Setup is a 24 core vm + 24G of ram, running debian 12 with top of
master as of writing this e-mail, commit being:
commit 1c41041124bd14dd6610da256a3da4e5b74ce6b1 (HEAD -> master,
origin/master, origin/HEAD)
Merge: b8cc56d0414e 9fd00df05e81
Author: Linus Torvalds <torva...@linux-foundation.org>
Date: Sat Nov 4 16:25:36 2023 -1000
Merge tag 'i3c/for-6.7' of
git://git.kernel.org/pub/scm/linux/kernel/git/i3c/linux
I ran into it when trying to poke around at the fine-grained inode hash patch.
Reproduction instructions are here: https://people.freebsd.org/~mjg/fstree.tgz
Running it destabilizes the system after few minutes, for example
hanging my ssh connection (interestingly running stuff over a serial
port works fine). hung task detector reports a kernel thread and one
of the workers, traces at the end of the e-mail.
I can create trees just fine if I limit it to one worker at a time.
However, traversing them (again 20 workers, each with a dedicated
tree) once more runs into trouble.
There are no issues when running this against xfs and ext4.
Side note: mkfs.bcachefs warns about xfs being on the volume, but is
perfectly happy to format an existing bcachefs partition -- perhaps
this should also ask if it was intended?
traces:
INFO: task kworker/u48:0:11 blocked for more than 120 seconds.
Not tainted 6.6.0+ #386
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:kworker/u48:0 state:D stack:0 pid:11 tgid:11 ppid:2
flags:0x00004000
Workqueue: btree_update btree_interior_update_work
Call Trace:
<TASK>
__schedule+0x3b0/0xaf0
? bch2_path_get+0x5e/0x560
? __pfx_bch2_six_check_for_deadlock+0x10/0x10
schedule+0x2e/0xd0
six_lock_slowpath.constprop.0+0x10b/0x2e0
btree_interior_update_work+0x8a3/0x9e0
? btree_interior_update_work+0x842/0x9e0
process_one_work+0x165/0x330
worker_thread+0x2f1/0x410
? __pfx_worker_thread+0x10/0x10
kthread+0xe1/0x110
? __pfx_kthread+0x10/0x10
ret_from_fork+0x2d/0x50
? __pfx_kthread+0x10/0x10
ret_from_fork_asm+0x1b/0x30
</TASK>
INFO: task createtree:3024 blocked for more than 120 seconds.
Not tainted 6.6.0+ #386
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:createtree state:D stack:0 pid:3024 tgid:3024
ppid:3011 flags:0x00000002
Call Trace:
<TASK>
__schedule+0x3b0/0xaf0
? __pfx_bch2_six_check_for_deadlock+0x10/0x10
schedule+0x2e/0xd0
six_lock_slowpath.constprop.0+0x10b/0x2e0
bch2_btree_node_get+0x33d/0x4c0
? bch2_dirent_create+0x238/0x3e0
bch2_btree_path_traverse_one+0x212/0x8c0
? bch2_btree_path_traverse_one+0xb2/0x8c0
? bch2_dirent_create+0x238/0x3e0
bch2_btree_iter_peek_slot+0x106/0x6d0
? btree_path_get_locks.constprop.0+0x3a/0x150
? bch2_path_get+0x404/0x560
? bch2_dirent_hash+0xd6/0x150
? bch2_dirent_create+0xf6/0x3e0
bch2_dirent_create+0x238/0x3e0
? bch2_create_trans+0x4d0/0x6c0
bch2_create_trans+0x518/0x6c0
? chacha_block_generic+0x6f/0xb0
__bch2_create+0x1be/0x4e0
? bch2_trans_iter_init_outlined+0x112/0x180
? d_splice_alias+0x8e/0x2b0
? bch2_create+0x26/0x60
bch2_create+0x26/0x60
path_openat+0xe9f/0x11d0
do_filp_open+0xb4/0x160
? kmem_cache_alloc+0x15c/0x2b0
? _raw_spin_unlock+0xa/0x30
do_sys_openat2+0x91/0xc0
__x64_sys_openat+0x6a/0xa0
do_syscall_64+0x32/0xf0
entry_SYSCALL_64_after_hwframe+0x6e/0x76
RIP: 0033:0x7fe7b1bdfe01
RSP: 002b:00007ffee4ae0ea0 EFLAGS: 00000202 ORIG_RAX: 0000000000000101
RAX: ffffffffffffffda RBX: 0000000000000042 RCX: 00007fe7b1bdfe01
RDX: 0000000000000042 RSI: 00007ffee4ae0f30 RDI: 00000000ffffff9c
RBP: 00007ffee4ae0f30 R08: 0000000000000000 R09: 0000000000000064
R10: 00000000000001ff R11: 0000000000000202 R12: 0000000000000263
R13: 00000000000003e8 R14: 00007ffee4ae267c R15: 000055b2c0e97010
</TASK>
--
Mateusz Guzik <mjguzik gmail.com>
