zbd/012: Test requeuing of zoned writes and queue freezing

[bvanassche]

Test concurrent requeuing of zoned writes and request queue freezing. While this test passes with kernel 6.9, it triggers a hang with kernels 6.10..6.12. This shows that this hang is a regression introduced by the zone write plugging code.

sysrq: Show Blocked State
task:(udev-worker) state:D stack:0 pid:75392 tgid:75392 ppid:2178 flags:0x00000006
Call Trace:

__schedule+0x3e8/0x1410
schedule+0x27/0xf0
blk_mq_freeze_queue_wait+0x6f/0xa0
queue_attr_store+0x60/0xc0
kernfs_fop_write_iter+0x13e/0x1f0
vfs_write+0x25b/0x420
ksys_write+0x65/0xe0
do_syscall_64+0x82/0x160
entry_SYSCALL_64_after_hwframe+0x76/0x7e

[bvanassche]

@kawasaki @damien-lemoal Please take a look.

[damien-lemoal]

Bart,

The zone write plugging fixes that I posted and that Jens applied should fix this. Please try again with these patches to check.

[bvanassche]

Thanks for having taken a look Damien. I have already verified that your patches fix the reported issue. Since the kernel versions referred to above have been released and cannot be modified anymore, I think the references to kernel versions above are correct.

[damien-lemoal]

Yes. The fixes are marked for stable so they will be backported to stable 6.12. 6.10 and 6.11 are already EOL so they will not get the backports.

[bvanassche]

With "6.10..6.12" I want to refer to kernel versions 6.10.0, 6.11.0 and 6.12.0. In other words, I want to exclude later versions that include backports. Please let me know if you want me to include this clarification in the patch description.

[kawasaki]

@bvanassche Thanks for this PR. The code looks good to me. I will wait for the kernel side fix gets upstreamed (v6.13-rc3 or rc4, hopefully?) then merge this PR.

[bvanassche]

I think that Linus just pulled Damien's kernel patches :-) See also https://lore.kernel.org/linux-block/d2acd3cb-188c-4ad5-91db-efbc6e50a1c1@kernel.dk/.

[kawasaki]

Yes, the fix is now upstreamed and tagged with v6.13-rc3. I have merged this PR. Thanks!

[yizhanglinux]

@kawasaki @bvanassche To run this case, seems we also need _io_uring_enable/_io_uring_restore for the fio tests.

[bvanassche]

I will switch to libaio since io_uring does not preserve the write submission order.