Re: [BUG] staging: r8188eu: KASAN: slab-out-of-bounds in rtw_cmd_thread

[Solomon Tan <wjsota@gmail.com>]

On Mon, Apr 25, 2022 at 06:41:38PM +0300, Dan Carpenter wrote:
> On Sun, Apr 24, 2022 at 06:06:32PM +0300, Pavel Skripkin wrote:
> > Hi Solomon,
> > 
> > On 4/24/22 15:11, Solomon Tan wrote:
> > > On Sun, Apr 24, 2022 at 12:00:12PM +0200, Michael Straube wrote:
> > > > Hi,
> > > > 
> > > > It looks like
> > > > commit 0afaa121813e ("staging: r8188eu: use in-kernel ieee channel")
> > > > intoduced a. See KASAN output below.
> > > > 
> > > > That commit replaced the use of struct rtw_ieee80211_channel with struct
> > > > ieee80211_channel.
> > > > 
> > > > There are several calls to memcpy that used sizeof(struct
> > > > rtw_ieee80211_channel)
> > > > and now use sizeof(struct ieee80211_channel) but the sizes of these two
> > > > structures are not equal.
> > > > 
> > > 
> > > Oh no. When does this issue get triggered?
> > > 
> > > > regards,
> > > > Michael
> > > > 
> > > > dmesg:
> > > > 
> > > >  ==================================================================
> > > > [  422.214237] BUG: KASAN: slab-out-of-bounds in rtw_cmd_thread+0x1e8/0x430
> > > > [r8188eu]
> > > > [  422.214277] Write of size 3600 at addr ffff8881e149d200 by task
> > > > RTW_CMD_THREAD/2563
> > > > 
> > > > [  422.214289] CPU: 11 PID: 2563 Comm: RTW_CMD_THREAD Tainted: G C OE
> > > > 5.18.0-rc2-staging+ #47 94e3ca73bebf5b7fec506721475e4fff2a023bb9
> > > > [  422.214301] Hardware name: Gigabyte Technology Co., Ltd. B550M S2H/B550M
> > > > S2H, BIOS F15a 02/16/2022
> > > > [  422.214309] Call Trace:
> > > > [  422.214313]  <TASK>
> > > > [  422.214317]  dump_stack_lvl+0x45/0x5b
> > > > [  422.214327]  print_report.cold+0x5e/0x5dc
> > > > [  422.214335]  ? kasan_set_track+0x21/0x30
> > > > [  422.214342]  ? kasan_set_free_info+0x20/0x40
> > > > [  422.214349]  ? rtw_cmd_thread+0x1e8/0x430 [r8188eu
> > > > 91924fe1575bf49b9b37985ffde2c585d847446d]
> > > > [  422.214386]  kasan_report+0xab/0x120
> > > > [  422.214394]  ? rtw_cmd_thread+0x1e8/0x430 [r8188eu
> > > > 91924fe1575bf49b9b37985ffde2c585d847446d]
> > > > [  422.214430]  kasan_check_range+0xf6/0x1d0
> > > > [  422.214436]  memcpy+0x39/0x60
> > > > [  422.214442]  rtw_cmd_thread+0x1e8/0x430 [r8188eu
> > > > 91924fe1575bf49b9b37985ffde2c585d847446d]
> > > > [  422.214479]  ? rtw_setassocsta_cmdrsp_callback+0xd0/0xd0 [r8188eu
> > > > 91924fe1575bf49b9b37985ffde2c585d847446d]
> > > > [  422.214516]  kthread+0x15d/0x190
> > > > [  422.214523]  ? kthread_complete_and_exit+0x20/0x20
> > > > [  422.214531]  ret_from_fork+0x22/0x30
> > > > [  422.214540]  </TASK>
> > > 
> > > Sorry, I am not familiar with KASAN. How should I interpret this output?
> > > I see the paragraph above has references to rtw_cmd_thread. I assume
> > > that is its way of indicating that rtw_cmd_thread is the cause of the
> > > problem, but the one below refers to other functions. I'm not sure where
> > > I should start looking. I would start looking at `rtw_sitesurvey_cmd` and
> > > `rtw_scan_ch_decision`, which call the memcpy on the
> > > rtw_ieee80211_channel structure, but they are not on the call trace.
> > > 
> > 
> > drivers/staging/r8188eu/core/rtw_cmd.c:276: memcpy() call.
> > 
> 
> What git hash are you on?  Is that the line:
> 
> 	memcpy(&psurveyPara->ch[i], &ch[i], sizeof(struct ieee80211_channel));
> 
> Are you sure that's the line which crashes?  According to Smatch and my
> simple grep that's dead code because "ch" is always NULL.
>

I don't think Pavel is referring to that line as it belongs to the
`rtw_sitesurvey_cmd` function. That func didnt appear in the dmesg output.
Rather I think Pavel refers to line 267, which is
`memcpy(pcmdbuf, pcmd->parmbuf, pcmd->cmdsz);` from function
`rtw_cmd_thread`. Changing rtw_ieee80211_channel to ieee80211_channel
must have messed up that line somewhere but I just can't figure out what
or why.
 
> > As Michael said the sizes of structures do not mach and the memcpy writes
> > below allocated buffer.
> 
> I feel a bit bad for not spotting this in review because I was looking
> for that kind of bug.  I still don't immediately spot where the bug is.

Sorry I put you in this position, Dan.

> 
> regards,
> dan carpenter