On Sat, Mar 15, 2014 at 11:46:26AM -0400, Theodore Ts'o wrote: > On Fri, Mar 14, 2014 at 01:13:19PM -0700, Darrick J. Wong wrote: > > Both cases above cause the (temporary) use of two extents to map the same lblk. > > _set_bmap() first adjusts next_extent.e_lblk to cover the lblk, then decrements > > extent.e_len so that "extent" no longer covers the lblk. _punch_extent() > > splits an extent by first inserting the second part of the extent and then > > shortening the original extent to reflect the punchout. > > > > These two cases seemed slightly suspect to me, but since e2fsprogs doesn't > > journal, changing the code to reduce one extent and enlarge the other opens the > > possibility that we could lose the lblk mapping entirely if something happens > > in between the two operations. I prefer "block still mapped but fsck is > > unhappy about redundant bookeepping" to "the block is gone", so I added the two > > fixes and let it go. > > OK, fair enough, I'll merge in this patch since it's fixing a real > bug, and I can't think of a good way to fix this issue without making > pretty massive changes. > > I'll note though that at the moment e2fsck doesn't have sophisticated > extent tree recovery support; so if the extent tree is corrupt, it > offers to zap the entire extent tree, instead of trying to fix up the > extent tree. That wasn't an issue because it was likely that if > absent bugs, the most likely case if the extent tree is corrupted, > there's not much you can do, so it wasn't worth it to add some code to > handle these cases. > > However, if a large number of users start using your FUSE server in > production, then making sure the right thing happens when they suffer > power failures start becoming more important --- but it may not be > trivial since libext2fs wasn't originally intended for that use case. > I am glad that you implemented it, since it's a great way to get > better testing for various corner cases. But that's different from > advertising that the FUSE server should be used in production use > cases; in particular, we might need to figure out some kind of > journalling system for FUSE, either using the ext4's internal journal, > or some user-space journalling system. I wasn't planning to advertise fuse2fs for production use. But perhaps it could cough out a few more warnings if you're not mounting ro, and/or default to ro? --D > > - Ted > -- > To unsubscribe from this list: send the line "unsubscribe linux-ext4" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
