xfs: Avoid races with cnt_btree lastrec updates

A concurrent file creation and little writing could unexpectedly return
-ENOSPC error since there is a race window that the allocator could get
the wrong agf->agf_longest.

Write file process steps:
1) Find the entry that best meets the conditions, then calculate the start
   address and length of the remaining part of the entry after allocation.
2) Delete this entry and update the -current- agf->agf_longest.
3) Insert the remaining unused parts of this entry based on the
   calculations in 1), and update the agf->agf_longest again if necessary.

Create file process steps:
1) Check whether there are free inodes in the inode chunk.
2) If there is no free inode, check whether there has space for creating
   inode chunks, perform the no-lock judgment first.
3) If the judgment succeeds, the judgment is performed again with agf lock
   held. Otherwire, an error is returned directly.

If the write process is in step 2) but not go to 3) yet, the create file
process goes to 2) at this time, it may be mistaken for no space,
resulting in the file system still has space but the file creation fails.

We have sent two different commits to the community in order to fix this
problem[1][2]. Unfortunately, both solutions have flaws. In [2], I
discussed with Dave and Darrick, realized that a better solution to this
problem requires the "last cnt record tracking" to be ripped out of the
generic btree code. And surprisingly, Dave directly provided his fix code.
This patch includes appropriate modifications based on his tmp-code to
address this issue.

The entire fix can be roughly divided into two parts:
1) Delete the code related to lastrec-update in the generic btree code.
2) Place the process of updating longest freespace with cntbt separately
   to the end of the cntbt modifications. Move the cursor to the rightmost
   firstly, and update the longest free extent based on the record.

Note that we can not update the longest with xfs_alloc_get_rec() after
find the longest record, as xfs_verify_agbno() may not pass because
pag->block_count is updated on the outside. Therefore, use
xfs_btree_get_rec() as a replacement.

[1] https://lore.kernel.org/all/20240419061848.1032366-2-yebin10@huawei.com
[2] https://lore.kernel.org/all/20240604071121.3981686-1-wozizhi@huawei.com

Reported by: Ye Bin <yebin10@huawei.com>

Signed-off-by: Zizhi Wo <wozizhi@huawei.com>
Reviewed-by: Darrick J. Wong <djwong@kernel.org>
Signed-off-by: Chandan Babu R <chandanbabu@kernel.org>

1 files changed, 114 insertions, 0 deletions

diff --git a/fs/xfs/libxfs/xfs_alloc.c b/fs/xfs/libxfs/xfs_alloc.c
index ef4f5972da5d..59326f84f6a5 100644
--- a/fs/xfs/libxfs/xfs_alloc.c
+++ b/fs/xfs/libxfs/xfs_alloc.c
@@ -467,6 +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.
@@ -490,6 +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;
 
@@ -578,6 +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.
 	 */
@@ -655,6 +751,10 @@ xfs_alloc_fixup_trees(
 			return -EFSCORRUPTED;
 		}
 	}
+
+	if (fixup_longest)
+		return xfs_alloc_fixup_longest(cnt_cur);
+
 	return 0;
 }
 
@@ -1957,6 +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;
@@ -2220,8 +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;
@@ -2230,6 +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)) {
@@ -2237,6 +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;