full-disclosure - Linux 2.4.x execve() file read race vulnerability

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

From: paul at starzetz.de (Paul Starzetz)
Subject: Linux 2.4.x execve() file read race vulnerability

Hi people,

again it is time to discover a funny bug inside the Linux execve() 
system call.

Details:
---------

While looking at the execve() code I've found the following piece of 
code (from fs/binfmt_elf.c):

static int load_elf_binary(struct linux_binprm * bprm, struct pt_regs * 
regs)
{
    struct file *interpreter = NULL; /* to shut gcc up */

[...]

    retval = kernel_read(bprm->file, elf_ex.e_phoff, (char *) 
elf_phdata, size);
    if (retval < 0)
        goto out_free_ph;

    retval = get_unused_fd();
    if (retval < 0)
        goto out_free_ph;
    get_file(bprm->file);
    fd_install(elf_exec_fileno = retval, bprm->file);

So, during the execution of new binary, the opened file descriptor to 
the executable is put into the file table of the current (the caller of 
execve()) process. This can be exploited creating a file sharing 
parent/child pair by means of the clone() syscall and reading the file 
descriptor from one of them.

Further, the check for shared files structure (in compute_creds() from 
exec.c) is made to late, so even the parent can successfully exit after 
playing games on that file descriptor and the child (if setuid) is 
executed under full privileges. I wrote a simple setuid binary dump 
utility so far, but further implications (due to the complexity of the 
execve() syscall) may be possible...

Lets illustrate the vulnerability:

paul@...gy:~> ls -l /bin/ping
-rws--x--x    1 root     root        29680 Oct 25  2001 /bin/ping

so the setuid ping binary can be only executed by anyone, but not read.

Now we start the suid dumper (while playing with the disk on another 
console like cat /usr/bin/* >/dev/null) :

paul@...gy:~> while true ; do ./suiddmp /bin/ping -c 1 127.0.0.1 ; if 
test $? -eq 1 ; then exit 1 ; fi; done 2>/dev/null | grep -A5 suc

and after few seconds:

Parent success stating:
uid 0 gid 0 mode 104711 inode 9788 size 29680
PING 127.0.0.1 (127.0.0.1) from 127.0.0.1 : 56(84) bytes of data.
64 bytes from 127.0.0.1: icmp_seq=1 ttl=64 time=94 usec

--- 127.0.0.1 ping statistics ---

paul@...gy:~> ls -l
total 7132
-rwxr-xr-x    1 paul     users       29680 Jun 26 19:17 suid.dump
[...]

paul@...gy:~> ./suid.dump
Usage: ping [-LRUbdfnqrvVaA] [-c count] [-i interval] [-w deadline]
            [-p pattern] [-s packetsize] [-t ttl] [-I interface or address]
            [-M mtu discovery hint] [-S sndbuf]
            [ -T timestamp option ] [ -Q tos ] [hop1 ...] destination

Obviously the setuid binary has been duplicated :-) (but with no setuid 
flag of course).

Source also available at:

http://www.starzetz.com/paul/suiddmp.c

/ih

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: suiddmp.c
Url: http://lists.grok.org.uk/pipermail/full-disclosure/attachments/20030626/1402768e/suiddmp.c