[PATCH] fix oops in namespace.c

[PATCH] fix oops in namespace.c

[Andries.Brouwer]

Number four in the series of namespace.c fixes:

A familar type of Oops: d_path() can return an error ENAMETOOLONG,
and if we fail to test a segfault occurs.

So we must test. What we do is a different matter.
Rather arbitrarily I return the string " (too long)"
for use in /proc/mounts.

Andries

--- namespace.c~	Mon May 19 01:50:48 2003
+++ namespace.c	Mon May 19 01:51:24 2003
@@ -217,6 +217,8 @@
 	if (!path_buf)
 		return -ENOMEM;
 	path = d_path(mnt->mnt_root, mnt, path_buf, PAGE_SIZE);
+	if (IS_ERR(path))
+		path = " (too long)";
 
 	mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
 	seq_putc(m, ' ');

Re: [PATCH] fix oops in namespace.c

[viro]

On Mon, May 19, 2003 at 12:57:37AM +0200, Andries.Brouwer@cwi.nl wrote:
> Number four in the series of namespace.c fixes:
> 
> A familar type of Oops: d_path() can return an error ENAMETOOLONG,
> and if we fail to test a segfault occurs.
> 
> So we must test. What we do is a different matter.
> Rather arbitrarily I return the string " (too long)"
> for use in /proc/mounts.

There's a better fix.  Since we are going to use seq_...() anyway,
we both have a buffer and length.  Moreover, if there's not enough
space to store the result, generic seq_file code will take care
about expanding buffer and calling us again.

IOW, we need to add seq_path(seq_file *, vfsmount *, dentry *) that would
DTRT.  That would have a benefit of avoiding extra allocation/freeing.

I'll send such patch in ~20 minutes.  BTW, it's also needed for /proc/swaps
and can be used for /proc/<pid>/maps if we switch it to seq_file (worth
doing anyway).

PS: math.psu.edu account is alive, but I'm not checking it often these
days, so Cc there is not too useful.