Btrfs: Add more synchronization before creating a snapshot

File data checksums are only done during writepage, so we have to make sure
all pages are written when the snapshot is taken.  This also adds some
locking so that new writes don't race in and add new dirty pages.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

4 files changed, 13 insertions, 1 deletions

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index cd75c906048f..03e1c4ad341b 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -333,6 +333,7 @@ struct btrfs_root {
 	struct inode *inode;
 	struct kobject root_kobj;
 	struct completion kobj_unregister;
+	struct rw_semaphore snap_sem;
 	u64 objectid;
 	u64 last_trans;
 	u32 blocksize;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index c25ef0a68f18..6c953a0e0aa3 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -300,6 +300,7 @@ static int __setup_root(int blocksize,
 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 	init_completion(&root->kobj_unregister);
+	init_rwsem(&root->snap_sem);
 	root->defrag_running = 0;
 	root->defrag_level = 0;
 	root->root_key.objectid = objectid;
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index acef8cb7fb79..77f1950b0aa0 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -160,6 +160,7 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 	num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
 			inode->i_blkbits;
 
+	down_read(&BTRFS_I(inode)->root->snap_sem);
 	end_of_last_block = start_pos + (num_blocks << inode->i_blkbits) - 1;
 	lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 	mutex_lock(&root->fs_info->fs_mutex);
@@ -250,6 +251,7 @@ out_unlock:
 	mutex_unlock(&root->fs_info->fs_mutex);
 	unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
 	free_extent_map(em);
+	up_read(&BTRFS_I(inode)->root->snap_sem);
 	return err;
 }
 
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 6b3e4404dc6a..1ace6d11e097 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -686,6 +686,7 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 	if ((offset & (blocksize - 1)) == 0)
 		goto out;
 
+	down_read(&BTRFS_I(inode)->root->snap_sem);
 	ret = -ENOMEM;
 	page = grab_cache_page(mapping, index);
 	if (!page)
@@ -704,6 +705,7 @@ static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 
 	unlock_page(page);
 	page_cache_release(page);
+	up_read(&BTRFS_I(inode)->root->snap_sem);
 out:
 	return ret;
 }
@@ -1668,6 +1670,7 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 	int ret = -EINVAL;
 	u64 page_start;
 
+	down_read(&BTRFS_I(inode)->root->snap_sem);
 	lock_page(page);
 	wait_on_page_writeback(page);
 	size = i_size_read(inode);
@@ -1688,6 +1691,7 @@ int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
 	ret = btrfs_cow_one_page(inode, page, end);
 
 out_unlock:
+	up_read(&BTRFS_I(inode)->root->snap_sem);
 	unlock_page(page);
 	return ret;
 }
@@ -1851,6 +1855,10 @@ static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
 	if (!root->ref_cows)
 		return -EINVAL;
 
+	down_write(&root->snap_sem);
+	freeze_bdev(root->fs_info->sb->s_bdev);
+	thaw_bdev(root->fs_info->sb->s_bdev, root->fs_info->sb);
+
 	mutex_lock(&root->fs_info->fs_mutex);
 	trans = btrfs_start_transaction(root, 1);
 	BUG_ON(!trans);
@@ -1894,12 +1902,12 @@ static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
 	ret = btrfs_inc_root_ref(trans, root);
 	if (ret)
 		goto fail;
-
 fail:
 	err = btrfs_commit_transaction(trans, root);
 	if (err && !ret)
 		ret = err;
 	mutex_unlock(&root->fs_info->fs_mutex);
+	up_write(&root->snap_sem);
 	btrfs_btree_balance_dirty(root);
 	return ret;
 }