Skip to contentSkip to conversation

memory corruption after "Fix race when checking i_size on direct i/o read"

PATCH In review J. Bruce Fields 2 participants 5 messages Feb 8, 2014 – Feb 11, 2014

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

I'm getting warnings about corruption of the kmalloc-32 slab on an nfs server while running some regression tests.

A bisect lands on 9fe55eea7e4b444bafc42fa0000cc2d1d2847275 "Fix race when checking i_size on direct i/o read". Is there any known problem here?

It doesn't reproduce immediately and it's possible I could have made a mistake on the bisect.

I'll verify and try to narrow down the reproducer.

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

#1 lore raw
Quoted 9 lines · bfields

I'm getting warnings about corruption of the kmalloc-32 slab on an nfs server while running some regression tests.

A bisect lands on 9fe55eea7e4b444bafc42fa0000cc2d1d2847275 "Fix race when checking i_size on direct i/o read". Is there any known problem here?

It doesn't reproduce immediately and it's possible I could have made a mistake on the bisect.

Argh, apologies, yes--I'm able to reproduce the corruption before that commit. I'll keep looking....

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

#2 lore raw

PATCH 1/1 nfsd4: fix acl buffer overrun

From: "J. Bruce Fields" <[email protected]>

4ac7249ea5a0ceef9f8269f63f33cc873c3fac61 "nfsd: use get_acl and ->set_acl" forgets to set the size in the case get_acl() succeeds, so _posix_to_nfsv4_one() can then write past the end of its allocation. Symptoms were slab corruption warnings.

Also, some minor cleanup while we're here. (Among other things, note that the first few lines guarantee that pacl is non-NULL.)

---
 fs/nfsd/nfs4acl.c |    9 +++------
 1 file changed, 3 insertions(+), 6 deletions(-)
Quoted 8 lines · J. Bruce Fields
Quoted 7 lines · bfields

I'm getting warnings about corruption of the kmalloc-32 slab on an nfs server while running some regression tests.

A bisect lands on 9fe55eea7e4b444bafc42fa0000cc2d1d2847275 "Fix race when checking i_size on direct i/o read". Is there any known problem here?

Quoted 5 lines
Quoted 2 lines

It doesn't reproduce immediately and it's possible I could have made a mistake on the bisect.

Argh, apologies, yes--I'm able to reproduce the corruption before that commit. I'll keep looking....

Fortunately my mistake was at the end of the bisect so the offending commit was nearby (and clearly labeled "nfsd:..."). This fixes the regression I was seeing. I intend to include it in a pull request for 3.14 soon.

fs/nfsd/nfs4acl.c +3 -6
index d3a5871..d190e33 100644
@@ -151,17 +151,15 @@ nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry,
 		pacl = posix_acl_from_mode(inode->i_mode, GFP_KERNEL);
 		if (IS_ERR(pacl))
 			return PTR_ERR(pacl);
-		/* allocate for worst case: one (deny, allow) pair each: */
-		size += 2 * pacl->a_count;
 	}
+	/* allocate for worst case: one (deny, allow) pair each: */
+	size += 2 * pacl->a_count;
 
 	if (S_ISDIR(inode->i_mode)) {
 		flags = NFS4_ACL_DIR;
 		dpacl = get_acl(inode, ACL_TYPE_DEFAULT);
 		if (dpacl)
 			size += 2 * dpacl->a_count;
-	} else {
-		dpacl = NULL;
 	}
 
 	*acl = nfs4_acl_new(size);
@@ -170,8 +168,7 @@ nfsd4_get_nfs4_acl(struct svc_rqst *rqstp, struct dentry *dentry,
 		goto out;
 	}
 
-	if (pacl)
-		_posix_to_nfsv4_one(pacl, *acl, flags & ~NFS4_ACL_TYPE_DEFAULT);
+	_posix_to_nfsv4_one(pacl, *acl, flags & ~NFS4_ACL_TYPE_DEFAULT);
 
 	if (dpacl)
 		_posix_to_nfsv4_one(dpacl, *acl, flags | NFS4_ACL_TYPE_DEFAULT);
Signature · 1 line
1.7.9.5
#3 lore raw

Re: [PATCH] nfsd4: fix acl buffer overrun

Looks good and sorry for the regression.

#4 lore raw

Re: [PATCH] nfsd4: fix acl buffer overrun

Quoted 1 line · Christoph Hellwig

Looks good and sorry for the regression.

Well, my fault for not reviewing the patch, I saw it and didn't give it more than a skim. Not sure I would have caught that, though.

The only thing I can think of that might have made it easier to catch was breaking up the patch (v4 could probably have been done separately from v2/v3?). But I still probably would've overlooked it....

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