af_unix: Fix data-races around user->unix_inflight.
authorKuniyuki Iwashima <kuniyu@amazon.com>
Sat, 2 Sep 2023 00:27:05 +0000 (17:27 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 4 Sep 2023 10:06:16 +0000 (11:06 +0100)
user->unix_inflight is changed under spin_lock(unix_gc_lock),
but too_many_unix_fds() reads it locklessly.

Let's annotate the write/read accesses to user->unix_inflight.

BUG: KCSAN: data-race in unix_attach_fds / unix_inflight

write to 0xffffffff8546f2d0 of 8 bytes by task 44798 on cpu 1:
 unix_inflight+0x157/0x180 net/unix/scm.c:66
 unix_attach_fds+0x147/0x1e0 net/unix/scm.c:123
 unix_scm_to_skb net/unix/af_unix.c:1827 [inline]
 unix_dgram_sendmsg+0x46a/0x14f0 net/unix/af_unix.c:1950
 unix_seqpacket_sendmsg net/unix/af_unix.c:2308 [inline]
 unix_seqpacket_sendmsg+0xba/0x130 net/unix/af_unix.c:2292
 sock_sendmsg_nosec net/socket.c:725 [inline]
 sock_sendmsg+0x148/0x160 net/socket.c:748
 ____sys_sendmsg+0x4e4/0x610 net/socket.c:2494
 ___sys_sendmsg+0xc6/0x140 net/socket.c:2548
 __sys_sendmsg+0x94/0x140 net/socket.c:2577
 __do_sys_sendmsg net/socket.c:2586 [inline]
 __se_sys_sendmsg net/socket.c:2584 [inline]
 __x64_sys_sendmsg+0x45/0x50 net/socket.c:2584
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x3b/0x90 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x6e/0xd8

read to 0xffffffff8546f2d0 of 8 bytes by task 44814 on cpu 0:
 too_many_unix_fds net/unix/scm.c:101 [inline]
 unix_attach_fds+0x54/0x1e0 net/unix/scm.c:110
 unix_scm_to_skb net/unix/af_unix.c:1827 [inline]
 unix_dgram_sendmsg+0x46a/0x14f0 net/unix/af_unix.c:1950
 unix_seqpacket_sendmsg net/unix/af_unix.c:2308 [inline]
 unix_seqpacket_sendmsg+0xba/0x130 net/unix/af_unix.c:2292
 sock_sendmsg_nosec net/socket.c:725 [inline]
 sock_sendmsg+0x148/0x160 net/socket.c:748
 ____sys_sendmsg+0x4e4/0x610 net/socket.c:2494
 ___sys_sendmsg+0xc6/0x140 net/socket.c:2548
 __sys_sendmsg+0x94/0x140 net/socket.c:2577
 __do_sys_sendmsg net/socket.c:2586 [inline]
 __se_sys_sendmsg net/socket.c:2584 [inline]
 __x64_sys_sendmsg+0x45/0x50 net/socket.c:2584
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x3b/0x90 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x6e/0xd8

value changed: 0x000000000000000c -> 0x000000000000000d

Reported by Kernel Concurrency Sanitizer on:
CPU: 0 PID: 44814 Comm: systemd-coredum Not tainted 6.4.0-11989-g6843306689af #6
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014

Fixes: 712f4aad406b ("unix: properly account for FDs passed over unix sockets")
Reported-by: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Acked-by: Willy Tarreau <w@1wt.eu>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/unix/scm.c

index e9dde71..6ff628f 100644 (file)
@@ -64,7 +64,7 @@ void unix_inflight(struct user_struct *user, struct file *fp)
                /* Paired with READ_ONCE() in wait_for_unix_gc() */
                WRITE_ONCE(unix_tot_inflight, unix_tot_inflight + 1);
        }
-       user->unix_inflight++;
+       WRITE_ONCE(user->unix_inflight, user->unix_inflight + 1);
        spin_unlock(&unix_gc_lock);
 }
 
@@ -85,7 +85,7 @@ void unix_notinflight(struct user_struct *user, struct file *fp)
                /* Paired with READ_ONCE() in wait_for_unix_gc() */
                WRITE_ONCE(unix_tot_inflight, unix_tot_inflight - 1);
        }
-       user->unix_inflight--;
+       WRITE_ONCE(user->unix_inflight, user->unix_inflight - 1);
        spin_unlock(&unix_gc_lock);
 }
 
@@ -99,7 +99,7 @@ static inline bool too_many_unix_fds(struct task_struct *p)
 {
        struct user_struct *user = current_user();
 
-       if (unlikely(user->unix_inflight > task_rlimit(p, RLIMIT_NOFILE)))
+       if (unlikely(READ_ONCE(user->unix_inflight) > task_rlimit(p, RLIMIT_NOFILE)))
                return !capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN);
        return false;
 }