Btrfs: Do fsync log replay when mount -o ro, except when on readonly media
authorChris Mason <chris.mason@oracle.com>
Wed, 19 Nov 2008 20:13:35 +0000 (15:13 -0500)
committerChris Mason <chris.mason@oracle.com>
Wed, 19 Nov 2008 20:13:35 +0000 (15:13 -0500)
fsync log replay can change the filesystem, so it cannot be delayed until
mount -o rw,remount, and it can't be forgotten entirely.  So, this patch
changes btrfs to do with reiserfs, ext3 and xfs do, which is to do the
log replay even when mounted readonly.

On a readonly device if log replay is required, the mount is aborted.

Getting all of this right had required fixing up some of the error
handling in open_ctree.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
fs/btrfs/disk-io.c

index 8d03e4a3c4e934a876f390f609599e3bf2bc6889..c8dcb47b6d7dc62151eb1bdaf462d9572ec50c00 100644 (file)
@@ -518,7 +518,6 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
                        extent_submit_bio_hook_t *submit_bio_done)
 {
        struct async_submit_bio *async;
-       int limit = btrfs_async_submit_limit(fs_info);
 
        async = kmalloc(sizeof(*async), GFP_NOFS);
        if (!async)
@@ -541,6 +540,7 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
        atomic_inc(&fs_info->nr_async_submits);
        btrfs_queue_worker(&fs_info->workers, &async->work);
 #if 0
+       int limit = btrfs_async_submit_limit(fs_info);
        if (atomic_read(&fs_info->nr_async_submits) > limit) {
                wait_event_timeout(fs_info->async_submit_wait,
                           (atomic_read(&fs_info->nr_async_submits) < limit),
@@ -1732,13 +1732,15 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        if (!fs_info->transaction_kthread)
                goto fail_cleaner;
 
-       if (sb->s_flags & MS_RDONLY)
-               goto read_fs_root;
-
        if (btrfs_super_log_root(disk_super) != 0) {
                u32 blocksize;
                u64 bytenr = btrfs_super_log_root(disk_super);
 
+               if (fs_devices->rw_devices == 0) {
+                       printk("Btrfs log replay required on RO media\n");
+                       err = -EIO;
+                       goto fail_trans_kthread;
+               }
                blocksize =
                     btrfs_level_size(tree_root,
                                      btrfs_super_log_root_level(disk_super));
@@ -1756,21 +1758,32 @@ struct btrfs_root *open_ctree(struct super_block *sb,
                BUG_ON(ret);
        }
 
-       ret = btrfs_cleanup_reloc_trees(tree_root);
-       BUG_ON(ret);
+       if (!(sb->s_flags & MS_RDONLY)) {
+               ret = btrfs_cleanup_reloc_trees(tree_root);
+               BUG_ON(ret);
+       }
 
-read_fs_root:
        location.objectid = BTRFS_FS_TREE_OBJECTID;
        location.type = BTRFS_ROOT_ITEM_KEY;
        location.offset = (u64)-1;
 
        fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
        if (!fs_info->fs_root)
-               goto fail_cleaner;
+               goto fail_trans_kthread;
        return tree_root;
 
+fail_trans_kthread:
+       kthread_stop(fs_info->transaction_kthread);
 fail_cleaner:
        kthread_stop(fs_info->cleaner_kthread);
+
+       /*
+        * make sure we're done with the btree inode before we stop our
+        * kthreads
+        */
+       filemap_write_and_wait(fs_info->btree_inode->i_mapping);
+       invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
+
 fail_extent_root:
        free_extent_buffer(extent_root->node);
 fail_tree_root:
@@ -1778,6 +1791,7 @@ fail_tree_root:
 fail_chunk_root:
        free_extent_buffer(chunk_root->node);
 fail_sys_array:
+       free_extent_buffer(dev_root->node);
 fail_sb_buffer:
        btrfs_stop_workers(&fs_info->fixup_workers);
        btrfs_stop_workers(&fs_info->delalloc_workers);
@@ -1786,6 +1800,7 @@ fail_sb_buffer:
        btrfs_stop_workers(&fs_info->endio_write_workers);
        btrfs_stop_workers(&fs_info->submit_workers);
 fail_iput:
+       invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
        iput(fs_info->btree_inode);
 fail:
        btrfs_close_devices(fs_info->fs_devices);