linux-kernel - Re: [PATCH][next] mtd: rawnand: ensure return variable is initialized

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20210607085711.65c64c58@xps13>
Date:   Mon, 7 Jun 2021 08:57:11 +0200
From:   Miquel Raynal <miquel.raynal@...tlin.com>
To:     Dan Carpenter <dan.carpenter@...cle.com>
Cc:     Colin King <colin.king@...onical.com>,
        Richard Weinberger <richard@....at>,
        Vignesh Raghavendra <vigneshr@...com>,
        linux-mtd@...ts.infradead.org, kernel-janitors@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH][next] mtd: rawnand: ensure return variable is
 initialized

Hi Dan,

Dan Carpenter <dan.carpenter@...cle.com> wrote on Tue, 1 Jun 2021
15:14:02 +0300:

> On Thu, May 27, 2021 at 05:03:09PM +0200, Miquel Raynal wrote:
> > Hi Colin,
> > 
> > Colin King <colin.king@...onical.com> wrote on Thu, 27 May 2021
> > 15:50:48 +0100:
> >   
> > > From: Colin Ian King <colin.king@...onical.com>
> > > 
> > > Currently there are corner cases where spec_times is NULL and
> > > chip->parameters.onfi or when best_mode is zero where ret is  
> > 
> >                        ^
> > something is missing here, the sentence is not clear
> >   
> > > not assigned a value and an uninitialized return value can be
> > > returned. Fix this by ensuring ret is initialized to -EINVAL.  
> > 
> > I don't see how this situation can happen.
> > 
> > In both cases, no matter the value of best_mode, the for loop will
> > always execute at least one time (mode 0) so ret will be populated.
> > 
> > Maybe the robot does not know that best_mode cannot be negative and
> > should be defined unsigned, but the current patch is invalid.
> >  
> 
> People think list counter unsigned is a good idea, but it's a terrible
> idea and has caused hundreds of bugs for me to fix/report over the
> years.  *grumble*.
> 
> Anyway, I was revisiting this code because it showed up as a Smatch
> warning and the bug appears to be real.
> 
> 	best_mode = fls(chip->parameters.onfi->sdr_timing_modes) - 1;
> 
> The "onfi->sdr_timing_modes" comes from the hardware in nand_onfi_detect()
> and nothing checks that it is non-zero so "best_mode = fls(0) - 1;" is
> negative and "ret" is uninitialized.

In the ONFI specification, the sdr_timing_mode field is defined as
follow:

SDR timing mode support
BIT  VALUE MEANING
6-15 N/A   Reserved (0)
5    1     supports timing mode 5
4    1     supports timing mode 4
3    1     supports timing mode 3
2    1     supports timing mode 2
1    1     supports timing mode 1
0    1     supports timing mode 0, shall be 1

IOW sdr_timing_modes *cannot* be 0, or it is a truly deep and crazily
impacting hardware bug (so far I am not aware of any chip not returning
the right timing mode 0 value). Hence my proposal to turn best_mode as
unsigned. I honestly don't know what is the best option here and am
fully open to other suggestions to silence the robot.

Thanks,
Miquèl