Re: KCSAN: data-race in __alloc_file / __alloc_file

From: Eric Dumazet <edumazet@google.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>,
	syzbot <syzbot+3ef049d50587836c0606@syzkaller.appspotmail.com>,
	Marco Elver <elver@google.com>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	syzkaller-bugs <syzkaller-bugs@googlegroups.com>,
	Al Viro <viro@zeniv.linux.org.uk>
Subject: Re: KCSAN: data-race in __alloc_file / __alloc_file
Date: Fri, 8 Nov 2019 10:02:22 -0800	[thread overview]
Message-ID: <CANn89iLN758CpQPKcd++NLdj62LS-ekiEUV91VREzMsamLn9bw@mail.gmail.com> (raw)
In-Reply-To: <CANn89i+RrngUr11_iOYDuqDvAZnPfG3ieJR025M78uhiwEPuvQ@mail.gmail.com>

On Fri, Nov 8, 2019 at 9:55 AM Eric Dumazet <edumazet@google.com> wrote:
>
> On Fri, Nov 8, 2019 at 9:53 AM Eric Dumazet <edumazet@google.com> wrote:
> >
> > On Fri, Nov 8, 2019 at 9:39 AM Linus Torvalds
> > <torvalds@linux-foundation.org> wrote:
> >
> > > I'd hope that there is some way to mark the cases we know about where
> > > we just have a flag. I'm not sure what KCSAN uses right now - is it
> > > just the "volatile" that makes KCSAN ignore it, or are there other
> > > ways to do it?
> >
> > I dunno, Marco will comment on this.
> >
> > I personally like WRITE_ONCE() since it adds zero overhead on generated code,
> > and is the facto accessor we used for many years (before KCSAN was conceived)
> >
>
> BTW, I would love an efficient ADD_ONCE(variable, value)
>
> Using WRITE_ONCE(variable, variable + value) is not good, since it can
> not use the
> optimized instructions operating directly on memory.

Another interesting KCSAN report :

static inline s64 percpu_counter_read_positive(struct percpu_counter *fbc)
{
        s64 ret = fbc->count;   // data-race ....

        barrier();              /* Prevent reloads of fbc->count */
        if (ret >= 0)
                return ret;
        return 0;
}

How was this code supposed to work at all on 32bit arches ???

Using READ_ONCE(fbc->count) alone will not help.

BUG: KCSAN: data-race in ip6_dst_gc / ip6_dst_gc

read to 0xffff88811dd6298c of 4 bytes by task 10060 on cpu 1:
 dst_entries_get_fast include/net/dst_ops.h:47 [inline]
 ip6_dst_gc+0xf6/0x220 net/ipv6/route.c:3167
 dst_alloc+0x104/0x149 net/core/dst.c:85
 ip6_dst_alloc+0x3d/0x80 net/ipv6/route.c:353
 ip6_rt_cache_alloc+0x8d/0x340 net/ipv6/route.c:1338
 ip6_pol_route+0x4ec/0x5c0 net/ipv6/route.c:2217
 ip6_pol_route_output+0x48/0x60 net/ipv6/route.c:2452
 fib6_rule_lookup+0x95/0x470 net/ipv6/fib6_rules.c:113
 ip6_route_output_flags_noref+0x16b/0x230 net/ipv6/route.c:2484
 ip6_route_output_flags+0x50/0x1a0 net/ipv6/route.c:2497
 ip6_dst_lookup_tail+0x25d/0xc30 net/ipv6/ip6_output.c:1052
 ip6_dst_lookup_flow+0x68/0x120 net/ipv6/ip6_output.c:1153
 rawv6_sendmsg+0x82c/0x21e0 net/ipv6/raw.c:928
 inet_sendmsg+0x6d/0x90 net/ipv4/af_inet.c:807
 sock_sendmsg_nosec net/socket.c:637 [inline]
 sock_sendmsg+0x9f/0xc0 net/socket.c:657
 kernel_sendmsg+0x4d/0x70 net/socket.c:677
 sock_no_sendpage+0xda/0x110 net/core/sock.c:2742
 kernel_sendpage+0x7b/0xc0 net/socket.c:3682
 sock_sendpage+0x6c/0x90 net/socket.c:935
 pipe_to_sendpage+0x102/0x180 fs/splice.c:449
 splice_from_pipe_feed fs/splice.c:500 [inline]
 __splice_from_pipe+0x248/0x480 fs/splice.c:624
 splice_from_pipe+0xbb/0x100 fs/splice.c:659
 generic_splice_sendpage+0x45/0x60 fs/splice.c:829
 do_splice_from fs/splice.c:848 [inline]
 direct_splice_actor+0xa0/0xc0 fs/splice.c:1020
 splice_direct_to_actor+0x215/0x510 fs/splice.c:975
 do_splice_direct+0x161/0x1e0 fs/splice.c:1063
 do_sendfile+0x384/0x7f0 fs/read_write.c:1464
 __do_sys_sendfile64 fs/read_write.c:1525 [inline]
 __se_sys_sendfile64 fs/read_write.c:1511 [inline]
 __x64_sys_sendfile64+0x12a/0x140 fs/read_write.c:1511
 do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290
 entry_SYSCALL_64_after_hwframe+0x44/0xa9