Re: [syzbot] [fs?] kernel BUG in sctp_getsockopt_peeloff_common

On Mon, Dec 1, 2025 at 5:09 PM syzbot
<[email protected]> wrote:
>
> syzbot has bisected this issue to:
>
> commit 457528eb27c3a3053181939ca65998477cc39c49
> Author: Christian Brauner <[email protected]>
> Date:   Sun Nov 23 16:33:47 2025 +0000
>
>     net/sctp: convert sctp_getsockopt_peeloff_common() to FD_PREPARE()
>
> bisection log:  https://syzkaller.appspot.com/x/bisect.txt?x=1136a512580000
> start commit:   7d31f578f323 Add linux-next specific files for 20251128
> git tree:       linux-next
> final oops:     https://syzkaller.appspot.com/x/report.txt?x=1336a512580000
> console output: https://syzkaller.appspot.com/x/log.txt?x=1536a512580000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=6336d8e94a7c517d
> dashboard link: https://syzkaller.appspot.com/bug?extid=984a5c208d87765b2ee7
> syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=16a2322c580000
> C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=12a3c512580000
>
> Reported-by: [email protected]
> Fixes: 457528eb27c3 ("net/sctp: convert sctp_getsockopt_peeloff_common() to FD_PREPARE()")
This commit seems to no longer exist.

But I triggered a similar call trace with FAULT_INJECTION on net-next.git:

 [] FAULT_INJECTION: forcing a failure.
 [] Call Trace:
 []  <TASK>
 []  dump_stack_lvl+0x180/0x1b0
 []  should_fail_ex+0x520/0x650
 []  should_failslab+0xc2/0x120
 []  kmem_cache_alloc_lru_noprof+0x7a/0x780
 []  d_alloc_pseudo+0x1d/0xc0
 []  alloc_file_pseudo+0xbe/0x220
 []  sock_alloc_file+0x53/0x220
 []  __sys_socket+0x1be/0x320
 [] VFS_BUG_ON_INODE(inode_state_read_once(inode) & I_CLEAR)
encountered for inode ffff888054f9a900
 [] ------------[ cut here ]------------
 [] kernel BUG at fs/inode.c:1971!
 [] Call Trace:
 []  <TASK>
 []  iput+0x35/0x40
 []  __sock_release+0x20b/0x270
 []  __sys_socket+0x276/0x320
 []  __x64_sys_socket+0x72/0xb0

which was caused by:

commit 245f0d1c622b0183ce4f44b3e39aeacf78fae594
Author: Christian Brauner <[email protected]>
Date:   Sun Nov 23 17:33:48 2025 +0100

    net/socket: convert sock_map_fd() to FD_ADD()

static int sock_map_fd(struct socket *sock, int flags)
{
        int fd;

        fd = FD_ADD(flags, sock_alloc_file(sock, flags, NULL));
        if (fd < 0)
                sock_release(sock);
        return fd;
}

The allocation failure in sock_alloc_file() will call sock_release(),
and it should not be called again in sock_map_fd().

It could be fixed by:

diff --git a/net/socket.c b/net/socket.c
index 809ef372727b..0c2b03cec83d 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -503,12 +503,13 @@ EXPORT_SYMBOL(sock_alloc_file);

 static int sock_map_fd(struct socket *sock, int flags)
 {
-       int fd;
+       struct file *file;

-       fd = FD_ADD(flags, sock_alloc_file(sock, flags, NULL));
-       if (fd < 0)
-               sock_release(sock);
-       return fd;
+       file = sock_alloc_file(sock, flags, NULL);
+       if (IS_ERR(file))
+               return PTR_ERR(file);
+
+       return FD_ADD(flags, file);
 }