Re: [PATCH] tty: n_gsm: check error while registering tty devices

[Jiri Slaby]

On 07. 04. 21, 4:16, Hillf Danton wrote:
> Add the error path for registering tty devices and roll back in case of error
> in bid to avoid the UAF like the below one reported.
>
>   ------------[ cut here ]------------
>   refcount_t: underflow; use-after-free.
>   WARNING: CPU: 1 PID: 8923 at lib/refcount.c:28
>   refcount_warn_saturate+0x1cf/0x210 -origin/lib/refcount.c:28
>   Modules linked in:
>   CPU: 1 PID: 8923 Comm: executor Not tainted 5.12.0-rc5+ #8
>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
>   1.13.0-1ubuntu1.1 04/01/2014
>   RIP: 0010:refcount_warn_saturate+0x1cf/0x210 -origin/lib/refcount.c:28
>   Code: 4f ff ff ff e8 32 fa b5 fe 48 c7 c7 3d f8 f6 86 e8 d6 ab c6 fe
>   c6 05 7c 34 67 04 01 48 c7 c7 68 f8 6d 86 31 c0 e8 81 2e 9d fe <0f> 0b
>   e9 22 ff ff ff e8 05 fa b5 fe 48 c7 c7 3e f8 f6 86 e8 a9 ab
>   RSP: 0018:ffffc90001633c60 EFLAGS: 00010246
>   RAX: 15d08b2e34b77800 RBX: 0000000000000003 RCX: ffff88804c056c80
>   RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000000
>   RBP: 0000000000000003 R08: ffffffff813767aa R09: 0001ffffffffffff
>   R10: 0001ffffffffffff R11: ffff88804c056c80 R12: ffff888040b7d000
>   R13: ffff88804c206938 R14: ffff88804c206900 R15: ffff888041b18488
>   FS:  00000000022c9940(0000) GS:ffff88807ec00000(0000) knlGS:0000000000000000
>   CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>   CR2: 00007f9f9b122008 CR3: 0000000044b4b000 CR4: 0000000000750ee0
>   PKRU: 55555554
>   Call Trace:
>    __refcount_sub_and_test -origin/./include/linux/refcount.h:283 [inline]
>    __refcount_dec_and_test -origin/./include/linux/refcount.h:315 [inline]
>    refcount_dec_and_test -origin/./include/linux/refcount.h:333 [inline]
>    kref_put -origin/./include/linux/kref.h:64 [inline]
>    kobject_put+0x17b/0x180 -origin/lib/kobject.c:753
>    cdev_del+0x4b/0x50 -origin/fs/char_dev.c:597
>    tty_unregister_device+0x99/0xd0 -origin/drivers/tty/tty_io.c:3343
>    gsmld_detach_gsm -origin/drivers/tty/n_gsm.c:2409 [inline]
>    gsmld_close+0x6c/0x140 -origin/drivers/tty/n_gsm.c:2478
>    tty_ldisc_close -origin/drivers/tty/tty_ldisc.c:488 [inline]
>    tty_ldisc_kill -origin/drivers/tty/tty_ldisc.c:636 [inline]
>    tty_ldisc_release+0x1b6/0x400 -origin/drivers/tty/tty_ldisc.c:809
>    tty_release_struct+0x19/0xb0 -origin/drivers/tty/tty_io.c:1714
>    tty_release+0x9ad/0xa00 -origin/drivers/tty/tty_io.c:1885

Yes, the fix makes sense. But could you elaborate in the commit log when 
this happens? I only wonder how real this is. I assume you inject faults 
to allocations?

> Reported-and-tested-by: Hao Sun <sunhao.th@xxxxxxxxx>
> Cc: Jiri Slaby <jslaby@xxxxxxx>

Use my MAINTAINERS e-mail please.

> Signed-off-by: Hillf Danton <hdanton@xxxxxxxx>
> ---
>
> --- a/drivers/tty/n_gsm.c
> +++ b/drivers/tty/n_gsm.c
> @@ -2384,8 +2384,18 @@ static int gsmld_attach_gsm(struct tty_s
>   		/* Don't register device 0 - this is the control channel and not
>   		   a usable tty interface */
>   		base = mux_num_to_base(gsm); /* Base for this MUX */
> -		for (i = 1; i < NUM_DLCI; i++)
> -			tty_register_device(gsm_tty_driver, base + i, NULL);
> +		for (i = 1; i < NUM_DLCI; i++) {
> +			struct device *dev;
> +
> +			dev = tty_register_device(gsm_tty_driver,
> +							base + i, NULL);
> +			if (IS_ERR(dev)) {
> +				for (i--; i >= 1; i--)
> +					tty_unregister_device(gsm_tty_driver,
> +								base + i);
> +				return PTR_ERR(dev);
> +			}
> +		}
>   	}
>   	return ret;
>   }
> --

thanks,
--
js
suse labs