6.6-stable review patch. If anyone has any objections, please let me know.
------------------
From: Filipe Manana fdmanana@suse.com
[ Upstream commit 5630f7557de61264ccb4f031d4734a1a97eaed16 ]
When we are logging a directory and the log context indicates that we are logging a new name for some other file (that is or was inside that directory), we skip logging the inodes for new dentries in the directory.
This is ok most of the time, but if after the rename or link operation that triggered the logging of that directory, we have an explicit fsync of that directory without the directory inode being evicted and reloaded, we end up never logging the inodes for the new dentries that we found during the new name logging, as the next directory fsync will only process dentries that were added after the last time we logged the directory (we are doing an incremental directory logging).
So make sure we always log new dentries for a directory even if we are in a context of logging a new name.
We started skipping logging inodes for new dentries as of commit c48792c6ee7a ("btrfs: do not log new dentries when logging that a new name exists") and it was fine back then, because when logging a directory we always iterated over all the directory entries (for leaves changed in the current transaction) so a subsequent fsync would always log anything that was previously skipped while logging a directory when logging a new name (with btrfs_log_new_name()). But later support for incrementally logging a directory was added in commit dc2872247ec0 ("btrfs: keep track of the last logged keys when logging a directory"), to avoid checking all dir items every time we log a directory, so the check to skip dentry logging added in the first commit should have been removed when the incremental support for logging a directory was added.
A test case for fstests will follow soon.
Reported-by: Vyacheslav Kovalevsky slava.kovalevskiy.2014@gmail.com Link: https://lore.kernel.org/linux-btrfs/84c4e713-85d6-42b9-8dcf-0722ed26cb05@gma... Fixes: dc2872247ec0 ("btrfs: keep track of the last logged keys when logging a directory") Reviewed-by: Boris Burkov boris@bur.io Signed-off-by: Filipe Manana fdmanana@suse.com Signed-off-by: David Sterba dsterba@suse.com Signed-off-by: Sasha Levin sashal@kernel.org --- fs/btrfs/tree-log.c | 8 -------- 1 file changed, 8 deletions(-)
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c index 256116a6e07c6..ca01f2daac6e4 100644 --- a/fs/btrfs/tree-log.c +++ b/fs/btrfs/tree-log.c @@ -5526,14 +5526,6 @@ static int log_new_dir_dentries(struct btrfs_trans_handle *trans, struct btrfs_inode *curr_inode = start_inode; int ret = 0;
- /* - * If we are logging a new name, as part of a link or rename operation, - * don't bother logging new dentries, as we just want to log the names - * of an inode and that any new parents exist. - */ - if (ctx->logging_new_name) - return 0; - path = btrfs_alloc_path(); if (!path) return -ENOMEM;