linux-kernel - Re: [PATCH v2] firmware_loader: Block path traversal

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: <CAG48ez3A=NZ9GqkQv9U6871ciNc+Yy=AvPfm3UgeXfMyh=0+oQ@mail.gmail.com>
Date: Sat, 24 Aug 2024 03:34:20 +0200
From: Jann Horn <jannh@...gle.com>
To: Danilo Krummrich <dakr@...nel.org>
Cc: Luis Chamberlain <mcgrof@...nel.org>, Russ Weight <russ.weight@...ux.dev>, 
	Danilo Krummrich <dakr@...hat.com>, Greg Kroah-Hartman <gregkh@...uxfoundation.org>, 
	"Rafael J. Wysocki" <rafael@...nel.org>, linux-kernel@...r.kernel.org, stable@...r.kernel.org
Subject: Re: [PATCH v2] firmware_loader: Block path traversal

On Sat, Aug 24, 2024 at 2:31 AM Danilo Krummrich <dakr@...nel.org> wrote:
> On Fri, Aug 23, 2024 at 08:38:55PM +0200, Jann Horn wrote:
> > Fix it by rejecting any firmware names containing ".." path components.
[...]
> > +/*
> > + * Reject firmware file names with ".." path components.
> > + * There are drivers that construct firmware file names from device-supplied
> > + * strings, and we don't want some device to be able to tell us "I would like to
> > + * be sent my firmware from ../../../etc/shadow, please".
> > + *
> > + * Search for ".." surrounded by either '/' or start/end of string.
> > + *
> > + * This intentionally only looks at the firmware name, not at the firmware base
> > + * directory or at symlink contents.
> > + */
> > +static bool name_contains_dotdot(const char *name)
> > +{
> > +     size_t name_len = strlen(name);
> > +     size_t i;
> > +
> > +     if (name_len < 2)
> > +             return false;
> > +     for (i = 0; i < name_len - 1; i++) {
> > +             /* do we see a ".." sequence? */
> > +             if (name[i] != '.' || name[i+1] != '.')
> > +                     continue;
> > +
> > +             /* is it a path component? */
> > +             if ((i == 0 || name[i-1] == '/') &&
> > +                 (i == name_len - 2 || name[i+2] == '/'))
> > +                     return true;
> > +     }
> > +     return false;
> > +}
>
> Why do you open code it, instead of using strstr() and strncmp() like you did
> in v1? I think your approach from v1 read way better.

The code in v1 was kinda sloppy - it was probably good enough for this
check, but not good enough to put in a function called
name_contains_dotdot() that is documented to exactly search for any
".." components.

Basically, the precise regex we have to search for is something like
/(^|/)\.\.($|/)/

To implement that by searching for substrings like in v1, we'd have to
search for each possible combination of the capture groups in the
regex, which gives the following four (pow(2,2)) patterns:

<start>..<end>
<start>../
/..<end>
/../

So written like in v1, that'd look something like:

if (strcmp(name, "..") == 0 || strncmp(name, "../", 3) == 0 ||
strstr(name, "/../") != NULL || (name_len >= 3 &&
strcmp(name+name_len-3, "/..") == 0)))
  return true;

Compared to that, I prefer the code I wrote in v2, since it is less
repetitive. But if you want, I can change it to the expression I wrote
just now.