xfs: support repairing metadata btrees rooted in metadir inodes

Adapt the repair code so that we can stage a new btree in the data fork
area of a metadir inode and reap the old blocks.  We already have nearly
all of the infrastructure; the only parts that were missing were the
metadata inode reservation handling.

Signed-off-by: "Darrick J. Wong" <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>

fs/xfs/scrub/newbt.c

@@ -19,6 +19,8 @@
 #include "xfs_rmap.h"
 #include "xfs_ag.h"
 #include "xfs_defer.h"
+#include "xfs_metafile.h"
+#include "xfs_quota.h"
 #include "scrub/scrub.h"
 #include "scrub/common.h"
 #include "scrub/trace.h"
@@ -120,6 +122,43 @@ xrep_newbt_init_inode(
 	return 0;
 }
 
+/*
+ * Initialize accounting resources for staging a new metadata inode btree.
+ * If the metadata file has a space reservation, the caller must adjust that
+ * reservation when committing the new ondisk btree.
+ */
+int
+xrep_newbt_init_metadir_inode(
+	struct xrep_newbt		*xnr,
+	struct xfs_scrub		*sc)
+{
+	struct xfs_owner_info		oinfo;
+	struct xfs_ifork		*ifp;
+
+	ASSERT(xfs_is_metadir_inode(sc->ip));
+
+	xfs_rmap_ino_bmbt_owner(&oinfo, sc->ip->i_ino, XFS_DATA_FORK);
+
+	ifp = kmem_cache_zalloc(xfs_ifork_cache, XCHK_GFP_FLAGS);
+	if (!ifp)
+		return -ENOMEM;
+
+	/*
+	 * Allocate new metadir btree blocks with XFS_AG_RESV_NONE because the
+	 * inode metadata space reservations can only account allocated space
+	 * to the i_nblocks.  We do not want to change the inode core fields
+	 * until we're ready to commit the new tree, so we allocate the blocks
+	 * as if they were regular file blocks.  This exposes us to a higher
+	 * risk of the repair being cancelled due to ENOSPC.
+	 */
+	xrep_newbt_init_ag(xnr, sc, &oinfo,
+			XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
+			XFS_AG_RESV_NONE);
+	xnr->ifake.if_fork = ifp;
+	xnr->ifake.if_fork_size = xfs_inode_fork_size(sc->ip, XFS_DATA_FORK);
+	return 0;
+}
+
 /*
  * Initialize accounting resources for staging a new btree.  Callers are
  * expected to add their own reservations (and clean them up) manually.
@@ -224,6 +263,7 @@ xrep_newbt_alloc_ag_blocks(
 	int			error = 0;
 
 	ASSERT(sc->sa.pag != NULL);
+	ASSERT(xnr->resv != XFS_AG_RESV_METAFILE);
 
 	while (nr_blocks > 0) {
 		struct xfs_alloc_arg	args = {
@@ -297,6 +337,8 @@ xrep_newbt_alloc_file_blocks(
 	struct xfs_mount	*mp = sc->mp;
 	int			error = 0;
 
+	ASSERT(xnr->resv != XFS_AG_RESV_METAFILE);
+
 	while (nr_blocks > 0) {
 		struct xfs_alloc_arg	args = {
 			.tp		= sc->tp,

fs/xfs/scrub/newbt.h

@@ -63,6 +63,7 @@ void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct xfs_scrub *sc,
 		enum xfs_ag_resv_type resv);
 int xrep_newbt_init_inode(struct xrep_newbt *xnr, struct xfs_scrub *sc,
 		int whichfork, const struct xfs_owner_info *oinfo);
+int xrep_newbt_init_metadir_inode(struct xrep_newbt *xnr, struct xfs_scrub *sc);
 int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
 int xrep_newbt_add_extent(struct xrep_newbt *xnr, struct xfs_perag *pag,
 		xfs_agblock_t agbno, xfs_extlen_t len);

fs/xfs/scrub/reap.c

@@ -33,6 +33,7 @@
 #include "xfs_attr.h"
 #include "xfs_attr_remote.h"
 #include "xfs_defer.h"
+#include "xfs_metafile.h"
 #include "scrub/scrub.h"
 #include "scrub/common.h"
 #include "scrub/trace.h"
@@ -390,6 +391,8 @@ xreap_agextent_iter(
 	xfs_fsblock_t		fsbno;
 	int			error = 0;
 
+	ASSERT(rs->resv != XFS_AG_RESV_METAFILE);
+
 	fsbno = xfs_agbno_to_fsb(sc->sa.pag, agbno);
 
 	/*
@@ -675,6 +678,44 @@ xrep_reap_fsblocks(
 	return 0;
 }
 
+/*
+ * Dispose of every block of an old metadata btree that used to be rooted in a
+ * metadata directory file.
+ */
+int
+xrep_reap_metadir_fsblocks(
+	struct xfs_scrub		*sc,
+	struct xfsb_bitmap		*bitmap)
+{
+	/*
+	 * Reap old metadir btree blocks with XFS_AG_RESV_NONE because the old
+	 * blocks are no longer mapped by the inode, and inode metadata space
+	 * reservations can only account freed space to the i_nblocks.
+	 */
+	struct xfs_owner_info		oinfo;
+	struct xreap_state		rs = {
+		.sc			= sc,
+		.oinfo			= &oinfo,
+		.resv			= XFS_AG_RESV_NONE,
+	};
+	int				error;
+
+	ASSERT(xfs_has_rmapbt(sc->mp));
+	ASSERT(sc->ip != NULL);
+	ASSERT(xfs_is_metadir_inode(sc->ip));
+
+	xfs_rmap_ino_bmbt_owner(&oinfo, sc->ip->i_ino, XFS_DATA_FORK);
+
+	error = xfsb_bitmap_walk(bitmap, xreap_fsmeta_extent, &rs);
+	if (error)
+		return error;
+
+	if (xreap_dirty(&rs))
+		return xrep_defer_finish(sc);
+
+	return 0;
+}
+
 /*
  * Metadata files are not supposed to share blocks with anything else.
  * If blocks are shared, we remove the reverse mapping (thus reducing the

fs/xfs/scrub/reap.h

@@ -14,6 +14,8 @@ int xrep_reap_agblocks(struct xfs_scrub *sc, struct xagb_bitmap *bitmap,
 int xrep_reap_fsblocks(struct xfs_scrub *sc, struct xfsb_bitmap *bitmap,
 		const struct xfs_owner_info *oinfo);
 int xrep_reap_ifork(struct xfs_scrub *sc, struct xfs_inode *ip, int whichfork);
+int xrep_reap_metadir_fsblocks(struct xfs_scrub *sc,
+		struct xfsb_bitmap *bitmap);
 
 /* Buffer cache scan context. */
 struct xrep_bufscan {