Skip to contentSkip to conversation

cifs: Clean up an error code in cifs_root_iget()

PATCH In review Dan Carpenter 2 participants 3 messages Feb 28, 2019

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

PATCH 1/1 cifs: Clean up an error code in cifs_root_iget()

This patch silences a Smatch warning:

fs/cifs/inode.c:1094 cifs_root_iget() warn: passing zero to 'ERR_PTR'

The shouldn't have a noticeable effect on runtime, it's basically a cleanup. The code is checking to ensure that cifs_get_inode_info_unix() returns -EOPNOTSUPP when we have the UNIX extensions. From the patch description in commit b5b374eab11e ("Workaround Mac server problem") this affects Macs.

Presumably most of the time "rc" is zero, which means we return ERR_PTR(0) which is NULL. This cifs_root_iget() function is only called from cifs_read_super() and if we return NULL that causes d_make_root() to return NULL so in the end we fail with -ENOMEM.

After this patch is applied we instead return with -EINVAL.

---
 fs/cifs/inode.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
fs/cifs/inode.c +3 -1
index 0f53ecb071ac..e40c554bb2f3 100644
@@ -1080,8 +1080,10 @@ struct inode *cifs_root_iget(struct super_block *sb)
 	if (tcon->unix_ext) {
 		rc = cifs_get_inode_info_unix(&inode, path, sb, xid);
 		/* some servers mistakenly claim POSIX support */
-		if (rc != -EOPNOTSUPP)
+		if (rc != -EOPNOTSUPP) {
+			rc = -EINVAL;
 			goto iget_no_retry;
+		}
 		cifs_dbg(VFS, "server does not support POSIX extensions");
 		tcon->unix_ext = false;
 	}
Signature · 1 line
2.17.1
#1 lore raw

Won't that change behavior and cause the success case (server returned 0, and does support POSIX and &inode pointer is ok) to now return -EINVAL.

Before if rc was 0 and inode was filled in we wouldn't execute this else block:

    } else if (rc) {
        iget_failed(inode);
        inode = ERR_PTR(rc);
    }

But with your change, won't we return EINVAL for the success case?

How about just changing:

iget_no_retry:

    if (!inode) {
        inode = ERR_PTR(rc);
        goto out;
    }

to

iget_no_retry:

    if (!inode) {
        if (rc = 0)
              rc = -EINVAL;
        inode = ERR_PTR(rc);
        goto out;
    }
Quoted 40 lines · Dan Carpenter via samba-technical

This patch silences a Smatch warning:

fs/cifs/inode.c:1094 cifs_root_iget() warn: passing zero to 'ERR_PTR'

The shouldn't have a noticeable effect on runtime, it's basically a cleanup. The code is checking to ensure that cifs_get_inode_info_unix() returns -EOPNOTSUPP when we have the UNIX extensions. From the patch description in commit b5b374eab11e ("Workaround Mac server problem") this affects Macs.

Presumably most of the time "rc" is zero, which means we return ERR_PTR(0) which is NULL. This cifs_root_iget() function is only called from cifs_read_super() and if we return NULL that causes d_make_root() to return NULL so in the end we fail with -ENOMEM.

After this patch is applied we instead return with -EINVAL.

---
 fs/cifs/inode.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
fs/cifs/inode.c +3 -2
index 0f53ecb071ac..e40c554bb2f3 100644
@@ -1080,8 +1080,10 @@ struct inode *cifs_root_iget(struct super_block *sb)
        if (tcon->unix_ext) {
                rc = cifs_get_inode_info_unix(&inode, path, sb, xid);
                /* some servers mistakenly claim POSIX support */
-               if (rc != -EOPNOTSUPP)
+               if (rc != -EOPNOTSUPP) {
+                       rc = -EINVAL;
                        goto iget_no_retry;
+               }
                cifs_dbg(VFS, "server does not support POSIX extensions");
                tcon->unix_ext = false;
        }
--
Quoted 3 lines

2.17.1

--
Thanks,

Steve

#2 lore raw
Quoted 4 lines · Steve French

Won't that change behavior and cause the success case (server returned 0, and does support POSIX and &inode pointer is ok) to now return -EINVAL.

You're right. I'm really sorry. I don't know why I misread the code...

Also looking at it now, this was a false positive in Smatch. I have been working on that code in the past couple days and I don't have the warning in my latest build.

regards,
dan carpenter