Message-ID: <CAFcO6XP2MpiAsF7YXYjgh7FMq+hyzFJjK8iBf=ccZ2B6BpNvOg@mail.gmail.com>
Date:   Tue, 27 Sep 2022 01:16:48 +0800
From:   butt3rflyh4ck <butterflyhuangxx@...il.com>
To:     Takashi Iwai <tiwai@...e.de>
Cc:     perex@...ex.cz, tiwai@...e.com,
        pierre-louis.bossart@...ux.intel.com, broonie@...nel.org,
        alsa-devel@...a-project.org, LKML <linux-kernel@...r.kernel.org>
Subject: Re: A divide error bug in snd_pcm_write

The latest kernel upstream.
Yes, but using mmap, you can map the runtime->status page, and then
copy the data through memcpy to overwrite the status->state data, or
even more, which is incredible.
this is debug log:
````
gef➤  c
Continuing.
Cannot remove breakpoints because program is no longer writable.
Further execution is probably impossible.

Thread 2 hit Hardware access (read/write) watchpoint 7: *0xffff88804c5cb000

Old value = 0x0
New value = 0x7665642f
0xffffffff828501a9 in snd_pcm_write (file=<optimized out>,
buf=0x20000000 "/dev/snd/pcmC#D#p", count=0x18,
offset=0xffffc9000a81bf08) at sound/core/pcm_native.c:3494
3494 if (runtime->status->state == SNDRV_PCM_STATE_OPEN ||

   0xffffffff82850199 <snd_pcm_write+25> (bad)  [rsi+riz*2+0x48]
   0xffffffff8285019d <snd_pcm_write+29> mov    eax, DWORD PTR [rbx+0x100]
   0xffffffff828501a3 <snd_pcm_write+35> test   DWORD PTR [rax], 0xfffffff7
 → 0xffffffff828501a9 <snd_pcm_write+41> je     0xffffffff82850216
<snd_pcm_write+150> NOT taken [Reason: !(Z)]
   0xffffffff828501ab <snd_pcm_write+43> mov    rax, rdx
   0xffffffff828501ae <snd_pcm_write+46> mov    rcx, rdx
   0xffffffff828501b1 <snd_pcm_write+49> xor    edx, edx
   0xffffffff828501b3 <snd_pcm_write+51> div    QWORD PTR [rbx+0x90]
   0xffffffff828501ba <snd_pcm_write+58> test   rdx, rdx
[!] Command 'context' failed to execute properly, reason: argument of
type 'NoneType' is not iterable
gef➤  p/x 0xffff88804c5cb000
$14 = 0xffff88804c5cb000
gef➤  x/10gx  0xffff88804c5cb000
0xffff88804c5cb000: 0x646e732f7665642f 0x234423436d63702f
0xffff88804c5cb010: 0x0000000000000070 0x0000000000000000
0xffff88804c5cb020: 0x0000000000000000 0x0000000000000000
0xffff88804c5cb030: 0x0000000000000000 0x0000000000000000
0xffff88804c5cb040: 0x0000000000000000 0x0000000000000000
gef➤  x/s  0xffff88804c5cb000
0xffff88804c5cb000: "/dev/snd/pcmC#D#p"
gef➤  p/x *(struct snd_pcm_mmap_status *)0xffff88804c5cb000
$15 = {
  state = 0x7665642f,
  pad1 = 0x646e732f,
  __pad1 = 0xffff88804c5cb008,
  hw_ptr = 0x234423436d63702f,
  __pad2 = 0xffff88804c5cb010,
  tstamp = {
    tv_sec = 0x70,
    tv_nsec = 0x0
  },
  suspended_state = 0x0,
  pad3 = 0x0,
  audio_tstamp = {
    tv_sec = 0x0,
    tv_nsec = 0x0
  }
}
gef➤
````

Regards,
 butt3rflyh4ck.




On Mon, Sep 26, 2022 at 6:21 PM Takashi Iwai <tiwai@...e.de> wrote:
>
> On Mon, 26 Sep 2022 11:26:17 +0200,
> butt3rflyh4ck wrote:
> >
> > Hi, there is a divide error bug in snd_pcm_write in
> > sound/core/pcm_native.c in the latest kernel.
> >
> > ##Root Cause
> > When open the device of /dev/snd/pcmC#D#p, there would attach a
> > runtime to pcm->substream via snd_pcm_open_substream. see the code
> > below:
> > ```
> > int snd_pcm_attach_substream(struct snd_pcm *pcm, int stream,
> >     struct file *file,
> >     struct snd_pcm_substream **rsubstream)
> > {
> > ......
> >
> >  runtime = kzalloc(sizeof(*runtime), GFP_KERNEL);
> > if (runtime == NULL)
> > return -ENOMEM;
> >
> > size = PAGE_ALIGN(sizeof(struct snd_pcm_mmap_status));
> > runtime->status = alloc_pages_exact(size, GFP_KERNEL);
> > if (runtime->status == NULL) {
> > kfree(runtime);
> > return -ENOMEM;
> > }
> > memset(runtime->status, 0, size);
> >
> > size = PAGE_ALIGN(sizeof(struct snd_pcm_mmap_control));
> > runtime->control = alloc_pages_exact(size, GFP_KERNEL);
> > if (runtime->control == NULL) {
> > free_pages_exact(runtime->status,
> >       PAGE_ALIGN(sizeof(struct snd_pcm_mmap_status)));
> > kfree(runtime);
> > return -ENOMEM;
> > }
> > memset(runtime->control, 0, size);
> >
> > init_waitqueue_head(&runtime->sleep);
> > init_waitqueue_head(&runtime->tsleep);
> >
> > runtime->status->state = SNDRV_PCM_STATE_OPEN;
> > mutex_init(&runtime->buffer_mutex);
> > atomic_set(&runtime->buffer_accessing, 0);
> >
> > substream->runtime = runtime;
> > substream->private_data = pcm->private_data;
> > substream->ref_count = 1;
> > substream->f_flags = file->f_flags;
> > substream->pid = get_pid(task_pid(current));
> > pstr->substream_opened++;
> > *rsubstream = substream;
> > return 0;
> > }
> > ```
> > It would kzmalloc a new runtime. And initialize runtime simply.
> > If we write some data to the device. it would call snd_pcm_write or
> > snd_pcm_writev. and read some date from the device, it would call
> > snd_pcm_read or snd_pcm_readv.
> > Anyway, the four function would use data of runtime, but some data of
> > runtime is NULL not be initialized, see the code below:
> > ```
> > static ssize_t snd_pcm_write(struct file *file, const char __user *buf,
> >     size_t count, loff_t * offset)
> > {
> > struct snd_pcm_file *pcm_file;
> > struct snd_pcm_substream *substream;
> > struct snd_pcm_runtime *runtime;
> > snd_pcm_sframes_t result;
> >
> > pcm_file = file->private_data;
> > substream = pcm_file->substream;
> > if (PCM_RUNTIME_CHECK(substream))
> > return -ENXIO;
> > runtime = substream->runtime;
> > if (runtime->status->state == SNDRV_PCM_STATE_OPEN ||
> >    runtime->status->state == SNDRV_PCM_STATE_DISCONNECTED)
> > return -EBADFD;
> > if (!frame_aligned(runtime, count))    ///////  [1]
> > return -EINVAL;
> > count = bytes_to_frames(runtime, count);    /////// [2]
> > result = snd_pcm_lib_write(substream, buf, count);
> > if (result > 0)
> > result = frames_to_bytes(runtime, result);
> > return result;
> > }
> > ```
> > [1] call frame_aligned to aligned.
> > ```
> > static inline int frame_aligned(struct snd_pcm_runtime *runtime, ssize_t bytes)
> > {
> > return bytes % runtime->byte_align == 0;
> > }
> > ```
> > but runtime->byte_align is NULL.
> >
> > [2] call bytes_to_frames.
> > ```
> > static inline ssize_t frames_to_bytes(struct snd_pcm_runtime *runtime,
> > snd_pcm_sframes_t size)
> > {
> > return size * runtime->frame_bits / 8;
> > }
> > ```
> > but runtime->frame_bits is NULL.
> >
> > ##reproduce it
> > [ 1189.305083][ T4656] divide error: 0000 [#1] PREEMPT SMP
> > [ 1189.305600][ T4656] CPU: 1 PID: 4656 Comm: snd_pcm_write Not
> > tainted 6.0.0-rc7 #16
> > [ 1189.306157][ T4656] Hardware name: QEMU Standard PC (i440FX + PIIX,
> > 1996), BIOS 1.13.0-1ubuntu1 04/01/2014
> > [ 1189.306760][ T4656] RIP: 0010:snd_pcm_write+0x33/0xa0
> > [ 1189.307155][ T4656] Code: 8b 38 48 85 ff 74 72 48 8b 9f c0 00 00 00
> > 48 85 db 74 66 48 8b 83 00 01 00 00 f7 00 f7 ff ff ff 74 6b 48 89 d0
> > 48 89 d1 31 d2 <48> f7 b3 91
> > [ 1189.308553][ T4656] RSP: 0018:ffffc9000adc7e68 EFLAGS: 00010246
> > [ 1189.309034][ T4656] RAX: 0000000000000000 RBX: ffff888048ec2000
> > RCX: 0000000000000000
> > [ 1189.309583][ T4656] RDX: 0000000000000000 RSI: 0000000000000000
> > RDI: ffff888046fc9c00
> > [ 1189.310163][ T4656] RBP: 0000000000000000 R08: 0000000000000000
> > R09: 0000000000020026
> > [ 1189.310679][ T4656] R10: 0000000000000001 R11: 0000000000000000
> > R12: 0000000000000000
> > [ 1189.311226][ T4656] R13: ffffc9000adc7f08 R14: 0000000000000000
> > R15: 0000000000000000
> > [ 1189.311754][ T4656] FS:  00000000012d8880(0000)
> > GS:ffff88807ec00000(0000) knlGS:0000000000000000
> > [ 1189.312350][ T4656] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > [ 1189.312780][ T4656] CR2: 0000000020000000 CR3: 000000004b496000
> > CR4: 00000000000006e0
> > [ 1189.313234][ T4656] Call Trace:
> > [ 1189.313424][ T4656]  <TASK>
> > [ 1189.313597][ T4656]  vfs_write+0xe6/0x4d0
> > [ 1189.313836][ T4656]  ksys_write+0x60/0xe0
> > [ 1189.314071][ T4656]  do_syscall_64+0x35/0xb0
> > [ 1189.314324][ T4656]  entry_SYSCALL_64_after_hwframe+0x63/0xcd
> > [ 1189.314734][ T4656] RIP: 0033:0x44dc3d
> > [ 1189.315007][ T4656] Code: 28 c3 e8 36 29 00 00 66 0f 1f 44 00 00 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 f8
> > [ 1189.316338][ T4656] RSP: 002b:00007ffcbdf5ef38 EFLAGS: 00000246
> > ORIG_RAX: 0000000000000001
> > [ 1189.317012][ T4656] RAX: ffffffffffffffda RBX: 0000000000400530
> > RCX: 000000000044dc3d
> > [ 1189.317693][ T4656] RDX: 0000000000000000 RSI: 0000000000000000
> > RDI: 0000000000000003
> > [ 1189.318172][ T4656] RBP: 00007ffcbdf5ef50 R08: 0000000082000000
> > R09: 0000000000000000
> > [ 1189.318648][ T4656] R10: 000000000000ffff R11: 0000000000000246
> > R12: 00000000004030a0
> > [ 1189.319182][ T4656] R13: 0000000000000000 R14: 00000000004c5018
> > R15: 0000000000000000
>
> The question is how the code passes the check before [1]:
>
>  if (runtime->status->state == SNDRV_PCM_STATE_OPEN ||
>      runtime->status->state == SNDRV_PCM_STATE_DISCONNECTED)
>         return -EBADFD;
>
> The uninitialized state should have been with SNDRV_PCM_STATE_OPEN,
> and runtime->byte_align is set up at snd_pcm_hw_params() followed by
> snd_pcm-set_state() to chage the state to SNDRV_PCM_STATE_SETUP.
>
> Which kernel version are you testing?
>
>
> Takashi



--
Active Defense Lab of Venustech

Powered by blists - more mailing lists