Merge tag 'xfs-6.4-rc5-fixes' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux

Pull xfs fixes from Dave Chinner:
 "These are a set of regression fixes discovered on recent kernels. I
  was hoping to send this to you a week and half ago, but events out of
  my control delayed finalising the changes until early this week.

  Whilst the diffstat looks large for this stage of the merge window, a
  large chunk of it comes from moving the guts of one function from one
  file to another i.e. it's the same code, it is just run in a different
  context where it is safe to hold a specific lock. Otherwise the
  individual changes are relatively small and straigtht forward.

  Summary:

   - Propagate unlinked inode list corruption back up to log recovery
     (regression fix)

   - improve corruption detection for AGFL entries, AGFL indexes and
     XEFI extents (syzkaller fuzzer oops report)

   - Avoid double perag reference release (regression fix)

   - Improve extent merging detection in scrub (regression fix)

   - Fix a new undefined high bit shift (regression fix)

   - Fix for AGF vs inode cluster buffer deadlock (regression fix)"

* tag 'xfs-6.4-rc5-fixes' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux:
  xfs: collect errors from inodegc for unlinked inode recovery
  xfs: validate block number being freed before adding to xefi
  xfs: validity check agbnos on the AGFL
  xfs: fix agf/agfl verification on v4 filesystems
  xfs: fix double xfs_perag_rele() in xfs_filestream_pick_ag()
  xfs: fix broken logic when detecting mergeable bmap records
  xfs: Fix undefined behavior of shift into sign bit
  xfs: fix AGF vs inode cluster buffer deadlock
  xfs: defered work could create precommits
  xfs: restore allocation trylock iteration
  xfs: buffer pins need to hold a buffer reference

fs/xfs/libxfs/xfs_ag.c

@@ -984,7 +984,10 @@ xfs_ag_shrink_space(
 		if (err2 != -ENOSPC)
 			goto resv_err;
 
-		__xfs_free_extent_later(*tpp, args.fsbno, delta, NULL, true);
+		err2 = __xfs_free_extent_later(*tpp, args.fsbno, delta, NULL,
+				true);
+		if (err2)
+			goto resv_err;
 
 		/*
 		 * Roll the transaction before trying to re-init the per-ag

fs/xfs/libxfs/xfs_alloc.c

@@ -628,6 +628,25 @@ xfs_alloc_fixup_trees(
 	return 0;
 }
 
+/*
+ * We do not verify the AGFL contents against AGF-based index counters here,
+ * even though we may have access to the perag that contains shadow copies. We
+ * don't know if the AGF based counters have been checked, and if they have they
+ * still may be inconsistent because they haven't yet been reset on the first
+ * allocation after the AGF has been read in.
+ *
+ * This means we can only check that all agfl entries contain valid or null
+ * values because we can't reliably determine the active range to exclude
+ * NULLAGBNO as a valid value.
+ *
+ * However, we can't even do that for v4 format filesystems because there are
+ * old versions of mkfs out there that does not initialise the AGFL to known,
+ * verifiable values. HEnce we can't tell the difference between a AGFL block
+ * allocated by mkfs and a corrupted AGFL block here on v4 filesystems.
+ *
+ * As a result, we can only fully validate AGFL block numbers when we pull them
+ * from the freelist in xfs_alloc_get_freelist().
+ */
 static xfs_failaddr_t
 xfs_agfl_verify(
 	struct xfs_buf	*bp)
@@ -637,12 +656,6 @@ xfs_agfl_verify(
 	__be32		*agfl_bno = xfs_buf_to_agfl_bno(bp);
 	int		i;
 
-	/*
-	 * There is no verification of non-crc AGFLs because mkfs does not
-	 * initialise the AGFL to zero or NULL. Hence the only valid part of the
-	 * AGFL is what the AGF says is active. We can't get to the AGF, so we
-	 * can't verify just those entries are valid.
-	 */
 	if (!xfs_has_crc(mp))
 		return NULL;
 
@@ -2321,12 +2334,16 @@ xfs_free_agfl_block(
 }
 
 /*
- * Check the agfl fields of the agf for inconsistency or corruption. The purpose
- * is to detect an agfl header padding mismatch between current and early v5
- * kernels. This problem manifests as a 1-slot size difference between the
- * on-disk flcount and the active [first, last] range of a wrapped agfl. This
- * may also catch variants of agfl count corruption unrelated to padding. Either
- * way, we'll reset the agfl and warn the user.
+ * Check the agfl fields of the agf for inconsistency or corruption.
+ *
+ * The original purpose was to detect an agfl header padding mismatch between
+ * current and early v5 kernels. This problem manifests as a 1-slot size
+ * difference between the on-disk flcount and the active [first, last] range of
+ * a wrapped agfl.
+ *
+ * However, we need to use these same checks to catch agfl count corruptions
+ * unrelated to padding. This could occur on any v4 or v5 filesystem, so either
+ * way, we need to reset the agfl and warn the user.
  *
  * Return true if a reset is required before the agfl can be used, false
  * otherwise.
@@ -2342,10 +2359,6 @@ xfs_agfl_needs_reset(
 	int			agfl_size = xfs_agfl_size(mp);
 	int			active;
 
-	/* no agfl header on v4 supers */
-	if (!xfs_has_crc(mp))
-		return false;
-
 	/*
 	 * The agf read verifier catches severe corruption of these fields.
 	 * Repeat some sanity checks to cover a packed -> unpacked mismatch if
@@ -2418,7 +2431,7 @@ xfs_agfl_reset(
  * the real allocation can proceed. Deferring the free disconnects freeing up
  * the AGFL slot from freeing the block.
  */
-STATIC void
+static int
 xfs_defer_agfl_block(
 	struct xfs_trans		*tp,
 	xfs_agnumber_t			agno,
@@ -2437,17 +2450,21 @@ xfs_defer_agfl_block(
 	xefi->xefi_blockcount = 1;
 	xefi->xefi_owner = oinfo->oi_owner;
 
+	if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, xefi->xefi_startblock)))
+		return -EFSCORRUPTED;
+
 	trace_xfs_agfl_free_defer(mp, agno, 0, agbno, 1);
 
 	xfs_extent_free_get_group(mp, xefi);
 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_AGFL_FREE, &xefi->xefi_list);
+	return 0;
 }
 
 /*
  * Add the extent to the list of extents to be free at transaction end.
  * The list is maintained sorted (by block number).
  */
-void
+int
 __xfs_free_extent_later(
 	struct xfs_trans		*tp,
 	xfs_fsblock_t			bno,
@@ -2474,6 +2491,9 @@ __xfs_free_extent_later(
 #endif
 	ASSERT(xfs_extfree_item_cache != NULL);
 
+	if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbext(mp, bno, len)))
+		return -EFSCORRUPTED;
+
 	xefi = kmem_cache_zalloc(xfs_extfree_item_cache,
 			       GFP_KERNEL | __GFP_NOFAIL);
 	xefi->xefi_startblock = bno;
@@ -2497,6 +2517,7 @@ __xfs_free_extent_later(
 
 	xfs_extent_free_get_group(mp, xefi);
 	xfs_defer_add(tp, XFS_DEFER_OPS_TYPE_FREE, &xefi->xefi_list);
+	return 0;
 }
 
 #ifdef DEBUG
@@ -2657,7 +2678,9 @@ xfs_alloc_fix_freelist(
 			goto out_agbp_relse;
 
 		/* defer agfl frees */
-		xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo);
+		error = xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo);
+		if (error)
+			goto out_agbp_relse;
 	}
 
 	targs.tp = tp;
@@ -2767,6 +2790,9 @@ xfs_alloc_get_freelist(
 	 */
 	agfl_bno = xfs_buf_to_agfl_bno(agflbp);
 	bno = be32_to_cpu(agfl_bno[be32_to_cpu(agf->agf_flfirst)]);
+	if (XFS_IS_CORRUPT(tp->t_mountp, !xfs_verify_agbno(pag, bno)))
+		return -EFSCORRUPTED;
+
 	be32_add_cpu(&agf->agf_flfirst, 1);
 	xfs_trans_brelse(tp, agflbp);
 	if (be32_to_cpu(agf->agf_flfirst) == xfs_agfl_size(mp))
@@ -2889,6 +2915,19 @@ xfs_alloc_put_freelist(
 	return 0;
 }
 
+/*
+ * Verify the AGF is consistent.
+ *
+ * We do not verify the AGFL indexes in the AGF are fully consistent here
+ * because of issues with variable on-disk structure sizes. Instead, we check
+ * the agfl indexes for consistency when we initialise the perag from the AGF
+ * information after a read completes.
+ *
+ * If the index is inconsistent, then we mark the perag as needing an AGFL
+ * reset. The first AGFL update performed then resets the AGFL indexes and
+ * refills the AGFL with known good free blocks, allowing the filesystem to
+ * continue operating normally at the cost of a few leaked free space blocks.
+ */
 static xfs_failaddr_t
 xfs_agf_verify(
 	struct xfs_buf		*bp)
@@ -2962,7 +3001,6 @@ xfs_agf_verify(
 		return __this_address;
 
 	return NULL;
-
 }
 
 static void
@@ -3187,7 +3225,8 @@ xfs_alloc_vextent_check_args(
  */
 static int
 xfs_alloc_vextent_prepare_ag(
-	struct xfs_alloc_arg	*args)
+	struct xfs_alloc_arg	*args,
+	uint32_t		flags)
 {
 	bool			need_pag = !args->pag;
 	int			error;
@@ -3196,7 +3235,7 @@ xfs_alloc_vextent_prepare_ag(
 		args->pag = xfs_perag_get(args->mp, args->agno);
 
 	args->agbp = NULL;
-	error = xfs_alloc_fix_freelist(args, 0);
+	error = xfs_alloc_fix_freelist(args, flags);
 	if (error) {
 		trace_xfs_alloc_vextent_nofix(args);
 		if (need_pag)
@@ -3336,7 +3375,7 @@ xfs_alloc_vextent_this_ag(
 		return error;
 	}
 
-	error = xfs_alloc_vextent_prepare_ag(args);
+	error = xfs_alloc_vextent_prepare_ag(args, 0);
 	if (!error && args->agbp)
 		error = xfs_alloc_ag_vextent_size(args);
 
@@ -3380,7 +3419,7 @@ xfs_alloc_vextent_iterate_ags(
 	for_each_perag_wrap_range(mp, start_agno, restart_agno,
 			mp->m_sb.sb_agcount, agno, args->pag) {
 		args->agno = agno;
-		error = xfs_alloc_vextent_prepare_ag(args);
+		error = xfs_alloc_vextent_prepare_ag(args, flags);
 		if (error)
 			break;
 		if (!args->agbp) {
@@ -3546,7 +3585,7 @@ xfs_alloc_vextent_exact_bno(
 		return error;
 	}
 
-	error = xfs_alloc_vextent_prepare_ag(args);
+	error = xfs_alloc_vextent_prepare_ag(args, 0);
 	if (!error && args->agbp)
 		error = xfs_alloc_ag_vextent_exact(args);
 
@@ -3587,7 +3626,7 @@ xfs_alloc_vextent_near_bno(
 	if (needs_perag)
 		args->pag = xfs_perag_grab(mp, args->agno);
 
-	error = xfs_alloc_vextent_prepare_ag(args);
+	error = xfs_alloc_vextent_prepare_ag(args, 0);
 	if (!error && args->agbp)
 		error = xfs_alloc_ag_vextent_near(args);
 

fs/xfs/libxfs/xfs_alloc.h

@@ -230,7 +230,7 @@ xfs_buf_to_agfl_bno(
 	return bp->b_addr;
 }
 
-void __xfs_free_extent_later(struct xfs_trans *tp, xfs_fsblock_t bno,
+int __xfs_free_extent_later(struct xfs_trans *tp, xfs_fsblock_t bno,
 		xfs_filblks_t len, const struct xfs_owner_info *oinfo,
 		bool skip_discard);
 
@@ -254,14 +254,14 @@ void xfs_extent_free_get_group(struct xfs_mount *mp,
 #define XFS_EFI_ATTR_FORK	(1U << 1) /* freeing attr fork block */
 #define XFS_EFI_BMBT_BLOCK	(1U << 2) /* freeing bmap btree block */
 
-static inline void
+static inline int
 xfs_free_extent_later(
 	struct xfs_trans		*tp,
 	xfs_fsblock_t			bno,
 	xfs_filblks_t			len,
 	const struct xfs_owner_info	*oinfo)
 {
-	__xfs_free_extent_later(tp, bno, len, oinfo, false);
+	return __xfs_free_extent_later(tp, bno, len, oinfo, false);
 }
 
 

fs/xfs/libxfs/xfs_bmap.c

@@ -572,8 +572,12 @@ xfs_bmap_btree_to_extents(
 	cblock = XFS_BUF_TO_BLOCK(cbp);
 	if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
 		return error;
+
 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, whichfork);
-	xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo);
+	error = xfs_free_extent_later(cur->bc_tp, cbno, 1, &oinfo);
+	if (error)
+		return error;
+
 	ip->i_nblocks--;
 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 	xfs_trans_binval(tp, cbp);
@@ -5230,10 +5234,12 @@ xfs_bmap_del_extent_real(
 		if (xfs_is_reflink_inode(ip) && whichfork == XFS_DATA_FORK) {
 			xfs_refcount_decrease_extent(tp, del);
 		} else {
-			__xfs_free_extent_later(tp, del->br_startblock,
+			error = __xfs_free_extent_later(tp, del->br_startblock,
 					del->br_blockcount, NULL,
 					(bflags & XFS_BMAPI_NODISCARD) ||
 					del->br_state == XFS_EXT_UNWRITTEN);
+			if (error)
+				goto done;
 		}
 	}
 

fs/xfs/libxfs/xfs_bmap_btree.c

@@ -268,11 +268,14 @@ xfs_bmbt_free_block(
 	struct xfs_trans	*tp = cur->bc_tp;
 	xfs_fsblock_t		fsbno = XFS_DADDR_TO_FSB(mp, xfs_buf_daddr(bp));
 	struct xfs_owner_info	oinfo;
+	int			error;
 
 	xfs_rmap_ino_bmbt_owner(&oinfo, ip->i_ino, cur->bc_ino.whichfork);
-	xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
-	ip->i_nblocks--;
+	error = xfs_free_extent_later(cur->bc_tp, fsbno, 1, &oinfo);
+	if (error)
+		return error;
 
+	ip->i_nblocks--;
 	xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
 	return 0;

fs/xfs/libxfs/xfs_ialloc.c

@@ -1834,7 +1834,7 @@ xfs_dialloc(
  * might be sparse and only free the regions that are allocated as part of the
  * chunk.
  */
-STATIC void
+static int
 xfs_difree_inode_chunk(
 	struct xfs_trans		*tp,
 	xfs_agnumber_t			agno,
@@ -1851,10 +1851,10 @@ xfs_difree_inode_chunk(
 
 	if (!xfs_inobt_issparse(rec->ir_holemask)) {
 		/* not sparse, calculate extent info directly */
-		xfs_free_extent_later(tp, XFS_AGB_TO_FSB(mp, agno, sagbno),
-				  M_IGEO(mp)->ialloc_blks,
-				  &XFS_RMAP_OINFO_INODES);
-		return;
+		return xfs_free_extent_later(tp,
+				XFS_AGB_TO_FSB(mp, agno, sagbno),
+				M_IGEO(mp)->ialloc_blks,
+				&XFS_RMAP_OINFO_INODES);
 	}
 
 	/* holemask is only 16-bits (fits in an unsigned long) */
@@ -1871,6 +1871,8 @@ xfs_difree_inode_chunk(
 						XFS_INOBT_HOLEMASK_BITS);
 	nextbit = startidx + 1;
 	while (startidx < XFS_INOBT_HOLEMASK_BITS) {
+		int error;
+
 		nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
 					     nextbit);
 		/*
@@ -1896,8 +1898,11 @@ xfs_difree_inode_chunk(
 
 		ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
 		ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
-		xfs_free_extent_later(tp, XFS_AGB_TO_FSB(mp, agno, agbno),
-				  contigblk, &XFS_RMAP_OINFO_INODES);
+		error = xfs_free_extent_later(tp,
+				XFS_AGB_TO_FSB(mp, agno, agbno),
+				contigblk, &XFS_RMAP_OINFO_INODES);
+		if (error)
+			return error;
 
 		/* reset range to current bit and carry on... */
 		startidx = endidx = nextbit;
@@ -1905,6 +1910,7 @@ xfs_difree_inode_chunk(
 next:
 		nextbit++;
 	}
+	return 0;
 }
 
 STATIC int
@@ -2003,7 +2009,9 @@ xfs_difree_inobt(
 			goto error0;
 		}
 
-		xfs_difree_inode_chunk(tp, pag->pag_agno, &rec);
+		error = xfs_difree_inode_chunk(tp, pag->pag_agno, &rec);
+		if (error)
+			goto error0;
 	} else {
 		xic->deleted = false;
 

fs/xfs/libxfs/xfs_log_format.h

@@ -324,7 +324,6 @@ struct xfs_inode_log_format_32 {
 #define XFS_ILOG_DOWNER	0x200	/* change the data fork owner on replay */
 #define XFS_ILOG_AOWNER	0x400	/* change the attr fork owner on replay */
 
-
 /*
  * The timestamps are dirty, but not necessarily anything else in the inode
  * core.  Unlike the other fields above this one must never make it to disk
@@ -333,6 +332,14 @@ struct xfs_inode_log_format_32 {
  */
 #define XFS_ILOG_TIMESTAMP	0x4000
 
+/*
+ * The version field has been changed, but not necessarily anything else of
+ * interest. This must never make it to disk - it is used purely to ensure that
+ * the inode item ->precommit operation can update the fsync flag triggers
+ * in the inode item correctly.
+ */
+#define XFS_ILOG_IVERSION	0x8000
+
 #define	XFS_ILOG_NONCORE	(XFS_ILOG_DDATA | XFS_ILOG_DEXT | \
 				 XFS_ILOG_DBROOT | XFS_ILOG_DEV | \
 				 XFS_ILOG_ADATA | XFS_ILOG_AEXT | \

fs/xfs/libxfs/xfs_refcount.c

@@ -1151,8 +1151,10 @@ xfs_refcount_adjust_extents(
 				fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
 						cur->bc_ag.pag->pag_agno,
 						tmp.rc_startblock);
-				xfs_free_extent_later(cur->bc_tp, fsbno,
+				error = xfs_free_extent_later(cur->bc_tp, fsbno,
 						  tmp.rc_blockcount, NULL);
+				if (error)
+					goto out_error;
 			}
 
 			(*agbno) += tmp.rc_blockcount;
@@ -1210,8 +1212,10 @@ xfs_refcount_adjust_extents(
 			fsbno = XFS_AGB_TO_FSB(cur->bc_mp,
 					cur->bc_ag.pag->pag_agno,
 					ext.rc_startblock);
-			xfs_free_extent_later(cur->bc_tp, fsbno,
+			error = xfs_free_extent_later(cur->bc_tp, fsbno,
 					ext.rc_blockcount, NULL);
+			if (error)
+				goto out_error;
 		}
 
 skip:
@@ -1976,7 +1980,10 @@ xfs_refcount_recover_cow_leftovers(
 				rr->rr_rrec.rc_blockcount);
 
 		/* Free the block. */
-		xfs_free_extent_later(tp, fsb, rr->rr_rrec.rc_blockcount, NULL);
+		error = xfs_free_extent_later(tp, fsb,
+				rr->rr_rrec.rc_blockcount, NULL);
+		if (error)
+			goto out_trans;
 
 		error = xfs_trans_commit(tp);
 		if (error)

fs/xfs/libxfs/xfs_trans_inode.c

@@ -40,9 +40,8 @@ xfs_trans_ijoin(
 	iip->ili_lock_flags = lock_flags;
 	ASSERT(!xfs_iflags_test(ip, XFS_ISTALE));
 
-	/*
-	 * Get a log_item_desc to point at the new item.
-	 */
+	/* Reset the per-tx dirty context and add the item to the tx. */
+	iip->ili_dirty_flags = 0;
 	xfs_trans_add_item(tp, &iip->ili_item);
 }
 
@@ -76,17 +75,10 @@ xfs_trans_ichgtime(
 /*
  * This is called to mark the fields indicated in fieldmask as needing to be
  * logged when the transaction is committed.  The inode must already be
- * associated with the given transaction.
- *
- * The values for fieldmask are defined in xfs_inode_item.h.  We always log all
- * of the core inode if any of it has changed, and we always log all of the
- * inline data/extents/b-tree root if any of them has changed.
- *
- * Grab and pin the cluster buffer associated with this inode to avoid RMW
- * cycles at inode writeback time. Avoid the need to add error handling to every
- * xfs_trans_log_inode() call by shutting down on read error.  This will cause
- * transactions to fail and everything to error out, just like if we return a
- * read error in a dirty transaction and cancel it.
+ * associated with the given transaction. All we do here is record where the
+ * inode was dirtied and mark the transaction and inode log item dirty;
+ * everything else is done in the ->precommit log item operation after the
+ * changes in the transaction have been completed.
  */
 void
 xfs_trans_log_inode(
@@ -96,7 +88,6 @@ xfs_trans_log_inode(
 {
 	struct xfs_inode_log_item *iip = ip->i_itemp;
 	struct inode		*inode = VFS_I(ip);
-	uint			iversion_flags = 0;
 
 	ASSERT(iip);
 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
@@ -104,18 +95,6 @@ xfs_trans_log_inode(
 
 	tp->t_flags |= XFS_TRANS_DIRTY;
 
-	/*
-	 * Don't bother with i_lock for the I_DIRTY_TIME check here, as races
-	 * don't matter - we either will need an extra transaction in 24 hours
-	 * to log the timestamps, or will clear already cleared fields in the
-	 * worst case.
-	 */
-	if (inode->i_state & I_DIRTY_TIME) {
-		spin_lock(&inode->i_lock);
-		inode->i_state &= ~I_DIRTY_TIME;
-		spin_unlock(&inode->i_lock);
-	}
-
 	/*
 	 * First time we log the inode in a transaction, bump the inode change
 	 * counter if it is configured for this to occur. While we have the
@@ -128,86 +107,10 @@ xfs_trans_log_inode(
 	if (!test_and_set_bit(XFS_LI_DIRTY, &iip->ili_item.li_flags)) {
 		if (IS_I_VERSION(inode) &&
 		    inode_maybe_inc_iversion(inode, flags & XFS_ILOG_CORE))
-			iversion_flags = XFS_ILOG_CORE;
-	}
-
-	/*
-	 * If we're updating the inode core or the timestamps and it's possible
-	 * to upgrade this inode to bigtime format, do so now.
-	 */
-	if ((flags & (XFS_ILOG_CORE | XFS_ILOG_TIMESTAMP)) &&
-	    xfs_has_bigtime(ip->i_mount) &&
-	    !xfs_inode_has_bigtime(ip)) {
-		ip->i_diflags2 |= XFS_DIFLAG2_BIGTIME;
-		flags |= XFS_ILOG_CORE;
-	}
-
-	/*
-	 * Inode verifiers do not check that the extent size hint is an integer
-	 * multiple of the rt extent size on a directory with both rtinherit
-	 * and extszinherit flags set.  If we're logging a directory that is
-	 * misconfigured in this way, clear the hint.
-	 */
-	if ((ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
-	    (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) &&
-	    (ip->i_extsize % ip->i_mount->m_sb.sb_rextsize) > 0) {
-		ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
-				   XFS_DIFLAG_EXTSZINHERIT);
-		ip->i_extsize = 0;
-		flags |= XFS_ILOG_CORE;
+			flags |= XFS_ILOG_IVERSION;
 	}
 
-	/*
-	 * Record the specific change for fdatasync optimisation. This allows
-	 * fdatasync to skip log forces for inodes that are only timestamp
-	 * dirty.
-	 */
-	spin_lock(&iip->ili_lock);
-	iip->ili_fsync_fields |= flags;
-
-	if (!iip->ili_item.li_buf) {
-		struct xfs_buf	*bp;
-		int		error;
-
-		/*
-		 * We hold the ILOCK here, so this inode is not going to be
-		 * flushed while we are here. Further, because there is no
-		 * buffer attached to the item, we know that there is no IO in
-		 * progress, so nothing will clear the ili_fields while we read
-		 * in the buffer. Hence we can safely drop the spin lock and
-		 * read the buffer knowing that the state will not change from
-		 * here.
-		 */
-		spin_unlock(&iip->ili_lock);
-		error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &bp);
-		if (error) {
-			xfs_force_shutdown(ip->i_mount, SHUTDOWN_META_IO_ERROR);
-			return;
-		}
-
-		/*
-		 * We need an explicit buffer reference for the log item but
-		 * don't want the buffer to remain attached to the transaction.
-		 * Hold the buffer but release the transaction reference once
-		 * we've attached the inode log item to the buffer log item
-		 * list.
-		 */
-		xfs_buf_hold(bp);
-		spin_lock(&iip->ili_lock);
-		iip->ili_item.li_buf = bp;
-		bp->b_flags |= _XBF_INODES;
-		list_add_tail(&iip->ili_item.li_bio_list, &bp->b_li_list);
-		xfs_trans_brelse(tp, bp);
-	}
-
-	/*
-	 * Always OR in the bits from the ili_last_fields field.  This is to
-	 * coordinate with the xfs_iflush() and xfs_buf_inode_iodone() routines
-	 * in the eventual clearing of the ili_fields bits.  See the big comment
-	 * in xfs_iflush() for an explanation of this coordination mechanism.
-	 */
-	iip->ili_fields |= (flags | iip->ili_last_fields | iversion_flags);
-	spin_unlock(&iip->ili_lock);
+	iip->ili_dirty_flags |= flags;
 }
 
 int

fs/xfs/scrub/bmap.c

@@ -769,14 +769,14 @@ xchk_are_bmaps_contiguous(
  * mapping or false if there are no more mappings.  Caller must ensure that
  * @info.icur is zeroed before the first call.
  */
-static int
+static bool
 xchk_bmap_iext_iter(
 	struct xchk_bmap_info	*info,
 	struct xfs_bmbt_irec	*irec)
 {
 	struct xfs_bmbt_irec	got;
 	struct xfs_ifork	*ifp;
-	xfs_filblks_t		prev_len;
+	unsigned int		nr = 0;
 
 	ifp = xfs_ifork_ptr(info->sc->ip, info->whichfork);
 
@@ -790,12 +790,12 @@ xchk_bmap_iext_iter(
 				irec->br_startoff);
 		return false;
 	}
+	nr++;
 
 	/*
 	 * Iterate subsequent iextent records and merge them with the one
 	 * that we just read, if possible.
 	 */
-	prev_len = irec->br_blockcount;
 	while (xfs_iext_peek_next_extent(ifp, &info->icur, &got)) {
 		if (!xchk_are_bmaps_contiguous(irec, &got))
 			break;
@@ -805,20 +805,21 @@ xchk_bmap_iext_iter(
 					got.br_startoff);
 			return false;
 		}
-
-		/*
-		 * Notify the user of mergeable records in the data or attr
-		 * forks.  CoW forks only exist in memory so we ignore them.
-		 */
-		if (info->whichfork != XFS_COW_FORK &&
-		    prev_len + got.br_blockcount > BMBT_BLOCKCOUNT_MASK)
-			xchk_ino_set_preen(info->sc, info->sc->ip->i_ino);
+		nr++;
 
 		irec->br_blockcount += got.br_blockcount;
-		prev_len = got.br_blockcount;
 		xfs_iext_next(ifp, &info->icur);
 	}
 
+	/*
+	 * If the merged mapping could be expressed with fewer bmbt records
+	 * than we actually found, notify the user that this fork could be
+	 * optimized.  CoW forks only exist in memory so we ignore them.
+	 */
+	if (nr > 1 && info->whichfork != XFS_COW_FORK &&
+	    howmany_64(irec->br_blockcount, XFS_MAX_BMBT_EXTLEN) < nr)
+		xchk_ino_set_preen(info->sc, info->sc->ip->i_ino);
+
 	return true;
 }
 

fs/xfs/scrub/scrub.h

@@ -105,10 +105,10 @@ struct xfs_scrub {
 };
 
 /* XCHK state flags grow up from zero, XREP state flags grown down from 2^31 */
-#define XCHK_TRY_HARDER		(1 << 0)  /* can't get resources, try again */
-#define XCHK_FSGATES_DRAIN	(1 << 2)  /* defer ops draining enabled */
-#define XCHK_NEED_DRAIN		(1 << 3)  /* scrub needs to drain defer ops */
-#define XREP_ALREADY_FIXED	(1 << 31) /* checking our repair work */
+#define XCHK_TRY_HARDER		(1U << 0)  /* can't get resources, try again */
+#define XCHK_FSGATES_DRAIN	(1U << 2)  /* defer ops draining enabled */
+#define XCHK_NEED_DRAIN		(1U << 3)  /* scrub needs to drain defer ops */
+#define XREP_ALREADY_FIXED	(1U << 31) /* checking our repair work */
 
 /*
  * The XCHK_FSGATES* flags reflect functionality in the main filesystem that

fs/xfs/xfs_buf_item.c

@@ -452,10 +452,18 @@ xfs_buf_item_format(
  * This is called to pin the buffer associated with the buf log item in memory
  * so it cannot be written out.
  *
- * We also always take a reference to the buffer log item here so that the bli
- * is held while the item is pinned in memory. This means that we can
- * unconditionally drop the reference count a transaction holds when the
- * transaction is completed.
+ * We take a reference to the buffer log item here so that the BLI life cycle
+ * extends at least until the buffer is unpinned via xfs_buf_item_unpin() and
+ * inserted into the AIL.
+ *
+ * We also need to take a reference to the buffer itself as the BLI unpin
+ * processing requires accessing the buffer after the BLI has dropped the final
+ * BLI reference. See xfs_buf_item_unpin() for an explanation.
+ * If unpins race to drop the final BLI reference and only the
+ * BLI owns a reference to the buffer, then the loser of the race can have the
+ * buffer fgreed from under it (e.g. on shutdown). Taking a buffer reference per
+ * pin count ensures the life cycle of the buffer extends for as
+ * long as we hold the buffer pin reference in xfs_buf_item_unpin().
  */
 STATIC void
 xfs_buf_item_pin(
@@ -470,13 +478,30 @@ xfs_buf_item_pin(
 
 	trace_xfs_buf_item_pin(bip);
 
+	xfs_buf_hold(bip->bli_buf);
 	atomic_inc(&bip->bli_refcount);
 	atomic_inc(&bip->bli_buf->b_pin_count);
 }
 
 /*
- * This is called to unpin the buffer associated with the buf log item which
- * was previously pinned with a call to xfs_buf_item_pin().
+ * This is called to unpin the buffer associated with the buf log item which was
+ * previously pinned with a call to xfs_buf_item_pin().  We enter this function
+ * with a buffer pin count, a buffer reference and a BLI reference.
+ *
+ * We must drop the BLI reference before we unpin the buffer because the AIL
+ * doesn't acquire a BLI reference whenever it accesses it. Therefore if the
+ * refcount drops to zero, the bli could still be AIL resident and the buffer
+ * submitted for I/O at any point before we return. This can result in IO
+ * completion freeing the buffer while we are still trying to access it here.
+ * This race condition can also occur in shutdown situations where we abort and
+ * unpin buffers from contexts other that journal IO completion.
+ *
+ * Hence we have to hold a buffer reference per pin count to ensure that the
+ * buffer cannot be freed until we have finished processing the unpin operation.
+ * The reference is taken in xfs_buf_item_pin(), and we must hold it until we
+ * are done processing the buffer state. In the case of an abort (remove =
+ * true) then we re-use the current pin reference as the IO reference we hand
+ * off to IO failure handling.
  */
 STATIC void
 xfs_buf_item_unpin(
@@ -493,24 +518,18 @@ xfs_buf_item_unpin(
 
 	trace_xfs_buf_item_unpin(bip);
 
-	/*
-	 * Drop the bli ref associated with the pin and grab the hold required
-	 * for the I/O simulation failure in the abort case. We have to do this
-	 * before the pin count drops because the AIL doesn't acquire a bli
-	 * reference. Therefore if the refcount drops to zero, the bli could
-	 * still be AIL resident and the buffer submitted for I/O (and freed on
-	 * completion) at any point before we return. This can be removed once
-	 * the AIL properly holds a reference on the bli.
-	 */
 	freed = atomic_dec_and_test(&bip->bli_refcount);
-	if (freed && !stale && remove)
-		xfs_buf_hold(bp);
 	if (atomic_dec_and_test(&bp->b_pin_count))
 		wake_up_all(&bp->b_waiters);
 
-	 /* nothing to do but drop the pin count if the bli is active */
-	if (!freed)
+	/*
+	 * Nothing to do but drop the buffer pin reference if the BLI is
+	 * still active.
+	 */
+	if (!freed) {
+		xfs_buf_rele(bp);
 		return;
+	}
 
 	if (stale) {
 		ASSERT(bip->bli_flags & XFS_BLI_STALE);
@@ -522,6 +541,15 @@ xfs_buf_item_unpin(
 
 		trace_xfs_buf_item_unpin_stale(bip);
 
+		/*
+		 * The buffer has been locked and referenced since it was marked
+		 * stale so we own both lock and reference exclusively here. We
+		 * do not need the pin reference any more, so drop it now so
+		 * that we only have one reference to drop once item completion
+		 * processing is complete.
+		 */
+		xfs_buf_rele(bp);
+
 		/*
 		 * If we get called here because of an IO error, we may or may
 		 * not have the item on the AIL. xfs_trans_ail_delete() will
@@ -538,16 +566,30 @@ xfs_buf_item_unpin(
 			ASSERT(bp->b_log_item == NULL);
 		}
 		xfs_buf_relse(bp);
-	} else if (remove) {
+		return;
+	}
+
+	if (remove) {
 		/*
-		 * The buffer must be locked and held by the caller to simulate
-		 * an async I/O failure. We acquired the hold for this case
-		 * before the buffer was unpinned.
+		 * We need to simulate an async IO failures here to ensure that
+		 * the correct error completion is run on this buffer. This
+		 * requires a reference to the buffer and for the buffer to be
+		 * locked. We can safely pass ownership of the pin reference to
+		 * the IO to ensure that nothing can free the buffer while we
+		 * wait for the lock and then run the IO failure completion.
 		 */
 		xfs_buf_lock(bp);
 		bp->b_flags |= XBF_ASYNC;
 		xfs_buf_ioend_fail(bp);
+		return;
 	}
+
+	/*
+	 * BLI has no more active references - it will be moved to the AIL to
+	 * manage the remaining BLI/buffer life cycle. There is nothing left for
+	 * us to do here so drop the pin reference to the buffer.
+	 */
+	xfs_buf_rele(bp);
 }
 
 STATIC uint

fs/xfs/xfs_filestream.c

@@ -78,7 +78,6 @@ xfs_filestream_pick_ag(
 		*longest = 0;
 		err = xfs_bmap_longest_free_extent(pag, NULL, longest);
 		if (err) {
-			xfs_perag_rele(pag);
 			if (err != -EAGAIN)
 				break;
 			/* Couldn't lock the AGF, skip this AG. */

fs/xfs/xfs_icache.c

@@ -454,6 +454,27 @@ xfs_inodegc_queue_all(
 	return ret;
 }
 
+/* Wait for all queued work and collect errors */
+static int
+xfs_inodegc_wait_all(
+	struct xfs_mount	*mp)
+{
+	int			cpu;
+	int			error = 0;
+
+	flush_workqueue(mp->m_inodegc_wq);
+	for_each_online_cpu(cpu) {
+		struct xfs_inodegc	*gc;
+
+		gc = per_cpu_ptr(mp->m_inodegc, cpu);
+		if (gc->error && !error)
+			error = gc->error;
+		gc->error = 0;
+	}
+
+	return error;
+}
+
 /*
  * Check the validity of the inode we just found it the cache
  */
@@ -1491,15 +1512,14 @@ xfs_blockgc_free_space(
 	if (error)
 		return error;
 
-	xfs_inodegc_flush(mp);
-	return 0;
+	return xfs_inodegc_flush(mp);
 }
 
 /*
  * Reclaim all the free space that we can by scheduling the background blockgc
  * and inodegc workers immediately and waiting for them all to clear.
  */
-void
+int
 xfs_blockgc_flush_all(
 	struct xfs_mount	*mp)
 {
@@ -1520,7 +1540,7 @@ xfs_blockgc_flush_all(
 	for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
 		flush_delayed_work(&pag->pag_blockgc_work);
 
-	xfs_inodegc_flush(mp);
+	return xfs_inodegc_flush(mp);
 }
 
 /*
@@ -1842,13 +1862,17 @@ xfs_inodegc_set_reclaimable(
  * This is the last chance to make changes to an otherwise unreferenced file
  * before incore reclamation happens.
  */
-static void
+static int
 xfs_inodegc_inactivate(
 	struct xfs_inode	*ip)
 {
+	int			error;
+
 	trace_xfs_inode_inactivating(ip);
-	xfs_inactive(ip);
+	error = xfs_inactive(ip);
 	xfs_inodegc_set_reclaimable(ip);
+	return error;
+
 }
 
 void
@@ -1880,8 +1904,12 @@ xfs_inodegc_worker(
 
 	WRITE_ONCE(gc->shrinker_hits, 0);
 	llist_for_each_entry_safe(ip, n, node, i_gclist) {
+		int	error;
+
 		xfs_iflags_set(ip, XFS_INACTIVATING);
-		xfs_inodegc_inactivate(ip);
+		error = xfs_inodegc_inactivate(ip);
+		if (error && !gc->error)
+			gc->error = error;
 	}
 
 	memalloc_nofs_restore(nofs_flag);
@@ -1905,13 +1933,13 @@ xfs_inodegc_push(
  * Force all currently queued inode inactivation work to run immediately and
  * wait for the work to finish.
  */
-void
+int
 xfs_inodegc_flush(
 	struct xfs_mount	*mp)
 {
 	xfs_inodegc_push(mp);
 	trace_xfs_inodegc_flush(mp, __return_address);
-	flush_workqueue(mp->m_inodegc_wq);
+	return xfs_inodegc_wait_all(mp);
 }
 
 /*

fs/xfs/xfs_icache.h

@@ -62,7 +62,7 @@ int xfs_blockgc_free_dquots(struct xfs_mount *mp, struct xfs_dquot *udqp,
 		unsigned int iwalk_flags);
 int xfs_blockgc_free_quota(struct xfs_inode *ip, unsigned int iwalk_flags);
 int xfs_blockgc_free_space(struct xfs_mount *mp, struct xfs_icwalk *icm);
-void xfs_blockgc_flush_all(struct xfs_mount *mp);
+int xfs_blockgc_flush_all(struct xfs_mount *mp);
 
 void xfs_inode_set_eofblocks_tag(struct xfs_inode *ip);
 void xfs_inode_clear_eofblocks_tag(struct xfs_inode *ip);
@@ -80,7 +80,7 @@ void xfs_blockgc_start(struct xfs_mount *mp);
 
 void xfs_inodegc_worker(struct work_struct *work);
 void xfs_inodegc_push(struct xfs_mount *mp);
-void xfs_inodegc_flush(struct xfs_mount *mp);
+int xfs_inodegc_flush(struct xfs_mount *mp);
 void xfs_inodegc_stop(struct xfs_mount *mp);
 void xfs_inodegc_start(struct xfs_mount *mp);
 void xfs_inodegc_cpu_dead(struct xfs_mount *mp, unsigned int cpu);

fs/xfs/xfs_inode.c

@@ -1620,16 +1620,7 @@ xfs_inactive_ifree(
 	 */
 	xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_ICOUNT, -1);
 
-	/*
-	 * Just ignore errors at this point.  There is nothing we can do except
-	 * to try to keep going. Make sure it's not a silent error.
-	 */
-	error = xfs_trans_commit(tp);
-	if (error)
-		xfs_notice(mp, "%s: xfs_trans_commit returned error %d",
-			__func__, error);
-
-	return 0;
+	return xfs_trans_commit(tp);
 }
 
 /*
@@ -1693,12 +1684,12 @@ xfs_inode_needs_inactive(
  * now be truncated.  Also, we clear all of the read-ahead state
  * kept for the inode here since the file is now closed.
  */
-void
+int
 xfs_inactive(
 	xfs_inode_t	*ip)
 {
 	struct xfs_mount	*mp;
-	int			error;
+	int			error = 0;
 	int			truncate = 0;
 
 	/*
@@ -1736,7 +1727,7 @@ xfs_inactive(
 		 * reference to the inode at this point anyways.
 		 */
 		if (xfs_can_free_eofblocks(ip, true))
-			xfs_free_eofblocks(ip);
+			error = xfs_free_eofblocks(ip);
 
 		goto out;
 	}
@@ -1773,7 +1764,7 @@ xfs_inactive(
 	/*
 	 * Free the inode.
 	 */
-	xfs_inactive_ifree(ip);
+	error = xfs_inactive_ifree(ip);
 
 out:
 	/*
@@ -1781,6 +1772,7 @@ xfs_inactive(
 	 * the attached dquots.
 	 */
 	xfs_qm_dqdetach(ip);
+	return error;
 }
 
 /*

fs/xfs/xfs_inode.h

@@ -470,7 +470,7 @@ enum layout_break_reason {
 	(xfs_has_grpid((pip)->i_mount) || (VFS_I(pip)->i_mode & S_ISGID))
 
 int		xfs_release(struct xfs_inode *ip);
-void		xfs_inactive(struct xfs_inode *ip);
+int		xfs_inactive(struct xfs_inode *ip);
 int		xfs_lookup(struct xfs_inode *dp, const struct xfs_name *name,
 			   struct xfs_inode **ipp, struct xfs_name *ci_name);
 int		xfs_create(struct mnt_idmap *idmap,

fs/xfs/xfs_inode_item.c

@@ -29,6 +29,153 @@ static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)
 	return container_of(lip, struct xfs_inode_log_item, ili_item);
 }
 
+static uint64_t
+xfs_inode_item_sort(
+	struct xfs_log_item	*lip)
+{
+	return INODE_ITEM(lip)->ili_inode->i_ino;
+}
+
+/*
+ * Prior to finally logging the inode, we have to ensure that all the
+ * per-modification inode state changes are applied. This includes VFS inode
+ * state updates, format conversions, verifier state synchronisation and
+ * ensuring the inode buffer remains in memory whilst the inode is dirty.
+ *
+ * We have to be careful when we grab the inode cluster buffer due to lock
+ * ordering constraints. The unlinked inode modifications (xfs_iunlink_item)
+ * require AGI -> inode cluster buffer lock order. The inode cluster buffer is
+ * not locked until ->precommit, so it happens after everything else has been
+ * modified.
+ *
+ * Further, we have AGI -> AGF lock ordering, and with O_TMPFILE handling we
+ * have AGI -> AGF -> iunlink item -> inode cluster buffer lock order. Hence we
+ * cannot safely lock the inode cluster buffer in xfs_trans_log_inode() because
+ * it can be called on a inode (e.g. via bumplink/droplink) before we take the
+ * AGF lock modifying directory blocks.
+ *
+ * Rather than force a complete rework of all the transactions to call
+ * xfs_trans_log_inode() once and once only at the end of every transaction, we
+ * move the pinning of the inode cluster buffer to a ->precommit operation. This
+ * matches how the xfs_iunlink_item locks the inode cluster buffer, and it
+ * ensures that the inode cluster buffer locking is always done last in a
+ * transaction. i.e. we ensure the lock order is always AGI -> AGF -> inode
+ * cluster buffer.
+ *
+ * If we return the inode number as the precommit sort key then we'll also
+ * guarantee that the order all inode cluster buffer locking is the same all the
+ * inodes and unlink items in the transaction.
+ */
+static int
+xfs_inode_item_precommit(
+	struct xfs_trans	*tp,
+	struct xfs_log_item	*lip)
+{
+	struct xfs_inode_log_item *iip = INODE_ITEM(lip);
+	struct xfs_inode	*ip = iip->ili_inode;
+	struct inode		*inode = VFS_I(ip);
+	unsigned int		flags = iip->ili_dirty_flags;
+
+	/*
+	 * Don't bother with i_lock for the I_DIRTY_TIME check here, as races
+	 * don't matter - we either will need an extra transaction in 24 hours
+	 * to log the timestamps, or will clear already cleared fields in the
+	 * worst case.
+	 */
+	if (inode->i_state & I_DIRTY_TIME) {
+		spin_lock(&inode->i_lock);
+		inode->i_state &= ~I_DIRTY_TIME;
+		spin_unlock(&inode->i_lock);
+	}
+
+	/*
+	 * If we're updating the inode core or the timestamps and it's possible
+	 * to upgrade this inode to bigtime format, do so now.
+	 */
+	if ((flags & (XFS_ILOG_CORE | XFS_ILOG_TIMESTAMP)) &&
+	    xfs_has_bigtime(ip->i_mount) &&
+	    !xfs_inode_has_bigtime(ip)) {
+		ip->i_diflags2 |= XFS_DIFLAG2_BIGTIME;
+		flags |= XFS_ILOG_CORE;
+	}
+
+	/*
+	 * Inode verifiers do not check that the extent size hint is an integer
+	 * multiple of the rt extent size on a directory with both rtinherit
+	 * and extszinherit flags set.  If we're logging a directory that is
+	 * misconfigured in this way, clear the hint.
+	 */
+	if ((ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&
+	    (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) &&
+	    (ip->i_extsize % ip->i_mount->m_sb.sb_rextsize) > 0) {
+		ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |
+				   XFS_DIFLAG_EXTSZINHERIT);
+		ip->i_extsize = 0;
+		flags |= XFS_ILOG_CORE;
+	}
+
+	/*
+	 * Record the specific change for fdatasync optimisation. This allows
+	 * fdatasync to skip log forces for inodes that are only timestamp
+	 * dirty. Once we've processed the XFS_ILOG_IVERSION flag, convert it
+	 * to XFS_ILOG_CORE so that the actual on-disk dirty tracking
+	 * (ili_fields) correctly tracks that the version has changed.
+	 */
+	spin_lock(&iip->ili_lock);
+	iip->ili_fsync_fields |= (flags & ~XFS_ILOG_IVERSION);
+	if (flags & XFS_ILOG_IVERSION)
+		flags = ((flags & ~XFS_ILOG_IVERSION) | XFS_ILOG_CORE);
+
+	if (!iip->ili_item.li_buf) {
+		struct xfs_buf	*bp;
+		int		error;
+
+		/*
+		 * We hold the ILOCK here, so this inode is not going to be
+		 * flushed while we are here. Further, because there is no
+		 * buffer attached to the item, we know that there is no IO in
+		 * progress, so nothing will clear the ili_fields while we read
+		 * in the buffer. Hence we can safely drop the spin lock and
+		 * read the buffer knowing that the state will not change from
+		 * here.
+		 */
+		spin_unlock(&iip->ili_lock);
+		error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &bp);
+		if (error)
+			return error;
+
+		/*
+		 * We need an explicit buffer reference for the log item but
+		 * don't want the buffer to remain attached to the transaction.
+		 * Hold the buffer but release the transaction reference once
+		 * we've attached the inode log item to the buffer log item
+		 * list.
+		 */
+		xfs_buf_hold(bp);
+		spin_lock(&iip->ili_lock);
+		iip->ili_item.li_buf = bp;
+		bp->b_flags |= _XBF_INODES;
+		list_add_tail(&iip->ili_item.li_bio_list, &bp->b_li_list);
+		xfs_trans_brelse(tp, bp);
+	}
+
+	/*
+	 * Always OR in the bits from the ili_last_fields field.  This is to
+	 * coordinate with the xfs_iflush() and xfs_buf_inode_iodone() routines
+	 * in the eventual clearing of the ili_fields bits.  See the big comment
+	 * in xfs_iflush() for an explanation of this coordination mechanism.
+	 */
+	iip->ili_fields |= (flags | iip->ili_last_fields);
+	spin_unlock(&iip->ili_lock);
+
+	/*
+	 * We are done with the log item transaction dirty state, so clear it so
+	 * that it doesn't pollute future transactions.
+	 */
+	iip->ili_dirty_flags = 0;
+	return 0;
+}
+
 /*
  * The logged size of an inode fork is always the current size of the inode
  * fork. This means that when an inode fork is relogged, the size of the logged
@@ -662,6 +809,8 @@ xfs_inode_item_committing(
 }
 
 static const struct xfs_item_ops xfs_inode_item_ops = {
+	.iop_sort	= xfs_inode_item_sort,
+	.iop_precommit	= xfs_inode_item_precommit,
 	.iop_size	= xfs_inode_item_size,
 	.iop_format	= xfs_inode_item_format,
 	.iop_pin	= xfs_inode_item_pin,

fs/xfs/xfs_inode_item.h

@@ -17,6 +17,7 @@ struct xfs_inode_log_item {
 	struct xfs_log_item	ili_item;	   /* common portion */
 	struct xfs_inode	*ili_inode;	   /* inode ptr */
 	unsigned short		ili_lock_flags;	   /* inode lock flags */
+	unsigned int		ili_dirty_flags;   /* dirty in current tx */
 	/*
 	 * The ili_lock protects the interactions between the dirty state and
 	 * the flush state of the inode log item. This allows us to do atomic

fs/xfs/xfs_log_recover.c

@@ -2711,7 +2711,9 @@ xlog_recover_iunlink_bucket(
 			 * just to flush the inodegc queue and wait for it to
 			 * complete.
 			 */
-			xfs_inodegc_flush(mp);
+			error = xfs_inodegc_flush(mp);
+			if (error)
+				break;
 		}
 
 		prev_agino = agino;
@@ -2719,10 +2721,15 @@ xlog_recover_iunlink_bucket(
 	}
 
 	if (prev_ip) {
+		int	error2;
+
 		ip->i_prev_unlinked = prev_agino;
 		xfs_irele(prev_ip);
+
+		error2 = xfs_inodegc_flush(mp);
+		if (error2 && !error)
+			return error2;
 	}
-	xfs_inodegc_flush(mp);
 	return error;
 }
 
@@ -2789,7 +2796,6 @@ xlog_recover_iunlink_ag(
 			 * bucket and remaining inodes on it unreferenced and
 			 * unfreeable.
 			 */
-			xfs_inodegc_flush(pag->pag_mount);
 			xlog_recover_clear_agi_bucket(pag, bucket);
 		}
 	}
@@ -2806,13 +2812,6 @@ xlog_recover_process_iunlinks(
 
 	for_each_perag(log->l_mp, agno, pag)
 		xlog_recover_iunlink_ag(pag);
-
-	/*
-	 * Flush the pending unlinked inodes to ensure that the inactivations
-	 * are fully completed on disk and the incore inodes can be reclaimed
-	 * before we signal that recovery is complete.
-	 */
-	xfs_inodegc_flush(log->l_mp);
 }
 
 STATIC void

fs/xfs/xfs_mount.h

@@ -62,6 +62,7 @@ struct xfs_error_cfg {
 struct xfs_inodegc {
 	struct llist_head	list;
 	struct delayed_work	work;
+	int			error;
 
 	/* approximate count of inodes in the list */
 	unsigned int		items;

fs/xfs/xfs_reflink.c

@@ -616,8 +616,10 @@ xfs_reflink_cancel_cow_blocks(
 			xfs_refcount_free_cow_extent(*tpp, del.br_startblock,
 					del.br_blockcount);
 
-			xfs_free_extent_later(*tpp, del.br_startblock,
+			error = xfs_free_extent_later(*tpp, del.br_startblock,
 					  del.br_blockcount, NULL);
+			if (error)
+				break;
 
 			/* Roll the transaction */
 			error = xfs_defer_finish(tpp);

fs/xfs/xfs_super.c

@@ -1100,6 +1100,7 @@ xfs_inodegc_init_percpu(
 #endif
 		init_llist_head(&gc->list);
 		gc->items = 0;
+		gc->error = 0;
 		INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
 	}
 	return 0;

fs/xfs/xfs_trans.c

@@ -290,7 +290,9 @@ xfs_trans_alloc(
 		 * Do not perform a synchronous scan because callers can hold
 		 * other locks.
 		 */
-		xfs_blockgc_flush_all(mp);
+		error = xfs_blockgc_flush_all(mp);
+		if (error)
+			return error;
 		want_retry = false;
 		goto retry;
 	}
@@ -970,6 +972,11 @@ __xfs_trans_commit(
 		error = xfs_defer_finish_noroll(&tp);
 		if (error)
 			goto out_unreserve;
+
+		/* Run precommits from final tx in defer chain. */
+		error = xfs_trans_run_precommits(tp);
+		if (error)
+			goto out_unreserve;
 	}
 
 	/*