Re: KCSAN: data-race in __alloc_file / __alloc_file

[Eric Dumazet <eric.dumazet@gmail.com>]

On 11/8/19 5:16 AM, syzbot wrote:
> Hello,
> 
> syzbot found the following crash on:
> 
> HEAD commit:    05f22368 x86, kcsan: Enable KCSAN for x86
> git tree:       https://github.com/google/ktsan.git kcsan
> console output: https://syzkaller.appspot.com/x/log.txt?x=10d7fd88e00000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=87d111955f40591f
> dashboard link: https://syzkaller.appspot.com/bug?extid=3ef049d50587836c0606
> compiler:       gcc (GCC) 9.0.0 20181231 (experimental)
> 
> Unfortunately, I don't have any reproducer for this crash yet.
> 
> IMPORTANT: if you fix the bug, please add the following tag to the commit:
> Reported-by: syzbot+3ef049d50587836c0606@syzkaller.appspotmail.com
> 
> ==================================================================
> BUG: KCSAN: data-race in __alloc_file / __alloc_file
> 
> write to 0xffff8880bb157398 of 4 bytes by task 10993 on cpu 0:
>  get_cred include/linux/cred.h:253 [inline]
>  __alloc_file+0x74/0x210 fs/file_table.c:105
>  alloc_empty_file+0x8f/0x180 fs/file_table.c:151
>  alloc_file+0x4e/0x2b0 fs/file_table.c:193
>  alloc_file_pseudo+0x11c/0x1b0 fs/file_table.c:232
>  anon_inode_getfile fs/anon_inodes.c:91 [inline]
>  anon_inode_getfile+0x103/0x1d0 fs/anon_inodes.c:74
>  __do_sys_perf_event_open+0xd32/0x1ac0 kernel/events/core.c:11100
>  __se_sys_perf_event_open kernel/events/core.c:10867 [inline]
>  __x64_sys_perf_event_open+0x70/0x90 kernel/events/core.c:10867
>  do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290
>  entry_SYSCALL_64_after_hwframe+0x44/0xa9
> 
> write to 0xffff8880bb157398 of 4 bytes by task 11004 on cpu 1:
>  get_cred include/linux/cred.h:253 [inline]
>  __alloc_file+0x74/0x210 fs/file_table.c:105
>  alloc_empty_file+0x8f/0x180 fs/file_table.c:151
>  path_openat+0x74/0x36e0 fs/namei.c:3514
>  do_filp_open+0x11e/0x1b0 fs/namei.c:3555
>  do_sys_open+0x3b3/0x4f0 fs/open.c:1097
>  __do_sys_open fs/open.c:1115 [inline]
>  __se_sys_open fs/open.c:1110 [inline]
>  __x64_sys_open+0x55/0x70 fs/open.c:1110
>  do_syscall_64+0xcc/0x370 arch/x86/entry/common.c:290
>  entry_SYSCALL_64_after_hwframe+0x44/0xa9
> 
> Reported by Kernel Concurrency Sanitizer on:
> CPU: 1 PID: 11004 Comm: syz-executor.5 Not tainted 5.4.0-rc3+ #0
> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
> ==================================================================
> 
> 
> ---
> This bug is generated by a bot. It may contain errors.
> See https://goo.gl/tpsmEJ for more information about syzbot.
> syzbot engineers can be reached at syzkaller@googlegroups.com.
> 
> syzbot will keep track of this bug report. See:
> https://goo.gl/tpsmEJ#status for how to communicate with syzbot.

Linus, what do you think of the following fix ?

I also took the opportunity avoiding dirtying a cache line if this was possible.

diff --git a/include/linux/cred.h b/include/linux/cred.h
index 18639c069263fbe79dfd5a36163c656dca5da220..01b5b7d4e054ddca0df676dc1ceb068e5d71a3f8 100644
--- a/include/linux/cred.h
+++ b/include/linux/cred.h
@@ -250,7 +250,14 @@ static inline const struct cred *get_cred(const struct cred *cred)
        if (!cred)
                return cred;
        validate_creds(cred);
-       nonconst_cred->non_rcu = 0;
+
+       /*
+        * Avoid dirtying one cache line. The WRITE_ONCE() also pairs
+        * with itself, since we run without protection of a lock.
+        */
+       if (READ_ONCE(nonconst_cred->non_rcu))
+               WRITE_ONCE(nonconst_cred->non_rcu, 0);
+
        return get_new_cred(nonconst_cred);
 }
 
@@ -262,7 +269,14 @@ static inline const struct cred *get_cred_rcu(const struct cred *cred)
        if (!atomic_inc_not_zero(&nonconst_cred->usage))
                return NULL;
        validate_creds(cred);
-       nonconst_cred->non_rcu = 0;
+
+       /*
+        * Avoid dirtying one cache line. The WRITE_ONCE() also pairs
+        * with itself, since we run without protection of a lock.
+        */
+       if (READ_ONCE(nonconst_cred->non_rcu))
+               WRITE_ONCE(nonconst_cred->non_rcu, 0);
+
        return cred;
 }