xfs: remove old vmap cache
[linux-2.6.git] / fs / super.c
index 0f9d17f..f35ac60 100644 (file)
@@ -28,7 +28,6 @@
 #include <linux/blkdev.h>
 #include <linux/quotaops.h>
 #include <linux/namei.h>
-#include <linux/buffer_head.h>         /* for fsync_super() */
 #include <linux/mount.h>
 #include <linux/security.h>
 #include <linux/syscalls.h>
@@ -38,7 +37,6 @@
 #include <linux/kobject.h>
 #include <linux/mutex.h>
 #include <linux/file.h>
-#include <linux/async.h>
 #include <asm/uaccess.h>
 #include "internal.h"
 
@@ -56,7 +54,7 @@ DEFINE_SPINLOCK(sb_lock);
 static struct super_block *alloc_super(struct file_system_type *type)
 {
        struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
-       static struct super_operations default_op;
+       static const struct super_operations default_op;
 
        if (s) {
                if (security_sb_alloc(s)) {
@@ -64,15 +62,11 @@ static struct super_block *alloc_super(struct file_system_type *type)
                        s = NULL;
                        goto out;
                }
-               INIT_LIST_HEAD(&s->s_dirty);
-               INIT_LIST_HEAD(&s->s_io);
-               INIT_LIST_HEAD(&s->s_more_io);
                INIT_LIST_HEAD(&s->s_files);
                INIT_LIST_HEAD(&s->s_instances);
                INIT_HLIST_HEAD(&s->s_anon);
                INIT_LIST_HEAD(&s->s_inodes);
                INIT_LIST_HEAD(&s->s_dentry_lru);
-               INIT_LIST_HEAD(&s->s_async_list);
                init_rwsem(&s->s_umount);
                mutex_init(&s->s_lock);
                lockdep_set_class(&s->s_umount, &type->s_umount_key);
@@ -174,7 +168,7 @@ int __put_super_and_need_restart(struct super_block *sb)
  *     Drops a temporary reference, frees superblock if there's no
  *     references left.
  */
-static void put_super(struct super_block *sb)
+void put_super(struct super_block *sb)
 {
        spin_lock(&sb_lock);
        __put_super(sb);
@@ -208,6 +202,34 @@ void deactivate_super(struct super_block *s)
 EXPORT_SYMBOL(deactivate_super);
 
 /**
+ *     deactivate_locked_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.
+ */
+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);
+               fs->kill_sb(s);
+               put_filesystem(fs);
+               put_super(s);
+       } else {
+               up_write(&s->s_umount);
+       }
+}
+
+EXPORT_SYMBOL(deactivate_locked_super);
+
+/**
  *     grab_super - acquire an active reference
  *     @s: reference we are trying to make active
  *
@@ -257,37 +279,6 @@ void unlock_super(struct super_block * sb)
 EXPORT_SYMBOL(lock_super);
 EXPORT_SYMBOL(unlock_super);
 
-/*
- * Write out and wait upon all dirty data associated with this
- * superblock.  Filesystem data as well as the underlying block
- * device.  Takes the superblock lock.  Requires a second blkdev
- * flush by the caller to complete the operation.
- */
-void __fsync_super(struct super_block *sb)
-{
-       sync_inodes_sb(sb, 0);
-       vfs_dq_sync(sb);
-       lock_super(sb);
-       if (sb->s_dirt && sb->s_op->write_super)
-               sb->s_op->write_super(sb);
-       unlock_super(sb);
-       if (sb->s_op->sync_fs)
-               sb->s_op->sync_fs(sb, 1);
-       sync_blockdev(sb->s_bdev);
-       sync_inodes_sb(sb, 1);
-}
-
-/*
- * Write out and wait upon all dirty data associated with this
- * superblock.  Filesystem data as well as the underlying block
- * device.  Takes the superblock lock.
- */
-int fsync_super(struct super_block *sb)
-{
-       __fsync_super(sb);
-       return sync_blockdev(sb->s_bdev);
-}
-
 /**
  *     generic_shutdown_super  -       common helper for ->kill_sb()
  *     @sb: superblock to kill
@@ -309,21 +300,13 @@ void generic_shutdown_super(struct super_block *sb)
 
        if (sb->s_root) {
                shrink_dcache_for_umount(sb);
-               fsync_super(sb);
-               lock_super(sb);
+               sync_filesystem(sb);
+               get_fs_excl();
                sb->s_flags &= ~MS_ACTIVE;
 
-               /*
-                * wait for asynchronous fs operations to finish before going further
-                */
-               async_synchronize_full_domain(&sb->s_async_list);
-
                /* bad name - it should be evict_inodes() */
                invalidate_inodes(sb);
-               lock_kernel();
 
-               if (sop->write_super && sb->s_dirt)
-                       sop->write_super(sb);
                if (sop->put_super)
                        sop->put_super(sb);
 
@@ -333,9 +316,7 @@ void generic_shutdown_super(struct super_block *sb)
                           "Self-destruct in 5 seconds.  Have a nice day...\n",
                           sb->s_id);
                }
-
-               unlock_kernel();
-               unlock_super(sb);
+               put_fs_excl();
        }
        spin_lock(&sb_lock);
        /* should be initialized for __put_super_and_need_restart() */
@@ -412,16 +393,14 @@ void drop_super(struct super_block *sb)
 
 EXPORT_SYMBOL(drop_super);
 
-static inline void write_super(struct super_block *sb)
-{
-       lock_super(sb);
-       if (sb->s_root && sb->s_dirt)
-               if (sb->s_op->write_super)
-                       sb->s_op->write_super(sb);
-       unlock_super(sb);
-}
-
-/*
+/**
+ * sync_supers - helper for periodic superblock writeback
+ *
+ * Call the write_super method if present on all dirty superblocks in
+ * the system.  This is for the periodic writeback used by most older
+ * filesystems.  For data integrity superblock writeback use
+ * sync_filesystems() instead.
+ *
  * Note: check the dirty flag before waiting, so we don't
  * hold up the sync while mounting a device. (The newly
  * mounted device won't need syncing.)
@@ -433,12 +412,15 @@ void sync_supers(void)
        spin_lock(&sb_lock);
 restart:
        list_for_each_entry(sb, &super_blocks, s_list) {
-               if (sb->s_dirt) {
+               if (sb->s_op->write_super && sb->s_dirt) {
                        sb->s_count++;
                        spin_unlock(&sb_lock);
+
                        down_read(&sb->s_umount);
-                       write_super(sb);
+                       if (sb->s_root && sb->s_dirt)
+                               sb->s_op->write_super(sb);
                        up_read(&sb->s_umount);
+
                        spin_lock(&sb_lock);
                        if (__put_super_and_need_restart(sb))
                                goto restart;
@@ -447,60 +429,6 @@ restart:
        spin_unlock(&sb_lock);
 }
 
-/*
- * Call the ->sync_fs super_op against all filesystems which are r/w and
- * which implement it.
- *
- * This operation is careful to avoid the livelock which could easily happen
- * if two or more filesystems are being continuously dirtied.  s_need_sync_fs
- * is used only here.  We set it against all filesystems and then clear it as
- * we sync them.  So redirtied filesystems are skipped.
- *
- * But if process A is currently running sync_filesystems and then process B
- * calls sync_filesystems as well, process B will set all the s_need_sync_fs
- * flags again, which will cause process A to resync everything.  Fix that with
- * a local mutex.
- *
- * (Fabian) Avoid sync_fs with clean fs & wait mode 0
- */
-void sync_filesystems(int wait)
-{
-       struct super_block *sb;
-       static DEFINE_MUTEX(mutex);
-
-       mutex_lock(&mutex);             /* Could be down_interruptible */
-       spin_lock(&sb_lock);
-       list_for_each_entry(sb, &super_blocks, s_list) {
-               if (!sb->s_op->sync_fs)
-                       continue;
-               if (sb->s_flags & MS_RDONLY)
-                       continue;
-               sb->s_need_sync_fs = 1;
-       }
-
-restart:
-       list_for_each_entry(sb, &super_blocks, s_list) {
-               if (!sb->s_need_sync_fs)
-                       continue;
-               sb->s_need_sync_fs = 0;
-               if (sb->s_flags & MS_RDONLY)
-                       continue;       /* hm.  Was remounted r/o meanwhile */
-               sb->s_count++;
-               spin_unlock(&sb_lock);
-               down_read(&sb->s_umount);
-               async_synchronize_full_domain(&sb->s_async_list);
-               if (sb->s_root && (wait || sb->s_dirt))
-                       sb->s_op->sync_fs(sb, wait);
-               up_read(&sb->s_umount);
-               /* restart only when sb is no longer on the list */
-               spin_lock(&sb_lock);
-               if (__put_super_and_need_restart(sb))
-                       goto restart;
-       }
-       spin_unlock(&sb_lock);
-       mutex_unlock(&mutex);
-}
-
 /**
  *     get_super - get the superblock of a device
  *     @bdev: device to get the superblock for
@@ -537,6 +465,48 @@ rescan:
 }
 
 EXPORT_SYMBOL(get_super);
+
+/**
+ * get_active_super - get an active reference to the superblock of a device
+ * @bdev: device to get the superblock for
+ *
+ * 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.
+ */
+struct super_block *get_active_super(struct block_device *bdev)
+{
+       struct super_block *sb;
+
+       if (!bdev)
+               return NULL;
+
+       spin_lock(&sb_lock);
+       list_for_each_entry(sb, &super_blocks, s_list) {
+               if (sb->s_bdev != bdev)
+                       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);
+                               return sb;
+                       }
+                       spin_unlock(&sb_lock);
+               }
+               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)
 {
@@ -587,45 +557,6 @@ out:
 }
 
 /**
- *     mark_files_ro - mark all files read-only
- *     @sb: superblock in question
- *
- *     All files are marked read-only.  We don't care about pending
- *     delete files so this should be used in 'force' mode only.
- */
-
-static void mark_files_ro(struct super_block *sb)
-{
-       struct file *f;
-
-retry:
-       file_list_lock();
-       list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
-               struct vfsmount *mnt;
-               if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
-                      continue;
-               if (!file_count(f))
-                       continue;
-               if (!(f->f_mode & FMODE_WRITE))
-                       continue;
-               f->f_mode &= ~FMODE_WRITE;
-               if (file_check_writeable(f) != 0)
-                       continue;
-               file_release_write(f);
-               mnt = mntget(f->f_path.mnt);
-               file_list_unlock();
-               /*
-                * This can sleep, so we can't hold
-                * the file_list_lock() spinlock.
-                */
-               mnt_drop_write(mnt);
-               mntput(mnt);
-               goto retry;
-       }
-       file_list_unlock();
-}
-
-/**
  *     do_remount_sb - asks filesystem to change mount options.
  *     @sb:    superblock in question
  *     @flags: numeric part of options
@@ -637,20 +568,27 @@ retry:
 int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
 {
        int retval;
-       int remount_rw;
-       
+       int remount_rw, remount_ro;
+
+       if (sb->s_frozen != SB_UNFROZEN)
+               return -EBUSY;
+
 #ifdef CONFIG_BLOCK
        if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
                return -EACCES;
 #endif
+
        if (flags & MS_RDONLY)
                acct_auto_close(sb);
        shrink_dcache_sb(sb);
-       fsync_super(sb);
+       sync_filesystem(sb);
+
+       remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
+       remount_rw = !(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 (remount_ro) {
                if (force)
                        mark_files_ro(sb);
                else if (!fs_may_remount_ro(sb))
@@ -659,22 +597,29 @@ int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
                if (retval < 0 && retval != -ENOSYS)
                        return -EBUSY;
        }
-       remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY);
 
        if (sb->s_op->remount_fs) {
-               lock_super(sb);
                retval = sb->s_op->remount_fs(sb, &flags, data);
-               unlock_super(sb);
                if (retval)
                        return retval;
        }
        sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
        if (remount_rw)
                vfs_dq_quota_on_remount(sb);
+       /*
+        * 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;
 }
 
-static void do_emergency_remount(unsigned long foo)
+static void do_emergency_remount(struct work_struct *work)
 {
        struct super_block *sb;
 
@@ -682,27 +627,33 @@ static void do_emergency_remount(unsigned long foo)
        list_for_each_entry(sb, &super_blocks, s_list) {
                sb->s_count++;
                spin_unlock(&sb_lock);
-               down_read(&sb->s_umount);
+               down_write(&sb->s_umount);
                if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) {
                        /*
                         * ->remount_fs needs lock_kernel().
                         *
                         * What lock protects sb->s_flags??
                         */
-                       lock_kernel();
                        do_remount_sb(sb, MS_RDONLY, NULL, 1);
-                       unlock_kernel();
                }
-               drop_super(sb);
+               up_write(&sb->s_umount);
+               put_super(sb);
                spin_lock(&sb_lock);
        }
        spin_unlock(&sb_lock);
+       kfree(work);
        printk("Emergency Remount complete\n");
 }
 
 void emergency_remount(void)
 {
-       pdflush_operation(do_emergency_remount, 0);
+       struct work_struct *work;
+
+       work = kmalloc(sizeof(*work), GFP_ATOMIC);
+       if (work) {
+               INIT_WORK(work, do_emergency_remount);
+               schedule_work(work);
+       }
 }
 
 /*
@@ -712,6 +663,7 @@ void emergency_remount(void)
 
 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)
 {
@@ -722,7 +674,9 @@ int set_anon_super(struct super_block *s, void *data)
        if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
                return -ENOMEM;
        spin_lock(&unnamed_dev_lock);
-       error = ida_get_new(&unnamed_dev_ida, &dev);
+       error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
+       if (!error)
+               unnamed_dev_start = dev + 1;
        spin_unlock(&unnamed_dev_lock);
        if (error == -EAGAIN)
                /* We raced and lost with another CPU. */
@@ -733,6 +687,8 @@ int set_anon_super(struct super_block *s, void *data)
        if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
                spin_lock(&unnamed_dev_lock);
                ida_remove(&unnamed_dev_ida, dev);
+               if (unnamed_dev_start > dev)
+                       unnamed_dev_start = dev;
                spin_unlock(&unnamed_dev_lock);
                return -EMFILE;
        }
@@ -749,6 +705,8 @@ void kill_anon_super(struct super_block *sb)
        generic_shutdown_super(sb);
        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);
 }
 
@@ -763,11 +721,56 @@ void kill_litter_super(struct super_block *sb)
 
 EXPORT_SYMBOL(kill_litter_super);
 
+static int ns_test_super(struct super_block *sb, void *data)
+{
+       return sb->s_fs_info == data;
+}
+
+static int ns_set_super(struct super_block *sb, void *data)
+{
+       sb->s_fs_info = 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 super_block *sb;
+
+       sb = sget(fs_type, ns_test_super, ns_set_super, data);
+       if (IS_ERR(sb))
+               return PTR_ERR(sb);
+
+       if (!sb->s_root) {
+               int err;
+               sb->s_flags = flags;
+               err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
+               if (err) {
+                       deactivate_locked_super(sb);
+                       return err;
+               }
+
+               sb->s_flags |= MS_ACTIVE;
+       }
+
+       simple_set_mnt(mnt, sb);
+       return 0;
+}
+
+EXPORT_SYMBOL(get_sb_ns);
+
 #ifdef CONFIG_BLOCK
 static int set_bdev_super(struct super_block *s, void *data)
 {
        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;
 }
 
@@ -798,16 +801,20 @@ int get_sb_bdev(struct file_system_type *fs_type,
         * will protect the lockfs code from trying to start a snapshot
         * while we are mounting
         */
-       down(&bdev->bd_mount_sem);
+       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_bdev_super, set_bdev_super, bdev);
-       up(&bdev->bd_mount_sem);
+       mutex_unlock(&bdev->bd_fsfreeze_mutex);
        if (IS_ERR(s))
                goto error_s;
 
        if (s->s_root) {
                if ((flags ^ s->s_flags) & MS_RDONLY) {
-                       up_write(&s->s_umount);
-                       deactivate_super(s);
+                       deactivate_locked_super(s);
                        error = -EBUSY;
                        goto error_bdev;
                }
@@ -822,8 +829,7 @@ int get_sb_bdev(struct file_system_type *fs_type,
                sb_set_blocksize(s, block_size(bdev));
                error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
                if (error) {
-                       up_write(&s->s_umount);
-                       deactivate_super(s);
+                       deactivate_locked_super(s);
                        goto error;
                }
 
@@ -831,7 +837,8 @@ int get_sb_bdev(struct file_system_type *fs_type,
                bdev->bd_super = s;
        }
 
-       return simple_set_mnt(mnt, s);
+       simple_set_mnt(mnt, s);
+       return 0;
 
 error_s:
        error = PTR_ERR(s);
@@ -848,7 +855,7 @@ void kill_block_super(struct super_block *sb)
        struct block_device *bdev = sb->s_bdev;
        fmode_t mode = sb->s_mode;
 
-       bdev->bd_super = 0;
+       bdev->bd_super = NULL;
        generic_shutdown_super(sb);
        sync_blockdev(bdev);
        close_bdev_exclusive(bdev, mode);
@@ -872,12 +879,12 @@ int get_sb_nodev(struct file_system_type *fs_type,
 
        error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
        if (error) {
-               up_write(&s->s_umount);
-               deactivate_super(s);
+               deactivate_locked_super(s);
                return error;
        }
        s->s_flags |= MS_ACTIVE;
-       return simple_set_mnt(mnt, s);
+       simple_set_mnt(mnt, s);
+       return 0;
 }
 
 EXPORT_SYMBOL(get_sb_nodev);
@@ -902,14 +909,15 @@ int get_sb_single(struct file_system_type *fs_type,
                s->s_flags = flags;
                error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
                if (error) {
-                       up_write(&s->s_umount);
-                       deactivate_super(s);
+                       deactivate_locked_super(s);
                        return error;
                }
                s->s_flags |= MS_ACTIVE;
+       } else {
+               do_remount_sb(s, flags, data, 0);
        }
-       do_remount_sb(s, flags, data, 0);
-       return simple_set_mnt(mnt, s);
+       simple_set_mnt(mnt, s);
+       return 0;
 }
 
 EXPORT_SYMBOL(get_sb_single);
@@ -929,6 +937,9 @@ vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void
        if (!mnt)
                goto out;
 
+       if (flags & MS_KERNMOUNT)
+               mnt->mnt_flags = MNT_INTERNAL;
+
        if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
                secdata = alloc_secdata();
                if (!secdata)
@@ -948,6 +959,16 @@ vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void
        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.
+        */
+       WARN((mnt->mnt_sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
+               "negative value (%lld)\n", type->name, mnt->mnt_sb->s_maxbytes);
+
        mnt->mnt_mountpoint = mnt->mnt_root;
        mnt->mnt_parent = mnt;
        up_write(&mnt->mnt_sb->s_umount);
@@ -955,8 +976,7 @@ vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void
        return mnt;
 out_sb:
        dput(mnt->mnt_root);
-       up_write(&mnt->mnt_sb->s_umount);
-       deactivate_super(mnt->mnt_sb);
+       deactivate_locked_super(mnt->mnt_sb);
 out_free_secdata:
        free_secdata(secdata);
 out_mnt: