From 8c594ea81d7abbbffdda447b127f8ba8d76f319d Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Mon, 20 Apr 2009 15:50:10 -0400 Subject: Btrfs: use the right node in reada_for_balance reada_for_balance was using the wrong index into the path node array, so it wasn't reading the right blocks. We never directly used the results of the read done by this function because the btree search is started over at the end. This fixes reada_for_balance to reada in the correct node and to avoid searching past the last slot in the node. It also makes sure to hold the parent lock while we are finding the nodes to read. Signed-off-by: Chris Mason --- fs/btrfs/ctree.c | 17 ++++++++++++----- 1 file changed, 12 insertions(+), 5 deletions(-) (limited to 'fs/btrfs/ctree.c') diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c index e5b2533b691..a99f1c2a710 100644 --- a/fs/btrfs/ctree.c +++ b/fs/btrfs/ctree.c @@ -1325,12 +1325,12 @@ static noinline int reada_for_balance(struct btrfs_root *root, int ret = 0; int blocksize; - parent = path->nodes[level - 1]; + parent = path->nodes[level + 1]; if (!parent) return 0; nritems = btrfs_header_nritems(parent); - slot = path->slots[level]; + slot = path->slots[level + 1]; blocksize = btrfs_level_size(root, level); if (slot > 0) { @@ -1341,7 +1341,7 @@ static noinline int reada_for_balance(struct btrfs_root *root, block1 = 0; free_extent_buffer(eb); } - if (slot < nritems) { + if (slot + 1 < nritems) { block2 = btrfs_node_blockptr(parent, slot + 1); gen = btrfs_node_ptr_generation(parent, slot + 1); eb = btrfs_find_tree_block(root, block2, blocksize); @@ -1351,7 +1351,11 @@ static noinline int reada_for_balance(struct btrfs_root *root, } if (block1 || block2) { ret = -EAGAIN; + + /* release the whole path */ btrfs_release_path(root, path); + + /* read the blocks */ if (block1) readahead_tree_block(root, block1, blocksize, 0); if (block2) @@ -1361,7 +1365,7 @@ static noinline int reada_for_balance(struct btrfs_root *root, eb = read_tree_block(root, block1, blocksize, 0); free_extent_buffer(eb); } - if (block1) { + if (block2) { eb = read_tree_block(root, block2, blocksize, 0); free_extent_buffer(eb); } @@ -1481,12 +1485,15 @@ read_block_for_search(struct btrfs_trans_handle *trans, * of the btree by dropping locks before * we read. */ - btrfs_release_path(NULL, p); + btrfs_unlock_up_safe(p, level + 1); + btrfs_set_path_blocking(p); + if (tmp) free_extent_buffer(tmp); if (p->reada) reada_for_search(root, p, level, slot, key->objectid); + btrfs_release_path(NULL, p); tmp = read_tree_block(root, blocknr, blocksize, gen); if (tmp) free_extent_buffer(tmp); -- cgit v1.2.3-70-g09d2 From 76a05b35a320e8c968d0fec8f512a1acae227309 Mon Sep 17 00:00:00 2001 From: Chris Mason Date: Thu, 14 May 2009 13:24:30 -0400 Subject: Btrfs: Don't loop forever on metadata IO failures When a btrfs metadata read fails, the first thing we try to do is find a good copy on another mirror of the block. If this fails, read_tree_block() ends up returning a buffer that isn't up to date. The btrfs btree reading code was reworked to drop locks and repeat the search when IO was done, but the changes didn't add a check for failed reads. The end result was looping forever on buffers that were never going to become up to date. Signed-off-by: Chris Mason --- fs/btrfs/ctree.c | 39 ++++++++++++++++++++++++++++++++++++--- 1 file changed, 36 insertions(+), 3 deletions(-) (limited to 'fs/btrfs/ctree.c') diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c index a99f1c2a710..fedf8b9f03a 100644 --- a/fs/btrfs/ctree.c +++ b/fs/btrfs/ctree.c @@ -1469,6 +1469,7 @@ read_block_for_search(struct btrfs_trans_handle *trans, u32 blocksize; struct extent_buffer *b = *eb_ret; struct extent_buffer *tmp; + int ret; blocknr = btrfs_node_blockptr(b, slot); gen = btrfs_node_ptr_generation(b, slot); @@ -1476,6 +1477,10 @@ read_block_for_search(struct btrfs_trans_handle *trans, 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; } @@ -1483,7 +1488,9 @@ read_block_for_search(struct btrfs_trans_handle *trans, /* * reduce lock contention at high levels * of the btree by dropping locks before - * we read. + * we read. Don't release the lock on the current + * level because we need to walk this node to figure + * out which blocks to read. */ btrfs_unlock_up_safe(p, level + 1); btrfs_set_path_blocking(p); @@ -1494,10 +1501,21 @@ read_block_for_search(struct btrfs_trans_handle *trans, reada_for_search(root, p, level, slot, key->objectid); btrfs_release_path(NULL, p); + + ret = -EAGAIN; tmp = read_tree_block(root, blocknr, blocksize, gen); - if (tmp) + if (tmp) { + /* + * If the read above didn't mark this buffer up to date, + * it will never end up being up to date. Set ret to EIO now + * and give up so that our caller doesn't loop forever + * on our EAGAINs. + */ + if (!btrfs_buffer_uptodate(tmp, 0)) + ret = -EIO; free_extent_buffer(tmp); - return -EAGAIN; + } + return ret; } /* @@ -1696,6 +1714,9 @@ cow_done: if (ret == -EAGAIN) goto again; + if (ret == -EIO) + goto done; + if (!p->skip_locking) { int lret; @@ -1738,6 +1759,8 @@ done: */ if (!p->leave_spinning) btrfs_set_path_blocking(p); + if (ret < 0) + btrfs_release_path(root, p); return ret; } @@ -4212,6 +4235,11 @@ again: if (ret == -EAGAIN) goto again; + if (ret < 0) { + btrfs_release_path(root, path); + goto done; + } + if (!path->skip_locking) { ret = btrfs_try_spin_lock(next); if (!ret) { @@ -4246,6 +4274,11 @@ again: if (ret == -EAGAIN) goto again; + if (ret < 0) { + btrfs_release_path(root, path); + goto done; + } + if (!path->skip_locking) { btrfs_assert_tree_locked(path->nodes[level]); ret = btrfs_try_spin_lock(next); -- cgit v1.2.3-70-g09d2