Btrfs: check if previous transaction aborted to avoid fs corruption
authorFilipe Manana <fdmanana@suse.com>
Wed, 12 Aug 2015 10:54:35 +0000 (11:54 +0100)
committerChris Mason <clm@fb.com>
Wed, 19 Aug 2015 21:27:31 +0000 (14:27 -0700)
While we are committing a transaction, it's possible the previous one is
still finishing its commit and therefore we wait for it to finish first.
However we were not checking if that previous transaction ended up getting
aborted after we waited for it to commit, so we ended up committing the
current transaction which can lead to fs corruption because the new
superblock can point to trees that have had one or more nodes/leafs that
were never durably persisted.
The following sequence diagram exemplifies how this is possible:

          CPU 0                                                        CPU 1

  transaction N starts

  (...)

  btrfs_commit_transaction(N)

    cur_trans->state = TRANS_STATE_COMMIT_START;
    (...)
    cur_trans->state = TRANS_STATE_COMMIT_DOING;
    (...)

    cur_trans->state = TRANS_STATE_UNBLOCKED;
    root->fs_info->running_transaction = NULL;

                                                              btrfs_start_transaction()
                                                                 --> starts transaction N + 1

    btrfs_write_and_wait_transaction(trans, root);
      --> starts writing all new or COWed ebs created
          at transaction N

                                                              creates some new ebs, COWs some
                                                              existing ebs but doesn't COW or
                                                              deletes eb X

                                                              btrfs_commit_transaction(N + 1)
                                                                (...)
                                                                cur_trans->state = TRANS_STATE_COMMIT_START;
                                                                (...)
                                                                wait_for_commit(root, prev_trans);
                                                                  --> prev_trans == transaction N

    btrfs_write_and_wait_transaction() continues
    writing ebs
       --> fails writing eb X, we abort transaction N
           and set bit BTRFS_FS_STATE_ERROR on
           fs_info->fs_state, so no new transactions
           can start after setting that bit

       cleanup_transaction()
         btrfs_cleanup_one_transaction()
           wakes up task at CPU 1

                                                                continues, doesn't abort because
                                                                cur_trans->aborted (transaction N + 1)
                                                                is zero, and no checks for bit
                                                                BTRFS_FS_STATE_ERROR in fs_info->fs_state
                                                                are made

                                                                btrfs_write_and_wait_transaction(trans, root);
                                                                  --> succeeds, no errors during writeback

                                                                write_ctree_super(trans, root, 0);
                                                                  --> succeeds
                                                                  --> we have now a superblock that points us
                                                                      to some root that uses eb X, which was
                                                                      never written to disk

In this scenario future attempts to read eb X from disk results in an
error message like "parent transid verify failed on X wanted Y found Z".

So fix this by aborting the current transaction if after waiting for the
previous transaction we verify that it was aborted.

Cc: stable@vger.kernel.org
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: Liu Bo <bo.li.liu@oracle.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/transaction.c

index 20267d47dbcd501dda238e7ff1ce0381137473a7..68ad89e237132c9e6b3bceed8ad0bba7d67357c6 100644 (file)
@@ -1895,8 +1895,11 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
                        spin_unlock(&root->fs_info->trans_lock);
 
                        wait_for_commit(root, prev_trans);
+                       ret = prev_trans->aborted;
 
                        btrfs_put_transaction(prev_trans);
+                       if (ret)
+                               goto cleanup_transaction;
                } else {
                        spin_unlock(&root->fs_info->trans_lock);
                }