Re: BUG: Bad page map in process udevd (anon_vma: (null)) in 2.6.38-rc4

[Arnaldo Carvalho de Melo <acme@redhat.com>]

Em Fri, Feb 18, 2011 at 10:48:18AM -0800, Linus Torvalds escreveu:
> On Fri, Feb 18, 2011 at 10:08 AM, Eric W. Biederman
> <ebiederm@xmission.com> wrote:
> >
> > I am still getting programs segfaulting but that is happening on other
> > machines running on older kernels so I am going to chalk that up to a
> > buggy test and a false positive.
> 
> Ok.
> 
> > I am have OOM problems getting my tests run to complete.  On a good
> > day that happens about 1 time in 3 right now.  I'm guess I will have
> > to turn off DEBUG_PAGEALLOC to get everything to complete.
> > DEBUG_PAGEALLOC causes us to use more memory doesn't it?
> 
> It does use a bit more memory, but it shouldn't be _that_ noticeable.
> The real cost of DEBUG_PAGEALLOC is all the crazy page table
> operations and TLB flushes we do for each allocation/deallocation. So
> DEBUG_PAGEALLOC is very CPU-intensive, but it shouldn't have _that_
> much of a memory overhead - just some trivial overhead due to not
> being able to use largepages for the normal kernel identity mappings.
> 
> But there might be some other interaction with OOM that I haven't thought about.
> 
> > The most interesting thing I have right now is a networking lockdep
> > issue.  Does anyone know what is going on there?
> 
> This seems to be a fairly straightforward bug.
> 
> In net/ipv4/inet_timewait_sock.c we have this:
> 
>   /* These are always called from BH context.  See callers in
>    * tcp_input.c to verify this.
>    */
> 
>   /* This is for handling early-kills of TIME_WAIT sockets. */
>   void inet_twsk_deschedule(struct inet_timewait_sock *tw,
>                             struct inet_timewait_death_row *twdr)
>   {
>           spin_lock(&twdr->death_lock);
>           ..
> 
> and the intention is clearly that that spin_lock is BH-safe because
> it's called from BH context.
> 
> Except that clearly isn't true. It's called from a worker thread:
> 
> > stack backtrace:
> > Pid: 10833, comm: kworker/u:1 Not tainted 2.6.38-rc4-359399.2010AroraKernelBeta.fc14.x86_64 #1
> > Call Trace:
> >  [<ffffffff81460e69>] ? inet_twsk_deschedule+0x29/0xa0
> >  [<ffffffff81460fd6>] ? inet_twsk_purge+0xf6/0x180
> >  [<ffffffff81460f10>] ? inet_twsk_purge+0x30/0x180
> >  [<ffffffff814760fc>] ? tcp_sk_exit_batch+0x1c/0x20
> >  [<ffffffff8141c1d3>] ? ops_exit_list.clone.0+0x53/0x60
> >  [<ffffffff8141c520>] ? cleanup_net+0x100/0x1b0
> >  [<ffffffff81068c47>] ? process_one_work+0x187/0x4b0
> >  [<ffffffff81068be1>] ? process_one_work+0x121/0x4b0
> >  [<ffffffff8141c420>] ? cleanup_net+0x0/0x1b0
> >  [<ffffffff8106a65c>] ? worker_thread+0x15c/0x330
> 
> so it can deadlock with a BH happening at the same time, afaik.
> 
> The code (and comment) is all from 2005, it looks like the BH->worker
> thread has broken the code. But somebody who knows that code better
> should take a deeper look at it.
> 
> Added acme to the cc, since the code is attributed to him back in 2005
> ;). Although I don't know how active he's been in networking lately
> (seems to be all perf-related). Whatever, it can't hurt.

Original code is ANK's, I just made it possible to use with DCCP, and
yeah, the smiley is appropriate, something 6 years old and the world
around it changing continually... well, thanks for the git blame ;-)

- Arnaldo