2 * Routines for AJP13 dissection
3 * Copyright 2002, Christopher K. St. John <cks@distributopia.com>
5 * $Id: packet-ajp13.c,v 1.9 2003/01/27 22:19:10 deniel Exp $
7 * Ethereal - Network traffic analyzer
8 * By Gerald Combs <gerald@ethereal.com>
9 * Copyright 1998 Gerald Combs
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License
13 * as published by the Free Software Foundation; either version 2
14 * of the License, or (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
36 #include <epan/packet.h>
37 #include <epan/conversation.h>
38 #include "packet-tcp.h"
42 /* IMPORTANT IMPLEMENTATION NOTES
44 * You need to be looking at: jk/doc/AJP13.html in the
45 * jakarta-tomcat-connectors repository.
47 * If you're an ethereal dissector guru, then you can skip the rest of
48 * this. I'm writing it all down because I've written 3 dissectors so
49 * far and every time I've forgotten it all and had to re-learn it
50 * from scratch. Not this time, damnit.
52 * Dissector routines get called in two phases:
54 * The first phase is an in-order traversal of every incoming
55 * frame. Since we know it's in-order, we can set up a "conversational
56 * state" that records context-sensitive stuff like "was there a
57 * content-length in the previous request". During this first pass
58 * through the data, the "tree" parameter might be null, or not. For
59 * the regular gui-based ethereal, it's null, which means we don't
60 * actually display the dissected data in the gui quite yet. For the
61 * text based interface, we might do the parsing and display both in
64 * The second phase happens when the data is actually displayed. In
65 * this pase the "tree" param is non-null, so you've got a hook to
66 * hang the parsed-out display data on. Since there might be gigabytes
67 * worth of capture data, the display code only calls the dissector
68 * for the stuff the user actually clicks on. So you have to assume
69 * the dissector is getting called on random frames, you can't depend
70 * on ordering anymore.
72 * But some parts of the AJP13 capture stream are context sensitive.
73 * That's no big deal during the first in-order pass, but the second
74 * phase requires us to display any random frame correctly. So during
75 * the first in-order phase we create a per-frame user data structure
76 * and attach it to the frame using p_add_proto_data.
78 * Since AJP13 is a TCP/IP based protocol, writing a dissector for it
79 * requires addressing several other issues:
81 * 1) TCP/IP segments can get retransmitted or be sent out of
82 * order. Users don't normally care, because the low-level kernel
83 * networking code takes care of reassembling them properly. But we're
84 * looking at raw network packets, aren't we? The stuff on the
85 * wire. Ethereal has been getting better and better at helping
86 * dissectors with this. I'm a little fuzzy on the details, but my
87 * uderstanding is that ethereal now contains a fairly substantial
88 * user-space TCP/IP stack so it can re-assemble the data. But I might
89 * be wrong. Since AJP13 is going to be used either on the loopback
90 * interface or on a LAN, it isn't likely to be a big issues anyway.
92 * 2) AJP13 packets (PDU's or protocol data unit's in
93 * networking-speak) don't necessarily line up with TCP segments. That
94 * is, one TCP segment can have more than one AJP13 PDU, or one AJP13
95 * PDU can stretch across multiple TCP segments. Assembling them is
96 * obviously possible, but a royal pain. During the "phase one"
97 * in-order pass you have to keep track of a bunch of offsets and
98 * store which PDU goes with which TCP segment. Luckly, recent
99 * (0.9.4+) versions of ethereal provide the "tcp_dissect_pdus()"
100 * function that takes care of much of the work. See the comments in
101 * packet-tcp.c, the example code in packet-dns.c, or check the
102 * ethereal-dev archives for details.
104 * 3) Ethereal isn't guaranteed to see all the data. I'm a little
105 * unclear on all the possible failure modes, but it comes down to: a)
106 * Not your fault: it's an imperfect world, we're eavesdroppers, and
107 * stuff happens. We might totally miss packets or get garbled
108 * data. Or b) Totally your fault: you turn on the capture during the
109 * middle of an AJP13 conversation and the capture starts out with
110 * half an AJP13 PDU. This code doesn't currently handle either case
111 * very well, but you can get arbitrarily clever. Like: put in tests
112 * to see if this packet has reasonable field values, and if it
113 * doesn't, walk the offset ahead until we see a matching magic number
114 * field, then re-test. But we don't do that now, and since we're
115 * using tcp_dissect_pdu's, I'm not sure how to do it.
121 * Request/response header codes. Common headers are stored as ints in
122 * an effort to improve performance. Why can't we just have one big
126 static const value_string req_header_codes[] = {
128 { 0x02, "accept-charset" },
129 { 0x03, "accept-encoding" },
130 { 0x04, "accept-language" },
131 { 0x05, "authorization" },
132 { 0x06, "connection" },
133 { 0x07, "content-type" },
134 { 0x08, "content-length" },
140 { 0x0E, "user-agent" },
144 static const value_string rsp_header_codes[] = {
145 { 0x01, "Content-Type" },
146 { 0x02, "Content-Language" },
147 { 0x03, "Content-Length" },
149 { 0x05, "Last-Modified" },
150 { 0x06, "Location" },
151 { 0x07, "Set-Cookie" },
152 { 0x08, "Set-Cookie2" },
153 { 0x09, "Servlet-Engine" },
155 { 0x0B, "WWW-Authenticate" },
159 static const value_string mtype_codes[] = {
162 { 2, "FORWARD REQUEST" },
163 { 3, "SEND BODY CHUNK" },
164 { 4, "SEND HEADERS" },
165 { 5, "END RESPONSE" },
166 { 6, "GET BODY CHUNK" },
171 static const value_string http_method_codes[] = {
188 { 17, "VERSION-CONTROL" },
191 { 20, "UNCHECKOUT" },
197 static int proto_ajp13 = -1;
198 static int hf_ajp13_magic = -1;
199 static int hf_ajp13_len = -1;
200 static int hf_ajp13_code = -1;
201 static int hf_ajp13_method = -1;
202 static int hf_ajp13_ver = -1;
203 static int hf_ajp13_uri = -1;
204 static int hf_ajp13_raddr = -1;
205 static int hf_ajp13_rhost = -1;
206 static int hf_ajp13_srv = -1;
207 static int hf_ajp13_port = -1;
208 static int hf_ajp13_sslp = -1;
209 static int hf_ajp13_nhdr = -1;
210 static int hf_ajp13_hname = -1;
211 static int hf_ajp13_hval = -1;
212 static int hf_ajp13_rlen = -1;
213 static int hf_ajp13_reusep = -1;
214 static int hf_ajp13_rstatus= -1;
215 static int hf_ajp13_rsmsg = -1;
216 static int hf_ajp13_data = -1;
217 static gint ett_ajp13 = -1;
220 typedef struct ajp13_conv_data {
222 int was_get_body_chunk;
226 typedef struct ajp13_frame_data {
232 /* ajp13, in sort of a belt-and-suspenders move, encodes strings with
233 * both a leading length field, and a trailing null. Mostly, see
234 * AJPv13.html. The returned length _includes_ the trailing null, if
237 * XXX - is there a tvbuff routine to handle this?
240 get_nstring(tvbuff_t *tvb, gint offset, guint8* cbuf, size_t cbuflen)
245 len = tvb_get_ntohs(tvb, offset);
252 if (copylen > cbuflen - 1)
253 copylen = cbuflen - 1;
254 tvb_memcpy(tvb, cbuf, offset+2, copylen);
255 cbuf[copylen] = '\0';
263 /* dissect a response. more work to do here.
266 display_rsp(tvbuff_t *tvb, packet_info *pinfo, proto_tree *ajp13_tree)
268 gchar* msg_code = NULL;
271 char mcode_buf[1024];
277 proto_tree_add_item(ajp13_tree, hf_ajp13_magic, tvb, pos, 2, 0);
283 proto_tree_add_item(ajp13_tree, hf_ajp13_len, tvb, pos, 2, 0);
288 mcode = tvb_get_guint8(tvb, pos);
289 msg_code = val_to_str(mcode, mtype_codes, "UNKNOWN");
290 sprintf(mcode_buf, "(%d) %s", mcode, msg_code);
292 proto_tree_add_string(ajp13_tree, hf_ajp13_code, tvb, pos, 1, mcode_buf);
295 if(check_col(pinfo->cinfo, COL_INFO))
296 col_append_str(pinfo->cinfo, COL_INFO, msg_code);
300 proto_tree_add_item(ajp13_tree, hf_ajp13_reusep, tvb, pos, 1, 0);
303 } else if (mcode == 4) {
305 guint8 rsmsg_bytes[8*1024]; /* DANGER WILL ROBINSON */
310 /* HTTP RESPONSE STATUS CODE
312 rcode_num = tvb_get_ntohs(tvb, pos);
314 proto_tree_add_item(ajp13_tree, hf_ajp13_rstatus, tvb, pos, 2, 0);
316 if(check_col(pinfo->cinfo, COL_INFO))
317 col_append_fstr(pinfo->cinfo, COL_INFO, ":%d", rcode_num);
319 /* HTTP RESPONSE STATUS MESSAGE
321 rsmsg_len = get_nstring(tvb, pos, rsmsg_bytes, sizeof rsmsg_bytes);
324 proto_tree_add_item(ajp13_tree, hf_ajp13_rsmsg, tvb, pos, rsmsg_len, 0);
326 /* dangerous assumption that we can just %s out raw bytes */
327 if(check_col(pinfo->cinfo, COL_INFO))
328 col_append_fstr(pinfo->cinfo, COL_INFO, " %s", rsmsg_bytes);
332 nhdr = tvb_get_ntohs(tvb, pos);
334 proto_tree_add_item(ajp13_tree, hf_ajp13_nhdr, tvb, pos, 2, 0);
339 for(i=0; i<nhdr; i++) {
349 guint8 hname_bytes[1024];
350 gchar hname_value[8*1024];
354 hcd = tvb_get_guint8(tvb, pos);
358 hid = tvb_get_guint8(tvb, pos);
361 hname = val_to_str(hid, rsp_header_codes, "UNKNOWN");
362 /* Content-Length header (encoded by 0x08) is special */
366 int hname_len = get_nstring(tvb, pos, hname_bytes, sizeof hname_bytes);
369 hname = (gchar*)hname_bytes; /* VERY EVIL */
377 hval_len = get_nstring(tvb, pos, hval, sizeof hval);
382 sprintf(hname_value, "%s : %s", hname, hval);
383 proto_tree_add_string(ajp13_tree, hf_ajp13_hval, tvb, orig_pos, dp, hname_value);
387 } else if (mcode == 6) {
389 proto_tree_add_item(ajp13_tree, hf_ajp13_rlen, tvb, pos, 2, 0);
393 /* MESSAGE DATA (COPOUT)
396 proto_tree_add_item(ajp13_tree, hf_ajp13_data, tvb, pos+2, -1, 0);
402 /* dissect a request body. see AJPv13.html, but the idea is that these
403 * packets, unlike all other packets, have no type field. you just
404 * sort of have to know that they're coming based on the previous
408 display_req_body(tvbuff_t *tvb, proto_tree *ajp13_tree)
410 /*printf("ajp13:display_req_body()\n");*/
414 guint8 body_bytes[128*1024]; /* DANGER WILL ROBINSON */
420 proto_tree_add_item(ajp13_tree, hf_ajp13_magic, tvb, pos, 2, 0);
425 proto_tree_add_item(ajp13_tree, hf_ajp13_len, tvb, pos, 2, 0);
430 body_len = get_nstring(tvb, pos, body_bytes, sizeof body_bytes);
431 proto_tree_add_item(ajp13_tree, hf_ajp13_data, tvb, pos+2, body_len-1, 0);
437 /* note that even if ajp13_tree is null on the first pass, we still
438 * need to dissect the packet in order to determine if there is a
439 * content-length, and thus if there is a subsequent automatic
440 * request-body transmitted in the next request packet. if there is a
441 * content-length, we record the fact in the conversation context.
442 * ref the top of this file for comments explaining the multi-pass
446 display_req_forward(tvbuff_t *tvb, packet_info *pinfo,
447 proto_tree *ajp13_tree,
467 proto_tree_add_item(ajp13_tree, hf_ajp13_magic, tvb, pos, 2, 0);
471 proto_tree_add_item(ajp13_tree, hf_ajp13_len, tvb, pos, 2, 0);
476 cod = tvb_get_guint8(tvb, 4);
478 gchar* msg_code = NULL;
479 char mcode_buf[1024];
480 msg_code = val_to_str(cod, mtype_codes, "UNKNOWN");
481 sprintf(mcode_buf, "(%d) %s", cod, msg_code);
482 proto_tree_add_string(ajp13_tree, hf_ajp13_code, tvb, pos, 1, mcode_buf);
486 /* HTTP METHOD (ENCODED AS INTEGER)
489 gchar* meth_code = NULL;
490 meth = tvb_get_guint8(tvb, pos);
491 meth_code = val_to_str(meth, http_method_codes, "UNKNOWN");
493 char mcode_buf[1024];
494 sprintf(mcode_buf, "(%d) %s", meth, meth_code);
495 proto_tree_add_string(ajp13_tree, hf_ajp13_method, tvb, pos, 1, mcode_buf);
497 if(check_col(pinfo->cinfo, COL_INFO))
498 col_append_str(pinfo->cinfo, COL_INFO, meth_code);
502 /* HTTP VERSION STRING
504 ver_len = get_nstring(tvb, pos, ver, sizeof ver);
505 pos+=2; /* skip over size */
507 proto_tree_add_item(ajp13_tree, hf_ajp13_ver, tvb, pos, ver_len, 0);
508 pos=pos+ver_len; /* skip over chars + trailing null */
512 uri_len = get_nstring(tvb, pos, uri, sizeof uri);
513 pos+=2; /* skip over size */
515 proto_tree_add_item(ajp13_tree, hf_ajp13_uri, tvb, pos, uri_len, 0);
516 pos=pos+uri_len; /* skip over chars + trailing null */
519 if(check_col(pinfo->cinfo, COL_INFO))
520 col_append_fstr(pinfo->cinfo, COL_INFO, " %s %s", uri, ver);
525 raddr_len = get_nstring(tvb, pos, raddr, sizeof raddr);
526 pos+=2; /* skip over size */
528 proto_tree_add_item(ajp13_tree, hf_ajp13_raddr, tvb, pos, raddr_len, 0);
529 pos=pos+raddr_len; /* skip over chars + trailing null */
533 rhost_len = get_nstring(tvb, pos, rhost, sizeof rhost);
534 pos+=2; /* skip over size */
536 proto_tree_add_item(ajp13_tree, hf_ajp13_rhost, tvb, pos, rhost_len, 0);
537 pos=pos+rhost_len; /* skip over chars + trailing null */
541 srv_len = get_nstring(tvb, pos, srv, sizeof srv);
542 pos+=2; /* skip over size */
544 proto_tree_add_item(ajp13_tree, hf_ajp13_srv, tvb, pos, srv_len, 0);
545 pos=pos+srv_len; /* skip over chars + trailing null */
550 proto_tree_add_item(ajp13_tree, hf_ajp13_port, tvb, pos, 2, 0);
556 proto_tree_add_item(ajp13_tree, hf_ajp13_sslp, tvb, pos, 1, 0);
561 nhdr = tvb_get_ntohs(tvb, pos);
564 proto_tree_add_item(ajp13_tree, hf_ajp13_nhdr, tvb, pos, 2, 0);
566 cd->content_length = 0;
570 for(i=0; i<nhdr; i++) {
580 guint8 hname_bytes[1024];
584 hcd = tvb_get_guint8(tvb, pos);
588 hid = tvb_get_guint8(tvb, pos);
591 hname = val_to_str(hid, req_header_codes, "UNKNOWN");
595 int hname_len = get_nstring(tvb, pos, hname_bytes, sizeof hname_bytes);
598 hname = (gchar*)hname_bytes; /* VERY EVIL */
606 hval_len = get_nstring(tvb, pos, hval, sizeof hval);
611 proto_tree_add_string_format(ajp13_tree, hf_ajp13_hval,
612 tvb, orig_pos, dp, hname,
613 "%s: %s", hname, hval);
617 cd->content_length = cl;
624 /* main dissector function. ethereal calls it for segments in both
628 dissect_ajp13_tcp_pdu(tvbuff_t *tvb, packet_info *pinfo, proto_tree *tree)
632 conversation_t *conv = NULL;
633 ajp13_conv_data *cd = NULL;
634 proto_tree *ajp13_tree = NULL;
635 ajp13_frame_data* fd = NULL;
637 /* conversational state really only does us good during the first
640 conv = find_conversation(&pinfo->src, &pinfo->dst, pinfo->ptype,
641 pinfo->srcport, pinfo->destport, 0);
643 conv = conversation_new(&pinfo->src, &pinfo->dst, pinfo->ptype,
644 pinfo->srcport, pinfo->destport, 0);
645 cd = (ajp13_conv_data*)malloc(sizeof(ajp13_conv_data));
646 cd->content_length = 0;
647 cd->was_get_body_chunk = 0;
648 conversation_add_proto_data(conv, proto_ajp13, cd);
650 cd = (ajp13_conv_data*)conversation_get_proto_data(conv, proto_ajp13);
653 /* we use the per segment user data to record the conversational
654 * state for use later on when we're called out of order (see
655 * comments at top of this file)
657 fd = (ajp13_frame_data*)p_get_proto_data(pinfo->fd, proto_ajp13);
659 /*printf("ajp13:dissect_ajp13_common():no frame data, adding");*/
660 /* since there's no per-packet user data, this must be the first
661 * time we've see the packet, and it must be the first "in order"
662 * pass through the data.
664 fd = (ajp13_frame_data*)malloc(sizeof(ajp13_frame_data));
665 p_add_proto_data(pinfo->fd, proto_ajp13, fd);
666 fd->is_request_body = 0;
667 if (cd->content_length) {
668 /* this is screwy, see AJPv13.html. the idea is that if the
669 * request has a body (as determined by the content-length
670 * header), then there's always an immediate follow-up PDU with
671 * no GET_BODY_CHUNK from the container.
673 fd->is_request_body = 1;
677 if (check_col(pinfo->cinfo, COL_INFO))
678 col_clear(pinfo->cinfo, COL_INFO);
680 mag = tvb_get_ntohs(tvb, 0);
681 len = tvb_get_ntohs(tvb, 2);
683 if (check_col(pinfo->cinfo, COL_PROTOCOL))
684 col_set_str(pinfo->cinfo, COL_PROTOCOL, "AJP13");
685 if (check_col(pinfo->cinfo, COL_INFO)) {
686 if (mag == 0x1234 && !fd->is_request_body)
687 col_append_fstr(pinfo->cinfo, COL_INFO, "%d:REQ:", conv->index);
688 else if (mag == 0x1234 && fd->is_request_body)
689 col_append_fstr(pinfo->cinfo, COL_INFO, "%d:REQ:Body", conv->index);
690 else if (mag == 0x4142)
691 col_append_fstr(pinfo->cinfo, COL_INFO, "%d:RSP:", conv->index);
693 col_set_str(pinfo->cinfo, COL_INFO, "AJP13 Error?");
698 ti = proto_tree_add_item(tree, proto_ajp13, tvb, 0, tvb_length(tvb), FALSE);
699 ajp13_tree = proto_item_add_subtree(ti, ett_ajp13);
704 if (fd->is_request_body)
705 display_req_body(tvb, ajp13_tree);
707 display_req_forward(tvb, pinfo, ajp13_tree, cd);
709 } else if (mag == 0x4142) {
711 display_rsp(tvb, pinfo, ajp13_tree);
718 /* given the first chunk of the AJP13 pdu, extract out and return the
719 * packet length. see comments in packet-tcp.c:tcp_dissect_pdus().
722 get_ajp13_pdu_len(tvbuff_t *tvb, int offset)
726 magic = tvb_get_ntohs(tvb, offset);
727 plen = tvb_get_ntohs(tvb, offset+2);
734 /* Code to actually dissect the packets.
737 dissect_ajp13(tvbuff_t *tvb, packet_info *pinfo, proto_tree *tree)
739 /* Set up structures needed to add the protocol subtree and manage it
741 tcp_dissect_pdus(tvb, pinfo, tree,
742 TRUE, /* desegment or not */
743 4, /* magic + length */
744 get_ajp13_pdu_len, /* use first 4, calc data len */
745 dissect_ajp13_tcp_pdu); /* the naive dissector */
751 proto_register_ajp13(void)
753 static hf_register_info hf[] = {
755 { "Magic", "ajp13.magic", FT_BYTES, BASE_HEX, NULL, 0x0, "Magic Number",
759 { "Length", "ajp13.len", FT_UINT16, BASE_DEC, NULL, 0x0, "Data Length",
763 { "Code", "ajp13.code", FT_STRING, BASE_DEC, NULL, 0x0, "Type Code",
767 { "Method", "ajp13.method", FT_STRING, BASE_DEC, NULL, 0x0, "HTTP Method",
771 { "Version", "ajp13.ver", FT_STRING, BASE_DEC, NULL, 0x0, "HTTP Version",
775 { "URI", "ajp13.uri", FT_STRING, BASE_DEC, NULL, 0x0, "HTTP URI",
779 { "RADDR", "ajp13.raddr", FT_STRING, BASE_DEC, NULL, 0x0, "Remote Address",
783 { "RHOST", "ajp13.rhost", FT_STRING, BASE_DEC, NULL, 0x0, "Remote Host",
787 { "SRV", "ajp13.srv", FT_STRING, BASE_DEC, NULL, 0x0, "Server",
791 { "PORT", "ajp13.port", FT_UINT16, BASE_DEC, NULL, 0x0, "Port",
795 { "SSLP", "ajp13.sslp", FT_UINT8, BASE_DEC, NULL, 0x0, "Is SSL?",
799 { "NHDR", "ajp13.nhdr", FT_UINT16, BASE_DEC, NULL, 0x0, "Num Headers",
803 { "HNAME", "ajp13.hname", FT_STRING, BASE_DEC, NULL, 0x0, "Header Name",
807 { "HVAL", "ajp13.hval", FT_STRING, BASE_DEC, NULL, 0x0, "Header Value",
811 { "RLEN", "ajp13.rlen", FT_UINT16, BASE_DEC, NULL, 0x0, "Requested Length",
815 { "REUSEP", "ajp13.reusep", FT_UINT8, BASE_DEC, NULL, 0x0, "Reuse Connection?",
819 { "RSTATUS", "ajp13.rstatus", FT_UINT16, BASE_DEC, NULL, 0x0, "HTTP Status Code",
823 { "RSMSG", "ajp13.rmsg", FT_STRING, BASE_DEC, NULL, 0x0, "HTTP Status Message",
827 { "Data", "ajp13.data", FT_STRING, BASE_DEC, NULL, 0x0, "Data",
832 static gint *ett[] = {
836 /* Register the protocol name and description
838 proto_ajp13 = proto_register_protocol("Apache JServ Protocol v1.3", "AJP13", "ajp13");
840 proto_register_field_array(proto_ajp13, hf, array_length(hf));
841 proto_register_subtree_array(ett, array_length(ett));
847 proto_reg_handoff_ajp13(void)
849 dissector_handle_t ajp13_handle;
850 ajp13_handle = create_dissector_handle(dissect_ajp13, proto_ajp13);
851 dissector_add("tcp.port", 8009, ajp13_handle);