1 files changed, 131 insertions, 7 deletions

diff --git a/fs/xfs/scrub/scrub.c b/fs/xfs/scrub/scrub.c
index ad6eda4fa123..c5999c28c20c 100644
--- a/fs/xfs/scrub/scrub.c
+++ b/fs/xfs/scrub/scrub.c
@@ -44,11 +44,16 @@
 #include "xfs_rmap_btree.h"
 #include "xfs_quota.h"
 #include "xfs_qm.h"
+#include "xfs_errortag.h"
+#include "xfs_error.h"
+#include "xfs_log.h"
+#include "xfs_trans_priv.h"
 #include "scrub/xfs_scrub.h"
 #include "scrub/scrub.h"
 #include "scrub/common.h"
 #include "scrub/trace.h"
 #include "scrub/btree.h"
+#include "scrub/repair.h"
 
 /*
  * Online Scrub and Repair
@@ -122,6 +127,24 @@
  * XCORRUPT flag; btree query function errors are noted by setting the
  * XFAIL flag and deleting the cursor to prevent further attempts to
  * cross-reference with a defective btree.
+ *
+ * If a piece of metadata proves corrupt or suboptimal, the userspace
+ * program can ask the kernel to apply some tender loving care (TLC) to
+ * the metadata object by setting the REPAIR flag and re-calling the
+ * scrub ioctl.  "Corruption" is defined by metadata violating the
+ * on-disk specification; operations cannot continue if the violation is
+ * left untreated.  It is possible for XFS to continue if an object is
+ * "suboptimal", however performance may be degraded.  Repairs are
+ * usually performed by rebuilding the metadata entirely out of
+ * redundant metadata.  Optimizing, on the other hand, can sometimes be
+ * done without rebuilding entire structures.
+ *
+ * Generally speaking, the repair code has the following code structure:
+ * Lock -> scrub -> repair -> commit -> re-lock -> re-scrub -> unlock.
+ * The first check helps us figure out if we need to rebuild or simply
+ * optimize the structure so that the rebuild knows what to do.  The
+ * second check evaluates the completeness of the repair; that is what
+ * is reported to userspace.
  */
 
 /*
@@ -157,7 +180,10 @@ xfs_scrub_teardown(
 {
 	xfs_scrub_ag_free(sc, &sc->sa);
 	if (sc->tp) {
-		xfs_trans_cancel(sc->tp);
+		if (error == 0 && (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR))
+			error = xfs_trans_commit(sc->tp);
+		else
+			xfs_trans_cancel(sc->tp);
 		sc->tp = NULL;
 	}
 	if (sc->ip) {
@@ -184,126 +210,150 @@ static const struct xfs_scrub_meta_ops meta_scrub_ops[] = {
 		.type	= ST_NONE,
 		.setup	= xfs_scrub_setup_fs,
 		.scrub	= xfs_scrub_probe,
+		.repair = xfs_repair_probe,
 	},
 	[XFS_SCRUB_TYPE_SB] = {		/* superblock */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_fs,
 		.scrub	= xfs_scrub_superblock,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_AGF] = {	/* agf */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_fs,
 		.scrub	= xfs_scrub_agf,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_AGFL]= {	/* agfl */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_fs,
 		.scrub	= xfs_scrub_agfl,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_AGI] = {	/* agi */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_fs,
 		.scrub	= xfs_scrub_agi,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_BNOBT] = {	/* bnobt */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_ag_allocbt,
 		.scrub	= xfs_scrub_bnobt,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_CNTBT] = {	/* cntbt */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_ag_allocbt,
 		.scrub	= xfs_scrub_cntbt,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_INOBT] = {	/* inobt */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_ag_iallocbt,
 		.scrub	= xfs_scrub_inobt,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_FINOBT] = {	/* finobt */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_ag_iallocbt,
 		.scrub	= xfs_scrub_finobt,
 		.has	= xfs_sb_version_hasfinobt,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_RMAPBT] = {	/* rmapbt */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_ag_rmapbt,
 		.scrub	= xfs_scrub_rmapbt,
 		.has	= xfs_sb_version_hasrmapbt,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_REFCNTBT] = {	/* refcountbt */
 		.type	= ST_PERAG,
 		.setup	= xfs_scrub_setup_ag_refcountbt,
 		.scrub	= xfs_scrub_refcountbt,
 		.has	= xfs_sb_version_hasreflink,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_INODE] = {	/* inode record */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_inode,
 		.scrub	= xfs_scrub_inode,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_BMBTD] = {	/* inode data fork */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_inode_bmap,
 		.scrub	= xfs_scrub_bmap_data,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_BMBTA] = {	/* inode attr fork */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_inode_bmap,
 		.scrub	= xfs_scrub_bmap_attr,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_BMBTC] = {	/* inode CoW fork */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_inode_bmap,
 		.scrub	= xfs_scrub_bmap_cow,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_DIR] = {	/* directory */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_directory,
 		.scrub	= xfs_scrub_directory,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_XATTR] = {	/* extended attributes */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_xattr,
 		.scrub	= xfs_scrub_xattr,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_SYMLINK] = {	/* symbolic link */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_symlink,
 		.scrub	= xfs_scrub_symlink,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_PARENT] = {	/* parent pointers */
 		.type	= ST_INODE,
 		.setup	= xfs_scrub_setup_parent,
 		.scrub	= xfs_scrub_parent,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_RTBITMAP] = {	/* realtime bitmap */
 		.type	= ST_FS,
 		.setup	= xfs_scrub_setup_rt,
 		.scrub	= xfs_scrub_rtbitmap,
 		.has	= xfs_sb_version_hasrealtime,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_RTSUM] = {	/* realtime summary */
 		.type	= ST_FS,
 		.setup	= xfs_scrub_setup_rt,
 		.scrub	= xfs_scrub_rtsummary,
 		.has	= xfs_sb_version_hasrealtime,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_UQUOTA] = {	/* user quota */
 		.type	= ST_FS,
 		.setup	= xfs_scrub_setup_quota,
 		.scrub	= xfs_scrub_quota,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_GQUOTA] = {	/* group quota */
 		.type	= ST_FS,
 		.setup	= xfs_scrub_setup_quota,
 		.scrub	= xfs_scrub_quota,
+		.repair	= xfs_repair_notsupported,
 	},
 	[XFS_SCRUB_TYPE_PQUOTA] = {	/* project quota */
 		.type	= ST_FS,
 		.setup	= xfs_scrub_setup_quota,
 		.scrub	= xfs_scrub_quota,
+		.repair	= xfs_repair_notsupported,
 	},
 };
 
@@ -383,15 +433,54 @@ xfs_scrub_validate_inputs(
 	if (!xfs_sb_version_hasextflgbit(&mp->m_sb))
 		goto out;
 
-	/* We don't know how to repair anything yet. */
-	if (sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)
-		goto out;
+	/*
+	 * We only want to repair read-write v5+ filesystems.  Defer the check
+	 * for ops->repair until after our scrub confirms that we need to
+	 * perform repairs so that we avoid failing due to not supporting
+	 * repairing an object that doesn't need repairs.
+	 */
+	if (sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) {
+		error = -EOPNOTSUPP;
+		if (!xfs_sb_version_hascrc(&mp->m_sb))
+			goto out;
+
+		error = -EROFS;
+		if (mp->m_flags & XFS_MOUNT_RDONLY)
+			goto out;
+	}
 
 	error = 0;
 out:
 	return error;
 }
 
+#ifdef CONFIG_XFS_ONLINE_REPAIR
+static inline void xfs_scrub_postmortem(struct xfs_scrub_context *sc)
+{
+	/*
+	 * Userspace asked us to repair something, we repaired it, rescanned
+	 * it, and the rescan says it's still broken.  Scream about this in
+	 * the system logs.
+	 */
+	if ((sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) &&
+	    (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
+				 XFS_SCRUB_OFLAG_XCORRUPT)))
+		xfs_repair_failure(sc->mp);
+}
+#else
+static inline void xfs_scrub_postmortem(struct xfs_scrub_context *sc)
+{
+	/*
+	 * Userspace asked us to scrub something, it's broken, and we have no
+	 * way of fixing it.  Scream in the logs.
+	 */
+	if (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
+				XFS_SCRUB_OFLAG_XCORRUPT))
+		xfs_alert_ratelimited(sc->mp,
+				"Corruption detected during scrub.");
+}
+#endif /* CONFIG_XFS_ONLINE_REPAIR */
+
 /* Dispatch metadata scrubbing. */
 int
 xfs_scrub_metadata(
@@ -401,6 +490,7 @@ xfs_scrub_metadata(
 	struct xfs_scrub_context	sc;
 	struct xfs_mount		*mp = ip->i_mount;
 	bool				try_harder = false;
+	bool				already_fixed = false;
 	int				error = 0;
 
 	BUILD_BUG_ON(sizeof(meta_scrub_ops) !=
@@ -450,10 +540,44 @@ retry_op:
 	} else if (error)
 		goto out_teardown;
 
-	if (sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
-			    XFS_SCRUB_OFLAG_XCORRUPT))
-		xfs_alert_ratelimited(mp, "Corruption detected during scrub.");
+	if ((sc.sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) && !already_fixed) {
+		bool needs_fix;
+
+		/* Let debug users force us into the repair routines. */
+		if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_FORCE_SCRUB_REPAIR))
+			sc.sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+
+		needs_fix = (sc.sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
+						XFS_SCRUB_OFLAG_XCORRUPT |
+						XFS_SCRUB_OFLAG_PREEN));
+		/*
+		 * If userspace asked for a repair but it wasn't necessary,
+		 * report that back to userspace.
+		 */
+		if (!needs_fix) {
+			sc.sm->sm_flags |= XFS_SCRUB_OFLAG_NO_REPAIR_NEEDED;
+			goto out_nofix;
+		}
+
+		/*
+		 * If it's broken, userspace wants us to fix it, and we haven't
+		 * already tried to fix it, then attempt a repair.
+		 */
+		error = xfs_repair_attempt(ip, &sc, &already_fixed);
+		if (error == -EAGAIN) {
+			if (sc.try_harder)
+				try_harder = true;
+			error = xfs_scrub_teardown(&sc, ip, 0);
+			if (error) {
+				xfs_repair_failure(mp);
+				goto out;
+			}
+			goto retry_op;
+		}
+	}
 
+out_nofix:
+	xfs_scrub_postmortem(&sc);
 out_teardown:
 	error = xfs_scrub_teardown(&sc, ip, error);
 out: