Re: fs/proc/base.c: text md5sums; tgid vs tid; and INF vs ONE?

[Arvid Brodin <Arvid.Brodin@xdin.com>]

On 2012-10-30 22:22, Al Viro wrote:
> On Tue, Oct 30, 2012 at 09:02:33PM +0000, Arvid Brodin wrote:
> 
>> +config PROC_TEXT_MD5SUM
>> +	bool "/proc/<pid>/text_md5sum support"
>> +	depends on PROC_FS
>> +	select CRYPTO
>> +	select CRYPTO_MD5
>> +	help
>> +	  Read /proc/<pid>/text_md5sum to get the kernel to perform an MD5
>> +	  checksum over the process' text segment and print the result. Can be
>> +	  used to make sure a process' code has not been tampered with.
> 
> Sorry, but this is pointless.  Any attacker capable of modifying the code
> will be just as capable of modifying pointers to functions in data segment.
> IOW, you are not making sure of anything useful.

On 2012-10-30 22:23, Cyrill Gorcunov wrote:
> I don't think this increments security by any means. start/end-code are rather
> informative fields which are set when program being started, so one can ptrace
> it, alloc new exec area, put evil code there, tuneup cs:ip and restore original
> program contents, you won't even notice that.

You are both correct of course. Actually, I was kind of sloppy when I wrote the
Kconfig help text. The following more accurately describes the intended use. Would
this make the patch more acceptable?

+config PROC_TEXT_MD5SUM
+	bool "/proc/<pid>/text_md5sum support"
+	depends on PROC_FS
+	select CRYPTO
+	select CRYPTO_MD5
+	help
+	  Read /proc/<pid>/text_md5sum to get the kernel to perform an MD5
+	  checksum over the process' text segment and print the result. This
+	  can detect some cases where the system RAM has been disturbed by
+	  e.g. EMC or cosmic radiation (on systems where ECC is not available).
+	  It might also detect some accidental or malicious modifications of
+	  executables, where the perpetrator has not bothered to cover up the
+	  tracks.


-- 
Arvid Brodin | Consultant (Linux)
XDIN AB | Knarrarnäsgatan 7 | SE-164 40 Kista | Sweden | xdin.com