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
  ...

1 files changed, 163 insertions, 192 deletions

diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c
index 107bf2a2f344..aac83f9d6107 100644
--- a/fs/xfs/xfs_reflink.c
+++ b/fs/xfs/xfs_reflink.c
@@ -179,7 +179,7 @@ xfs_reflink_trim_around_shared(
 	int			error = 0;
 
 	/* Holes, unwritten, and delalloc extents cannot be shared */
-	if (!xfs_is_cow_inode(ip) || !xfs_bmap_is_real_extent(irec)) {
+	if (!xfs_is_cow_inode(ip) || !xfs_bmap_is_written_extent(irec)) {
 		*shared = false;
 		return 0;
 	}
@@ -655,7 +655,7 @@ xfs_reflink_end_cow_extent(
 	 * preallocations can leak into the range we are called upon, and we
 	 * need to skip them.
 	 */
-	if (!xfs_bmap_is_real_extent(&got)) {
+	if (!xfs_bmap_is_written_extent(&got)) {
 		*end_fsb = del.br_startoff;
 		goto out_cancel;
 	}
@@ -984,40 +984,28 @@ xfs_reflink_ag_has_free_space(
 }
 
 /*
- * Unmap a range of blocks from a file, then map other blocks into the hole.
- * The range to unmap is (destoff : destoff + srcioff + irec->br_blockcount).
- * The extent irec is mapped into dest at irec->br_startoff.
+ * Remap the given extent into the file.  The dmap blockcount will be set to
+ * the number of blocks that were actually remapped.
  */
 STATIC int
 xfs_reflink_remap_extent(
 	struct xfs_inode	*ip,
-	struct xfs_bmbt_irec	*irec,
-	xfs_fileoff_t		destoff,
+	struct xfs_bmbt_irec	*dmap,
 	xfs_off_t		new_isize)
 {
+	struct xfs_bmbt_irec	smap;
 	struct xfs_mount	*mp = ip->i_mount;
-	bool			real_extent = xfs_bmap_is_real_extent(irec);
 	struct xfs_trans	*tp;
-	unsigned int		resblks;
-	struct xfs_bmbt_irec	uirec;
-	xfs_filblks_t		rlen;
-	xfs_filblks_t		unmap_len;
 	xfs_off_t		newlen;
+	int64_t			qres, qdelta;
+	unsigned int		resblks;
+	bool			smap_real;
+	bool			dmap_written = xfs_bmap_is_written_extent(dmap);
+	int			nimaps;
 	int			error;
 
-	unmap_len = irec->br_startoff + irec->br_blockcount - destoff;
-	trace_xfs_reflink_punch_range(ip, destoff, unmap_len);
-
-	/* No reflinking if we're low on space */
-	if (real_extent) {
-		error = xfs_reflink_ag_has_free_space(mp,
-				XFS_FSB_TO_AGNO(mp, irec->br_startblock));
-		if (error)
-			goto out;
-	}
-
 	/* Start a rolling transaction to switch the mappings */
-	resblks = XFS_EXTENTADD_SPACE_RES(ip->i_mount, XFS_DATA_FORK);
+	resblks = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, 0, &tp);
 	if (error)
 		goto out;
@@ -1025,87 +1013,147 @@ xfs_reflink_remap_extent(
 	xfs_ilock(ip, XFS_ILOCK_EXCL);
 	xfs_trans_ijoin(tp, ip, 0);
 
-	/* If we're not just clearing space, then do we have enough quota? */
-	if (real_extent) {
-		error = xfs_trans_reserve_quota_nblks(tp, ip,
-				irec->br_blockcount, 0, XFS_QMOPT_RES_REGBLKS);
+	/*
+	 * Read what's currently mapped in the destination file into smap.
+	 * If smap isn't a hole, we will have to remove it before we can add
+	 * dmap to the destination file.
+	 */
+	nimaps = 1;
+	error = xfs_bmapi_read(ip, dmap->br_startoff, dmap->br_blockcount,
+			&smap, &nimaps, 0);
+	if (error)
+		goto out_cancel;
+	ASSERT(nimaps == 1 && smap.br_startoff == dmap->br_startoff);
+	smap_real = xfs_bmap_is_real_extent(&smap);
+
+	/*
+	 * We can only remap as many blocks as the smaller of the two extent
+	 * maps, because we can only remap one extent at a time.
+	 */
+	dmap->br_blockcount = min(dmap->br_blockcount, smap.br_blockcount);
+	ASSERT(dmap->br_blockcount == smap.br_blockcount);
+
+	trace_xfs_reflink_remap_extent_dest(ip, &smap);
+
+	/*
+	 * Two extents mapped to the same physical block must not have
+	 * different states; that's filesystem corruption.  Move on to the next
+	 * extent if they're both holes or both the same physical extent.
+	 */
+	if (dmap->br_startblock == smap.br_startblock) {
+		if (dmap->br_state != smap.br_state)
+			error = -EFSCORRUPTED;
+		goto out_cancel;
+	}
+
+	/* If both extents are unwritten, leave them alone. */
+	if (dmap->br_state == XFS_EXT_UNWRITTEN &&
+	    smap.br_state == XFS_EXT_UNWRITTEN)
+		goto out_cancel;
+
+	/* No reflinking if the AG of the dest mapping is low on space. */
+	if (dmap_written) {
+		error = xfs_reflink_ag_has_free_space(mp,
+				XFS_FSB_TO_AGNO(mp, dmap->br_startblock));
 		if (error)
 			goto out_cancel;
 	}
 
-	trace_xfs_reflink_remap(ip, irec->br_startoff,
-				irec->br_blockcount, irec->br_startblock);
-
-	/* Unmap the old blocks in the data fork. */
-	rlen = unmap_len;
-	while (rlen) {
-		ASSERT(tp->t_firstblock == NULLFSBLOCK);
-		error = __xfs_bunmapi(tp, ip, destoff, &rlen, 0, 1);
+	/*
+	 * Compute quota reservation if we think the quota block counter for
+	 * this file could increase.
+	 *
+	 * Adding a written extent to the extent map can cause a bmbt split,
+	 * and removing a mapped extent from the extent can cause a bmbt split.
+	 * The two operations cannot both cause a split since they operate on
+	 * the same index in the bmap btree, so we only need a reservation for
+	 * one bmbt split if either thing is happening.
+	 *
+	 * If we are mapping a written extent into the file, we need to have
+	 * enough quota block count reservation to handle the blocks in that
+	 * extent.  We log only the delta to the quota block counts, so if the
+	 * extent we're unmapping also has blocks allocated to it, we don't
+	 * need a quota reservation for the extent itself.
+	 *
+	 * Note that if we're replacing a delalloc reservation with a written
+	 * extent, we have to take the full quota reservation because removing
+	 * the delalloc reservation gives the block count back to the quota
+	 * count.  This is suboptimal, but the VFS flushed the dest range
+	 * before we started.  That should have removed all the delalloc
+	 * reservations, but we code defensively.
+	 */
+	qres = qdelta = 0;
+	if (smap_real || dmap_written)
+		qres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
+	if (!smap_real && dmap_written)
+		qres += dmap->br_blockcount;
+	if (qres > 0) {
+		error = xfs_trans_reserve_quota_nblks(tp, ip, qres, 0,
+				XFS_QMOPT_RES_REGBLKS);
 		if (error)
 			goto out_cancel;
+	}
 
+	if (smap_real) {
 		/*
-		 * Trim the extent to whatever got unmapped.
-		 * Remember, bunmapi works backwards.
+		 * If the extent we're unmapping is backed by storage (written
+		 * or not), unmap the extent and drop its refcount.
 		 */
-		uirec.br_startblock = irec->br_startblock + rlen;
-		uirec.br_startoff = irec->br_startoff + rlen;
-		uirec.br_blockcount = unmap_len - rlen;
-		uirec.br_state = irec->br_state;
-		unmap_len = rlen;
-
-		/* If this isn't a real mapping, we're done. */
-		if (!real_extent || uirec.br_blockcount == 0)
-			goto next_extent;
+		xfs_bmap_unmap_extent(tp, ip, &smap);
+		xfs_refcount_decrease_extent(tp, &smap);
+		qdelta -= smap.br_blockcount;
+	} else if (smap.br_startblock == DELAYSTARTBLOCK) {
+		xfs_filblks_t	len = smap.br_blockcount;
 
-		trace_xfs_reflink_remap(ip, uirec.br_startoff,
-				uirec.br_blockcount, uirec.br_startblock);
-
-		/* Update the refcount tree */
-		xfs_refcount_increase_extent(tp, &uirec);
-
-		/* Map the new blocks into the data fork. */
-		xfs_bmap_map_extent(tp, ip, &uirec);
+		/*
+		 * If the extent we're unmapping is a delalloc reservation,
+		 * we can use the regular bunmapi function to release the
+		 * incore state.  Dropping the delalloc reservation takes care
+		 * of the quota reservation for us.
+		 */
+		error = __xfs_bunmapi(NULL, ip, smap.br_startoff, &len, 0, 1);
+		if (error)
+			goto out_cancel;
+		ASSERT(len == 0);
+	}
 
-		/* Update quota accounting. */
-		xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT,
-				uirec.br_blockcount);
+	/*
+	 * If the extent we're sharing is backed by written storage, increase
+	 * its refcount and map it into the file.
+	 */
+	if (dmap_written) {
+		xfs_refcount_increase_extent(tp, dmap);
+		xfs_bmap_map_extent(tp, ip, dmap);
+		qdelta += dmap->br_blockcount;
+	}
 
-		/* Update dest isize if needed. */
-		newlen = XFS_FSB_TO_B(mp,
-				uirec.br_startoff + uirec.br_blockcount);
-		newlen = min_t(xfs_off_t, newlen, new_isize);
-		if (newlen > i_size_read(VFS_I(ip))) {
-			trace_xfs_reflink_update_inode_size(ip, newlen);
-			i_size_write(VFS_I(ip), newlen);
-			ip->i_d.di_size = newlen;
-			xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
-		}
+	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, qdelta);
 
-next_extent:
-		/* Process all the deferred stuff. */
-		error = xfs_defer_finish(&tp);
-		if (error)
-			goto out_cancel;
+	/* Update dest isize if needed. */
+	newlen = XFS_FSB_TO_B(mp, dmap->br_startoff + dmap->br_blockcount);
+	newlen = min_t(xfs_off_t, newlen, new_isize);
+	if (newlen > i_size_read(VFS_I(ip))) {
+		trace_xfs_reflink_update_inode_size(ip, newlen);
+		i_size_write(VFS_I(ip), newlen);
+		ip->i_d.di_size = newlen;
+		xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 	}
 
+	/* Commit everything and unlock. */
 	error = xfs_trans_commit(tp);
-	xfs_iunlock(ip, XFS_ILOCK_EXCL);
-	if (error)
-		goto out;
-	return 0;
+	goto out_unlock;
 
 out_cancel:
 	xfs_trans_cancel(tp);
+out_unlock:
 	xfs_iunlock(ip, XFS_ILOCK_EXCL);
 out:
-	trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
+	if (error)
+		trace_xfs_reflink_remap_extent_error(ip, error, _RET_IP_);
 	return error;
 }
 
-/*
- * Iteratively remap one file's extents (and holes) to another's.
- */
+/* Remap a range of one file to the other. */
 int
 xfs_reflink_remap_blocks(
 	struct xfs_inode	*src,
@@ -1116,25 +1164,22 @@ xfs_reflink_remap_blocks(
 	loff_t			*remapped)
 {
 	struct xfs_bmbt_irec	imap;
-	xfs_fileoff_t		srcoff;
-	xfs_fileoff_t		destoff;
+	struct xfs_mount	*mp = src->i_mount;
+	xfs_fileoff_t		srcoff = XFS_B_TO_FSBT(mp, pos_in);
+	xfs_fileoff_t		destoff = XFS_B_TO_FSBT(mp, pos_out);
 	xfs_filblks_t		len;
-	xfs_filblks_t		range_len;
 	xfs_filblks_t		remapped_len = 0;
 	xfs_off_t		new_isize = pos_out + remap_len;
 	int			nimaps;
 	int			error = 0;
 
-	destoff = XFS_B_TO_FSBT(src->i_mount, pos_out);
-	srcoff = XFS_B_TO_FSBT(src->i_mount, pos_in);
-	len = XFS_B_TO_FSB(src->i_mount, remap_len);
+	len = min_t(xfs_filblks_t, XFS_B_TO_FSB(mp, remap_len),
+			XFS_MAX_FILEOFF);
 
-	/* drange = (destoff, destoff + len); srange = (srcoff, srcoff + len) */
-	while (len) {
-		uint		lock_mode;
+	trace_xfs_reflink_remap_blocks(src, srcoff, len, dest, destoff);
 
-		trace_xfs_reflink_remap_blocks_loop(src, srcoff, len,
-				dest, destoff);
+	while (len > 0) {
+		unsigned int	lock_mode;
 
 		/* Read extent from the source file */
 		nimaps = 1;
@@ -1143,18 +1188,25 @@ xfs_reflink_remap_blocks(
 		xfs_iunlock(src, lock_mode);
 		if (error)
 			break;
-		ASSERT(nimaps == 1);
-
-		trace_xfs_reflink_remap_imap(src, srcoff, len, XFS_DATA_FORK,
-				&imap);
+		/*
+		 * The caller supposedly flushed all dirty pages in the source
+		 * file range, which means that writeback should have allocated
+		 * or deleted all delalloc reservations in that range.  If we
+		 * find one, that's a good sign that something is seriously
+		 * wrong here.
+		 */
+		ASSERT(nimaps == 1 && imap.br_startoff == srcoff);
+		if (imap.br_startblock == DELAYSTARTBLOCK) {
+			ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
+			error = -EFSCORRUPTED;
+			break;
+		}
 
-		/* Translate imap into the destination file. */
-		range_len = imap.br_startoff + imap.br_blockcount - srcoff;
-		imap.br_startoff += destoff - srcoff;
+		trace_xfs_reflink_remap_extent_src(src, &imap);
 
-		/* Clear dest from destoff to the end of imap and map it in. */
-		error = xfs_reflink_remap_extent(dest, &imap, destoff,
-				new_isize);
+		/* Remap into the destination file at the given offset. */
+		imap.br_startoff = destoff;
+		error = xfs_reflink_remap_extent(dest, &imap, new_isize);
 		if (error)
 			break;
 
@@ -1164,10 +1216,10 @@ xfs_reflink_remap_blocks(
 		}
 
 		/* Advance drange/srange */
-		srcoff += range_len;
-		destoff += range_len;
-		len -= range_len;
-		remapped_len += range_len;
+		srcoff += imap.br_blockcount;
+		destoff += imap.br_blockcount;
+		len -= imap.br_blockcount;
+		remapped_len += imap.br_blockcount;
 	}
 
 	if (error)
@@ -1178,81 +1230,6 @@ xfs_reflink_remap_blocks(
 }
 
 /*
- * Grab the exclusive iolock for a data copy from src to dest, making sure to
- * abide vfs locking order (lowest pointer value goes first) and breaking the
- * layout leases before proceeding.  The loop is needed because we cannot call
- * the blocking break_layout() with the iolocks held, and therefore have to
- * back out both locks.
- */
-static int
-xfs_iolock_two_inodes_and_break_layout(
-	struct inode		*src,
-	struct inode		*dest)
-{
-	int			error;
-
-	if (src > dest)
-		swap(src, dest);
-
-retry:
-	/* Wait to break both inodes' layouts before we start locking. */
-	error = break_layout(src, true);
-	if (error)
-		return error;
-	if (src != dest) {
-		error = break_layout(dest, true);
-		if (error)
-			return error;
-	}
-
-	/* Lock one inode and make sure nobody got in and leased it. */
-	inode_lock(src);
-	error = break_layout(src, false);
-	if (error) {
-		inode_unlock(src);
-		if (error == -EWOULDBLOCK)
-			goto retry;
-		return error;
-	}
-
-	if (src == dest)
-		return 0;
-
-	/* Lock the other inode and make sure nobody got in and leased it. */
-	inode_lock_nested(dest, I_MUTEX_NONDIR2);
-	error = break_layout(dest, false);
-	if (error) {
-		inode_unlock(src);
-		inode_unlock(dest);
-		if (error == -EWOULDBLOCK)
-			goto retry;
-		return error;
-	}
-
-	return 0;
-}
-
-/* Unlock both inodes after they've been prepped for a range clone. */
-void
-xfs_reflink_remap_unlock(
-	struct file		*file_in,
-	struct file		*file_out)
-{
-	struct inode		*inode_in = file_inode(file_in);
-	struct xfs_inode	*src = XFS_I(inode_in);
-	struct inode		*inode_out = file_inode(file_out);
-	struct xfs_inode	*dest = XFS_I(inode_out);
-	bool			same_inode = (inode_in == inode_out);
-
-	xfs_iunlock(dest, XFS_MMAPLOCK_EXCL);
-	if (!same_inode)
-		xfs_iunlock(src, XFS_MMAPLOCK_EXCL);
-	inode_unlock(inode_out);
-	if (!same_inode)
-		inode_unlock(inode_in);
-}
-
-/*
  * If we're reflinking to a point past the destination file's EOF, we must
  * zero any speculative post-EOF preallocations that sit between the old EOF
  * and the destination file offset.
@@ -1314,18 +1291,12 @@ xfs_reflink_remap_prep(
 	struct xfs_inode	*src = XFS_I(inode_in);
 	struct inode		*inode_out = file_inode(file_out);
 	struct xfs_inode	*dest = XFS_I(inode_out);
-	bool			same_inode = (inode_in == inode_out);
-	ssize_t			ret;
+	int			ret;
 
 	/* Lock both files against IO */
-	ret = xfs_iolock_two_inodes_and_break_layout(inode_in, inode_out);
+	ret = xfs_ilock2_io_mmap(src, dest);
 	if (ret)
 		return ret;
-	if (same_inode)
-		xfs_ilock(src, XFS_MMAPLOCK_EXCL);
-	else
-		xfs_lock_two_inodes(src, XFS_MMAPLOCK_EXCL, dest,
-				XFS_MMAPLOCK_EXCL);
 
 	/* Check file eligibility and prepare for block sharing. */
 	ret = -EINVAL;
@@ -1339,7 +1310,7 @@ xfs_reflink_remap_prep(
 
 	ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
 			len, remap_flags);
-	if (ret < 0 || *len == 0)
+	if (ret || *len == 0)
 		goto out_unlock;
 
 	/* Attach dquots to dest inode before changing block map */
@@ -1374,9 +1345,9 @@ xfs_reflink_remap_prep(
 	if (ret)
 		goto out_unlock;
 
-	return 1;
+	return 0;
 out_unlock:
-	xfs_reflink_remap_unlock(file_in, file_out);
+	xfs_iunlock2_io_mmap(src, dest);
 	return ret;
 }