[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20120723010447.GA23410@dztty>
Date: Mon, 23 Jul 2012 02:04:47 +0100
From: Djalal Harouni <tixxdz@...ndz.org>
To: Oleg Nesterov <oleg@...hat.com>
Cc: linux-kernel@...r.kernel.org, kernel-hardening@...ts.openwall.com,
Al Viro <viro@...iv.linux.org.uk>,
Andrew Morton <akpm@...ux-foundation.org>,
Vasiliy Kulikov <segoon@...nwall.com>,
WANG Cong <xiyou.wangcong@...il.com>,
Solar Designer <solar@...nwall.com>,
Kees Cook <keescook@...omium.org>
Subject: Re: [PATCH] proc: do not allow negative offsets on
/proc/<pid>/environ
Hi Oleg,
On Sun, Jul 22, 2012 at 10:00:49PM +0200, Oleg Nesterov wrote:
> On 07/22, Djalal Harouni wrote:
> >
> > __mem_open() which is called by both /proc/<pid>/environ and
> > /proc/<pid>/mem ->open() handlers will allow the use of negative offsets.
> > /proc/<pid>/mem has negative offsets but not /proc/<pid>/environ.
>
> Probablt the patch makes sense, but I can't understand the changelog...
>
> > Allowing negative offsets on /proc/<pid>/environ can turn it to act like
> > /proc/<pid>/mem. A negative offset will pass the
> > fs/read_write.c:lseek_execute() and the environ_read() checks and will
> > point to another VMA.
>
> which VMA?
It depends on the offset. Please see below.
> environ_read() can only read the memory from [env_start, env_end], and
> it should check *ppos anyway to ensure it doesn't read something else.
Yes I agree, but currently that's not the case, there are no checks on *ppos.
So if you pass a negative offset you will be able to read from an arbitrary
address.
I'll send another patch tomorrow to add the checks for *ppos.
Since negative offsets are allowed we can pass it to lseek():
1) ->llseek()
-> generic_file_llseek()
-> generic_file_llseek_size()
-> lseek_execute()
inside fs/read_write.c:lseek_execute() we pass the two checks and
file->f_pos will be updated.
2) ->read()
-> environ_read()
inside environ_read() there is only a one check:
int this_len = mm->env_end - (mm->env_start + src);
if (this_len <= 0)
break;
Here 'src' is 'src = *ppos' the negative offset converted to unsigned long
and (mm->env_start + src) can overflow and point to another VMA.
int this_len = mm->env_end - (mm->env_start + src)
'this_len' will be positive and we pass that check.
I also don't like the truncation of the result to 'int this_len'
A quick example to reproduce it:
New kernels /proc/<pid>/stat include 'mm->env_start', third number from
the end.
To read the .text area from 0x00400000:
0x00400000 - (mm->env_start == 140733359794601) = negative_offset
$ ./mem_environ /proc/$(pidof cat)/environ 140733359794601 | hexdump -C -v
00000000 7f 45 4c 46 02 01 01 00 00 00 00 00 00 00 00 00 |.ELF............|
00000010 02 00 3e 00 01 00 00 00 a0 17 40 00 00 00 00 00 |..>.......@.....|
00000020 40 00 00 00 00 00 00 00 40 c5 00 00 00 00 00 00 |@.......@.......|
...
mem_environ is just a program that calculats the negative offset,
open(/proc/<pid>/environ), lseek() and read().
The source is attached, just run this command to test it:
$ ./mem_environ /proc/self/environ 0x0 | hexdump -C -v
In rare cases it will not work, I don't know why.
> > static int mem_open(struct inode *inode, struct file *file)
> > {
> > - return __mem_open(inode, file, PTRACE_MODE_ATTACH);
> > + int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
> > + if (!ret)
> > + /* OK to pass negative loff_t, we can catch out-of-range */
> > + file->f_mode |= FMODE_UNSIGNED_OFFSET;
> > +
> > + return ret;
>
> I guess you can set FMODE_UNSIGNED_OFFSET unconditionally, it doesn't
> matter if __mem_open() fails. But I won't insist.
Sure.
> Oleg.
>
Thanks Oleg. BTW should I resend the patch with a better changelog entry ?
I'll also add another patch to check the offsets inside environ_read().
--
tixxdz
http://opendz.org
View attachment "mem_environ.c" of type "text/x-c" (1766 bytes)
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
