[SR-Users] offset exceeds message size crash

Hi, I'm using kamailio 1.5.4-notls and I'm experimenting crashes when an UAC sends an INVITE with a content-length greater then the effective body length. The error messages written on the logs is: CRITICAL:core:del_lump: offset exceeds message size (1266 > 1161) aborting... and this is the backtrace of the generated core file: #0 0x00002ad718ab307b in raise () from /lib/libc.so.6 (gdb) bt #0 0x00002ad718ab307b in raise () from /lib/libc.so.6 #1 0x00002ad718ab484e in abort () from /lib/libc.so.6 #2 0x0000000000418f53 in del_lump (msg=0x66de00, offset=1266, len=12, type=HDR_OTHER_T) at data_lump.c:292 #3 0x00002ad71a8145ba in alter_mediaip (msg=0x66de00, body=<value optimized out>, oldip=0x7fff81fe6700, oldpf=<value optimized out>, newip=0x7fff81fe66e0, newpf=2, preserve=0) at nathelper.c:1857 #4 0x00002ad71a821a3a in force_rtp_proxy (msg=0x66de00, str1=<value optimized out>, str2=<value optimized out>, offer=<value optimized out>) at nathelper.c:2871 #5 0x00002ad71a8238df in rtpproxy_offer1_f (msg=0x66de00, str1=0x65f370 "cof", str2=<value optimized out>) at nathelper.c:2391 #6 0x000000000040cc5a in do_action (a=0x65f400, msg=0x66de00) at action.c:874 #7 0x000000000040f19f in run_action_list (a=<value optimized out>, msg=0x66de00) at action.c:145 #8 0x0000000000454155 in eval_expr (e=0x65f4d0, msg=0x66de00, val=0x0) at route.c:1171 #9 0x0000000000453bd7 in eval_expr (e=0x65f518, msg=0x66de00, val=0x0) at route.c:1488 #10 0x0000000000453b7f in eval_expr (e=0x65f560, msg=0x66de00, val=0x0) at route.c:1493 #11 0x000000000040c4c9 in do_action (a=0x65ffe8, msg=0x66de00) at action.c:729 #12 0x000000000040f19f in run_action_list (a=<value optimized out>, msg=0x66de00) at action.c:145 #13 0x000000000040dbc9 in do_action (a=0x660528, msg=0x66de00) at action.c:746 #14 0x000000000040f19f in run_action_list (a=<value optimized out>, msg=0x66de00) at action.c:145 #15 0x000000000040dbc9 in do_action (a=0x6606c8, msg=0x66de00) at action.c:746 #16 0x000000000040f19f in run_action_list (a=<value optimized out>, msg=0x66de00) at action.c:145 #17 0x000000000040dac5 in do_action (a=0x656790, msg=0x66de00) at action.c:120 #18 0x000000000040f19f in run_action_list (a=<value optimized out>, msg=0x66de00) at action.c:145 #19 0x000000000040dbc9 in do_action (a=0x656860, msg=0x66de00) at action.c:746 #20 0x000000000040f19f in run_action_list (a=<value optimized out>, msg=0x66de00) at action.c:145 #21 0x000000000040dac5 in do_action (a=0x6560b0, msg=0x66de00) at action.c:120 #22 0x000000000040f19f in run_action_list (a=<value optimized out>, msg=0x66de00) at action.c:145 #23 0x000000000040f4f3 in run_top_route (a=0x64b870, msg=0x66de00) at action.c:120 #24 0x0000000000444e90 in receive_msg ( buf=0x619a20 "INVITE sip:xxxxxxxx@xxxxxxxxxxxx SIP/2.0\r\nVia: SIP/2.0/UDP xxx.xxx.xxx.xxx:xxxx;branch=z9hG4bK-d8754z-24245342621eb55b-1---d8754z-;rport\r\nMax-Forwards: 69\r\nContact: <sip:xxxxxxxx@xxx.xxx.xxx.xxx"..., len=1161, rcv_info=0x7fff81fe86e0) at receive.c:175 #25 0x0000000000479254 in udp_rcv_loop () at udp_server.c:449 #26 0x0000000000427237 in main (argc=7, argv=0x7fff81fe88e8) at main.c:774 I couldn't get to reproduce this behavior in my test development (it has newer version of glibc) in which I only get the messages: ERROR:core:anchor_lump: offset exceeds message size (1125 > 714)... ERROR:nathelper:force_rtp_proxy: anchor_lump failed Looking into nathelper code, extract_body function I found that the body->len value is taken from Content-Length header, so i added the following piece of code: --- nhelpr_funcs.c.orig 2010-09-02 14:04:09.891649254 +0200 +++ nhelpr_funcs.c 2010-09-02 14:17:40.183747107 +0200 @@ -196,6 +196,12 @@ LM_ERR("message body has length zero\n"); goto error; } + + if (body->len + body->s > msg->buf + msg->len) { + LM_ERR("content-length exceeds packet-length by %d\n", + (body->len + body->s) - (msg->buf + msg->len)); + body->len=strlen(body->s); + } /* no need for parse_headers(msg, EOH), get_body will * parse everything */ This way if the Content-Length header is greater then the effective body length body->len is corrected with the real value. This solved for the moment, but I'm not sure if this is a good approach and I still don't understand why in the test platform I cannot reproduce the crash. Regards, Federico Cabiddu