ext3: Avoid filesystem corruption after a crash under heavy delete load

It can happen that ext3_free_branches calls ext3_forget() for an indirect block
in an earlier transaction than a transaction in which we clear pointer to this
indirect block. Thus if we crash before a transaction clearing the block
pointer is committed, we will see indirect block pointing to already freed
blocks and complain during orphan list cleanup.

The fix is simple: Make sure ext3_forget() is called in the transaction
doing block pointer clearing.

This is a backport of an ext4 fix by Amir G. <amir73il@users.sourceforge.net>

Signed-off-by: Jan Kara <jack@suse.cz>

1 files changed, 25 insertions, 21 deletions

diff --git a/fs/ext3/inode.c b/fs/ext3/inode.c
index a786db403efc..436e5bbccbc2 100644
--- a/fs/ext3/inode.c
+++ b/fs/ext3/inode.c
@@ -2270,27 +2270,6 @@ static void ext3_free_branches(handle_t *handle, struct inode *inode,
                                            depth);
 
                         /*
-                         * We've probably journalled the indirect block several
-                         * times during the truncate.  But it's no longer
-                         * needed and we now drop it from the transaction via
-                         * journal_revoke().
-                         *
-                         * That's easy if it's exclusively part of this
-                         * transaction.  But if it's part of the committing
-                         * transaction then journal_forget() will simply
-                         * brelse() it.  That means that if the underlying
-                         * block is reallocated in ext3_get_block(),
-                         * unmap_underlying_metadata() will find this block
-                         * and will try to get rid of it.  damn, damn.
-                         *
-                         * If this block has already been committed to the
-                         * journal, a revoke record will be written.  And
-                         * revoke records must be emitted *before* clearing
-                         * this block's bit in the bitmaps.
-                         */
-                        ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
-
-                        /*
                          * Everything below this this pointer has been
                          * released.  Now let this top-of-subtree go.
                          *
@@ -2313,6 +2292,31 @@ static void ext3_free_branches(handle_t *handle, struct inode *inode,
                                 truncate_restart_transaction(handle, inode);
                         }
 
+                        /*
+                         * We've probably journalled the indirect block several
+                         * times during the truncate.  But it's no longer
+                         * needed and we now drop it from the transaction via
+                         * journal_revoke().
+                         *
+                         * That's easy if it's exclusively part of this
+                         * transaction.  But if it's part of the committing
+                         * transaction then journal_forget() will simply
+                         * brelse() it.  That means that if the underlying
+                         * block is reallocated in ext3_get_block(),
+                         * unmap_underlying_metadata() will find this block
+                         * and will try to get rid of it.  damn, damn. Thus
+                         * we don't allow a block to be reallocated until
+                         * a transaction freeing it has fully committed.
+                         *
+                         * We also have to make sure journal replay after a
+                         * crash does not overwrite non-journaled data blocks
+                         * with old metadata when the block got reallocated for
+                         * data.  Thus we have to store a revoke record for a
+                         * block in the same transaction in which we free the
+                         * block.
+                         */
+                        ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
+
                         ext3_free_blocks(handle, inode, nr, 1);
 
                         if (parent_bh) {