wake up s_wait_unfrozen when ->freeze_fs fails
[linux-2.6.git] / fs / super.c
index aff046b..6015c02 100644 (file)
 
 #include <linux/module.h>
 #include <linux/slab.h>
-#include <linux/init.h>
-#include <linux/smp_lock.h>
 #include <linux/acct.h>
 #include <linux/blkdev.h>
-#include <linux/quotaops.h>
-#include <linux/namei.h>
 #include <linux/mount.h>
 #include <linux/security.h>
-#include <linux/syscalls.h>
-#include <linux/vfs.h>
 #include <linux/writeback.h>           /* for the emergency remount stuff */
 #include <linux/idr.h>
-#include <linux/kobject.h>
 #include <linux/mutex.h>
-#include <linux/file.h>
-#include <asm/uaccess.h>
+#include <linux/backing-dev.h>
+#include <linux/rculist_bl.h>
+#include <linux/cleancache.h>
 #include "internal.h"
 
 
 LIST_HEAD(super_blocks);
 DEFINE_SPINLOCK(sb_lock);
 
+/*
+ * One thing we have to be careful of with a per-sb shrinker is that we don't
+ * drop the last active reference to the superblock from within the shrinker.
+ * If that happens we could trigger unregistering the shrinker from within the
+ * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
+ * take a passive reference to the superblock to avoid this from occurring.
+ */
+static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
+{
+       struct super_block *sb;
+       int     fs_objects = 0;
+       int     total_objects;
+
+       sb = container_of(shrink, struct super_block, s_shrink);
+
+       /*
+        * Deadlock avoidance.  We may hold various FS locks, and we don't want
+        * to recurse into the FS that called us in clear_inode() and friends..
+        */
+       if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
+               return -1;
+
+       if (!grab_super_passive(sb))
+               return !sc->nr_to_scan ? 0 : -1;
+
+       if (sb->s_op && sb->s_op->nr_cached_objects)
+               fs_objects = sb->s_op->nr_cached_objects(sb);
+
+       total_objects = sb->s_nr_dentry_unused +
+                       sb->s_nr_inodes_unused + fs_objects + 1;
+
+       if (sc->nr_to_scan) {
+               int     dentries;
+               int     inodes;
+
+               /* proportion the scan between the caches */
+               dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
+                                                       total_objects;
+               inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
+                                                       total_objects;
+               if (fs_objects)
+                       fs_objects = (sc->nr_to_scan * fs_objects) /
+                                                       total_objects;
+               /*
+                * prune the dcache first as the icache is pinned by it, then
+                * prune the icache, followed by the filesystem specific caches
+                */
+               prune_dcache_sb(sb, dentries);
+               prune_icache_sb(sb, inodes);
+
+               if (fs_objects && sb->s_op->free_cached_objects) {
+                       sb->s_op->free_cached_objects(sb, fs_objects);
+                       fs_objects = sb->s_op->nr_cached_objects(sb);
+               }
+               total_objects = sb->s_nr_dentry_unused +
+                               sb->s_nr_inodes_unused + fs_objects;
+       }
+
+       total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
+       drop_super(sb);
+       return total_objects;
+}
+
 /**
  *     alloc_super     -       create new superblock
  *     @type:  filesystem type superblock should belong to
@@ -62,11 +119,30 @@ static struct super_block *alloc_super(struct file_system_type *type)
                        s = NULL;
                        goto out;
                }
+#ifdef CONFIG_SMP
+               s->s_files = alloc_percpu(struct list_head);
+               if (!s->s_files) {
+                       security_sb_free(s);
+                       kfree(s);
+                       s = NULL;
+                       goto out;
+               } else {
+                       int i;
+
+                       for_each_possible_cpu(i)
+                               INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
+               }
+#else
                INIT_LIST_HEAD(&s->s_files);
-               INIT_LIST_HEAD(&s->s_instances);
-               INIT_HLIST_HEAD(&s->s_anon);
+#endif
+               s->s_bdi = &default_backing_dev_info;
+               INIT_HLIST_NODE(&s->s_instances);
+               INIT_HLIST_BL_HEAD(&s->s_anon);
                INIT_LIST_HEAD(&s->s_inodes);
                INIT_LIST_HEAD(&s->s_dentry_lru);
+               INIT_LIST_HEAD(&s->s_inode_lru);
+               spin_lock_init(&s->s_inode_lru_lock);
+               INIT_LIST_HEAD(&s->s_mounts);
                init_rwsem(&s->s_umount);
                mutex_init(&s->s_lock);
                lockdep_set_class(&s->s_umount, &type->s_umount_key);
@@ -92,18 +168,22 @@ static struct super_block *alloc_super(struct file_system_type *type)
                 * subclass.
                 */
                down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
-               s->s_count = S_BIAS;
+               s->s_count = 1;
                atomic_set(&s->s_active, 1);
                mutex_init(&s->s_vfs_rename_mutex);
+               lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
                mutex_init(&s->s_dquot.dqio_mutex);
                mutex_init(&s->s_dquot.dqonoff_mutex);
                init_rwsem(&s->s_dquot.dqptr_sem);
                init_waitqueue_head(&s->s_wait_unfrozen);
                s->s_maxbytes = MAX_NON_LFS;
-               s->dq_op = sb_dquot_ops;
-               s->s_qcop = sb_quotactl_ops;
                s->s_op = &default_op;
                s->s_time_gran = 1000000000;
+               s->cleancache_poolid = -1;
+
+               s->s_shrink.seeks = DEFAULT_SEEKS;
+               s->s_shrink.shrink = prune_super;
+               s->s_shrink.batch = 1024;
        }
 out:
        return s;
@@ -117,7 +197,11 @@ out:
  */
 static inline void destroy_super(struct super_block *s)
 {
+#ifdef CONFIG_SMP
+       free_percpu(s->s_files);
+#endif
        security_sb_free(s);
+       WARN_ON(!list_empty(&s->s_mounts));
        kfree(s->s_subtype);
        kfree(s->s_options);
        kfree(s);
@@ -126,39 +210,14 @@ static inline void destroy_super(struct super_block *s)
 /* Superblock refcounting  */
 
 /*
- * Drop a superblock's refcount.  Returns non-zero if the superblock was
- * destroyed.  The caller must hold sb_lock.
+ * Drop a superblock's refcount.  The caller must hold sb_lock.
  */
-static int __put_super(struct super_block *sb)
+static void __put_super(struct super_block *sb)
 {
-       int ret = 0;
-
        if (!--sb->s_count) {
+               list_del_init(&sb->s_list);
                destroy_super(sb);
-               ret = 1;
        }
-       return ret;
-}
-
-/*
- * Drop a superblock's refcount.
- * Returns non-zero if the superblock is about to be destroyed and
- * at least is already removed from super_blocks list, so if we are
- * making a loop through super blocks then we need to restart.
- * The caller must hold sb_lock.
- */
-int __put_super_and_need_restart(struct super_block *sb)
-{
-       /* check for race with generic_shutdown_super() */
-       if (list_empty(&sb->s_list)) {
-               /* super block is removed, need to restart... */
-               __put_super(sb);
-               return 1;
-       }
-       /* can't be the last, since s_list is still in use */
-       sb->s_count--;
-       BUG_ON(sb->s_count == 0);
-       return 0;
 }
 
 /**
@@ -168,7 +227,7 @@ int __put_super_and_need_restart(struct super_block *sb)
  *     Drops a temporary reference, frees superblock if there's no
  *     references left.
  */
-void put_super(struct super_block *sb)
+static void put_super(struct super_block *sb)
 {
        spin_lock(&sb_lock);
        __put_super(sb);
@@ -177,57 +236,57 @@ void put_super(struct super_block *sb)
 
 
 /**
- *     deactivate_super        -       drop an active reference to superblock
+ *     deactivate_locked_super -       drop an active reference to superblock
  *     @s: superblock to deactivate
  *
- *     Drops an active reference to superblock, acquiring a temprory one if
- *     there is no active references left.  In that case we lock superblock,
+ *     Drops an active reference to superblock, converting it into a temprory
+ *     one if there is no other active references left.  In that case we
  *     tell fs driver to shut it down and drop the temporary reference we
  *     had just acquired.
+ *
+ *     Caller holds exclusive lock on superblock; that lock is released.
  */
-void deactivate_super(struct super_block *s)
+void deactivate_locked_super(struct super_block *s)
 {
        struct file_system_type *fs = s->s_type;
-       if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
-               s->s_count -= S_BIAS-1;
-               spin_unlock(&sb_lock);
-               vfs_dq_off(s, 0);
-               down_write(&s->s_umount);
+       if (atomic_dec_and_test(&s->s_active)) {
+               cleancache_flush_fs(s);
                fs->kill_sb(s);
+
+               /* caches are now gone, we can safely kill the shrinker now */
+               unregister_shrinker(&s->s_shrink);
+
+               /*
+                * We need to call rcu_barrier so all the delayed rcu free
+                * inodes are flushed before we release the fs module.
+                */
+               rcu_barrier();
                put_filesystem(fs);
                put_super(s);
+       } else {
+               up_write(&s->s_umount);
        }
 }
 
-EXPORT_SYMBOL(deactivate_super);
+EXPORT_SYMBOL(deactivate_locked_super);
 
 /**
- *     deactivate_locked_super -       drop an active reference to superblock
+ *     deactivate_super        -       drop an active reference to superblock
  *     @s: superblock to deactivate
  *
- *     Equivalent of up_write(&s->s_umount); deactivate_super(s);, except that
- *     it does not unlock it until it's all over.  As the result, it's safe to
- *     use to dispose of new superblock on ->get_sb() failure exits - nobody
- *     will see the sucker until it's all over.  Equivalent using up_write +
- *     deactivate_super is safe for that purpose only if superblock is either
- *     safe to use or has NULL ->s_root when we unlock.
+ *     Variant of deactivate_locked_super(), except that superblock is *not*
+ *     locked by caller.  If we are going to drop the final active reference,
+ *     lock will be acquired prior to that.
  */
-void deactivate_locked_super(struct super_block *s)
+void deactivate_super(struct super_block *s)
 {
-       struct file_system_type *fs = s->s_type;
-       if (atomic_dec_and_lock(&s->s_active, &sb_lock)) {
-               s->s_count -= S_BIAS-1;
-               spin_unlock(&sb_lock);
-               vfs_dq_off(s, 0);
-               fs->kill_sb(s);
-               put_filesystem(fs);
-               put_super(s);
-       } else {
-               up_write(&s->s_umount);
+        if (!atomic_add_unless(&s->s_active, -1, 1)) {
+               down_write(&s->s_umount);
+               deactivate_locked_super(s);
        }
 }
 
-EXPORT_SYMBOL(deactivate_locked_super);
+EXPORT_SYMBOL(deactivate_super);
 
 /**
  *     grab_super - acquire an active reference
@@ -242,37 +301,63 @@ EXPORT_SYMBOL(deactivate_locked_super);
  */
 static int grab_super(struct super_block *s) __releases(sb_lock)
 {
+       if (atomic_inc_not_zero(&s->s_active)) {
+               spin_unlock(&sb_lock);
+               return 1;
+       }
+       /* it's going away */
        s->s_count++;
        spin_unlock(&sb_lock);
+       /* wait for it to die */
        down_write(&s->s_umount);
-       if (s->s_root) {
-               spin_lock(&sb_lock);
-               if (s->s_count > S_BIAS) {
-                       atomic_inc(&s->s_active);
-                       s->s_count--;
-                       spin_unlock(&sb_lock);
-                       return 1;
-               }
-               spin_unlock(&sb_lock);
-       }
        up_write(&s->s_umount);
        put_super(s);
-       yield();
        return 0;
 }
 
 /*
+ *     grab_super_passive - acquire a passive reference
+ *     @s: reference we are trying to grab
+ *
+ *     Tries to acquire a passive reference. This is used in places where we
+ *     cannot take an active reference but we need to ensure that the
+ *     superblock does not go away while we are working on it. It returns
+ *     false if a reference was not gained, and returns true with the s_umount
+ *     lock held in read mode if a reference is gained. On successful return,
+ *     the caller must drop the s_umount lock and the passive reference when
+ *     done.
+ */
+bool grab_super_passive(struct super_block *sb)
+{
+       spin_lock(&sb_lock);
+       if (hlist_unhashed(&sb->s_instances)) {
+               spin_unlock(&sb_lock);
+               return false;
+       }
+
+       sb->s_count++;
+       spin_unlock(&sb_lock);
+
+       if (down_read_trylock(&sb->s_umount)) {
+               if (sb->s_root && (sb->s_flags & MS_BORN))
+                       return true;
+               up_read(&sb->s_umount);
+       }
+
+       put_super(sb);
+       return false;
+}
+
+/*
  * Superblock locking.  We really ought to get rid of these two.
  */
 void lock_super(struct super_block * sb)
 {
-       get_fs_excl();
        mutex_lock(&sb->s_lock);
 }
 
 void unlock_super(struct super_block * sb)
 {
-       put_fs_excl();
        mutex_unlock(&sb->s_lock);
 }
 
@@ -297,31 +382,27 @@ void generic_shutdown_super(struct super_block *sb)
 {
        const struct super_operations *sop = sb->s_op;
 
-
        if (sb->s_root) {
                shrink_dcache_for_umount(sb);
                sync_filesystem(sb);
-               get_fs_excl();
                sb->s_flags &= ~MS_ACTIVE;
 
-               /* bad name - it should be evict_inodes() */
-               invalidate_inodes(sb);
+               fsnotify_unmount_inodes(&sb->s_inodes);
+
+               evict_inodes(sb);
 
                if (sop->put_super)
                        sop->put_super(sb);
 
-               /* Forget any remaining inodes */
-               if (invalidate_inodes(sb)) {
+               if (!list_empty(&sb->s_inodes)) {
                        printk("VFS: Busy inodes after unmount of %s. "
                           "Self-destruct in 5 seconds.  Have a nice day...\n",
                           sb->s_id);
                }
-               put_fs_excl();
        }
        spin_lock(&sb_lock);
        /* should be initialized for __put_super_and_need_restart() */
-       list_del_init(&sb->s_list);
-       list_del(&sb->s_instances);
+       hlist_del_init(&sb->s_instances);
        spin_unlock(&sb_lock);
        up_write(&sb->s_umount);
 }
@@ -341,13 +422,14 @@ struct super_block *sget(struct file_system_type *type,
                        void *data)
 {
        struct super_block *s = NULL;
+       struct hlist_node *node;
        struct super_block *old;
        int err;
 
 retry:
        spin_lock(&sb_lock);
        if (test) {
-               list_for_each_entry(old, &type->fs_supers, s_instances) {
+               hlist_for_each_entry(old, node, &type->fs_supers, s_instances) {
                        if (!test(old, data))
                                continue;
                        if (!grab_super(old))
@@ -355,6 +437,12 @@ retry:
                        if (s) {
                                up_write(&s->s_umount);
                                destroy_super(s);
+                               s = NULL;
+                       }
+                       down_write(&old->s_umount);
+                       if (unlikely(!(old->s_flags & MS_BORN))) {
+                               deactivate_locked_super(old);
+                               goto retry;
                        }
                        return old;
                }
@@ -377,9 +465,10 @@ retry:
        s->s_type = type;
        strlcpy(s->s_id, type->name, sizeof(s->s_id));
        list_add_tail(&s->s_list, &super_blocks);
-       list_add(&s->s_instances, &type->fs_supers);
+       hlist_add_head(&s->s_instances, &type->fs_supers);
        spin_unlock(&sb_lock);
        get_filesystem(type);
+       register_shrinker(&s->s_shrink);
        return s;
 }
 
@@ -407,28 +496,103 @@ EXPORT_SYMBOL(drop_super);
  */
 void sync_supers(void)
 {
-       struct super_block *sb;
+       struct super_block *sb, *p = NULL;
 
        spin_lock(&sb_lock);
-restart:
        list_for_each_entry(sb, &super_blocks, s_list) {
+               if (hlist_unhashed(&sb->s_instances))
+                       continue;
                if (sb->s_op->write_super && sb->s_dirt) {
                        sb->s_count++;
                        spin_unlock(&sb_lock);
 
                        down_read(&sb->s_umount);
-                       if (sb->s_root && sb->s_dirt)
+                       if (sb->s_root && sb->s_dirt && (sb->s_flags & MS_BORN))
                                sb->s_op->write_super(sb);
                        up_read(&sb->s_umount);
 
                        spin_lock(&sb_lock);
-                       if (__put_super_and_need_restart(sb))
-                               goto restart;
+                       if (p)
+                               __put_super(p);
+                       p = sb;
                }
        }
+       if (p)
+               __put_super(p);
+       spin_unlock(&sb_lock);
+}
+
+/**
+ *     iterate_supers - call function for all active superblocks
+ *     @f: function to call
+ *     @arg: argument to pass to it
+ *
+ *     Scans the superblock list and calls given function, passing it
+ *     locked superblock and given argument.
+ */
+void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
+{
+       struct super_block *sb, *p = NULL;
+
+       spin_lock(&sb_lock);
+       list_for_each_entry(sb, &super_blocks, s_list) {
+               if (hlist_unhashed(&sb->s_instances))
+                       continue;
+               sb->s_count++;
+               spin_unlock(&sb_lock);
+
+               down_read(&sb->s_umount);
+               if (sb->s_root && (sb->s_flags & MS_BORN))
+                       f(sb, arg);
+               up_read(&sb->s_umount);
+
+               spin_lock(&sb_lock);
+               if (p)
+                       __put_super(p);
+               p = sb;
+       }
+       if (p)
+               __put_super(p);
+       spin_unlock(&sb_lock);
+}
+
+/**
+ *     iterate_supers_type - call function for superblocks of given type
+ *     @type: fs type
+ *     @f: function to call
+ *     @arg: argument to pass to it
+ *
+ *     Scans the superblock list and calls given function, passing it
+ *     locked superblock and given argument.
+ */
+void iterate_supers_type(struct file_system_type *type,
+       void (*f)(struct super_block *, void *), void *arg)
+{
+       struct super_block *sb, *p = NULL;
+       struct hlist_node *node;
+
+       spin_lock(&sb_lock);
+       hlist_for_each_entry(sb, node, &type->fs_supers, s_instances) {
+               sb->s_count++;
+               spin_unlock(&sb_lock);
+
+               down_read(&sb->s_umount);
+               if (sb->s_root && (sb->s_flags & MS_BORN))
+                       f(sb, arg);
+               up_read(&sb->s_umount);
+
+               spin_lock(&sb_lock);
+               if (p)
+                       __put_super(p);
+               p = sb;
+       }
+       if (p)
+               __put_super(p);
        spin_unlock(&sb_lock);
 }
 
+EXPORT_SYMBOL(iterate_supers_type);
+
 /**
  *     get_super - get the superblock of a device
  *     @bdev: device to get the superblock for
@@ -437,7 +601,7 @@ restart:
  *     mounted on the device given. %NULL is returned if no match is found.
  */
 
-struct super_block * get_super(struct block_device *bdev)
+struct super_block *get_super(struct block_device *bdev)
 {
        struct super_block *sb;
 
@@ -447,17 +611,20 @@ struct super_block * get_super(struct block_device *bdev)
        spin_lock(&sb_lock);
 rescan:
        list_for_each_entry(sb, &super_blocks, s_list) {
+               if (hlist_unhashed(&sb->s_instances))
+                       continue;
                if (sb->s_bdev == bdev) {
                        sb->s_count++;
                        spin_unlock(&sb_lock);
                        down_read(&sb->s_umount);
-                       if (sb->s_root)
+                       /* still alive? */
+                       if (sb->s_root && (sb->s_flags & MS_BORN))
                                return sb;
                        up_read(&sb->s_umount);
-                       /* restart only when sb is no longer on the list */
+                       /* nope, got unmounted */
                        spin_lock(&sb_lock);
-                       if (__put_super_and_need_restart(sb))
-                               goto rescan;
+                       __put_super(sb);
+                       goto rescan;
                }
        }
        spin_unlock(&sb_lock);
@@ -472,7 +639,7 @@ EXPORT_SYMBOL(get_super);
  *
  * Scans the superblock list and finds the superblock of the file system
  * mounted on the device given.  Returns the superblock with an active
- * reference and s_umount held exclusively or %NULL if none was found.
+ * reference or %NULL if none was found.
  */
 struct super_block *get_active_super(struct block_device *bdev)
 {
@@ -481,81 +648,49 @@ struct super_block *get_active_super(struct block_device *bdev)
        if (!bdev)
                return NULL;
 
+restart:
        spin_lock(&sb_lock);
        list_for_each_entry(sb, &super_blocks, s_list) {
-               if (sb->s_bdev != bdev)
+               if (hlist_unhashed(&sb->s_instances))
                        continue;
-
-               sb->s_count++;
-               spin_unlock(&sb_lock);
-               down_write(&sb->s_umount);
-               if (sb->s_root) {
-                       spin_lock(&sb_lock);
-                       if (sb->s_count > S_BIAS) {
-                               atomic_inc(&sb->s_active);
-                               sb->s_count--;
-                               spin_unlock(&sb_lock);
+               if (sb->s_bdev == bdev) {
+                       if (grab_super(sb)) /* drops sb_lock */
                                return sb;
-                       }
-                       spin_unlock(&sb_lock);
+                       else
+                               goto restart;
                }
-               up_write(&sb->s_umount);
-               put_super(sb);
-               yield();
-               spin_lock(&sb_lock);
        }
        spin_unlock(&sb_lock);
        return NULL;
 }
  
-struct super_block * user_get_super(dev_t dev)
+struct super_block *user_get_super(dev_t dev)
 {
        struct super_block *sb;
 
        spin_lock(&sb_lock);
 rescan:
        list_for_each_entry(sb, &super_blocks, s_list) {
+               if (hlist_unhashed(&sb->s_instances))
+                       continue;
                if (sb->s_dev ==  dev) {
                        sb->s_count++;
                        spin_unlock(&sb_lock);
                        down_read(&sb->s_umount);
-                       if (sb->s_root)
+                       /* still alive? */
+                       if (sb->s_root && (sb->s_flags & MS_BORN))
                                return sb;
                        up_read(&sb->s_umount);
-                       /* restart only when sb is no longer on the list */
+                       /* nope, got unmounted */
                        spin_lock(&sb_lock);
-                       if (__put_super_and_need_restart(sb))
-                               goto rescan;
+                       __put_super(sb);
+                       goto rescan;
                }
        }
        spin_unlock(&sb_lock);
        return NULL;
 }
 
-SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf)
-{
-        struct super_block *s;
-        struct ustat tmp;
-        struct kstatfs sbuf;
-       int err = -EINVAL;
-
-        s = user_get_super(new_decode_dev(dev));
-        if (s == NULL)
-                goto out;
-       err = vfs_statfs(s->s_root, &sbuf);
-       drop_super(s);
-       if (err)
-               goto out;
-
-        memset(&tmp,0,sizeof(struct ustat));
-        tmp.f_tfree = sbuf.f_bfree;
-        tmp.f_tinode = sbuf.f_ffree;
-
-        err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0;
-out:
-       return err;
-}
-
 /**
  *     do_remount_sb - asks filesystem to change mount options.
  *     @sb:    superblock in question
@@ -568,7 +703,7 @@ out:
 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
 {
        int retval;
-       int remount_rw;
+       int remount_ro;
 
        if (sb->s_frozen != SB_UNFROZEN)
                return -EBUSY;
@@ -583,51 +718,78 @@ int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
        shrink_dcache_sb(sb);
        sync_filesystem(sb);
 
+       remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
+
        /* If we are remounting RDONLY and current sb is read/write,
           make sure there are no rw files opened */
-       if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) {
-               if (force)
+       if (remount_ro) {
+               if (force) {
                        mark_files_ro(sb);
-               else if (!fs_may_remount_ro(sb))
-                       return -EBUSY;
-               retval = vfs_dq_off(sb, 1);
-               if (retval < 0 && retval != -ENOSYS)
-                       return -EBUSY;
+               } else {
+                       retval = sb_prepare_remount_readonly(sb);
+                       if (retval)
+                               return retval;
+               }
        }
-       remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
 
        if (sb->s_op->remount_fs) {
                retval = sb->s_op->remount_fs(sb, &flags, data);
-               if (retval)
-                       return retval;
+               if (retval) {
+                       if (!force)
+                               goto cancel_readonly;
+                       /* If forced remount, go ahead despite any errors */
+                       WARN(1, "forced remount of a %s fs returned %i\n",
+                            sb->s_type->name, retval);
+               }
        }
        sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
-       if (remount_rw)
-               vfs_dq_quota_on_remount(sb);
+       /* Needs to be ordered wrt mnt_is_readonly() */
+       smp_wmb();
+       sb->s_readonly_remount = 0;
+
+       /*
+        * Some filesystems modify their metadata via some other path than the
+        * bdev buffer cache (eg. use a private mapping, or directories in
+        * pagecache, etc). Also file data modifications go via their own
+        * mappings. So If we try to mount readonly then copy the filesystem
+        * from bdev, we could get stale data, so invalidate it to give a best
+        * effort at coherency.
+        */
+       if (remount_ro && sb->s_bdev)
+               invalidate_bdev(sb->s_bdev);
        return 0;
+
+cancel_readonly:
+       sb->s_readonly_remount = 0;
+       return retval;
 }
 
 static void do_emergency_remount(struct work_struct *work)
 {
-       struct super_block *sb;
+       struct super_block *sb, *p = NULL;
 
        spin_lock(&sb_lock);
        list_for_each_entry(sb, &super_blocks, s_list) {
+               if (hlist_unhashed(&sb->s_instances))
+                       continue;
                sb->s_count++;
                spin_unlock(&sb_lock);
                down_write(&sb->s_umount);
-               if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
+               if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
+                   !(sb->s_flags & MS_RDONLY)) {
                        /*
-                        * ->remount_fs needs lock_kernel().
-                        *
                         * What lock protects sb->s_flags??
                         */
                        do_remount_sb(sb, MS_RDONLY, NULL, 1);
                }
                up_write(&sb->s_umount);
-               put_super(sb);
                spin_lock(&sb_lock);
+               if (p)
+                       __put_super(p);
+               p = sb;
        }
+       if (p)
+               __put_super(p);
        spin_unlock(&sb_lock);
        kfree(work);
        printk("Emergency Remount complete\n");
@@ -653,7 +815,7 @@ static DEFINE_IDA(unnamed_dev_ida);
 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
 static int unnamed_dev_start = 0; /* don't bother trying below it */
 
-int set_anon_super(struct super_block *s, void *data)
+int get_anon_bdev(dev_t *p)
 {
        int dev;
        int error;
@@ -680,23 +842,38 @@ int set_anon_super(struct super_block *s, void *data)
                spin_unlock(&unnamed_dev_lock);
                return -EMFILE;
        }
-       s->s_dev = MKDEV(0, dev & MINORMASK);
+       *p = MKDEV(0, dev & MINORMASK);
        return 0;
 }
+EXPORT_SYMBOL(get_anon_bdev);
 
-EXPORT_SYMBOL(set_anon_super);
-
-void kill_anon_super(struct super_block *sb)
+void free_anon_bdev(dev_t dev)
 {
-       int slot = MINOR(sb->s_dev);
-
-       generic_shutdown_super(sb);
+       int slot = MINOR(dev);
        spin_lock(&unnamed_dev_lock);
        ida_remove(&unnamed_dev_ida, slot);
        if (slot < unnamed_dev_start)
                unnamed_dev_start = slot;
        spin_unlock(&unnamed_dev_lock);
 }
+EXPORT_SYMBOL(free_anon_bdev);
+
+int set_anon_super(struct super_block *s, void *data)
+{
+       int error = get_anon_bdev(&s->s_dev);
+       if (!error)
+               s->s_bdi = &noop_backing_dev_info;
+       return error;
+}
+
+EXPORT_SYMBOL(set_anon_super);
+
+void kill_anon_super(struct super_block *sb)
+{
+       dev_t dev = sb->s_dev;
+       generic_shutdown_super(sb);
+       free_anon_bdev(dev);
+}
 
 EXPORT_SYMBOL(kill_anon_super);
 
@@ -720,15 +897,14 @@ static int ns_set_super(struct super_block *sb, void *data)
        return set_anon_super(sb, NULL);
 }
 
-int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
-       int (*fill_super)(struct super_block *, void *, int),
-       struct vfsmount *mnt)
+struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
+       void *data, int (*fill_super)(struct super_block *, void *, int))
 {
        struct super_block *sb;
 
        sb = sget(fs_type, ns_test_super, ns_set_super, data);
        if (IS_ERR(sb))
-               return PTR_ERR(sb);
+               return ERR_CAST(sb);
 
        if (!sb->s_root) {
                int err;
@@ -736,17 +912,16 @@ int get_sb_ns(struct file_system_type *fs_type, int flags, void *data,
                err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
                if (err) {
                        deactivate_locked_super(sb);
-                       return err;
+                       return ERR_PTR(err);
                }
 
                sb->s_flags |= MS_ACTIVE;
        }
 
-       simple_set_mnt(mnt, sb);
-       return 0;
+       return dget(sb->s_root);
 }
 
-EXPORT_SYMBOL(get_sb_ns);
+EXPORT_SYMBOL(mount_ns);
 
 #ifdef CONFIG_BLOCK
 static int set_bdev_super(struct super_block *s, void *data)
@@ -767,22 +942,21 @@ static int test_bdev_super(struct super_block *s, void *data)
        return (void *)s->s_bdev == data;
 }
 
-int get_sb_bdev(struct file_system_type *fs_type,
+struct dentry *mount_bdev(struct file_system_type *fs_type,
        int flags, const char *dev_name, void *data,
-       int (*fill_super)(struct super_block *, void *, int),
-       struct vfsmount *mnt)
+       int (*fill_super)(struct super_block *, void *, int))
 {
        struct block_device *bdev;
        struct super_block *s;
-       fmode_t mode = FMODE_READ;
+       fmode_t mode = FMODE_READ | FMODE_EXCL;
        int error = 0;
 
        if (!(flags & MS_RDONLY))
                mode |= FMODE_WRITE;
 
-       bdev = open_bdev_exclusive(dev_name, mode, fs_type);
+       bdev = blkdev_get_by_path(dev_name, mode, fs_type);
        if (IS_ERR(bdev))
-               return PTR_ERR(bdev);
+               return ERR_CAST(bdev);
 
        /*
         * once the super is inserted into the list by sget, s_umount
@@ -807,11 +981,20 @@ int get_sb_bdev(struct file_system_type *fs_type,
                        goto error_bdev;
                }
 
-               close_bdev_exclusive(bdev, mode);
+               /*
+                * s_umount nests inside bd_mutex during
+                * __invalidate_device().  blkdev_put() acquires
+                * bd_mutex and can't be called under s_umount.  Drop
+                * s_umount temporarily.  This is safe as we're
+                * holding an active reference.
+                */
+               up_write(&s->s_umount);
+               blkdev_put(bdev, mode);
+               down_write(&s->s_umount);
        } else {
                char b[BDEVNAME_SIZE];
 
-               s->s_flags = flags;
+               s->s_flags = flags | MS_NOSEC;
                s->s_mode = mode;
                strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
                sb_set_blocksize(s, block_size(bdev));
@@ -825,18 +1008,16 @@ int get_sb_bdev(struct file_system_type *fs_type,
                bdev->bd_super = s;
        }
 
-       simple_set_mnt(mnt, s);
-       return 0;
+       return dget(s->s_root);
 
 error_s:
        error = PTR_ERR(s);
 error_bdev:
-       close_bdev_exclusive(bdev, mode);
+       blkdev_put(bdev, mode);
 error:
-       return error;
+       return ERR_PTR(error);
 }
-
-EXPORT_SYMBOL(get_sb_bdev);
+EXPORT_SYMBOL(mount_bdev);
 
 void kill_block_super(struct super_block *sb)
 {
@@ -846,174 +1027,212 @@ void kill_block_super(struct super_block *sb)
        bdev->bd_super = NULL;
        generic_shutdown_super(sb);
        sync_blockdev(bdev);
-       close_bdev_exclusive(bdev, mode);
+       WARN_ON_ONCE(!(mode & FMODE_EXCL));
+       blkdev_put(bdev, mode | FMODE_EXCL);
 }
 
 EXPORT_SYMBOL(kill_block_super);
 #endif
 
-int get_sb_nodev(struct file_system_type *fs_type,
+struct dentry *mount_nodev(struct file_system_type *fs_type,
        int flags, void *data,
-       int (*fill_super)(struct super_block *, void *, int),
-       struct vfsmount *mnt)
+       int (*fill_super)(struct super_block *, void *, int))
 {
        int error;
        struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
 
        if (IS_ERR(s))
-               return PTR_ERR(s);
+               return ERR_CAST(s);
 
        s->s_flags = flags;
 
        error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
        if (error) {
                deactivate_locked_super(s);
-               return error;
+               return ERR_PTR(error);
        }
        s->s_flags |= MS_ACTIVE;
-       simple_set_mnt(mnt, s);
-       return 0;
+       return dget(s->s_root);
 }
-
-EXPORT_SYMBOL(get_sb_nodev);
+EXPORT_SYMBOL(mount_nodev);
 
 static int compare_single(struct super_block *s, void *p)
 {
        return 1;
 }
 
-int get_sb_single(struct file_system_type *fs_type,
+struct dentry *mount_single(struct file_system_type *fs_type,
        int flags, void *data,
-       int (*fill_super)(struct super_block *, void *, int),
-       struct vfsmount *mnt)
+       int (*fill_super)(struct super_block *, void *, int))
 {
        struct super_block *s;
        int error;
 
        s = sget(fs_type, compare_single, set_anon_super, NULL);
        if (IS_ERR(s))
-               return PTR_ERR(s);
+               return ERR_CAST(s);
        if (!s->s_root) {
                s->s_flags = flags;
                error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
                if (error) {
                        deactivate_locked_super(s);
-                       return error;
+                       return ERR_PTR(error);
                }
                s->s_flags |= MS_ACTIVE;
        } else {
                do_remount_sb(s, flags, data, 0);
        }
-       simple_set_mnt(mnt, s);
-       return 0;
+       return dget(s->s_root);
 }
+EXPORT_SYMBOL(mount_single);
 
-EXPORT_SYMBOL(get_sb_single);
-
-struct vfsmount *
-vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
+struct dentry *
+mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
 {
-       struct vfsmount *mnt;
+       struct dentry *root;
+       struct super_block *sb;
        char *secdata = NULL;
-       int error;
-
-       if (!type)
-               return ERR_PTR(-ENODEV);
-
-       error = -ENOMEM;
-       mnt = alloc_vfsmnt(name);
-       if (!mnt)
-               goto out;
+       int error = -ENOMEM;
 
        if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
                secdata = alloc_secdata();
                if (!secdata)
-                       goto out_mnt;
+                       goto out;
 
                error = security_sb_copy_data(data, secdata);
                if (error)
                        goto out_free_secdata;
        }
 
-       error = type->get_sb(type, flags, name, data, mnt);
-       if (error < 0)
+       root = type->mount(type, flags, name, data);
+       if (IS_ERR(root)) {
+               error = PTR_ERR(root);
                goto out_free_secdata;
-       BUG_ON(!mnt->mnt_sb);
+       }
+       sb = root->d_sb;
+       BUG_ON(!sb);
+       WARN_ON(!sb->s_bdi);
+       WARN_ON(sb->s_bdi == &default_backing_dev_info);
+       sb->s_flags |= MS_BORN;
 
-       error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata);
-       if (error)
-               goto out_sb;
+       error = security_sb_kern_mount(sb, flags, secdata);
+       if (error)
+               goto out_sb;
 
        /*
         * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
         * but s_maxbytes was an unsigned long long for many releases. Throw
         * this warning for a little while to try and catch filesystems that
-        * violate this rule. This warning should be either removed or
-        * converted to a BUG() in 2.6.34.
+        * violate this rule.
         */
-       WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
-               "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
+       WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
+               "negative value (%lld)\n", type->name, sb->s_maxbytes);
 
-       mnt->mnt_mountpoint = mnt->mnt_root;
-       mnt->mnt_parent = mnt;
-       up_write(&mnt->mnt_sb->s_umount);
+       up_write(&sb->s_umount);
        free_secdata(secdata);
-       return mnt;
+       return root;
 out_sb:
-       dput(mnt->mnt_root);
-       deactivate_locked_super(mnt->mnt_sb);
+       dput(root);
+       deactivate_locked_super(sb);
 out_free_secdata:
        free_secdata(secdata);
-out_mnt:
-       free_vfsmnt(mnt);
 out:
        return ERR_PTR(error);
 }
 
-EXPORT_SYMBOL_GPL(vfs_kern_mount);
-
-static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
+/**
+ * freeze_super - lock the filesystem and force it into a consistent state
+ * @sb: the super to lock
+ *
+ * Syncs the super to make sure the filesystem is consistent and calls the fs's
+ * freeze_fs.  Subsequent calls to this without first thawing the fs will return
+ * -EBUSY.
+ */
+int freeze_super(struct super_block *sb)
 {
-       int err;
-       const char *subtype = strchr(fstype, '.');
-       if (subtype) {
-               subtype++;
-               err = -EINVAL;
-               if (!subtype[0])
-                       goto err;
-       } else
-               subtype = "";
+       int ret;
 
-       mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
-       err = -ENOMEM;
-       if (!mnt->mnt_sb->s_subtype)
-               goto err;
-       return mnt;
+       atomic_inc(&sb->s_active);
+       down_write(&sb->s_umount);
+       if (sb->s_frozen) {
+               deactivate_locked_super(sb);
+               return -EBUSY;
+       }
 
- err:
-       mntput(mnt);
-       return ERR_PTR(err);
-}
+       if (!(sb->s_flags & MS_BORN)) {
+               up_write(&sb->s_umount);
+               return 0;       /* sic - it's "nothing to do" */
+       }
 
-struct vfsmount *
-do_kern_mount(const char *fstype, int flags, const char *name, void *data)
-{
-       struct file_system_type *type = get_fs_type(fstype);
-       struct vfsmount *mnt;
-       if (!type)
-               return ERR_PTR(-ENODEV);
-       mnt = vfs_kern_mount(type, flags, name, data);
-       if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
-           !mnt->mnt_sb->s_subtype)
-               mnt = fs_set_subtype(mnt, fstype);
-       put_filesystem(type);
-       return mnt;
+       if (sb->s_flags & MS_RDONLY) {
+               sb->s_frozen = SB_FREEZE_TRANS;
+               smp_wmb();
+               up_write(&sb->s_umount);
+               return 0;
+       }
+
+       sb->s_frozen = SB_FREEZE_WRITE;
+       smp_wmb();
+
+       sync_filesystem(sb);
+
+       sb->s_frozen = SB_FREEZE_TRANS;
+       smp_wmb();
+
+       sync_blockdev(sb->s_bdev);
+       if (sb->s_op->freeze_fs) {
+               ret = sb->s_op->freeze_fs(sb);
+               if (ret) {
+                       printk(KERN_ERR
+                               "VFS:Filesystem freeze failed\n");
+                       sb->s_frozen = SB_UNFROZEN;
+                       smp_wmb();
+                       wake_up(&sb->s_wait_unfrozen);
+                       deactivate_locked_super(sb);
+                       return ret;
+               }
+       }
+       up_write(&sb->s_umount);
+       return 0;
 }
-EXPORT_SYMBOL_GPL(do_kern_mount);
+EXPORT_SYMBOL(freeze_super);
 
-struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
+/**
+ * thaw_super -- unlock filesystem
+ * @sb: the super to thaw
+ *
+ * Unlocks the filesystem and marks it writeable again after freeze_super().
+ */
+int thaw_super(struct super_block *sb)
 {
-       return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
-}
+       int error;
+
+       down_write(&sb->s_umount);
+       if (sb->s_frozen == SB_UNFROZEN) {
+               up_write(&sb->s_umount);
+               return -EINVAL;
+       }
 
-EXPORT_SYMBOL_GPL(kern_mount_data);
+       if (sb->s_flags & MS_RDONLY)
+               goto out;
+
+       if (sb->s_op->unfreeze_fs) {
+               error = sb->s_op->unfreeze_fs(sb);
+               if (error) {
+                       printk(KERN_ERR
+                               "VFS:Filesystem thaw failed\n");
+                       sb->s_frozen = SB_FREEZE_TRANS;
+                       up_write(&sb->s_umount);
+                       return error;
+               }
+       }
+
+out:
+       sb->s_frozen = SB_UNFROZEN;
+       smp_wmb();
+       wake_up(&sb->s_wait_unfrozen);
+       deactivate_locked_super(sb);
+
+       return 0;
+}
+EXPORT_SYMBOL(thaw_super);