xfs: use per-AG reservations for the finobt

Source kernel commit: 76d771b4cbe33c581bd6ca2710c120be51172440

Currently we try to rely on the global reserved block pool for block
allocations for the free inode btree, but I have customer reports
(fairly complex workload, need to find an easier reproducer) where that
is not enough as the AG where we free an inode that requires a new
finobt block is entirely full.  This causes us to cancel a dirty
transaction and thus a file system shutdown.

I think the right way to guard against this is to treat the finot the same
way as the refcount btree and have a per-AG reservations for the possible
worst case size of it, and the patch below implements that.

Note that this could increase mount times with large finobt trees.  In
an ideal world we would have added a field for the number of finobt
fields to the AGI, similar to what we did for the refcount blocks.
We should do add it next time we rev the AGI or AGF format by adding
new fields.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Eric Sandeen <sandeen@sandeen.net>

4 files changed, 132 insertions, 9 deletions

diff --git a/include/xfs_mount.h b/include/xfs_mount.h
index 6f302748f2..5ff24ebf3a 100644
--- a/include/xfs_mount.h
+++ b/include/xfs_mount.h
@@ -78,6 +78,7 @@ typedef struct xfs_mount {
 	uint			m_ag_max_usable; /* max space per AG */
 	struct radix_tree_root	m_perag_tree;
 	uint			m_flags;	/* global mount flags */
+	bool			m_inotbt_nores; /* no per-AG finobt resv. */
 	uint			m_qflags;	/* quota status flags */
 	uint			m_attroffset;	/* inode attribute offset */
 	int			m_ialloc_inos;	/* inodes in inode allocation */
diff --git a/libxfs/xfs_ag_resv.c b/libxfs/xfs_ag_resv.c
index 54888c244b..4b51851497 100644
--- a/libxfs/xfs_ag_resv.c
+++ b/libxfs/xfs_ag_resv.c
@@ -38,6 +38,7 @@
 #include "xfs_rmap_btree.h"
 #include "xfs_btree.h"
 #include "xfs_refcount_btree.h"
+#include "xfs_ialloc_btree.h"
 
 /*
  * Per-AG Block Reservations
@@ -209,6 +210,9 @@ __xfs_ag_resv_init(
 	if (error) {
 		trace_xfs_ag_resv_init_error(pag->pag_mount, pag->pag_agno,
 				error, _RET_IP_);
+		xfs_warn(mp,
+"Per-AG reservation for AG %u failed.  Filesystem may run out of space.",
+				pag->pag_agno);
 		return error;
 	}
 
@@ -227,6 +231,8 @@ int
 xfs_ag_resv_init(
 	struct xfs_perag		*pag)
 {
+	struct xfs_mount		*mp = pag->pag_mount;
+	xfs_agnumber_t			agno = pag->pag_agno;
 	xfs_extlen_t			ask;
 	xfs_extlen_t			used;
 	int				error = 0;
@@ -235,23 +241,45 @@ xfs_ag_resv_init(
 	if (pag->pag_meta_resv.ar_asked == 0) {
 		ask = used = 0;
 
-		error = xfs_refcountbt_calc_reserves(pag->pag_mount,
-				pag->pag_agno, &ask, &used);
+		error = xfs_refcountbt_calc_reserves(mp, agno, &ask, &used);
 		if (error)
 			goto out;
 
-		error = __xfs_ag_resv_init(pag, XFS_AG_RESV_METADATA,
-				ask, used);
+		error = xfs_finobt_calc_reserves(mp, agno, &ask, &used);
 		if (error)
 			goto out;
+
+		error = __xfs_ag_resv_init(pag, XFS_AG_RESV_METADATA,
+				ask, used);
+		if (error) {
+			/*
+			 * Because we didn't have per-AG reservations when the
+			 * finobt feature was added we might not be able to
+			 * reserve all needed blocks.  Warn and fall back to the
+			 * old and potentially buggy code in that case, but
+			 * ensure we do have the reservation for the refcountbt.
+			 */
+			ask = used = 0;
+
+			mp->m_inotbt_nores = true;
+
+			error = xfs_refcountbt_calc_reserves(mp, agno, &ask,
+					&used);
+			if (error)
+				goto out;
+
+			error = __xfs_ag_resv_init(pag, XFS_AG_RESV_METADATA,
+					ask, used);
+			if (error)
+				goto out;
+		}
 	}
 
 	/* Create the AGFL metadata reservation */
 	if (pag->pag_agfl_resv.ar_asked == 0) {
 		ask = used = 0;
 
-		error = xfs_rmapbt_calc_reserves(pag->pag_mount, pag->pag_agno,
-				&ask, &used);
+		error = xfs_rmapbt_calc_reserves(mp, agno, &ask, &used);
 		if (error)
 			goto out;
 
@@ -260,9 +288,16 @@ xfs_ag_resv_init(
 			goto out;
 	}
 
+#ifdef DEBUG
+	/* need to read in the AGF for the ASSERT below to work */
+	error = xfs_alloc_pagf_init(pag->pag_mount, NULL, pag->pag_agno, 0);
+	if (error)
+		return error;
+
 	ASSERT(xfs_perag_resv(pag, XFS_AG_RESV_METADATA)->ar_reserved +
 	       xfs_perag_resv(pag, XFS_AG_RESV_AGFL)->ar_reserved <=
 	       pag->pagf_freeblks + pag->pagf_flcount);
+#endif
 out:
 	return error;
 }
diff --git a/libxfs/xfs_ialloc_btree.c b/libxfs/xfs_ialloc_btree.c
index 251732626f..7b4be76630 100644
--- a/libxfs/xfs_ialloc_btree.c
+++ b/libxfs/xfs_ialloc_btree.c
@@ -81,11 +81,12 @@ xfs_finobt_set_root(
 }
 
 STATIC int
-xfs_inobt_alloc_block(
+__xfs_inobt_alloc_block(
 	struct xfs_btree_cur	*cur,
 	union xfs_btree_ptr	*start,
 	union xfs_btree_ptr	*new,
-	int			*stat)
+	int			*stat,
+	enum xfs_ag_resv_type	resv)
 {
 	xfs_alloc_arg_t		args;		/* block allocation args */
 	int			error;		/* error return value */
@@ -102,6 +103,7 @@ xfs_inobt_alloc_block(
 	args.maxlen = 1;
 	args.prod = 1;
 	args.type = XFS_ALLOCTYPE_NEAR_BNO;
+	args.resv = resv;
 
 	error = xfs_alloc_vextent(&args);
 	if (error) {
@@ -122,6 +124,27 @@ xfs_inobt_alloc_block(
 }
 
 STATIC int
+xfs_inobt_alloc_block(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*start,
+	union xfs_btree_ptr	*new,
+	int			*stat)
+{
+	return __xfs_inobt_alloc_block(cur, start, new, stat, XFS_AG_RESV_NONE);
+}
+
+STATIC int
+xfs_finobt_alloc_block(
+	struct xfs_btree_cur	*cur,
+	union xfs_btree_ptr	*start,
+	union xfs_btree_ptr	*new,
+	int			*stat)
+{
+	return __xfs_inobt_alloc_block(cur, start, new, stat,
+			XFS_AG_RESV_METADATA);
+}
+
+STATIC int
 xfs_inobt_free_block(
 	struct xfs_btree_cur	*cur,
 	struct xfs_buf		*bp)
@@ -327,7 +350,7 @@ static const struct xfs_btree_ops xfs_finobt_ops = {
 
 	.dup_cursor		= xfs_inobt_dup_cursor,
 	.set_root		= xfs_finobt_set_root,
-	.alloc_block		= xfs_inobt_alloc_block,
+	.alloc_block		= xfs_finobt_alloc_block,
 	.free_block		= xfs_inobt_free_block,
 	.get_minrecs		= xfs_inobt_get_minrecs,
 	.get_maxrecs		= xfs_inobt_get_maxrecs,
@@ -479,3 +502,64 @@ xfs_inobt_rec_check_count(
 	return 0;
 }
 #endif	/* DEBUG */
+
+static xfs_extlen_t
+xfs_inobt_max_size(
+	struct xfs_mount	*mp)
+{
+	/* Bail out if we're uninitialized, which can happen in mkfs. */
+	if (mp->m_inobt_mxr[0] == 0)
+		return 0;
+
+	return xfs_btree_calc_size(mp, mp->m_inobt_mnr,
+		(uint64_t)mp->m_sb.sb_agblocks * mp->m_sb.sb_inopblock /
+				XFS_INODES_PER_CHUNK);
+}
+
+static int
+xfs_inobt_count_blocks(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_btnum_t		btnum,
+	xfs_extlen_t		*tree_blocks)
+{
+	struct xfs_buf		*agbp;
+	struct xfs_btree_cur	*cur;
+	int			error;
+
+	error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
+	if (error)
+		return error;
+
+	cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno, btnum);
+	error = xfs_btree_count_blocks(cur, tree_blocks);
+	xfs_btree_del_cursor(cur, error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
+	xfs_buf_relse(agbp);
+
+	return error;
+}
+
+/*
+ * Figure out how many blocks to reserve and how many are used by this btree.
+ */
+int
+xfs_finobt_calc_reserves(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_extlen_t		*ask,
+	xfs_extlen_t		*used)
+{
+	xfs_extlen_t		tree_len = 0;
+	int			error;
+
+	if (!xfs_sb_version_hasfinobt(&mp->m_sb))
+		return 0;
+
+	error = xfs_inobt_count_blocks(mp, agno, XFS_BTNUM_FINO, &tree_len);
+	if (error)
+		return error;
+
+	*ask += xfs_inobt_max_size(mp);
+	*used += tree_len;
+	return 0;
+}
diff --git a/libxfs/xfs_ialloc_btree.h b/libxfs/xfs_ialloc_btree.h
index bd88453217..aa81e2e63f 100644
--- a/libxfs/xfs_ialloc_btree.h
+++ b/libxfs/xfs_ialloc_btree.h
@@ -72,4 +72,7 @@ int xfs_inobt_rec_check_count(struct xfs_mount *,
 #define xfs_inobt_rec_check_count(mp, rec)	0
 #endif	/* DEBUG */
 
+int xfs_finobt_calc_reserves(struct xfs_mount *mp, xfs_agnumber_t agno,
+		xfs_extlen_t *ask, xfs_extlen_t *used);
+
 #endif	/* __XFS_IALLOC_BTREE_H__ */