INFO: task hung in do_get_write_access on v7.3-rc2

INFO: task hung in do_get_write_access on v7.3-rc2

[CJ]

Hi,


syzbot reported "INFO: task hung in do_get_write_access" on Linux 6.1. The same failure is still
reproducible on v7.3-rc2 with the adapted reproducer described below.


The reproducer writes the PCI command register of the block controller that backs the root ext4 filesystem, disabling normal I/O, and then keeps buffered writes and delalloc activity running. The adapted C program discovers the backing PCI function dynamically instead of relying on a fixed guest BDF. It also lowers hung_task_timeout_secs to 20 so the stall is captured inside the replay window; no fault injection is used.


The target console reports tasks blocked in I/O wait in do_get_write_access, reached through jbd2_journal_get_write_access and the ext4 inode-dirty path.


An in-flight journal transaction waits for metadata I/O that no longer completes. The report establishes the blocked path, not the final device-state root cause.


The candidate changes /proc/sys/kernel/hung_task_timeout_secs to 20. This is a detection-window setting, not fault injection, but it is an environment change. Attempts 1 and 2 were machine failure/timeout; attempt 3 produced the evidence.


This is a follow-up for the existing syzbot report https://syzkaller.appspot.com/bug?extid=9e3d85789d3996cc829c, not a
bisected root cause. The target report shares
18 key frames with the source crash report.


Reproducer:
  C reproducer: https://pastebin.com/raw/ABvzeXMS


console output: https://pastebin.com/raw/HUs8ymdz
kernel config: https://syzkaller.appspot.com/text?tag=KernelConfig&x=31dfefc4a14efea3


Kernel:
  HEAD commit: df2908090cda368b01ff43709f51890076c56157
  git tree: linux-stable
  kernel version: 7.3.0-rc2
  tested tag: v7.3-rc2


Let me know if you need more details or testing.


Thanks.