuzz test.

First a block would get connected:
* LOCK(mempool.cs)
* BlockConnected (fuzz test runs in same thread)
* LOCK(m_tx_download_mutex)

Then a later iteration of the LIMITED_WHILE would send a TX:
* LOCK(m_tx_download_mutex)
* LOCK(mempool.cs)

causing a false positive deadlock. Normally, the BlockConnected
callback would run in a different thread and no deadlock is reported.

Fix this by launching a thread that runs the callback and is
immediately joined.

801d36f55b

DrahtBot added the label Fuzzing on May 13, 2026
DrahtBot commented at 4:05 PM on May 13, 2026: contributor


The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.



Code Coverage & Benchmarks

For details see: https://corecheck.dev/bitcoin/bitcoin/pulls/35284.



Reviews

See the guideline for information on the review process.

Type Reviewers

ACK maflcko, sedited

If your review is incorrectly listed, please copy-paste <code></code> into the comment that the bot should ignore.



Conflicts

Reviewers, this pull request conflicts with the following ones:
- #33966 (refactor: disentangle miner startup defaults from runtime options by Sjors)
If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.
Crypt-iQ commented at 4:06 PM on May 13, 2026: contributor

Leaving as draft while I run this to verify there are no timeouts over extended periods of time and/or with multiple cores.

cc @maflcko, let me know if you want co-author as these are mostly your suggestions.
maflcko commented at 4:22 PM on May 13, 2026: member

lgtm ACK 801d36f55b6d421f26d2376327699a04bd22e602

(fwiw)
Crypt-iQ marked this as ready for review on May 14, 2026
Crypt-iQ commented at 11:26 AM on May 14, 2026: contributor

Leaving as draft while I run this to verify there are no timeouts over extended periods of time and/or with multiple cores.

Ran for about 16 hours on 20 cores and no crashes / timeouts, so took out of draft
sedited approved
sedited commented at 12:12 PM on May 14, 2026: contributor

ACK 801d36f55b6d421f26d2376327699a04bd22e602
sedited merged this on May 14, 2026
sedited closed this on May 14, 2026
rustaceanrob referenced this in commit a4ca01c5dc on May 14, 2026
maflcko commented at 2:04 PM on May 15, 2026: member

Fix this by launching a thread that runs the callback and is immediately joined. I compared this PR to another branch (Crypt-iQ@0028847) that uses a ThreadPool and found this PR to be more stable and ~4% quicker.

Just to clarify: You tested with AFL? Because when I use libFuzzer, the pool one is faster 20%-40% (depending on whether sanitizers are enabled or not)
Crypt-iQ commented at 2:20 PM on May 15, 2026: contributor

Just to clarify: You tested with AFL? Because when I use libFuzzer, the pool one is faster 20%-40% (depending on whether sanitizers are enabled or not)

Yeah, I tested with AFL on Debian without sanitizers

Type	Reviewers
ACK	maflcko, sedited

Contributors

Labels

fuzz: use ImmediateBackgroundTaskRunner to silence DEBUG_LOCKORDER #35284

Code Coverage & Benchmarks

Reviews

Conflicts