Re: Crash in btrfs_uuid_tree_iterate during mount
From: Chris Mason <clm@fb.com>
Date: 2016-08-08 14:16:33
On 08/08/2016 06:49 AM, Nikolay Borisov wrote:
On 08/05/2016 06:12 PM, Chris Mason wrote:
Hello Chris,
Indeed it seems that btrfs_uuid_iter_rem returned a ENOENT:
callq 0xffffffffa081f450 <btrfs_uuid_tree_rem>
mov %eax,%r13d
je 0xffffffffa081f882 <btrfs_uuid_tree_iterate+514> ; if uuid_iter_rem returned -ENOENT; else fall through.
I checked and r13d is not being touched between the invocation of
btrfs_uuid_iter_rem and the btrfs_next_item:
RIP: ffffffffa081f774 RSP: ffff88034e507e20 RFLAGS: 00010246
RAX: 0000000000000000 RBX: 0000160000000000 RCX: ffff880000000000
RDX: 0000000000000001 RSI: ffff8801e3abd140 RDI: ffff88046f027f00
RBP: ffff88034e507ea8 R8: 000060fb80001760 R9: ffffffffa07ac1de
R10: ffffe8ffffd41760 R11: ffffea00078eaf40 R12: ffff8801b98ab750
R13: 00000000fffffffe R14: ffff8801e3abd140 R15: ffff880049586000
ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
r13 is clearly -ENOENT. So your assumption was correct.Fantastic, thanks again for digging through it. Making the patch is much easier than testing the patch in this case. If you can trigger this semi-reliably, we can add some debugging to make sure we're not papering over some other problem. -chris