]> err.no Git - linux-2.6/blobdiff - fs/xfs/xfs_mount.c
[CIFS] have cifs_get_spnego_key get the hostname from TCP_Server_Info
[linux-2.6] / fs / xfs / xfs_mount.c
index 3bed0cf0d8afa2eaa87a98b00be0a10034c701b7..ebdb76da527c81d6c021e0ffe995ef1395393c9e 100644 (file)
@@ -139,7 +139,7 @@ xfs_mount_init(void)
        AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
        spinlock_init(&mp->m_sb_lock, "xfs_sb");
        mutex_init(&mp->m_ilock);
-       initnsema(&mp->m_growlock, 1, "xfs_grow");
+       mutex_init(&mp->m_growlock);
        /*
         * Initialize the AIL.
         */
@@ -157,14 +157,8 @@ xfs_mount_init(void)
  */
 void
 xfs_mount_free(
-       xfs_mount_t     *mp,
-       int             remove_bhv)
+       xfs_mount_t     *mp)
 {
-       if (mp->m_ihash)
-               xfs_ihash_free(mp);
-       if (mp->m_chash)
-               xfs_chash_free(mp);
-
        if (mp->m_perag) {
                int     agno;
 
@@ -180,7 +174,7 @@ xfs_mount_free(
        AIL_LOCK_DESTROY(&mp->m_ail_lock);
        spinlock_destroy(&mp->m_sb_lock);
        mutex_destroy(&mp->m_ilock);
-       freesema(&mp->m_growlock);
+       mutex_destroy(&mp->m_growlock);
        if (mp->m_quotainfo)
                XFS_QM_DONE(mp);
 
@@ -191,17 +185,30 @@ xfs_mount_free(
        if (mp->m_logname != NULL)
                kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
 
-       if (remove_bhv) {
-               struct bhv_vfs  *vfsp = XFS_MTOVFS(mp);
-
-               bhv_remove_all_vfsops(vfsp, 0);
-               VFS_REMOVEBHV(vfsp, &mp->m_bhv);
-       }
-
        xfs_icsb_destroy_counters(mp);
-       kmem_free(mp, sizeof(xfs_mount_t));
 }
 
+/*
+ * Check size of device based on the (data/realtime) block count.
+ * Note: this check is used by the growfs code as well as mount.
+ */
+int
+xfs_sb_validate_fsb_count(
+       xfs_sb_t        *sbp,
+       __uint64_t      nblocks)
+{
+       ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
+       ASSERT(sbp->sb_blocklog >= BBSHIFT);
+
+#if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
+       if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
+               return E2BIG;
+#else                  /* Limited by UINT_MAX of sectors */
+       if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
+               return E2BIG;
+#endif
+       return 0;
+}
 
 /*
  * Check the validity of the SB found.
@@ -284,18 +291,8 @@ xfs_mount_validate_sb(
                return XFS_ERROR(EFSCORRUPTED);
        }
 
-       ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
-       ASSERT(sbp->sb_blocklog >= BBSHIFT);
-
-#if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
-       if (unlikely(
-           (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
-           (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
-#else                  /* Limited by UINT_MAX of sectors */
-       if (unlikely(
-           (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
-           (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
-#endif
+       if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
+           xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
                xfs_fs_mount_cmn_err(flags,
                        "file system too large to be mounted on this system.");
                return XFS_ERROR(E2BIG);
@@ -331,9 +328,19 @@ xfs_mount_validate_sb(
        return 0;
 }
 
+STATIC void
+xfs_initialize_perag_icache(
+       xfs_perag_t     *pag)
+{
+       if (!pag->pag_ici_init) {
+               rwlock_init(&pag->pag_ici_lock);
+               INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
+               pag->pag_ici_init = 1;
+       }
+}
+
 xfs_agnumber_t
 xfs_initialize_perag(
-       bhv_vfs_t       *vfs,
        xfs_mount_t     *mp,
        xfs_agnumber_t  agcount)
 {
@@ -351,7 +358,7 @@ xfs_initialize_perag(
        /* Clear the mount flag if no inode can overflow 32 bits
         * on this filesystem, or if specifically requested..
         */
-       if ((vfs->vfs_flag & VFS_32BITINODES) && ino > max_inum) {
+       if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
                mp->m_flags |= XFS_MOUNT_32BITINODES;
        } else {
                mp->m_flags &= ~XFS_MOUNT_32BITINODES;
@@ -385,48 +392,92 @@ xfs_initialize_perag(
                        pag->pagi_inodeok = 1;
                        if (index < max_metadata)
                                pag->pagf_metadata = 1;
+                       xfs_initialize_perag_icache(pag);
                }
        } else {
                /* Setup default behavior for smaller filesystems */
                for (index = 0; index < agcount; index++) {
                        pag = &mp->m_perag[index];
                        pag->pagi_inodeok = 1;
+                       xfs_initialize_perag_icache(pag);
                }
        }
        return index;
 }
 
+void
+xfs_sb_from_disk(
+       xfs_sb_t        *to,
+       xfs_dsb_t       *from)
+{
+       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);
+       to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
+       to->sb_rextents = be64_to_cpu(from->sb_rextents);
+       memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
+       to->sb_logstart = be64_to_cpu(from->sb_logstart);
+       to->sb_rootino = be64_to_cpu(from->sb_rootino);
+       to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
+       to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
+       to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
+       to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
+       to->sb_agcount = be32_to_cpu(from->sb_agcount);
+       to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
+       to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
+       to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
+       to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
+       to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
+       to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
+       memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
+       to->sb_blocklog = from->sb_blocklog;
+       to->sb_sectlog = from->sb_sectlog;
+       to->sb_inodelog = from->sb_inodelog;
+       to->sb_inopblog = from->sb_inopblog;
+       to->sb_agblklog = from->sb_agblklog;
+       to->sb_rextslog = from->sb_rextslog;
+       to->sb_inprogress = from->sb_inprogress;
+       to->sb_imax_pct = from->sb_imax_pct;
+       to->sb_icount = be64_to_cpu(from->sb_icount);
+       to->sb_ifree = be64_to_cpu(from->sb_ifree);
+       to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
+       to->sb_frextents = be64_to_cpu(from->sb_frextents);
+       to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
+       to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
+       to->sb_qflags = be16_to_cpu(from->sb_qflags);
+       to->sb_flags = from->sb_flags;
+       to->sb_shared_vn = from->sb_shared_vn;
+       to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
+       to->sb_unit = be32_to_cpu(from->sb_unit);
+       to->sb_width = be32_to_cpu(from->sb_width);
+       to->sb_dirblklog = from->sb_dirblklog;
+       to->sb_logsectlog = from->sb_logsectlog;
+       to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
+       to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
+       to->sb_features2 = be32_to_cpu(from->sb_features2);
+}
+
 /*
- * xfs_xlatesb
+ * Copy in core superblock to ondisk one.
  *
- *     data       - on disk version of sb
- *     sb         - a superblock
- *     dir        - conversion direction: <0 - convert sb to buf
- *                                        >0 - convert buf to sb
- *     fields     - which fields to copy (bitmask)
+ * The fields argument is mask of superblock fields to copy.
  */
 void
-xfs_xlatesb(
-       void            *data,
-       xfs_sb_t        *sb,
-       int             dir,
+xfs_sb_to_disk(
+       xfs_dsb_t       *to,
+       xfs_sb_t        *from,
        __int64_t       fields)
 {
-       xfs_caddr_t     buf_ptr;
-       xfs_caddr_t     mem_ptr;
+       xfs_caddr_t     to_ptr = (xfs_caddr_t)to;
+       xfs_caddr_t     from_ptr = (xfs_caddr_t)from;
        xfs_sb_field_t  f;
        int             first;
        int             size;
 
-       ASSERT(dir);
        ASSERT(fields);
-
        if (!fields)
                return;
 
-       buf_ptr = (xfs_caddr_t)data;
-       mem_ptr = (xfs_caddr_t)sb;
-
        while (fields) {
                f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
                first = xfs_sb_info[f].offset;
@@ -435,26 +486,20 @@ xfs_xlatesb(
                ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
 
                if (size == 1 || xfs_sb_info[f].type == 1) {
-                       if (dir > 0) {
-                               memcpy(mem_ptr + first, buf_ptr + first, size);
-                       } else {
-                               memcpy(buf_ptr + first, mem_ptr + first, size);
-                       }
+                       memcpy(to_ptr + first, from_ptr + first, size);
                } else {
                        switch (size) {
                        case 2:
-                               INT_XLATE(*(__uint16_t*)(buf_ptr+first),
-                                         *(__uint16_t*)(mem_ptr+first),
-                                         dir, ARCH_CONVERT);
+                               *(__be16 *)(to_ptr + first) =
+                                       cpu_to_be16(*(__u16 *)(from_ptr + first));
                                break;
                        case 4:
-                               INT_XLATE(*(__uint32_t*)(buf_ptr+first),
-                                         *(__uint32_t*)(mem_ptr+first),
-                                         dir, ARCH_CONVERT);
+                               *(__be32 *)(to_ptr + first) =
+                                       cpu_to_be32(*(__u32 *)(from_ptr + first));
                                break;
                        case 8:
-                               INT_XLATE(*(__uint64_t*)(buf_ptr+first),
-                                         *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT);
+                               *(__be64 *)(to_ptr + first) =
+                                       cpu_to_be64(*(__u64 *)(from_ptr + first));
                                break;
                        default:
                                ASSERT(0);
@@ -476,7 +521,6 @@ xfs_readsb(xfs_mount_t *mp, int flags)
        unsigned int    sector_size;
        unsigned int    extra_flags;
        xfs_buf_t       *bp;
-       xfs_sb_t        *sbp;
        int             error;
 
        ASSERT(mp->m_sb_bp == NULL);
@@ -504,8 +548,7 @@ xfs_readsb(xfs_mount_t *mp, int flags)
         * Initialize the mount structure from the superblock.
         * But first do some basic consistency checking.
         */
-       sbp = XFS_BUF_TO_SBP(bp);
-       xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS);
+       xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
 
        error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
        if (error) {
@@ -632,6 +675,64 @@ xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
                                        sbp->sb_inopblock);
        mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
 }
+
+/*
+ * xfs_initialize_perag_data
+ *
+ * Read in each per-ag structure so we can count up the number of
+ * allocated inodes, free inodes and used filesystem blocks as this
+ * information is no longer persistent in the superblock. Once we have
+ * this information, write it into the in-core superblock structure.
+ */
+STATIC int
+xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
+{
+       xfs_agnumber_t  index;
+       xfs_perag_t     *pag;
+       xfs_sb_t        *sbp = &mp->m_sb;
+       uint64_t        ifree = 0;
+       uint64_t        ialloc = 0;
+       uint64_t        bfree = 0;
+       uint64_t        bfreelst = 0;
+       uint64_t        btree = 0;
+       int             error;
+       int             s;
+
+       for (index = 0; index < agcount; index++) {
+               /*
+                * read the agf, then the agi. This gets us
+                * all the inforamtion we need and populates the
+                * per-ag structures for us.
+                */
+               error = xfs_alloc_pagf_init(mp, NULL, index, 0);
+               if (error)
+                       return error;
+
+               error = xfs_ialloc_pagi_init(mp, NULL, index);
+               if (error)
+                       return error;
+               pag = &mp->m_perag[index];
+               ifree += pag->pagi_freecount;
+               ialloc += pag->pagi_count;
+               bfree += pag->pagf_freeblks;
+               bfreelst += pag->pagf_flcount;
+               btree += pag->pagf_btreeblks;
+       }
+       /*
+        * Overwrite incore superblock counters with just-read data
+        */
+       s = XFS_SB_LOCK(mp);
+       sbp->sb_ifree = ifree;
+       sbp->sb_icount = ialloc;
+       sbp->sb_fdblocks = bfree + bfreelst + btree;
+       XFS_SB_UNLOCK(mp, s);
+
+       /* Fixup the per-cpu counters as well. */
+       xfs_icsb_reinit_counters(mp);
+
+       return 0;
+}
+
 /*
  * xfs_mountfs
  *
@@ -646,7 +747,6 @@ xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
  */
 int
 xfs_mountfs(
-       bhv_vfs_t       *vfsp,
        xfs_mount_t     *mp,
        int             mfsi_flags)
 {
@@ -656,7 +756,7 @@ xfs_mountfs(
        bhv_vnode_t     *rvp = NULL;
        int             readio_log, writeio_log;
        xfs_daddr_t     d;
-       __uint64_t      ret64;
+       __uint64_t      resblks;
        __int64_t       update_flags;
        uint            quotamount, quotaflags;
        int             agno;
@@ -778,8 +878,6 @@ xfs_mountfs(
                        goto error1;
                }
                uuid_mounted=1;
-               ret64 = uuid_hash64(&sbp->sb_uuid);
-               memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
        }
 
        /*
@@ -801,16 +899,6 @@ xfs_mountfs(
                writeio_log = mp->m_writeio_log;
        }
 
-       /*
-        * Set the number of readahead buffers to use based on
-        * physical memory size.
-        */
-       if (xfs_physmem <= 4096)                /* <= 16MB */
-               mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
-       else if (xfs_physmem <= 8192)   /* <= 32MB */
-               mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
-       else
-               mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
        if (sbp->sb_blocklog > readio_log) {
                mp->m_readio_log = sbp->sb_blocklog;
        } else {
@@ -825,15 +913,12 @@ xfs_mountfs(
        mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
 
        /*
-        * Set the inode cluster size based on the physical memory
-        * size.  This may still be overridden by the file system
+        * Set the inode cluster size.
+        * This may still be overridden by the file system
         * block size if it is larger than the chosen cluster size.
         */
-       if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
-               mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
-       } else {
-               mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
-       }
+       mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
+
        /*
         * Set whether we're using inode alignment.
         */
@@ -917,16 +1002,6 @@ xfs_mountfs(
         */
        uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
 
-       /*
-        *  The vfs structure needs to have a file system independent
-        *  way of checking for the invariant file system ID.  Since it
-        *  can't look at mount structures it has a pointer to the data
-        *  in the mount structure.
-        *
-        *  File systems that don't support user level file handles (i.e.
-        *  all of them except for XFS) will leave vfs_altfsid as NULL.
-        */
-       vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
        mp->m_dmevmask = 0;     /* not persistent; set after each mount */
 
        xfs_dir_mount(mp);
@@ -941,13 +1016,6 @@ xfs_mountfs(
         */
        xfs_trans_init(mp);
 
-       /*
-        * Allocate and initialize the inode hash table for this
-        * file system.
-        */
-       xfs_ihash_init(mp);
-       xfs_chash_init(mp);
-
        /*
         * Allocate and initialize the per-ag data.
         */
@@ -955,7 +1023,7 @@ xfs_mountfs(
        mp->m_perag =
                kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
 
-       mp->m_maxagi = xfs_initialize_perag(vfsp, mp, sbp->sb_agcount);
+       mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
 
        /*
         * log's mount-time initialization. Perform 1st part recovery if needed
@@ -975,6 +1043,34 @@ xfs_mountfs(
                goto error2;
        }
 
+       /*
+        * Now the log is mounted, we know if it was an unclean shutdown or
+        * not. If it was, with the first phase of recovery has completed, we
+        * have consistent AG blocks on disk. We have not recovered EFIs yet,
+        * but they are recovered transactionally in the second recovery phase
+        * later.
+        *
+        * Hence we can safely re-initialise incore superblock counters from
+        * the per-ag data. These may not be correct if the filesystem was not
+        * cleanly unmounted, so we need to wait for recovery to finish before
+        * doing this.
+        *
+        * If the filesystem was cleanly unmounted, then we can trust the
+        * values in the superblock to be correct and we don't need to do
+        * anything here.
+        *
+        * If we are currently making the filesystem, the initialisation will
+        * fail as the perag data is in an undefined state.
+        */
+
+       if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
+           !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
+            !mp->m_sb.sb_inprogress) {
+               error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
+               if (error) {
+                       goto error2;
+               }
+       }
        /*
         * Get and sanity-check the root inode.
         * Save the pointer to it in the mount structure.
@@ -1018,7 +1114,7 @@ xfs_mountfs(
         * If fs is not mounted readonly, then update the superblock
         * unit and width changes.
         */
-       if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
+       if (update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY))
                xfs_mount_log_sbunit(mp, update_flags);
 
        /*
@@ -1044,6 +1140,23 @@ xfs_mountfs(
        if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
                goto error4;
 
+       /*
+        * Now we are mounted, reserve a small amount of unused space for
+        * privileged transactions. This is needed so that transaction
+        * space required for critical operations can dip into this pool
+        * when at ENOSPC. This is needed for operations like create with
+        * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
+        * are not allowed to use this reserved space.
+        *
+        * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
+        * This may drive us straight to ENOSPC on mount, but that implies
+        * we were already there on the last unmount.
+        */
+       resblks = mp->m_sb.sb_dblocks;
+       do_div(resblks, 20);
+       resblks = min_t(__uint64_t, resblks, 1024);
+       xfs_reserve_blocks(mp, &resblks, NULL);
+
        return 0;
 
  error4:
@@ -1054,8 +1167,6 @@ xfs_mountfs(
  error3:
        xfs_log_unmount_dealloc(mp);
  error2:
-       xfs_ihash_free(mp);
-       xfs_chash_free(mp);
        for (agno = 0; agno < sbp->sb_agcount; agno++)
                if (mp->m_perag[agno].pagb_list)
                        kmem_free(mp->m_perag[agno].pagb_list,
@@ -1079,11 +1190,19 @@ xfs_mountfs(
 int
 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
 {
-       struct bhv_vfs  *vfsp = XFS_MTOVFS(mp);
-#if defined(DEBUG) || defined(INDUCE_IO_ERROR)
-       int64_t         fsid;
-#endif
+       __uint64_t      resblks;
 
+       /*
+        * We can potentially deadlock here if we have an inode cluster
+        * that has been freed has it's buffer still pinned in memory because
+        * the transaction is still sitting in a iclog. The stale inodes
+        * on that buffer will have their flush locks held until the
+        * transaction hits the disk and the callbacks run. the inode
+        * flush takes the flush lock unconditionally and with nothing to
+        * push out the iclog we will never get that unlocked. hence we
+        * need to force the log first.
+        */
+       xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
        xfs_iflush_all(mp);
 
        XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
@@ -1100,10 +1219,26 @@ xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
                xfs_binval(mp->m_rtdev_targp);
        }
 
-       xfs_unmountfs_writesb(mp);
+       /*
+        * Unreserve any blocks we have so that when we unmount we don't account
+        * the reserved free space as used. This is really only necessary for
+        * lazy superblock counting because it trusts the incore superblock
+        * counters to be aboslutely correct on clean unmount.
+        *
+        * We don't bother correcting this elsewhere for lazy superblock
+        * counting because on mount of an unclean filesystem we reconstruct the
+        * correct counter value and this is irrelevant.
+        *
+        * For non-lazy counter filesystems, this doesn't matter at all because
+        * we only every apply deltas to the superblock and hence the incore
+        * value does not matter....
+        */
+       resblks = 0;
+       xfs_reserve_blocks(mp, &resblks, NULL);
 
+       xfs_log_sbcount(mp, 1);
+       xfs_unmountfs_writesb(mp);
        xfs_unmountfs_wait(mp);                 /* wait for async bufs */
-
        xfs_log_unmount(mp);                    /* Done! No more fs ops. */
 
        xfs_freesb(mp);
@@ -1118,21 +1253,17 @@ xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
                xfs_uuid_unmount(mp);
 
 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
-       /*
-        * clear all error tags on this filesystem
-        */
-       memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
-       xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
+       xfs_errortag_clearall(mp, 0);
 #endif
-       XFS_IODONE(vfsp);
-       xfs_mount_free(mp, 1);
+       XFS_IODONE(mp);
+       xfs_mount_free(mp);
        return 0;
 }
 
 void
 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
 {
-       if (mp->m_logdev_targp != mp->m_ddev_targp)
+       if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
                xfs_free_buftarg(mp->m_logdev_targp, 1);
        if (mp->m_rtdev_targp)
                xfs_free_buftarg(mp->m_rtdev_targp, 1);
@@ -1149,35 +1280,99 @@ xfs_unmountfs_wait(xfs_mount_t *mp)
        xfs_wait_buftarg(mp->m_ddev_targp);
 }
 
+int
+xfs_fs_writable(xfs_mount_t *mp)
+{
+       return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
+               (mp->m_flags & XFS_MOUNT_RDONLY));
+}
+
+/*
+ * xfs_log_sbcount
+ *
+ * Called either periodically to keep the on disk superblock values
+ * roughly up to date or from unmount to make sure the values are
+ * correct on a clean unmount.
+ *
+ * Note this code can be called during the process of freezing, so
+ * we may need to use the transaction allocator which does not not
+ * block when the transaction subsystem is in its frozen state.
+ */
+int
+xfs_log_sbcount(
+       xfs_mount_t     *mp,
+       uint            sync)
+{
+       xfs_trans_t     *tp;
+       int             error;
+
+       if (!xfs_fs_writable(mp))
+               return 0;
+
+       xfs_icsb_sync_counters(mp);
+
+       /*
+        * we don't need to do this if we are updating the superblock
+        * counters on every modification.
+        */
+       if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
+               return 0;
+
+       tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
+       error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
+                                       XFS_DEFAULT_LOG_COUNT);
+       if (error) {
+               xfs_trans_cancel(tp, 0);
+               return error;
+       }
+
+       xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
+       if (sync)
+               xfs_trans_set_sync(tp);
+       xfs_trans_commit(tp, 0);
+
+       return 0;
+}
+
+STATIC void
+xfs_mark_shared_ro(
+       xfs_mount_t     *mp,
+       xfs_buf_t       *bp)
+{
+       xfs_dsb_t       *sb = XFS_BUF_TO_SBP(bp);
+       __uint16_t      version;
+
+       if (!(sb->sb_flags & XFS_SBF_READONLY))
+               sb->sb_flags |= XFS_SBF_READONLY;
+
+       version = be16_to_cpu(sb->sb_versionnum);
+       if ((version & XFS_SB_VERSION_NUMBITS) != XFS_SB_VERSION_4 ||
+           !(version & XFS_SB_VERSION_SHAREDBIT))
+               version |= XFS_SB_VERSION_SHAREDBIT;
+       sb->sb_versionnum = cpu_to_be16(version);
+}
+
 int
 xfs_unmountfs_writesb(xfs_mount_t *mp)
 {
        xfs_buf_t       *sbp;
-       xfs_sb_t        *sb;
        int             error = 0;
 
        /*
         * skip superblock write if fs is read-only, or
         * if we are doing a forced umount.
         */
-       sbp = xfs_getsb(mp, 0);
-       if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
+       if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
                XFS_FORCED_SHUTDOWN(mp))) {
 
-               xfs_icsb_sync_counters(mp);
+               sbp = xfs_getsb(mp, 0);
 
                /*
                 * mark shared-readonly if desired
                 */
-               sb = XFS_BUF_TO_SBP(sbp);
-               if (mp->m_mk_sharedro) {
-                       if (!(sb->sb_flags & XFS_SBF_READONLY))
-                               sb->sb_flags |= XFS_SBF_READONLY;
-                       if (!XFS_SB_VERSION_HASSHARED(sb))
-                               XFS_SB_VERSION_ADDSHARED(sb);
-                       xfs_fs_cmn_err(CE_NOTE, mp,
-                               "Unmounting, marking shared read-only");
-               }
+               if (mp->m_mk_sharedro)
+                       xfs_mark_shared_ro(mp, sbp);
+
                XFS_BUF_UNDONE(sbp);
                XFS_BUF_UNREAD(sbp);
                XFS_BUF_UNDELAYWRITE(sbp);
@@ -1192,8 +1387,8 @@ xfs_unmountfs_writesb(xfs_mount_t *mp)
                                          mp, sbp, XFS_BUF_ADDR(sbp));
                if (error && mp->m_mk_sharedro)
                        xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting.  Filesystem may not be marked shared readonly");
+               xfs_buf_relse(sbp);
        }
-       xfs_buf_relse(sbp);
        return error;
 }
 
@@ -1211,7 +1406,6 @@ xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
        int             first;
        int             last;
        xfs_mount_t     *mp;
-       xfs_sb_t        *sbp;
        xfs_sb_field_t  f;
 
        ASSERT(fields);
@@ -1219,13 +1413,12 @@ xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
                return;
        mp = tp->t_mountp;
        bp = xfs_trans_getsb(tp, mp, 0);
-       sbp = XFS_BUF_TO_SBP(bp);
        first = sizeof(xfs_sb_t);
        last = 0;
 
        /* translate/copy */
 
-       xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
+       xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
 
        /* find modified range */
 
@@ -1653,7 +1846,7 @@ xfs_mount_log_sbunit(
                return;
        }
        xfs_mod_sb(tp, fields);
-       xfs_trans_commit(tp, 0, NULL);
+       xfs_trans_commit(tp, 0);
 }
 
 
@@ -1734,11 +1927,13 @@ xfs_icsb_cpu_notify(
                        per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
        switch (action) {
        case CPU_UP_PREPARE:
+       case CPU_UP_PREPARE_FROZEN:
                /* Easy Case - initialize the area and locks, and
                 * then rebalance when online does everything else for us. */
                memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
                break;
        case CPU_ONLINE:
+       case CPU_ONLINE_FROZEN:
                xfs_icsb_lock(mp);
                xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
                xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
@@ -1746,6 +1941,7 @@ xfs_icsb_cpu_notify(
                xfs_icsb_unlock(mp);
                break;
        case CPU_DEAD:
+       case CPU_DEAD_FROZEN:
                /* Disable all the counters, then fold the dead cpu's
                 * count into the total on the global superblock and
                 * re-enable the counters. */