fuzz: reset connman state in p2p targets #35808

1. 
   HowHsu commented at 11:23 AM on July 26, 2026: contributor

   Resets ConnmanTestMsg at the start of each input in cmpctblock and p2p_handshake, matching the other reused-connman fuzz targets and preventing sticky CConnman state from leaking between corpus inputs.

   Before this, deterministic-fuzz-coverage showed single inputs were stable, but all-input directory runs were not:

   cmpctblock, src/net.cpp:4172
   - Branch (4172:9): [True: 1.21k, False: 33.0k]
   + Branch (4172:9): [True: 613, False: 33.6k]
   - Branch (4172:72): [True: 901, False: 311]
   + Branch (4172:72): [True: 497, False: 116]
   

   p2p_handshake, src/net.cpp:4172
   - Branch (4172:9): [True: 98, False: 1.67k]
   + Branch (4172:9): [True: 743, False: 1.03k]
   - Branch (4172:72): [True: 90, False: 8]
   + Branch (4172:72): [True: 612, False: 131]
   

   With the resets, deterministic-fuzz-coverage passed for both cmpctblock and p2p_handshake.

2. DrahtBot added the label Fuzzing on Jul 26, 2026
3. 
   DrahtBot commented at 11:23 AM on July 26, 2026: contributor

   <!--e57a25ab6845829454e8d69fc972939a-->

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

   <!--006a51241073e994b41acfe9ec718e94-->

   Code Coverage & Benchmarks

   For details see: https://corecheck.dev/bitcoin/bitcoin/pulls/35808.

   <!--021abf342d371248e50ceaed478a90ca-->

   Reviews

   See the guideline and AI policy for information on the review process.

   Type	Reviewers
   ACK	nervana21

   If your review is incorrectly listed, please copy-paste <code>&lt;!--meta-tag:bot-skip--&gt;</code> into the comment that the bot should ignore.

   <!--5faf32d7da4f0f540f40219e4f7537a3-->

4. in src/test/fuzz/p2p_handshake.cpp:109 in fd403f9474

    105 | @@ -106,4 +106,5 @@ FUZZ_TARGET(p2p_handshake, .init = ::initialize)
    106 |      }
    107 |  
    108 |      node.connman->StopNodes();
    109 | +    connman.Reset();
   

   ---

   ---

   

   nervana21 commented at 6:26 PM on August 8, 2026:

   fd403f947432067678a9d5e866d06df3e59ad0e1: fuzz: reset connman state in p2p targets

   All other fuzz callers of connman.Reset() are at the start of the input (process_message.cpp, process_messages.cpp, p2p_private_broadcast.cpp).

   Is there a good reason to abandon that pattern here? If not, I'd prefer moving both of these Reset calls to the start of their respective harnesses to stay consistent with the rest of the codebase.

   ---

   

   HowHsu commented at 7:21 AM on August 13, 2026:

   Updated

5. 
   nervana21 commented at 6:26 PM on August 8, 2026: contributor

   Concept ACK

6. HowHsu force-pushed on Aug 13, 2026
7. HowHsu requested review from nervana21 on Aug 13, 2026
8. 
   nervana21 commented at 1:19 PM on August 13, 2026: contributor

   tACK d015b73c90646a1510b4c24447b3d011cd1188b6

9. 
   nervana21 commented at 1:22 PM on August 13, 2026: contributor

   Nit. Commit body still says Reset happens after stopping test nodes. PR description still says after StopNodes().

10. 
    HowHsu commented at 1:46 PM on August 13, 2026: contributor

    Nit. Commit body still says Reset happens after stopping test nodes. PR description still says after StopNodes().

    My bad, I'll update that.

11. fuzz: reset connman state in p2p targets

    Clear ConnmanTestMsg state at the start of each input in cmpctblock and p2p_handshake so sticky CConnman state does not leak between fuzz inputs.

    d29b22d078
12. HowHsu force-pushed on Aug 13, 2026
13. 
    nervana21 commented at 3:26 PM on August 13, 2026: contributor

    re-tACK d29b22d07897058eb69c862ecea8b00325229de2

    No code change. Commit message and PR description now match start-of-input Reset.

    Thanks for addressing nits :)

Contributors

HowHsuDrahtBotnervana21

Review Requested
nervana21