]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
xfs: verify superblocks as they are read from disk
authorDave Chinner <dchinner@redhat.com>
Mon, 12 Nov 2012 11:54:03 +0000 (22:54 +1100)
committerBen Myers <bpm@sgi.com>
Fri, 16 Nov 2012 03:34:07 +0000 (21:34 -0600)
Add a superblock verify callback function and pass it into the
buffer read functions. Remove the now redundant verification code
that is currently in use.

Adding verification shows that secondary superblocks never have
their "sb_inprogress" flag cleared by mkfs.xfs, so when validating
the secondary superblocks during a grow operation we have to avoid
checking this field. Even if we fix mkfs, we will still have to
ignore this field for verification purposes unless a version of mkfs
that does not have this bug was used.

Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Phil White <pwhite@sgi.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
fs/xfs/xfs_fsops.c
fs/xfs/xfs_log_recover.c
fs/xfs/xfs_mount.c
fs/xfs/xfs_mount.h

index f35f8d7731f0a2fbde84671f1aa27c61130b8253..cb65b067ed3138e72e6f03ba34c4cebf114709fe 100644 (file)
@@ -444,7 +444,8 @@ xfs_growfs_data_private(
                if (agno < oagcount) {
                        error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
                                  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
-                                 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
+                                 XFS_FSS_TO_BB(mp, 1), 0, &bp,
+                                 xfs_sb_read_verify);
                } else {
                        bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
                                  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
@@ -462,6 +463,7 @@ xfs_growfs_data_private(
                        break;
                }
                xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
+
                /*
                 * If we get an error writing out the alternate superblocks,
                 * just issue a warning and continue.  The real work is
index eb1e29ff0c7c4a6a31c7dec180f271418a61f989..924a4bc3d49a2f186de13c8171fc25d015e88aa0 100644 (file)
@@ -3692,13 +3692,14 @@ xlog_do_recover(
 
        /*
         * Now that we've finished replaying all buffer and inode
-        * updates, re-read in the superblock.
+        * updates, re-read in the superblock and reverify it.
         */
        bp = xfs_getsb(log->l_mp, 0);
        XFS_BUF_UNDONE(bp);
        ASSERT(!(XFS_BUF_ISWRITE(bp)));
        XFS_BUF_READ(bp);
        XFS_BUF_UNASYNC(bp);
+       bp->b_iodone = xfs_sb_read_verify;
        xfsbdstrat(log->l_mp, bp);
        error = xfs_buf_iowait(bp);
        if (error) {
@@ -3710,7 +3711,7 @@ xlog_do_recover(
 
        /* Convert superblock from on-disk format */
        sbp = &log->l_mp->m_sb;
-       xfs_sb_from_disk(log->l_mp, XFS_BUF_TO_SBP(bp));
+       xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
        ASSERT(sbp->sb_magicnum == XFS_SB_MAGIC);
        ASSERT(xfs_sb_good_version(sbp));
        xfs_buf_relse(bp);
index df6d0b2aade107ace47d51fb1f1253881a4344c1..bff18d73c6105deb3d38054301fc9b8158da62a6 100644 (file)
@@ -304,9 +304,8 @@ STATIC int
 xfs_mount_validate_sb(
        xfs_mount_t     *mp,
        xfs_sb_t        *sbp,
-       int             flags)
+       bool            check_inprogress)
 {
-       int             loud = !(flags & XFS_MFSI_QUIET);
 
        /*
         * If the log device and data device have the
@@ -316,21 +315,18 @@ xfs_mount_validate_sb(
         * a volume filesystem in a non-volume manner.
         */
        if (sbp->sb_magicnum != XFS_SB_MAGIC) {
-               if (loud)
-                       xfs_warn(mp, "bad magic number");
+               xfs_warn(mp, "bad magic number");
                return XFS_ERROR(EWRONGFS);
        }
 
        if (!xfs_sb_good_version(sbp)) {
-               if (loud)
-                       xfs_warn(mp, "bad version");
+               xfs_warn(mp, "bad version");
                return XFS_ERROR(EWRONGFS);
        }
 
        if (unlikely(
            sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
-               if (loud)
-                       xfs_warn(mp,
+               xfs_warn(mp,
                "filesystem is marked as having an external log; "
                "specify logdev on the mount command line.");
                return XFS_ERROR(EINVAL);
@@ -338,8 +334,7 @@ xfs_mount_validate_sb(
 
        if (unlikely(
            sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
-               if (loud)
-                       xfs_warn(mp,
+               xfs_warn(mp,
                "filesystem is marked as having an internal log; "
                "do not specify logdev on the mount command line.");
                return XFS_ERROR(EINVAL);
@@ -373,8 +368,7 @@ xfs_mount_validate_sb(
            sbp->sb_dblocks == 0                                        ||
            sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)                      ||
            sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
-               if (loud)
-                       XFS_CORRUPTION_ERROR("SB sanity check failed",
+               XFS_CORRUPTION_ERROR("SB sanity check failed",
                                XFS_ERRLEVEL_LOW, mp, sbp);
                return XFS_ERROR(EFSCORRUPTED);
        }
@@ -383,12 +377,10 @@ xfs_mount_validate_sb(
         * Until this is fixed only page-sized or smaller data blocks work.
         */
        if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
-               if (loud) {
-                       xfs_warn(mp,
+               xfs_warn(mp,
                "File system with blocksize %d bytes. "
                "Only pagesize (%ld) or less will currently work.",
                                sbp->sb_blocksize, PAGE_SIZE);
-               }
                return XFS_ERROR(ENOSYS);
        }
 
@@ -402,23 +394,20 @@ xfs_mount_validate_sb(
        case 2048:
                break;
        default:
-               if (loud)
-                       xfs_warn(mp, "inode size of %d bytes not supported",
+               xfs_warn(mp, "inode size of %d bytes not supported",
                                sbp->sb_inodesize);
                return XFS_ERROR(ENOSYS);
        }
 
        if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
            xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
-               if (loud)
-                       xfs_warn(mp,
+               xfs_warn(mp,
                "file system too large to be mounted on this system.");
                return XFS_ERROR(EFBIG);
        }
 
-       if (unlikely(sbp->sb_inprogress)) {
-               if (loud)
-                       xfs_warn(mp, "file system busy");
+       if (check_inprogress && sbp->sb_inprogress) {
+               xfs_warn(mp, "Offline file system operation in progress!");
                return XFS_ERROR(EFSCORRUPTED);
        }
 
@@ -426,9 +415,7 @@ xfs_mount_validate_sb(
         * Version 1 directory format has never worked on Linux.
         */
        if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
-               if (loud)
-                       xfs_warn(mp,
-                               "file system using version 1 directory format");
+               xfs_warn(mp, "file system using version 1 directory format");
                return XFS_ERROR(ENOSYS);
        }
 
@@ -521,11 +508,9 @@ out_unwind:
 
 void
 xfs_sb_from_disk(
-       struct xfs_mount        *mp,
+       struct xfs_sb   *to,
        xfs_dsb_t       *from)
 {
-       struct xfs_sb *to = &mp->m_sb;
-
        to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
        to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
        to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
@@ -627,6 +612,50 @@ xfs_sb_to_disk(
        }
 }
 
+void
+xfs_sb_read_verify(
+       struct xfs_buf  *bp)
+{
+       struct xfs_mount *mp = bp->b_target->bt_mount;
+       struct xfs_sb   sb;
+       int             error;
+
+       xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp));
+
+       /*
+        * Only check the in progress field for the primary superblock as
+        * mkfs.xfs doesn't clear it from secondary superblocks.
+        */
+       error = xfs_mount_validate_sb(mp, &sb, bp->b_bn == XFS_SB_DADDR);
+       if (error)
+               xfs_buf_ioerror(bp, error);
+       bp->b_iodone = NULL;
+       xfs_buf_ioend(bp, 0);
+}
+
+/*
+ * We may be probed for a filesystem match, so we may not want to emit
+ * messages when the superblock buffer is not actually an XFS superblock.
+ * If we find an XFS superblock, the run a normal, noisy mount because we are
+ * really going to mount it and want to know about errors.
+ */
+void
+xfs_sb_quiet_read_verify(
+       struct xfs_buf  *bp)
+{
+       struct xfs_sb   sb;
+
+       xfs_sb_from_disk(&sb, XFS_BUF_TO_SBP(bp));
+
+       if (sb.sb_magicnum == XFS_SB_MAGIC) {
+               /* XFS filesystem, verify noisily! */
+               xfs_sb_read_verify(bp);
+               return;
+       }
+       /* quietly fail */
+       xfs_buf_ioerror(bp, EFSCORRUPTED);
+}
+
 /*
  * xfs_readsb
  *
@@ -652,7 +681,9 @@ xfs_readsb(xfs_mount_t *mp, int flags)
 
 reread:
        bp = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
-                                       BTOBB(sector_size), 0, NULL);
+                                  BTOBB(sector_size), 0,
+                                  loud ? xfs_sb_read_verify
+                                       : xfs_sb_quiet_read_verify);
        if (!bp) {
                if (loud)
                        xfs_warn(mp, "SB buffer read failed");
@@ -667,15 +698,8 @@ reread:
 
        /*
         * Initialize the mount structure from the superblock.
-        * But first do some basic consistency checking.
         */
-       xfs_sb_from_disk(mp, XFS_BUF_TO_SBP(bp));
-       error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
-       if (error) {
-               if (loud)
-                       xfs_warn(mp, "SB validate failed");
-               goto release_buf;
-       }
+       xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
 
        /*
         * We must be able to do sector-sized and sector-aligned IO.
index dc306a09f56fa8338d055b2fa5f843975c4ab520..de9089acc610bf9c0f6737e74be56a372cefda7d 100644 (file)
@@ -385,10 +385,11 @@ extern void       xfs_set_low_space_thresholds(struct xfs_mount *);
 
 #endif /* __KERNEL__ */
 
+extern void    xfs_sb_read_verify(struct xfs_buf *);
 extern void    xfs_mod_sb(struct xfs_trans *, __int64_t);
 extern int     xfs_initialize_perag(struct xfs_mount *, xfs_agnumber_t,
                                        xfs_agnumber_t *);
-extern void    xfs_sb_from_disk(struct xfs_mount *, struct xfs_dsb *);
+extern void    xfs_sb_from_disk(struct xfs_sb *, struct xfs_dsb *);
 extern void    xfs_sb_to_disk(struct xfs_dsb *, struct xfs_sb *, __int64_t);
 
 #endif /* __XFS_MOUNT_H__ */