On Mon, Jul 22, 2024 at 12:28:42PM -0700, Kuniyuki Iwashima wrote:
> syzkaller reported a warning in bcm_connect() below. [0]
>
> The repro calls connect() to vxcan1, removes vxcan1, and calls
> connect() with ifindex == 0.
>
> Calling connect() for a BCM socket allocates a proc entry.
> Then, bcm_sk(sk)->bound is set to 1 to prevent further connect().
>
> However, removing the bound device resets bcm_sk(sk)->bound to 0
> in bcm_notify().
>
> The 2nd connect() tries to allocate a proc entry with the same
> name and sets NULL to bcm_sk(sk)->bcm_proc_read, leaking the
> original proc entry.
>
> Since the proc entry is available only for connect()ed sockets,
> let's clean up the entry when the bound netdev is unregistered.
>
> [0]:
> proc_dir_entry 'can-bcm/2456' already registered
> WARNING: CPU: 1 PID: 394 at fs/proc/generic.c:376 proc_register+0x645/0x8f0 fs/proc/generic.c:375
> Modules linked in:
> CPU: 1 PID: 394 Comm: syz-executor403 Not tainted 6.10.0-rc7-g852e42cc2dd4
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/01/2014
> RIP: 0010:proc_register+0x645/0x8f0 fs/proc/generic.c:375
> Code: 00 00 00 00 00 48 85 ed 0f 85 97 02 00 00 4d 85 f6 0f 85 9f 02 00 00 48 c7 c7 9b cb cf 87 48 89 de 4c 89 fa e8 1c 6f eb fe 90 <0f> 0b 90 90 48 c7 c7 98 37 99 89 e8 cb 7e 22 05 bb 00 00 00 10 48
> RSP: 0018:ffa0000000cd7c30 EFLAGS: 00010246
> RAX: 9e129be1950f0200 RBX: ff1100011b51582c RCX: ff1100011857cd80
> RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000002
> RBP: 0000000000000000 R08: ffd400000000000f R09: ff1100013e78cac0
> R10: ffac800000cd7980 R11: ff1100013e12b1f0 R12: 0000000000000000
> R13: 0000000000000000 R14: 0000000000000000 R15: ff1100011a99a2ec
> FS: 00007fbd7086f740(0000) GS:ff1100013fd00000(0000) knlGS:0000000000000000
> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00000000200071c0 CR3: 0000000118556004 CR4: 0000000000771ef0
> DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> DR3: 0000000000000000 DR6: 00000000fffe07f0 DR7: 0000000000000400
> PKRU: 55555554
> Call Trace:
> <TASK>
> proc_create_net_single+0x144/0x210 fs/proc/proc_net.c:220
> bcm_connect+0x472/0x840 net/can/bcm.c:1673
> __sys_connect_file net/socket.c:2049 [inline]
> __sys_connect+0x5d2/0x690 net/socket.c:2066
> __do_sys_connect net/socket.c:2076 [inline]
> __se_sys_connect net/socket.c:2073 [inline]
> __x64_sys_connect+0x8f/0x100 net/socket.c:2073
> do_syscall_x64 arch/x86/entry/common.c:52 [inline]
> do_syscall_64+0xd9/0x1c0 arch/x86/entry/common.c:83
> entry_SYSCALL_64_after_hwframe+0x4b/0x53
> RIP: 0033:0x7fbd708b0e5d
> Code: ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 73 9f 1b 00 f7 d8 64 89 01 48
> RSP: 002b:00007fff8cd33f08 EFLAGS: 00000246 ORIG_RAX: 000000000000002a
> RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007fbd708b0e5d
> RDX: 0000000000000010 RSI: 0000000020000040 RDI: 0000000000000003
> RBP: 0000000000000000 R08: 0000000000000040 R09: 0000000000000040
> R10: 0000000000000040 R11: 0000000000000246 R12: 00007fff8cd34098
> R13: 0000000000401280 R14: 0000000000406de8 R15: 00007fbd70ab9000
> </TASK>
> remove_proc_entry: removing non-empty directory 'net/can-bcm', leaking at least '2456'
>
> Fixes: ffd980f976e7 ("[CAN]: Add broadcast manager (bcm) protocol")
> Reported-by: syzkaller <syzkaller@xxxxxxxxxxxxxxxx>
> Signed-off-by: Kuniyuki Iwashima <kuniyu@xxxxxxxxxx>
Thanks,
I agree that the problem was introduced by the cited commit
and is resolved by this patch.
Reviewed-by: Simon Horman <horms@xxxxxxxxxx>
> ---
> net/can/bcm.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/net/can/bcm.c b/net/can/bcm.c
> index 27d5fcf0eac9..46d3ec3aa44b 100644
> --- a/net/can/bcm.c
> +++ b/net/can/bcm.c
> @@ -1470,6 +1470,10 @@ static void bcm_notify(struct bcm_sock *bo, unsigned long msg,
>
> /* remove device reference, if this is our bound device */
> if (bo->bound && bo->ifindex == dev->ifindex) {
> +#if IS_ENABLED(CONFIG_PROC_FS)
> + if (sock_net(sk)->can.bcmproc_dir && bo->bcm_proc_read)
> + remove_proc_entry(bo->procname, sock_net(sk)->can.bcmproc_dir);
> +#endif
As a fix this looks good. But I wonder if it is worth following up
with a helper for the above as it inlines #if logic and now appears twice.
> bo->bound = 0;
> bo->ifindex = 0;
> notify_enodev = 1;
> --
> 2.30.2
>
>
