tty: kmalloc size WARNING in vc_do_resize

tty: kmalloc size WARNING in vc_do_resize

[Dmitry Vyukov]

Hello,

The following program triggers kmalloc size WARNING in vc_do_resize:

// autogenerated by syzkaller (http://github.com/google/syzkaller)
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <termios.h>
#include <sys/ioctl.h>

int main()
{
  int fd = open("/dev/tty1", O_RDWR);
  struct winsize ws;
  ws.ws_row = 0x1000;
  ws.ws_col = 0x5dc;
  ws.ws_xpixel = 0x2;
  ws.ws_ypixel = 0x0;
  ioctl(fd, TIOCSWINSZ, &ws);
  return 0;
}


------------[ cut here ]------------
WARNING: CPU: 3 PID: 7642 at mm/page_alloc.c:2999
__alloc_pages_nodemask+0x7d2/0x1760()
Modules linked in:
CPU: 3 PID: 7642 Comm: a.out Not tainted 4.4.0+ #276
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
 00000000ffffffff ffff88006d24f610 ffffffff82999e2d 0000000000000000
 ffff880060d9af80 ffffffff86475560 ffff88006d24f650 ffffffff81352089
 ffffffff816721e2 ffffffff86475560 0000000000000bb7 00000000024240c0
Call Trace:
 [<     inline     >] __dump_stack lib/dump_stack.c:15
 [<ffffffff82999e2d>] dump_stack+0x6f/0xa2 lib/dump_stack.c:50
 [<ffffffff81352089>] warn_slowpath_common+0xd9/0x140 kernel/panic.c:482
 [<ffffffff813522b9>] warn_slowpath_null+0x29/0x30 kernel/panic.c:515
 [<     inline     >] __alloc_pages_slowpath mm/page_alloc.c:2999
 [<ffffffff816721e2>] __alloc_pages_nodemask+0x7d2/0x1760 mm/page_alloc.c:3253
 [<ffffffff8174a799>] alloc_pages_current+0xe9/0x450 mm/mempolicy.c:2090
 [<     inline     >] alloc_pages include/linux/gfp.h:459
 [<ffffffff8166df66>] alloc_kmem_pages+0x16/0x100 mm/page_alloc.c:3433
 [<ffffffff816c698f>] kmalloc_order+0x1f/0x80 mm/slab_common.c:1008
 [<ffffffff816c6a0f>] kmalloc_order_trace+0x1f/0x140 mm/slab_common.c:1019
 [<     inline     >] kmalloc_large include/linux/slab.h:395
 [<ffffffff8175b624>] __kmalloc+0x2f4/0x340 mm/slub.c:3557
 [<     inline     >] kmalloc include/linux/slab.h:468
 [<ffffffff82d47800>] vc_do_resize+0x2c0/0x1140 drivers/tty/vt/vt.c:874
 [<ffffffff82d4878a>] vt_resize+0xaa/0xe0 drivers/tty/vt/vt.c:993
 [<     inline     >] tiocswinsz drivers/tty/tty_io.c:2357
 [<ffffffff82cf22b3>] tty_ioctl+0x1083/0x2160 drivers/tty/tty_io.c:2869
 [<     inline     >] vfs_ioctl fs/ioctl.c:43
 [<ffffffff817efdac>] do_vfs_ioctl+0x18c/0xfb0 fs/ioctl.c:674
 [<     inline     >] SYSC_ioctl fs/ioctl.c:689
 [<ffffffff817f0c5f>] SyS_ioctl+0x8f/0xc0 fs/ioctl.c:680
 [<ffffffff86336c36>] entry_SYSCALL_64_fastpath+0x16/0x7a
arch/x86/entry/entry_64.S:185
---[ end trace cc28f7cc9d447282 ]---


I think that either the kmalloc should use __GFP_NOWARN or
vc_do_resize should do stricter size check.

On commit 30f05309bde49295e02e45c7e615f73aa4e0ccc2.

Re: tty: kmalloc size WARNING in vc_do_resize

[One Thousand Gnomes]

> I think that either the kmalloc should use __GFP_NOWARN or
> vc_do_resize should do stricter size check.

vc_do_resize doesn't know enough to do a stricter size check. There are
not many methods it calls but those are:

vgacon_resize doesn't error when stupid things happen because of a silly
hack for legacy svgatextmode tools.

sisusbcon_resize does the right checks but we can get in a situation
where we unplug during a resize so attempting to resize and size back
might fail.

fbcon_resize does the right thing, but again can error if we go
size/resize/size back.

That means we really have to guess or would have to add 'max size' info
the structures. Right now it's clamped to 1Gbyte which is a 32767 x 32767
character display and probably ludicrous. Clamping to 4048 x 1024 chars
would cover displays up to 16K x 6K even with the 4x6 font (which is
insane on a 4K display, let alone 4 of them).

Alan

Re: tty: kmalloc size WARNING in vc_do_resize

[Dmitry Vyukov]

On Mon, Jan 25, 2016 at 5:12 PM, One Thousand Gnomes
<gnomes@lxorguk.ukuu.org.uk> wrote:
>> I think that either the kmalloc should use __GFP_NOWARN or
>> vc_do_resize should do stricter size check.
>
> vc_do_resize doesn't know enough to do a stricter size check. There are
> not many methods it calls but those are:
>
> vgacon_resize doesn't error when stupid things happen because of a silly
> hack for legacy svgatextmode tools.
>
> sisusbcon_resize does the right checks but we can get in a situation
> where we unplug during a resize so attempting to resize and size back
> might fail.
>
> fbcon_resize does the right thing, but again can error if we go
> size/resize/size back.
>
> That means we really have to guess or would have to add 'max size' info
> the structures. Right now it's clamped to 1Gbyte which is a 32767 x 32767
> character display and probably ludicrous. Clamping to 4048 x 1024 chars
> would cover displays up to 16K x 6K even with the 4x6 font (which is
> insane on a 4K display, let alone 4 of them).


OK, then kmalloc call needs to include __GFP_NOWARN for cases when
size is too large (large than 8MB).