netdev - Re: [PATCH] net: ax88179: add proper error handling of usb read errors

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] [day] [month] [year] [list]

Message-ID: <0ece0ee4-baca-bff6-89e0-ad10483a5f8e@suse.com>
Date:   Thu, 14 Apr 2022 11:13:25 +0200
From:   Oliver Neukum <oneukum@...e.com>
To:     Dan Carpenter <dan.carpenter@...cle.com>,
        Oliver Neukum <oneukum@...e.com>
CC:     David Kahurani <k.kahurani@...il.com>, netdev@...r.kernel.org,
        syzbot <syzbot+d3dbdf31fbe9d8f5f311@...kaller.appspotmail.com>,
        davem@...emloft.net, jgg@...pe.ca, kuba@...nel.org,
        linux-kernel@...r.kernel.org, linux-usb@...r.kernel.org,
        Phillip Potter <phil@...lpotter.co.uk>,
        syzkaller-bugs@...glegroups.com, arnd@...db.de,
        Pavel Skripkin <paskripkin@...il.com>
Subject: Re: [PATCH] net: ax88179: add proper error handling of usb read
 errors



On 14.04.22 10:21, Dan Carpenter wrote:
> On Thu, Apr 14, 2022 at 09:31:56AM +0200, Oliver Neukum wrote:
>>
>> On 13.04.22 17:32, Dan Carpenter wrote:
>>> Bug: buffer partially filled.  Information leak.
>>>
>>> If you return the bytes then the only correct way to write error
>>> handling is:
>>>
>>> 	if (ret < 0)
>>> 		return ret;
>>> 	if (ret != size)
>>> 		return -EIO;
>>>
>> You have to make up your mind on whether you ever need to read
>> answer of a length not known before you try it. The alternative of
>> passing a pointer to an integer for length is worse.
> How is it worse?  Can you give an example, so I will write a static
> checker rule for it?
My favorite example would be:

int usb_bulk_msg(struct usb_device *usb_dev, unsigned int pipe,
                void *data, int len, int *actual_length, int timeout)

The actual_length parameter is horrible. Now, there are situations like this
where this pattern is unavoidable, because in addition to an error you
can get a partial completion of an operation.

Do I see it correctly that you would prefer this pattern even if
you could report either an error or a result in the function?
> There used to be more APIs that consistently caused bug after bug where
> we mixed positives success values with negative error codes.  We
> converted some bad offenders to return the positive as a parameter
> and I was really happy about that.
>
> Another example I used to see a lot is request_irq() saved to an
> unsigned.  These days I think GCC warns about that?  Maybe the build
> bots get to it before I do.
>
That needs to be checked for.. In fact while we are at it, do we check
for integer overflows?

    Regards
        Oliver