As part of testing log recovery with dm_log_writes, Amir Goldstein
discovered an error in the deferred ops recovery that lead to corruption
of the filesystem metadata if a reflink+rmap filesystem happened to shut
down midway through a CoW remap:
"This is what happens [after failed log recovery]:
"Phase 1 - find and verify superblock...
"Phase 2 - using internal log
" - zero log...
" - scan filesystem freespace and inode maps...
" - found root inode chunk
"Phase 3 - for each AG...
" - scan (but don't clear) agi unlinked lists...
" - process known inodes and perform inode discovery...
" - agno = 0
"data fork in regular inode 134 claims CoW block 376
"correcting nextents for inode 134
"bad data fork in inode 134
"would have cleared inode 134"
Hou Tao dissected the log contents of exactly such a crash:
"According to the implementation of xfs_defer_finish(), these ops should
be completed in the following sequence:
"Have been done:
"(1) CUI: Oper (160)
"(2) BUI: Oper (161)
"(3) CUD: Oper (194), for CUI Oper (160)
"(4) RUI A: Oper (197), free rmap [0x155, 2, -9]
"Should be done:
"(5) BUD: for BUI Oper (161)
"(6) RUI B: add rmap [0x155, 2, 137]
"(7) RUD: for RUI A
"(8) RUD: for RUI B
"Actually be done by xlog_recover_process_intents()
"(5) BUD: for BUI Oper (161)
"(6) RUI B: add rmap [0x155, 2, 137]
"(7) RUD: for RUI B
"(8) RUD: for RUI A
"So the rmap entry [0x155, 2, -9] for COW should be freed firstly,
then a new rmap entry [0x155, 2, 137] will be added. However, as we can see
from the log record in post_mount.log (generated after umount) and the trace
print, the new rmap entry [0x155, 2, 137] are added firstly, then the rmap
entry [0x155, 2, -9] are freed."
When reconstructing the internal log state from the log items found on
disk, it's required that deferred ops replay in exactly the same order
that they would have had the filesystem not gone down. However,
replaying unfinished deferred ops can create /more/ deferred ops. These
new deferred ops are finished in the wrong order. This causes fs
corruption and replay crashes, so let's create a single defer_ops to
handle the subsequent ops created during replay, then use one single
transaction at the end of log recovery to ensure that everything is
replayed in the same order as they're supposed to be.
Reported-by: Amir Goldstein <amir73il@gmail.com>
Analyzed-by: Hou Tao <houtao1@huawei.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Tested-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Provide a mechanism for higher levels to create CUI/CUD items, submit
them to the log, and a stub function to deal with recovered CUI items.
These parts will be connected to the refcountbt in a later patch.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Create refcount update intent/done log items to record redo
information in the log. Because we need to roll transactions between
updating the bmbt mapping and updating the reverse mapping, we also
have to track the status of the metadata updates that will be recorded
in the post-roll transactions, just in case we crash before committing
the final transaction. This mechanism enables log recovery to finish
what was already started.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Darrick J. Wong