Skip to contentSkip to conversation

btrfs: Don't panic when we can't find a root key

PATCH Superseded Qu Wenruo 5 participants 5 messages Feb 25, 2019 Btrfs file system

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Qu Wenruo

author
#0 lore raw

PATCH 1/1 btrfs: Don't panic when we can't find a root key

When we failed to find a root key in btrfs_update_root(), we just panic.

That's definitely not cool, fix it by aborting current transaction and return an error value.

---
 fs/btrfs/root-tree.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
fs/btrfs/root-tree.c +8 -4
index 65bda0682928..c48a3e18866d 100644
@@ -137,11 +137,15 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
 		goto out;
 	}
 
-	if (ret != 0) {
+	if (ret > 0) {
 		btrfs_print_leaf(path->nodes[0]);
-		btrfs_crit(fs_info, "unable to update root key %llu %u %llu",
-			   key->objectid, key->type, key->offset);
-		BUG_ON(1);
+		btrfs_crit(fs_info,
+			"unable to find root key (%llu %u %llu) in tree %llu",
+			   key->objectid, key->type, key->offset,
+			   root->root_key.objectid);
+		ret = -ENOENT;
+		btrfs_abort_transaction(trans, ret);
+		goto out;
 	}
 
 	l = path->nodes[0];
Signature · 1 line
2.20.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 7 lines · Qu Wenruo

When we failed to find a root key in btrfs_update_root(), we just panic.

That's definitely not cool, fix it by aborting current transaction and return an error value.

Looks good.

Quoted 31 lines
---
 fs/btrfs/root-tree.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
fs/btrfs/root-tree.c +8 -5
index 65bda0682928..c48a3e18866d 100644
@@ -137,11 +137,15 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
                goto out;
        }
 
-       if (ret != 0) {
+       if (ret > 0) {
                btrfs_print_leaf(path->nodes[0]);
-               btrfs_crit(fs_info, "unable to update root key %llu %u %llu",
-                          key->objectid, key->type, key->offset);
-               BUG_ON(1);
+               btrfs_crit(fs_info,
+                       "unable to find root key (%llu %u %llu) in tree %llu",
+                          key->objectid, key->type, key->offset,
+                          root->root_key.objectid);
+               ret = -ENOENT;
+               btrfs_abort_transaction(trans, ret);
+               goto out;
        }
 
        l = path->nodes[0];
--

2.20.1

Signature · 3 lines
Filipe David Manana,

“Whether you think you can, or you think you can't — you're right.”
#3 lore raw
In reply to Qu Wenruo
Quoted 15 lines · Qu Wenruo

When we failed to find a root key in btrfs_update_root(), we just panic.

That's definitely not cool, fix it by aborting current transaction and return an error value.

---
 fs/btrfs/root-tree.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
fs/btrfs/root-tree.c +0 -0
index 65bda0682928..c48a3e18866d 100644

fs/btrfs/root-tree.c int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root · line 141

Quoted 6 lines
fs/btrfs/root-tree.c +1 -1
@@ -137,11 +137,15 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
 		goto out;
 	}
 
-	if (ret != 0) {
+	if (ret > 0) {
 		btrfs_print_leaf(path->nodes[0]);

You leave the leaf dump here, which I believe was related to the BUG_ON so there's information for later analysis. This is not usually done where transaction is aborted and error returned. I'm not saying it's not worth keeping it here, as the logic that would lead to this error is in the 'never happens' category. But we know such things happen due to the bitflips and random memory corruptions so a dump might make sense.

If you think this makes sense, please put it to the changelog and write a comment before the leaf dump. This will be a good pattern to follow as there are many instances of leaf_dump/BUG to clean up or review.

#4 lore raw
Quoted 30 lines · David Sterba
Quoted 15 lines · Qu Wenruo

When we failed to find a root key in btrfs_update_root(), we just panic.

That's definitely not cool, fix it by aborting current transaction and return an error value.

---
 fs/btrfs/root-tree.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
fs/btrfs/root-tree.c +0 -0
index 65bda0682928..c48a3e18866d 100644

fs/btrfs/root-tree.c int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root · line 141

Quoted 6 lines
fs/btrfs/root-tree.c +1 -1
@@ -137,11 +137,15 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
 		goto out;
 	}
 
-	if (ret != 0) {
+	if (ret > 0) {
 		btrfs_print_leaf(path->nodes[0]);

You leave the leaf dump here, which I believe was related to the BUG_ON so there's information for later analysis. This is not usually done where transaction is aborted and error returned. I'm not saying it's not worth keeping it here, as the logic that would lead to this error is in the 'never happens' category. But we know such things happen due to the bitflips and random memory corruptions so a dump might make sense.

Quoted 3 lines

If you think this makes sense, please put it to the changelog and write a comment before the leaf dump. This will be a good pattern to follow as there are many instances of leaf_dump/BUG to clean up or review.

I think the dump makes sense, but didn't think so deep to make it as an example for later cleanup.

If we're going to use this as an example for later cleanups. I won't keep it.

The message itself should be enough to locate the cause, and the tree dump doesn't help as much as we expected.

So I'm going to remove it in next update.

Thanks,
Qu