Merge tag 'xfs-6.11-merge-3' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux

Pull xfs updates from Chandan Babu:
 "Major changes in this release are limited to enabling FITRIM on
  realtime devices and Byte-based grant head log reservation tracking.

  The remaining changes are limited to fixes and cleanups included in
  this pull request.

  Core:

   - Enable FITRIM on the realtime device

   - Introduce byte-based grant head log reservation tracking instead of
     physical log location tracking.

     This allows grant head to track a full 64 bit bytes space and hence
     overcome the limit of 4GB indexing that has been present until now

  Fixes:

   - xfs_flush_unmap_range() and xfs_prepare_shift() should consider RT
     extents in the flush unmap range

   - Implement bounds check when traversing log operations during log
     replay

   - Prevent out of bounds access when traversing a directory data block

   - Prevent incorrect ENOSPC when concurrently performing file creation
     and file writes

   - Fix rtalloc rotoring when delalloc is in use

  Cleanups:

   - Clean up I/O path inode locking helpers and the page fault handler

   - xfs: hoist inode operations to libxfs in anticipation of the
     metadata inode directory feature, which maintains a directory tree
     of metadata inodes. This will be necessary for further enhancements
     to the realtime feature, subvolume support

   - Clean up some warts in the extent freeing log intent code

   - Clean up the refcount and rmap intent code before adding support
     for realtime devices

   - Provide the correct email address for sysfs ABI documentation"

* tag 'xfs-6.11-merge-3' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux: (80 commits)
  xfs: fix rtalloc rotoring when delalloc is in use
  xfs: get rid of xfs_ag_resv_rmapbt_alloc
  xfs: skip flushing log items during push
  xfs: grant heads track byte counts, not LSNs
  xfs: pass the full grant head to accounting functions
  xfs: track log space pinned by the AIL
  xfs: collapse xlog_state_set_callback in caller
  xfs: l_last_sync_lsn is really AIL state
  xfs: ensure log tail is always up to date
  xfs: background AIL push should target physical space
  xfs: AIL doesn't need manual pushing
  xfs: move and rename xfs_trans_committed_bulk
  xfs: fix the contact address for the sysfs ABI documentation
  xfs: Avoid races with cnt_btree lastrec updates
  xfs: move xfs_refcount_update_defer_add to xfs_refcount_item.c
  xfs: simplify usage of the rcur local variable in xfs_refcount_finish_one
  xfs: don't bother calling xfs_refcount_finish_one_cleanup in xfs_refcount_finish_one
  xfs: reuse xfs_refcount_update_cancel_item
  xfs: add a ci_entry helper
  xfs: remove xfs_trans_set_refcount_flags
  ...

1 files changed, 139 insertions, 96 deletions

diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c
index 6c55a6e88eba..59326f84f6a5 100644
--- a/fs/xfs/libxfs/xfs_alloc.c
+++ b/fs/xfs/libxfs/xfs_alloc.c
@@ -27,6 +27,7 @@
 #include "xfs_ag_resv.h"
 #include "xfs_bmap.h"
 #include "xfs_health.h"
+#include "xfs_extfree_item.h"
 
 struct kmem_cache	*xfs_extfree_item_cache;
 
@@ -466,6 +467,97 @@ xfs_alloc_fix_len(
 }
 
 /*
+ * Determine if the cursor points to the block that contains the right-most
+ * block of records in the by-count btree. This block contains the largest
+ * contiguous free extent in the AG, so if we modify a record in this block we
+ * need to call xfs_alloc_fixup_longest() once the modifications are done to
+ * ensure the agf->agf_longest field is kept up to date with the longest free
+ * extent tracked by the by-count btree.
+ */
+static bool
+xfs_alloc_cursor_at_lastrec(
+	struct xfs_btree_cur	*cnt_cur)
+{
+	struct xfs_btree_block	*block;
+	union xfs_btree_ptr	ptr;
+	struct xfs_buf		*bp;
+
+	block = xfs_btree_get_block(cnt_cur, 0, &bp);
+
+	xfs_btree_get_sibling(cnt_cur, block, &ptr, XFS_BB_RIGHTSIB);
+	return xfs_btree_ptr_is_null(cnt_cur, &ptr);
+}
+
+/*
+ * Find the rightmost record of the cntbt, and return the longest free space
+ * recorded in it. Simply set both the block number and the length to their
+ * maximum values before searching.
+ */
+static int
+xfs_cntbt_longest(
+	struct xfs_btree_cur	*cnt_cur,
+	xfs_extlen_t		*longest)
+{
+	struct xfs_alloc_rec_incore irec;
+	union xfs_btree_rec	    *rec;
+	int			    stat = 0;
+	int			    error;
+
+	memset(&cnt_cur->bc_rec, 0xFF, sizeof(cnt_cur->bc_rec));
+	error = xfs_btree_lookup(cnt_cur, XFS_LOOKUP_LE, &stat);
+	if (error)
+		return error;
+	if (!stat) {
+		/* totally empty tree */
+		*longest = 0;
+		return 0;
+	}
+
+	error = xfs_btree_get_rec(cnt_cur, &rec, &stat);
+	if (error)
+		return error;
+	if (XFS_IS_CORRUPT(cnt_cur->bc_mp, !stat)) {
+		xfs_btree_mark_sick(cnt_cur);
+		return -EFSCORRUPTED;
+	}
+
+	xfs_alloc_btrec_to_irec(rec, &irec);
+	*longest = irec.ar_blockcount;
+	return 0;
+}
+
+/*
+ * Update the longest contiguous free extent in the AG from the by-count cursor
+ * that is passed to us. This should be done at the end of any allocation or
+ * freeing operation that touches the longest extent in the btree.
+ *
+ * Needing to update the longest extent can be determined by calling
+ * xfs_alloc_cursor_at_lastrec() after the cursor is positioned for record
+ * modification but before the modification begins.
+ */
+static int
+xfs_alloc_fixup_longest(
+	struct xfs_btree_cur	*cnt_cur)
+{
+	struct xfs_perag	*pag = cnt_cur->bc_ag.pag;
+	struct xfs_buf		*bp = cnt_cur->bc_ag.agbp;
+	struct xfs_agf		*agf = bp->b_addr;
+	xfs_extlen_t		longest = 0;
+	int			error;
+
+	/* Lookup last rec in order to update AGF. */
+	error = xfs_cntbt_longest(cnt_cur, &longest);
+	if (error)
+		return error;
+
+	pag->pagf_longest = longest;
+	agf->agf_longest = cpu_to_be32(pag->pagf_longest);
+	xfs_alloc_log_agf(cnt_cur->bc_tp, bp, XFS_AGF_LONGEST);
+
+	return 0;
+}
+
+/*
  * Update the two btrees, logically removing from freespace the extent
  * starting at rbno, rlen blocks.  The extent is contained within the
  * actual (current) free extent fbno for flen blocks.
@@ -489,6 +581,7 @@ xfs_alloc_fixup_trees(
 	xfs_extlen_t	nflen1=0;	/* first new free length */
 	xfs_extlen_t	nflen2=0;	/* second new free length */
 	struct xfs_mount *mp;
+	bool		fixup_longest = false;
 
 	mp = cnt_cur->bc_mp;
 
@@ -577,6 +670,10 @@ xfs_alloc_fixup_trees(
 		nfbno2 = rbno + rlen;
 		nflen2 = (fbno + flen) - nfbno2;
 	}
+
+	if (xfs_alloc_cursor_at_lastrec(cnt_cur))
+		fixup_longest = true;
+
 	/*
 	 * Delete the entry from the by-size btree.
 	 */
@@ -654,6 +751,10 @@ xfs_alloc_fixup_trees(
 			return -EFSCORRUPTED;
 		}
 	}
+
+	if (fixup_longest)
+		return xfs_alloc_fixup_longest(cnt_cur);
+
 	return 0;
 }
 
@@ -1932,7 +2033,7 @@ out_nominleft:
 /*
  * Free the extent starting at agno/bno for length.
  */
-STATIC int
+int
 xfs_free_ag_extent(
 	struct xfs_trans		*tp,
 	struct xfs_buf			*agbp,
@@ -1956,6 +2057,7 @@ xfs_free_ag_extent(
 	int				i;
 	int				error;
 	struct xfs_perag		*pag = agbp->b_pag;
+	bool				fixup_longest = false;
 
 	bno_cur = cnt_cur = NULL;
 	mp = tp->t_mountp;
@@ -2219,8 +2321,13 @@ xfs_free_ag_extent(
 	}
 	xfs_btree_del_cursor(bno_cur, XFS_BTREE_NOERROR);
 	bno_cur = NULL;
+
 	/*
 	 * In all cases we need to insert the new freespace in the by-size tree.
+	 *
+	 * If this new freespace is being inserted in the block that contains
+	 * the largest free space in the btree, make sure we also fix up the
+	 * agf->agf-longest tracker field.
 	 */
 	if ((error = xfs_alloc_lookup_eq(cnt_cur, nbno, nlen, &i)))
 		goto error0;
@@ -2229,6 +2336,8 @@ xfs_free_ag_extent(
 		error = -EFSCORRUPTED;
 		goto error0;
 	}
+	if (xfs_alloc_cursor_at_lastrec(cnt_cur))
+		fixup_longest = true;
 	if ((error = xfs_btree_insert(cnt_cur, &i)))
 		goto error0;
 	if (XFS_IS_CORRUPT(mp, i != 1)) {
@@ -2236,6 +2345,12 @@ xfs_free_ag_extent(
 		error = -EFSCORRUPTED;
 		goto error0;
 	}
+	if (fixup_longest) {
+		error = xfs_alloc_fixup_longest(cnt_cur);
+		if (error)
+			goto error0;
+	}
+
 	xfs_btree_del_cursor(cnt_cur, XFS_BTREE_NOERROR);
 	cnt_cur = NULL;
 
@@ -2422,32 +2537,6 @@ xfs_alloc_space_available(
 	return true;
 }
 
-int
-xfs_free_agfl_block(
-	struct xfs_trans	*tp,
-	xfs_agnumber_t		agno,
-	xfs_agblock_t		agbno,
-	struct xfs_buf		*agbp,
-	struct xfs_owner_info	*oinfo)
-{
-	int			error;
-	struct xfs_buf		*bp;
-
-	error = xfs_free_ag_extent(tp, agbp, agno, agbno, 1, oinfo,
-				   XFS_AG_RESV_AGFL);
-	if (error)
-		return error;
-
-	error = xfs_trans_get_buf(tp, tp->t_mountp->m_ddev_targp,
-			XFS_AGB_TO_DADDR(tp->t_mountp, agno, agbno),
-			tp->t_mountp->m_bsize, 0, &bp);
-	if (error)
-		return error;
-	xfs_trans_binval(tp, bp);
-
-	return 0;
-}
-
 /*
  * Check the agfl fields of the agf for inconsistency or corruption.
  *
@@ -2536,48 +2625,6 @@ xfs_agfl_reset(
 }
 
 /*
- * Defer an AGFL block free. This is effectively equivalent to
- * xfs_free_extent_later() with some special handling particular to AGFL blocks.
- *
- * Deferring AGFL frees helps prevent log reservation overruns due to too many
- * allocation operations in a transaction. AGFL frees are prone to this problem
- * because for one they are always freed one at a time. Further, an immediate
- * AGFL block free can cause a btree join and require another block free before
- * the real allocation can proceed. Deferring the free disconnects freeing up
- * the AGFL slot from freeing the block.
- */
-static int
-xfs_defer_agfl_block(
-	struct xfs_trans		*tp,
-	xfs_agnumber_t			agno,
-	xfs_agblock_t			agbno,
-	struct xfs_owner_info		*oinfo)
-{
-	struct xfs_mount		*mp = tp->t_mountp;
-	struct xfs_extent_free_item	*xefi;
-	xfs_fsblock_t			fsbno = XFS_AGB_TO_FSB(mp, agno, agbno);
-
-	ASSERT(xfs_extfree_item_cache != NULL);
-	ASSERT(oinfo != NULL);
-
-	if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbno(mp, fsbno)))
-		return -EFSCORRUPTED;
-
-	xefi = kmem_cache_zalloc(xfs_extfree_item_cache,
-			       GFP_KERNEL | __GFP_NOFAIL);
-	xefi->xefi_startblock = fsbno;
-	xefi->xefi_blockcount = 1;
-	xefi->xefi_owner = oinfo->oi_owner;
-	xefi->xefi_agresv = XFS_AG_RESV_AGFL;
-
-	trace_xfs_agfl_free_defer(mp, agno, 0, agbno, 1);
-
-	xfs_extent_free_get_group(mp, xefi);
-	xfs_defer_add(tp, &xefi->xefi_list, &xfs_agfl_free_defer_type);
-	return 0;
-}
-
-/*
  * Add the extent to the list of extents to be free at transaction end.
  * The list is maintained sorted (by block number).
  */
@@ -2588,28 +2635,15 @@ xfs_defer_extent_free(
 	xfs_filblks_t			len,
 	const struct xfs_owner_info	*oinfo,
 	enum xfs_ag_resv_type		type,
-	bool				skip_discard,
+	unsigned int			free_flags,
 	struct xfs_defer_pending	**dfpp)
 {
 	struct xfs_extent_free_item	*xefi;
 	struct xfs_mount		*mp = tp->t_mountp;
-#ifdef DEBUG
-	xfs_agnumber_t			agno;
-	xfs_agblock_t			agbno;
 
-	ASSERT(bno != NULLFSBLOCK);
-	ASSERT(len > 0);
 	ASSERT(len <= XFS_MAX_BMBT_EXTLEN);
 	ASSERT(!isnullstartblock(bno));
-	agno = XFS_FSB_TO_AGNO(mp, bno);
-	agbno = XFS_FSB_TO_AGBNO(mp, bno);
-	ASSERT(agno < mp->m_sb.sb_agcount);
-	ASSERT(agbno < mp->m_sb.sb_agblocks);
-	ASSERT(len < mp->m_sb.sb_agblocks);
-	ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
-#endif
-	ASSERT(xfs_extfree_item_cache != NULL);
-	ASSERT(type != XFS_AG_RESV_AGFL);
+	ASSERT(!(free_flags & ~XFS_FREE_EXTENT_ALL_FLAGS));
 
 	if (XFS_IS_CORRUPT(mp, !xfs_verify_fsbext(mp, bno, len)))
 		return -EFSCORRUPTED;
@@ -2619,7 +2653,7 @@ xfs_defer_extent_free(
 	xefi->xefi_startblock = bno;
 	xefi->xefi_blockcount = (xfs_extlen_t)len;
 	xefi->xefi_agresv = type;
-	if (skip_discard)
+	if (free_flags & XFS_FREE_EXTENT_SKIP_DISCARD)
 		xefi->xefi_flags |= XFS_EFI_SKIP_DISCARD;
 	if (oinfo) {
 		ASSERT(oinfo->oi_offset == 0);
@@ -2632,12 +2666,8 @@ xfs_defer_extent_free(
 	} else {
 		xefi->xefi_owner = XFS_RMAP_OWN_NULL;
 	}
-	trace_xfs_bmap_free_defer(mp,
-			XFS_FSB_TO_AGNO(tp->t_mountp, bno), 0,
-			XFS_FSB_TO_AGBNO(tp->t_mountp, bno), len);
 
-	xfs_extent_free_get_group(mp, xefi);
-	*dfpp = xfs_defer_add(tp, &xefi->xefi_list, &xfs_extent_free_defer_type);
+	xfs_extent_free_defer_add(tp, xefi, dfpp);
 	return 0;
 }
 
@@ -2648,11 +2678,11 @@ xfs_free_extent_later(
 	xfs_filblks_t			len,
 	const struct xfs_owner_info	*oinfo,
 	enum xfs_ag_resv_type		type,
-	bool				skip_discard)
+	unsigned int			free_flags)
 {
 	struct xfs_defer_pending	*dontcare = NULL;
 
-	return xfs_defer_extent_free(tp, bno, len, oinfo, type, skip_discard,
+	return xfs_defer_extent_free(tp, bno, len, oinfo, type, free_flags,
 			&dontcare);
 }
 
@@ -2677,13 +2707,13 @@ xfs_free_extent_later(
 int
 xfs_alloc_schedule_autoreap(
 	const struct xfs_alloc_arg	*args,
-	bool				skip_discard,
+	unsigned int			free_flags,
 	struct xfs_alloc_autoreap	*aarp)
 {
 	int				error;
 
 	error = xfs_defer_extent_free(args->tp, args->fsbno, args->len,
-			&args->oinfo, args->resv, skip_discard, &aarp->dfp);
+			&args->oinfo, args->resv, free_flags, &aarp->dfp);
 	if (error)
 		return error;
 
@@ -2895,8 +2925,21 @@ xfs_alloc_fix_freelist(
 		if (error)
 			goto out_agbp_relse;
 
-		/* defer agfl frees */
-		error = xfs_defer_agfl_block(tp, args->agno, bno, &targs.oinfo);
+		/*
+		 * Defer the AGFL block free.
+		 *
+		 * This helps to prevent log reservation overruns due to too
+		 * many allocation operations in a transaction. AGFL frees are
+		 * prone to this problem because for one they are always freed
+		 * one at a time.  Further, an immediate AGFL block free can
+		 * cause a btree join and require another block free before the
+		 * real allocation can proceed.
+		 * Deferring the free disconnects freeing up the AGFL slot from
+		 * freeing the block.
+		 */
+		error = xfs_free_extent_later(tp,
+				XFS_AGB_TO_FSB(mp, args->agno, bno), 1,
+				&targs.oinfo, XFS_AG_RESV_AGFL, 0);
 		if (error)
 			goto out_agbp_relse;
 	}