OSDN Git Service

btrfs: account for non-CoW'd blocks in btrfs_abort_transaction
authorJeff Mahoney <jeffm@suse.com>
Wed, 8 Jun 2016 04:36:38 +0000 (00:36 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 27 Jul 2016 16:47:33 +0000 (09:47 -0700)
commit 64c12921e11b3a0c10d088606e328c58e29274d8 upstream.

The test for !trans->blocks_used in btrfs_abort_transaction is
insufficient to determine whether it's safe to drop the transaction
handle on the floor.  btrfs_cow_block, informed by should_cow_block,
can return blocks that have already been CoW'd in the current
transaction.  trans->blocks_used is only incremented for new block
allocations. If an operation overlaps the blocks in the current
transaction entirely and must abort the transaction, we'll happily
let it clean up the trans handle even though it may have modified
the blocks and will commit an incomplete operation.

In the long-term, I'd like to do closer tracking of when the fs
is actually modified so we can still recover as gracefully as possible,
but that approach will need some discussion.  In the short term,
since this is the only code using trans->blocks_used, let's just
switch it to a bool indicating whether any blocks were used and set
it when should_cow_block returns false.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/btrfs/ctree.c
fs/btrfs/extent-tree.c
fs/btrfs/super.c
fs/btrfs/transaction.h

index 5b8e235..0f2b7c6 100644 (file)
@@ -1551,6 +1551,7 @@ noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
                       trans->transid, root->fs_info->generation);
 
        if (!should_cow_block(trans, root, buf)) {
+               trans->dirty = true;
                *cow_ret = buf;
                return 0;
        }
@@ -2773,8 +2774,10 @@ again:
                         * then we don't want to set the path blocking,
                         * so we test it here
                         */
-                       if (!should_cow_block(trans, root, b))
+                       if (!should_cow_block(trans, root, b)) {
+                               trans->dirty = true;
                                goto cow_done;
+                       }
 
                        /*
                         * must have write locks on this node and the
index 2368cac..47cdc6f 100644 (file)
@@ -7856,7 +7856,7 @@ btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
                set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
                         buf->start + buf->len - 1, GFP_NOFS);
        }
-       trans->blocks_used++;
+       trans->dirty = true;
        /* this returns a buffer locked for blocking */
        return buf;
 }
index fe609b8..5d34a06 100644 (file)
@@ -239,7 +239,7 @@ void __btrfs_abort_transaction(struct btrfs_trans_handle *trans,
        trans->aborted = errno;
        /* Nothing used. The other threads that have joined this
         * transaction may be able to continue. */
-       if (!trans->blocks_used && list_empty(&trans->new_bgs)) {
+       if (!trans->dirty && list_empty(&trans->new_bgs)) {
                const char *errstr;
 
                errstr = btrfs_decode_error(errno);
index 64c8221..1e87292 100644 (file)
@@ -110,7 +110,6 @@ struct btrfs_trans_handle {
        u64 chunk_bytes_reserved;
        unsigned long use_count;
        unsigned long blocks_reserved;
-       unsigned long blocks_used;
        unsigned long delayed_ref_updates;
        struct btrfs_transaction *transaction;
        struct btrfs_block_rsv *block_rsv;
@@ -121,6 +120,7 @@ struct btrfs_trans_handle {
        bool can_flush_pending_bgs;
        bool reloc_reserved;
        bool sync;
+       bool dirty;
        unsigned int type;
        /*
         * this root is only needed to validate that the root passed to