On Tue, Nov 02, 2021 at 10:42:58AM +0800, Zhongwei Cai wrote: > ext4_inode_datasync_dirty will call read_lock(&journal->j_state_lock) in > journal mode, which is unnecessary in read path (As far as I know, the > IOMAP_F_DIRTY flag set in the if branch is only used in write path, > making it unnecessary in read path. Please correct me if I'm wrong). IOMAP_F_DIRTY isn't conditional on the type of lookup being done. If the inode is dirty in a way that O_DSYNC would require it to be flushed to make the data stable, iomap should be told that it is dirty, even on read lookups... e.g. iomap_swapfile_activate() uses IOMAP_REPORT as the flags for extent mapping iteration passed to iomap_swapfile_iter(). THis then checks: /* No uncommitted metadata or shared blocks. */ if (iomap->flags & IOMAP_F_DIRTY) return iomap_swapfile_fail(isi, "is not committed"); IOWs, we expect the IOMAP_F_DIRTY flag to be set on all types of iomap mapping calls if the inode is dirty, not just IOMAP_WRITE calls. Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
