> + if (*seq != READ_ONCE(ifp->if_seq)) {
> + if (!xfs_iext_lookup_extent(ip, ifp, offset_fsb,
> + &icur, &timap) ||
> + timap.br_startoff > offset_fsb) {
> + ASSERT(0);
> + error = -EFSCORRUPTED;
> goto trans_cancel;
> }
> + xfs_trim_extent(imap, timap.br_startoff,
> + timap.br_blockcount);
> + count_fsb = imap->br_blockcount;
> + map_start_fsb = imap->br_startoff;
I very much disagree with this logic.
xfs_bmapi_write will always do a xfs_iext_lookup_extent for the
passed in blocks anyway. So this trimming logic should move into
xfs_bmapi_write to ensure it does the right thing, instead of papering
over the logic in xfs_bmapi_write in the caller with another lookup.
I think the improved XFS_BMAPI_DELALLOC handling in this patch:
http://git.infradead.org/users/hch/xfs.git/commitdiff/553fed9ce7695df081779c6a9a205af4142b2a06
is the right answer, as writeback really should only ever convert
existing delalloc extents, and I'd much rather rely on that rather than
piling hacks of hacks to feed the right range into a function that must
do a lookup in the extent tree anyway.
