mkfs on zvol: sleeping function called from invalid context

[dechamps]

Using latest master (0443477), Linux 3.2.28. Steps to reproduce:

# dd if=/dev/zero of=/tmp/file bs=1 count=1 seek=1024M
# zpool create foo /tmp/file
# zfs create -V 512M foo/vol
# mkfs.ext4 /dev/zvol/foo/vol 

This results in the following kernel message:

BUG: sleeping function called from invalid context at mm/slub.c:935
in_atomic(): 1, irqs_disabled(): 1, pid: 7556, name: mkfs.ext4
Pid: 7556, comm: mkfs.ext4 Tainted: P           O 3.2.28-zfsdev-std-ipv6-64 #6
Call Trace:
 [<ffffffff81093c21>] __might_sleep+0xe1/0x110
 [<ffffffff8117e8a7>] __kmalloc+0xd7/0x1a0
 [<ffffffffa00fe7fa>] kmem_alloc_debug+0x25a/0x530 [spl]
 [<ffffffffa010134b>] task_alloc+0x1eb/0x4a0 [spl]
 [<ffffffffa0102b9b>] __taskq_dispatch+0x1eb/0x460 [spl]
 [<ffffffffa02dc3cb>] zvol_request+0x10b/0x1a0 [zfs]
 [<ffffffff8162ccbe>] __blk_run_queue+0x1e/0x20
 [<ffffffff816330fe>] queue_unplugged+0x4e/0xe0
 [<ffffffff8163336c>] blk_flush_plug_list+0x1dc/0x260
 [<ffffffff81633645>] blk_queue_bio+0x205/0x370
 [<ffffffff816314ba>] generic_make_request+0xca/0x100
 [<ffffffff81631568>] submit_bio+0x78/0xf0
 [<ffffffff811bbaab>] submit_bh+0xeb/0x130
 [<ffffffff811bf1b0>] __block_write_full_page+0x220/0x3d0
 [<ffffffff811bf431>] block_write_full_page_endio+0xd1/0x110
 [<ffffffff811bf485>] block_write_full_page+0x15/0x20
 [<ffffffff811c3188>] blkdev_writepage+0x18/0x20
 [<ffffffff81141bca>] __writepage+0x1a/0x50
 [<ffffffff81142766>] write_cache_pages+0x1f6/0x4f0
 [<ffffffff81142ab4>] generic_writepages+0x54/0x80
 [<ffffffff81142b04>] do_writepages+0x24/0x40
 [<ffffffff81139ddb>] __filemap_fdatawrite_range+0x5b/0x60
 [<ffffffff81139e3a>] filemap_write_and_wait_range+0x5a/0x80
 [<ffffffff811c33f5>] blkdev_fsync+0x25/0x60
 [<ffffffff811b9afb>] vfs_fsync_range+0x2b/0x40
 [<ffffffff811b9b2c>] vfs_fsync+0x1c/0x20
 [<ffffffff811b9d6a>] do_fsync+0x3a/0x60
 [<ffffffff811b9dc0>] sys_fsync+0x10/0x20
 [<ffffffff81c1ecab>] system_call_fastpath+0x16/0x1b

[behlendorf]

This indicate that we called kmem_alloc() under a spin lock with KM_SLEEP(). Looking at the code I just don't see how this is possible unless you were running an older SPL. Commit openzfs/spl@9b51f21 should have resolved any lingering issues like this.

[dechamps]

That's possible. Unfortunately I don't have access to the system on which it occurred anymore so I'm gonna assume you're right and close this.