Merge branch 'for-next' into for-linus
[safe/jmp/linux-2.6] / fs / gfs2 / ops_fstype.c
index 3eb49ed..c1309ed 100644 (file)
@@ -17,7 +17,8 @@
 #include <linux/namei.h>
 #include <linux/mount.h>
 #include <linux/gfs2_ondisk.h>
-#include <linux/lm_interface.h>
+#include <linux/slow-work.h>
+#include <linux/quotaops.h>
 
 #include "gfs2.h"
 #include "incore.h"
@@ -33,6 +34,7 @@
 #include "log.h"
 #include "quota.h"
 #include "dir.h"
+#include "trace_gfs2.h"
 
 #define DO 0
 #define UNDO 1
@@ -56,21 +58,14 @@ static void gfs2_tune_init(struct gfs2_tune *gt)
        spin_lock_init(&gt->gt_spin);
 
        gt->gt_incore_log_blocks = 1024;
-       gt->gt_log_flush_secs = 60;
-       gt->gt_recoverd_secs = 60;
        gt->gt_logd_secs = 1;
        gt->gt_quota_simul_sync = 64;
        gt->gt_quota_warn_period = 10;
        gt->gt_quota_scale_num = 1;
        gt->gt_quota_scale_den = 1;
-       gt->gt_quota_cache_secs = 300;
-       gt->gt_quota_quantum = 60;
        gt->gt_new_files_jdata = 0;
        gt->gt_max_readahead = 1 << 18;
-       gt->gt_stall_secs = 600;
        gt->gt_complain_secs = 10;
-       gt->gt_statfs_quantum = 30;
-       gt->gt_statfs_slow = 0;
 }
 
 static struct gfs2_sbd *init_sbd(struct super_block *sb)
@@ -86,7 +81,8 @@ static struct gfs2_sbd *init_sbd(struct super_block *sb)
 
        gfs2_tune_init(&sdp->sd_tune);
 
-       mutex_init(&sdp->sd_inum_mutex);
+       init_waitqueue_head(&sdp->sd_glock_wait);
+       atomic_set(&sdp->sd_glock_disposal, 0);
        spin_lock_init(&sdp->sd_statfs_spin);
 
        spin_lock_init(&sdp->sd_rindex_spin);
@@ -99,7 +95,6 @@ static struct gfs2_sbd *init_sbd(struct super_block *sb)
        mutex_init(&sdp->sd_jindex_mutex);
 
        INIT_LIST_HEAD(&sdp->sd_quota_list);
-       spin_lock_init(&sdp->sd_quota_spin);
        mutex_init(&sdp->sd_quota_mutex);
        init_waitqueue_head(&sdp->sd_quota_wait);
        INIT_LIST_HEAD(&sdp->sd_trunc_list);
@@ -237,6 +232,7 @@ static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
 
        memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
        memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
+       memcpy(sb->sb_uuid, str->sb_uuid, 16);
 }
 
 /**
@@ -274,11 +270,6 @@ static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
        lock_page(page);
 
        bio = bio_alloc(GFP_NOFS, 1);
-       if (unlikely(!bio)) {
-               __free_page(page);
-               return -ENOBUFS;
-       }
-
        bio->bi_sector = sector * (sb->s_blocksize >> 9);
        bio->bi_bdev = sb->s_bdev;
        bio_add_page(bio, page, PAGE_SIZE, 0);
@@ -298,15 +289,15 @@ static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
        __free_page(page);
        return 0;
 }
+
 /**
  * gfs2_read_sb - Read super block
  * @sdp: The GFS2 superblock
- * @gl: the glock for the superblock (assumed to be held)
  * @silent: Don't print message if mount fails
  *
  */
 
-static int gfs2_read_sb(struct gfs2_sbd *sdp, struct gfs2_glock *gl, int silent)
+static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
 {
        u32 hash_blocks, ind_blocks, leaf_blocks;
        u32 tmp_blocks;
@@ -526,18 +517,18 @@ static int init_sb(struct gfs2_sbd *sdp, int silent)
                return ret;
        }
 
-       ret = gfs2_read_sb(sdp, sb_gh.gh_gl, silent);
+       ret = gfs2_read_sb(sdp, silent);
        if (ret) {
                fs_err(sdp, "can't read superblock: %d\n", ret);
                goto out;
        }
 
        /* Set up the buffer cache and SB for real */
-       if (sdp->sd_sb.sb_bsize < bdev_hardsect_size(sb->s_bdev)) {
+       if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
                ret = -EINVAL;
                fs_err(sdp, "FS block size (%u) is too small for device "
                       "block size (%u)\n",
-                      sdp->sd_sb.sb_bsize, bdev_hardsect_size(sb->s_bdev));
+                      sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
                goto out;
        }
        if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
@@ -629,13 +620,13 @@ static int map_journal_extents(struct gfs2_sbd *sdp)
        return rc;
 }
 
-static void gfs2_lm_others_may_mount(struct gfs2_sbd *sdp)
+static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
 {
-       if (!sdp->sd_lockstruct.ls_ops->lm_others_may_mount)
-               return;
-       if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
-               sdp->sd_lockstruct.ls_ops->lm_others_may_mount(
-                                       sdp->sd_lockstruct.ls_lockspace);
+       char *message = "FIRSTMOUNT=Done";
+       char *envp[] = { message, NULL };
+       struct lm_lockstruct *ls = &sdp->sd_lockstruct;
+       ls->ls_first_done = 1;
+       kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
 }
 
 /**
@@ -683,6 +674,7 @@ static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
                        break;
 
                INIT_LIST_HEAD(&jd->extent_list);
+               slow_work_init(&jd->jd_work, &gfs2_recover_ops);
                jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
                if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
                        if (!jd->jd_inode)
@@ -708,14 +700,13 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
 {
        struct inode *master = sdp->sd_master_dir->d_inode;
        struct gfs2_holder ji_gh;
-       struct task_struct *p;
        struct gfs2_inode *ip;
        int jindex = 1;
        int error = 0;
 
        if (undo) {
                jindex = 0;
-               goto fail_recoverd;
+               goto fail_jinode_gh;
        }
 
        sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
@@ -733,7 +724,7 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
                goto fail;
        }
 
-       error = -EINVAL;
+       error = -EUSERS;
        if (!gfs2_jindex_size(sdp)) {
                fs_err(sdp, "no journals!\n");
                goto fail_jindex;
@@ -783,6 +774,7 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
                /* Map the extents for this journal's blocks */
                map_journal_extents(sdp);
        }
+       trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
 
        if (sdp->sd_lockstruct.ls_first) {
                unsigned int x;
@@ -795,7 +787,7 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
                        }
                }
 
-               gfs2_lm_others_may_mount(sdp);
+               gfs2_others_may_mount(sdp);
        } else if (!sdp->sd_args.ar_spectator) {
                error = gfs2_recover_journal(sdp->sd_jdesc);
                if (error) {
@@ -808,18 +800,8 @@ static int init_journal(struct gfs2_sbd *sdp, int undo)
        gfs2_glock_dq_uninit(&ji_gh);
        jindex = 0;
 
-       p = kthread_run(gfs2_recoverd, sdp, "gfs2_recoverd");
-       error = IS_ERR(p);
-       if (error) {
-               fs_err(sdp, "can't start recoverd thread: %d\n", error);
-               goto fail_jinode_gh;
-       }
-       sdp->sd_recoverd_process = p;
-
        return 0;
 
-fail_recoverd:
-       kthread_stop(sdp->sd_recoverd_process);
 fail_jinode_gh:
        if (!sdp->sd_args.ar_spectator)
                gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
@@ -849,21 +831,12 @@ static int init_inodes(struct gfs2_sbd *sdp, int undo)
        if (error)
                goto fail;
 
-       /* Read in the master inode number inode */
-       sdp->sd_inum_inode = gfs2_lookup_simple(master, "inum");
-       if (IS_ERR(sdp->sd_inum_inode)) {
-               error = PTR_ERR(sdp->sd_inum_inode);
-               fs_err(sdp, "can't read in inum inode: %d\n", error);
-               goto fail_journal;
-       }
-
-
        /* Read in the master statfs inode */
        sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
        if (IS_ERR(sdp->sd_statfs_inode)) {
                error = PTR_ERR(sdp->sd_statfs_inode);
                fs_err(sdp, "can't read in statfs inode: %d\n", error);
-               goto fail_inum;
+               goto fail_journal;
        }
 
        /* Read in the resource index inode */
@@ -892,8 +865,6 @@ fail_rindex:
        iput(sdp->sd_rindex);
 fail_statfs:
        iput(sdp->sd_statfs_inode);
-fail_inum:
-       iput(sdp->sd_inum_inode);
 fail_journal:
        init_journal(sdp, UNDO);
 fail:
@@ -921,20 +892,12 @@ static int init_per_node(struct gfs2_sbd *sdp, int undo)
                return error;
        }
 
-       sprintf(buf, "inum_range%u", sdp->sd_jdesc->jd_jid);
-       sdp->sd_ir_inode = gfs2_lookup_simple(pn, buf);
-       if (IS_ERR(sdp->sd_ir_inode)) {
-               error = PTR_ERR(sdp->sd_ir_inode);
-               fs_err(sdp, "can't find local \"ir\" file: %d\n", error);
-               goto fail;
-       }
-
        sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
        sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
        if (IS_ERR(sdp->sd_sc_inode)) {
                error = PTR_ERR(sdp->sd_sc_inode);
                fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
-               goto fail_ir_i;
+               goto fail;
        }
 
        sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
@@ -948,27 +911,16 @@ static int init_per_node(struct gfs2_sbd *sdp, int undo)
        iput(pn);
        pn = NULL;
 
-       ip = GFS2_I(sdp->sd_ir_inode);
-       error = gfs2_glock_nq_init(ip->i_gl,
-                                  LM_ST_EXCLUSIVE, 0,
-                                  &sdp->sd_ir_gh);
-       if (error) {
-               fs_err(sdp, "can't lock local \"ir\" file: %d\n", error);
-               goto fail_qc_i;
-       }
-
        ip = GFS2_I(sdp->sd_sc_inode);
-       error = gfs2_glock_nq_init(ip->i_gl,
-                                  LM_ST_EXCLUSIVE, 0,
+       error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
                                   &sdp->sd_sc_gh);
        if (error) {
                fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
-               goto fail_ir_gh;
+               goto fail_qc_i;
        }
 
        ip = GFS2_I(sdp->sd_qc_inode);
-       error = gfs2_glock_nq_init(ip->i_gl,
-                                  LM_ST_EXCLUSIVE, 0,
+       error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
                                   &sdp->sd_qc_gh);
        if (error) {
                fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
@@ -981,14 +933,10 @@ fail_qc_gh:
        gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
 fail_ut_gh:
        gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
-fail_ir_gh:
-       gfs2_glock_dq_uninit(&sdp->sd_ir_gh);
 fail_qc_i:
        iput(sdp->sd_qc_inode);
 fail_ut_i:
        iput(sdp->sd_sc_inode);
-fail_ir_i:
-       iput(sdp->sd_ir_inode);
 fail:
        if (pn)
                iput(pn);
@@ -1004,7 +952,6 @@ static int init_threads(struct gfs2_sbd *sdp, int undo)
                goto fail_quotad;
 
        sdp->sd_log_flush_time = jiffies;
-       sdp->sd_jindex_refresh_time = jiffies;
 
        p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
        error = IS_ERR(p);
@@ -1032,10 +979,29 @@ fail:
        return error;
 }
 
+static const match_table_t nolock_tokens = {
+       { Opt_jid, "jid=%d\n", },
+       { Opt_err, NULL },
+};
+
+static void nolock_put_lock(struct kmem_cache *cachep, struct gfs2_glock *gl)
+{
+       struct gfs2_sbd *sdp = gl->gl_sbd;
+       kmem_cache_free(cachep, gl);
+       if (atomic_dec_and_test(&sdp->sd_glock_disposal))
+               wake_up(&sdp->sd_glock_wait);
+}
+
+static const struct lm_lockops nolock_ops = {
+       .lm_proto_name = "lock_nolock",
+       .lm_put_lock = nolock_put_lock,
+       .lm_tokens = &nolock_tokens,
+};
+
 /**
  * gfs2_lm_mount - mount a locking protocol
  * @sdp: the filesystem
- * @args: mount arguements
+ * @args: mount arguments
  * @silent: if 1, don't complain if the FS isn't a GFS2 fs
  *
  * Returns: errno
@@ -1043,31 +1009,69 @@ fail:
 
 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
 {
-       char *proto = sdp->sd_proto_name;
-       char *table = sdp->sd_table_name;
-       int flags = LM_MFLAG_CONV_NODROP;
-       int error;
+       const struct lm_lockops *lm;
+       struct lm_lockstruct *ls = &sdp->sd_lockstruct;
+       struct gfs2_args *args = &sdp->sd_args;
+       const char *proto = sdp->sd_proto_name;
+       const char *table = sdp->sd_table_name;
+       const char *fsname;
+       char *o, *options;
+       int ret;
 
-       if (sdp->sd_args.ar_spectator)
-               flags |= LM_MFLAG_SPECTATOR;
+       if (!strcmp("lock_nolock", proto)) {
+               lm = &nolock_ops;
+               sdp->sd_args.ar_localflocks = 1;
+               sdp->sd_args.ar_localcaching = 1;
+#ifdef CONFIG_GFS2_FS_LOCKING_DLM
+       } else if (!strcmp("lock_dlm", proto)) {
+               lm = &gfs2_dlm_ops;
+#endif
+       } else {
+               printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
+               return -ENOENT;
+       }
 
        fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
 
-       error = gfs2_mount_lockproto(proto, table, sdp->sd_args.ar_hostdata,
-                                    gfs2_glock_cb, sdp,
-                                    GFS2_MIN_LVB_SIZE, flags,
-                                    &sdp->sd_lockstruct, &sdp->sd_kobj);
-       if (error) {
-               fs_info(sdp, "can't mount proto=%s, table=%s, hostdata=%s\n",
-                       proto, table, sdp->sd_args.ar_hostdata);
-               goto out;
-       }
+       ls->ls_ops = lm;
+       ls->ls_first = 1;
 
-       if (gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_ops) ||
-           gfs2_assert_warn(sdp, sdp->sd_lockstruct.ls_lvb_size >=
-                                 GFS2_MIN_LVB_SIZE)) {
-               gfs2_unmount_lockproto(&sdp->sd_lockstruct);
-               goto out;
+       for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
+               substring_t tmp[MAX_OPT_ARGS];
+               int token, option;
+
+               if (!o || !*o)
+                       continue;
+
+               token = match_token(o, *lm->lm_tokens, tmp);
+               switch (token) {
+               case Opt_jid:
+                       ret = match_int(&tmp[0], &option);
+                       if (ret || option < 0) 
+                               goto hostdata_error;
+                       ls->ls_jid = option;
+                       break;
+               case Opt_id:
+                       /* Obsolete, but left for backward compat purposes */
+                       break;
+               case Opt_first:
+                       ret = match_int(&tmp[0], &option);
+                       if (ret || (option != 0 && option != 1))
+                               goto hostdata_error;
+                       ls->ls_first = option;
+                       break;
+               case Opt_nodir:
+                       ret = match_int(&tmp[0], &option);
+                       if (ret || (option != 0 && option != 1))
+                               goto hostdata_error;
+                       ls->ls_nodir = option;
+                       break;
+               case Opt_err:
+               default:
+hostdata_error:
+                       fs_info(sdp, "unknown hostdata (%s)\n", o);
+                       return -EINVAL;
+               }
        }
 
        if (sdp->sd_args.ar_spectator)
@@ -1076,22 +1080,36 @@ static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
                snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
                         sdp->sd_lockstruct.ls_jid);
 
-       fs_info(sdp, "Joined cluster. Now mounting FS...\n");
-
-       if ((sdp->sd_lockstruct.ls_flags & LM_LSFLAG_LOCAL) &&
-           !sdp->sd_args.ar_ignore_local_fs) {
-               sdp->sd_args.ar_localflocks = 1;
-               sdp->sd_args.ar_localcaching = 1;
+       fsname = strchr(table, ':');
+       if (fsname)
+               fsname++;
+       if (lm->lm_mount == NULL) {
+               fs_info(sdp, "Now mounting FS...\n");
+               return 0;
        }
-
-out:
-       return error;
+       ret = lm->lm_mount(sdp, fsname);
+       if (ret == 0)
+               fs_info(sdp, "Joined cluster. Now mounting FS...\n");
+       return ret;
 }
 
 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
 {
-       if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
-               gfs2_unmount_lockproto(&sdp->sd_lockstruct);
+       const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
+       if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
+           lm->lm_unmount)
+               lm->lm_unmount(sdp);
+}
+
+void gfs2_online_uevent(struct gfs2_sbd *sdp)
+{
+       struct super_block *sb = sdp->sd_vfs;
+       char ro[20];
+       char spectator[20];
+       char *envp[] = { ro, spectator, NULL };
+       sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
+       sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
+       kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
 }
 
 /**
@@ -1103,7 +1121,7 @@ void gfs2_lm_unmount(struct gfs2_sbd *sdp)
  * Returns: errno
  */
 
-static int fill_super(struct super_block *sb, void *data, int silent)
+static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
 {
        struct gfs2_sbd *sdp;
        struct gfs2_holder mount_gh;
@@ -1114,24 +1132,23 @@ static int fill_super(struct super_block *sb, void *data, int silent)
                printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
                return -ENOMEM;
        }
+       sdp->sd_args = *args;
 
-       sdp->sd_args.ar_quota = GFS2_QUOTA_DEFAULT;
-       sdp->sd_args.ar_data = GFS2_DATA_DEFAULT;
-
-       error = gfs2_mount_args(sdp, &sdp->sd_args, data);
-       if (error) {
-               printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
-               goto fail;
-       }
-
-       if (sdp->sd_args.ar_spectator)
+       if (sdp->sd_args.ar_spectator) {
                 sb->s_flags |= MS_RDONLY;
+               set_bit(SDF_NORECOVERY, &sdp->sd_flags);
+       }
        if (sdp->sd_args.ar_posix_acl)
                sb->s_flags |= MS_POSIXACL;
+       if (sdp->sd_args.ar_nobarrier)
+               set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
 
        sb->s_magic = GFS2_MAGIC;
        sb->s_op = &gfs2_super_ops;
        sb->s_export_op = &gfs2_export_ops;
+       sb->s_xattr = gfs2_xattr_handlers;
+       sb->s_qcop = &gfs2_quotactl_ops;
+       sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
        sb->s_time_gran = 1;
        sb->s_maxbytes = MAX_LFS_FILESIZE;
 
@@ -1143,6 +1160,17 @@ static int fill_super(struct super_block *sb, void *data, int silent)
                                GFS2_BASIC_BLOCK_SHIFT;
        sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
 
+       sdp->sd_tune.gt_log_flush_secs = sdp->sd_args.ar_commit;
+       sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
+       if (sdp->sd_args.ar_statfs_quantum) {
+               sdp->sd_tune.gt_statfs_slow = 0;
+               sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
+       }
+       else {
+               sdp->sd_tune.gt_statfs_slow = 1;
+               sdp->sd_tune.gt_statfs_quantum = 30;
+       }
+
        error = init_names(sdp, silent);
        if (error)
                goto fail;
@@ -1192,7 +1220,7 @@ static int fill_super(struct super_block *sb, void *data, int silent)
        }
 
        gfs2_glock_dq_uninit(&mount_gh);
-
+       gfs2_online_uevent(sdp);
        return 0;
 
 fail_threads:
@@ -1206,14 +1234,15 @@ fail_sb:
                dput(sdp->sd_root_dir);
        if (sdp->sd_master_dir)
                dput(sdp->sd_master_dir);
+       if (sb->s_root)
+               dput(sb->s_root);
        sb->s_root = NULL;
 fail_locking:
        init_locking(sdp, &mount_gh, UNDO);
 fail_lm:
+       invalidate_inodes(sb);
        gfs2_gl_hash_clear(sdp);
        gfs2_lm_unmount(sdp);
-       while (invalidate_inodes(sb))
-               yield();
 fail_sys:
        gfs2_sys_fs_del(sdp);
 fail:
@@ -1223,46 +1252,159 @@ fail:
        return error;
 }
 
-static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
-                      const char *dev_name, void *data, struct vfsmount *mnt)
+static int set_gfs2_super(struct super_block *s, void *data)
 {
-       return get_sb_bdev(fs_type, flags, dev_name, data, fill_super, mnt);
+       s->s_bdev = data;
+       s->s_dev = s->s_bdev->bd_dev;
+
+       /*
+        * We set the bdi here to the queue backing, file systems can
+        * overwrite this in ->fill_super()
+        */
+       s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
+       return 0;
 }
 
-static struct super_block *get_gfs2_sb(const char *dev_name)
+static int test_gfs2_super(struct super_block *s, void *ptr)
+{
+       struct block_device *bdev = ptr;
+       return (bdev == s->s_bdev);
+}
+
+/**
+ * gfs2_get_sb - Get the GFS2 superblock
+ * @fs_type: The GFS2 filesystem type
+ * @flags: Mount flags
+ * @dev_name: The name of the device
+ * @data: The mount arguments
+ * @mnt: The vfsmnt for this mount
+ *
+ * Q. Why not use get_sb_bdev() ?
+ * A. We need to select one of two root directories to mount, independent
+ *    of whether this is the initial, or subsequent, mount of this sb
+ *
+ * Returns: 0 or -ve on error
+ */
+
+static int gfs2_get_sb(struct file_system_type *fs_type, int flags,
+                      const char *dev_name, void *data, struct vfsmount *mnt)
 {
-       struct super_block *sb;
-       struct nameidata nd;
+       struct block_device *bdev;
+       struct super_block *s;
+       fmode_t mode = FMODE_READ;
        int error;
+       struct gfs2_args args;
+       struct gfs2_sbd *sdp;
 
-       error = path_lookup(dev_name, LOOKUP_FOLLOW, &nd);
+       if (!(flags & MS_RDONLY))
+               mode |= FMODE_WRITE;
+
+       bdev = open_bdev_exclusive(dev_name, mode, fs_type);
+       if (IS_ERR(bdev))
+               return PTR_ERR(bdev);
+
+       /*
+        * once the super is inserted into the list by sget, s_umount
+        * will protect the lockfs code from trying to start a snapshot
+        * while we are mounting
+        */
+       mutex_lock(&bdev->bd_fsfreeze_mutex);
+       if (bdev->bd_fsfreeze_count > 0) {
+               mutex_unlock(&bdev->bd_fsfreeze_mutex);
+               error = -EBUSY;
+               goto error_bdev;
+       }
+       s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
+       mutex_unlock(&bdev->bd_fsfreeze_mutex);
+       error = PTR_ERR(s);
+       if (IS_ERR(s))
+               goto error_bdev;
+
+       memset(&args, 0, sizeof(args));
+       args.ar_quota = GFS2_QUOTA_DEFAULT;
+       args.ar_data = GFS2_DATA_DEFAULT;
+       args.ar_commit = 60;
+       args.ar_statfs_quantum = 30;
+       args.ar_quota_quantum = 60;
+       args.ar_errors = GFS2_ERRORS_DEFAULT;
+
+       error = gfs2_mount_args(&args, data);
        if (error) {
-               printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
-                      dev_name, error);
-               return NULL;
+               printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
+               if (s->s_root)
+                       goto error_super;
+               deactivate_locked_super(s);
+               return error;
+       }
+
+       if (s->s_root) {
+               error = -EBUSY;
+               if ((flags ^ s->s_flags) & MS_RDONLY)
+                       goto error_super;
+               close_bdev_exclusive(bdev, mode);
+       } else {
+               char b[BDEVNAME_SIZE];
+
+               s->s_flags = flags;
+               s->s_mode = mode;
+               strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
+               sb_set_blocksize(s, block_size(bdev));
+               error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
+               if (error) {
+                       deactivate_locked_super(s);
+                       return error;
+               }
+               s->s_flags |= MS_ACTIVE;
+               bdev->bd_super = s;
        }
-       sb = nd.path.dentry->d_inode->i_sb;
-       if (sb && (sb->s_type == &gfs2_fs_type))
-               atomic_inc(&sb->s_active);
+
+       sdp = s->s_fs_info;
+       mnt->mnt_sb = s;
+       if (args.ar_meta)
+               mnt->mnt_root = dget(sdp->sd_master_dir);
        else
-               sb = NULL;
-       path_put(&nd.path);
-       return sb;
+               mnt->mnt_root = dget(sdp->sd_root_dir);
+       return 0;
+
+error_super:
+       deactivate_locked_super(s);
+error_bdev:
+       close_bdev_exclusive(bdev, mode);
+       return error;
+}
+
+static int set_meta_super(struct super_block *s, void *ptr)
+{
+       return -EINVAL;
 }
 
 static int gfs2_get_sb_meta(struct file_system_type *fs_type, int flags,
                            const char *dev_name, void *data, struct vfsmount *mnt)
 {
-       struct super_block *sb = NULL;
+       struct super_block *s;
        struct gfs2_sbd *sdp;
+       struct path path;
+       int error;
 
-       sb = get_gfs2_sb(dev_name);
-       if (!sb) {
+       error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
+       if (error) {
+               printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
+                      dev_name, error);
+               return error;
+       }
+       s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
+                path.dentry->d_inode->i_sb->s_bdev);
+       path_put(&path);
+       if (IS_ERR(s)) {
                printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
-               return -ENOENT;
+               return PTR_ERR(s);
+       }
+       if ((flags ^ s->s_flags) & MS_RDONLY) {
+               deactivate_locked_super(s);
+               return -EBUSY;
        }
-       sdp = sb->s_fs_info;
-       mnt->mnt_sb = sb;
+       sdp = s->s_fs_info;
+       mnt->mnt_sb = s;
        mnt->mnt_root = dget(sdp->sd_master_dir);
        return 0;
 }