Re: [PATCH] drm: assure aux_dev is nonzero before using it

From: Jani Nikula
Date: Fri May 24 2019 - 04:39:17 EST

On Thu, 23 May 2019, tcamuso <tcamuso@xxxxxxxxxx> wrote:
> From Daniel Kwon <dkwon@xxxxxxxxxx>
>
> The system was crashed due to invalid memory access while trying to access
> auxiliary device.
>
> crash> bt
> PID: 9863 TASK: ffff89d1bdf11040 CPU: 1 COMMAND: "ipmitool"
> #0 [ffff89cedd7f3868] machine_kexec at ffffffffb0663674
> #1 [ffff89cedd7f38c8] __crash_kexec at ffffffffb071cf62
> #2 [ffff89cedd7f3998] crash_kexec at ffffffffb071d050
> #3 [ffff89cedd7f39b0] oops_end at ffffffffb0d6d758
> #4 [ffff89cedd7f39d8] no_context at ffffffffb0d5bcde
> #5 [ffff89cedd7f3a28] __bad_area_nosemaphore at ffffffffb0d5bd75
> #6 [ffff89cedd7f3a78] bad_area at ffffffffb0d5c085
> #7 [ffff89cedd7f3aa0] __do_page_fault at ffffffffb0d7080c
> #8 [ffff89cedd7f3b10] do_page_fault at ffffffffb0d70905
> #9 [ffff89cedd7f3b40] page_fault at ffffffffb0d6c758
> [exception RIP: drm_dp_aux_dev_get_by_minor+0x3d]
> RIP: ffffffffc0a589bd RSP: ffff89cedd7f3bf0 RFLAGS: 00010246
> RAX: 0000000000000000 RBX: 0000000000000000 RCX: ffff89cedd7f3fd8
> RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffffffc0a613e0
> RBP: ffff89cedd7f3bf8 R8: ffff89f1bcbabbd0 R9: 0000000000000000
> R10: ffff89f1be7a1cc0 R11: 0000000000000000 R12: 0000000000000000
> R13: ffff89f1b32a2830 R14: ffff89d18fadfa00 R15: 0000000000000000
> ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
> RIP: 00002b45f0d80d30 RSP: 00007ffc416066a0 RFLAGS: 00010246
> RAX: 0000000000000002 RBX: 000056062e212d80 RCX: 00007ffc41606810
> RDX: 0000000000000000 RSI: 0000000000000002 RDI: 00007ffc41606ec0
> RBP: 0000000000000000 R8: 000056062dfed229 R9: 00002b45f0cdf14d
> R10: 0000000000000002 R11: 0000000000000246 R12: 00007ffc41606ec0
> R13: 00007ffc41606ed0 R14: 00007ffc41606ee0 R15: 0000000000000000
> ORIG_RAX: 0000000000000002 CS: 0033 SS: 002b
>
> ----------------------------------------------------------------------------
>
> It was trying to open '/dev/ipmi0', but as no entry in aux_dir, it returned
> NULL from 'idr_find()'. This drm_dp_aux_dev_get_by_minor() should have done a
> check on this, but had failed to do it.

I think the better question is, *why* does the idr_find() return NULL? I
don't think it should, under any circumstances. I fear adding the check
here papers over some other problem, taking us further away from the
root cause.

Also, can you reproduce this on a recent upstream kernel? The aux device
nodes were introduced in kernel v4.6. Whatever you reproduced on v3.10
is pretty much irrelevant for upstream.


BR,
Jani.




>
> ----------------------------------------------------------------------------
> /usr/src/debug/kernel-3.10.0-957.12.1.el7/linux-3.10.0-957.12.1.el7.x86_64/include/linux/idr.h: 114
> 114 struct idr_layer *hint = rcu_dereference_raw(idr->hint);
> 0xffffffffc0a58998 <drm_dp_aux_dev_get_by_minor+0x18>: mov 0x8a41(%rip),%rax # 0xffffffffc0a613e0 <aux_idr>
> /usr/src/debug/kernel-3.10.0-957.12.1.el7/linux-3.10.0-957.12.1.el7.x86_64/include/linux/idr.h: 116
> 116 if (hint && (id & ~IDR_MASK) == hint->prefix)
> 117 return rcu_dereference_raw(hint->ary[id & IDR_MASK]);
> 0xffffffffc0a5899f <drm_dp_aux_dev_get_by_minor+0x1f>: test %rax,%rax
> 0xffffffffc0a589a2 <drm_dp_aux_dev_get_by_minor+0x22>: je 0xffffffffc0a589ac <drm_dp_aux_dev_get_by_minor+0x2c>
> 0xffffffffc0a589a4 <drm_dp_aux_dev_get_by_minor+0x24>: mov %ebx,%edx
> 0xffffffffc0a589a6 <drm_dp_aux_dev_get_by_minor+0x26>: xor %dl,%dl
> 0xffffffffc0a589a8 <drm_dp_aux_dev_get_by_minor+0x28>: cmp (%rax),%edx
> 0xffffffffc0a589aa <drm_dp_aux_dev_get_by_minor+0x2a>: je 0xffffffffc0a589f0 <drm_dp_aux_dev_get_by_minor+0x70>
> /usr/src/debug/kernel-3.10.0-957.12.1.el7/linux-3.10.0-957.12.1.el7.x86_64/include/linux/idr.h: 119
> 119 return idr_find_slowpath(idr, id);
> 0xffffffffc0a589ac <drm_dp_aux_dev_get_by_minor+0x2c>: mov %ebx,%esi
> 0xffffffffc0a589ae <drm_dp_aux_dev_get_by_minor+0x2e>: mov $0xffffffffc0a613e0,%rdi
> 0xffffffffc0a589b5 <drm_dp_aux_dev_get_by_minor+0x35>: callq 0xffffffffb09771b0 <idr_find_slowpath>
> 0xffffffffc0a589ba <drm_dp_aux_dev_get_by_minor+0x3a>: mov %rax,%rbx
> /usr/src/debug/kernel-3.10.0-957.12.1.el7/linux-3.10.0-957.12.1.el7.x86_64/arch/x86/include/asm/atomic.h: 25
> 25 return ACCESS_ONCE((v)->counter);
> 0xffffffffc0a589bd <drm_dp_aux_dev_get_by_minor+0x3d>: mov 0x18(%rbx),%edx
>
> crash> struct file.f_path 0xffff89d18fadfa00
> f_path = {
> mnt = 0xffff89f23feaa620,
> dentry = 0xffff89f1be7a1cc0
> }
> crash> files -d 0xffff89f1be7a1cc0
> DENTRY INODE SUPERBLK TYPE PATH
> ffff89f1be7a1cc0 ffff89f1b32a2830 ffff89d293aa8800 CHR /dev/ipmi0
>
> crash> struct inode.i_rdev ffff89f1b32a2830
> i_rdev = 0xf200000
> crash> eval (0xfffff & 0xf200000)
> hexadecimal: 0
> decimal: 0
> octal: 0
> binary: 0000000000000000000000000000000000000000000000000000000000000000
> ----------------------------------------------------------------------------
>
> As the index value was 0 and aux_idr had value 0 for all, it can have value
> NULL from idr_find() function, but the below function doesn't check and just
> tries to use it.
>
> ----------------------------------------------------------------------------
> crash> aux_idr
> aux_idr = $8 = {
> hint = 0x0,
> top = 0x0,
> id_free = 0x0,
> layers = 0x0,
> id_free_cnt = 0x0,
> cur = 0x0,
> lock = {
> {
> rlock = {
> raw_lock = {
> val = {
> counter = 0x0
> }
> }
> }
> }
> }
> }
>
> crash> edis -f drm_dp_aux_dev_get_by_minor
> /usr/src/debug/kernel-3.10.0-957.12.1.el7/linux-3.10.0-957.12.1.el7.x86_64/drivers/gpu/drm/drm_dp_aux_dev.c: 57
>
> 56 static struct drm_dp_aux_dev *drm_dp_aux_dev_get_by_minor(unsigned index)
> 57 {
> 58 struct drm_dp_aux_dev *aux_dev = NULL;
> 59
> 60 mutex_lock(&aux_idr_mutex);
> 61 aux_dev = idr_find(&aux_idr, index);
> 62 if (!kref_get_unless_zero(&aux_dev->refcount))
> 63 aux_dev = NULL;
> 64 mutex_unlock(&aux_idr_mutex);
> 65
> 66 return aux_dev;
> 67 }
> ----------------------------------------------------------------------------
>
> To avoid this kinds of situation, we should make a safeguard for the returned
> value. Changing the line 62 with the below would do.
>
> 62 if (aux_dev && !kref_get_unless_zero(&aux_dev->refcount))
> ^^^^^^^^^^
> From Tony Camuso <tcamuso@xxxxxxxxxx>
> I built a patched kernel for several architectures.
> Booted the kernel, and ran the following for 100 iterations.
> rmmod ipmi kmods to remove /dev/ipmi0.
> Invoked ipmitool
> insmod ipmi kmods
> Did not see any crashes or call traces.
>
> Suggested-by: Daniel Kwon <dkwon@xxxxxxxxxx>
> Signed-off-by: Tony Camuso <tcamuso@xxxxxxxxxx>
> ---
> drivers/gpu/drm/drm_dp_aux_dev.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/drm_dp_aux_dev.c b/drivers/gpu/drm/drm_dp_aux_dev.c
> index 0e4f25d63fd2d..0b11210c882ee 100644
> --- a/drivers/gpu/drm/drm_dp_aux_dev.c
> +++ b/drivers/gpu/drm/drm_dp_aux_dev.c
> @@ -60,7 +60,7 @@ static struct drm_dp_aux_dev *drm_dp_aux_dev_get_by_minor(unsigned index)
>
> mutex_lock(&aux_idr_mutex);
> aux_dev = idr_find(&aux_idr, index);
> - if (!kref_get_unless_zero(&aux_dev->refcount))
> + if (aux_dev && !kref_get_unless_zero(&aux_dev->refcount))
> aux_dev = NULL;
> mutex_unlock(&aux_idr_mutex);

--
Jani Nikula, Intel Open Source Graphics Center