Re: [syzbot] [mptcp?] general protection fault in proc_scheduler

On Sat, Jan 4, 2025 at 9:21 PM Al Viro [off-list ref] wrote:

On Sat, Jan 04, 2025 at 08:11:49PM +0100, Matthieu Baerts wrote:

quoted

quoted

quoted

+       if (S_ISREG(file_inode(file)->i_mode))

                ^^^^^^^^^

quoted

quoted

quoted

+               return;

... won't help, since the file in question *is* a regular file.  IOW, it's
a wrong predicate here.

On my side, it looks like I'm not able to reproduce the issue with this
patch. Without it, it is very easy to reproduce it. (But I don't know if
there are other consequences that would avoid the issue to happen: when
looking at the logs, with the patch, I don't have heaps of "Process
accounting resumed" messages that I had before.)

Unsurprisingly so, since it rejects all regular files due to a typo;
fix that and you'll see that the oops is still there.

The real issue (and the one that affects more than just this scenario) is
the use of current->nsproxy->net to get to the damn thing.

According to grep, we have many other places directly reading
current->nsproxy->net_ns
For instance in net/sctp/sysctl.c
Should we change them all ?

Perhaps an alternative would be to add a generic check in
proc_sys_call_handler()

diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
index 27a283d85a6e7df1a7edbfb513ce75832363e2e6..84968b10ce86e7fd88c6e3c43f52b601394b056f

100644

--- a/fs/proc/proc_sysctl.c
+++ b/fs/proc/proc_sysctl.c

@@ -576,6 +576,8 @@ static ssize_t proc_sys_call_handler(struct kiocb

*iocb, struct iov_iter *iter,
        error = -EINVAL;
        if (!table->proc_handler)
                goto out;
+       if (unlikely(current->flags & PF_EXITING))
+               goto out;

        /* don't even try if the size is too large */
        error = -ENOMEM;


Thanks.