Re: [PATCH] drm/atomic-helper: prevent uaf in wait_for_vblanks

From: José Pekkarinen
Date: Wed Sep 20 2023 - 03:43:35 EST


On 2023-09-19 19:56, Michel Dänzer wrote:
On 9/18/23 18:53, José Pekkarinen wrote:
Kasan reported the following in my system:

[ 3935.321003] ==================================================================
[ 3935.321022] BUG: KASAN: slab-use-after-free in drm_atomic_helper_wait_for_vblanks.part.0+0x116/0x450 [drm_kms_helper]
[ 3935.321124] Read of size 1 at addr ffff88818a6f8009 by task kworker/u16:3/5268

[ 3935.321124] CPU: 7 PID: 5268 Comm: kworker/u16:3 Not tainted 6.6.0-rc2+ #1
[ 3935.321124] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
[ 3935.321124] Workqueue: events_unbound commit_work [drm_kms_helper]
[ 3935.321124] Call Trace:
[ 3935.321124] <TASK>
[ 3935.321124] dump_stack_lvl+0x43/0x60
[ 3935.321124] print_report+0xcf/0x660
[ 3935.321124] ? remove_entity_load_avg+0xdc/0x100
[ 3935.321124] ? __virt_addr_valid+0xd9/0x160
[ 3935.321124] ? drm_atomic_helper_wait_for_vblanks.part.0+0x116/0x450 [drm_kms_helper]
[ 3935.321124] kasan_report+0xda/0x110
[ 3935.321124] ? drm_atomic_helper_wait_for_vblanks.part.0+0x116/0x450 [drm_kms_helper]
[ 3935.321124] drm_atomic_helper_wait_for_vblanks.part.0+0x116/0x450 [drm_kms_helper]
[ 3935.321124] ? __pfx_drm_atomic_helper_wait_for_vblanks.part.0+0x10/0x10 [drm_kms_helper]
[ 3935.321124] ? complete_all+0x48/0x100
[ 3935.321124] ? _raw_spin_unlock_irqrestore+0x19/0x40
[ 3935.321124] ? preempt_count_sub+0x14/0xc0
[ 3935.321124] ? _raw_spin_unlock_irqrestore+0x23/0x40
[ 3935.321124] ? drm_atomic_helper_commit_hw_done+0x1ac/0x240 [drm_kms_helper]
[ 3935.321124] drm_atomic_helper_commit_tail+0x82/0x90 [drm_kms_helper]
[ 3935.321124] commit_tail+0x15c/0x1d0 [drm_kms_helper]
[ 3935.323185] process_one_work+0x31a/0x610
[ 3935.323185] worker_thread+0x38e/0x5f0
[ 3935.323185] ? __pfx_worker_thread+0x10/0x10
[ 3935.323185] kthread+0x184/0x1c0
[ 3935.323185] ? __pfx_kthread+0x10/0x10
[ 3935.323185] ret_from_fork+0x30/0x50
[ 3935.323185] ? __pfx_kthread+0x10/0x10
[ 3935.323185] ret_from_fork_asm+0x1b/0x30
[ 3935.323185] </TASK>

[ 3935.323185] Allocated by task 3751:
[ 3935.323185] kasan_save_stack+0x2f/0x50
[ 3935.323185] kasan_set_track+0x21/0x30
[ 3935.323185] __kasan_kmalloc+0xa6/0xb0
[ 3935.323185] drm_atomic_helper_crtc_duplicate_state+0x42/0x70 [drm_kms_helper]
[ 3935.323185] drm_atomic_get_crtc_state+0xc3/0x1e0 [drm]
[ 3935.323185] page_flip_common+0x42/0x160 [drm_kms_helper]
[ 3935.323185] drm_atomic_helper_page_flip+0x6b/0xf0 [drm_kms_helper]
[ 3935.323185] drm_mode_page_flip_ioctl+0x8ad/0x900 [drm]
[ 3935.323185] drm_ioctl_kernel+0x169/0x240 [drm]
[ 3935.323185] drm_ioctl+0x399/0x6b0 [drm]
[ 3935.324772] __x64_sys_ioctl+0xc5/0x100
[ 3935.324772] do_syscall_64+0x5b/0xc0
[ 3935.324772] entry_SYSCALL_64_after_hwframe+0x6e/0xd8

[ 3935.324772] Freed by task 3751:
[ 3935.324772] kasan_save_stack+0x2f/0x50
[ 3935.324772] kasan_set_track+0x21/0x30
[ 3935.324772] kasan_save_free_info+0x27/0x40
[ 3935.324772] ____kasan_slab_free+0x166/0x1c0
[ 3935.324772] slab_free_freelist_hook+0x9f/0x1e0
[ 3935.324772] __kmem_cache_free+0x187/0x2d0
[ 3935.324772] drm_atomic_state_default_clear+0x226/0x5e0 [drm]
[ 3935.324772] __drm_atomic_state_free+0xc8/0x130 [drm]
[ 3935.324772] drm_atomic_helper_update_plane+0x17d/0x1b0 [drm_kms_helper]
[ 3935.324772] drm_mode_cursor_universal+0x2a4/0x4d0 [drm]
[ 3935.324772] drm_mode_cursor_common+0x1cf/0x430 [drm]
[ 3935.324772] drm_mode_cursor_ioctl+0xc6/0x100 [drm]
[ 3935.326167] drm_ioctl_kernel+0x169/0x240 [drm]
[ 3935.326167] drm_ioctl+0x399/0x6b0 [drm]
[ 3935.326614] __x64_sys_ioctl+0xc5/0x100
[ 3935.326614] do_syscall_64+0x5b/0xc0
[ 3935.326614] entry_SYSCALL_64_after_hwframe+0x6e/0xd8

[ 3935.326614] The buggy address belongs to the object at ffff88818a6f8000
which belongs to the cache kmalloc-512 of size 512
[ 3935.326614] The buggy address is located 9 bytes inside of
freed 512-byte region [ffff88818a6f8000, ffff88818a6f8200)

[ 3935.326614] The buggy address belongs to the physical page:
[ 3935.326614] page:00000000b0fb0816 refcount:1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x18a6f8
[ 3935.326614] head:00000000b0fb0816 order:3 entire_mapcount:0 nr_pages_mapped:0 pincount:0
[ 3935.326614] anon flags: 0x17ffffc0000840(slab|head|node=0|zone=2|lastcpupid=0x1fffff)
[ 3935.326614] page_type: 0xffffffff()
[ 3935.326614] raw: 0017ffffc0000840 ffff888100042c80 0000000000000000 dead000000000001
[ 3935.326614] raw: 0000000000000000 0000000080200020 00000001ffffffff 0000000000000000
[ 3935.326614] page dumped because: kasan: bad access detected

[ 3935.326614] Memory state around the buggy address:
[ 3935.326614] ffff88818a6f7f00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 3935.326614] ffff88818a6f7f80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 3935.326614] >ffff88818a6f8000: fa fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 3935.326772] ^
[ 3935.326772] ffff88818a6f8080: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 3935.326772] ffff88818a6f8100: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 3935.326772] ==================================================================

This suggest there may be some situation where a
struct drm_crtc_state is referenced after already
being freed by drm_atomic_state_default_clear. This
patch will check the new_crtc_state is not null before
using it.

Signed-off-by: José Pekkarinen <jose.pekkarinen@xxxxxxxxxxx>
---
drivers/gpu/drm/drm_atomic_helper.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
index 292e38eb6218..cc75d387a542 100644
--- a/drivers/gpu/drm/drm_atomic_helper.c
+++ b/drivers/gpu/drm/drm_atomic_helper.c
@@ -1647,7 +1647,7 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,
return;

for_each_oldnew_crtc_in_state(old_state, crtc, old_crtc_state, new_crtc_state, i) {
- if (!new_crtc_state->active)
+ if (new_crtc_state && !new_crtc_state->active)
continue;

ret = drm_crtc_vblank_get(crtc);

I'm not quite seeing the connection between this change and the KASAN report.

If new_crtc_state was NULL, I would have expected a normal NULL
pointer dereference oops, not a KASAN report.

The KASAN report instead indicates that new_crtc_state isn't NULL, but
points to memory which has already been freed. This could be e.g. due
to incorrect reference counting somewhere else.

Hi Michel,

Thanks for your comments, I found the kasan report also
in this version of the patch, so I'm setting bigger traps to
the bug to find out what is the issue behind. While what you
say it is true, I may invite you to take a look to
drm_atomic_state_default_clear, you'll quickly notice that
after destroying the crtc_state, it set state pointers to null,
so unless something very bad happened in my system, at the
time of executing drm_atomic_helper_wait_for_vblanks, if state
is freed, its pointer should be null also.

Best regards.

José.