Btrfs: don't loop forever on bad btree blocks

When btrfs discovers the generation number in a btree block is
incorrect, it can loop forever without forcing the RAID
code to try a valid mirror, and without returning EIO.

This changes things to properly kick out the EIO.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
This commit is contained in:
Chris Mason 2010-10-24 11:01:27 -04:00
Родитель 6b5b817f10
Коммит cb44921a09
1 изменённых файлов: 28 добавлений и 9 удалений

Просмотреть файл

@ -1577,13 +1577,33 @@ read_block_for_search(struct btrfs_trans_handle *trans,
blocksize = btrfs_level_size(root, level - 1);
tmp = btrfs_find_tree_block(root, blocknr, blocksize);
if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
/*
* we found an up to date block without sleeping, return
* right away
*/
*eb_ret = tmp;
return 0;
if (tmp) {
if (btrfs_buffer_uptodate(tmp, 0)) {
if (btrfs_buffer_uptodate(tmp, gen)) {
/*
* we found an up to date block without
* sleeping, return
* right away
*/
*eb_ret = tmp;
return 0;
}
/* the pages were up to date, but we failed
* the generation number check. Do a full
* read for the generation number that is correct.
* We must do this without dropping locks so
* we can trust our generation number
*/
free_extent_buffer(tmp);
tmp = read_tree_block(root, blocknr, blocksize, gen);
if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
*eb_ret = tmp;
return 0;
}
free_extent_buffer(tmp);
btrfs_release_path(NULL, p);
return -EIO;
}
}
/*
@ -1596,8 +1616,7 @@ read_block_for_search(struct btrfs_trans_handle *trans,
btrfs_unlock_up_safe(p, level + 1);
btrfs_set_path_blocking(p);
if (tmp)
free_extent_buffer(tmp);
free_extent_buffer(tmp);
if (p->reada)
reada_for_search(root, p, level, slot, key->objectid);