After investigating crashes on generic/176 it turned that the culprit in fact is the random failure induced by generic/019. As it happens, if on unmount the filesystem is in BTRFS_FS_STATE_ERROR then btrfs_error_commit_super is called. This unveiled 2 bugs: 1. btrfs_destroy_delalloc_inodes's implementation was completely bogus, since it only called btrfs_invalidate_inodes which only pruned dentries and didn't do anything to free any inodes with pending delalloc bytes. Once this is fixed with the use of invalide_inode_pages2 the second bug transpired. 2. The last call ot run_delayed_iputs is made before btrfs_cleanup_transaction is called. The latter in turn could queue up more delayed iputs resulting from invalidates_inode_pages2.
This series fixes the problem by first fixing btrfs_destroy_delalloc_inode to properly cleanup delalloc inodes and as a result cleans up the code a bit. I've given it a good bashing through xfstest (4 full xfstest cycles + 100 iterations of generic/475 since it was hitting some early assertion failures, which are fixed in the final version) so am pretty confident in the change. Nikolay Borisov (5): btrfs: Unexport btrfs_alloc_delalloc_work btrfs: Split btrfs_del_delalloc_inode into 2 functions btrfs: Add assert in __btrfs_del_delalloc_inode btrfs: Fix delalloc inodes invalidation during transaction abort btrfs: Unexport and rename btrfs_invalidate_inodes fs/btrfs/ctree.h | 12 +---- fs/btrfs/disk-io.c | 22 ++++---- fs/btrfs/inode.c | 152 ++++++++++++++++++++++++++++++----------------------- 3 files changed, 99 insertions(+), 87 deletions(-) -- 2.7.4 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
