Re: [PATCH 1/1] block: print warning when invalid domain set to ioprio

From: Niklas Cassel
Date: Wed Jan 31 2024 - 07:53:09 EST


On Wed, Jan 31, 2024 at 08:14:01PM +0800, zhaoyang.huang wrote:
> From: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
>
> Since few caller check IOPRIO_PRIO_VALUE's return value and bio_set_prio
> is a open macro to set bio->ioprio directly.It is confused for the developer
> who run across kernel panic[1] but can find nothing in previous kernel log.
> Add a pr_err here to dump the information.
>
> [1]
> Here is the kernel panic I run across which caused by a out of bounds check
> introduced by CONFIG_FOTIFY_SOURCE.
>
> [exception_serialno]:
> [exception_kernel_version]:
> [exception_reboot_reason]: kernel_crash
> [exception_panic_reason]: UBSAN: array index out of bounds: Fatal exception
> [exception_time]: 1970-01-01_08-00-23
> [exception_file_info]: not-bugon
> [exception_task_id]: 409
> [exception_task_family]: [f2fs_ckpt-254:4, 409][kthreadd, 2]
> [exception_pc_symbol]: [<ffffffc080736974>] dd_request_merge+0x100/0x110
> [exception_stack_info]: [<ffffffc07a27e274>] get_exception_stack_info+0x124/0x2d8 [sysdump]gc
> [<ffffffc07a27e670>] prepare_exception_info+0x158/0x1d4 [sysdump]gc
> [<ffffffc07a280128>] sysdump_panic_event+0x5d8/0x748 [sysdump]gc
> [<ffffffc08016a508>] notifier_call_chain+0x98/0x17cgc
> [<ffffffc08016a9b4>] atomic_notifier_call_chain+0x44/0x68gc
> [<ffffffc0810f0eb4>] panic+0x194/0x37cgc
> [<ffffffc0800a638c>] die+0x300/0x310gc
> [<ffffffc0800a77e8>] ubsan_handler+0x34/0x4cgc
> [<ffffffc0800960a8>] brk_handler+0x9c/0x11cgc
> [<ffffffc0800bf998>] do_debug_exception+0xb0/0x140gc
> [<ffffffc0810f8bf0>] el1_dbg+0x58/0x74gc
> [<ffffffc0810f89f4>] el1h_64_sync_handler+0x3c/0x90gc
> [<ffffffc080091298>] el1h_64_sync+0x68/0x6cgc
> [<ffffffc080736974>] dd_request_merge+0x100/0x110gc //out of bound
> here caused by the value of class transferred from ioprio
> [<ffffffc080707f28>] elv_merge+0x248/0x270gc
> [<ffffffc0807146e8>] blk_mq_sched_try_merge+0x4c/0x20cgc
> [<ffffffc080736824>] dd_bio_merge+0x64/0xb4gc
> [<ffffffc080723e3c>] blk_mq_sched_bio_merge+0x68/0x144gc
> [<ffffffc08071b944>] blk_mq_submit_bio+0x2e8/0x6c0gc
> [<ffffffc08070dd3c>] __submit_bio+0xbc/0x1b0gc
> [<ffffffc08070c440>] submit_bio_noacct_nocheck+0xe4/0x2f0gc
> [<ffffffc08070c8e4>] submit_bio_noacct+0x298/0x3d8gc
> [<ffffffc08070caf8>] submit_bio+0xd4/0xf0gc
> [<ffffffc080642644>] f2fs_submit_write_bio+0xcc/0x49cgc
> [<ffffffc0806442d4>] __submit_merged_bio+0x48/0x13cgc
> [<ffffffc080641de4>] __submit_merged_write_cond+0x18c/0x1f8gc
> [<ffffffc080641c4c>] f2fs_submit_merged_write+0x2c/0x38
> [<ffffffc080655724>] f2fs_sync_node_pages+0x6e0/0x740gc
> [<ffffffc08063946c>] f2fs_write_checkpoint+0x4c0/0x97cgc
> [<ffffffc08063b37c>] __checkpoint_and_complete_reqs+0x88/0x248gc
> [<ffffffc08063ad70>] issue_checkpoint_thread+0x94/0xf4gc
> [<ffffffc080167c20>] kthread+0x110/0x1b8gc
>
> Signed-off-by: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
> ---
> include/uapi/linux/ioprio.h | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/include/uapi/linux/ioprio.h b/include/uapi/linux/ioprio.h
> index bee2bdb0eedb..73c420a0df72 100644
> --- a/include/uapi/linux/ioprio.h
> +++ b/include/uapi/linux/ioprio.h
> @@ -112,8 +112,11 @@ static __always_inline __u16 ioprio_value(int prioclass, int priolevel,
> {
> if (IOPRIO_BAD_VALUE(prioclass, IOPRIO_NR_CLASSES) ||
> IOPRIO_BAD_VALUE(priolevel, IOPRIO_NR_LEVELS) ||
> - IOPRIO_BAD_VALUE(priohint, IOPRIO_NR_HINTS))
> + IOPRIO_BAD_VALUE(priohint, IOPRIO_NR_HINTS)) {
> + pr_err("%s: get a invalid domain in class %d, level %d, hint %d\n",
> + __func__, prioclass, priolevel, priohint);
> return IOPRIO_CLASS_INVALID << IOPRIO_CLASS_SHIFT;
> + }
>
> return (prioclass << IOPRIO_CLASS_SHIFT) |
> (priohint << IOPRIO_HINT_SHIFT) | priolevel;
> --
> 2.25.1
>

Adding linux-block to CC.

pr_err() is a kernel function for printing.
ioprio_value() is a function in a uapi header, so this function will be
used by user space programs.

There is a reason:
$ git grep pr_err include/uapi/

Gives no results.



I think you should fix mq-deadline instead.
It looks like the problem comes from:
ioprio_value() will set class to IOPRIO_CLASS_INVALID (value 7),
if the user specified an class/level/hint that was invalid.

ioprio_class_to_prio[] array in mq-deadline.c does currently not have an
entry in to translate IOPRIO_CLASS_INVALID (7) to a valid DD_*_PRIO value.

Although, why does this I/O even reach the scheduler, shouldn't this I/O
get rejected even earlier?

Both io_uring and libaio will call ioprio_check_cap(), which should fail
the I/O before it reaches the I/O scheduler, but in your case, you are
submitting the I/O from the filesystem.

Should we perhaps add a call to ioprio_check_cap() or similar in some
path used to submit I/O by filesystems?


Kind regards,
Niklas