On Mon, Aug 31, 2009 at 12:47:49PM +0200, Jens Axboe wrote: > It's because ext4 writepages sets ->range_start and wb_writeback() is > range cyclic, then the next iteration will have the previous end point > as the starting point. Looks like we need to clear ->range_start in > wb_writeback(), the better place is probably to do that in > fs/fs-writeback.c:generic_sync_wb_inodes() right after the > writeback_single_inode() call. This, btw, should be no different than > the current code, weird/correct or not :-) Hmm, or we could have ext4_da_writepages save and restore ->range_start. One of the things that's never been well documented is exactly what the semantics are of the various fields in the wbc struct, and who is allowed to modify which fields when. If you have some time, it would be great if you could document the rules filesystems should be following with respect to the wbc struct, and then we can audit each filesystem to make sure they follow those rules. One of the things which is a bit scary about how the many wbc flags work is that each time a filesystem wants some particular behavior, it seems like we need to dive into writeback code, and figure out some combination of flags/settings that make the page writeback code do what we wants, and sometimes it's not clear whether that was a designed-in semantic of the interface, or just something that happened to work given the current implementation. In any case, if one of the rules is that the filesystems' writepages command shouldn't be modifying range_start, we can fix this problem up by saving and restore range_start inside ext4_da_writepages(). - Ted -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
