[Linux] klogd 1.3-22 buffer overflow

From: Michal Zalewski (lcamtufat_private)
Date: Thu Sep 10 1998 - 01:26:06 PDT

  • Next message: Michal Zalewski: "Re: [Linux] klogd 1.3-22 buffer overflow"

    Good morning,
    
    This time - buffer overflow in Linux klogd daemon from sysklogd-1.3
    package (up to release 22 - affects Red Hat 5.x and Slackware 3.x, no data
    about other distributions).
    
    The problem:
    
    Kernel messages are stored in 4 kB cyclic printk ring. Klogd reads this
    buffer using /proc/kmsg to 4 kB long buffer, that's good. But then, data
    is split into lines, by copying data until '\n' is reached. What a pity,
    line buffer is only 1 kB long - sometimes, it's not enough... Exploitable?
    Could be...
    
    Impact:
    
    To exploit this security hole, we have to generate very long kernel
    message (or a lot of short messages with no '\n' inside). There are two
    potential ways of doing this:
    
    a) In kernel source (or any of installed modules), find printk not
       terminated with '\n'. There are some old, obscure messages both in
       2.0.xx and 2.1.xxx. Yep, but what now? You have to generate it :-S
       It's especially easy when poking with strange network packets
       (so it's possible to perform remote DoS attack). Unfortunately, DoS
       if probably all you can do - enjoy SEGV in klogd daemon, or (better?),
       by overwriting fd to /proc/kmsg lyingo on the stack, increase LA and
       generate enormous amount of error messages like 'Cannot read /proc
       filesystem', apparently from kernel.
    
    b) ...or, in kernel (2.1.xxx is more interesting), locate any printk with
       %s in format string, where substituted string depends in some way on
       luser (process/filename?). Then, you should be able to parse arbitrary
       shellcode into buffer, obtaining root privledges.
    
    Solution:
    
    In klog.c, at the beginning, there are two '#define's. First one is
    responsible for main buffer size - don't change it, 4096 should be ok. The
    next one is line buffer size - hmm, replace 1024 with 4096, for example...
    Or, better, implement some range checking ;>
    
    Quick vunerability test:
    
    -- gcc -c -O3 test.c; insmod test; rmmod test --
    #define MODULE
    #define __KERNEL__
    
    #include <linux/module.h>
    #include <linux/kernel.h>
    #include <linux/types.h>
    #include <linux/string.h>
    #include <linux/malloc.h>
    #include <asm/unistd.h>
    #include <linux/version.h>
    #include <asm/string.h>
    
    int init_module(void) {
      printk("INSERT_ABOUT_2000_BYTES_OF_JUNK_HERE\n"); return 0;
    }
    
    void cleanup_module(void) {}
    --
    
    Modify this source by increasing amount of junk after printk, compile,
    insmod and watch out what happened to klogd.
    
    _______________________________________________________________________
    Michal Zalewski [lcamtufat_private] [ENSI / marchew] [dione.ids.pl SYSADM]
    [http://linux.lepszy.od.kobiety.pl/~lcamtuf/] <=--=> bash$ :(){ :|:&};:
    [voice phone: +48 (0) 22 813 25 86] ? [pager (MetroBip): 0 642 222 813]
    Iterowac jest rzecza ludzka, wykonywac rekursywnie - boska [P. Deutsch]
    



    This archive was generated by hypermail 2b30 : Fri Apr 13 2001 - 14:22:44 PDT