Re: [PATCH] iproute2: Conforming to -D_FORTIFY_SOURCE=2 restrictions

On Wed, 19 Oct 2011 13:30:51 +0200
Eric Dumazet [off-list ref] wrote:

Le mercredi 19 octobre 2011 à 17:15 +0800, Bin Li a écrit :

quoted

Stephen,

 You can reproduce this issue in 2.6.37 like below. And the previous
gdb log is after the install the debuginfo package in SUSE.

# ip -6 xfrm state add src 3ffe:501:ffff:ff03:21a:64ff:fe12:e4c1 dst
3ffe:501:ffff:ff05:200:ff:fe00:c1c1 proto ah spi 0x1000 mode transport
auth md5 "TAHITEST89ABCDEF"

*** buffer overflow detected ***: ip terminated
======= Backtrace: =========
/lib/libc.so.6(__fortify_fail+0x40)[0xb76d0070]
/lib/libc.so.6(+0xe8e27)[0xb76cde27]
/lib/libc.so.6(+0xe8317)[0xb76cd317]
ip[0x806d6c4]
ip(do_xfrm_state+0x120)[0x806dc70]
ip(do_xfrm+0x81)[0x806ad51]
ip[0x804c355]
ip(main+0x476)[0x804caa6]
/lib/libc.so.6(__libc_start_main+0xfe)[0xb75fbc2e]
ip[0x804c261]
======= Memory map: ========
08048000-08087000 r-xp 00000000 08:01 4465       /sbin/ip
08087000-08088000 r--p 0003e000 08:01 4465       /sbin/ip
08088000-0808a000 rw-p 0003f000 08:01 4465       /sbin/ip
0808a000-080ad000 rw-p 00000000 00:00 0          [heap]
b75c6000-b75e2000 r-xp 00000000 08:01 131084     /lib/libgcc_s.so.1
b75e2000-b75e3000 r--p 0001b000 08:01 131084     /lib/libgcc_s.so.1
b75e3000-b75e4000 rw-p 0001c000 08:01 131084     /lib/libgcc_s.so.1
b75e4000-b75e5000 rw-p 00000000 00:00 0
b75e5000-b774b000 r-xp 00000000 08:01 131375     /lib/libc-2.11.3.so
b774b000-b774c000 ---p 00166000 08:01 131375     /lib/libc-2.11.3.so
b774c000-b774e000 r--p 00166000 08:01 131375     /lib/libc-2.11.3.so
b774e000-b774f000 rw-p 00168000 08:01 131375     /lib/libc-2.11.3.so
b774f000-b7752000 rw-p 00000000 00:00 0
b7752000-b7755000 r-xp 00000000 08:01 131428     /lib/libdl-2.11.3.so
b7755000-b7756000 r--p 00002000 08:01 131428     /lib/libdl-2.11.3.so
b7756000-b7757000 rw-p 00003000 08:01 131428     /lib/libdl-2.11.3.so
b7774000-b7775000 rw-p 00000000 00:00 0
b7775000-b7794000 r-xp 00000000 08:01 154467     /lib/ld-2.11.3.so
b7794000-b7795000 r--p 0001e000 08:01 154467     /lib/ld-2.11.3.so
b7795000-b7796000 rw-p 0001f000 08:01 154467     /lib/ld-2.11.3.so
bfa02000-bfa23000 rw-p 00000000 00:00 0          [stack]
ffffe000-fffff000 r-xp 00000000 00:00 0          [vdso]
Aborted

And If without -D_FORTIFY_SOURCE=2 in gcc, it works fine, so It's a
bug in iproute2 which is not conforming to -D_FORTIFY_SOURCE=2
restrictions.

FORTIFY assumes we cant copy a string on alg.u.alg.alg_key !

This completely precludes 0-sized arrays

struct xfrm_algo {
        char            alg_name[64];
        unsigned int    alg_key_len;    /* in bits */
        char            alg_key[0];
};

struct {
      union {
          struct xfrm_algo alg;
          struct xfrm_algo_aead aead;
          struct xfrm_algo_auth auth;
      } u;
      char buf[XFRM_ALGO_KEY_BUF_SIZE];
} alg = {};

I would say its a FORTIFY bug. This kind of construct is perfectly
valid.

Maybe it will handle flexible style arrays.
See also:
   http://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html

At this time, I won't accept the patch that uses alloca() just to deal
with this FORTIFY bug.