mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [BUG] WARNING: possible recursive locking in hfsplus_get_block (fs/hfsplus/extents.c:264, v7.4-rc3)
@ 2026-11-17  3:01 Zheng Zheng
  2026-11-17 20:50 ` Viacheslav Dubeyko
  0 siblings, 1 reply; 2+ messages in thread
From: Zheng Zheng @ 2026-11-17  3:01 UTC (permalink / raw)
  To: slava; +Cc: glaubitz, frank.li, linux-fsdevel, linux-kernel, syzkaller-bugs

Hi Viacheslav,

We found a reproducible lockdep recursive-locking warning in hfsplus on
v7.4-rc3 with syzkaller. The same signature is present on v7.4-rc2.

The following is the key excerpt from the rc3 external verification log
(`rc3/logs/external-verification/run1-singleproc-serial.log`, hit at 114.00 s):

  [  114.001303][T10445] WARNING: possible recursive locking detected
  [  114.003629][T10445] 7.4.0-rc3-kernsight-kasan #1 Tainted: G            E
  [  114.005869][T10445] --------------------------------------------
  [  114.008132][T10445] syz-executor/10445 is trying to acquire lock:
  [  114.010406][T10445] ff1100017d7545c0
(&HFSPLUS_I(inode)->extents_lock){+.+.}-{4:4},
                          at: hfsplus_get_block+0x282/0xab0
  [  114.021835][T10445] but task is already holding lock:
  [  114.024125][T10445] ff1100017d753300
(&HFSPLUS_I(inode)->extents_lock){+.+.}-{4:4},
                          at: hfsplus_file_truncate+0x20d/0xd70
  [  114.044862][T10445]   lock(&HFSPLUS_I(inode)->extents_lock);
  [  114.058492][T10445]   lock(&HFSPLUS_I(inode)->extents_lock);
  [  114.058944][T10445]  *** DEADLOCK ***
  [  114.058944][T10445]  May be due to missing lock nesting notation
  [  114.060170][T10445] locks held by syz-executor/10445: 6, on CPU#3:
  [  114.063858][T10445]  #4: ff1100017d753300
(&HFSPLUS_I(inode)->extents_lock){+.+.}-{4:4},
                          at: hfsplus_file_truncate+0x20d/0xd70
  [  114.064748][T10445]  #5: ff1100017d26e8f0 (&sbi->alloc_mutex){+.+.}-{4:4},
                          at: hfsplus_block_free+0xf3/0x690
  [  114.068038][T10445] Call Trace:
  [  114.068542][T10445]  dump_stack_lvl+0x116/0x1b0
  [  114.068921][T10445]  print_deadlock_bug+0x293/0x3d0
  [  114.069329][T10445]  __lock_acquire+0x153c/0x2110
  [  114.070151][T10445]  lock_acquire+0x1bf/0x370
  [  114.071749][T10445]  __mutex_lock+0x178/0x1dd0
  [  114.075318][T10445]  hfsplus_get_block+0x282/0xab0
  [  114.075713][T10445]  block_read_full_folio+0x305/0x850
  [  114.076997][T10445]  filemap_read_folio+0xc4/0x2a0
  [  114.077824][T10445]  do_read_cache_folio+0x255/0x5a0
  [  114.078746][T10445]  read_cache_page+0x62/0x150
  [  114.079124][T10445]  hfsplus_block_free+0x156/0x690
  [  114.079520][T10445]  hfsplus_free_extents+0x345/0x4a0
  [  114.079928][T10445]  hfsplus_file_truncate+0x59b/0xd70
  [  114.159730][T10445]  hfsplus_delete_inode+0x196/0x230
  [  114.160139][T10445]  hfsplus_unlink+0x7b5/0x980
  [  114.162948][T10445]  vfs_unlink+0x30d/0xc90
  [  114.169449][T10445]  filename_unlinkat+0x3ee/0x680
  [  114.171418][T10445]  __x64_sys_unlink+0x4c/0xa0
  [  114.171796][T10445]  do_syscall_64+0x132/0x7c0
  [  114.172170][T10445]  entry_SYSCALL_64_after_hwframe+0x77/0x7f

What seems to happen:

1. hfsplus_file_truncate() on the unlinked file holds that file inode's
   extents_lock (extents.c:579).
2. hfsplus_free_extents() -> hfsplus_block_free() reads the allocation
   bitmap through sbi->alloc_file->i_mapping while the file inode's
   extents_lock is still held (bitmap.c:183-185).
3. The bitmap read enters hfsplus_get_block() for the allocation-file inode
   and takes its own extents_lock (extents.c:264).

The two mutexes are different instances (the external-log addresses differ:
ff1100017d7545c0 vs ff1100017d753300) but share the same lockdep class,
because extents_lock is initialized in a common inode-initialization path.
lockdep therefore reports recursive locking.

Is this a lockdep false positive caused by the shared class, or is there a
real lock-ordering issue? If it is a false positive, should the
allocation-file inode use a separate lockdep class (or a nested annotation)
for extents_lock? Alternatively, should hfsplus_block_free() run outside the
file inode's extents_lock?

Reproduction:

  kernel: v7.4-rc3, CONFIG_KASAN=y, CONFIG_PROVE_LOCKING=y,
          CONFIG_HFSPLUS_FS=y, booted with oops=panic panic_on_warn=1
  guest:  Debian trixie x86_64, 6 GiB RAM, 4 vCPU, KVM
  syzkaller: 853b104bb454141a5b0fc0e4218ec78e9f34ffb1

  in the guest, as root:
  syz-execprog -procs=1 -threaded -repeat=1 \
      -executor ./syz-executor repro.kernsight.prog

The reproducer is a 56-entry state-reconstruction recipe. The main fuzzing
pipeline verified it 2/2 on freshly replayed VMs; the crash was reproduced
during crash-log replay at entry 55/55. An external fresh-VM run also
reproduced it 2/2 with -procs=1 -repeat=1 (114.00 s and 110.75 s). Note that
the recipe needs strict single-process ordering; 8 parallel workers did not
reproduce it.

How to reproduce reliably on mainline (v7.4-rc3)
------------------------------------------------
We reproduced this on mainline v7.4-rc3 with the following recipe. No
local HFS+ patches are applied; the source tree is vanilla v7.4-rc3.

1. Build the mainline v7.4-rc3 kernel with the configuration in
   rc3/kernel.config. The essential options are CONFIG_KASAN=y,
   CONFIG_PROVE_LOCKING=y, CONFIG_HFSPLUS_FS=y and CONFIG_DEBUG_INFO=y.
   The bzImage used here has sha256
   0c16409f8aec508ac44a63e631c2bd352536401f3fd0856df1d839fb953a1882.

2. Boot a Debian trixie x86_64 guest (6 GiB RAM, 4 vCPU, KVM) with the QEMU
   command line in rc3/environment.txt. Keep the serial console; the lockdep
   report appears there (and in dmesg). panic_on_warn=1 is not required to
   observe this lockdep report.

3. Copy syz-execprog and syz-executor built from syzkaller commit
   853b104bb454141a5b0fc0e4218ec78e9f34ffb1, plus
   rc3/reproducer/repro.kernsight.prog, into the guest.

4. In the guest, as root, run:

     syz-execprog -procs=1 -threaded -repeat=1 \
         -executor ./syz-executor repro.kernsight.prog

5. Expect "WARNING: possible recursive locking detected" after roughly
   110-115 seconds, with:

     hfsplus_get_block+0x282/0xab0
      hfsplus_file_truncate+0x20d/0xd70

   We observed this 2/2 on fresh VMs at 114.00 s and 110.75 s.

Important details for stable reproduction:

- Use -procs=1. With -procs=8 (or -repeat=0) the 56-program state recipe is
  interleaved and did not reproduce in a 20-minute run.
- Use the full repro.kernsight.prog recipe. The minimal repro.prog alone is
  state-dependent and does not reproduce by itself.
- The report is a lockdep recursive-locking warning, so serial/dmesg is the
  reliable success signal; it is not a KASAN panic.

A repository with the syzkaller programs, kernel configs and full logs is
available at:

  https://github.com/Zheng-zheng-codes/hfsplus-extents-lock-deadlock

The repository keeps the v7.4-rc2 and v7.4-rc3 results in separate `rc2/`
and `rc3/` directories so the two kernels are not mixed.

Repository layout:

- `README.md` at the repository root
  Overview of the report and an explicit rc2/rc3 split.

- `rc2/` -- original v7.4-rc2 results
  - `README.md`: v7.4-rc2 report and reproduction notes.
  - `environment.txt`: rc2 kernel image/hash, QEMU command line and
    syzkaller revision.
  - `kernel.config`: v7.4-rc2 kernel configuration.
  - `reproducer/repro.kernsight.prog`: rc2 pipeline-verified state recipe.
  - `reproducer/repro.priming.prog`: rc2 priming portion.
  - `reproducer/repro.prog`: rc2 triggering testcase.
  - `logs/log0..log3`, `logs/report0..report3`: original rc2 syzkaller
    execution logs and crash reports.
  - `logs/machineInfo0..machineInfo3`, `logs/title-stat.json`: guest
    environment and title statistics.
  - `logs/crash-excerpt.txt`: rc2 crash excerpt.

- `rc3/` -- new v7.4-rc3 results (the results in this report)
  - `README.md`: v7.4-rc3 report, locking analysis and reproduction notes.
  - `environment.txt`: rc3 kernel image/hash, guest image/hash, QEMU command
    line and syzkaller revision.
  - `kernel.config`: v7.4-rc3 kernel configuration.
  - `reproducer/repro.kernsight.prog`: rc3 56-entry state recipe.
  - `reproducer/repro.priming.prog`: rc3 priming portion.
  - `reproducer/repro.prog`: rc3 triggering testcase.
  - `reproducer/repro.report`, `repro.stats`, `repro.outcomes.jsonl`,
    `repro.context.json`: pipeline verification evidence, including the
    2/2 fresh-VM result.
  - `logs/log-0.txt`, `logs/report-0.txt`: original rc3 syzkaller log and
    crash report.
  - `logs/machineInfo-0.txt`, `logs/title-stat.json`: rc3 guest environment
    and title statistics.
  - `logs/external-verification/run1-singleproc-serial.log` and
    `run2-singleproc-serial.log`: full serial consoles of the two
    independent rc3 fresh-VM reproductions (114.00 s and 110.75 s).
  - `logs/external-verification/*-run.log` and `*-execprog-out.log`: the
    corresponding run records and syz-execprog output.
  - `SHA256SUMS`: checksums for the rc3 inputs and logs.

When reproducing, note the difference:
- rc2 reproduces from its own recipe in `rc2/reproducer/` (see
  `rc2/README.md` for the exact command used in that testing round).
- rc3 must be run with strict single-process ordering:
    syz-execprog -procs=1 -threaded -repeat=1 \
        -executor ./syz-executor repro.kernsight.prog

Best regards,
Zheng Zheng

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-17 20:50 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-17  3:01 [BUG] WARNING: possible recursive locking in hfsplus_get_block (fs/hfsplus/extents.c:264, v7.4-rc3) Zheng Zheng
2026-11-17 20:50 ` Viacheslav Dubeyko

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®