ext4: make sure to revoke all the freeable blocks in ext4_free_blocks
authorDaeho Jeong <daeho.jeong@samsung.com>
Sun, 21 Feb 2016 23:31:41 +0000 (18:31 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Sun, 21 Feb 2016 23:31:41 +0000 (18:31 -0500)
Now, ext4_free_blocks() doesn't revoke data blocks of per-file data
journalled inode and it can cause file data inconsistency problems.
Even though data blocks of per-file data journalled inode are already
forgotten by jbd2_journal_invalidatepage() in advance of invoking
ext4_free_blocks(), we still need to revoke the data blocks here.
Moreover some of the metadata blocks, which are not found by
sb_find_get_block(), are still needed to be revoked, but this is also
missing here.

Signed-off-by: Daeho Jeong <daeho.jeong@samsung.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
fs/ext4/mballoc.c

index 4424b7bf8ac64e431a11570e247a1d828634d29e..f6ff47838670b08a338c1cc32704c18aa08d8f72 100644 (file)
@@ -4694,16 +4694,6 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
                            inode, bh, block);
        }
 
-       /*
-        * We need to make sure we don't reuse the freed block until
-        * after the transaction is committed, which we can do by
-        * treating the block as metadata, below.  We make an
-        * exception if the inode is to be written in writeback mode
-        * since writeback mode has weak data consistency guarantees.
-        */
-       if (!ext4_should_writeback_data(inode))
-               flags |= EXT4_FREE_BLOCKS_METADATA;
-
        /*
         * If the extent to be freed does not begin on a cluster
         * boundary, we need to deal with partial clusters at the
@@ -4738,14 +4728,13 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
 
        if (!bh && (flags & EXT4_FREE_BLOCKS_FORGET)) {
                int i;
+               int is_metadata = flags & EXT4_FREE_BLOCKS_METADATA;
 
                for (i = 0; i < count; i++) {
                        cond_resched();
-                       bh = sb_find_get_block(inode->i_sb, block + i);
-                       if (!bh)
-                               continue;
-                       ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
-                                   inode, bh, block + i);
+                       if (is_metadata)
+                               bh = sb_find_get_block(inode->i_sb, block + i);
+                       ext4_forget(handle, is_metadata, inode, bh, block + i);
                }
        }
 
@@ -4819,12 +4808,17 @@ do_more:
        if (err)
                goto error_return;
 
-       if ((flags & EXT4_FREE_BLOCKS_METADATA) && ext4_handle_valid(handle)) {
+       /*
+        * We need to make sure we don't reuse the freed block until after the
+        * transaction is committed. We make an exception if the inode is to be
+        * written in writeback mode since writeback mode has weak data
+        * consistency guarantees.
+        */
+       if (ext4_handle_valid(handle) &&
+           ((flags & EXT4_FREE_BLOCKS_METADATA) ||
+            !ext4_should_writeback_data(inode))) {
                struct ext4_free_data *new_entry;
                /*
-                * blocks being freed are metadata. these blocks shouldn't
-                * be used until this transaction is committed
-                *
                 * We use __GFP_NOFAIL because ext4_free_blocks() is not allowed
                 * to fail.
                 */