summaryrefslogtreecommitdiff
path: root/fs/xfs/xfs_bmap_util.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2020-08-07 10:57:29 -0700
committerLinus Torvalds <torvalds@linux-foundation.org>2020-08-07 10:57:29 -0700
commit5631c5e0eb9035d92ceb20fcd9cdb7779a3f5cc7 (patch)
tree4c293828f0220bb009d0fe18e66e209bef35e4ef /fs/xfs/xfs_bmap_util.c
parente51418191f5a741b5f94764798c81bf69dec4806 (diff)
parent818d5a91559ffe1e1f2095dcbbdb96c13fdb94ec (diff)
Merge tag 'xfs-5.9-merge-7' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux
Pull xfs updates from Darrick Wong: "There are quite a few changes in this release, the most notable of which is that we've made inode flushing fully asynchronous, and we no longer block memory reclaim on this. Furthermore, we have fixed a long-standing bug in the quota code where soft limit warnings and inode limits were never tracked properly. Moving further down the line, the reflink control loops have been redesigned to behave more efficiently; and numerous small bugs have been fixed (see below). The xattr and quota code have been extensively refactored in preparation for more new features coming down the line. Finally, the behavior of DAX between ext4 and xfs has been stabilized, which gets us a step closer to removing the experimental tag from that feature. We have a few new contributors this time around. Welcome, all! I anticipate a second pull request next week for a few small bugfixes that have been trickling in, but this is it for big changes. Summary: - Fix some btree block pingponging problems when swapping extents - Redesign the reflink copy loop so that we only run one remapping operation per transaction. This helps us avoid running out of block reservation on highly deduped filesystems. - Take the MMAPLOCK around filemap_map_pages. - Make inode reclaim fully async so that we avoid stalling processes on flushing inodes to disk. - Reduce inode cluster buffer RMW cycles by attaching the buffer to dirty inodes so we won't let go of the cluster buffer when we know we're going to need it soon. - Add some more checks to the realtime bitmap file scrubber. - Don't trip false lockdep warnings in fs freeze. - Remove various redundant lines of code. - Remove unnecessary calls to xfs_perag_{get,put}. - Preserve I_VERSION state across remounts. - Fix an unmount hang due to AIL going to sleep with a non-empty delwri buffer list. - Fix an error in the inode allocation space reservation macro that caused regressions in generic/531. - Fix a potential livelock when dquot flush fails because the dquot buffer is locked. - Fix a miscalculation when reserving inode quota that could cause users to exceed a hardlimit. - Refactor struct xfs_dquot to use native types for incore fields instead of abusing the ondisk struct for this purpose. This will eventually enable proper y2038+ support, but for now it merely cleans up the quota function declarations. - Actually increment the quota softlimit warning counter so that soft failures turn into hard(er) failures when they exceed the softlimit warning counter limits set by the administrator. - Split incore dquot state flags into their own field and namespace, to avoid mixing them with quota type flags. - Create a new quota type flags namespace so that we can make it obvious when a quota function takes a quota type (user, group, project) as an argument. - Rename the ondisk dquot flags field to type, as that more accurately represents what we store in it. - Drop our bespoke memory allocation flags in favor of GFP_*. - Rearrange the xattr functions so that we no longer mix metadata updates and transaction management (e.g. rolling complex transactions) in the same functions. This work will prepare us for atomic xattr operations (itself a prerequisite for directory backrefs) in future release cycles. - Support FS_DAX_FL (aka FS_XFLAG_DAX) via GETFLAGS/SETFLAGS" * tag 'xfs-5.9-merge-7' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux: (117 commits) fs/xfs: Support that ioctl(SETXFLAGS/GETXFLAGS) can set/get inode DAX on XFS. xfs: Lift -ENOSPC handler from xfs_attr_leaf_addname xfs: Simplify xfs_attr_node_addname xfs: Simplify xfs_attr_leaf_addname xfs: Add helper function xfs_attr_node_removename_rmt xfs: Add helper function xfs_attr_node_removename_setup xfs: Add remote block helper functions xfs: Add helper function xfs_attr_leaf_mark_incomplete xfs: Add helpers xfs_attr_is_shortform and xfs_attr_set_shortform xfs: Remove xfs_trans_roll in xfs_attr_node_removename xfs: Remove unneeded xfs_trans_roll_inode calls xfs: Add helper function xfs_attr_node_shrink xfs: Pull up xfs_attr_rmtval_invalidate xfs: Refactor xfs_attr_rmtval_remove xfs: Pull up trans roll in xfs_attr3_leaf_clearflag xfs: Factor out xfs_attr_rmtval_invalidate xfs: Pull up trans roll from xfs_attr3_leaf_setflag xfs: Refactor xfs_attr_try_sf_addname xfs: Split apart xfs_attr_leaf_addname xfs: Pull up trans handling in xfs_attr3_leaf_flipflags ...
Diffstat (limited to 'fs/xfs/xfs_bmap_util.c')
-rw-r--r--fs/xfs/xfs_bmap_util.c18
1 files changed, 9 insertions, 9 deletions
diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
index 30525861c596..73cafc843cd7 100644
--- a/fs/xfs/xfs_bmap_util.c
+++ b/fs/xfs/xfs_bmap_util.c
@@ -1567,6 +1567,7 @@ xfs_swap_extents(
int lock_flags;
uint64_t f;
int resblks = 0;
+ unsigned int flags = 0;
/*
* Lock the inodes against other IO, page faults and truncate to
@@ -1630,17 +1631,16 @@ xfs_swap_extents(
resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
/*
- * Handle the corner case where either inode might straddle the
- * btree format boundary. If so, the inode could bounce between
- * btree <-> extent format on unmap -> remap cycles, freeing and
- * allocating a bmapbt block each time.
+ * If either inode straddles a bmapbt block allocation boundary,
+ * the rmapbt algorithm triggers repeated allocs and frees as
+ * extents are remapped. This can exhaust the block reservation
+ * prematurely and cause shutdown. Return freed blocks to the
+ * transaction reservation to counter this behavior.
*/
- if (ipnext == (XFS_IFORK_MAXEXT(ip, w) + 1))
- resblks += XFS_IFORK_MAXEXT(ip, w);
- if (tipnext == (XFS_IFORK_MAXEXT(tip, w) + 1))
- resblks += XFS_IFORK_MAXEXT(tip, w);
+ flags |= XFS_TRANS_RES_FDBLKS;
}
- error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
+ error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
+ &tp);
if (error)
goto out_unlock;