From: Siddharth Chintamaneni <sidchintamaneni@gmail.com>
To: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Cc: Barret Rhoden <brho@google.com>,
bpf@vger.kernel.org, alexei.starovoitov@gmail.com,
daniel@iogearbox.net, olsajiri@gmail.com, andrii@kernel.org,
yonghong.song@linux.dev, rjsu26@vt.edu, sairoop@vt.edu,
miloc@vt.edu,
syzbot+8bdfc2c53fb2b63e1871@syzkaller.appspotmail.com
Subject: Re: [PATCH v3 bpf-next 1/2] bpf: Patch to Fix deadlocks in queue and stack maps
Date: Thu, 16 May 2024 11:31:38 -0400 [thread overview]
Message-ID: <CAE5sdEiR2nNMjyZBwEBuFovwW+YWStP1P_nZRJUOre5K1Z0AaQ@mail.gmail.com> (raw)
In-Reply-To: <CAP01T77yXv+trVCryMDK-9VghnRrNQpoSpp_Z-OLmQz9omHRGQ@mail.gmail.com>
On Thu, 16 May 2024 at 10:34, Kumar Kartikeya Dwivedi <memxor@gmail.com> wrote:
>
> On Thu, 16 May 2024 at 16:05, Barret Rhoden <brho@google.com> wrote:
> >
> > On 5/14/24 08:40, Siddharth Chintamaneni wrote:
> > [...]
> > > +static inline int map_lock_inc(struct bpf_queue_stack *qs)
> > > +{
> > > + unsigned long flags;
> > > +
> > > + preempt_disable();
> > > + local_irq_save(flags);
> > > + if (unlikely(__this_cpu_inc_return(*(qs->map_locked)) != 1)) {
> > > + __this_cpu_dec(*(qs->map_locked));
> > > + local_irq_restore(flags);
> > > + preempt_enable();
> > > + return -EBUSY;
> > > + }
> > > +
> > > + local_irq_restore(flags);
> > > + preempt_enable();
> >
> > it looks like you're taking the approach from kernel/bpf/hashtab.c to
> > use a per-cpu lock before grabbing the real lock. but in the success
> > case here (where you incremented the percpu counter), you're enabling
> > irqs and preemption.
> >
> > what happens if you get preempted right after this? you've left the
> > per-cpu bit set, but then you run on another cpu.
>
> Great catch, that's a bug. It's not a problem when BPF programs call
> this, as migration is disabled for them (but it's questionable whether
> we should keep preemption enabled between map_inc/dec increasing the
> chances of conflicts on the same CPU), but it's certainly a problem
> from the syscall path.
>
I was also thinking from the BPF programs perspective as migration is
disabled on them. I will fix this.
> >
> > possible alternative: instead of splitting the overall lock into "grab
> > percpu lock, then grab real lock", have a single function for both,
> > similar to htab_lock_bucket(). and keep irqs and preemption off from
> > the moment you start attempting the overall lock until you completely
> > unlock.
>
> +1.
>
> >
> > barret
> >
next prev parent reply other threads:[~2024-05-16 15:31 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-05-14 12:40 [PATCH v3 bpf-next 2/2] selftests/bpf: Added selftests to check deadlocks in queue and stack map Siddharth Chintamaneni
2024-05-14 12:40 ` [PATCH v3 bpf-next 1/2] bpf: Patch to Fix deadlocks in queue and stack maps Siddharth Chintamaneni
2024-05-15 17:32 ` Kumar Kartikeya Dwivedi
2024-05-16 14:04 ` Barret Rhoden
2024-05-16 14:34 ` Kumar Kartikeya Dwivedi
2024-05-16 15:31 ` Siddharth Chintamaneni [this message]
2024-05-17 1:53 ` Hou Tao
2024-05-17 3:32 ` Siddharth Chintamaneni
2024-05-15 17:02 ` [PATCH v3 bpf-next 2/2] selftests/bpf: Added selftests to check deadlocks in queue and stack map Kumar Kartikeya Dwivedi
2024-05-15 17:44 ` Siddharth Chintamaneni
2024-05-15 17:56 ` Kumar Kartikeya Dwivedi
2024-05-15 17:58 ` Kumar Kartikeya Dwivedi
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAE5sdEiR2nNMjyZBwEBuFovwW+YWStP1P_nZRJUOre5K1Z0AaQ@mail.gmail.com \
--to=sidchintamaneni@gmail.com \
--cc=alexei.starovoitov@gmail.com \
--cc=andrii@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=brho@google.com \
--cc=daniel@iogearbox.net \
--cc=memxor@gmail.com \
--cc=miloc@vt.edu \
--cc=olsajiri@gmail.com \
--cc=rjsu26@vt.edu \
--cc=sairoop@vt.edu \
--cc=syzbot+8bdfc2c53fb2b63e1871@syzkaller.appspotmail.com \
--cc=yonghong.song@linux.dev \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).