Redispatch ZIOs in deep call graphs

[ryao]

The below excerpt of a backtrace is from a ztest failure when running ZoL's
ztest:

/#453 0x00007f03c8060b35 in vdev_queue_io_to_issue (vq=vq@entry=0x99f8a8) at ../../module/zfs/vdev_queue.c:706
/#454 0x00007f03c806106e in vdev_queue_io (zio=zio@entry=0x7f0350003de0) at ../../module/zfs/vdev_queue.c:747
/#455 0x00007f03c80818c1 in zio_vdev_io_start (zio=0x7f0350003de0) at ../../module/zfs/zio.c:2659
/#456 0x00007f03c807f243 in __zio_execute (zio=0x7f0350003de0) at ../../module/zfs/zio.c:1399
/#457 zio_nowait (zio=0x7f0350003de0) at ../../module/zfs/zio.c:1456
/#458 0x00007f03c805f71b in vdev_mirror_io_start (zio=0x7f0350003a10) at ../../module/zfs/vdev_mirror.c:374
/#459 0x00007f03c807f243 in __zio_execute (zio=0x7f0350003a10) at ../../module/zfs/zio.c:1399
/#460 zio_nowait (zio=0x7f0350003a10) at ../../module/zfs/zio.c:1456
/#461 0x00007f03c806464c in vdev_raidz_io_start (zio=0x7f0350003380) at ../../module/zfs/vdev_raidz.c:1607
/#462 0x00007f03c807f243 in __zio_execute (zio=0x7f0350003380) at ../../module/zfs/zio.c:1399
/#463 zio_nowait (zio=0x7f0350003380) at ../../module/zfs/zio.c:1456
/#464 0x00007f03c805f71b in vdev_mirror_io_start (zio=0x7f0350002fb0) at ../../module/zfs/vdev_mirror.c:374
/#465 0x00007f03c807f243 in __zio_execute (zio=0x7f0350002fb0) at ../../module/zfs/zio.c:1399
/#466 zio_nowait (zio=0x7f0350002fb0) at ../../module/zfs/zio.c:1456
/#467 0x00007f03c805ed43 in vdev_mirror_io_done (zio=0x7f033957ebf0) at ../../module/zfs/vdev_mirror.c:499
/#468 0x00007f03c807a0c0 in zio_vdev_io_done (zio=0x7f033957ebf0) at ../../module/zfs/zio.c:2707
/#469 0x00007f03c808285b in __zio_execute (zio=0x7f033957ebf0) at ../../module/zfs/zio.c:1399
/#470 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f0390001330, pio=0x7f033957ebf0) at ../../module/zfs/zio.c:547
/#471 zio_done (zio=0x7f0390001330) at ../../module/zfs/zio.c:3278
/#472 0x00007f03c808285b in __zio_execute (zio=0x7f0390001330) at ../../module/zfs/zio.c:1399
/#473 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f03b4013a00, pio=0x7f0390001330) at ../../module/zfs/zio.c:547
/#474 zio_done (zio=0x7f03b4013a00) at ../../module/zfs/zio.c:3278
/#475 0x00007f03c808285b in __zio_execute (zio=0x7f03b4013a00) at ../../module/zfs/zio.c:1399
/#476 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f03b4014210, pio=0x7f03b4013a00) at ../../module/zfs/zio.c:547
/#477 zio_done (zio=0x7f03b4014210) at ../../module/zfs/zio.c:3278
/#478 0x00007f03c808285b in __zio_execute (zio=0x7f03b4014210) at ../../module/zfs/zio.c:1399
/#479 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f03b4014620, pio=0x7f03b4014210) at ../../module/zfs/zio.c:547
/#480 zio_done (zio=0x7f03b4014620) at ../../module/zfs/zio.c:3278
/#481 0x00007f03c807a6d3 in __zio_execute (zio=0x7f03b4014620) at ../../module/zfs/zio.c:1399
/#482 zio_execute (zio=zio@entry=0x7f03b4014620) at ../../module/zfs/zio.c:1337
/#483 0x00007f03c8060b35 in vdev_queue_io_to_issue (vq=vq@entry=0x99f8a8) at ../../module/zfs/vdev_queue.c:706
/#484 0x00007f03c806106e in vdev_queue_io (zio=zio@entry=0x7f0350002be0) at ../../module/zfs/vdev_queue.c:747
/#485 0x00007f03c80818c1 in zio_vdev_io_start (zio=0x7f0350002be0) at ../../module/zfs/zio.c:2659
/#486 0x00007f03c807f243 in __zio_execute (zio=0x7f0350002be0) at ../../module/zfs/zio.c:1399
/#487 zio_nowait (zio=0x7f0350002be0) at ../../module/zfs/zio.c:1456
/#488 0x00007f03c805f71b in vdev_mirror_io_start (zio=0x7f0350002810) at ../../module/zfs/vdev_mirror.c:374
/#489 0x00007f03c807f243 in __zio_execute (zio=0x7f0350002810) at ../../module/zfs/zio.c:1399
/#490 zio_nowait (zio=0x7f0350002810) at ../../module/zfs/zio.c:1456
/#491 0x00007f03c8064593 in vdev_raidz_io_start (zio=0x7f0350001270) at ../../module/zfs/vdev_raidz.c:1591
/#492 0x00007f03c807f243 in __zio_execute (zio=0x7f0350001270) at ../../module/zfs/zio.c:1399
/#493 zio_nowait (zio=0x7f0350001270) at ../../module/zfs/zio.c:1456
/#494 0x00007f03c805f71b in vdev_mirror_io_start (zio=0x7f0350001e60) at ../../module/zfs/vdev_mirror.c:374
/#495 0x00007f03c807f243 in __zio_execute (zio=0x7f0350001e60) at ../../module/zfs/zio.c:1399
/#496 zio_nowait (zio=0x7f0350001e60) at ../../module/zfs/zio.c:1456
/#497 0x00007f03c805ed43 in vdev_mirror_io_done (zio=0x7f033a0c39c0) at ../../module/zfs/vdev_mirror.c:499
/#498 0x00007f03c807a0c0 in zio_vdev_io_done (zio=0x7f033a0c39c0) at ../../module/zfs/zio.c:2707
/#499 0x00007f03c808285b in __zio_execute (zio=0x7f033a0c39c0) at ../../module/zfs/zio.c:1399
/#500 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f03a8003c00, pio=0x7f033a0c39c0) at ../../module/zfs/zio.c:547
/#501 zio_done (zio=0x7f03a8003c00) at ../../module/zfs/zio.c:3278
/#502 0x00007f03c808285b in __zio_execute (zio=0x7f03a8003c00) at ../../module/zfs/zio.c:1399
/#503 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f038800c400, pio=0x7f03a8003c00) at ../../module/zfs/zio.c:547
/#504 zio_done (zio=0x7f038800c400) at ../../module/zfs/zio.c:3278
/#505 0x00007f03c808285b in __zio_execute (zio=0x7f038800c400) at ../../module/zfs/zio.c:1399
/#506 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f038800da00, pio=0x7f038800c400) at ../../module/zfs/zio.c:547
/#507 zio_done (zio=0x7f038800da00) at ../../module/zfs/zio.c:3278
/#508 0x00007f03c808285b in __zio_execute (zio=0x7f038800da00) at ../../module/zfs/zio.c:1399
/#509 zio_notify_parent (wait=ZIO_WAIT_DONE, zio=0x7f038800fd80, pio=0x7f038800da00) at ../../module/zfs/zio.c:547
/#510 zio_done (zio=0x7f038800fd80) at ../../module/zfs/zio.c:3278
/#511 0x00007f03c807a6d3 in __zio_execute (zio=0x7f038800fd80) at ../../module/zfs/zio.c:1399
/#512 zio_execute (zio=zio@entry=0x7f038800fd80) at ../../module/zfs/zio.c:1337
/#513 0x00007f03c8060b35 in vdev_queue_io_to_issue (vq=vq@entry=0x99f8a8) at ../../module/zfs/vdev_queue.c:706
/#514 0x00007f03c806119d in vdev_queue_io_done (zio=zio@entry=0x7f03a0010950) at ../../module/zfs/vdev_queue.c:775
/#515 0x00007f03c807a0e8 in zio_vdev_io_done (zio=0x7f03a0010950) at ../../module/zfs/zio.c:2686
/#516 0x00007f03c807a6d3 in __zio_execute (zio=0x7f03a0010950) at ../../module/zfs/zio.c:1399
/#517 zio_execute (zio=0x7f03a0010950) at ../../module/zfs/zio.c:1337
/#518 0x00007f03c7fcd0c4 in taskq_thread (arg=0x966d50) at ../../lib/libzpool/taskq.c:215
/#519 0x00007f03c7fc7937 in zk_thread_helper (arg=0x967e90) at ../../lib/libzpool/kernel.c:135
/#520 0x00007f03c78890a3 in start_thread (arg=0x7f03c2703700) at pthread_create.c:309
/#521 0x00007f03c75c50fd in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:111

This occurred when ztest was simulating a scrub under heavy I/O load. Under
those circumstances, it was possible for a mix of noop I/Os for aggregation
continuity and the I/O elevator to generate arbitrarily deep recursion.

This patch modifies ZFS to propapage a recursion counter inside the zio_t
objects such that IOs will be redispatched upon reaching a given recursion
depth. We can detect long call chains and dispatch to another ZIO taskq. We
cut in-line when we do this to minimize the potential for taskq exhaustion that
can prevent a zio from notifying its parent.

Signed-off-by: Richard Yao [email protected]

[behlendorf]

This is a possible fix for #2293. While this does appear to be a legitimate but it's something which never occurs in practice. We've only ever seen this while ztest to my knowledge so I'm reluctant to add this additional complexity.

[behlendorf]

@ryao it sounds as if the preferred fix for this upstream was to just redispatch the zio. Let's plan on merging the upstream fix to address this issue unless for some reason it turns out to be problematic.