[6.15-rc0 LBS(?) regression] XFS mount hangs in set_blocksize()

[6.15-rc0 LBS(?) regression] XFS mount hangs in set_blocksize()

[Dave Chinner]

Hi folks,

Just saw this on a test run after upgrading to a current TOT kernel
+ linux-xfs/for-next. A mount being run from check-parallel hungs
with this stack trace:

 task:mount           state:D stack:13064 pid:4005317 tgid:4005317 ppid:3498345 task_flags:0x400100 flags:0x00004002
 Call Trace:
  <TASK>
  __schedule+0x63b/0xad0
  schedule+0x6d/0xf0
  io_schedule+0x48/0x70
  folio_wait_bit_common+0x180/0x290
  __folio_lock+0x17/0x20
  truncate_inode_pages_range+0x364/0x3e0
  truncate_inode_pages+0x15/0x20
  set_blocksize+0x193/0x1b0
  xfs_setsize_buftarg+0x24/0x60
  xfs_setup_devices+0x1d/0xd0
  xfs_fs_fill_super+0x350/0x870
  get_tree_bdev_flags+0x121/0x1a0
  get_tree_bdev+0x10/0x20
  xfs_fs_get_tree+0x15/0x20
  vfs_get_tree+0x28/0xe0
  vfs_cmd_create+0x5f/0xd0
  vfs_fsconfig_locked+0x50/0x130
  __se_sys_fsconfig+0x34b/0x3d0
  __x64_sys_fsconfig+0x25/0x30
  x64_sys_call+0x3c3/0x2f70
  do_syscall_64+0x68/0x130
  entry_SYSCALL_64_after_hwframe+0x76/0x7e

This has come from xfs/032 which is testing all supported sector and
block sizes on XFS, so it is likely exercising LBS functionality at
both the block device and filesystem layers. I haven't seen this in
6.14 at all, so it happening within a couple of test runs on a
current TOT implies a regression introduced in the current merge
window.

This doesn't fail every time it is run, but I'm reporting this on
the first failure so I don't yet know how frequently it happens yet.

-Dave.
-- 
Dave Chinner
david@fromorbit.com