Re: 2.6.22-rc6(mm1) Unable to handle kernel NULL pointer dereference- git-bisect result

From: Linus Torvalds
Date: Sun Jul 08 2007 - 14:31:19 EST




On Sun, 8 Jul 2007, Linus Torvalds wrote:
>
> No.
>
> notify_change() does *not* do permission checks for
> ATTR_CTIME/MTIME/ATIME.
>
> It does them for the "xxx_SET" attributes, but MTIME/ATIME is expected to
> change when other things change, so notify_change() expects that those
> _other_ changes have been validated from a security standpoint!
>
> utimes() is special.

This might be an acceptable patch.

Al? What do you think? Basically, it says that

- the file owner can always set MTIME/ATIME, of course

- non-owners can set it only when they have write permissions, and if it
was a file descriptor, the only way for us to know that they have write
permissions is if it's opened writably, which is hopefully equivalent
to that MAY_WRITE test (except the MAY_WRITE test was done at _open_
time).

This would seem to be the minimal change, and I think it's right.

Linus

---
fs/utimes.c | 13 ++++++++++---
1 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/fs/utimes.c b/fs/utimes.c
index 480f7c8..b3c8895 100644
--- a/fs/utimes.c
+++ b/fs/utimes.c
@@ -106,9 +106,16 @@ long do_utimes(int dfd, char __user *filename, struct timespec *times, int flags
if (IS_IMMUTABLE(inode))
goto dput_and_out;

- if (current->fsuid != inode->i_uid &&
- (error = vfs_permission(&nd, MAY_WRITE)) != 0)
- goto dput_and_out;
+ if (current->fsuid != inode->i_uid) {
+ if (f) {
+ if (!(f->f_mode & FMODE_WRITE))
+ goto dput_and_out;
+ } else {
+ error = vfs_permission(&nd, MAY_WRITE);
+ if (error)
+ goto dput_and_out;
+ }
+ }
}
mutex_lock(&inode->i_mutex);
error = notify_change(dentry, &newattrs);
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/