[1/2] Btrfs: be more precise on errors when getting an inode from disk

Commit Message

Filipe Manana June 9, 2016, 4:53 p.m. UTC

From: Filipe Manana <[e-mail]>

When we attempt to read an inode from disk, we end up always returning an
-ESTALE error to the caller regardless of the actual failure reason, which
can be an out of memory problem (when allocating a path), some error found
when reading from the fs/subvolume btree (like a genuine IO error) or the
inode does not exists. So lets start returning the real error code to the
callers so that they don't treat all -ESTALE errors as meaning that the
inode does not exists (such as during orphan cleanup). This will also be
needed for a subsequent patch in the same series dealing with a special
fsync case.

Signed-off-by: Filipe Manana <[e-mail]>
---
 fs/btrfs/inode.c | 27 ++++++++++++++++++---------
 1 file changed, 18 insertions(+), 9 deletions(-)

Comments

NeilBrown July 22, 2016, 1:08 a.m. UTC | #1

On Fri, Jun 10 2016, [e-mail] wrote:

> From: Filipe Manana <[e-mail]>
>
> When we attempt to read an inode from disk, we end up always returning an
> -ESTALE error to the caller regardless of the actual failure reason, which
> can be an out of memory problem (when allocating a path), some error found
> when reading from the fs/subvolume btree (like a genuine IO error) or the
> inode does not exists. So lets start returning the real error code to the
> callers so that they don't treat all -ESTALE errors as meaning that the
> inode does not exists (such as during orphan cleanup). This will also be
> needed for a subsequent patch in the same series dealing with a special
> fsync case.
>
> Signed-off-by: Filipe Manana <[e-mail]>

SNIP

> @@ -5594,7 +5602,8 @@ struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
>  		} else {
>  			unlock_new_inode(inode);
>  			iput(inode);
> -			inode = ERR_PTR(-ESTALE);
> +			ASSERT(ret < 0);
> +			inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
>  		}

Just a heads-up.  This change breaks NFS :-(

The change in error code percolates up the call chain:

 nfs4_pufh->fh_verify->nfsd_set_fh_dentry->exportfs_decode_fh
    ->btrfs_fh_to_dentry->ntrfs_get_dentry->btrfs_iget

and nfsd returns NFS4ERR_NOENT to the client instead of NFS4ERR_STALE,
and the client doesn't handle that quite the same way.

This doesn't mean that the change is wrong, but it could mean we need to
fix something else in the path to sanitize the error code.

nfsd_set_fh_dentry already has

	error = nfserr_stale;
	if (PTR_ERR(exp) == -ENOENT)
		return error;

	if (IS_ERR(exp))
		return nfserrno(PTR_ERR(exp));

for a different error case, so duplicating that would work, but I doubt
it is best.  At the very least we should check for valid errors, not
specific invalid ones.

Bruce: do you have an opinion where we should make sure that PUTFH (and
various other requests) returns a valid error code?

Thanks,
NeilBrown

J. Bruce Fields July 22, 2016, 1:59 a.m. UTC | #2

On Fri, Jul 22, 2016 at 11:08:17AM +1000, NeilBrown wrote:
> On Fri, Jun 10 2016, [e-mail] wrote:
> 
> > From: Filipe Manana <[e-mail]>
> >
> > When we attempt to read an inode from disk, we end up always returning an
> > -ESTALE error to the caller regardless of the actual failure reason, which
> > can be an out of memory problem (when allocating a path), some error found
> > when reading from the fs/subvolume btree (like a genuine IO error) or the
> > inode does not exists. So lets start returning the real error code to the
> > callers so that they don't treat all -ESTALE errors as meaning that the
> > inode does not exists (such as during orphan cleanup). This will also be
> > needed for a subsequent patch in the same series dealing with a special
> > fsync case.
> >
> > Signed-off-by: Filipe Manana <[e-mail]>
> 
> SNIP
> 
> > @@ -5594,7 +5602,8 @@ struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
> >  		} else {
> >  			unlock_new_inode(inode);
> >  			iput(inode);
> > -			inode = ERR_PTR(-ESTALE);
> > +			ASSERT(ret < 0);
> > +			inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
> >  		}
> 
> Just a heads-up.  This change breaks NFS :-(
> 
> The change in error code percolates up the call chain:
> 
>  nfs4_pufh->fh_verify->nfsd_set_fh_dentry->exportfs_decode_fh
>     ->btrfs_fh_to_dentry->ntrfs_get_dentry->btrfs_iget
> 
> and nfsd returns NFS4ERR_NOENT to the client instead of NFS4ERR_STALE,
> and the client doesn't handle that quite the same way.
> 
> This doesn't mean that the change is wrong, but it could mean we need to
> fix something else in the path to sanitize the error code.
> 
> nfsd_set_fh_dentry already has
> 
> 	error = nfserr_stale;
> 	if (PTR_ERR(exp) == -ENOENT)
> 		return error;
> 
> 	if (IS_ERR(exp))
> 		return nfserrno(PTR_ERR(exp));
> 
> for a different error case, so duplicating that would work, but I doubt
> it is best.  At the very least we should check for valid errors, not
> specific invalid ones.
> 
> Bruce: do you have an opinion where we should make sure that PUTFH (and
> various other requests) returns a valid error code?

Uh, I guess not.  Maybe exportfs_decode_fh?

Though my kneejerk reaction is to be cranky and wonder why btrfs
suddenly needs a different convention for decode_fh().

--b.
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index e5558d9..1ae0fc8 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3419,10 +3419,10 @@  int btrfs_orphan_cleanup(struct btrfs_root *root)
 		found_key.offset = 0;
 		inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
 		ret = PTR_ERR_OR_ZERO(inode);
-		if (ret && ret != -ESTALE)
+		if (ret && ret != -ENOENT)
 			goto out;
 
-		if (ret == -ESTALE && root == root->fs_info->tree_root) {
+		if (ret == -ENOENT && root == root->fs_info->tree_root) {
 			struct btrfs_root *dead_root;
 			struct btrfs_fs_info *fs_info = root->fs_info;
 			int is_dead_root = 0;
@@ -3458,7 +3458,7 @@  int btrfs_orphan_cleanup(struct btrfs_root *root)
 		 * Inode is already gone but the orphan item is still there,
 		 * kill the orphan item.
 		 */
-		if (ret == -ESTALE) {
+		if (ret == -ENOENT) {
 			trans = btrfs_start_transaction(root, 1);
 			if (IS_ERR(trans)) {
 				ret = PTR_ERR(trans);
@@ -3617,7 +3617,7 @@  static noinline int acls_after_inode_item(struct extent_buffer *leaf,
 /*
  * read an inode from the btree into the in-memory inode
  */
-static void btrfs_read_locked_inode(struct inode *inode)
+static int btrfs_read_locked_inode(struct inode *inode)
 {
 	struct btrfs_path *path;
 	struct extent_buffer *leaf;
@@ -3636,14 +3636,19 @@  static void btrfs_read_locked_inode(struct inode *inode)
 		filled = true;
 
 	path = btrfs_alloc_path();
-	if (!path)
+	if (!path) {
+		ret = -ENOMEM;
 		goto make_bad;
+	}
 
 	memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 
 	ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
-	if (ret)
+	if (ret) {
+		if (ret > 0)
+			ret = -ENOENT;
 		goto make_bad;
+	}
 
 	leaf = path->nodes[0];
 
@@ -3796,11 +3801,12 @@  cache_acl:
 	}
 
 	btrfs_update_iflags(inode);
-	return;
+	return 0;
 
 make_bad:
 	btrfs_free_path(path);
 	make_bad_inode(inode);
+	return ret;
 }
 
 /*
@@ -5585,7 +5591,9 @@  struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
 		return ERR_PTR(-ENOMEM);
 
 	if (inode->i_state & I_NEW) {
-		btrfs_read_locked_inode(inode);
+		int ret;
+
+		ret = btrfs_read_locked_inode(inode);
 		if (!is_bad_inode(inode)) {
 			inode_tree_add(inode);
 			unlock_new_inode(inode);
@@ -5594,7 +5602,8 @@  struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
 		} else {
 			unlock_new_inode(inode);
 			iput(inode);
-			inode = ERR_PTR(-ESTALE);
+			ASSERT(ret < 0);
+			inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
 		}
 	}