Make inode64 a remountable option
authorCarlos Maiolino <cmaiolino@redhat.com>
Fri, 17 Aug 2012 21:19:38 +0000 (18:19 -0300)
committerBen Myers <bpm@sgi.com>
Wed, 26 Sep 2012 20:41:39 +0000 (15:41 -0500)
Actually, there is no reason about why a user must umount and mount a
XFS filesystem to enable 'inode64' option. So, this patch makes this a
remountable option.

Signed-off-by: Carlos Maiolino <cmaiolino@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Ben Myers <bpm@sgi.com>
fs/xfs/xfs_super.c

index 19e2380..c416a01 100644 (file)
@@ -120,12 +120,13 @@ mempool_t *xfs_ioend_pool;
  * in the future, too.
  */
 enum {
-       Opt_barrier, Opt_nobarrier, Opt_err
+       Opt_barrier, Opt_nobarrier, Opt_inode64, Opt_err
 };
 
 static const match_table_t tokens = {
        {Opt_barrier, "barrier"},
        {Opt_nobarrier, "nobarrier"},
+       {Opt_inode64, "inode64"},
        {Opt_err, NULL}
 };
 
@@ -1031,6 +1032,30 @@ xfs_restore_resvblks(struct xfs_mount *mp)
        xfs_reserve_blocks(mp, &resblks, NULL);
 }
 
+STATIC void
+xfs_set_inode64(struct xfs_mount *mp)
+{
+       int i = 0;
+
+       for (i = 0; i < mp->m_sb.sb_agcount; i++) {
+               struct xfs_perag        *pag;
+
+               pag = xfs_perag_get(mp, i);
+               pag->pagi_inodeok = 1;
+               pag->pagf_metadata = 0;
+               xfs_perag_put(pag);
+       }
+
+       /* There is no need for lock protection on m_flags,
+        * the rw_semaphore of the VFS superblock is locked
+        * during mount/umount/remount operations, so this is
+        * enough to avoid concurency on the m_flags field
+        */
+       mp->m_flags &= ~(XFS_MOUNT_32BITINODES |
+                        XFS_MOUNT_SMALL_INUMS);
+       mp->m_maxagi = i;
+}
+
 STATIC int
 xfs_fs_remount(
        struct super_block      *sb,
@@ -1056,6 +1081,9 @@ xfs_fs_remount(
                case Opt_nobarrier:
                        mp->m_flags &= ~XFS_MOUNT_BARRIER;
                        break;
+               case Opt_inode64:
+                       xfs_set_inode64(mp);
+                       break;
                default:
                        /*
                         * Logically we would return an error here to prevent