blob: a77ab79bda986caab67aadeef6d82caa205e3300 [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * HTTP protocol analyzer
3 *
Willy Tarreau7c669d72008-06-20 15:04:11 +02004 * Copyright 2000-2008 Willy Tarreau <w@1wt.eu>
Willy Tarreaubaaee002006-06-26 02:48:02 +02005 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
13#include <ctype.h>
14#include <errno.h>
15#include <fcntl.h>
16#include <stdio.h>
17#include <stdlib.h>
18#include <string.h>
19#include <syslog.h>
Willy Tarreau42250582007-04-01 01:30:43 +020020#include <time.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020021
22#include <sys/socket.h>
23#include <sys/stat.h>
24#include <sys/types.h>
25
Willy Tarreau2dd0d472006-06-29 17:53:05 +020026#include <common/appsession.h>
27#include <common/compat.h>
28#include <common/config.h>
Willy Tarreaua4cd1f52006-12-16 19:57:26 +010029#include <common/debug.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020030#include <common/memory.h>
31#include <common/mini-clist.h>
32#include <common/standard.h>
Willy Tarreau0c303ee2008-07-07 00:09:58 +020033#include <common/ticks.h>
Willy Tarreau2dd0d472006-06-29 17:53:05 +020034#include <common/time.h>
35#include <common/uri_auth.h>
36#include <common/version.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020037
38#include <types/capture.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020039#include <types/global.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020040
Willy Tarreau8797c062007-05-07 00:55:35 +020041#include <proto/acl.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020042#include <proto/backend.h>
43#include <proto/buffers.h>
Willy Tarreau91861262007-10-17 17:06:05 +020044#include <proto/dumpstats.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020045#include <proto/fd.h>
46#include <proto/log.h>
Willy Tarreau58f10d72006-12-04 02:26:12 +010047#include <proto/hdr_idx.h>
Willy Tarreaub6866442008-07-14 23:54:42 +020048#include <proto/proto_tcp.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020049#include <proto/proto_http.h>
Willy Tarreau7f062c42009-03-05 18:43:00 +010050#include <proto/proxy.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020051#include <proto/queue.h>
Willy Tarreau7f062c42009-03-05 18:43:00 +010052#include <proto/server.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020053#include <proto/session.h>
Willy Tarreaucff64112008-11-03 06:26:53 +010054#include <proto/stream_interface.h>
Willy Tarreau2d212792008-08-27 21:41:35 +020055#include <proto/stream_sock.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020056#include <proto/task.h>
57
Willy Tarreau6d1a9882007-01-07 02:03:04 +010058#ifdef CONFIG_HAP_TCPSPLICE
59#include <libtcpsplice.h>
60#endif
Willy Tarreaubaaee002006-06-26 02:48:02 +020061
Willy Tarreau58f10d72006-12-04 02:26:12 +010062#define DEBUG_PARSE_NO_SPEEDUP
63#undef DEBUG_PARSE_NO_SPEEDUP
64
Willy Tarreau976f1ee2006-12-17 10:06:03 +010065/* This is used to perform a quick jump as an alternative to a break/continue
66 * instruction. The first argument is the label for normal operation, and the
67 * second one is the break/continue instruction in the no_speedup mode.
68 */
69
70#ifdef DEBUG_PARSE_NO_SPEEDUP
71#define QUICK_JUMP(x,y) y
72#else
73#define QUICK_JUMP(x,y) goto x
74#endif
75
Willy Tarreau1c47f852006-07-09 08:22:27 +020076/* This is used by remote monitoring */
Willy Tarreau0f772532006-12-23 20:51:41 +010077const char HTTP_200[] =
Willy Tarreau1c47f852006-07-09 08:22:27 +020078 "HTTP/1.0 200 OK\r\n"
79 "Cache-Control: no-cache\r\n"
80 "Connection: close\r\n"
81 "Content-Type: text/html\r\n"
82 "\r\n"
83 "<html><body><h1>200 OK</h1>\nHAProxy: service ready.\n</body></html>\n";
84
Willy Tarreau0f772532006-12-23 20:51:41 +010085const struct chunk http_200_chunk = {
86 .str = (char *)&HTTP_200,
87 .len = sizeof(HTTP_200)-1
88};
89
Willy Tarreaub463dfb2008-06-07 23:08:56 +020090const char *HTTP_301 =
91 "HTTP/1.0 301 Moved Permantenly\r\n"
92 "Cache-Control: no-cache\r\n"
93 "Connection: close\r\n"
94 "Location: "; /* not terminated since it will be concatenated with the URL */
95
Willy Tarreau0f772532006-12-23 20:51:41 +010096const char *HTTP_302 =
97 "HTTP/1.0 302 Found\r\n"
98 "Cache-Control: no-cache\r\n"
99 "Connection: close\r\n"
100 "Location: "; /* not terminated since it will be concatenated with the URL */
101
102/* same as 302 except that the browser MUST retry with the GET method */
103const char *HTTP_303 =
104 "HTTP/1.0 303 See Other\r\n"
105 "Cache-Control: no-cache\r\n"
106 "Connection: close\r\n"
107 "Location: "; /* not terminated since it will be concatenated with the URL */
108
Willy Tarreaubaaee002006-06-26 02:48:02 +0200109/* Warning: this one is an sprintf() fmt string, with <realm> as its only argument */
110const char *HTTP_401_fmt =
111 "HTTP/1.0 401 Unauthorized\r\n"
112 "Cache-Control: no-cache\r\n"
113 "Connection: close\r\n"
Willy Tarreau791d66d2006-07-08 16:53:38 +0200114 "Content-Type: text/html\r\n"
Willy Tarreaubaaee002006-06-26 02:48:02 +0200115 "WWW-Authenticate: Basic realm=\"%s\"\r\n"
116 "\r\n"
117 "<html><body><h1>401 Unauthorized</h1>\nYou need a valid user and password to access this content.\n</body></html>\n";
118
Willy Tarreau0f772532006-12-23 20:51:41 +0100119
120const int http_err_codes[HTTP_ERR_SIZE] = {
121 [HTTP_ERR_400] = 400,
122 [HTTP_ERR_403] = 403,
123 [HTTP_ERR_408] = 408,
124 [HTTP_ERR_500] = 500,
125 [HTTP_ERR_502] = 502,
126 [HTTP_ERR_503] = 503,
127 [HTTP_ERR_504] = 504,
128};
129
Willy Tarreau80587432006-12-24 17:47:20 +0100130static const char *http_err_msgs[HTTP_ERR_SIZE] = {
Willy Tarreau0f772532006-12-23 20:51:41 +0100131 [HTTP_ERR_400] =
Willy Tarreau80587432006-12-24 17:47:20 +0100132 "HTTP/1.0 400 Bad request\r\n"
Willy Tarreau0f772532006-12-23 20:51:41 +0100133 "Cache-Control: no-cache\r\n"
134 "Connection: close\r\n"
135 "Content-Type: text/html\r\n"
136 "\r\n"
137 "<html><body><h1>400 Bad request</h1>\nYour browser sent an invalid request.\n</body></html>\n",
138
139 [HTTP_ERR_403] =
140 "HTTP/1.0 403 Forbidden\r\n"
141 "Cache-Control: no-cache\r\n"
142 "Connection: close\r\n"
143 "Content-Type: text/html\r\n"
144 "\r\n"
145 "<html><body><h1>403 Forbidden</h1>\nRequest forbidden by administrative rules.\n</body></html>\n",
146
147 [HTTP_ERR_408] =
148 "HTTP/1.0 408 Request Time-out\r\n"
149 "Cache-Control: no-cache\r\n"
150 "Connection: close\r\n"
151 "Content-Type: text/html\r\n"
152 "\r\n"
153 "<html><body><h1>408 Request Time-out</h1>\nYour browser didn't send a complete request in time.\n</body></html>\n",
154
155 [HTTP_ERR_500] =
156 "HTTP/1.0 500 Server Error\r\n"
157 "Cache-Control: no-cache\r\n"
158 "Connection: close\r\n"
159 "Content-Type: text/html\r\n"
160 "\r\n"
161 "<html><body><h1>500 Server Error</h1>\nAn internal server error occured.\n</body></html>\n",
162
163 [HTTP_ERR_502] =
164 "HTTP/1.0 502 Bad Gateway\r\n"
165 "Cache-Control: no-cache\r\n"
166 "Connection: close\r\n"
167 "Content-Type: text/html\r\n"
168 "\r\n"
169 "<html><body><h1>502 Bad Gateway</h1>\nThe server returned an invalid or incomplete response.\n</body></html>\n",
170
171 [HTTP_ERR_503] =
172 "HTTP/1.0 503 Service Unavailable\r\n"
173 "Cache-Control: no-cache\r\n"
174 "Connection: close\r\n"
175 "Content-Type: text/html\r\n"
176 "\r\n"
177 "<html><body><h1>503 Service Unavailable</h1>\nNo server is available to handle this request.\n</body></html>\n",
178
179 [HTTP_ERR_504] =
180 "HTTP/1.0 504 Gateway Time-out\r\n"
181 "Cache-Control: no-cache\r\n"
182 "Connection: close\r\n"
183 "Content-Type: text/html\r\n"
184 "\r\n"
185 "<html><body><h1>504 Gateway Time-out</h1>\nThe server didn't respond in time.\n</body></html>\n",
186
187};
188
Willy Tarreau80587432006-12-24 17:47:20 +0100189/* We must put the messages here since GCC cannot initialize consts depending
190 * on strlen().
191 */
192struct chunk http_err_chunks[HTTP_ERR_SIZE];
193
Willy Tarreau42250582007-04-01 01:30:43 +0200194#define FD_SETS_ARE_BITFIELDS
195#ifdef FD_SETS_ARE_BITFIELDS
196/*
197 * This map is used with all the FD_* macros to check whether a particular bit
198 * is set or not. Each bit represents an ACSII code. FD_SET() sets those bytes
199 * which should be encoded. When FD_ISSET() returns non-zero, it means that the
200 * byte should be encoded. Be careful to always pass bytes from 0 to 255
201 * exclusively to the macros.
202 */
203fd_set hdr_encode_map[(sizeof(fd_set) > (256/8)) ? 1 : ((256/8) / sizeof(fd_set))];
204fd_set url_encode_map[(sizeof(fd_set) > (256/8)) ? 1 : ((256/8) / sizeof(fd_set))];
205
206#else
207#error "Check if your OS uses bitfields for fd_sets"
208#endif
209
Willy Tarreau80587432006-12-24 17:47:20 +0100210void init_proto_http()
211{
Willy Tarreau42250582007-04-01 01:30:43 +0200212 int i;
213 char *tmp;
Willy Tarreau80587432006-12-24 17:47:20 +0100214 int msg;
Willy Tarreau42250582007-04-01 01:30:43 +0200215
Willy Tarreau80587432006-12-24 17:47:20 +0100216 for (msg = 0; msg < HTTP_ERR_SIZE; msg++) {
217 if (!http_err_msgs[msg]) {
218 Alert("Internal error: no message defined for HTTP return code %d. Aborting.\n", msg);
219 abort();
220 }
221
222 http_err_chunks[msg].str = (char *)http_err_msgs[msg];
223 http_err_chunks[msg].len = strlen(http_err_msgs[msg]);
224 }
Willy Tarreau42250582007-04-01 01:30:43 +0200225
226 /* initialize the log header encoding map : '{|}"#' should be encoded with
227 * '#' as prefix, as well as non-printable characters ( <32 or >= 127 ).
228 * URL encoding only requires '"', '#' to be encoded as well as non-
229 * printable characters above.
230 */
231 memset(hdr_encode_map, 0, sizeof(hdr_encode_map));
232 memset(url_encode_map, 0, sizeof(url_encode_map));
233 for (i = 0; i < 32; i++) {
234 FD_SET(i, hdr_encode_map);
235 FD_SET(i, url_encode_map);
236 }
237 for (i = 127; i < 256; i++) {
238 FD_SET(i, hdr_encode_map);
239 FD_SET(i, url_encode_map);
240 }
241
242 tmp = "\"#{|}";
243 while (*tmp) {
244 FD_SET(*tmp, hdr_encode_map);
245 tmp++;
246 }
247
248 tmp = "\"#";
249 while (*tmp) {
250 FD_SET(*tmp, url_encode_map);
251 tmp++;
252 }
Willy Tarreau332f8bf2007-05-13 21:36:56 +0200253
254 /* memory allocations */
255 pool2_requri = create_pool("requri", REQURI_LEN, MEM_F_SHARED);
Willy Tarreau086b3b42007-05-13 21:45:51 +0200256 pool2_capture = create_pool("capture", CAPTURE_LEN, MEM_F_SHARED);
Willy Tarreau80587432006-12-24 17:47:20 +0100257}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200258
Willy Tarreau53b6c742006-12-17 13:37:46 +0100259/*
260 * We have 26 list of methods (1 per first letter), each of which can have
261 * up to 3 entries (2 valid, 1 null).
262 */
263struct http_method_desc {
264 http_meth_t meth;
265 int len;
266 const char text[8];
267};
268
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100269const struct http_method_desc http_methods[26][3] = {
Willy Tarreau53b6c742006-12-17 13:37:46 +0100270 ['C' - 'A'] = {
271 [0] = { .meth = HTTP_METH_CONNECT , .len=7, .text="CONNECT" },
272 },
273 ['D' - 'A'] = {
274 [0] = { .meth = HTTP_METH_DELETE , .len=6, .text="DELETE" },
275 },
276 ['G' - 'A'] = {
277 [0] = { .meth = HTTP_METH_GET , .len=3, .text="GET" },
278 },
279 ['H' - 'A'] = {
280 [0] = { .meth = HTTP_METH_HEAD , .len=4, .text="HEAD" },
281 },
282 ['P' - 'A'] = {
283 [0] = { .meth = HTTP_METH_POST , .len=4, .text="POST" },
284 [1] = { .meth = HTTP_METH_PUT , .len=3, .text="PUT" },
285 },
286 ['T' - 'A'] = {
287 [0] = { .meth = HTTP_METH_TRACE , .len=5, .text="TRACE" },
288 },
289 /* rest is empty like this :
290 * [1] = { .meth = HTTP_METH_NONE , .len=0, .text="" },
291 */
292};
293
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100294/* It is about twice as fast on recent architectures to lookup a byte in a
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +0200295 * table than to perform a boolean AND or OR between two tests. Refer to
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100296 * RFC2616 for those chars.
297 */
298
299const char http_is_spht[256] = {
300 [' '] = 1, ['\t'] = 1,
301};
302
303const char http_is_crlf[256] = {
304 ['\r'] = 1, ['\n'] = 1,
305};
306
307const char http_is_lws[256] = {
308 [' '] = 1, ['\t'] = 1,
309 ['\r'] = 1, ['\n'] = 1,
310};
311
312const char http_is_sep[256] = {
313 ['('] = 1, [')'] = 1, ['<'] = 1, ['>'] = 1,
314 ['@'] = 1, [','] = 1, [';'] = 1, [':'] = 1,
315 ['"'] = 1, ['/'] = 1, ['['] = 1, [']'] = 1,
316 ['{'] = 1, ['}'] = 1, ['?'] = 1, ['='] = 1,
317 [' '] = 1, ['\t'] = 1, ['\\'] = 1,
318};
319
320const char http_is_ctl[256] = {
321 [0 ... 31] = 1,
322 [127] = 1,
323};
324
325/*
326 * A token is any ASCII char that is neither a separator nor a CTL char.
327 * Do not overwrite values in assignment since gcc-2.95 will not handle
328 * them correctly. Instead, define every non-CTL char's status.
329 */
330const char http_is_token[256] = {
331 [' '] = 0, ['!'] = 1, ['"'] = 0, ['#'] = 1,
332 ['$'] = 1, ['%'] = 1, ['&'] = 1, ['\''] = 1,
333 ['('] = 0, [')'] = 0, ['*'] = 1, ['+'] = 1,
334 [','] = 0, ['-'] = 1, ['.'] = 1, ['/'] = 0,
335 ['0'] = 1, ['1'] = 1, ['2'] = 1, ['3'] = 1,
336 ['4'] = 1, ['5'] = 1, ['6'] = 1, ['7'] = 1,
337 ['8'] = 1, ['9'] = 1, [':'] = 0, [';'] = 0,
338 ['<'] = 0, ['='] = 0, ['>'] = 0, ['?'] = 0,
339 ['@'] = 0, ['A'] = 1, ['B'] = 1, ['C'] = 1,
340 ['D'] = 1, ['E'] = 1, ['F'] = 1, ['G'] = 1,
341 ['H'] = 1, ['I'] = 1, ['J'] = 1, ['K'] = 1,
342 ['L'] = 1, ['M'] = 1, ['N'] = 1, ['O'] = 1,
343 ['P'] = 1, ['Q'] = 1, ['R'] = 1, ['S'] = 1,
344 ['T'] = 1, ['U'] = 1, ['V'] = 1, ['W'] = 1,
345 ['X'] = 1, ['Y'] = 1, ['Z'] = 1, ['['] = 0,
346 ['\\'] = 0, [']'] = 0, ['^'] = 1, ['_'] = 1,
347 ['`'] = 1, ['a'] = 1, ['b'] = 1, ['c'] = 1,
348 ['d'] = 1, ['e'] = 1, ['f'] = 1, ['g'] = 1,
349 ['h'] = 1, ['i'] = 1, ['j'] = 1, ['k'] = 1,
350 ['l'] = 1, ['m'] = 1, ['n'] = 1, ['o'] = 1,
351 ['p'] = 1, ['q'] = 1, ['r'] = 1, ['s'] = 1,
352 ['t'] = 1, ['u'] = 1, ['v'] = 1, ['w'] = 1,
353 ['x'] = 1, ['y'] = 1, ['z'] = 1, ['{'] = 0,
354 ['|'] = 1, ['}'] = 0, ['~'] = 1,
355};
356
357
Willy Tarreau4b89ad42007-03-04 18:13:58 +0100358/*
359 * An http ver_token is any ASCII which can be found in an HTTP version,
360 * which includes 'H', 'T', 'P', '/', '.' and any digit.
361 */
362const char http_is_ver_token[256] = {
363 ['.'] = 1, ['/'] = 1,
364 ['0'] = 1, ['1'] = 1, ['2'] = 1, ['3'] = 1, ['4'] = 1,
365 ['5'] = 1, ['6'] = 1, ['7'] = 1, ['8'] = 1, ['9'] = 1,
366 ['H'] = 1, ['P'] = 1, ['T'] = 1,
367};
368
369
Willy Tarreaubaaee002006-06-26 02:48:02 +0200370#ifdef DEBUG_FULL
Willy Tarreau67f0eea2008-08-10 22:55:22 +0200371static char *cli_stnames[4] = { "DAT", "SHR", "SHW", "CLS" };
Willy Tarreaubaaee002006-06-26 02:48:02 +0200372#endif
373
Willy Tarreau4af6f3a2007-03-18 22:36:26 +0100374/*
375 * Adds a header and its CRLF at the tail of buffer <b>, just before the last
376 * CRLF. Text length is measured first, so it cannot be NULL.
377 * The header is also automatically added to the index <hdr_idx>, and the end
378 * of headers is automatically adjusted. The number of bytes added is returned
379 * on success, otherwise <0 is returned indicating an error.
380 */
381int http_header_add_tail(struct buffer *b, struct http_msg *msg,
382 struct hdr_idx *hdr_idx, const char *text)
383{
384 int bytes, len;
385
386 len = strlen(text);
387 bytes = buffer_insert_line2(b, b->data + msg->eoh, text, len);
388 if (!bytes)
389 return -1;
390 msg->eoh += bytes;
391 return hdr_idx_add(len, 1, hdr_idx, hdr_idx->tail);
392}
393
394/*
395 * Adds a header and its CRLF at the tail of buffer <b>, just before the last
396 * CRLF. <len> bytes are copied, not counting the CRLF. If <text> is NULL, then
397 * the buffer is only opened and the space reserved, but nothing is copied.
398 * The header is also automatically added to the index <hdr_idx>, and the end
399 * of headers is automatically adjusted. The number of bytes added is returned
400 * on success, otherwise <0 is returned indicating an error.
401 */
402int http_header_add_tail2(struct buffer *b, struct http_msg *msg,
403 struct hdr_idx *hdr_idx, const char *text, int len)
404{
405 int bytes;
406
407 bytes = buffer_insert_line2(b, b->data + msg->eoh, text, len);
408 if (!bytes)
409 return -1;
410 msg->eoh += bytes;
411 return hdr_idx_add(len, 1, hdr_idx, hdr_idx->tail);
412}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200413
414/*
Willy Tarreauaa9dce32007-03-18 23:50:16 +0100415 * Checks if <hdr> is exactly <name> for <len> chars, and ends with a colon.
416 * If so, returns the position of the first non-space character relative to
417 * <hdr>, or <end>-<hdr> if not found before. If no value is found, it tries
418 * to return a pointer to the place after the first space. Returns 0 if the
419 * header name does not match. Checks are case-insensitive.
420 */
421int http_header_match2(const char *hdr, const char *end,
422 const char *name, int len)
423{
424 const char *val;
425
426 if (hdr + len >= end)
427 return 0;
428 if (hdr[len] != ':')
429 return 0;
430 if (strncasecmp(hdr, name, len) != 0)
431 return 0;
432 val = hdr + len + 1;
433 while (val < end && HTTP_IS_SPHT(*val))
434 val++;
435 if ((val >= end) && (len + 2 <= end - hdr))
436 return len + 2; /* we may replace starting from second space */
437 return val - hdr;
438}
439
Willy Tarreau33a7e692007-06-10 19:45:56 +0200440/* Find the end of the header value contained between <s> and <e>.
441 * See RFC2616, par 2.2 for more information. Note that it requires
442 * a valid header to return a valid result.
443 */
444const char *find_hdr_value_end(const char *s, const char *e)
445{
446 int quoted, qdpair;
447
448 quoted = qdpair = 0;
449 for (; s < e; s++) {
450 if (qdpair) qdpair = 0;
451 else if (quoted && *s == '\\') qdpair = 1;
452 else if (quoted && *s == '"') quoted = 0;
453 else if (*s == '"') quoted = 1;
454 else if (*s == ',') return s;
455 }
456 return s;
457}
458
459/* Find the first or next occurrence of header <name> in message buffer <sol>
460 * using headers index <idx>, and return it in the <ctx> structure. This
461 * structure holds everything necessary to use the header and find next
462 * occurrence. If its <idx> member is 0, the header is searched from the
463 * beginning. Otherwise, the next occurrence is returned. The function returns
464 * 1 when it finds a value, and 0 when there is no more.
465 */
466int http_find_header2(const char *name, int len,
467 const char *sol, struct hdr_idx *idx,
468 struct hdr_ctx *ctx)
469{
Willy Tarreau33a7e692007-06-10 19:45:56 +0200470 const char *eol, *sov;
471 int cur_idx;
472
473 if (ctx->idx) {
474 /* We have previously returned a value, let's search
475 * another one on the same line.
476 */
477 cur_idx = ctx->idx;
478 sol = ctx->line;
479 sov = sol + ctx->val + ctx->vlen;
480 eol = sol + idx->v[cur_idx].len;
481
482 if (sov >= eol)
483 /* no more values in this header */
484 goto next_hdr;
485
486 /* values remaining for this header, skip the comma */
487 sov++;
488 while (sov < eol && http_is_lws[(unsigned char)*sov])
489 sov++;
490
491 goto return_hdr;
492 }
493
494 /* first request for this header */
495 sol += hdr_idx_first_pos(idx);
496 cur_idx = hdr_idx_first_idx(idx);
497
498 while (cur_idx) {
499 eol = sol + idx->v[cur_idx].len;
500
Willy Tarreau1ad7c6d2007-06-10 21:42:55 +0200501 if (len == 0) {
502 /* No argument was passed, we want any header.
503 * To achieve this, we simply build a fake request. */
504 while (sol + len < eol && sol[len] != ':')
505 len++;
506 name = sol;
507 }
508
Willy Tarreau33a7e692007-06-10 19:45:56 +0200509 if ((len < eol - sol) &&
510 (sol[len] == ':') &&
511 (strncasecmp(sol, name, len) == 0)) {
512
513 sov = sol + len + 1;
514 while (sov < eol && http_is_lws[(unsigned char)*sov])
515 sov++;
516 return_hdr:
517 ctx->line = sol;
518 ctx->idx = cur_idx;
519 ctx->val = sov - sol;
520
521 eol = find_hdr_value_end(sov, eol);
522 ctx->vlen = eol - sov;
523 return 1;
524 }
525 next_hdr:
526 sol = eol + idx->v[cur_idx].cr + 1;
527 cur_idx = idx->v[cur_idx].next;
528 }
529 return 0;
530}
531
532int http_find_header(const char *name,
533 const char *sol, struct hdr_idx *idx,
534 struct hdr_ctx *ctx)
535{
536 return http_find_header2(name, strlen(name), sol, idx, ctx);
537}
538
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100539/* This function handles a server error at the stream interface level. The
540 * stream interface is assumed to be already in a closed state. An optional
541 * message is copied into the input buffer, and an HTTP status code stored.
542 * The error flags are set to the values in arguments. Any pending request
Willy Tarreau6f0aa472009-03-08 20:33:29 +0100543 * in this buffer will be lost.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200544 */
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100545static void http_server_error(struct session *t, struct stream_interface *si,
546 int err, int finst, int status, const struct chunk *msg)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200547{
Willy Tarreau6f0aa472009-03-08 20:33:29 +0100548 buffer_erase(si->ob);
549 buffer_erase(si->ib);
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100550 buffer_write_ena(si->ib);
Willy Tarreau0f772532006-12-23 20:51:41 +0100551 if (status > 0 && msg) {
Willy Tarreau3bac9ff2007-03-18 17:31:28 +0100552 t->txn.status = status;
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100553 buffer_write(si->ib, msg->str, msg->len);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200554 }
555 if (!(t->flags & SN_ERR_MASK))
556 t->flags |= err;
557 if (!(t->flags & SN_FINST_MASK))
558 t->flags |= finst;
559}
560
Willy Tarreau80587432006-12-24 17:47:20 +0100561/* This function returns the appropriate error location for the given session
562 * and message.
563 */
564
565struct chunk *error_message(struct session *s, int msgnum)
566{
Willy Tarreaue2e27a52007-04-01 00:01:37 +0200567 if (s->be->errmsg[msgnum].str)
568 return &s->be->errmsg[msgnum];
Willy Tarreau80587432006-12-24 17:47:20 +0100569 else if (s->fe->errmsg[msgnum].str)
570 return &s->fe->errmsg[msgnum];
571 else
572 return &http_err_chunks[msgnum];
573}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200574
Willy Tarreau53b6c742006-12-17 13:37:46 +0100575/*
576 * returns HTTP_METH_NONE if there is nothing valid to read (empty or non-text
577 * string), HTTP_METH_OTHER for unknown methods, or the identified method.
578 */
579static http_meth_t find_http_meth(const char *str, const int len)
580{
581 unsigned char m;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100582 const struct http_method_desc *h;
Willy Tarreau53b6c742006-12-17 13:37:46 +0100583
584 m = ((unsigned)*str - 'A');
585
586 if (m < 26) {
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100587 for (h = http_methods[m]; h->len > 0; h++) {
588 if (unlikely(h->len != len))
Willy Tarreau53b6c742006-12-17 13:37:46 +0100589 continue;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100590 if (likely(memcmp(str, h->text, h->len) == 0))
Willy Tarreau53b6c742006-12-17 13:37:46 +0100591 return h->meth;
Willy Tarreau53b6c742006-12-17 13:37:46 +0100592 };
593 return HTTP_METH_OTHER;
594 }
595 return HTTP_METH_NONE;
596
597}
598
Willy Tarreau21d2af32008-02-14 20:25:24 +0100599/* Parse the URI from the given transaction (which is assumed to be in request
600 * phase) and look for the "/" beginning the PATH. If not found, return NULL.
601 * It is returned otherwise.
602 */
603static char *
604http_get_path(struct http_txn *txn)
605{
606 char *ptr, *end;
607
608 ptr = txn->req.sol + txn->req.sl.rq.u;
609 end = ptr + txn->req.sl.rq.u_l;
610
611 if (ptr >= end)
612 return NULL;
613
614 /* RFC2616, par. 5.1.2 :
615 * Request-URI = "*" | absuri | abspath | authority
616 */
617
618 if (*ptr == '*')
619 return NULL;
620
621 if (isalpha((unsigned char)*ptr)) {
622 /* this is a scheme as described by RFC3986, par. 3.1 */
623 ptr++;
624 while (ptr < end &&
625 (isalnum((unsigned char)*ptr) || *ptr == '+' || *ptr == '-' || *ptr == '.'))
626 ptr++;
627 /* skip '://' */
628 if (ptr == end || *ptr++ != ':')
629 return NULL;
630 if (ptr == end || *ptr++ != '/')
631 return NULL;
632 if (ptr == end || *ptr++ != '/')
633 return NULL;
634 }
635 /* skip [user[:passwd]@]host[:[port]] */
636
637 while (ptr < end && *ptr != '/')
638 ptr++;
639
640 if (ptr == end)
641 return NULL;
642
643 /* OK, we got the '/' ! */
644 return ptr;
645}
646
Willy Tarreauefb453c2008-10-26 20:49:47 +0100647/* Returns a 302 for a redirectable request. This may only be called just after
648 * the stream interface has moved to SI_ST_ASS. Unprocessable requests are
649 * left unchanged and will follow normal proxy processing.
650 */
Willy Tarreau55a8d0e2008-11-30 18:47:21 +0100651void perform_http_redirect(struct session *s, struct stream_interface *si)
Willy Tarreauefb453c2008-10-26 20:49:47 +0100652{
653 struct http_txn *txn;
654 struct chunk rdr;
655 char *path;
656 int len;
657
658 /* 1: create the response header */
659 rdr.len = strlen(HTTP_302);
660 rdr.str = trash;
661 memcpy(rdr.str, HTTP_302, rdr.len);
662
663 /* 2: add the server's prefix */
664 if (rdr.len + s->srv->rdr_len > sizeof(trash))
665 return;
666
667 memcpy(rdr.str + rdr.len, s->srv->rdr_pfx, s->srv->rdr_len);
668 rdr.len += s->srv->rdr_len;
669
670 /* 3: add the request URI */
671 txn = &s->txn;
672 path = http_get_path(txn);
673 if (!path)
674 return;
675
676 len = txn->req.sl.rq.u_l + (txn->req.sol+txn->req.sl.rq.u) - path;
677 if (rdr.len + len > sizeof(trash) - 4) /* 4 for CRLF-CRLF */
678 return;
679
680 memcpy(rdr.str + rdr.len, path, len);
681 rdr.len += len;
682 memcpy(rdr.str + rdr.len, "\r\n\r\n", 4);
683 rdr.len += 4;
684
685 /* prepare to return without error. */
Willy Tarreau99126c32008-11-27 10:30:51 +0100686 si->shutr(si);
687 si->shutw(si);
Willy Tarreauefb453c2008-10-26 20:49:47 +0100688 si->err_type = SI_ET_NONE;
689 si->err_loc = NULL;
690 si->state = SI_ST_CLO;
691
692 /* send the message */
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100693 http_server_error(s, si, SN_ERR_PRXCOND, SN_FINST_C, 302, &rdr);
Willy Tarreauefb453c2008-10-26 20:49:47 +0100694
695 /* FIXME: we should increase a counter of redirects per server and per backend. */
696 if (s->srv)
Willy Tarreau7f062c42009-03-05 18:43:00 +0100697 srv_inc_sess_ctr(s->srv);
Willy Tarreauefb453c2008-10-26 20:49:47 +0100698}
699
Willy Tarreau0cac36f2008-11-30 20:44:17 +0100700/* Return the error message corresponding to si->err_type. It is assumed
Willy Tarreauefb453c2008-10-26 20:49:47 +0100701 * that the server side is closed. Note that err_type is actually a
702 * bitmask, where almost only aborts may be cumulated with other
703 * values. We consider that aborted operations are more important
704 * than timeouts or errors due to the fact that nobody else in the
705 * logs might explain incomplete retries. All others should avoid
706 * being cumulated. It should normally not be possible to have multiple
707 * aborts at once, but just in case, the first one in sequence is reported.
708 */
Willy Tarreau0cac36f2008-11-30 20:44:17 +0100709void http_return_srv_error(struct session *s, struct stream_interface *si)
Willy Tarreauefb453c2008-10-26 20:49:47 +0100710{
Willy Tarreau0cac36f2008-11-30 20:44:17 +0100711 int err_type = si->err_type;
Willy Tarreauefb453c2008-10-26 20:49:47 +0100712
713 if (err_type & SI_ET_QUEUE_ABRT)
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100714 http_server_error(s, si, SN_ERR_CLICL, SN_FINST_Q,
715 503, error_message(s, HTTP_ERR_503));
Willy Tarreauefb453c2008-10-26 20:49:47 +0100716 else if (err_type & SI_ET_CONN_ABRT)
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100717 http_server_error(s, si, SN_ERR_CLICL, SN_FINST_C,
718 503, error_message(s, HTTP_ERR_503));
Willy Tarreauefb453c2008-10-26 20:49:47 +0100719 else if (err_type & SI_ET_QUEUE_TO)
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100720 http_server_error(s, si, SN_ERR_SRVTO, SN_FINST_Q,
721 503, error_message(s, HTTP_ERR_503));
Willy Tarreauefb453c2008-10-26 20:49:47 +0100722 else if (err_type & SI_ET_QUEUE_ERR)
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100723 http_server_error(s, si, SN_ERR_SRVCL, SN_FINST_Q,
724 503, error_message(s, HTTP_ERR_503));
Willy Tarreauefb453c2008-10-26 20:49:47 +0100725 else if (err_type & SI_ET_CONN_TO)
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100726 http_server_error(s, si, SN_ERR_SRVTO, SN_FINST_C,
727 503, error_message(s, HTTP_ERR_503));
Willy Tarreauefb453c2008-10-26 20:49:47 +0100728 else if (err_type & SI_ET_CONN_ERR)
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100729 http_server_error(s, si, SN_ERR_SRVCL, SN_FINST_C,
730 503, error_message(s, HTTP_ERR_503));
Willy Tarreauefb453c2008-10-26 20:49:47 +0100731 else /* SI_ET_CONN_OTHER and others */
Willy Tarreau2d3d94c2008-11-30 20:20:08 +0100732 http_server_error(s, si, SN_ERR_INTERNAL, SN_FINST_C,
733 500, error_message(s, HTTP_ERR_500));
Willy Tarreauefb453c2008-10-26 20:49:47 +0100734}
735
Willy Tarreau42250582007-04-01 01:30:43 +0200736extern const char sess_term_cond[8];
737extern const char sess_fin_state[8];
738extern const char *monthname[12];
739const char sess_cookie[4] = "NIDV"; /* No cookie, Invalid cookie, cookie for a Down server, Valid cookie */
740const char sess_set_cookie[8] = "N1I3PD5R"; /* No set-cookie, unknown, Set-Cookie Inserted, unknown,
741 Set-cookie seen and left unchanged (passive), Set-cookie Deleted,
742 unknown, Set-cookie Rewritten */
Willy Tarreau332f8bf2007-05-13 21:36:56 +0200743struct pool_head *pool2_requri;
Willy Tarreau086b3b42007-05-13 21:45:51 +0200744struct pool_head *pool2_capture;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100745
Willy Tarreau42250582007-04-01 01:30:43 +0200746/*
747 * send a log for the session when we have enough info about it.
748 * Will not log if the frontend has no log defined.
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100749 */
Willy Tarreau55a8d0e2008-11-30 18:47:21 +0100750void http_sess_log(struct session *s)
Willy Tarreau42250582007-04-01 01:30:43 +0200751{
752 char pn[INET6_ADDRSTRLEN + strlen(":65535")];
753 struct proxy *fe = s->fe;
754 struct proxy *be = s->be;
755 struct proxy *prx_log;
756 struct http_txn *txn = &s->txn;
757 int tolog;
758 char *uri, *h;
759 char *svid;
Willy Tarreaufe944602007-10-25 10:34:16 +0200760 struct tm tm;
Willy Tarreau42250582007-04-01 01:30:43 +0200761 static char tmpline[MAX_SYSLOG_LEN];
Willy Tarreau70089872008-06-13 21:12:51 +0200762 int t_request;
Willy Tarreau42250582007-04-01 01:30:43 +0200763 int hdr;
764
765 if (fe->logfac1 < 0 && fe->logfac2 < 0)
766 return;
767 prx_log = fe;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +0100768
Willy Tarreau42250582007-04-01 01:30:43 +0200769 if (s->cli_addr.ss_family == AF_INET)
770 inet_ntop(AF_INET,
771 (const void *)&((struct sockaddr_in *)&s->cli_addr)->sin_addr,
772 pn, sizeof(pn));
773 else
774 inet_ntop(AF_INET6,
775 (const void *)&((struct sockaddr_in6 *)(&s->cli_addr))->sin6_addr,
776 pn, sizeof(pn));
777
Willy Tarreaub7f694f2008-06-22 17:18:02 +0200778 get_localtime(s->logs.accept_date.tv_sec, &tm);
Willy Tarreau42250582007-04-01 01:30:43 +0200779
780 /* FIXME: let's limit ourselves to frontend logging for now. */
781 tolog = fe->to_log;
782
783 h = tmpline;
784 if (fe->to_log & LW_REQHDR &&
785 txn->req.cap &&
786 (h < tmpline + sizeof(tmpline) - 10)) {
787 *(h++) = ' ';
788 *(h++) = '{';
789 for (hdr = 0; hdr < fe->nb_req_cap; hdr++) {
790 if (hdr)
791 *(h++) = '|';
792 if (txn->req.cap[hdr] != NULL)
793 h = encode_string(h, tmpline + sizeof(tmpline) - 7,
794 '#', hdr_encode_map, txn->req.cap[hdr]);
795 }
796 *(h++) = '}';
797 }
798
799 if (fe->to_log & LW_RSPHDR &&
800 txn->rsp.cap &&
801 (h < tmpline + sizeof(tmpline) - 7)) {
802 *(h++) = ' ';
803 *(h++) = '{';
804 for (hdr = 0; hdr < fe->nb_rsp_cap; hdr++) {
805 if (hdr)
806 *(h++) = '|';
807 if (txn->rsp.cap[hdr] != NULL)
808 h = encode_string(h, tmpline + sizeof(tmpline) - 4,
809 '#', hdr_encode_map, txn->rsp.cap[hdr]);
810 }
811 *(h++) = '}';
812 }
813
814 if (h < tmpline + sizeof(tmpline) - 4) {
815 *(h++) = ' ';
816 *(h++) = '"';
817 uri = txn->uri ? txn->uri : "<BADREQ>";
818 h = encode_string(h, tmpline + sizeof(tmpline) - 1,
819 '#', url_encode_map, uri);
820 *(h++) = '"';
821 }
822 *h = '\0';
823
824 svid = (tolog & LW_SVID) ?
825 (s->data_source != DATA_SRC_STATS) ?
826 (s->srv != NULL) ? s->srv->id : "<NOSRV>" : "<STATS>" : "-";
827
Willy Tarreau70089872008-06-13 21:12:51 +0200828 t_request = -1;
829 if (tv_isge(&s->logs.tv_request, &s->logs.tv_accept))
830 t_request = tv_ms_elapsed(&s->logs.tv_accept, &s->logs.tv_request);
831
Willy Tarreau42250582007-04-01 01:30:43 +0200832 send_log(prx_log, LOG_INFO,
833 "%s:%d [%02d/%s/%04d:%02d:%02d:%02d.%03d]"
834 " %s %s/%s %d/%d/%d/%d/%s%d %d %s%lld"
Krzysztof Piotr Oledzki25b501a2008-01-06 16:36:16 +0100835 " %s %s %c%c%c%c %d/%d/%d/%d/%s%u %d/%d%s\n",
Willy Tarreau42250582007-04-01 01:30:43 +0200836 pn,
837 (s->cli_addr.ss_family == AF_INET) ?
838 ntohs(((struct sockaddr_in *)&s->cli_addr)->sin_port) :
839 ntohs(((struct sockaddr_in6 *)&s->cli_addr)->sin6_port),
Willy Tarreaufe944602007-10-25 10:34:16 +0200840 tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
Willy Tarreaub7f694f2008-06-22 17:18:02 +0200841 tm.tm_hour, tm.tm_min, tm.tm_sec, s->logs.accept_date.tv_usec/1000,
Willy Tarreau42250582007-04-01 01:30:43 +0200842 fe->id, be->id, svid,
Willy Tarreau70089872008-06-13 21:12:51 +0200843 t_request,
844 (s->logs.t_queue >= 0) ? s->logs.t_queue - t_request : -1,
Willy Tarreau42250582007-04-01 01:30:43 +0200845 (s->logs.t_connect >= 0) ? s->logs.t_connect - s->logs.t_queue : -1,
846 (s->logs.t_data >= 0) ? s->logs.t_data - s->logs.t_connect : -1,
847 (tolog & LW_BYTES) ? "" : "+", s->logs.t_close,
848 txn->status,
Willy Tarreau8b3977f2008-01-18 11:16:32 +0100849 (tolog & LW_BYTES) ? "" : "+", s->logs.bytes_out,
Willy Tarreau42250582007-04-01 01:30:43 +0200850 txn->cli_cookie ? txn->cli_cookie : "-",
851 txn->srv_cookie ? txn->srv_cookie : "-",
852 sess_term_cond[(s->flags & SN_ERR_MASK) >> SN_ERR_SHIFT],
853 sess_fin_state[(s->flags & SN_FINST_MASK) >> SN_FINST_SHIFT],
854 (be->options & PR_O_COOK_ANY) ? sess_cookie[(txn->flags & TX_CK_MASK) >> TX_CK_SHIFT] : '-',
855 (be->options & PR_O_COOK_ANY) ? sess_set_cookie[(txn->flags & TX_SCK_MASK) >> TX_SCK_SHIFT] : '-',
856 actconn, fe->feconn, be->beconn, s->srv ? s->srv->cur_sess : 0,
Krzysztof Piotr Oledzki25b501a2008-01-06 16:36:16 +0100857 (s->flags & SN_REDISP)?"+":"",
858 (s->conn_retries>0)?(be->conn_retries - s->conn_retries):be->conn_retries,
Willy Tarreau42250582007-04-01 01:30:43 +0200859 s->logs.srv_queue_size, s->logs.prx_queue_size, tmpline);
860
861 s->logs.logwait = 0;
862}
863
Willy Tarreau117f59e2007-03-04 18:17:17 +0100864
865/*
866 * Capture headers from message starting at <som> according to header list
867 * <cap_hdr>, and fill the <idx> structure appropriately.
868 */
869void capture_headers(char *som, struct hdr_idx *idx,
870 char **cap, struct cap_hdr *cap_hdr)
871{
872 char *eol, *sol, *col, *sov;
873 int cur_idx;
874 struct cap_hdr *h;
875 int len;
876
877 sol = som + hdr_idx_first_pos(idx);
878 cur_idx = hdr_idx_first_idx(idx);
879
880 while (cur_idx) {
881 eol = sol + idx->v[cur_idx].len;
882
883 col = sol;
884 while (col < eol && *col != ':')
885 col++;
886
887 sov = col + 1;
888 while (sov < eol && http_is_lws[(unsigned char)*sov])
889 sov++;
890
891 for (h = cap_hdr; h; h = h->next) {
892 if ((h->namelen == col - sol) &&
893 (strncasecmp(sol, h->name, h->namelen) == 0)) {
894 if (cap[h->index] == NULL)
895 cap[h->index] =
Willy Tarreaucf7f3202007-05-13 22:46:04 +0200896 pool_alloc2(h->pool);
Willy Tarreau117f59e2007-03-04 18:17:17 +0100897
898 if (cap[h->index] == NULL) {
899 Alert("HTTP capture : out of memory.\n");
900 continue;
901 }
902
903 len = eol - sov;
904 if (len > h->len)
905 len = h->len;
906
907 memcpy(cap[h->index], sov, len);
908 cap[h->index][len]=0;
909 }
910 }
911 sol = eol + idx->v[cur_idx].cr + 1;
912 cur_idx = idx->v[cur_idx].next;
913 }
914}
915
916
Willy Tarreau42250582007-04-01 01:30:43 +0200917/* either we find an LF at <ptr> or we jump to <bad>.
918 */
919#define EXPECT_LF_HERE(ptr, bad) do { if (unlikely(*(ptr) != '\n')) goto bad; } while (0)
920
921/* plays with variables <ptr>, <end> and <state>. Jumps to <good> if OK,
922 * otherwise to <http_msg_ood> with <state> set to <st>.
923 */
924#define EAT_AND_JUMP_OR_RETURN(good, st) do { \
925 ptr++; \
926 if (likely(ptr < end)) \
927 goto good; \
928 else { \
929 state = (st); \
930 goto http_msg_ood; \
931 } \
932 } while (0)
933
934
Willy Tarreaubaaee002006-06-26 02:48:02 +0200935/*
Willy Tarreaua15645d2007-03-18 16:22:39 +0100936 * This function parses a status line between <ptr> and <end>, starting with
Willy Tarreau8973c702007-01-21 23:58:29 +0100937 * parser state <state>. Only states HTTP_MSG_RPVER, HTTP_MSG_RPVER_SP,
938 * HTTP_MSG_RPCODE, HTTP_MSG_RPCODE_SP and HTTP_MSG_RPREASON are handled. Others
939 * will give undefined results.
940 * Note that it is upon the caller's responsibility to ensure that ptr < end,
941 * and that msg->sol points to the beginning of the response.
942 * If a complete line is found (which implies that at least one CR or LF is
943 * found before <end>, the updated <ptr> is returned, otherwise NULL is
944 * returned indicating an incomplete line (which does not mean that parts have
945 * not been updated). In the incomplete case, if <ret_ptr> or <ret_state> are
946 * non-NULL, they are fed with the new <ptr> and <state> values to be passed
947 * upon next call.
948 *
Willy Tarreau9cdde232007-05-02 20:58:19 +0200949 * This function was intentionally designed to be called from
Willy Tarreau8973c702007-01-21 23:58:29 +0100950 * http_msg_analyzer() with the lowest overhead. It should integrate perfectly
951 * within its state machine and use the same macros, hence the need for same
Willy Tarreau9cdde232007-05-02 20:58:19 +0200952 * labels and variable names. Note that msg->sol is left unchanged.
Willy Tarreau8973c702007-01-21 23:58:29 +0100953 */
Willy Tarreaue69eada2008-01-27 00:34:10 +0100954const char *http_parse_stsline(struct http_msg *msg, const char *msg_buf,
955 unsigned int state, const char *ptr, const char *end,
956 char **ret_ptr, unsigned int *ret_state)
Willy Tarreau8973c702007-01-21 23:58:29 +0100957{
Willy Tarreau8973c702007-01-21 23:58:29 +0100958 switch (state) {
959 http_msg_rpver:
960 case HTTP_MSG_RPVER:
Willy Tarreau4b89ad42007-03-04 18:13:58 +0100961 if (likely(HTTP_IS_VER_TOKEN(*ptr)))
Willy Tarreau8973c702007-01-21 23:58:29 +0100962 EAT_AND_JUMP_OR_RETURN(http_msg_rpver, HTTP_MSG_RPVER);
963
964 if (likely(HTTP_IS_SPHT(*ptr))) {
Willy Tarreaub326fcc2007-03-03 13:54:32 +0100965 msg->sl.st.v_l = (ptr - msg_buf) - msg->som;
Willy Tarreau8973c702007-01-21 23:58:29 +0100966 EAT_AND_JUMP_OR_RETURN(http_msg_rpver_sp, HTTP_MSG_RPVER_SP);
967 }
Willy Tarreau7552c032009-03-01 11:10:40 +0100968 state = HTTP_MSG_ERROR;
969 break;
970
Willy Tarreau8973c702007-01-21 23:58:29 +0100971 http_msg_rpver_sp:
972 case HTTP_MSG_RPVER_SP:
973 if (likely(!HTTP_IS_LWS(*ptr))) {
974 msg->sl.st.c = ptr - msg_buf;
975 goto http_msg_rpcode;
976 }
977 if (likely(HTTP_IS_SPHT(*ptr)))
978 EAT_AND_JUMP_OR_RETURN(http_msg_rpver_sp, HTTP_MSG_RPVER_SP);
979 /* so it's a CR/LF, this is invalid */
Willy Tarreau7552c032009-03-01 11:10:40 +0100980 state = HTTP_MSG_ERROR;
981 break;
Willy Tarreau8973c702007-01-21 23:58:29 +0100982
983 http_msg_rpcode:
984 case HTTP_MSG_RPCODE:
985 if (likely(!HTTP_IS_LWS(*ptr)))
986 EAT_AND_JUMP_OR_RETURN(http_msg_rpcode, HTTP_MSG_RPCODE);
987
988 if (likely(HTTP_IS_SPHT(*ptr))) {
989 msg->sl.st.c_l = (ptr - msg_buf) - msg->sl.st.c;
990 EAT_AND_JUMP_OR_RETURN(http_msg_rpcode_sp, HTTP_MSG_RPCODE_SP);
991 }
992
993 /* so it's a CR/LF, so there is no reason phrase */
994 msg->sl.st.c_l = (ptr - msg_buf) - msg->sl.st.c;
995 http_msg_rsp_reason:
996 /* FIXME: should we support HTTP responses without any reason phrase ? */
997 msg->sl.st.r = ptr - msg_buf;
998 msg->sl.st.r_l = 0;
999 goto http_msg_rpline_eol;
1000
1001 http_msg_rpcode_sp:
1002 case HTTP_MSG_RPCODE_SP:
1003 if (likely(!HTTP_IS_LWS(*ptr))) {
1004 msg->sl.st.r = ptr - msg_buf;
1005 goto http_msg_rpreason;
1006 }
1007 if (likely(HTTP_IS_SPHT(*ptr)))
1008 EAT_AND_JUMP_OR_RETURN(http_msg_rpcode_sp, HTTP_MSG_RPCODE_SP);
1009 /* so it's a CR/LF, so there is no reason phrase */
1010 goto http_msg_rsp_reason;
1011
1012 http_msg_rpreason:
1013 case HTTP_MSG_RPREASON:
1014 if (likely(!HTTP_IS_CRLF(*ptr)))
1015 EAT_AND_JUMP_OR_RETURN(http_msg_rpreason, HTTP_MSG_RPREASON);
1016 msg->sl.st.r_l = (ptr - msg_buf) - msg->sl.st.r;
1017 http_msg_rpline_eol:
1018 /* We have seen the end of line. Note that we do not
1019 * necessarily have the \n yet, but at least we know that we
1020 * have EITHER \r OR \n, otherwise the response would not be
1021 * complete. We can then record the response length and return
1022 * to the caller which will be able to register it.
1023 */
1024 msg->sl.st.l = ptr - msg->sol;
1025 return ptr;
1026
1027#ifdef DEBUG_FULL
1028 default:
1029 fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state);
1030 exit(1);
1031#endif
1032 }
1033
1034 http_msg_ood:
Willy Tarreau7552c032009-03-01 11:10:40 +01001035 /* out of valid data */
Willy Tarreau8973c702007-01-21 23:58:29 +01001036 if (ret_state)
1037 *ret_state = state;
1038 if (ret_ptr)
1039 *ret_ptr = (char *)ptr;
1040 return NULL;
Willy Tarreau8973c702007-01-21 23:58:29 +01001041}
1042
1043
1044/*
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001045 * This function parses a request line between <ptr> and <end>, starting with
1046 * parser state <state>. Only states HTTP_MSG_RQMETH, HTTP_MSG_RQMETH_SP,
1047 * HTTP_MSG_RQURI, HTTP_MSG_RQURI_SP and HTTP_MSG_RQVER are handled. Others
1048 * will give undefined results.
1049 * Note that it is upon the caller's responsibility to ensure that ptr < end,
1050 * and that msg->sol points to the beginning of the request.
1051 * If a complete line is found (which implies that at least one CR or LF is
1052 * found before <end>, the updated <ptr> is returned, otherwise NULL is
1053 * returned indicating an incomplete line (which does not mean that parts have
1054 * not been updated). In the incomplete case, if <ret_ptr> or <ret_state> are
1055 * non-NULL, they are fed with the new <ptr> and <state> values to be passed
1056 * upon next call.
1057 *
Willy Tarreau9cdde232007-05-02 20:58:19 +02001058 * This function was intentionally designed to be called from
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001059 * http_msg_analyzer() with the lowest overhead. It should integrate perfectly
1060 * within its state machine and use the same macros, hence the need for same
Willy Tarreau9cdde232007-05-02 20:58:19 +02001061 * labels and variable names. Note that msg->sol is left unchanged.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001062 */
Willy Tarreaue69eada2008-01-27 00:34:10 +01001063const char *http_parse_reqline(struct http_msg *msg, const char *msg_buf,
1064 unsigned int state, const char *ptr, const char *end,
1065 char **ret_ptr, unsigned int *ret_state)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001066{
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001067 switch (state) {
1068 http_msg_rqmeth:
1069 case HTTP_MSG_RQMETH:
1070 if (likely(HTTP_IS_TOKEN(*ptr)))
1071 EAT_AND_JUMP_OR_RETURN(http_msg_rqmeth, HTTP_MSG_RQMETH);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001072
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001073 if (likely(HTTP_IS_SPHT(*ptr))) {
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001074 msg->sl.rq.m_l = (ptr - msg_buf) - msg->som;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001075 EAT_AND_JUMP_OR_RETURN(http_msg_rqmeth_sp, HTTP_MSG_RQMETH_SP);
1076 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001077
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001078 if (likely(HTTP_IS_CRLF(*ptr))) {
1079 /* HTTP 0.9 request */
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001080 msg->sl.rq.m_l = (ptr - msg_buf) - msg->som;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001081 http_msg_req09_uri:
1082 msg->sl.rq.u = ptr - msg_buf;
1083 http_msg_req09_uri_e:
1084 msg->sl.rq.u_l = (ptr - msg_buf) - msg->sl.rq.u;
1085 http_msg_req09_ver:
1086 msg->sl.rq.v = ptr - msg_buf;
1087 msg->sl.rq.v_l = 0;
1088 goto http_msg_rqline_eol;
1089 }
Willy Tarreau7552c032009-03-01 11:10:40 +01001090 state = HTTP_MSG_ERROR;
1091 break;
1092
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001093 http_msg_rqmeth_sp:
1094 case HTTP_MSG_RQMETH_SP:
1095 if (likely(!HTTP_IS_LWS(*ptr))) {
1096 msg->sl.rq.u = ptr - msg_buf;
1097 goto http_msg_rquri;
1098 }
1099 if (likely(HTTP_IS_SPHT(*ptr)))
1100 EAT_AND_JUMP_OR_RETURN(http_msg_rqmeth_sp, HTTP_MSG_RQMETH_SP);
1101 /* so it's a CR/LF, meaning an HTTP 0.9 request */
1102 goto http_msg_req09_uri;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001103
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001104 http_msg_rquri:
1105 case HTTP_MSG_RQURI:
1106 if (likely(!HTTP_IS_LWS(*ptr)))
1107 EAT_AND_JUMP_OR_RETURN(http_msg_rquri, HTTP_MSG_RQURI);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001108
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001109 if (likely(HTTP_IS_SPHT(*ptr))) {
1110 msg->sl.rq.u_l = (ptr - msg_buf) - msg->sl.rq.u;
1111 EAT_AND_JUMP_OR_RETURN(http_msg_rquri_sp, HTTP_MSG_RQURI_SP);
1112 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001113
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001114 /* so it's a CR/LF, meaning an HTTP 0.9 request */
1115 goto http_msg_req09_uri_e;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001116
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001117 http_msg_rquri_sp:
1118 case HTTP_MSG_RQURI_SP:
1119 if (likely(!HTTP_IS_LWS(*ptr))) {
1120 msg->sl.rq.v = ptr - msg_buf;
1121 goto http_msg_rqver;
1122 }
1123 if (likely(HTTP_IS_SPHT(*ptr)))
1124 EAT_AND_JUMP_OR_RETURN(http_msg_rquri_sp, HTTP_MSG_RQURI_SP);
1125 /* so it's a CR/LF, meaning an HTTP 0.9 request */
1126 goto http_msg_req09_ver;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001127
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001128 http_msg_rqver:
1129 case HTTP_MSG_RQVER:
Willy Tarreau4b89ad42007-03-04 18:13:58 +01001130 if (likely(HTTP_IS_VER_TOKEN(*ptr)))
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001131 EAT_AND_JUMP_OR_RETURN(http_msg_rqver, HTTP_MSG_RQVER);
Willy Tarreau4b89ad42007-03-04 18:13:58 +01001132
1133 if (likely(HTTP_IS_CRLF(*ptr))) {
1134 msg->sl.rq.v_l = (ptr - msg_buf) - msg->sl.rq.v;
1135 http_msg_rqline_eol:
1136 /* We have seen the end of line. Note that we do not
1137 * necessarily have the \n yet, but at least we know that we
1138 * have EITHER \r OR \n, otherwise the request would not be
1139 * complete. We can then record the request length and return
1140 * to the caller which will be able to register it.
1141 */
1142 msg->sl.rq.l = ptr - msg->sol;
1143 return ptr;
1144 }
1145
1146 /* neither an HTTP_VER token nor a CRLF */
Willy Tarreau7552c032009-03-01 11:10:40 +01001147 state = HTTP_MSG_ERROR;
1148 break;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001149
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001150#ifdef DEBUG_FULL
1151 default:
1152 fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state);
1153 exit(1);
1154#endif
1155 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001156
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001157 http_msg_ood:
Willy Tarreau7552c032009-03-01 11:10:40 +01001158 /* out of valid data */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001159 if (ret_state)
1160 *ret_state = state;
1161 if (ret_ptr)
1162 *ret_ptr = (char *)ptr;
1163 return NULL;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001164}
Willy Tarreau58f10d72006-12-04 02:26:12 +01001165
1166
Willy Tarreau8973c702007-01-21 23:58:29 +01001167/*
1168 * This function parses an HTTP message, either a request or a response,
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001169 * depending on the initial msg->msg_state. It can be preempted everywhere
Willy Tarreau8973c702007-01-21 23:58:29 +01001170 * when data are missing and recalled at the exact same location with no
1171 * information loss. The header index is re-initialized when switching from
Willy Tarreau9cdde232007-05-02 20:58:19 +02001172 * MSG_R[PQ]BEFORE to MSG_RPVER|MSG_RQMETH. It modifies msg->sol among other
1173 * fields.
Willy Tarreau8973c702007-01-21 23:58:29 +01001174 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001175void http_msg_analyzer(struct buffer *buf, struct http_msg *msg, struct hdr_idx *idx)
1176{
Willy Tarreaue69eada2008-01-27 00:34:10 +01001177 unsigned int state; /* updated only when leaving the FSM */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001178 register char *ptr, *end; /* request pointers, to avoid dereferences */
Willy Tarreau58f10d72006-12-04 02:26:12 +01001179
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001180 state = msg->msg_state;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001181 ptr = buf->lr;
1182 end = buf->r;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001183
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001184 if (unlikely(ptr >= end))
1185 goto http_msg_ood;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001186
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001187 switch (state) {
Willy Tarreau8973c702007-01-21 23:58:29 +01001188 /*
1189 * First, states that are specific to the response only.
1190 * We check them first so that request and headers are
1191 * closer to each other (accessed more often).
1192 */
1193 http_msg_rpbefore:
1194 case HTTP_MSG_RPBEFORE:
1195 if (likely(HTTP_IS_TOKEN(*ptr))) {
1196 if (likely(ptr == buf->data)) {
1197 msg->sol = ptr;
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001198 msg->som = 0;
Willy Tarreau8973c702007-01-21 23:58:29 +01001199 } else {
1200#if PARSE_PRESERVE_EMPTY_LINES
1201 /* only skip empty leading lines, don't remove them */
1202 msg->sol = ptr;
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001203 msg->som = ptr - buf->data;
Willy Tarreau8973c702007-01-21 23:58:29 +01001204#else
1205 /* Remove empty leading lines, as recommended by
1206 * RFC2616. This takes a lot of time because we
1207 * must move all the buffer backwards, but this
1208 * is rarely needed. The method above will be
1209 * cleaner when we'll be able to start sending
1210 * the request from any place in the buffer.
1211 */
1212 buf->lr = ptr;
1213 buffer_replace2(buf, buf->data, buf->lr, NULL, 0);
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001214 msg->som = 0;
Willy Tarreau8973c702007-01-21 23:58:29 +01001215 msg->sol = buf->data;
1216 ptr = buf->data;
1217 end = buf->r;
1218#endif
1219 }
1220 hdr_idx_init(idx);
1221 state = HTTP_MSG_RPVER;
1222 goto http_msg_rpver;
1223 }
1224
1225 if (unlikely(!HTTP_IS_CRLF(*ptr)))
1226 goto http_msg_invalid;
1227
1228 if (unlikely(*ptr == '\n'))
1229 EAT_AND_JUMP_OR_RETURN(http_msg_rpbefore, HTTP_MSG_RPBEFORE);
1230 EAT_AND_JUMP_OR_RETURN(http_msg_rpbefore_cr, HTTP_MSG_RPBEFORE_CR);
1231 /* stop here */
1232
1233 http_msg_rpbefore_cr:
1234 case HTTP_MSG_RPBEFORE_CR:
1235 EXPECT_LF_HERE(ptr, http_msg_invalid);
1236 EAT_AND_JUMP_OR_RETURN(http_msg_rpbefore, HTTP_MSG_RPBEFORE);
1237 /* stop here */
1238
1239 http_msg_rpver:
1240 case HTTP_MSG_RPVER:
1241 case HTTP_MSG_RPVER_SP:
1242 case HTTP_MSG_RPCODE:
1243 case HTTP_MSG_RPCODE_SP:
1244 case HTTP_MSG_RPREASON:
Willy Tarreaua15645d2007-03-18 16:22:39 +01001245 ptr = (char *)http_parse_stsline(msg, buf->data, state, ptr, end,
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001246 &buf->lr, &msg->msg_state);
Willy Tarreau8973c702007-01-21 23:58:29 +01001247 if (unlikely(!ptr))
1248 return;
1249
1250 /* we have a full response and we know that we have either a CR
1251 * or an LF at <ptr>.
1252 */
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001253 //fprintf(stderr,"som=%d rq.l=%d *ptr=0x%02x\n", msg->som, msg->sl.st.l, *ptr);
Willy Tarreau8973c702007-01-21 23:58:29 +01001254 hdr_idx_set_start(idx, msg->sl.st.l, *ptr == '\r');
1255
1256 msg->sol = ptr;
1257 if (likely(*ptr == '\r'))
1258 EAT_AND_JUMP_OR_RETURN(http_msg_rpline_end, HTTP_MSG_RPLINE_END);
1259 goto http_msg_rpline_end;
1260
1261 http_msg_rpline_end:
1262 case HTTP_MSG_RPLINE_END:
1263 /* msg->sol must point to the first of CR or LF. */
1264 EXPECT_LF_HERE(ptr, http_msg_invalid);
1265 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_first, HTTP_MSG_HDR_FIRST);
1266 /* stop here */
1267
1268 /*
1269 * Second, states that are specific to the request only
1270 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001271 http_msg_rqbefore:
1272 case HTTP_MSG_RQBEFORE:
1273 if (likely(HTTP_IS_TOKEN(*ptr))) {
1274 if (likely(ptr == buf->data)) {
1275 msg->sol = ptr;
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001276 msg->som = 0;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001277 } else {
1278#if PARSE_PRESERVE_EMPTY_LINES
1279 /* only skip empty leading lines, don't remove them */
1280 msg->sol = ptr;
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001281 msg->som = ptr - buf->data;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001282#else
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001283 /* Remove empty leading lines, as recommended by
1284 * RFC2616. This takes a lot of time because we
1285 * must move all the buffer backwards, but this
1286 * is rarely needed. The method above will be
1287 * cleaner when we'll be able to start sending
1288 * the request from any place in the buffer.
Willy Tarreau58f10d72006-12-04 02:26:12 +01001289 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001290 buf->lr = ptr;
1291 buffer_replace2(buf, buf->data, buf->lr, NULL, 0);
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001292 msg->som = 0;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001293 msg->sol = buf->data;
1294 ptr = buf->data;
1295 end = buf->r;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001296#endif
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001297 }
Willy Tarreauf0d058e2007-01-25 12:03:42 +01001298 /* we will need this when keep-alive will be supported
1299 hdr_idx_init(idx);
1300 */
Willy Tarreau8973c702007-01-21 23:58:29 +01001301 state = HTTP_MSG_RQMETH;
1302 goto http_msg_rqmeth;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001303 }
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001304
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001305 if (unlikely(!HTTP_IS_CRLF(*ptr)))
1306 goto http_msg_invalid;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001307
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001308 if (unlikely(*ptr == '\n'))
1309 EAT_AND_JUMP_OR_RETURN(http_msg_rqbefore, HTTP_MSG_RQBEFORE);
1310 EAT_AND_JUMP_OR_RETURN(http_msg_rqbefore_cr, HTTP_MSG_RQBEFORE_CR);
Willy Tarreau8973c702007-01-21 23:58:29 +01001311 /* stop here */
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001312
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001313 http_msg_rqbefore_cr:
1314 case HTTP_MSG_RQBEFORE_CR:
1315 EXPECT_LF_HERE(ptr, http_msg_invalid);
1316 EAT_AND_JUMP_OR_RETURN(http_msg_rqbefore, HTTP_MSG_RQBEFORE);
Willy Tarreau8973c702007-01-21 23:58:29 +01001317 /* stop here */
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001318
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001319 http_msg_rqmeth:
1320 case HTTP_MSG_RQMETH:
1321 case HTTP_MSG_RQMETH_SP:
1322 case HTTP_MSG_RQURI:
1323 case HTTP_MSG_RQURI_SP:
1324 case HTTP_MSG_RQVER:
1325 ptr = (char *)http_parse_reqline(msg, buf->data, state, ptr, end,
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001326 &buf->lr, &msg->msg_state);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001327 if (unlikely(!ptr))
1328 return;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001329
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001330 /* we have a full request and we know that we have either a CR
1331 * or an LF at <ptr>.
1332 */
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001333 //fprintf(stderr,"som=%d rq.l=%d *ptr=0x%02x\n", msg->som, msg->sl.rq.l, *ptr);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001334 hdr_idx_set_start(idx, msg->sl.rq.l, *ptr == '\r');
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001335
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001336 msg->sol = ptr;
1337 if (likely(*ptr == '\r'))
1338 EAT_AND_JUMP_OR_RETURN(http_msg_rqline_end, HTTP_MSG_RQLINE_END);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001339 goto http_msg_rqline_end;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001340
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001341 http_msg_rqline_end:
1342 case HTTP_MSG_RQLINE_END:
1343 /* check for HTTP/0.9 request : no version information available.
1344 * msg->sol must point to the first of CR or LF.
1345 */
1346 if (unlikely(msg->sl.rq.v_l == 0))
1347 goto http_msg_last_lf;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001348
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001349 EXPECT_LF_HERE(ptr, http_msg_invalid);
1350 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_first, HTTP_MSG_HDR_FIRST);
Willy Tarreau8973c702007-01-21 23:58:29 +01001351 /* stop here */
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001352
Willy Tarreau8973c702007-01-21 23:58:29 +01001353 /*
1354 * Common states below
1355 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001356 http_msg_hdr_first:
1357 case HTTP_MSG_HDR_FIRST:
1358 msg->sol = ptr;
1359 if (likely(!HTTP_IS_CRLF(*ptr))) {
1360 goto http_msg_hdr_name;
1361 }
1362
1363 if (likely(*ptr == '\r'))
1364 EAT_AND_JUMP_OR_RETURN(http_msg_last_lf, HTTP_MSG_LAST_LF);
1365 goto http_msg_last_lf;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001366
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001367 http_msg_hdr_name:
1368 case HTTP_MSG_HDR_NAME:
1369 /* assumes msg->sol points to the first char */
1370 if (likely(HTTP_IS_TOKEN(*ptr)))
1371 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_name, HTTP_MSG_HDR_NAME);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001372
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001373 if (likely(*ptr == ':')) {
1374 msg->col = ptr - buf->data;
1375 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_sp, HTTP_MSG_HDR_L1_SP);
1376 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001377
Willy Tarreau32a4ec02009-04-02 11:35:18 +02001378 if (likely(msg->err_pos < -1) || *ptr == '\n')
1379 goto http_msg_invalid;
1380
1381 if (msg->err_pos == -1) /* capture error pointer */
1382 msg->err_pos = ptr - buf->data; /* >= 0 now */
1383
1384 /* and we still accept this non-token character */
1385 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_name, HTTP_MSG_HDR_NAME);
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001386
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001387 http_msg_hdr_l1_sp:
1388 case HTTP_MSG_HDR_L1_SP:
1389 /* assumes msg->sol points to the first char and msg->col to the colon */
1390 if (likely(HTTP_IS_SPHT(*ptr)))
1391 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_sp, HTTP_MSG_HDR_L1_SP);
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001392
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001393 /* header value can be basically anything except CR/LF */
1394 msg->sov = ptr - buf->data;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001395
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001396 if (likely(!HTTP_IS_CRLF(*ptr))) {
1397 goto http_msg_hdr_val;
1398 }
1399
1400 if (likely(*ptr == '\r'))
1401 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_lf, HTTP_MSG_HDR_L1_LF);
1402 goto http_msg_hdr_l1_lf;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001403
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001404 http_msg_hdr_l1_lf:
1405 case HTTP_MSG_HDR_L1_LF:
1406 EXPECT_LF_HERE(ptr, http_msg_invalid);
1407 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_lws, HTTP_MSG_HDR_L1_LWS);
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001408
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001409 http_msg_hdr_l1_lws:
1410 case HTTP_MSG_HDR_L1_LWS:
1411 if (likely(HTTP_IS_SPHT(*ptr))) {
1412 /* replace HT,CR,LF with spaces */
1413 for (; buf->data+msg->sov < ptr; msg->sov++)
1414 buf->data[msg->sov] = ' ';
1415 goto http_msg_hdr_l1_sp;
1416 }
Willy Tarreauaa9dce32007-03-18 23:50:16 +01001417 /* we had a header consisting only in spaces ! */
1418 msg->eol = buf->data + msg->sov;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001419 goto http_msg_complete_header;
1420
1421 http_msg_hdr_val:
1422 case HTTP_MSG_HDR_VAL:
1423 /* assumes msg->sol points to the first char, msg->col to the
1424 * colon, and msg->sov points to the first character of the
1425 * value.
1426 */
1427 if (likely(!HTTP_IS_CRLF(*ptr)))
1428 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_val, HTTP_MSG_HDR_VAL);
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001429
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001430 msg->eol = ptr;
1431 /* Note: we could also copy eol into ->eoh so that we have the
1432 * real header end in case it ends with lots of LWS, but is this
1433 * really needed ?
1434 */
1435 if (likely(*ptr == '\r'))
1436 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l2_lf, HTTP_MSG_HDR_L2_LF);
1437 goto http_msg_hdr_l2_lf;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001438
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001439 http_msg_hdr_l2_lf:
1440 case HTTP_MSG_HDR_L2_LF:
1441 EXPECT_LF_HERE(ptr, http_msg_invalid);
1442 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l2_lws, HTTP_MSG_HDR_L2_LWS);
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001443
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001444 http_msg_hdr_l2_lws:
1445 case HTTP_MSG_HDR_L2_LWS:
1446 if (unlikely(HTTP_IS_SPHT(*ptr))) {
1447 /* LWS: replace HT,CR,LF with spaces */
1448 for (; msg->eol < ptr; msg->eol++)
1449 *msg->eol = ' ';
1450 goto http_msg_hdr_val;
1451 }
1452 http_msg_complete_header:
1453 /*
1454 * It was a new header, so the last one is finished.
1455 * Assumes msg->sol points to the first char, msg->col to the
1456 * colon, msg->sov points to the first character of the value
1457 * and msg->eol to the first CR or LF so we know how the line
1458 * ends. We insert last header into the index.
1459 */
1460 /*
1461 fprintf(stderr,"registering %-2d bytes : ", msg->eol - msg->sol);
1462 write(2, msg->sol, msg->eol-msg->sol);
1463 fprintf(stderr,"\n");
1464 */
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001465
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001466 if (unlikely(hdr_idx_add(msg->eol - msg->sol, *msg->eol == '\r',
1467 idx, idx->tail) < 0))
1468 goto http_msg_invalid;
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001469
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001470 msg->sol = ptr;
1471 if (likely(!HTTP_IS_CRLF(*ptr))) {
1472 goto http_msg_hdr_name;
1473 }
1474
1475 if (likely(*ptr == '\r'))
1476 EAT_AND_JUMP_OR_RETURN(http_msg_last_lf, HTTP_MSG_LAST_LF);
1477 goto http_msg_last_lf;
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001478
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001479 http_msg_last_lf:
1480 case HTTP_MSG_LAST_LF:
1481 /* Assumes msg->sol points to the first of either CR or LF */
1482 EXPECT_LF_HERE(ptr, http_msg_invalid);
1483 ptr++;
1484 buf->lr = ptr;
1485 msg->eoh = msg->sol - buf->data;
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001486 msg->msg_state = HTTP_MSG_BODY;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001487 return;
1488#ifdef DEBUG_FULL
1489 default:
1490 fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state);
1491 exit(1);
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001492#endif
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001493 }
1494 http_msg_ood:
1495 /* out of data */
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001496 msg->msg_state = state;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001497 buf->lr = ptr;
1498 return;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001499
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001500 http_msg_invalid:
1501 /* invalid message */
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001502 msg->msg_state = HTTP_MSG_ERROR;
Willy Tarreau7552c032009-03-01 11:10:40 +01001503 buf->lr = ptr;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001504 return;
1505}
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001506
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02001507/* This function performs all the processing enabled for the current request.
Willy Tarreauedcf6682008-11-30 23:15:34 +01001508 * It returns 1 if the processing can continue on next analysers, or zero if it
1509 * needs more data, encounters an error, or wants to immediately abort the
1510 * request. It relies on buffers flags, and updates s->req->analysers. Its
1511 * behaviour is rather simple:
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001512 * - all enabled analysers are called in turn from the lower to the higher
1513 * bit.
Willy Tarreauedcf6682008-11-30 23:15:34 +01001514 * - the analyser must check for errors and timeouts, and react as expected.
1515 * It does not have to close anything upon error, the caller will.
1516 * - if the analyser does not have enough data, it must return 0without calling
1517 * other ones. It should also probably do a buffer_write_dis() to ensure
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001518 * that unprocessed data will not be forwarded. But that probably depends on
Willy Tarreauedcf6682008-11-30 23:15:34 +01001519 * the protocol.
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001520 * - if an analyser has enough data, it just has to pass on to the next
Willy Tarreauedcf6682008-11-30 23:15:34 +01001521 * analyser without using buffer_write_dis() (enabled by default).
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001522 * - if an analyser thinks it has no added value anymore staying here, it must
1523 * reset its bit from the analysers flags in order not to be called anymore.
1524 *
1525 * In the future, analysers should be able to indicate that they want to be
1526 * called after XXX bytes have been received (or transfered), and the min of
1527 * all's wishes will be used to ring back (unless a special condition occurs).
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001528 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001529int http_process_request(struct session *s, struct buffer *req)
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001530{
Willy Tarreau59234e92008-11-30 23:51:27 +01001531 /*
1532 * We will parse the partial (or complete) lines.
1533 * We will check the request syntax, and also join multi-line
1534 * headers. An index of all the lines will be elaborated while
1535 * parsing.
1536 *
1537 * For the parsing, we use a 28 states FSM.
1538 *
1539 * Here is the information we currently have :
Willy Tarreauf073a832009-03-01 23:21:47 +01001540 * req->data + msg->som = beginning of request
Willy Tarreau59234e92008-11-30 23:51:27 +01001541 * req->data + req->eoh = end of processed headers / start of current one
1542 * req->data + req->eol = end of current header or line (LF or CRLF)
1543 * req->lr = first non-visited byte
1544 * req->r = end of data
1545 */
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001546
Willy Tarreau59234e92008-11-30 23:51:27 +01001547 int cur_idx;
1548 struct http_txn *txn = &s->txn;
1549 struct http_msg *msg = &txn->req;
1550 struct proxy *cur_proxy;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001551
Willy Tarreau6bf17362009-02-24 10:48:35 +01001552 DPRINTF(stderr,"[%u] %s: session=%p b=%p, exp(r,w)=%u,%u bf=%08x bl=%d analysers=%02x\n",
1553 now_ms, __FUNCTION__,
1554 s,
1555 req,
1556 req->rex, req->wex,
1557 req->flags,
1558 req->l,
1559 req->analysers);
1560
Willy Tarreau59234e92008-11-30 23:51:27 +01001561 if (likely(req->lr < req->r))
1562 http_msg_analyzer(req, msg, &txn->hdr_idx);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001563
Willy Tarreau59234e92008-11-30 23:51:27 +01001564 /* 1: we might have to print this header in debug mode */
1565 if (unlikely((global.mode & MODE_DEBUG) &&
1566 (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) &&
1567 (msg->msg_state == HTTP_MSG_BODY || msg->msg_state == HTTP_MSG_ERROR))) {
1568 char *eol, *sol;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001569
Willy Tarreau59234e92008-11-30 23:51:27 +01001570 sol = req->data + msg->som;
1571 eol = sol + msg->sl.rq.l;
1572 debug_hdr("clireq", s, sol, eol);
Willy Tarreau45e73e32006-12-17 00:05:15 +01001573
Willy Tarreau59234e92008-11-30 23:51:27 +01001574 sol += hdr_idx_first_pos(&txn->hdr_idx);
1575 cur_idx = hdr_idx_first_idx(&txn->hdr_idx);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001576
Willy Tarreau59234e92008-11-30 23:51:27 +01001577 while (cur_idx) {
1578 eol = sol + txn->hdr_idx.v[cur_idx].len;
1579 debug_hdr("clihdr", s, sol, eol);
1580 sol = eol + txn->hdr_idx.v[cur_idx].cr + 1;
1581 cur_idx = txn->hdr_idx.v[cur_idx].next;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001582 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001583 }
1584
Willy Tarreau58f10d72006-12-04 02:26:12 +01001585
Willy Tarreau59234e92008-11-30 23:51:27 +01001586 /*
1587 * Now we quickly check if we have found a full valid request.
1588 * If not so, we check the FD and buffer states before leaving.
1589 * A full request is indicated by the fact that we have seen
1590 * the double LF/CRLF, so the state is HTTP_MSG_BODY. Invalid
1591 * requests are checked first.
1592 *
1593 */
Willy Tarreau58f10d72006-12-04 02:26:12 +01001594
Willy Tarreau59234e92008-11-30 23:51:27 +01001595 if (unlikely(msg->msg_state != HTTP_MSG_BODY)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01001596 /*
Willy Tarreau59234e92008-11-30 23:51:27 +01001597 * First, let's catch bad requests.
Willy Tarreau58f10d72006-12-04 02:26:12 +01001598 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001599 if (unlikely(msg->msg_state == HTTP_MSG_ERROR))
1600 goto return_bad_req;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001601
Willy Tarreau59234e92008-11-30 23:51:27 +01001602 /* 1: Since we are in header mode, if there's no space
1603 * left for headers, we won't be able to free more
1604 * later, so the session will never terminate. We
1605 * must terminate it now.
1606 */
1607 if (unlikely(req->flags & BF_FULL)) {
1608 /* FIXME: check if URI is set and return Status
1609 * 414 Request URI too long instead.
Willy Tarreau58f10d72006-12-04 02:26:12 +01001610 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001611 goto return_bad_req;
1612 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001613
Willy Tarreau59234e92008-11-30 23:51:27 +01001614 /* 2: have we encountered a read error ? */
1615 else if (req->flags & BF_READ_ERROR) {
1616 /* we cannot return any message on error */
1617 msg->msg_state = HTTP_MSG_ERROR;
1618 req->analysers = 0;
1619 s->fe->failed_req++;
1620 if (!(s->flags & SN_ERR_MASK))
1621 s->flags |= SN_ERR_CLICL;
1622 if (!(s->flags & SN_FINST_MASK))
1623 s->flags |= SN_FINST_R;
1624 return 0;
1625 }
Willy Tarreauf9839bd2008-08-27 23:57:16 +02001626
Willy Tarreau59234e92008-11-30 23:51:27 +01001627 /* 3: has the read timeout expired ? */
1628 else if (req->flags & BF_READ_TIMEOUT || tick_is_expired(req->analyse_exp, now_ms)) {
1629 /* read timeout : give up with an error message. */
1630 txn->status = 408;
1631 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_408));
1632 msg->msg_state = HTTP_MSG_ERROR;
1633 req->analysers = 0;
1634 s->fe->failed_req++;
1635 if (!(s->flags & SN_ERR_MASK))
1636 s->flags |= SN_ERR_CLITO;
1637 if (!(s->flags & SN_FINST_MASK))
1638 s->flags |= SN_FINST_R;
1639 return 0;
1640 }
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02001641
Willy Tarreau59234e92008-11-30 23:51:27 +01001642 /* 4: have we encountered a close ? */
1643 else if (req->flags & BF_SHUTR) {
1644 txn->status = 400;
1645 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_400));
1646 msg->msg_state = HTTP_MSG_ERROR;
1647 req->analysers = 0;
1648 s->fe->failed_req++;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02001649
Willy Tarreau59234e92008-11-30 23:51:27 +01001650 if (!(s->flags & SN_ERR_MASK))
1651 s->flags |= SN_ERR_CLICL;
1652 if (!(s->flags & SN_FINST_MASK))
1653 s->flags |= SN_FINST_R;
Willy Tarreaudafde432008-08-17 01:00:46 +02001654 return 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001655 }
1656
Willy Tarreau59234e92008-11-30 23:51:27 +01001657 buffer_write_dis(req);
Willy Tarreau1b194fe2009-03-21 21:10:04 +01001658 req->flags |= BF_READ_DONTWAIT; /* try to get back here ASAP */
1659
Willy Tarreau59234e92008-11-30 23:51:27 +01001660 /* just set the request timeout once at the beginning of the request */
1661 if (!tick_isset(req->analyse_exp))
1662 req->analyse_exp = tick_add_ifset(now_ms, s->fe->timeout.httpreq);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001663
Willy Tarreau59234e92008-11-30 23:51:27 +01001664 /* we're not ready yet */
1665 return 0;
1666 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001667
Willy Tarreau67f0eea2008-08-10 22:55:22 +02001668
Willy Tarreau59234e92008-11-30 23:51:27 +01001669 /****************************************************************
1670 * More interesting part now : we know that we have a complete *
1671 * request which at least looks like HTTP. We have an indicator *
1672 * of each header's length, so we can parse them quickly. *
1673 ****************************************************************/
Willy Tarreau9cdde232007-05-02 20:58:19 +02001674
Willy Tarreau59234e92008-11-30 23:51:27 +01001675 req->analysers &= ~AN_REQ_HTTP_HDR;
1676 req->analyse_exp = TICK_ETERNITY;
1677
1678 /* ensure we keep this pointer to the beginning of the message */
1679 msg->sol = req->data + msg->som;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001680
Willy Tarreau59234e92008-11-30 23:51:27 +01001681 /*
1682 * 1: identify the method
1683 */
1684 txn->meth = find_http_meth(&req->data[msg->som], msg->sl.rq.m_l);
1685
1686 /* we can make use of server redirect on GET and HEAD */
1687 if (txn->meth == HTTP_METH_GET || txn->meth == HTTP_METH_HEAD)
1688 s->flags |= SN_REDIRECTABLE;
Willy Tarreaufa7e1022008-10-19 07:30:41 +02001689
Willy Tarreau59234e92008-11-30 23:51:27 +01001690 /*
1691 * 2: check if the URI matches the monitor_uri.
1692 * We have to do this for every request which gets in, because
1693 * the monitor-uri is defined by the frontend.
1694 */
1695 if (unlikely((s->fe->monitor_uri_len != 0) &&
1696 (s->fe->monitor_uri_len == msg->sl.rq.u_l) &&
1697 !memcmp(&req->data[msg->sl.rq.u],
1698 s->fe->monitor_uri,
1699 s->fe->monitor_uri_len))) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01001700 /*
Willy Tarreau59234e92008-11-30 23:51:27 +01001701 * We have found the monitor URI
Willy Tarreau58f10d72006-12-04 02:26:12 +01001702 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001703 struct acl_cond *cond;
1704 cur_proxy = s->fe;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001705
Willy Tarreau59234e92008-11-30 23:51:27 +01001706 s->flags |= SN_MONITOR;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001707
Willy Tarreau59234e92008-11-30 23:51:27 +01001708 /* Check if we want to fail this monitor request or not */
1709 list_for_each_entry(cond, &cur_proxy->mon_fail_cond, list) {
1710 int ret = acl_exec_cond(cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreau11382812008-07-09 16:18:21 +02001711
Willy Tarreau59234e92008-11-30 23:51:27 +01001712 ret = acl_pass(ret);
1713 if (cond->pol == ACL_COND_UNLESS)
1714 ret = !ret;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001715
Willy Tarreau59234e92008-11-30 23:51:27 +01001716 if (ret) {
1717 /* we fail this request, let's return 503 service unavail */
1718 txn->status = 503;
1719 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_503));
1720 goto return_prx_cond;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001721 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001722 }
Willy Tarreaua5555ec2008-11-30 19:02:32 +01001723
Willy Tarreau59234e92008-11-30 23:51:27 +01001724 /* nothing to fail, let's reply normaly */
1725 txn->status = 200;
1726 stream_int_retnclose(req->prod, &http_200_chunk);
1727 goto return_prx_cond;
1728 }
1729
1730 /*
1731 * 3: Maybe we have to copy the original REQURI for the logs ?
1732 * Note: we cannot log anymore if the request has been
1733 * classified as invalid.
1734 */
1735 if (unlikely(s->logs.logwait & LW_REQ)) {
1736 /* we have a complete HTTP request that we must log */
1737 if ((txn->uri = pool_alloc2(pool2_requri)) != NULL) {
1738 int urilen = msg->sl.rq.l;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001739
Willy Tarreau59234e92008-11-30 23:51:27 +01001740 if (urilen >= REQURI_LEN)
1741 urilen = REQURI_LEN - 1;
1742 memcpy(txn->uri, &req->data[msg->som], urilen);
1743 txn->uri[urilen] = 0;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001744
Willy Tarreau59234e92008-11-30 23:51:27 +01001745 if (!(s->logs.logwait &= ~LW_REQ))
1746 s->do_log(s);
1747 } else {
1748 Alert("HTTP logging : out of memory.\n");
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001749 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001750 }
Willy Tarreau06619262006-12-17 08:37:22 +01001751
Willy Tarreau59234e92008-11-30 23:51:27 +01001752 /* 4. We may have to convert HTTP/0.9 requests to HTTP/1.0 */
1753 if (unlikely(msg->sl.rq.v_l == 0)) {
1754 int delta;
1755 char *cur_end;
1756 msg->sol = req->data + msg->som;
1757 cur_end = msg->sol + msg->sl.rq.l;
1758 delta = 0;
Willy Tarreau06619262006-12-17 08:37:22 +01001759
Willy Tarreau59234e92008-11-30 23:51:27 +01001760 if (msg->sl.rq.u_l == 0) {
1761 /* if no URI was set, add "/" */
1762 delta = buffer_replace2(req, cur_end, cur_end, " /", 2);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001763 cur_end += delta;
Willy Tarreau59234e92008-11-30 23:51:27 +01001764 msg->eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001765 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001766 /* add HTTP version */
1767 delta = buffer_replace2(req, cur_end, cur_end, " HTTP/1.0\r\n", 11);
1768 msg->eoh += delta;
1769 cur_end += delta;
1770 cur_end = (char *)http_parse_reqline(msg, req->data,
1771 HTTP_MSG_RQMETH,
1772 msg->sol, cur_end + 1,
1773 NULL, NULL);
1774 if (unlikely(!cur_end))
1775 goto return_bad_req;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001776
Willy Tarreau59234e92008-11-30 23:51:27 +01001777 /* we have a full HTTP/1.0 request now and we know that
1778 * we have either a CR or an LF at <ptr>.
Willy Tarreau58f10d72006-12-04 02:26:12 +01001779 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001780 hdr_idx_set_start(&txn->hdr_idx, msg->sl.rq.l, *cur_end == '\r');
1781 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001782
1783
Willy Tarreau59234e92008-11-30 23:51:27 +01001784 /* 5: we may need to capture headers */
1785 if (unlikely((s->logs.logwait & LW_REQHDR) && s->fe->req_cap))
1786 capture_headers(req->data + msg->som, &txn->hdr_idx,
1787 txn->req.cap, s->fe->req_cap);
Willy Tarreau11382812008-07-09 16:18:21 +02001788
Willy Tarreau59234e92008-11-30 23:51:27 +01001789 /*
1790 * 6: we will have to evaluate the filters.
1791 * As opposed to version 1.2, now they will be evaluated in the
1792 * filters order and not in the header order. This means that
1793 * each filter has to be validated among all headers.
1794 *
1795 * We can now check whether we want to switch to another
1796 * backend, in which case we will re-check the backend's
1797 * filters and various options. In order to support 3-level
1798 * switching, here's how we should proceed :
1799 *
1800 * a) run be.
1801 * if (switch) then switch ->be to the new backend.
1802 * b) run be if (be != fe).
1803 * There cannot be any switch from there, so ->be cannot be
1804 * changed anymore.
1805 *
1806 * => filters always apply to ->be, then ->be may change.
1807 *
1808 * The response path will be able to apply either ->be, or
1809 * ->be then ->fe filters in order to match the reverse of
1810 * the forward sequence.
1811 */
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001812
Willy Tarreau59234e92008-11-30 23:51:27 +01001813 do {
1814 struct acl_cond *cond;
1815 struct redirect_rule *rule;
1816 struct proxy *rule_set = s->be;
1817 cur_proxy = s->be;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001818
Willy Tarreau59234e92008-11-30 23:51:27 +01001819 /* first check whether we have some ACLs set to redirect this request */
1820 list_for_each_entry(rule, &cur_proxy->redirect_rules, list) {
1821 int ret = acl_exec_cond(rule->cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001822
Willy Tarreau59234e92008-11-30 23:51:27 +01001823 ret = acl_pass(ret);
1824 if (rule->cond->pol == ACL_COND_UNLESS)
1825 ret = !ret;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001826
Willy Tarreau59234e92008-11-30 23:51:27 +01001827 if (ret) {
1828 struct chunk rdr = { trash, 0 };
1829 const char *msg_fmt;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001830
Willy Tarreau59234e92008-11-30 23:51:27 +01001831 /* build redirect message */
1832 switch(rule->code) {
1833 case 303:
1834 rdr.len = strlen(HTTP_303);
1835 msg_fmt = HTTP_303;
1836 break;
1837 case 301:
1838 rdr.len = strlen(HTTP_301);
1839 msg_fmt = HTTP_301;
1840 break;
1841 case 302:
1842 default:
1843 rdr.len = strlen(HTTP_302);
1844 msg_fmt = HTTP_302;
1845 break;
1846 }
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001847
Willy Tarreau59234e92008-11-30 23:51:27 +01001848 if (unlikely(rdr.len > sizeof(trash)))
1849 goto return_bad_req;
1850 memcpy(rdr.str, msg_fmt, rdr.len);
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001851
Willy Tarreau59234e92008-11-30 23:51:27 +01001852 switch(rule->type) {
1853 case REDIRECT_TYPE_PREFIX: {
1854 const char *path;
1855 int pathlen;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001856
Willy Tarreau59234e92008-11-30 23:51:27 +01001857 path = http_get_path(txn);
1858 /* build message using path */
1859 if (path) {
1860 pathlen = txn->req.sl.rq.u_l + (txn->req.sol+txn->req.sl.rq.u) - path;
Willy Tarreau79da4692008-11-19 20:03:04 +01001861 if (rule->flags & REDIRECT_FLAG_DROP_QS) {
1862 int qs = 0;
1863 while (qs < pathlen) {
1864 if (path[qs] == '?') {
1865 pathlen = qs;
1866 break;
1867 }
1868 qs++;
1869 }
1870 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001871 } else {
1872 path = "/";
1873 pathlen = 1;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001874 }
1875
Willy Tarreau59234e92008-11-30 23:51:27 +01001876 if (rdr.len + rule->rdr_len + pathlen > sizeof(trash) - 4)
1877 goto return_bad_req;
1878
Willy Tarreaufe651a52008-11-19 21:15:17 +01001879 /* add prefix. Note that if prefix == "/", we don't want to
1880 * add anything, otherwise it makes it hard for the user to
1881 * configure a self-redirection.
1882 */
1883 if (rule->rdr_len != 1 || *rule->rdr_str != '/') {
1884 memcpy(rdr.str + rdr.len, rule->rdr_str, rule->rdr_len);
1885 rdr.len += rule->rdr_len;
1886 }
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001887
Willy Tarreau59234e92008-11-30 23:51:27 +01001888 /* add path */
1889 memcpy(rdr.str + rdr.len, path, pathlen);
1890 rdr.len += pathlen;
1891 break;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001892 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001893 case REDIRECT_TYPE_LOCATION:
1894 default:
1895 if (rdr.len + rule->rdr_len > sizeof(trash) - 4)
1896 goto return_bad_req;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001897
Willy Tarreau59234e92008-11-30 23:51:27 +01001898 /* add location */
1899 memcpy(rdr.str + rdr.len, rule->rdr_str, rule->rdr_len);
1900 rdr.len += rule->rdr_len;
1901 break;
1902 }
Willy Tarreau11382812008-07-09 16:18:21 +02001903
Willy Tarreau0140f252008-11-19 21:07:09 +01001904 if (rule->cookie_len) {
1905 memcpy(rdr.str + rdr.len, "\r\nSet-Cookie: ", 14);
1906 rdr.len += 14;
1907 memcpy(rdr.str + rdr.len, rule->cookie_str, rule->cookie_len);
1908 rdr.len += rule->cookie_len;
1909 memcpy(rdr.str + rdr.len, "\r\n", 2);
1910 rdr.len += 2;
1911 }
1912
Willy Tarreau59234e92008-11-30 23:51:27 +01001913 /* add end of headers */
1914 memcpy(rdr.str + rdr.len, "\r\n\r\n", 4);
1915 rdr.len += 4;
Willy Tarreau5c8e3e02007-05-07 00:58:25 +02001916
Willy Tarreau59234e92008-11-30 23:51:27 +01001917 txn->status = rule->code;
1918 /* let's log the request time */
1919 s->logs.tv_request = now;
1920 stream_int_retnclose(req->prod, &rdr);
1921 goto return_prx_cond;
Willy Tarreau5c8e3e02007-05-07 00:58:25 +02001922 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001923 }
Willy Tarreau5c8e3e02007-05-07 00:58:25 +02001924
Willy Tarreau59234e92008-11-30 23:51:27 +01001925 /* first check whether we have some ACLs set to block this request */
1926 list_for_each_entry(cond, &cur_proxy->block_cond, list) {
1927 int ret = acl_exec_cond(cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreau53b6c742006-12-17 13:37:46 +01001928
Willy Tarreau59234e92008-11-30 23:51:27 +01001929 ret = acl_pass(ret);
1930 if (cond->pol == ACL_COND_UNLESS)
1931 ret = !ret;
Willy Tarreauf1221aa2006-12-17 22:14:12 +01001932
Willy Tarreau59234e92008-11-30 23:51:27 +01001933 if (ret) {
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01001934 txn->status = 403;
Willy Tarreau06619262006-12-17 08:37:22 +01001935 /* let's log the request time */
Willy Tarreau59234e92008-11-30 23:51:27 +01001936 s->logs.tv_request = now;
1937 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_403));
Willy Tarreau06619262006-12-17 08:37:22 +01001938 goto return_prx_cond;
1939 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001940 }
1941
1942 /* try headers filters */
1943 if (rule_set->req_exp != NULL) {
1944 if (apply_filters_to_request(s, req, rule_set->req_exp) < 0)
1945 goto return_bad_req;
1946 }
1947
1948 if (!(s->flags & SN_BE_ASSIGNED) && (s->be != cur_proxy)) {
1949 /* to ensure correct connection accounting on
1950 * the backend, we count the connection for the
1951 * one managing the queue.
1952 */
1953 s->be->beconn++;
1954 if (s->be->beconn > s->be->beconn_max)
1955 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01001956 proxy_inc_be_ctr(s->be);
Willy Tarreau59234e92008-11-30 23:51:27 +01001957 s->flags |= SN_BE_ASSIGNED;
1958 }
Willy Tarreau06619262006-12-17 08:37:22 +01001959
Willy Tarreau59234e92008-11-30 23:51:27 +01001960 /* has the request been denied ? */
1961 if (txn->flags & TX_CLDENY) {
1962 /* no need to go further */
1963 txn->status = 403;
1964 /* let's log the request time */
1965 s->logs.tv_request = now;
1966 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_403));
1967 goto return_prx_cond;
1968 }
Willy Tarreau06619262006-12-17 08:37:22 +01001969
Willy Tarreau59234e92008-11-30 23:51:27 +01001970 /* We might have to check for "Connection:" */
1971 if (((s->fe->options | s->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO)) &&
1972 !(s->flags & SN_CONN_CLOSED)) {
1973 char *cur_ptr, *cur_end, *cur_next;
1974 int cur_idx, old_idx, delta, val;
1975 struct hdr_idx_elem *cur_hdr;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001976
Willy Tarreau59234e92008-11-30 23:51:27 +01001977 cur_next = req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
1978 old_idx = 0;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001979
Willy Tarreau59234e92008-11-30 23:51:27 +01001980 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
1981 cur_hdr = &txn->hdr_idx.v[cur_idx];
1982 cur_ptr = cur_next;
1983 cur_end = cur_ptr + cur_hdr->len;
1984 cur_next = cur_end + cur_hdr->cr + 1;
1985
1986 val = http_header_match2(cur_ptr, cur_end, "Connection", 10);
1987 if (val) {
1988 /* 3 possibilities :
1989 * - we have already set Connection: close,
1990 * so we remove this line.
1991 * - we have not yet set Connection: close,
1992 * but this line indicates close. We leave
1993 * it untouched and set the flag.
1994 * - we have not yet set Connection: close,
1995 * and this line indicates non-close. We
1996 * replace it.
1997 */
1998 if (s->flags & SN_CONN_CLOSED) {
1999 delta = buffer_replace2(req, cur_ptr, cur_next, NULL, 0);
2000 txn->req.eoh += delta;
2001 cur_next += delta;
2002 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
2003 txn->hdr_idx.used--;
2004 cur_hdr->len = 0;
2005 } else {
2006 if (strncasecmp(cur_ptr + val, "close", 5) != 0) {
2007 delta = buffer_replace2(req, cur_ptr + val, cur_end,
2008 "close", 5);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002009 cur_next += delta;
Willy Tarreau59234e92008-11-30 23:51:27 +01002010 cur_hdr->len += delta;
2011 txn->req.eoh += delta;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002012 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002013 s->flags |= SN_CONN_CLOSED;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002014 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002015 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002016 old_idx = cur_idx;
Willy Tarreau06619262006-12-17 08:37:22 +01002017 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002018 }
2019 /* add request headers from the rule sets in the same order */
2020 for (cur_idx = 0; cur_idx < rule_set->nb_reqadd; cur_idx++) {
2021 if (unlikely(http_header_add_tail(req,
2022 &txn->req,
2023 &txn->hdr_idx,
2024 rule_set->req_add[cur_idx])) < 0)
2025 goto return_bad_req;
2026 }
Willy Tarreaub2513902006-12-17 14:52:38 +01002027
Willy Tarreau59234e92008-11-30 23:51:27 +01002028 /* check if stats URI was requested, and if an auth is needed */
2029 if (rule_set->uri_auth != NULL &&
2030 (txn->meth == HTTP_METH_GET || txn->meth == HTTP_METH_HEAD)) {
2031 /* we have to check the URI and auth for this request.
2032 * FIXME!!! that one is rather dangerous, we want to
2033 * make it follow standard rules (eg: clear req->analysers).
2034 */
2035 if (stats_check_uri_auth(s, rule_set)) {
2036 req->analysers = 0;
2037 return 0;
Willy Tarreaub2513902006-12-17 14:52:38 +01002038 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002039 }
Willy Tarreaub2513902006-12-17 14:52:38 +01002040
Willy Tarreau59234e92008-11-30 23:51:27 +01002041 /* now check whether we have some switching rules for this request */
2042 if (!(s->flags & SN_BE_ASSIGNED)) {
2043 struct switching_rule *rule;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002044
Willy Tarreau59234e92008-11-30 23:51:27 +01002045 list_for_each_entry(rule, &cur_proxy->switching_rules, list) {
2046 int ret;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002047
Willy Tarreau59234e92008-11-30 23:51:27 +01002048 ret = acl_exec_cond(rule->cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreau11382812008-07-09 16:18:21 +02002049
Willy Tarreau59234e92008-11-30 23:51:27 +01002050 ret = acl_pass(ret);
2051 if (rule->cond->pol == ACL_COND_UNLESS)
2052 ret = !ret;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002053
Willy Tarreau59234e92008-11-30 23:51:27 +01002054 if (ret) {
2055 s->be = rule->be.backend;
2056 s->be->beconn++;
2057 if (s->be->beconn > s->be->beconn_max)
2058 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01002059 proxy_inc_be_ctr(s->be);
Willy Tarreau6e4261e2007-09-18 18:36:05 +02002060
Willy Tarreau59234e92008-11-30 23:51:27 +01002061 /* assign new parameters to the session from the new backend */
2062 s->rep->rto = s->req->wto = s->be->timeout.server;
2063 s->req->cto = s->be->timeout.connect;
2064 s->conn_retries = s->be->conn_retries;
Willy Tarreau32a4ec02009-04-02 11:35:18 +02002065 if (s->be->options2 & PR_O2_RSPBUG_OK)
2066 s->txn.rsp.err_pos = -1; /* let buggy responses pass */
Willy Tarreau59234e92008-11-30 23:51:27 +01002067 s->flags |= SN_BE_ASSIGNED;
2068 break;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002069 }
2070 }
Willy Tarreauf1221aa2006-12-17 22:14:12 +01002071 }
2072
Willy Tarreau59234e92008-11-30 23:51:27 +01002073 if (!(s->flags & SN_BE_ASSIGNED) && cur_proxy->defbe.be) {
2074 /* No backend was set, but there was a default
2075 * backend set in the frontend, so we use it and
2076 * loop again.
2077 */
2078 s->be = cur_proxy->defbe.be;
2079 s->be->beconn++;
2080 if (s->be->beconn > s->be->beconn_max)
2081 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01002082 proxy_inc_be_ctr(s->be);
Willy Tarreau58f10d72006-12-04 02:26:12 +01002083
Willy Tarreau59234e92008-11-30 23:51:27 +01002084 /* assign new parameters to the session from the new backend */
2085 s->rep->rto = s->req->wto = s->be->timeout.server;
2086 s->req->cto = s->be->timeout.connect;
2087 s->conn_retries = s->be->conn_retries;
Willy Tarreau32a4ec02009-04-02 11:35:18 +02002088 if (s->be->options2 & PR_O2_RSPBUG_OK)
2089 s->txn.rsp.err_pos = -1; /* let buggy responses pass */
Willy Tarreau59234e92008-11-30 23:51:27 +01002090 s->flags |= SN_BE_ASSIGNED;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01002091 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002092 } while (s->be != cur_proxy); /* we loop only if s->be has changed */
Willy Tarreau58f10d72006-12-04 02:26:12 +01002093
Willy Tarreau59234e92008-11-30 23:51:27 +01002094 if (!(s->flags & SN_BE_ASSIGNED)) {
2095 /* To ensure correct connection accounting on
2096 * the backend, we count the connection for the
2097 * one managing the queue.
Willy Tarreau06619262006-12-17 08:37:22 +01002098 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002099 s->be->beconn++;
2100 if (s->be->beconn > s->be->beconn_max)
2101 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01002102 proxy_inc_be_ctr(s->be);
Willy Tarreau59234e92008-11-30 23:51:27 +01002103 s->flags |= SN_BE_ASSIGNED;
2104 }
Willy Tarreau06619262006-12-17 08:37:22 +01002105
Willy Tarreau59234e92008-11-30 23:51:27 +01002106 /*
2107 * Right now, we know that we have processed the entire headers
2108 * and that unwanted requests have been filtered out. We can do
2109 * whatever we want with the remaining request. Also, now we
2110 * may have separate values for ->fe, ->be.
2111 */
Willy Tarreau06619262006-12-17 08:37:22 +01002112
Willy Tarreau59234e92008-11-30 23:51:27 +01002113 /*
2114 * If HTTP PROXY is set we simply get remote server address
2115 * parsing incoming request.
2116 */
2117 if ((s->be->options & PR_O_HTTP_PROXY) && !(s->flags & SN_ADDR_SET)) {
2118 url2sa(req->data + msg->sl.rq.u, msg->sl.rq.u_l, &s->srv_addr);
2119 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01002120
Willy Tarreau59234e92008-11-30 23:51:27 +01002121 /*
2122 * 7: the appsession cookie was looked up very early in 1.2,
2123 * so let's do the same now.
2124 */
Willy Tarreau58f10d72006-12-04 02:26:12 +01002125
Willy Tarreau59234e92008-11-30 23:51:27 +01002126 /* It needs to look into the URI */
2127 if (s->be->appsession_name) {
2128 get_srv_from_appsession(s, &req->data[msg->som], msg->sl.rq.l);
2129 }
Willy Tarreau45e73e32006-12-17 00:05:15 +01002130
Willy Tarreau59234e92008-11-30 23:51:27 +01002131 /*
2132 * 8: Now we can work with the cookies.
2133 * Note that doing so might move headers in the request, but
2134 * the fields will stay coherent and the URI will not move.
2135 * This should only be performed in the backend.
2136 */
Willy Tarreaufd39dda2008-10-17 12:01:58 +02002137 if ((s->be->cookie_name || s->be->appsession_name || s->fe->capture_name)
Willy Tarreau59234e92008-11-30 23:51:27 +01002138 && !(txn->flags & (TX_CLDENY|TX_CLTARPIT)))
2139 manage_client_side_cookies(s, req);
Willy Tarreau7ac51f62007-03-25 16:00:04 +02002140
Willy Tarreau59234e92008-11-30 23:51:27 +01002141 /*
2142 * 9: add X-Forwarded-For if either the frontend or the backend
2143 * asks for it.
2144 */
2145 if ((s->fe->options | s->be->options) & PR_O_FWDFOR) {
2146 if (s->cli_addr.ss_family == AF_INET) {
2147 /* Add an X-Forwarded-For header unless the source IP is
2148 * in the 'except' network range.
2149 */
2150 if ((!s->fe->except_mask.s_addr ||
2151 (((struct sockaddr_in *)&s->cli_addr)->sin_addr.s_addr & s->fe->except_mask.s_addr)
2152 != s->fe->except_net.s_addr) &&
2153 (!s->be->except_mask.s_addr ||
2154 (((struct sockaddr_in *)&s->cli_addr)->sin_addr.s_addr & s->be->except_mask.s_addr)
2155 != s->be->except_net.s_addr)) {
Willy Tarreau2a324282006-12-05 00:05:46 +01002156 int len;
Willy Tarreau59234e92008-11-30 23:51:27 +01002157 unsigned char *pn;
2158 pn = (unsigned char *)&((struct sockaddr_in *)&s->cli_addr)->sin_addr;
Ross Westaf72a1d2008-08-03 10:51:45 +02002159
2160 /* Note: we rely on the backend to get the header name to be used for
2161 * x-forwarded-for, because the header is really meant for the backends.
2162 * However, if the backend did not specify any option, we have to rely
2163 * on the frontend's header name.
2164 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002165 if (s->be->fwdfor_hdr_len) {
2166 len = s->be->fwdfor_hdr_len;
2167 memcpy(trash, s->be->fwdfor_hdr_name, len);
Ross Westaf72a1d2008-08-03 10:51:45 +02002168 } else {
Willy Tarreau59234e92008-11-30 23:51:27 +01002169 len = s->fe->fwdfor_hdr_len;
2170 memcpy(trash, s->fe->fwdfor_hdr_name, len);
2171 }
2172 len += sprintf(trash + len, ": %d.%d.%d.%d", pn[0], pn[1], pn[2], pn[3]);
Willy Tarreauedcf6682008-11-30 23:15:34 +01002173
Willy Tarreau4af6f3a2007-03-18 22:36:26 +01002174 if (unlikely(http_header_add_tail2(req, &txn->req,
2175 &txn->hdr_idx, trash, len)) < 0)
Willy Tarreau06619262006-12-17 08:37:22 +01002176 goto return_bad_req;
Willy Tarreau2a324282006-12-05 00:05:46 +01002177 }
2178 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002179 else if (s->cli_addr.ss_family == AF_INET6) {
2180 /* FIXME: for the sake of completeness, we should also support
2181 * 'except' here, although it is mostly useless in this case.
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002182 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002183 int len;
2184 char pn[INET6_ADDRSTRLEN];
2185 inet_ntop(AF_INET6,
2186 (const void *)&((struct sockaddr_in6 *)(&s->cli_addr))->sin6_addr,
2187 pn, sizeof(pn));
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002188
Willy Tarreau59234e92008-11-30 23:51:27 +01002189 /* Note: we rely on the backend to get the header name to be used for
2190 * x-forwarded-for, because the header is really meant for the backends.
2191 * However, if the backend did not specify any option, we have to rely
2192 * on the frontend's header name.
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002193 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002194 if (s->be->fwdfor_hdr_len) {
2195 len = s->be->fwdfor_hdr_len;
2196 memcpy(trash, s->be->fwdfor_hdr_name, len);
2197 } else {
2198 len = s->fe->fwdfor_hdr_len;
2199 memcpy(trash, s->fe->fwdfor_hdr_name, len);
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002200 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002201 len += sprintf(trash + len, ": %s", pn);
Willy Tarreauadfb8562008-08-11 15:24:42 +02002202
Willy Tarreau59234e92008-11-30 23:51:27 +01002203 if (unlikely(http_header_add_tail2(req, &txn->req,
2204 &txn->hdr_idx, trash, len)) < 0)
2205 goto return_bad_req;
2206 }
2207 }
2208
2209 /*
2210 * 10: add "Connection: close" if needed and not yet set.
2211 * Note that we do not need to add it in case of HTTP/1.0.
2212 */
2213 if (!(s->flags & SN_CONN_CLOSED) &&
2214 ((s->fe->options | s->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO))) {
2215 if ((unlikely(msg->sl.rq.v_l != 8) ||
2216 unlikely(req->data[msg->som + msg->sl.rq.v + 7] != '0')) &&
2217 unlikely(http_header_add_tail2(req, &txn->req, &txn->hdr_idx,
2218 "Connection: close", 17)) < 0)
2219 goto return_bad_req;
2220 s->flags |= SN_CONN_CLOSED;
2221 }
2222 /* Before we switch to data, was assignment set in manage_client_side_cookie?
2223 * If not assigned, perhaps we are balancing on url_param, but this is a
2224 * POST; and the parameters are in the body, maybe scan there to find our server.
2225 * (unless headers overflowed the buffer?)
2226 */
2227 if (!(s->flags & (SN_ASSIGNED|SN_DIRECT)) &&
2228 s->txn.meth == HTTP_METH_POST && s->be->url_param_name != NULL &&
2229 s->be->url_param_post_limit != 0 && !(req->flags & BF_FULL) &&
2230 memchr(msg->sol + msg->sl.rq.u, '?', msg->sl.rq.u_l) == NULL) {
2231 /* are there enough bytes here? total == l || r || rlim ?
2232 * len is unsigned, but eoh is int,
2233 * how many bytes of body have we received?
2234 * eoh is the first empty line of the header
2235 */
2236 /* already established CRLF or LF at eoh, move to start of message, find message length in buffer */
2237 unsigned long len = req->l - (msg->sol[msg->eoh] == '\r' ? msg->eoh + 2 : msg->eoh + 1);
2238
2239 /* If we have HTTP/1.1 and Expect: 100-continue, then abort.
2240 * We can't assume responsibility for the server's decision,
2241 * on this URI and header set. See rfc2616: 14.20, 8.2.3,
2242 * We also can't change our mind later, about which server to choose, so round robin.
2243 */
2244 if ((likely(msg->sl.rq.v_l == 8) && req->data[msg->som + msg->sl.rq.v + 7] == '1')) {
2245 struct hdr_ctx ctx;
2246 ctx.idx = 0;
2247 /* Expect is allowed in 1.1, look for it */
2248 http_find_header2("Expect", 6, msg->sol, &txn->hdr_idx, &ctx);
2249 if (ctx.idx != 0 &&
2250 unlikely(ctx.vlen == 12 && strncasecmp(ctx.line+ctx.val, "100-continue", 12) == 0))
2251 /* We can't reliablly stall and wait for data, because of
2252 * .NET clients that don't conform to rfc2616; so, no need for
2253 * the next block to check length expectations.
2254 * We could send 100 status back to the client, but then we need to
2255 * re-write headers, and send the message. And this isn't the right
2256 * place for that action.
2257 * TODO: support Expect elsewhere and delete this block.
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002258 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002259 goto end_check_maybe_wait_for_body;
2260 }
2261
2262 if (likely(len > s->be->url_param_post_limit)) {
2263 /* nothing to do, we got enough */
2264 } else {
2265 /* limit implies we are supposed to need this many bytes
2266 * to find the parameter. Let's see how many bytes we can wait for.
2267 */
2268 long long hint = len;
2269 struct hdr_ctx ctx;
2270 ctx.idx = 0;
2271 http_find_header2("Transfer-Encoding", 17, msg->sol, &txn->hdr_idx, &ctx);
2272 if (ctx.idx && ctx.vlen >= 7 && strncasecmp(ctx.line+ctx.val, "chunked", 7) == 0) {
2273 buffer_write_dis(req);
2274 req->analysers |= AN_REQ_HTTP_BODY;
2275 }
2276 else {
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002277 ctx.idx = 0;
Willy Tarreau59234e92008-11-30 23:51:27 +01002278 http_find_header2("Content-Length", 14, msg->sol, &txn->hdr_idx, &ctx);
2279 /* now if we have a length, we'll take the hint */
2280 if (ctx.idx) {
2281 /* We have Content-Length */
2282 if (strl2llrc(ctx.line+ctx.val,ctx.vlen, &hint))
2283 hint = 0; /* parse failure, untrusted client */
2284 else {
2285 if (hint > 0)
2286 msg->hdr_content_len = hint;
2287 else
2288 hint = 0; /* bad client, sent negative length */
2289 }
2290 }
2291 /* but limited to what we care about, maybe we don't expect any entity data (hint == 0) */
2292 if (s->be->url_param_post_limit < hint)
2293 hint = s->be->url_param_post_limit;
2294 /* now do we really need to buffer more data? */
2295 if (len < hint) {
Willy Tarreau3da77c52008-08-29 09:58:42 +02002296 buffer_write_dis(req);
Willy Tarreau2df28e82008-08-17 15:20:19 +02002297 req->analysers |= AN_REQ_HTTP_BODY;
Willy Tarreau41f40ed2008-08-21 10:05:00 +02002298 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002299 /* else... There are no body bytes to wait for */
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002300 }
2301 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002302 }
2303 end_check_maybe_wait_for_body:
Willy Tarreaubaaee002006-06-26 02:48:02 +02002304
Willy Tarreau59234e92008-11-30 23:51:27 +01002305 /*************************************************************
2306 * OK, that's finished for the headers. We have done what we *
2307 * could. Let's switch to the DATA state. *
2308 ************************************************************/
Willy Tarreaubaaee002006-06-26 02:48:02 +02002309
Willy Tarreau59234e92008-11-30 23:51:27 +01002310 buffer_set_rlim(req, BUFSIZE); /* no more rewrite needed */
2311 s->logs.tv_request = now;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002312
Willy Tarreau59234e92008-11-30 23:51:27 +01002313 /* When a connection is tarpitted, we use the tarpit timeout,
2314 * which may be the same as the connect timeout if unspecified.
2315 * If unset, then set it to zero because we really want it to
2316 * eventually expire. We build the tarpit as an analyser.
2317 */
2318 if (txn->flags & TX_CLTARPIT) {
Willy Tarreau6f0aa472009-03-08 20:33:29 +01002319 buffer_erase(s->req);
2320 /* wipe the request out so that we can drop the connection early
Willy Tarreau59234e92008-11-30 23:51:27 +01002321 * if the client closes first.
Willy Tarreau2a324282006-12-05 00:05:46 +01002322 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002323 buffer_write_dis(req);
2324 req->analysers |= AN_REQ_HTTP_TARPIT;
2325 req->analyse_exp = tick_add_ifset(now_ms, s->be->timeout.tarpit);
2326 if (!req->analyse_exp)
Willy Tarreau2ab85e62009-03-29 10:24:15 +02002327 req->analyse_exp = tick_add(now_ms, 0);
Willy Tarreau59234e92008-11-30 23:51:27 +01002328 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002329
Willy Tarreau59234e92008-11-30 23:51:27 +01002330 /* OK let's go on with the BODY now */
2331 return 1;
Willy Tarreau06619262006-12-17 08:37:22 +01002332
Willy Tarreau59234e92008-11-30 23:51:27 +01002333 return_bad_req: /* let's centralize all bad requests */
Willy Tarreauf073a832009-03-01 23:21:47 +01002334 if (unlikely(msg->msg_state == HTTP_MSG_ERROR)) {
2335 /* we detected a parsing error. We want to archive this request
2336 * in the dedicated proxy area for later troubleshooting.
2337 */
2338 struct error_snapshot *es = &s->fe->invalid_req;
2339 int maxlen = MIN(req->r - req->data + msg->som, sizeof(es->buf));
2340 memcpy(es->buf, req->data + msg->som, maxlen);
2341 es->pos = req->lr - req->data + msg->som;
2342 es->len = req->r - req->data + msg->som;
Willy Tarreaudefc52d2009-03-04 20:53:44 +01002343 es->when = date; // user-visible date
Willy Tarreauf073a832009-03-01 23:21:47 +01002344 es->sid = s->uniq_id;
2345 es->srv = s->srv;
2346 es->oe = s->be;
2347 es->src = s->cli_addr;
2348 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002349 txn->req.msg_state = HTTP_MSG_ERROR;
2350 txn->status = 400;
2351 req->analysers = 0;
2352 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_400));
2353 s->fe->failed_req++;
Willy Tarreauadfb8562008-08-11 15:24:42 +02002354
Willy Tarreau59234e92008-11-30 23:51:27 +01002355 return_prx_cond:
2356 if (!(s->flags & SN_ERR_MASK))
2357 s->flags |= SN_ERR_PRXCOND;
2358 if (!(s->flags & SN_FINST_MASK))
2359 s->flags |= SN_FINST_R;
Willy Tarreaudafde432008-08-17 01:00:46 +02002360 return 0;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002361}
Willy Tarreauadfb8562008-08-11 15:24:42 +02002362
Willy Tarreau60b85b02008-11-30 23:28:40 +01002363/* This function is an analyser which processes the HTTP tarpit. It always
2364 * returns zero, at the beginning because it prevents any other processing
2365 * from occurring, and at the end because it terminates the request.
2366 */
2367int http_process_tarpit(struct session *s, struct buffer *req)
2368{
2369 struct http_txn *txn = &s->txn;
2370
2371 /* This connection is being tarpitted. The CLIENT side has
2372 * already set the connect expiration date to the right
2373 * timeout. We just have to check that the client is still
2374 * there and that the timeout has not expired.
2375 */
2376 if ((req->flags & (BF_SHUTR|BF_READ_ERROR)) == 0 &&
2377 !tick_is_expired(req->analyse_exp, now_ms))
2378 return 0;
2379
2380 /* We will set the queue timer to the time spent, just for
2381 * logging purposes. We fake a 500 server error, so that the
2382 * attacker will not suspect his connection has been tarpitted.
2383 * It will not cause trouble to the logs because we can exclude
2384 * the tarpitted connections by filtering on the 'PT' status flags.
2385 */
Willy Tarreau60b85b02008-11-30 23:28:40 +01002386 s->logs.t_queue = tv_ms_elapsed(&s->logs.tv_accept, &now);
2387
2388 txn->status = 500;
2389 if (req->flags != BF_READ_ERROR)
2390 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_500));
2391
2392 req->analysers = 0;
2393 req->analyse_exp = TICK_ETERNITY;
2394
2395 s->fe->failed_req++;
2396 if (!(s->flags & SN_ERR_MASK))
2397 s->flags |= SN_ERR_PRXCOND;
2398 if (!(s->flags & SN_FINST_MASK))
2399 s->flags |= SN_FINST_T;
2400 return 0;
2401}
2402
Willy Tarreaud34af782008-11-30 23:36:37 +01002403/* This function is an analyser which processes the HTTP request body. It looks
2404 * for parameters to be used for the load balancing algorithm (url_param). It
2405 * must only be called after the standard HTTP request processing has occurred,
2406 * because it expects the request to be parsed. It returns zero if it needs to
2407 * read more data, or 1 once it has completed its analysis.
2408 */
2409int http_process_request_body(struct session *s, struct buffer *req)
2410{
2411 struct http_msg *msg = &s->txn.req;
2412 unsigned long body = msg->sol[msg->eoh] == '\r' ? msg->eoh + 2 : msg->eoh + 1;
2413 long long limit = s->be->url_param_post_limit;
2414 struct hdr_ctx ctx;
2415
2416 /* We have to parse the HTTP request body to find any required data.
2417 * "balance url_param check_post" should have been the only way to get
2418 * into this. We were brought here after HTTP header analysis, so all
2419 * related structures are ready.
2420 */
2421
2422 ctx.idx = 0;
2423
2424 /* now if we have a length, we'll take the hint */
2425 http_find_header2("Transfer-Encoding", 17, msg->sol, &s->txn.hdr_idx, &ctx);
2426 if (ctx.idx && ctx.vlen >= 7 && strncasecmp(ctx.line+ctx.val, "chunked", 7) == 0) {
2427 unsigned int chunk = 0;
2428 while (body < req->l && !HTTP_IS_CRLF(msg->sol[body])) {
2429 char c = msg->sol[body];
2430 if (ishex(c)) {
2431 unsigned int hex = toupper(c) - '0';
2432 if (hex > 9)
2433 hex -= 'A' - '9' - 1;
2434 chunk = (chunk << 4) | hex;
2435 } else
2436 break;
2437 body++;
2438 }
2439 if (body + 2 >= req->l) /* we want CRLF too */
2440 goto http_body_end; /* end of buffer? data missing! */
2441
2442 if (memcmp(msg->sol+body, "\r\n", 2) != 0)
2443 goto http_body_end; /* chunked encoding len ends with CRLF, and we don't have it yet */
2444
2445 body += 2; // skip CRLF
2446
2447 /* if we support more then one chunk here, we have to do it again when assigning server
2448 * 1. how much entity data do we have? new var
2449 * 2. should save entity_start, entity_cursor, elen & rlen in req; so we don't repeat scanning here
2450 * 3. test if elen > limit, or set new limit to elen if 0 (end of entity found)
2451 */
2452
2453 if (chunk < limit)
2454 limit = chunk; /* only reading one chunk */
2455 } else {
2456 if (msg->hdr_content_len < limit)
2457 limit = msg->hdr_content_len;
2458 }
2459
2460 http_body_end:
2461 /* we leave once we know we have nothing left to do. This means that we have
2462 * enough bytes, or that we know we'll not get any more (buffer full, read
2463 * buffer closed).
2464 */
2465 if (req->l - body >= limit || /* enough bytes! */
2466 req->flags & (BF_FULL | BF_READ_ERROR | BF_SHUTR | BF_READ_TIMEOUT) ||
2467 tick_is_expired(req->analyse_exp, now_ms)) {
2468 /* The situation will not evolve, so let's give up on the analysis. */
2469 s->logs.tv_request = now; /* update the request timer to reflect full request */
2470 req->analysers &= ~AN_REQ_HTTP_BODY;
2471 req->analyse_exp = TICK_ETERNITY;
2472 return 1;
2473 }
2474 else {
2475 /* Not enough data. We'll re-use the http-request
2476 * timeout here. Ideally, we should set the timeout
2477 * relative to the accept() date. We just set the
2478 * request timeout once at the beginning of the
2479 * request.
2480 */
2481 buffer_write_dis(req);
2482 if (!tick_isset(req->analyse_exp))
2483 req->analyse_exp = tick_add_ifset(now_ms, s->fe->timeout.httpreq);
2484 return 0;
2485 }
2486}
2487
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002488/* This function performs all the processing enabled for the current response.
Willy Tarreaudafde432008-08-17 01:00:46 +02002489 * It normally returns zero, but may return 1 if it absolutely needs to be
2490 * called again after other functions. It relies on buffers flags, and updates
Willy Tarreau2df28e82008-08-17 15:20:19 +02002491 * t->rep->analysers. It might make sense to explode it into several other
Willy Tarreau41f40ed2008-08-21 10:05:00 +02002492 * functions. It works like process_request (see indications above).
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002493 */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002494int process_response(struct session *t)
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002495{
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002496 struct http_txn *txn = &t->txn;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002497 struct buffer *req = t->req;
2498 struct buffer *rep = t->rep;
Willy Tarreauadfb8562008-08-11 15:24:42 +02002499
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02002500 DPRINTF(stderr,"[%u] %s: session=%p b=%p, exp(r,w)=%u,%u bf=%08x bl=%d analysers=%02x\n",
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002501 now_ms, __FUNCTION__,
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02002502 t,
2503 rep,
2504 rep->rex, rep->wex,
2505 rep->flags,
2506 rep->l,
2507 rep->analysers);
Willy Tarreau67f0eea2008-08-10 22:55:22 +02002508
Willy Tarreau2df28e82008-08-17 15:20:19 +02002509 if (rep->analysers & AN_RTR_HTTP_HDR) { /* receiving server headers */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002510 /*
2511 * Now parse the partial (or complete) lines.
2512 * We will check the response syntax, and also join multi-line
2513 * headers. An index of all the lines will be elaborated while
2514 * parsing.
2515 *
2516 * For the parsing, we use a 28 states FSM.
2517 *
2518 * Here is the information we currently have :
Willy Tarreaue393fe22008-08-16 22:18:07 +02002519 * rep->data + rep->som = beginning of response
2520 * rep->data + rep->eoh = end of processed headers / start of current one
2521 * rep->data + rep->eol = end of current header or line (LF or CRLF)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002522 * rep->lr = first non-visited byte
2523 * rep->r = end of data
2524 */
Willy Tarreau7f875f62008-08-11 17:35:01 +02002525
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002526 int cur_idx;
2527 struct http_msg *msg = &txn->rsp;
2528 struct proxy *cur_proxy;
2529
2530 if (likely(rep->lr < rep->r))
2531 http_msg_analyzer(rep, msg, &txn->hdr_idx);
2532
2533 /* 1: we might have to print this header in debug mode */
2534 if (unlikely((global.mode & MODE_DEBUG) &&
2535 (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) &&
2536 (msg->msg_state == HTTP_MSG_BODY || msg->msg_state == HTTP_MSG_ERROR))) {
2537 char *eol, *sol;
2538
2539 sol = rep->data + msg->som;
2540 eol = sol + msg->sl.rq.l;
2541 debug_hdr("srvrep", t, sol, eol);
2542
2543 sol += hdr_idx_first_pos(&txn->hdr_idx);
2544 cur_idx = hdr_idx_first_idx(&txn->hdr_idx);
2545
2546 while (cur_idx) {
2547 eol = sol + txn->hdr_idx.v[cur_idx].len;
2548 debug_hdr("srvhdr", t, sol, eol);
2549 sol = eol + txn->hdr_idx.v[cur_idx].cr + 1;
2550 cur_idx = txn->hdr_idx.v[cur_idx].next;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002551 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002552 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002553
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002554 /*
2555 * Now we quickly check if we have found a full valid response.
2556 * If not so, we check the FD and buffer states before leaving.
2557 * A full response is indicated by the fact that we have seen
2558 * the double LF/CRLF, so the state is HTTP_MSG_BODY. Invalid
2559 * responses are checked first.
2560 *
2561 * Depending on whether the client is still there or not, we
2562 * may send an error response back or not. Note that normally
2563 * we should only check for HTTP status there, and check I/O
2564 * errors somewhere else.
2565 */
2566
2567 if (unlikely(msg->msg_state != HTTP_MSG_BODY)) {
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002568 /* Invalid response */
2569 if (unlikely(msg->msg_state == HTTP_MSG_ERROR)) {
Willy Tarreauf073a832009-03-01 23:21:47 +01002570 /* we detected a parsing error. We want to archive this response
2571 * in the dedicated proxy area for later troubleshooting.
2572 */
2573 struct error_snapshot *es = &t->be->invalid_rep;
2574 int maxlen = MIN(rep->r - rep->data + msg->som, sizeof(es->buf));
2575 memcpy(es->buf, rep->data + msg->som, maxlen);
2576 es->pos = rep->lr - rep->data + msg->som;
2577 es->len = rep->r - rep->data + msg->som;
Willy Tarreaudefc52d2009-03-04 20:53:44 +01002578 es->when = date; // user-visible date
Willy Tarreauf073a832009-03-01 23:21:47 +01002579 es->sid = t->uniq_id;
2580 es->srv = t->srv;
2581 es->oe = t->fe;
2582 es->src = t->cli_addr;
2583
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002584 hdr_response_bad:
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002585 buffer_shutr_now(rep);
2586 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002587 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002588 t->srv->failed_resp++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002589 t->be->failed_resp++;
Willy Tarreau2df28e82008-08-17 15:20:19 +02002590 rep->analysers = 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002591 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002592 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002593 if (!(t->flags & SN_ERR_MASK))
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002594 t->flags |= SN_ERR_PRXCOND;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002595 if (!(t->flags & SN_FINST_MASK))
2596 t->flags |= SN_FINST_H;
Willy Tarreau461f6622008-08-15 23:43:19 +02002597
Willy Tarreaudafde432008-08-17 01:00:46 +02002598 return 0;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002599 }
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002600 /* too large response does not fit in buffer. */
2601 else if (rep->flags & BF_FULL) {
2602 goto hdr_response_bad;
2603 }
2604 /* read error */
2605 else if (rep->flags & BF_READ_ERROR) {
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002606 buffer_shutr_now(rep);
2607 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002608 if (t->srv)
2609 t->srv->failed_resp++;
2610 t->be->failed_resp++;
Willy Tarreau2df28e82008-08-17 15:20:19 +02002611 rep->analysers = 0;
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002612 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002613 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002614 if (!(t->flags & SN_ERR_MASK))
2615 t->flags |= SN_ERR_SRVCL;
2616 if (!(t->flags & SN_FINST_MASK))
2617 t->flags |= SN_FINST_H;
Willy Tarreaudafde432008-08-17 01:00:46 +02002618 return 0;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002619 }
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002620 /* read timeout : return a 504 to the client. */
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002621 else if (rep->flags & BF_READ_TIMEOUT) {
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002622 buffer_shutr_now(rep);
2623 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002624 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002625 t->srv->failed_resp++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002626 t->be->failed_resp++;
Willy Tarreau2df28e82008-08-17 15:20:19 +02002627 rep->analysers = 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002628 txn->status = 504;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002629 stream_int_return(rep->cons, error_message(t, HTTP_ERR_504));
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002630 if (!(t->flags & SN_ERR_MASK))
2631 t->flags |= SN_ERR_SRVTO;
2632 if (!(t->flags & SN_FINST_MASK))
2633 t->flags |= SN_FINST_H;
Willy Tarreaudafde432008-08-17 01:00:46 +02002634 return 0;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002635 }
Willy Tarreau8365f932009-03-15 23:11:49 +01002636 /* close from server */
2637 else if (rep->flags & BF_SHUTR) {
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002638 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002639 if (t->srv)
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002640 t->srv->failed_resp++;
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002641 t->be->failed_resp++;
2642 rep->analysers = 0;
2643 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002644 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002645 if (!(t->flags & SN_ERR_MASK))
2646 t->flags |= SN_ERR_SRVCL;
2647 if (!(t->flags & SN_FINST_MASK))
2648 t->flags |= SN_FINST_H;
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002649 return 0;
2650 }
Willy Tarreau8365f932009-03-15 23:11:49 +01002651 /* write error to client (we don't send any message then) */
2652 else if (rep->flags & BF_WRITE_ERROR) {
2653 buffer_shutr_now(rep);
2654 t->be->failed_resp++;
2655 rep->analysers = 0;
2656 if (!(t->flags & SN_ERR_MASK))
2657 t->flags |= SN_ERR_CLICL;
2658 if (!(t->flags & SN_FINST_MASK))
2659 t->flags |= SN_FINST_H;
2660 return 0;
2661 }
Willy Tarreau3da77c52008-08-29 09:58:42 +02002662 buffer_write_dis(rep);
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002663 return 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002664 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002665
Willy Tarreau21d2af32008-02-14 20:25:24 +01002666
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002667 /*****************************************************************
2668 * More interesting part now : we know that we have a complete *
2669 * response which at least looks like HTTP. We have an indicator *
2670 * of each header's length, so we can parse them quickly. *
2671 ****************************************************************/
Willy Tarreau21d2af32008-02-14 20:25:24 +01002672
Willy Tarreau2df28e82008-08-17 15:20:19 +02002673 rep->analysers &= ~AN_RTR_HTTP_HDR;
Willy Tarreaua7c52762008-08-16 18:40:18 +02002674
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002675 /* ensure we keep this pointer to the beginning of the message */
2676 msg->sol = rep->data + msg->som;
Willy Tarreau21d2af32008-02-14 20:25:24 +01002677
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002678 /*
2679 * 1: get the status code and check for cacheability.
2680 */
Willy Tarreau21d2af32008-02-14 20:25:24 +01002681
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002682 t->logs.logwait &= ~LW_RESP;
2683 txn->status = strl2ui(rep->data + msg->sl.st.c, msg->sl.st.c_l);
Willy Tarreau21d2af32008-02-14 20:25:24 +01002684
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002685 switch (txn->status) {
2686 case 200:
2687 case 203:
2688 case 206:
2689 case 300:
2690 case 301:
2691 case 410:
2692 /* RFC2616 @13.4:
2693 * "A response received with a status code of
2694 * 200, 203, 206, 300, 301 or 410 MAY be stored
2695 * by a cache (...) unless a cache-control
2696 * directive prohibits caching."
2697 *
2698 * RFC2616 @9.5: POST method :
2699 * "Responses to this method are not cacheable,
2700 * unless the response includes appropriate
2701 * Cache-Control or Expires header fields."
2702 */
2703 if (likely(txn->meth != HTTP_METH_POST) &&
2704 (t->be->options & (PR_O_CHK_CACHE|PR_O_COOK_NOC)))
2705 txn->flags |= TX_CACHEABLE | TX_CACHE_COOK;
2706 break;
2707 default:
2708 break;
2709 }
Willy Tarreau21d2af32008-02-14 20:25:24 +01002710
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002711 /*
2712 * 2: we may need to capture headers
2713 */
2714 if (unlikely((t->logs.logwait & LW_RSPHDR) && t->fe->rsp_cap))
2715 capture_headers(rep->data + msg->som, &txn->hdr_idx,
2716 txn->rsp.cap, t->fe->rsp_cap);
2717
2718 /*
2719 * 3: we will have to evaluate the filters.
2720 * As opposed to version 1.2, now they will be evaluated in the
2721 * filters order and not in the header order. This means that
2722 * each filter has to be validated among all headers.
2723 *
2724 * Filters are tried with ->be first, then with ->fe if it is
2725 * different from ->be.
2726 */
2727
2728 t->flags &= ~SN_CONN_CLOSED; /* prepare for inspection */
2729
2730 cur_proxy = t->be;
2731 while (1) {
2732 struct proxy *rule_set = cur_proxy;
2733
2734 /* try headers filters */
2735 if (rule_set->rsp_exp != NULL) {
2736 if (apply_filters_to_response(t, rep, rule_set->rsp_exp) < 0) {
2737 return_bad_resp:
Willy Tarreau8365f932009-03-15 23:11:49 +01002738 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002739 t->srv->failed_resp++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002740 cur_proxy->failed_resp++;
2741 return_srv_prx_502:
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002742 buffer_shutr_now(rep);
2743 buffer_shutw_now(req);
Willy Tarreau2df28e82008-08-17 15:20:19 +02002744 rep->analysers = 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002745 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002746 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002747 if (!(t->flags & SN_ERR_MASK))
2748 t->flags |= SN_ERR_PRXCOND;
2749 if (!(t->flags & SN_FINST_MASK))
2750 t->flags |= SN_FINST_H;
Willy Tarreaudafde432008-08-17 01:00:46 +02002751 return 0;
Willy Tarreau21d2af32008-02-14 20:25:24 +01002752 }
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002753 }
Willy Tarreau21d2af32008-02-14 20:25:24 +01002754
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002755 /* has the response been denied ? */
2756 if (txn->flags & TX_SVDENY) {
Willy Tarreau8365f932009-03-15 23:11:49 +01002757 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002758 t->srv->failed_secu++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002759 cur_proxy->denied_resp++;
2760 goto return_srv_prx_502;
Willy Tarreau51406232008-03-10 22:04:20 +01002761 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002762
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002763 /* We might have to check for "Connection:" */
2764 if (((t->fe->options | t->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO)) &&
2765 !(t->flags & SN_CONN_CLOSED)) {
2766 char *cur_ptr, *cur_end, *cur_next;
2767 int cur_idx, old_idx, delta, val;
2768 struct hdr_idx_elem *cur_hdr;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002769
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002770 cur_next = rep->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
2771 old_idx = 0;
Willy Tarreau541b5c22008-01-06 23:34:21 +01002772
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002773 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
2774 cur_hdr = &txn->hdr_idx.v[cur_idx];
2775 cur_ptr = cur_next;
2776 cur_end = cur_ptr + cur_hdr->len;
2777 cur_next = cur_end + cur_hdr->cr + 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002778
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002779 val = http_header_match2(cur_ptr, cur_end, "Connection", 10);
2780 if (val) {
2781 /* 3 possibilities :
2782 * - we have already set Connection: close,
2783 * so we remove this line.
2784 * - we have not yet set Connection: close,
2785 * but this line indicates close. We leave
2786 * it untouched and set the flag.
2787 * - we have not yet set Connection: close,
2788 * and this line indicates non-close. We
2789 * replace it.
2790 */
2791 if (t->flags & SN_CONN_CLOSED) {
2792 delta = buffer_replace2(rep, cur_ptr, cur_next, NULL, 0);
2793 txn->rsp.eoh += delta;
2794 cur_next += delta;
2795 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
2796 txn->hdr_idx.used--;
2797 cur_hdr->len = 0;
2798 } else {
2799 if (strncasecmp(cur_ptr + val, "close", 5) != 0) {
2800 delta = buffer_replace2(rep, cur_ptr + val, cur_end,
2801 "close", 5);
2802 cur_next += delta;
2803 cur_hdr->len += delta;
2804 txn->rsp.eoh += delta;
2805 }
2806 t->flags |= SN_CONN_CLOSED;
2807 }
2808 }
2809 old_idx = cur_idx;
Willy Tarreau541b5c22008-01-06 23:34:21 +01002810 }
2811 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002812
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002813 /* add response headers from the rule sets in the same order */
2814 for (cur_idx = 0; cur_idx < rule_set->nb_rspadd; cur_idx++) {
2815 if (unlikely(http_header_add_tail(rep, &txn->rsp, &txn->hdr_idx,
2816 rule_set->rsp_add[cur_idx])) < 0)
2817 goto return_bad_resp;
Willy Tarreau0bbc3cf2006-10-15 14:26:02 +02002818 }
2819
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002820 /* check whether we're already working on the frontend */
2821 if (cur_proxy == t->fe)
2822 break;
2823 cur_proxy = t->fe;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002824 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002825
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002826 /*
2827 * 4: check for server cookie.
2828 */
Willy Tarreaufd39dda2008-10-17 12:01:58 +02002829 if (t->be->cookie_name || t->be->appsession_name || t->fe->capture_name
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002830 || (t->be->options & PR_O_CHK_CACHE))
2831 manage_server_side_cookies(t, rep);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002832
Willy Tarreaubaaee002006-06-26 02:48:02 +02002833
Willy Tarreaua15645d2007-03-18 16:22:39 +01002834 /*
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002835 * 5: check for cache-control or pragma headers if required.
Willy Tarreaua15645d2007-03-18 16:22:39 +01002836 */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002837 if ((t->be->options & (PR_O_COOK_NOC | PR_O_CHK_CACHE)) != 0)
2838 check_response_for_cacheability(t, rep);
Willy Tarreaua15645d2007-03-18 16:22:39 +01002839
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002840 /*
2841 * 6: add server cookie in the response if needed
2842 */
2843 if ((t->srv) && !(t->flags & SN_DIRECT) && (t->be->options & PR_O_COOK_INS) &&
2844 (!(t->be->options & PR_O_COOK_POST) || (txn->meth == HTTP_METH_POST))) {
2845 int len;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002846
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002847 /* the server is known, it's not the one the client requested, we have to
2848 * insert a set-cookie here, except if we want to insert only on POST
2849 * requests and this one isn't. Note that servers which don't have cookies
2850 * (eg: some backup servers) will return a full cookie removal request.
2851 */
2852 len = sprintf(trash, "Set-Cookie: %s=%s; path=/",
2853 t->be->cookie_name,
2854 t->srv->cookie ? t->srv->cookie : "; Expires=Thu, 01-Jan-1970 00:00:01 GMT");
Willy Tarreaubaaee002006-06-26 02:48:02 +02002855
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002856 if (t->be->cookie_domain)
2857 len += sprintf(trash+len, "; domain=%s", t->be->cookie_domain);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002858
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002859 if (unlikely(http_header_add_tail2(rep, &txn->rsp, &txn->hdr_idx,
2860 trash, len)) < 0)
2861 goto return_bad_resp;
2862 txn->flags |= TX_SCK_INSERTED;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002863
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002864 /* Here, we will tell an eventual cache on the client side that we don't
2865 * want it to cache this reply because HTTP/1.0 caches also cache cookies !
2866 * Some caches understand the correct form: 'no-cache="set-cookie"', but
2867 * others don't (eg: apache <= 1.3.26). So we use 'private' instead.
2868 */
2869 if ((t->be->options & PR_O_COOK_NOC) && (txn->flags & TX_CACHEABLE)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02002870
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002871 txn->flags &= ~TX_CACHEABLE & ~TX_CACHE_COOK;
2872
2873 if (unlikely(http_header_add_tail2(rep, &txn->rsp, &txn->hdr_idx,
2874 "Cache-control: private", 22)) < 0)
2875 goto return_bad_resp;
Willy Tarreaua15645d2007-03-18 16:22:39 +01002876 }
2877 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002878
Willy Tarreaubaaee002006-06-26 02:48:02 +02002879
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002880 /*
2881 * 7: check if result will be cacheable with a cookie.
2882 * We'll block the response if security checks have caught
2883 * nasty things such as a cacheable cookie.
2884 */
2885 if (((txn->flags & (TX_CACHEABLE | TX_CACHE_COOK | TX_SCK_ANY)) ==
2886 (TX_CACHEABLE | TX_CACHE_COOK | TX_SCK_ANY)) &&
2887 (t->be->options & PR_O_CHK_CACHE)) {
2888
2889 /* we're in presence of a cacheable response containing
2890 * a set-cookie header. We'll block it as requested by
2891 * the 'checkcache' option, and send an alert.
Willy Tarreaua15645d2007-03-18 16:22:39 +01002892 */
Willy Tarreau8365f932009-03-15 23:11:49 +01002893 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002894 t->srv->failed_secu++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002895 t->be->denied_resp++;
Willy Tarreaua15645d2007-03-18 16:22:39 +01002896
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002897 Alert("Blocking cacheable cookie in response from instance %s, server %s.\n",
2898 t->be->id, t->srv?t->srv->id:"<dispatch>");
2899 send_log(t->be, LOG_ALERT,
2900 "Blocking cacheable cookie in response from instance %s, server %s.\n",
2901 t->be->id, t->srv?t->srv->id:"<dispatch>");
2902 goto return_srv_prx_502;
2903 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01002904
2905 /*
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002906 * 8: add "Connection: close" if needed and not yet set.
2907 * Note that we do not need to add it in case of HTTP/1.0.
Willy Tarreaua15645d2007-03-18 16:22:39 +01002908 */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002909 if (!(t->flags & SN_CONN_CLOSED) &&
2910 ((t->fe->options | t->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO))) {
2911 if ((unlikely(msg->sl.st.v_l != 8) ||
2912 unlikely(req->data[msg->som + 7] != '0')) &&
2913 unlikely(http_header_add_tail2(rep, &txn->rsp, &txn->hdr_idx,
2914 "Connection: close", 17)) < 0)
2915 goto return_bad_resp;
2916 t->flags |= SN_CONN_CLOSED;
2917 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01002918
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002919 /*************************************************************
2920 * OK, that's finished for the headers. We have done what we *
2921 * could. Let's switch to the DATA state. *
2922 ************************************************************/
Willy Tarreaubaaee002006-06-26 02:48:02 +02002923
Willy Tarreaue393fe22008-08-16 22:18:07 +02002924 buffer_set_rlim(rep, BUFSIZE); /* no more rewrite needed */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002925 t->logs.t_data = tv_ms_elapsed(&t->logs.tv_accept, &now);
Willy Tarreaua15645d2007-03-18 16:22:39 +01002926
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002927#ifdef CONFIG_HAP_TCPSPLICE
2928 if ((t->fe->options & t->be->options) & PR_O_TCPSPLICE) {
2929 /* TCP splicing supported by both FE and BE */
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02002930 tcp_splice_splicefd(rep->cons->fd, rep->prod->fd, 0);
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002931 }
2932#endif
2933 /* if the user wants to log as soon as possible, without counting
2934 * bytes from the server, then this is the right moment. We have
2935 * to temporarily assign bytes_out to log what we currently have.
2936 */
2937 if (t->fe->to_log && !(t->logs.logwait & LW_BYTES)) {
2938 t->logs.t_close = t->logs.t_data; /* to get a valid end date */
2939 t->logs.bytes_out = txn->rsp.eoh;
Willy Tarreaua5555ec2008-11-30 19:02:32 +01002940 t->do_log(t);
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002941 t->logs.bytes_out = 0;
2942 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01002943
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002944 /* Note: we must not try to cheat by jumping directly to DATA,
2945 * otherwise we would not let the client side wake up.
2946 */
Willy Tarreaua15645d2007-03-18 16:22:39 +01002947
Willy Tarreaudafde432008-08-17 01:00:46 +02002948 return 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002949 }
Willy Tarreaudafde432008-08-17 01:00:46 +02002950
Willy Tarreau2df28e82008-08-17 15:20:19 +02002951 /* Note: eventhough nobody should set an unknown flag, clearing them right now will
2952 * probably reduce one day's debugging session.
2953 */
2954#ifdef DEBUG_DEV
2955 if (rep->analysers & ~(AN_RTR_HTTP_HDR)) {
2956 fprintf(stderr, "FIXME !!!! unknown analysers flags %s:%d = 0x%08X\n",
2957 __FILE__, __LINE__, rep->analysers);
2958 ABORT_NOW();
2959 }
2960#endif
2961 rep->analysers &= AN_RTR_HTTP_HDR;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002962 return 0;
2963}
Willy Tarreaua15645d2007-03-18 16:22:39 +01002964
Willy Tarreaubaaee002006-06-26 02:48:02 +02002965/*
2966 * Produces data for the session <s> depending on its source. Expects to be
Willy Tarreau1ae3a052008-08-16 10:56:30 +02002967 * called with client socket shut down on input. Right now, only statistics can
Willy Tarreau72b179a2008-08-28 16:01:32 +02002968 * be produced. It stops by itself by unsetting the BF_HIJACK flag from the
2969 * buffer, which it uses to keep on being called when there is free space in
Willy Tarreau01bf8672008-12-07 18:03:29 +01002970 * the buffer, or simply by letting an empty buffer upon return.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002971 */
Willy Tarreau01bf8672008-12-07 18:03:29 +01002972void produce_content(struct session *s, struct buffer *rep)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002973{
Willy Tarreaubaaee002006-06-26 02:48:02 +02002974 if (s->data_source == DATA_SRC_NONE) {
Willy Tarreau01bf8672008-12-07 18:03:29 +01002975 buffer_stop_hijack(rep);
2976 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002977 }
2978 else if (s->data_source == DATA_SRC_STATS) {
Willy Tarreauc0dde7a2007-01-01 21:38:07 +01002979 /* dump server statistics */
Willy Tarreau0a464892008-12-07 18:30:00 +01002980 int ret = stats_dump_http(s, rep, s->be->uri_auth);
Willy Tarreau91861262007-10-17 17:06:05 +02002981 if (ret >= 0)
Willy Tarreau01bf8672008-12-07 18:03:29 +01002982 return;
Willy Tarreau91861262007-10-17 17:06:05 +02002983 /* -1 indicates an error */
Willy Tarreauc0dde7a2007-01-01 21:38:07 +01002984 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002985
Willy Tarreau91861262007-10-17 17:06:05 +02002986 /* unknown data source or internal error */
2987 s->txn.status = 500;
Willy Tarreau01bf8672008-12-07 18:03:29 +01002988 stream_int_retnclose(rep->cons, error_message(s, HTTP_ERR_500));
Willy Tarreau91861262007-10-17 17:06:05 +02002989 if (!(s->flags & SN_ERR_MASK))
2990 s->flags |= SN_ERR_PRXCOND;
2991 if (!(s->flags & SN_FINST_MASK))
2992 s->flags |= SN_FINST_R;
Willy Tarreau01bf8672008-12-07 18:03:29 +01002993 buffer_stop_hijack(rep);
2994 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002995}
2996
2997
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002998/* Iterate the same filter through all request headers.
2999 * Returns 1 if this filter can be stopped upon return, otherwise 0.
Willy Tarreaua15645d2007-03-18 16:22:39 +01003000 * Since it can manage the switch to another backend, it updates the per-proxy
3001 * DENY stats.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003002 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003003int apply_filter_to_req_headers(struct session *t, struct buffer *req, struct hdr_exp *exp)
Willy Tarreau58f10d72006-12-04 02:26:12 +01003004{
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003005 char term;
3006 char *cur_ptr, *cur_end, *cur_next;
3007 int cur_idx, old_idx, last_hdr;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003008 struct http_txn *txn = &t->txn;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003009 struct hdr_idx_elem *cur_hdr;
3010 int len, delta;
Willy Tarreau0f7562b2007-01-07 15:46:13 +01003011
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003012 last_hdr = 0;
3013
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003014 cur_next = req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003015 old_idx = 0;
3016
3017 while (!last_hdr) {
Willy Tarreau3d300592007-03-18 18:34:41 +01003018 if (unlikely(txn->flags & (TX_CLDENY | TX_CLTARPIT)))
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003019 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003020 else if (unlikely(txn->flags & TX_CLALLOW) &&
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003021 (exp->action == ACT_ALLOW ||
3022 exp->action == ACT_DENY ||
3023 exp->action == ACT_TARPIT))
3024 return 0;
3025
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003026 cur_idx = txn->hdr_idx.v[old_idx].next;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003027 if (!cur_idx)
3028 break;
3029
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003030 cur_hdr = &txn->hdr_idx.v[cur_idx];
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003031 cur_ptr = cur_next;
3032 cur_end = cur_ptr + cur_hdr->len;
3033 cur_next = cur_end + cur_hdr->cr + 1;
3034
3035 /* Now we have one header between cur_ptr and cur_end,
3036 * and the next header starts at cur_next.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003037 */
3038
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003039 /* The annoying part is that pattern matching needs
3040 * that we modify the contents to null-terminate all
3041 * strings before testing them.
3042 */
3043
3044 term = *cur_end;
3045 *cur_end = '\0';
3046
3047 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3048 switch (exp->action) {
3049 case ACT_SETBE:
3050 /* It is not possible to jump a second time.
3051 * FIXME: should we return an HTTP/500 here so that
3052 * the admin knows there's a problem ?
3053 */
3054 if (t->be != t->fe)
3055 break;
3056
3057 /* Swithing Proxy */
3058 t->be = (struct proxy *) exp->replace;
3059
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02003060 /* right now, the backend switch is not overly complicated
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003061 * because we have associated req_cap and rsp_cap to the
3062 * frontend, and the beconn will be updated later.
3063 */
3064
Willy Tarreaud7c30f92007-12-03 01:38:36 +01003065 t->rep->rto = t->req->wto = t->be->timeout.server;
3066 t->req->cto = t->be->timeout.connect;
Willy Tarreau6e4261e2007-09-18 18:36:05 +02003067 t->conn_retries = t->be->conn_retries;
Willy Tarreau32a4ec02009-04-02 11:35:18 +02003068 if (t->be->options2 & PR_O2_RSPBUG_OK)
3069 t->txn.rsp.err_pos = -1; /* let buggy responses pass */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003070 last_hdr = 1;
3071 break;
3072
3073 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003074 txn->flags |= TX_CLALLOW;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003075 last_hdr = 1;
3076 break;
3077
3078 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003079 txn->flags |= TX_CLDENY;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003080 last_hdr = 1;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003081 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003082 break;
3083
3084 case ACT_TARPIT:
Willy Tarreau3d300592007-03-18 18:34:41 +01003085 txn->flags |= TX_CLTARPIT;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003086 last_hdr = 1;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003087 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003088 break;
3089
3090 case ACT_REPLACE:
3091 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3092 delta = buffer_replace2(req, cur_ptr, cur_end, trash, len);
3093 /* FIXME: if the user adds a newline in the replacement, the
3094 * index will not be recalculated for now, and the new line
3095 * will not be counted as a new header.
3096 */
3097
3098 cur_end += delta;
3099 cur_next += delta;
3100 cur_hdr->len += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003101 txn->req.eoh += delta;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003102 break;
3103
3104 case ACT_REMOVE:
3105 delta = buffer_replace2(req, cur_ptr, cur_next, NULL, 0);
3106 cur_next += delta;
3107
3108 /* FIXME: this should be a separate function */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003109 txn->req.eoh += delta;
3110 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3111 txn->hdr_idx.used--;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003112 cur_hdr->len = 0;
3113 cur_end = NULL; /* null-term has been rewritten */
3114 break;
3115
3116 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01003117 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003118 if (cur_end)
3119 *cur_end = term; /* restore the string terminator */
Willy Tarreau58f10d72006-12-04 02:26:12 +01003120
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003121 /* keep the link from this header to next one in case of later
3122 * removal of next header.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003123 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003124 old_idx = cur_idx;
3125 }
3126 return 0;
3127}
3128
3129
3130/* Apply the filter to the request line.
3131 * Returns 0 if nothing has been done, 1 if the filter has been applied,
3132 * or -1 if a replacement resulted in an invalid request line.
Willy Tarreaua15645d2007-03-18 16:22:39 +01003133 * Since it can manage the switch to another backend, it updates the per-proxy
3134 * DENY stats.
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003135 */
3136int apply_filter_to_req_line(struct session *t, struct buffer *req, struct hdr_exp *exp)
3137{
3138 char term;
3139 char *cur_ptr, *cur_end;
3140 int done;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003141 struct http_txn *txn = &t->txn;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003142 int len, delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003143
Willy Tarreau58f10d72006-12-04 02:26:12 +01003144
Willy Tarreau3d300592007-03-18 18:34:41 +01003145 if (unlikely(txn->flags & (TX_CLDENY | TX_CLTARPIT)))
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003146 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003147 else if (unlikely(txn->flags & TX_CLALLOW) &&
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003148 (exp->action == ACT_ALLOW ||
3149 exp->action == ACT_DENY ||
3150 exp->action == ACT_TARPIT))
3151 return 0;
3152 else if (exp->action == ACT_REMOVE)
3153 return 0;
3154
3155 done = 0;
3156
Willy Tarreau9cdde232007-05-02 20:58:19 +02003157 cur_ptr = req->data + txn->req.som; /* should be equal to txn->sol */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003158 cur_end = cur_ptr + txn->req.sl.rq.l;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003159
3160 /* Now we have the request line between cur_ptr and cur_end */
3161
3162 /* The annoying part is that pattern matching needs
3163 * that we modify the contents to null-terminate all
3164 * strings before testing them.
3165 */
3166
3167 term = *cur_end;
3168 *cur_end = '\0';
3169
3170 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3171 switch (exp->action) {
3172 case ACT_SETBE:
3173 /* It is not possible to jump a second time.
3174 * FIXME: should we return an HTTP/500 here so that
3175 * the admin knows there's a problem ?
Willy Tarreau58f10d72006-12-04 02:26:12 +01003176 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003177 if (t->be != t->fe)
3178 break;
3179
3180 /* Swithing Proxy */
3181 t->be = (struct proxy *) exp->replace;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003182
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003183 /* right now, the backend switch is not too much complicated
3184 * because we have associated req_cap and rsp_cap to the
3185 * frontend, and the beconn will be updated later.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003186 */
3187
Willy Tarreaud7c30f92007-12-03 01:38:36 +01003188 t->rep->rto = t->req->wto = t->be->timeout.server;
3189 t->req->cto = t->be->timeout.connect;
Willy Tarreau6e4261e2007-09-18 18:36:05 +02003190 t->conn_retries = t->be->conn_retries;
Willy Tarreau32a4ec02009-04-02 11:35:18 +02003191 if (t->be->options2 & PR_O2_RSPBUG_OK)
3192 t->txn.rsp.err_pos = -1; /* let buggy responses pass */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003193 done = 1;
3194 break;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003195
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003196 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003197 txn->flags |= TX_CLALLOW;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003198 done = 1;
3199 break;
Willy Tarreaua496b602006-12-17 23:15:24 +01003200
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003201 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003202 txn->flags |= TX_CLDENY;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003203 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003204 done = 1;
3205 break;
Willy Tarreaua496b602006-12-17 23:15:24 +01003206
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003207 case ACT_TARPIT:
Willy Tarreau3d300592007-03-18 18:34:41 +01003208 txn->flags |= TX_CLTARPIT;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003209 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003210 done = 1;
3211 break;
Willy Tarreaua496b602006-12-17 23:15:24 +01003212
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003213 case ACT_REPLACE:
3214 *cur_end = term; /* restore the string terminator */
3215 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3216 delta = buffer_replace2(req, cur_ptr, cur_end, trash, len);
3217 /* FIXME: if the user adds a newline in the replacement, the
3218 * index will not be recalculated for now, and the new line
3219 * will not be counted as a new header.
3220 */
Willy Tarreaua496b602006-12-17 23:15:24 +01003221
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003222 txn->req.eoh += delta;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003223 cur_end += delta;
Willy Tarreaua496b602006-12-17 23:15:24 +01003224
Willy Tarreau9cdde232007-05-02 20:58:19 +02003225 txn->req.sol = req->data + txn->req.som; /* should be equal to txn->sol */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003226 cur_end = (char *)http_parse_reqline(&txn->req, req->data,
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003227 HTTP_MSG_RQMETH,
3228 cur_ptr, cur_end + 1,
3229 NULL, NULL);
3230 if (unlikely(!cur_end))
3231 return -1;
Willy Tarreaua496b602006-12-17 23:15:24 +01003232
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003233 /* we have a full request and we know that we have either a CR
3234 * or an LF at <ptr>.
3235 */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003236 txn->meth = find_http_meth(cur_ptr, txn->req.sl.rq.m_l);
3237 hdr_idx_set_start(&txn->hdr_idx, txn->req.sl.rq.l, *cur_end == '\r');
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003238 /* there is no point trying this regex on headers */
3239 return 1;
3240 }
3241 }
3242 *cur_end = term; /* restore the string terminator */
3243 return done;
3244}
Willy Tarreau97de6242006-12-27 17:18:38 +01003245
Willy Tarreau58f10d72006-12-04 02:26:12 +01003246
Willy Tarreau58f10d72006-12-04 02:26:12 +01003247
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003248/*
3249 * Apply all the req filters <exp> to all headers in buffer <req> of session <t>.
3250 * Returns 0 if everything is alright, or -1 in case a replacement lead to an
Willy Tarreaua15645d2007-03-18 16:22:39 +01003251 * unparsable request. Since it can manage the switch to another backend, it
3252 * updates the per-proxy DENY stats.
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003253 */
3254int apply_filters_to_request(struct session *t, struct buffer *req, struct hdr_exp *exp)
3255{
Willy Tarreau3d300592007-03-18 18:34:41 +01003256 struct http_txn *txn = &t->txn;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003257 /* iterate through the filters in the outer loop */
Willy Tarreau3d300592007-03-18 18:34:41 +01003258 while (exp && !(txn->flags & (TX_CLDENY|TX_CLTARPIT))) {
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003259 int ret;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003260
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003261 /*
3262 * The interleaving of transformations and verdicts
3263 * makes it difficult to decide to continue or stop
3264 * the evaluation.
3265 */
3266
Willy Tarreau3d300592007-03-18 18:34:41 +01003267 if ((txn->flags & TX_CLALLOW) &&
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003268 (exp->action == ACT_ALLOW || exp->action == ACT_DENY ||
3269 exp->action == ACT_TARPIT || exp->action == ACT_PASS)) {
3270 exp = exp->next;
3271 continue;
3272 }
3273
3274 /* Apply the filter to the request line. */
3275 ret = apply_filter_to_req_line(t, req, exp);
3276 if (unlikely(ret < 0))
3277 return -1;
3278
3279 if (likely(ret == 0)) {
3280 /* The filter did not match the request, it can be
3281 * iterated through all headers.
3282 */
3283 apply_filter_to_req_headers(t, req, exp);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003284 }
3285 exp = exp->next;
3286 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003287 return 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003288}
3289
3290
Willy Tarreaua15645d2007-03-18 16:22:39 +01003291
Willy Tarreau58f10d72006-12-04 02:26:12 +01003292/*
Willy Tarreau396d2c62007-11-04 19:30:00 +01003293 * Manage client-side cookie. It can impact performance by about 2% so it is
3294 * desirable to call it only when needed.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003295 */
3296void manage_client_side_cookies(struct session *t, struct buffer *req)
3297{
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003298 struct http_txn *txn = &t->txn;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003299 char *p1, *p2, *p3, *p4;
3300 char *del_colon, *del_cookie, *colon;
3301 int app_cookies;
3302
3303 appsess *asession_temp = NULL;
3304 appsess local_asession;
3305
3306 char *cur_ptr, *cur_end, *cur_next;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003307 int cur_idx, old_idx;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003308
Willy Tarreau2a324282006-12-05 00:05:46 +01003309 /* Iterate through the headers.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003310 * we start with the start line.
3311 */
Willy Tarreau83969f42007-01-22 08:55:47 +01003312 old_idx = 0;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003313 cur_next = req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003314
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003315 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003316 struct hdr_idx_elem *cur_hdr;
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003317 int val;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003318
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003319 cur_hdr = &txn->hdr_idx.v[cur_idx];
Willy Tarreau58f10d72006-12-04 02:26:12 +01003320 cur_ptr = cur_next;
3321 cur_end = cur_ptr + cur_hdr->len;
3322 cur_next = cur_end + cur_hdr->cr + 1;
3323
3324 /* We have one full header between cur_ptr and cur_end, and the
3325 * next header starts at cur_next. We're only interested in
3326 * "Cookie:" headers.
3327 */
3328
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003329 val = http_header_match2(cur_ptr, cur_end, "Cookie", 6);
3330 if (!val) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003331 old_idx = cur_idx;
3332 continue;
3333 }
3334
3335 /* Now look for cookies. Conforming to RFC2109, we have to support
3336 * attributes whose name begin with a '$', and associate them with
3337 * the right cookie, if we want to delete this cookie.
3338 * So there are 3 cases for each cookie read :
3339 * 1) it's a special attribute, beginning with a '$' : ignore it.
3340 * 2) it's a server id cookie that we *MAY* want to delete : save
3341 * some pointers on it (last semi-colon, beginning of cookie...)
3342 * 3) it's an application cookie : we *MAY* have to delete a previous
3343 * "special" cookie.
3344 * At the end of loop, if a "special" cookie remains, we may have to
3345 * remove it. If no application cookie persists in the header, we
3346 * *MUST* delete it
3347 */
3348
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003349 colon = p1 = cur_ptr + val; /* first non-space char after 'Cookie:' */
Willy Tarreau58f10d72006-12-04 02:26:12 +01003350
Willy Tarreau58f10d72006-12-04 02:26:12 +01003351 /* del_cookie == NULL => nothing to be deleted */
3352 del_colon = del_cookie = NULL;
3353 app_cookies = 0;
3354
3355 while (p1 < cur_end) {
3356 /* skip spaces and colons, but keep an eye on these ones */
3357 while (p1 < cur_end) {
3358 if (*p1 == ';' || *p1 == ',')
3359 colon = p1;
Willy Tarreau8f8e6452007-06-17 21:51:38 +02003360 else if (!isspace((unsigned char)*p1))
Willy Tarreau58f10d72006-12-04 02:26:12 +01003361 break;
3362 p1++;
3363 }
3364
3365 if (p1 == cur_end)
3366 break;
3367
3368 /* p1 is at the beginning of the cookie name */
3369 p2 = p1;
3370 while (p2 < cur_end && *p2 != '=')
3371 p2++;
3372
3373 if (p2 == cur_end)
3374 break;
3375
3376 p3 = p2 + 1; /* skips the '=' sign */
3377 if (p3 == cur_end)
3378 break;
3379
3380 p4 = p3;
Willy Tarreau8f8e6452007-06-17 21:51:38 +02003381 while (p4 < cur_end && !isspace((unsigned char)*p4) && *p4 != ';' && *p4 != ',')
Willy Tarreau58f10d72006-12-04 02:26:12 +01003382 p4++;
3383
3384 /* here, we have the cookie name between p1 and p2,
3385 * and its value between p3 and p4.
3386 * we can process it :
3387 *
3388 * Cookie: NAME=VALUE;
3389 * | || || |
3390 * | || || +--> p4
3391 * | || |+-------> p3
3392 * | || +--------> p2
3393 * | |+------------> p1
3394 * | +-------------> colon
3395 * +--------------------> cur_ptr
3396 */
3397
3398 if (*p1 == '$') {
3399 /* skip this one */
3400 }
3401 else {
3402 /* first, let's see if we want to capture it */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003403 if (t->fe->capture_name != NULL &&
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003404 txn->cli_cookie == NULL &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003405 (p4 - p1 >= t->fe->capture_namelen) &&
3406 memcmp(p1, t->fe->capture_name, t->fe->capture_namelen) == 0) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003407 int log_len = p4 - p1;
3408
Willy Tarreau086b3b42007-05-13 21:45:51 +02003409 if ((txn->cli_cookie = pool_alloc2(pool2_capture)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003410 Alert("HTTP logging : out of memory.\n");
3411 } else {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003412 if (log_len > t->fe->capture_len)
3413 log_len = t->fe->capture_len;
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003414 memcpy(txn->cli_cookie, p1, log_len);
3415 txn->cli_cookie[log_len] = 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003416 }
3417 }
3418
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003419 if ((p2 - p1 == t->be->cookie_len) && (t->be->cookie_name != NULL) &&
3420 (memcmp(p1, t->be->cookie_name, p2 - p1) == 0)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003421 /* Cool... it's the right one */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003422 struct server *srv = t->be->srv;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003423 char *delim;
3424
3425 /* if we're in cookie prefix mode, we'll search the delimitor so that we
3426 * have the server ID betweek p3 and delim, and the original cookie between
3427 * delim+1 and p4. Otherwise, delim==p4 :
3428 *
3429 * Cookie: NAME=SRV~VALUE;
3430 * | || || | |
3431 * | || || | +--> p4
3432 * | || || +--------> delim
3433 * | || |+-----------> p3
3434 * | || +------------> p2
3435 * | |+----------------> p1
3436 * | +-----------------> colon
3437 * +------------------------> cur_ptr
3438 */
3439
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003440 if (t->be->options & PR_O_COOK_PFX) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003441 for (delim = p3; delim < p4; delim++)
3442 if (*delim == COOKIE_DELIM)
3443 break;
3444 }
3445 else
3446 delim = p4;
3447
3448
3449 /* Here, we'll look for the first running server which supports the cookie.
3450 * This allows to share a same cookie between several servers, for example
3451 * to dedicate backup servers to specific servers only.
3452 * However, to prevent clients from sticking to cookie-less backup server
3453 * when they have incidentely learned an empty cookie, we simply ignore
3454 * empty cookies and mark them as invalid.
3455 */
3456 if (delim == p3)
3457 srv = NULL;
3458
3459 while (srv) {
Willy Tarreau92f2ab12007-02-02 22:14:47 +01003460 if (srv->cookie && (srv->cklen == delim - p3) &&
3461 !memcmp(p3, srv->cookie, delim - p3)) {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003462 if (srv->state & SRV_RUNNING || t->be->options & PR_O_PERSIST) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003463 /* we found the server and it's usable */
Willy Tarreau3d300592007-03-18 18:34:41 +01003464 txn->flags &= ~TX_CK_MASK;
3465 txn->flags |= TX_CK_VALID;
3466 t->flags |= SN_DIRECT | SN_ASSIGNED;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003467 t->srv = srv;
3468 break;
3469 } else {
3470 /* we found a server, but it's down */
Willy Tarreau3d300592007-03-18 18:34:41 +01003471 txn->flags &= ~TX_CK_MASK;
3472 txn->flags |= TX_CK_DOWN;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003473 }
3474 }
3475 srv = srv->next;
3476 }
3477
Willy Tarreau3d300592007-03-18 18:34:41 +01003478 if (!srv && !(txn->flags & TX_CK_DOWN)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003479 /* no server matched this cookie */
Willy Tarreau3d300592007-03-18 18:34:41 +01003480 txn->flags &= ~TX_CK_MASK;
3481 txn->flags |= TX_CK_INVALID;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003482 }
3483
3484 /* depending on the cookie mode, we may have to either :
3485 * - delete the complete cookie if we're in insert+indirect mode, so that
3486 * the server never sees it ;
3487 * - remove the server id from the cookie value, and tag the cookie as an
3488 * application cookie so that it does not get accidentely removed later,
3489 * if we're in cookie prefix mode
3490 */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003491 if ((t->be->options & PR_O_COOK_PFX) && (delim != p4)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003492 int delta; /* negative */
3493
3494 delta = buffer_replace2(req, p3, delim + 1, NULL, 0);
3495 p4 += delta;
3496 cur_end += delta;
3497 cur_next += delta;
3498 cur_hdr->len += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003499 txn->req.eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003500
3501 del_cookie = del_colon = NULL;
3502 app_cookies++; /* protect the header from deletion */
3503 }
3504 else if (del_cookie == NULL &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003505 (t->be->options & (PR_O_COOK_INS | PR_O_COOK_IND)) == (PR_O_COOK_INS | PR_O_COOK_IND)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003506 del_cookie = p1;
3507 del_colon = colon;
3508 }
3509 } else {
3510 /* now we know that we must keep this cookie since it's
3511 * not ours. But if we wanted to delete our cookie
3512 * earlier, we cannot remove the complete header, but we
3513 * can remove the previous block itself.
3514 */
3515 app_cookies++;
3516
3517 if (del_cookie != NULL) {
3518 int delta; /* negative */
3519
3520 delta = buffer_replace2(req, del_cookie, p1, NULL, 0);
3521 p4 += delta;
3522 cur_end += delta;
3523 cur_next += delta;
3524 cur_hdr->len += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003525 txn->req.eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003526 del_cookie = del_colon = NULL;
3527 }
3528 }
3529
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003530 if ((t->be->appsession_name != NULL) &&
3531 (memcmp(p1, t->be->appsession_name, p2 - p1) == 0)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003532 /* first, let's see if the cookie is our appcookie*/
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003533
Willy Tarreau58f10d72006-12-04 02:26:12 +01003534 /* Cool... it's the right one */
3535
3536 asession_temp = &local_asession;
3537
Willy Tarreau63963c62007-05-13 21:29:55 +02003538 if ((asession_temp->sessid = pool_alloc2(apools.sessid)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003539 Alert("Not enough memory process_cli():asession->sessid:malloc().\n");
3540 send_log(t->be, LOG_ALERT, "Not enough memory process_cli():asession->sessid:malloc().\n");
3541 return;
3542 }
3543
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003544 memcpy(asession_temp->sessid, p3, t->be->appsession_len);
3545 asession_temp->sessid[t->be->appsession_len] = 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003546 asession_temp->serverid = NULL;
Willy Tarreau51041c72007-09-09 21:56:53 +02003547
Willy Tarreau58f10d72006-12-04 02:26:12 +01003548 /* only do insert, if lookup fails */
Willy Tarreau51041c72007-09-09 21:56:53 +02003549 asession_temp = appsession_hash_lookup(&(t->be->htbl_proxy), asession_temp->sessid);
3550 if (asession_temp == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02003551 if ((asession_temp = pool_alloc2(pool2_appsess)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003552 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02003553 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003554 Alert("Not enough memory process_cli():asession:calloc().\n");
3555 send_log(t->be, LOG_ALERT, "Not enough memory process_cli():asession:calloc().\n");
3556 return;
3557 }
3558
3559 asession_temp->sessid = local_asession.sessid;
3560 asession_temp->serverid = local_asession.serverid;
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003561 asession_temp->request_count = 0;
Willy Tarreau51041c72007-09-09 21:56:53 +02003562 appsession_hash_insert(&(t->be->htbl_proxy), asession_temp);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003563 } else {
3564 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02003565 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003566 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01003567 if (asession_temp->serverid == NULL) {
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003568 /* TODO redispatch request */
Willy Tarreau58f10d72006-12-04 02:26:12 +01003569 Alert("Found Application Session without matching server.\n");
3570 } else {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003571 struct server *srv = t->be->srv;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003572 while (srv) {
3573 if (strcmp(srv->id, asession_temp->serverid) == 0) {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003574 if (srv->state & SRV_RUNNING || t->be->options & PR_O_PERSIST) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003575 /* we found the server and it's usable */
Willy Tarreau3d300592007-03-18 18:34:41 +01003576 txn->flags &= ~TX_CK_MASK;
3577 txn->flags |= TX_CK_VALID;
3578 t->flags |= SN_DIRECT | SN_ASSIGNED;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003579 t->srv = srv;
3580 break;
3581 } else {
Willy Tarreau3d300592007-03-18 18:34:41 +01003582 txn->flags &= ~TX_CK_MASK;
3583 txn->flags |= TX_CK_DOWN;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003584 }
3585 }
3586 srv = srv->next;
3587 }/* end while(srv) */
3588 }/* end else if server == NULL */
3589
Willy Tarreau0c303ee2008-07-07 00:09:58 +02003590 asession_temp->expire = tick_add_ifset(now_ms, t->be->timeout.appsession);
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003591 asession_temp->request_count++;
3592#if defined(DEBUG_HASH)
3593 Alert("manage_client_side_cookies\n");
3594 appsession_hash_dump(&(t->be->htbl_proxy));
3595#endif
Willy Tarreau58f10d72006-12-04 02:26:12 +01003596 }/* end if ((t->proxy->appsession_name != NULL) ... */
3597 }
3598
3599 /* we'll have to look for another cookie ... */
3600 p1 = p4;
3601 } /* while (p1 < cur_end) */
3602
3603 /* There's no more cookie on this line.
3604 * We may have marked the last one(s) for deletion.
3605 * We must do this now in two ways :
3606 * - if there is no app cookie, we simply delete the header ;
3607 * - if there are app cookies, we must delete the end of the
3608 * string properly, including the colon/semi-colon before
3609 * the cookie name.
3610 */
3611 if (del_cookie != NULL) {
3612 int delta;
3613 if (app_cookies) {
3614 delta = buffer_replace2(req, del_colon, cur_end, NULL, 0);
3615 cur_end = del_colon;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003616 cur_hdr->len += delta;
3617 } else {
3618 delta = buffer_replace2(req, cur_ptr, cur_next, NULL, 0);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003619
3620 /* FIXME: this should be a separate function */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003621 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3622 txn->hdr_idx.used--;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003623 cur_hdr->len = 0;
3624 }
Willy Tarreau45e73e32006-12-17 00:05:15 +01003625 cur_next += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003626 txn->req.eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003627 }
3628
3629 /* keep the link from this header to next one */
3630 old_idx = cur_idx;
3631 } /* end of cookie processing on this header */
3632}
3633
3634
Willy Tarreaua15645d2007-03-18 16:22:39 +01003635/* Iterate the same filter through all response headers contained in <rtr>.
3636 * Returns 1 if this filter can be stopped upon return, otherwise 0.
3637 */
3638int apply_filter_to_resp_headers(struct session *t, struct buffer *rtr, struct hdr_exp *exp)
3639{
3640 char term;
3641 char *cur_ptr, *cur_end, *cur_next;
3642 int cur_idx, old_idx, last_hdr;
3643 struct http_txn *txn = &t->txn;
3644 struct hdr_idx_elem *cur_hdr;
3645 int len, delta;
3646
3647 last_hdr = 0;
3648
3649 cur_next = rtr->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
3650 old_idx = 0;
3651
3652 while (!last_hdr) {
Willy Tarreau3d300592007-03-18 18:34:41 +01003653 if (unlikely(txn->flags & TX_SVDENY))
Willy Tarreaua15645d2007-03-18 16:22:39 +01003654 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003655 else if (unlikely(txn->flags & TX_SVALLOW) &&
Willy Tarreaua15645d2007-03-18 16:22:39 +01003656 (exp->action == ACT_ALLOW ||
3657 exp->action == ACT_DENY))
3658 return 0;
3659
3660 cur_idx = txn->hdr_idx.v[old_idx].next;
3661 if (!cur_idx)
3662 break;
3663
3664 cur_hdr = &txn->hdr_idx.v[cur_idx];
3665 cur_ptr = cur_next;
3666 cur_end = cur_ptr + cur_hdr->len;
3667 cur_next = cur_end + cur_hdr->cr + 1;
3668
3669 /* Now we have one header between cur_ptr and cur_end,
3670 * and the next header starts at cur_next.
3671 */
3672
3673 /* The annoying part is that pattern matching needs
3674 * that we modify the contents to null-terminate all
3675 * strings before testing them.
3676 */
3677
3678 term = *cur_end;
3679 *cur_end = '\0';
3680
3681 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3682 switch (exp->action) {
3683 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003684 txn->flags |= TX_SVALLOW;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003685 last_hdr = 1;
3686 break;
3687
3688 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003689 txn->flags |= TX_SVDENY;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003690 last_hdr = 1;
3691 break;
3692
3693 case ACT_REPLACE:
3694 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3695 delta = buffer_replace2(rtr, cur_ptr, cur_end, trash, len);
3696 /* FIXME: if the user adds a newline in the replacement, the
3697 * index will not be recalculated for now, and the new line
3698 * will not be counted as a new header.
3699 */
3700
3701 cur_end += delta;
3702 cur_next += delta;
3703 cur_hdr->len += delta;
3704 txn->rsp.eoh += delta;
3705 break;
3706
3707 case ACT_REMOVE:
3708 delta = buffer_replace2(rtr, cur_ptr, cur_next, NULL, 0);
3709 cur_next += delta;
3710
3711 /* FIXME: this should be a separate function */
3712 txn->rsp.eoh += delta;
3713 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3714 txn->hdr_idx.used--;
3715 cur_hdr->len = 0;
3716 cur_end = NULL; /* null-term has been rewritten */
3717 break;
3718
3719 }
3720 }
3721 if (cur_end)
3722 *cur_end = term; /* restore the string terminator */
3723
3724 /* keep the link from this header to next one in case of later
3725 * removal of next header.
3726 */
3727 old_idx = cur_idx;
3728 }
3729 return 0;
3730}
3731
3732
3733/* Apply the filter to the status line in the response buffer <rtr>.
3734 * Returns 0 if nothing has been done, 1 if the filter has been applied,
3735 * or -1 if a replacement resulted in an invalid status line.
3736 */
3737int apply_filter_to_sts_line(struct session *t, struct buffer *rtr, struct hdr_exp *exp)
3738{
3739 char term;
3740 char *cur_ptr, *cur_end;
3741 int done;
3742 struct http_txn *txn = &t->txn;
3743 int len, delta;
3744
3745
Willy Tarreau3d300592007-03-18 18:34:41 +01003746 if (unlikely(txn->flags & TX_SVDENY))
Willy Tarreaua15645d2007-03-18 16:22:39 +01003747 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003748 else if (unlikely(txn->flags & TX_SVALLOW) &&
Willy Tarreaua15645d2007-03-18 16:22:39 +01003749 (exp->action == ACT_ALLOW ||
3750 exp->action == ACT_DENY))
3751 return 0;
3752 else if (exp->action == ACT_REMOVE)
3753 return 0;
3754
3755 done = 0;
3756
Willy Tarreau9cdde232007-05-02 20:58:19 +02003757 cur_ptr = rtr->data + txn->rsp.som; /* should be equal to txn->sol */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003758 cur_end = cur_ptr + txn->rsp.sl.rq.l;
3759
3760 /* Now we have the status line between cur_ptr and cur_end */
3761
3762 /* The annoying part is that pattern matching needs
3763 * that we modify the contents to null-terminate all
3764 * strings before testing them.
3765 */
3766
3767 term = *cur_end;
3768 *cur_end = '\0';
3769
3770 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3771 switch (exp->action) {
3772 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003773 txn->flags |= TX_SVALLOW;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003774 done = 1;
3775 break;
3776
3777 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003778 txn->flags |= TX_SVDENY;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003779 done = 1;
3780 break;
3781
3782 case ACT_REPLACE:
3783 *cur_end = term; /* restore the string terminator */
3784 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3785 delta = buffer_replace2(rtr, cur_ptr, cur_end, trash, len);
3786 /* FIXME: if the user adds a newline in the replacement, the
3787 * index will not be recalculated for now, and the new line
3788 * will not be counted as a new header.
3789 */
3790
3791 txn->rsp.eoh += delta;
3792 cur_end += delta;
3793
Willy Tarreau9cdde232007-05-02 20:58:19 +02003794 txn->rsp.sol = rtr->data + txn->rsp.som; /* should be equal to txn->sol */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003795 cur_end = (char *)http_parse_stsline(&txn->rsp, rtr->data,
Willy Tarreau02785762007-04-03 14:45:44 +02003796 HTTP_MSG_RPVER,
Willy Tarreaua15645d2007-03-18 16:22:39 +01003797 cur_ptr, cur_end + 1,
3798 NULL, NULL);
3799 if (unlikely(!cur_end))
3800 return -1;
3801
3802 /* we have a full respnse and we know that we have either a CR
3803 * or an LF at <ptr>.
3804 */
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003805 txn->status = strl2ui(rtr->data + txn->rsp.sl.st.c, txn->rsp.sl.st.c_l);
Willy Tarreaua15645d2007-03-18 16:22:39 +01003806 hdr_idx_set_start(&txn->hdr_idx, txn->rsp.sl.rq.l, *cur_end == '\r');
3807 /* there is no point trying this regex on headers */
3808 return 1;
3809 }
3810 }
3811 *cur_end = term; /* restore the string terminator */
3812 return done;
3813}
3814
3815
3816
3817/*
3818 * Apply all the resp filters <exp> to all headers in buffer <rtr> of session <t>.
3819 * Returns 0 if everything is alright, or -1 in case a replacement lead to an
3820 * unparsable response.
3821 */
3822int apply_filters_to_response(struct session *t, struct buffer *rtr, struct hdr_exp *exp)
3823{
Willy Tarreau3d300592007-03-18 18:34:41 +01003824 struct http_txn *txn = &t->txn;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003825 /* iterate through the filters in the outer loop */
Willy Tarreau3d300592007-03-18 18:34:41 +01003826 while (exp && !(txn->flags & TX_SVDENY)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003827 int ret;
3828
3829 /*
3830 * The interleaving of transformations and verdicts
3831 * makes it difficult to decide to continue or stop
3832 * the evaluation.
3833 */
3834
Willy Tarreau3d300592007-03-18 18:34:41 +01003835 if ((txn->flags & TX_SVALLOW) &&
Willy Tarreaua15645d2007-03-18 16:22:39 +01003836 (exp->action == ACT_ALLOW || exp->action == ACT_DENY ||
3837 exp->action == ACT_PASS)) {
3838 exp = exp->next;
3839 continue;
3840 }
3841
3842 /* Apply the filter to the status line. */
3843 ret = apply_filter_to_sts_line(t, rtr, exp);
3844 if (unlikely(ret < 0))
3845 return -1;
3846
3847 if (likely(ret == 0)) {
3848 /* The filter did not match the response, it can be
3849 * iterated through all headers.
3850 */
3851 apply_filter_to_resp_headers(t, rtr, exp);
3852 }
3853 exp = exp->next;
3854 }
3855 return 0;
3856}
3857
3858
3859
3860/*
Willy Tarreau396d2c62007-11-04 19:30:00 +01003861 * Manage server-side cookies. It can impact performance by about 2% so it is
3862 * desirable to call it only when needed.
Willy Tarreaua15645d2007-03-18 16:22:39 +01003863 */
3864void manage_server_side_cookies(struct session *t, struct buffer *rtr)
3865{
3866 struct http_txn *txn = &t->txn;
3867 char *p1, *p2, *p3, *p4;
3868
3869 appsess *asession_temp = NULL;
3870 appsess local_asession;
3871
3872 char *cur_ptr, *cur_end, *cur_next;
3873 int cur_idx, old_idx, delta;
3874
Willy Tarreaua15645d2007-03-18 16:22:39 +01003875 /* Iterate through the headers.
3876 * we start with the start line.
3877 */
3878 old_idx = 0;
3879 cur_next = rtr->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
3880
3881 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
3882 struct hdr_idx_elem *cur_hdr;
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003883 int val;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003884
3885 cur_hdr = &txn->hdr_idx.v[cur_idx];
3886 cur_ptr = cur_next;
3887 cur_end = cur_ptr + cur_hdr->len;
3888 cur_next = cur_end + cur_hdr->cr + 1;
3889
3890 /* We have one full header between cur_ptr and cur_end, and the
3891 * next header starts at cur_next. We're only interested in
3892 * "Cookie:" headers.
3893 */
3894
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003895 val = http_header_match2(cur_ptr, cur_end, "Set-Cookie", 10);
3896 if (!val) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003897 old_idx = cur_idx;
3898 continue;
3899 }
3900
3901 /* OK, right now we know we have a set-cookie at cur_ptr */
Willy Tarreau3d300592007-03-18 18:34:41 +01003902 txn->flags |= TX_SCK_ANY;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003903
3904
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003905 /* maybe we only wanted to see if there was a set-cookie. Note that
3906 * the cookie capture is declared in the fronend.
3907 */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003908 if (t->be->cookie_name == NULL &&
3909 t->be->appsession_name == NULL &&
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003910 t->fe->capture_name == NULL)
Willy Tarreaua15645d2007-03-18 16:22:39 +01003911 return;
3912
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003913 p1 = cur_ptr + val; /* first non-space char after 'Set-Cookie:' */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003914
3915 while (p1 < cur_end) { /* in fact, we'll break after the first cookie */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003916 if (p1 == cur_end || *p1 == ';') /* end of cookie */
3917 break;
3918
3919 /* p1 is at the beginning of the cookie name */
3920 p2 = p1;
3921
3922 while (p2 < cur_end && *p2 != '=' && *p2 != ';')
3923 p2++;
3924
3925 if (p2 == cur_end || *p2 == ';') /* next cookie */
3926 break;
3927
3928 p3 = p2 + 1; /* skip the '=' sign */
3929 if (p3 == cur_end)
3930 break;
3931
3932 p4 = p3;
Willy Tarreau8f8e6452007-06-17 21:51:38 +02003933 while (p4 < cur_end && !isspace((unsigned char)*p4) && *p4 != ';')
Willy Tarreaua15645d2007-03-18 16:22:39 +01003934 p4++;
3935
3936 /* here, we have the cookie name between p1 and p2,
3937 * and its value between p3 and p4.
3938 * we can process it.
3939 */
3940
3941 /* first, let's see if we want to capture it */
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003942 if (t->fe->capture_name != NULL &&
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003943 txn->srv_cookie == NULL &&
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003944 (p4 - p1 >= t->fe->capture_namelen) &&
3945 memcmp(p1, t->fe->capture_name, t->fe->capture_namelen) == 0) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003946 int log_len = p4 - p1;
3947
Willy Tarreau086b3b42007-05-13 21:45:51 +02003948 if ((txn->srv_cookie = pool_alloc2(pool2_capture)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003949 Alert("HTTP logging : out of memory.\n");
3950 }
3951
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003952 if (log_len > t->fe->capture_len)
3953 log_len = t->fe->capture_len;
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003954 memcpy(txn->srv_cookie, p1, log_len);
3955 txn->srv_cookie[log_len] = 0;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003956 }
3957
3958 /* now check if we need to process it for persistence */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003959 if ((p2 - p1 == t->be->cookie_len) && (t->be->cookie_name != NULL) &&
3960 (memcmp(p1, t->be->cookie_name, p2 - p1) == 0)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003961 /* Cool... it's the right one */
Willy Tarreau3d300592007-03-18 18:34:41 +01003962 txn->flags |= TX_SCK_SEEN;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003963
3964 /* If the cookie is in insert mode on a known server, we'll delete
3965 * this occurrence because we'll insert another one later.
3966 * We'll delete it too if the "indirect" option is set and we're in
3967 * a direct access. */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003968 if (((t->srv) && (t->be->options & PR_O_COOK_INS)) ||
3969 ((t->flags & SN_DIRECT) && (t->be->options & PR_O_COOK_IND))) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003970 /* this header must be deleted */
3971 delta = buffer_replace2(rtr, cur_ptr, cur_next, NULL, 0);
3972 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3973 txn->hdr_idx.used--;
3974 cur_hdr->len = 0;
3975 cur_next += delta;
3976 txn->rsp.eoh += delta;
3977
Willy Tarreau3d300592007-03-18 18:34:41 +01003978 txn->flags |= TX_SCK_DELETED;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003979 }
3980 else if ((t->srv) && (t->srv->cookie) &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003981 (t->be->options & PR_O_COOK_RW)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003982 /* replace bytes p3->p4 with the cookie name associated
3983 * with this server since we know it.
3984 */
3985 delta = buffer_replace2(rtr, p3, p4, t->srv->cookie, t->srv->cklen);
3986 cur_hdr->len += delta;
3987 cur_next += delta;
3988 txn->rsp.eoh += delta;
3989
Willy Tarreau3d300592007-03-18 18:34:41 +01003990 txn->flags |= TX_SCK_INSERTED | TX_SCK_DELETED;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003991 }
3992 else if ((t->srv) && (t->srv->cookie) &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003993 (t->be->options & PR_O_COOK_PFX)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003994 /* insert the cookie name associated with this server
3995 * before existing cookie, and insert a delimitor between them..
3996 */
3997 delta = buffer_replace2(rtr, p3, p3, t->srv->cookie, t->srv->cklen + 1);
3998 cur_hdr->len += delta;
3999 cur_next += delta;
4000 txn->rsp.eoh += delta;
4001
4002 p3[t->srv->cklen] = COOKIE_DELIM;
Willy Tarreau3d300592007-03-18 18:34:41 +01004003 txn->flags |= TX_SCK_INSERTED | TX_SCK_DELETED;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004004 }
4005 }
4006 /* next, let's see if the cookie is our appcookie */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004007 else if ((t->be->appsession_name != NULL) &&
4008 (memcmp(p1, t->be->appsession_name, p2 - p1) == 0)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004009
4010 /* Cool... it's the right one */
4011
4012 size_t server_id_len = strlen(t->srv->id) + 1;
4013 asession_temp = &local_asession;
4014
Willy Tarreau63963c62007-05-13 21:29:55 +02004015 if ((asession_temp->sessid = pool_alloc2(apools.sessid)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004016 Alert("Not enough Memory process_srv():asession->sessid:malloc().\n");
4017 send_log(t->be, LOG_ALERT, "Not enough Memory process_srv():asession->sessid:malloc().\n");
4018 return;
4019 }
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004020 memcpy(asession_temp->sessid, p3, t->be->appsession_len);
4021 asession_temp->sessid[t->be->appsession_len] = 0;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004022 asession_temp->serverid = NULL;
4023
4024 /* only do insert, if lookup fails */
Ryan Warnick6d0b1fa2008-02-17 11:24:35 +01004025 asession_temp = appsession_hash_lookup(&(t->be->htbl_proxy), asession_temp->sessid);
4026 if (asession_temp == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02004027 if ((asession_temp = pool_alloc2(pool2_appsess)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004028 Alert("Not enough Memory process_srv():asession:calloc().\n");
4029 send_log(t->be, LOG_ALERT, "Not enough Memory process_srv():asession:calloc().\n");
4030 return;
4031 }
4032 asession_temp->sessid = local_asession.sessid;
4033 asession_temp->serverid = local_asession.serverid;
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004034 asession_temp->request_count = 0;
Willy Tarreau51041c72007-09-09 21:56:53 +02004035 appsession_hash_insert(&(t->be->htbl_proxy), asession_temp);
4036 } else {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004037 /* free wasted memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02004038 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau51041c72007-09-09 21:56:53 +02004039 }
4040
Willy Tarreaua15645d2007-03-18 16:22:39 +01004041 if (asession_temp->serverid == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02004042 if ((asession_temp->serverid = pool_alloc2(apools.serverid)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004043 Alert("Not enough Memory process_srv():asession->sessid:malloc().\n");
4044 send_log(t->be, LOG_ALERT, "Not enough Memory process_srv():asession->sessid:malloc().\n");
4045 return;
4046 }
4047 asession_temp->serverid[0] = '\0';
4048 }
4049
4050 if (asession_temp->serverid[0] == '\0')
4051 memcpy(asession_temp->serverid, t->srv->id, server_id_len);
4052
Willy Tarreau0c303ee2008-07-07 00:09:58 +02004053 asession_temp->expire = tick_add_ifset(now_ms, t->be->timeout.appsession);
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004054 asession_temp->request_count++;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004055#if defined(DEBUG_HASH)
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004056 Alert("manage_server_side_cookies\n");
Willy Tarreau51041c72007-09-09 21:56:53 +02004057 appsession_hash_dump(&(t->be->htbl_proxy));
Willy Tarreaua15645d2007-03-18 16:22:39 +01004058#endif
4059 }/* end if ((t->proxy->appsession_name != NULL) ... */
4060 break; /* we don't want to loop again since there cannot be another cookie on the same line */
4061 } /* we're now at the end of the cookie value */
4062
4063 /* keep the link from this header to next one */
4064 old_idx = cur_idx;
4065 } /* end of cookie processing on this header */
4066}
4067
4068
4069
4070/*
4071 * Check if response is cacheable or not. Updates t->flags.
4072 */
4073void check_response_for_cacheability(struct session *t, struct buffer *rtr)
4074{
4075 struct http_txn *txn = &t->txn;
4076 char *p1, *p2;
4077
4078 char *cur_ptr, *cur_end, *cur_next;
4079 int cur_idx;
4080
Willy Tarreau5df51872007-11-25 16:20:08 +01004081 if (!(txn->flags & TX_CACHEABLE))
Willy Tarreaua15645d2007-03-18 16:22:39 +01004082 return;
4083
4084 /* Iterate through the headers.
4085 * we start with the start line.
4086 */
4087 cur_idx = 0;
4088 cur_next = rtr->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
4089
4090 while ((cur_idx = txn->hdr_idx.v[cur_idx].next)) {
4091 struct hdr_idx_elem *cur_hdr;
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004092 int val;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004093
4094 cur_hdr = &txn->hdr_idx.v[cur_idx];
4095 cur_ptr = cur_next;
4096 cur_end = cur_ptr + cur_hdr->len;
4097 cur_next = cur_end + cur_hdr->cr + 1;
4098
4099 /* We have one full header between cur_ptr and cur_end, and the
4100 * next header starts at cur_next. We're only interested in
4101 * "Cookie:" headers.
4102 */
4103
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004104 val = http_header_match2(cur_ptr, cur_end, "Pragma", 6);
4105 if (val) {
4106 if ((cur_end - (cur_ptr + val) >= 8) &&
4107 strncasecmp(cur_ptr + val, "no-cache", 8) == 0) {
4108 txn->flags &= ~TX_CACHEABLE & ~TX_CACHE_COOK;
4109 return;
4110 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01004111 }
4112
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004113 val = http_header_match2(cur_ptr, cur_end, "Cache-control", 13);
4114 if (!val)
Willy Tarreaua15645d2007-03-18 16:22:39 +01004115 continue;
4116
4117 /* OK, right now we know we have a cache-control header at cur_ptr */
4118
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004119 p1 = cur_ptr + val; /* first non-space char after 'cache-control:' */
Willy Tarreaua15645d2007-03-18 16:22:39 +01004120
4121 if (p1 >= cur_end) /* no more info */
4122 continue;
4123
4124 /* p1 is at the beginning of the value */
4125 p2 = p1;
4126
Willy Tarreau8f8e6452007-06-17 21:51:38 +02004127 while (p2 < cur_end && *p2 != '=' && *p2 != ',' && !isspace((unsigned char)*p2))
Willy Tarreaua15645d2007-03-18 16:22:39 +01004128 p2++;
4129
4130 /* we have a complete value between p1 and p2 */
4131 if (p2 < cur_end && *p2 == '=') {
4132 /* we have something of the form no-cache="set-cookie" */
4133 if ((cur_end - p1 >= 21) &&
4134 strncasecmp(p1, "no-cache=\"set-cookie", 20) == 0
4135 && (p1[20] == '"' || p1[20] == ','))
Willy Tarreau3d300592007-03-18 18:34:41 +01004136 txn->flags &= ~TX_CACHE_COOK;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004137 continue;
4138 }
4139
4140 /* OK, so we know that either p2 points to the end of string or to a comma */
4141 if (((p2 - p1 == 7) && strncasecmp(p1, "private", 7) == 0) ||
4142 ((p2 - p1 == 8) && strncasecmp(p1, "no-store", 8) == 0) ||
4143 ((p2 - p1 == 9) && strncasecmp(p1, "max-age=0", 9) == 0) ||
4144 ((p2 - p1 == 10) && strncasecmp(p1, "s-maxage=0", 10) == 0)) {
Willy Tarreau3d300592007-03-18 18:34:41 +01004145 txn->flags &= ~TX_CACHEABLE & ~TX_CACHE_COOK;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004146 return;
4147 }
4148
4149 if ((p2 - p1 == 6) && strncasecmp(p1, "public", 6) == 0) {
Willy Tarreau3d300592007-03-18 18:34:41 +01004150 txn->flags |= TX_CACHEABLE | TX_CACHE_COOK;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004151 continue;
4152 }
4153 }
4154}
4155
4156
Willy Tarreau58f10d72006-12-04 02:26:12 +01004157/*
4158 * Try to retrieve a known appsession in the URI, then the associated server.
4159 * If the server is found, it's assigned to the session.
4160 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004161void get_srv_from_appsession(struct session *t, const char *begin, int len)
Willy Tarreau58f10d72006-12-04 02:26:12 +01004162{
Willy Tarreau3d300592007-03-18 18:34:41 +01004163 struct http_txn *txn = &t->txn;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004164 appsess *asession_temp = NULL;
4165 appsess local_asession;
4166 char *request_line;
4167
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004168 if (t->be->appsession_name == NULL ||
Willy Tarreaub326fcc2007-03-03 13:54:32 +01004169 (t->txn.meth != HTTP_METH_GET && t->txn.meth != HTTP_METH_POST) ||
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004170 (request_line = memchr(begin, ';', len)) == NULL ||
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004171 ((1 + t->be->appsession_name_len + 1 + t->be->appsession_len) > (begin + len - request_line)))
Willy Tarreau58f10d72006-12-04 02:26:12 +01004172 return;
4173
4174 /* skip ';' */
4175 request_line++;
4176
4177 /* look if we have a jsessionid */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004178 if (strncasecmp(request_line, t->be->appsession_name, t->be->appsession_name_len) != 0)
Willy Tarreau58f10d72006-12-04 02:26:12 +01004179 return;
4180
4181 /* skip jsessionid= */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004182 request_line += t->be->appsession_name_len + 1;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004183
4184 /* First try if we already have an appsession */
4185 asession_temp = &local_asession;
4186
Willy Tarreau63963c62007-05-13 21:29:55 +02004187 if ((asession_temp->sessid = pool_alloc2(apools.sessid)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01004188 Alert("Not enough memory process_cli():asession_temp->sessid:calloc().\n");
4189 send_log(t->be, LOG_ALERT, "Not enough Memory process_cli():asession_temp->sessid:calloc().\n");
4190 return;
4191 }
4192
4193 /* Copy the sessionid */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004194 memcpy(asession_temp->sessid, request_line, t->be->appsession_len);
4195 asession_temp->sessid[t->be->appsession_len] = 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004196 asession_temp->serverid = NULL;
4197
4198 /* only do insert, if lookup fails */
Ryan Warnick6d0b1fa2008-02-17 11:24:35 +01004199 asession_temp = appsession_hash_lookup(&(t->be->htbl_proxy), asession_temp->sessid);
4200 if (asession_temp == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02004201 if ((asession_temp = pool_alloc2(pool2_appsess)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01004202 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02004203 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004204 Alert("Not enough memory process_cli():asession:calloc().\n");
4205 send_log(t->be, LOG_ALERT, "Not enough memory process_cli():asession:calloc().\n");
4206 return;
4207 }
4208 asession_temp->sessid = local_asession.sessid;
4209 asession_temp->serverid = local_asession.serverid;
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004210 asession_temp->request_count=0;
Willy Tarreau51041c72007-09-09 21:56:53 +02004211 appsession_hash_insert(&(t->be->htbl_proxy), asession_temp);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004212 }
4213 else {
4214 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02004215 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004216 }
Willy Tarreau51041c72007-09-09 21:56:53 +02004217
Willy Tarreau0c303ee2008-07-07 00:09:58 +02004218 asession_temp->expire = tick_add_ifset(now_ms, t->be->timeout.appsession);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004219 asession_temp->request_count++;
Willy Tarreau51041c72007-09-09 21:56:53 +02004220
Willy Tarreau58f10d72006-12-04 02:26:12 +01004221#if defined(DEBUG_HASH)
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004222 Alert("get_srv_from_appsession\n");
Willy Tarreau51041c72007-09-09 21:56:53 +02004223 appsession_hash_dump(&(t->be->htbl_proxy));
Willy Tarreau58f10d72006-12-04 02:26:12 +01004224#endif
4225 if (asession_temp->serverid == NULL) {
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004226 /* TODO redispatch request */
Willy Tarreau58f10d72006-12-04 02:26:12 +01004227 Alert("Found Application Session without matching server.\n");
4228 } else {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004229 struct server *srv = t->be->srv;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004230 while (srv) {
4231 if (strcmp(srv->id, asession_temp->serverid) == 0) {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004232 if (srv->state & SRV_RUNNING || t->be->options & PR_O_PERSIST) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01004233 /* we found the server and it's usable */
Willy Tarreau3d300592007-03-18 18:34:41 +01004234 txn->flags &= ~TX_CK_MASK;
4235 txn->flags |= TX_CK_VALID;
4236 t->flags |= SN_DIRECT | SN_ASSIGNED;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004237 t->srv = srv;
4238 break;
4239 } else {
Willy Tarreau3d300592007-03-18 18:34:41 +01004240 txn->flags &= ~TX_CK_MASK;
4241 txn->flags |= TX_CK_DOWN;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004242 }
4243 }
4244 srv = srv->next;
4245 }
4246 }
4247}
4248
4249
Willy Tarreaub2513902006-12-17 14:52:38 +01004250/*
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004251 * In a GET or HEAD request, check if the requested URI matches the stats uri
4252 * for the current backend, and if an authorization has been passed and is valid.
Willy Tarreaub2513902006-12-17 14:52:38 +01004253 *
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004254 * It is assumed that the request is either a HEAD or GET and that the
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004255 * t->be->uri_auth field is valid. An HTTP/401 response may be sent, or
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004256 * produce_content() can be called to start sending data.
Willy Tarreaub2513902006-12-17 14:52:38 +01004257 *
4258 * Returns 1 if the session's state changes, otherwise 0.
4259 */
4260int stats_check_uri_auth(struct session *t, struct proxy *backend)
4261{
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004262 struct http_txn *txn = &t->txn;
Willy Tarreaub2513902006-12-17 14:52:38 +01004263 struct uri_auth *uri_auth = backend->uri_auth;
4264 struct user_auth *user;
4265 int authenticated, cur_idx;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004266 char *h;
Willy Tarreaub2513902006-12-17 14:52:38 +01004267
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004268 memset(&t->data_ctx.stats, 0, sizeof(t->data_ctx.stats));
4269
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004270 /* check URI size */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004271 if (uri_auth->uri_len > txn->req.sl.rq.u_l)
Willy Tarreaub2513902006-12-17 14:52:38 +01004272 return 0;
4273
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004274 h = t->req->data + txn->req.sl.rq.u;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004275
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004276 /* the URI is in h */
4277 if (memcmp(h, uri_auth->uri_prefix, uri_auth->uri_len) != 0)
Willy Tarreaub2513902006-12-17 14:52:38 +01004278 return 0;
4279
Willy Tarreaue7150cd2007-07-25 14:43:32 +02004280 h += uri_auth->uri_len;
4281 while (h <= t->req->data + txn->req.sl.rq.u + txn->req.sl.rq.u_l - 3) {
4282 if (memcmp(h, ";up", 3) == 0) {
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004283 t->data_ctx.stats.flags |= STAT_HIDE_DOWN;
Willy Tarreaue7150cd2007-07-25 14:43:32 +02004284 break;
4285 }
4286 h++;
4287 }
4288
4289 if (uri_auth->refresh) {
4290 h = t->req->data + txn->req.sl.rq.u + uri_auth->uri_len;
4291 while (h <= t->req->data + txn->req.sl.rq.u + txn->req.sl.rq.u_l - 10) {
4292 if (memcmp(h, ";norefresh", 10) == 0) {
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004293 t->data_ctx.stats.flags |= STAT_NO_REFRESH;
Willy Tarreaue7150cd2007-07-25 14:43:32 +02004294 break;
4295 }
4296 h++;
4297 }
4298 }
4299
Willy Tarreau55bb8452007-10-17 18:44:57 +02004300 h = t->req->data + txn->req.sl.rq.u + uri_auth->uri_len;
4301 while (h <= t->req->data + txn->req.sl.rq.u + txn->req.sl.rq.u_l - 4) {
4302 if (memcmp(h, ";csv", 4) == 0) {
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004303 t->data_ctx.stats.flags |= STAT_FMT_CSV;
Willy Tarreau55bb8452007-10-17 18:44:57 +02004304 break;
4305 }
4306 h++;
4307 }
4308
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004309 t->data_ctx.stats.flags |= STAT_SHOW_STAT | STAT_SHOW_INFO;
4310
Willy Tarreaub2513902006-12-17 14:52:38 +01004311 /* we are in front of a interceptable URI. Let's check
4312 * if there's an authentication and if it's valid.
4313 */
4314 user = uri_auth->users;
4315 if (!user) {
4316 /* no user auth required, it's OK */
4317 authenticated = 1;
4318 } else {
4319 authenticated = 0;
4320
4321 /* a user list is defined, we have to check.
4322 * skip 21 chars for "Authorization: Basic ".
4323 */
4324
4325 /* FIXME: this should move to an earlier place */
4326 cur_idx = 0;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004327 h = t->req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
4328 while ((cur_idx = txn->hdr_idx.v[cur_idx].next)) {
4329 int len = txn->hdr_idx.v[cur_idx].len;
Willy Tarreaub2513902006-12-17 14:52:38 +01004330 if (len > 14 &&
4331 !strncasecmp("Authorization:", h, 14)) {
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004332 txn->auth_hdr.str = h;
4333 txn->auth_hdr.len = len;
Willy Tarreaub2513902006-12-17 14:52:38 +01004334 break;
4335 }
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004336 h += len + txn->hdr_idx.v[cur_idx].cr + 1;
Willy Tarreaub2513902006-12-17 14:52:38 +01004337 }
4338
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004339 if (txn->auth_hdr.len < 21 ||
4340 memcmp(txn->auth_hdr.str + 14, " Basic ", 7))
Willy Tarreaub2513902006-12-17 14:52:38 +01004341 user = NULL;
4342
4343 while (user) {
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004344 if ((txn->auth_hdr.len == user->user_len + 14 + 7)
4345 && !memcmp(txn->auth_hdr.str + 14 + 7,
Willy Tarreaub2513902006-12-17 14:52:38 +01004346 user->user_pwd, user->user_len)) {
4347 authenticated = 1;
4348 break;
4349 }
4350 user = user->next;
4351 }
4352 }
4353
4354 if (!authenticated) {
Willy Tarreau0f772532006-12-23 20:51:41 +01004355 struct chunk msg;
Willy Tarreaub2513902006-12-17 14:52:38 +01004356
4357 /* no need to go further */
Willy Tarreau0f772532006-12-23 20:51:41 +01004358 msg.str = trash;
4359 msg.len = sprintf(trash, HTTP_401_fmt, uri_auth->auth_realm);
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01004360 txn->status = 401;
Willy Tarreaudded32d2008-11-30 19:48:07 +01004361 stream_int_retnclose(t->req->prod, &msg);
Willy Tarreau2df28e82008-08-17 15:20:19 +02004362 t->req->analysers = 0;
Willy Tarreaub2513902006-12-17 14:52:38 +01004363 if (!(t->flags & SN_ERR_MASK))
4364 t->flags |= SN_ERR_PRXCOND;
4365 if (!(t->flags & SN_FINST_MASK))
4366 t->flags |= SN_FINST_R;
4367 return 1;
4368 }
4369
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004370 /* The request is valid, the user is authenticated. Let's start sending
Willy Tarreaub2513902006-12-17 14:52:38 +01004371 * data.
4372 */
Willy Tarreauefb453c2008-10-26 20:49:47 +01004373 buffer_write_dis(t->req);
Willy Tarreau72b179a2008-08-28 16:01:32 +02004374 buffer_shutw_now(t->req);
4375 buffer_shutr_now(t->rep);
Willy Tarreau70089872008-06-13 21:12:51 +02004376 t->logs.tv_request = now;
Willy Tarreaub2513902006-12-17 14:52:38 +01004377 t->data_source = DATA_SRC_STATS;
4378 t->data_state = DATA_ST_INIT;
Willy Tarreau91e99932008-06-30 07:51:00 +02004379 t->task->nice = -32; /* small boost for HTTP statistics */
Willy Tarreau01bf8672008-12-07 18:03:29 +01004380 buffer_install_hijacker(t, t->rep, produce_content);
Willy Tarreaub2513902006-12-17 14:52:38 +01004381 return 1;
4382}
4383
4384
Willy Tarreaubaaee002006-06-26 02:48:02 +02004385/*
Willy Tarreau58f10d72006-12-04 02:26:12 +01004386 * Print a debug line with a header
4387 */
4388void debug_hdr(const char *dir, struct session *t, const char *start, const char *end)
4389{
4390 int len, max;
4391 len = sprintf(trash, "%08x:%s.%s[%04x:%04x]: ", t->uniq_id, t->be->id,
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02004392 dir, (unsigned short)t->req->prod->fd, (unsigned short)t->req->cons->fd);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004393 max = end - start;
4394 UBOUND(max, sizeof(trash) - len - 1);
4395 len += strlcpy2(trash + len, start, max + 1);
4396 trash[len++] = '\n';
4397 write(1, trash, len);
4398}
4399
4400
Willy Tarreau8797c062007-05-07 00:55:35 +02004401/************************************************************************/
4402/* The code below is dedicated to ACL parsing and matching */
4403/************************************************************************/
4404
4405
4406
4407
4408/* 1. Check on METHOD
4409 * We use the pre-parsed method if it is known, and store its number as an
4410 * integer. If it is unknown, we use the pointer and the length.
4411 */
Willy Tarreauae8b7962007-06-09 23:10:04 +02004412static int acl_parse_meth(const char **text, struct acl_pattern *pattern, int *opaque)
Willy Tarreau8797c062007-05-07 00:55:35 +02004413{
4414 int len, meth;
4415
Willy Tarreauae8b7962007-06-09 23:10:04 +02004416 len = strlen(*text);
4417 meth = find_http_meth(*text, len);
Willy Tarreau8797c062007-05-07 00:55:35 +02004418
4419 pattern->val.i = meth;
4420 if (meth == HTTP_METH_OTHER) {
Willy Tarreauae8b7962007-06-09 23:10:04 +02004421 pattern->ptr.str = strdup(*text);
Willy Tarreau8797c062007-05-07 00:55:35 +02004422 if (!pattern->ptr.str)
4423 return 0;
4424 pattern->len = len;
4425 }
4426 return 1;
4427}
4428
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004429static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004430acl_fetch_meth(struct proxy *px, struct session *l4, void *l7, int dir,
4431 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004432{
4433 int meth;
4434 struct http_txn *txn = l7;
4435
Willy Tarreaub6866442008-07-14 23:54:42 +02004436 if (!txn)
4437 return 0;
4438
Willy Tarreauc11416f2007-06-17 16:58:38 +02004439 if (txn->req.msg_state != HTTP_MSG_BODY)
4440 return 0;
4441
Willy Tarreau8797c062007-05-07 00:55:35 +02004442 meth = txn->meth;
4443 test->i = meth;
4444 if (meth == HTTP_METH_OTHER) {
Willy Tarreauc11416f2007-06-17 16:58:38 +02004445 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4446 /* ensure the indexes are not affected */
4447 return 0;
Willy Tarreau8797c062007-05-07 00:55:35 +02004448 test->len = txn->req.sl.rq.m_l;
4449 test->ptr = txn->req.sol;
4450 }
4451 test->flags = ACL_TEST_F_READ_ONLY | ACL_TEST_F_VOL_1ST;
4452 return 1;
4453}
4454
4455static int acl_match_meth(struct acl_test *test, struct acl_pattern *pattern)
4456{
Willy Tarreauc8d7c962007-06-17 08:20:33 +02004457 int icase;
4458
Willy Tarreau8797c062007-05-07 00:55:35 +02004459 if (test->i != pattern->val.i)
Willy Tarreau11382812008-07-09 16:18:21 +02004460 return ACL_PAT_FAIL;
Willy Tarreau8797c062007-05-07 00:55:35 +02004461
4462 if (test->i != HTTP_METH_OTHER)
Willy Tarreau11382812008-07-09 16:18:21 +02004463 return ACL_PAT_PASS;
Willy Tarreau8797c062007-05-07 00:55:35 +02004464
4465 /* Other method, we must compare the strings */
4466 if (pattern->len != test->len)
Willy Tarreau11382812008-07-09 16:18:21 +02004467 return ACL_PAT_FAIL;
Willy Tarreauc8d7c962007-06-17 08:20:33 +02004468
4469 icase = pattern->flags & ACL_PAT_F_IGNORE_CASE;
4470 if ((icase && strncasecmp(pattern->ptr.str, test->ptr, test->len) != 0) ||
4471 (!icase && strncmp(pattern->ptr.str, test->ptr, test->len) != 0))
Willy Tarreau11382812008-07-09 16:18:21 +02004472 return ACL_PAT_FAIL;
4473 return ACL_PAT_PASS;
Willy Tarreau8797c062007-05-07 00:55:35 +02004474}
4475
4476/* 2. Check on Request/Status Version
4477 * We simply compare strings here.
4478 */
Willy Tarreauae8b7962007-06-09 23:10:04 +02004479static int acl_parse_ver(const char **text, struct acl_pattern *pattern, int *opaque)
Willy Tarreau8797c062007-05-07 00:55:35 +02004480{
Willy Tarreauae8b7962007-06-09 23:10:04 +02004481 pattern->ptr.str = strdup(*text);
Willy Tarreau8797c062007-05-07 00:55:35 +02004482 if (!pattern->ptr.str)
4483 return 0;
Willy Tarreauae8b7962007-06-09 23:10:04 +02004484 pattern->len = strlen(*text);
Willy Tarreau8797c062007-05-07 00:55:35 +02004485 return 1;
4486}
4487
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004488static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004489acl_fetch_rqver(struct proxy *px, struct session *l4, void *l7, int dir,
4490 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004491{
4492 struct http_txn *txn = l7;
4493 char *ptr;
4494 int len;
4495
Willy Tarreaub6866442008-07-14 23:54:42 +02004496 if (!txn)
4497 return 0;
4498
Willy Tarreauc11416f2007-06-17 16:58:38 +02004499 if (txn->req.msg_state != HTTP_MSG_BODY)
4500 return 0;
4501
Willy Tarreau8797c062007-05-07 00:55:35 +02004502 len = txn->req.sl.rq.v_l;
4503 ptr = txn->req.sol + txn->req.sl.rq.v - txn->req.som;
4504
4505 while ((len-- > 0) && (*ptr++ != '/'));
4506 if (len <= 0)
4507 return 0;
4508
4509 test->ptr = ptr;
4510 test->len = len;
4511
4512 test->flags = ACL_TEST_F_READ_ONLY | ACL_TEST_F_VOL_1ST;
4513 return 1;
4514}
4515
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004516static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004517acl_fetch_stver(struct proxy *px, struct session *l4, void *l7, int dir,
4518 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004519{
4520 struct http_txn *txn = l7;
4521 char *ptr;
4522 int len;
4523
Willy Tarreaub6866442008-07-14 23:54:42 +02004524 if (!txn)
4525 return 0;
4526
Willy Tarreauc11416f2007-06-17 16:58:38 +02004527 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4528 return 0;
4529
Willy Tarreau8797c062007-05-07 00:55:35 +02004530 len = txn->rsp.sl.st.v_l;
4531 ptr = txn->rsp.sol;
4532
4533 while ((len-- > 0) && (*ptr++ != '/'));
4534 if (len <= 0)
4535 return 0;
4536
4537 test->ptr = ptr;
4538 test->len = len;
4539
4540 test->flags = ACL_TEST_F_READ_ONLY | ACL_TEST_F_VOL_1ST;
4541 return 1;
4542}
4543
4544/* 3. Check on Status Code. We manipulate integers here. */
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004545static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004546acl_fetch_stcode(struct proxy *px, struct session *l4, void *l7, int dir,
4547 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004548{
4549 struct http_txn *txn = l7;
4550 char *ptr;
4551 int len;
4552
Willy Tarreaub6866442008-07-14 23:54:42 +02004553 if (!txn)
4554 return 0;
4555
Willy Tarreauc11416f2007-06-17 16:58:38 +02004556 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4557 return 0;
4558
Willy Tarreau8797c062007-05-07 00:55:35 +02004559 len = txn->rsp.sl.st.c_l;
4560 ptr = txn->rsp.sol + txn->rsp.sl.st.c - txn->rsp.som;
4561
4562 test->i = __strl2ui(ptr, len);
4563 test->flags = ACL_TEST_F_VOL_1ST;
4564 return 1;
4565}
4566
4567/* 4. Check on URL/URI. A pointer to the URI is stored. */
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004568static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004569acl_fetch_url(struct proxy *px, struct session *l4, void *l7, int dir,
4570 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004571{
4572 struct http_txn *txn = l7;
4573
Willy Tarreaub6866442008-07-14 23:54:42 +02004574 if (!txn)
4575 return 0;
4576
Willy Tarreauc11416f2007-06-17 16:58:38 +02004577 if (txn->req.msg_state != HTTP_MSG_BODY)
4578 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004579
Willy Tarreauc11416f2007-06-17 16:58:38 +02004580 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4581 /* ensure the indexes are not affected */
4582 return 0;
4583
Willy Tarreau8797c062007-05-07 00:55:35 +02004584 test->len = txn->req.sl.rq.u_l;
4585 test->ptr = txn->req.sol + txn->req.sl.rq.u;
4586
Willy Tarreauf3d25982007-05-08 22:45:09 +02004587 /* we do not need to set READ_ONLY because the data is in a buffer */
4588 test->flags = ACL_TEST_F_VOL_1ST;
Willy Tarreau8797c062007-05-07 00:55:35 +02004589 return 1;
4590}
4591
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004592static int
4593acl_fetch_url_ip(struct proxy *px, struct session *l4, void *l7, int dir,
4594 struct acl_expr *expr, struct acl_test *test)
4595{
4596 struct http_txn *txn = l7;
4597
Willy Tarreaub6866442008-07-14 23:54:42 +02004598 if (!txn)
4599 return 0;
4600
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004601 if (txn->req.msg_state != HTTP_MSG_BODY)
4602 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004603
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004604 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4605 /* ensure the indexes are not affected */
4606 return 0;
4607
4608 /* Parse HTTP request */
4609 url2sa(txn->req.sol + txn->req.sl.rq.u, txn->req.sl.rq.u_l, &l4->srv_addr);
4610 test->ptr = (void *)&((struct sockaddr_in *)&l4->srv_addr)->sin_addr;
4611 test->i = AF_INET;
4612
4613 /*
4614 * If we are parsing url in frontend space, we prepare backend stage
4615 * to not parse again the same url ! optimization lazyness...
4616 */
4617 if (px->options & PR_O_HTTP_PROXY)
4618 l4->flags |= SN_ADDR_SET;
4619
4620 test->flags = ACL_TEST_F_READ_ONLY;
4621 return 1;
4622}
4623
4624static int
4625acl_fetch_url_port(struct proxy *px, struct session *l4, void *l7, int dir,
4626 struct acl_expr *expr, struct acl_test *test)
4627{
4628 struct http_txn *txn = l7;
4629
Willy Tarreaub6866442008-07-14 23:54:42 +02004630 if (!txn)
4631 return 0;
4632
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004633 if (txn->req.msg_state != HTTP_MSG_BODY)
4634 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004635
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004636 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4637 /* ensure the indexes are not affected */
4638 return 0;
4639
4640 /* Same optimization as url_ip */
4641 url2sa(txn->req.sol + txn->req.sl.rq.u, txn->req.sl.rq.u_l, &l4->srv_addr);
4642 test->i = ntohs(((struct sockaddr_in *)&l4->srv_addr)->sin_port);
4643
4644 if (px->options & PR_O_HTTP_PROXY)
4645 l4->flags |= SN_ADDR_SET;
4646
4647 test->flags = ACL_TEST_F_READ_ONLY;
4648 return 1;
4649}
4650
Willy Tarreauc11416f2007-06-17 16:58:38 +02004651/* 5. Check on HTTP header. A pointer to the beginning of the value is returned.
4652 * This generic function is used by both acl_fetch_chdr() and acl_fetch_shdr().
4653 */
Willy Tarreau33a7e692007-06-10 19:45:56 +02004654static int
Willy Tarreauc11416f2007-06-17 16:58:38 +02004655acl_fetch_hdr(struct proxy *px, struct session *l4, void *l7, char *sol,
Willy Tarreau33a7e692007-06-10 19:45:56 +02004656 struct acl_expr *expr, struct acl_test *test)
4657{
4658 struct http_txn *txn = l7;
4659 struct hdr_idx *idx = &txn->hdr_idx;
4660 struct hdr_ctx *ctx = (struct hdr_ctx *)test->ctx.a;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004661
Willy Tarreaub6866442008-07-14 23:54:42 +02004662 if (!txn)
4663 return 0;
4664
Willy Tarreau33a7e692007-06-10 19:45:56 +02004665 if (!(test->flags & ACL_TEST_F_FETCH_MORE))
4666 /* search for header from the beginning */
4667 ctx->idx = 0;
4668
Willy Tarreau33a7e692007-06-10 19:45:56 +02004669 if (http_find_header2(expr->arg.str, expr->arg_len, sol, idx, ctx)) {
4670 test->flags |= ACL_TEST_F_FETCH_MORE;
4671 test->flags |= ACL_TEST_F_VOL_HDR;
4672 test->len = ctx->vlen;
4673 test->ptr = (char *)ctx->line + ctx->val;
4674 return 1;
4675 }
4676
4677 test->flags &= ~ACL_TEST_F_FETCH_MORE;
4678 test->flags |= ACL_TEST_F_VOL_HDR;
4679 return 0;
4680}
4681
Willy Tarreau33a7e692007-06-10 19:45:56 +02004682static int
Willy Tarreauc11416f2007-06-17 16:58:38 +02004683acl_fetch_chdr(struct proxy *px, struct session *l4, void *l7, int dir,
4684 struct acl_expr *expr, struct acl_test *test)
4685{
4686 struct http_txn *txn = l7;
4687
Willy Tarreaub6866442008-07-14 23:54:42 +02004688 if (!txn)
4689 return 0;
4690
Willy Tarreauc11416f2007-06-17 16:58:38 +02004691 if (txn->req.msg_state != HTTP_MSG_BODY)
4692 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004693
Willy Tarreauc11416f2007-06-17 16:58:38 +02004694 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4695 /* ensure the indexes are not affected */
4696 return 0;
4697
4698 return acl_fetch_hdr(px, l4, txn, txn->req.sol, expr, test);
4699}
4700
4701static int
4702acl_fetch_shdr(struct proxy *px, struct session *l4, void *l7, int dir,
4703 struct acl_expr *expr, struct acl_test *test)
4704{
4705 struct http_txn *txn = l7;
4706
Willy Tarreaub6866442008-07-14 23:54:42 +02004707 if (!txn)
4708 return 0;
4709
Willy Tarreauc11416f2007-06-17 16:58:38 +02004710 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4711 return 0;
4712
4713 return acl_fetch_hdr(px, l4, txn, txn->rsp.sol, expr, test);
4714}
4715
4716/* 6. Check on HTTP header count. The number of occurrences is returned.
4717 * This generic function is used by both acl_fetch_chdr* and acl_fetch_shdr*.
4718 */
4719static int
4720acl_fetch_hdr_cnt(struct proxy *px, struct session *l4, void *l7, char *sol,
Willy Tarreau33a7e692007-06-10 19:45:56 +02004721 struct acl_expr *expr, struct acl_test *test)
4722{
4723 struct http_txn *txn = l7;
4724 struct hdr_idx *idx = &txn->hdr_idx;
4725 struct hdr_ctx ctx;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004726 int cnt;
Willy Tarreau8797c062007-05-07 00:55:35 +02004727
Willy Tarreaub6866442008-07-14 23:54:42 +02004728 if (!txn)
4729 return 0;
4730
Willy Tarreau33a7e692007-06-10 19:45:56 +02004731 ctx.idx = 0;
4732 cnt = 0;
4733 while (http_find_header2(expr->arg.str, expr->arg_len, sol, idx, &ctx))
4734 cnt++;
4735
4736 test->i = cnt;
4737 test->flags = ACL_TEST_F_VOL_HDR;
4738 return 1;
4739}
4740
Willy Tarreauc11416f2007-06-17 16:58:38 +02004741static int
4742acl_fetch_chdr_cnt(struct proxy *px, struct session *l4, void *l7, int dir,
4743 struct acl_expr *expr, struct acl_test *test)
4744{
4745 struct http_txn *txn = l7;
4746
Willy Tarreaub6866442008-07-14 23:54:42 +02004747 if (!txn)
4748 return 0;
4749
Willy Tarreauc11416f2007-06-17 16:58:38 +02004750 if (txn->req.msg_state != HTTP_MSG_BODY)
4751 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004752
Willy Tarreauc11416f2007-06-17 16:58:38 +02004753 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4754 /* ensure the indexes are not affected */
4755 return 0;
4756
4757 return acl_fetch_hdr_cnt(px, l4, txn, txn->req.sol, expr, test);
4758}
4759
4760static int
4761acl_fetch_shdr_cnt(struct proxy *px, struct session *l4, void *l7, int dir,
4762 struct acl_expr *expr, struct acl_test *test)
4763{
4764 struct http_txn *txn = l7;
4765
Willy Tarreaub6866442008-07-14 23:54:42 +02004766 if (!txn)
4767 return 0;
4768
Willy Tarreauc11416f2007-06-17 16:58:38 +02004769 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4770 return 0;
4771
4772 return acl_fetch_hdr_cnt(px, l4, txn, txn->rsp.sol, expr, test);
4773}
4774
Willy Tarreau33a7e692007-06-10 19:45:56 +02004775/* 7. Check on HTTP header's integer value. The integer value is returned.
4776 * FIXME: the type is 'int', it may not be appropriate for everything.
Willy Tarreauc11416f2007-06-17 16:58:38 +02004777 * This generic function is used by both acl_fetch_chdr* and acl_fetch_shdr*.
Willy Tarreau33a7e692007-06-10 19:45:56 +02004778 */
4779static int
Willy Tarreauc11416f2007-06-17 16:58:38 +02004780acl_fetch_hdr_val(struct proxy *px, struct session *l4, void *l7, char *sol,
Willy Tarreau33a7e692007-06-10 19:45:56 +02004781 struct acl_expr *expr, struct acl_test *test)
4782{
4783 struct http_txn *txn = l7;
4784 struct hdr_idx *idx = &txn->hdr_idx;
4785 struct hdr_ctx *ctx = (struct hdr_ctx *)test->ctx.a;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004786
Willy Tarreaub6866442008-07-14 23:54:42 +02004787 if (!txn)
4788 return 0;
4789
Willy Tarreau33a7e692007-06-10 19:45:56 +02004790 if (!(test->flags & ACL_TEST_F_FETCH_MORE))
4791 /* search for header from the beginning */
4792 ctx->idx = 0;
4793
Willy Tarreau33a7e692007-06-10 19:45:56 +02004794 if (http_find_header2(expr->arg.str, expr->arg_len, sol, idx, ctx)) {
4795 test->flags |= ACL_TEST_F_FETCH_MORE;
4796 test->flags |= ACL_TEST_F_VOL_HDR;
4797 test->i = strl2ic((char *)ctx->line + ctx->val, ctx->vlen);
4798 return 1;
4799 }
4800
4801 test->flags &= ~ACL_TEST_F_FETCH_MORE;
4802 test->flags |= ACL_TEST_F_VOL_HDR;
4803 return 0;
4804}
4805
Willy Tarreauc11416f2007-06-17 16:58:38 +02004806static int
4807acl_fetch_chdr_val(struct proxy *px, struct session *l4, void *l7, int dir,
4808 struct acl_expr *expr, struct acl_test *test)
4809{
4810 struct http_txn *txn = l7;
4811
Willy Tarreaub6866442008-07-14 23:54:42 +02004812 if (!txn)
4813 return 0;
4814
Willy Tarreauc11416f2007-06-17 16:58:38 +02004815 if (txn->req.msg_state != HTTP_MSG_BODY)
4816 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004817
Willy Tarreauc11416f2007-06-17 16:58:38 +02004818 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4819 /* ensure the indexes are not affected */
4820 return 0;
4821
4822 return acl_fetch_hdr_val(px, l4, txn, txn->req.sol, expr, test);
4823}
4824
4825static int
4826acl_fetch_shdr_val(struct proxy *px, struct session *l4, void *l7, int dir,
4827 struct acl_expr *expr, struct acl_test *test)
4828{
4829 struct http_txn *txn = l7;
4830
Willy Tarreaub6866442008-07-14 23:54:42 +02004831 if (!txn)
4832 return 0;
4833
Willy Tarreauc11416f2007-06-17 16:58:38 +02004834 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4835 return 0;
4836
4837 return acl_fetch_hdr_val(px, l4, txn, txn->rsp.sol, expr, test);
4838}
4839
Willy Tarreau737b0c12007-06-10 21:28:46 +02004840/* 8. Check on URI PATH. A pointer to the PATH is stored. The path starts at
4841 * the first '/' after the possible hostname, and ends before the possible '?'.
4842 */
4843static int
4844acl_fetch_path(struct proxy *px, struct session *l4, void *l7, int dir,
4845 struct acl_expr *expr, struct acl_test *test)
4846{
4847 struct http_txn *txn = l7;
4848 char *ptr, *end;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004849
Willy Tarreaub6866442008-07-14 23:54:42 +02004850 if (!txn)
4851 return 0;
4852
Willy Tarreauc11416f2007-06-17 16:58:38 +02004853 if (txn->req.msg_state != HTTP_MSG_BODY)
4854 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004855
Willy Tarreauc11416f2007-06-17 16:58:38 +02004856 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4857 /* ensure the indexes are not affected */
4858 return 0;
4859
Willy Tarreau21d2af32008-02-14 20:25:24 +01004860 end = txn->req.sol + txn->req.sl.rq.u + txn->req.sl.rq.u_l;
4861 ptr = http_get_path(txn);
4862 if (!ptr)
Willy Tarreau737b0c12007-06-10 21:28:46 +02004863 return 0;
4864
4865 /* OK, we got the '/' ! */
4866 test->ptr = ptr;
4867
4868 while (ptr < end && *ptr != '?')
4869 ptr++;
4870
4871 test->len = ptr - test->ptr;
4872
4873 /* we do not need to set READ_ONLY because the data is in a buffer */
4874 test->flags = ACL_TEST_F_VOL_1ST;
4875 return 1;
4876}
4877
4878
Willy Tarreau8797c062007-05-07 00:55:35 +02004879
4880/************************************************************************/
4881/* All supported keywords must be declared here. */
4882/************************************************************************/
4883
4884/* Note: must not be declared <const> as its list will be overwritten */
4885static struct acl_kw_list acl_kws = {{ },{
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004886 { "method", acl_parse_meth, acl_fetch_meth, acl_match_meth, ACL_USE_L7REQ_PERMANENT },
4887 { "req_ver", acl_parse_ver, acl_fetch_rqver, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4888 { "resp_ver", acl_parse_ver, acl_fetch_stver, acl_match_str, ACL_USE_L7RTR_VOLATILE },
4889 { "status", acl_parse_int, acl_fetch_stcode, acl_match_int, ACL_USE_L7RTR_PERMANENT },
Willy Tarreau8797c062007-05-07 00:55:35 +02004890
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004891 { "url", acl_parse_str, acl_fetch_url, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4892 { "url_beg", acl_parse_str, acl_fetch_url, acl_match_beg, ACL_USE_L7REQ_VOLATILE },
4893 { "url_end", acl_parse_str, acl_fetch_url, acl_match_end, ACL_USE_L7REQ_VOLATILE },
4894 { "url_sub", acl_parse_str, acl_fetch_url, acl_match_sub, ACL_USE_L7REQ_VOLATILE },
4895 { "url_dir", acl_parse_str, acl_fetch_url, acl_match_dir, ACL_USE_L7REQ_VOLATILE },
4896 { "url_dom", acl_parse_str, acl_fetch_url, acl_match_dom, ACL_USE_L7REQ_VOLATILE },
4897 { "url_reg", acl_parse_reg, acl_fetch_url, acl_match_reg, ACL_USE_L7REQ_VOLATILE },
4898 { "url_ip", acl_parse_ip, acl_fetch_url_ip, acl_match_ip, ACL_USE_L7REQ_VOLATILE },
4899 { "url_port", acl_parse_int, acl_fetch_url_port, acl_match_int, ACL_USE_L7REQ_VOLATILE },
Willy Tarreau8797c062007-05-07 00:55:35 +02004900
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004901 /* note: we should set hdr* to use ACL_USE_HDR_VOLATILE, and chdr* to use L7REQ_VOLATILE */
4902 { "hdr", acl_parse_str, acl_fetch_chdr, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4903 { "hdr_reg", acl_parse_reg, acl_fetch_chdr, acl_match_reg, ACL_USE_L7REQ_VOLATILE },
4904 { "hdr_beg", acl_parse_str, acl_fetch_chdr, acl_match_beg, ACL_USE_L7REQ_VOLATILE },
4905 { "hdr_end", acl_parse_str, acl_fetch_chdr, acl_match_end, ACL_USE_L7REQ_VOLATILE },
4906 { "hdr_sub", acl_parse_str, acl_fetch_chdr, acl_match_sub, ACL_USE_L7REQ_VOLATILE },
4907 { "hdr_dir", acl_parse_str, acl_fetch_chdr, acl_match_dir, ACL_USE_L7REQ_VOLATILE },
4908 { "hdr_dom", acl_parse_str, acl_fetch_chdr, acl_match_dom, ACL_USE_L7REQ_VOLATILE },
4909 { "hdr_cnt", acl_parse_int, acl_fetch_chdr_cnt,acl_match_int, ACL_USE_L7REQ_VOLATILE },
4910 { "hdr_val", acl_parse_int, acl_fetch_chdr_val,acl_match_int, ACL_USE_L7REQ_VOLATILE },
Willy Tarreauc11416f2007-06-17 16:58:38 +02004911
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004912 { "shdr", acl_parse_str, acl_fetch_shdr, acl_match_str, ACL_USE_L7RTR_VOLATILE },
4913 { "shdr_reg", acl_parse_reg, acl_fetch_shdr, acl_match_reg, ACL_USE_L7RTR_VOLATILE },
4914 { "shdr_beg", acl_parse_str, acl_fetch_shdr, acl_match_beg, ACL_USE_L7RTR_VOLATILE },
4915 { "shdr_end", acl_parse_str, acl_fetch_shdr, acl_match_end, ACL_USE_L7RTR_VOLATILE },
4916 { "shdr_sub", acl_parse_str, acl_fetch_shdr, acl_match_sub, ACL_USE_L7RTR_VOLATILE },
4917 { "shdr_dir", acl_parse_str, acl_fetch_shdr, acl_match_dir, ACL_USE_L7RTR_VOLATILE },
4918 { "shdr_dom", acl_parse_str, acl_fetch_shdr, acl_match_dom, ACL_USE_L7RTR_VOLATILE },
4919 { "shdr_cnt", acl_parse_int, acl_fetch_shdr_cnt,acl_match_int, ACL_USE_L7RTR_VOLATILE },
4920 { "shdr_val", acl_parse_int, acl_fetch_shdr_val,acl_match_int, ACL_USE_L7RTR_VOLATILE },
Willy Tarreau737b0c12007-06-10 21:28:46 +02004921
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004922 { "path", acl_parse_str, acl_fetch_path, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4923 { "path_reg", acl_parse_reg, acl_fetch_path, acl_match_reg, ACL_USE_L7REQ_VOLATILE },
4924 { "path_beg", acl_parse_str, acl_fetch_path, acl_match_beg, ACL_USE_L7REQ_VOLATILE },
4925 { "path_end", acl_parse_str, acl_fetch_path, acl_match_end, ACL_USE_L7REQ_VOLATILE },
4926 { "path_sub", acl_parse_str, acl_fetch_path, acl_match_sub, ACL_USE_L7REQ_VOLATILE },
4927 { "path_dir", acl_parse_str, acl_fetch_path, acl_match_dir, ACL_USE_L7REQ_VOLATILE },
4928 { "path_dom", acl_parse_str, acl_fetch_path, acl_match_dom, ACL_USE_L7REQ_VOLATILE },
Willy Tarreau737b0c12007-06-10 21:28:46 +02004929
Willy Tarreauf3d25982007-05-08 22:45:09 +02004930 { NULL, NULL, NULL, NULL },
4931
4932#if 0
Willy Tarreau8797c062007-05-07 00:55:35 +02004933 { "line", acl_parse_str, acl_fetch_line, acl_match_str },
4934 { "line_reg", acl_parse_reg, acl_fetch_line, acl_match_reg },
4935 { "line_beg", acl_parse_str, acl_fetch_line, acl_match_beg },
4936 { "line_end", acl_parse_str, acl_fetch_line, acl_match_end },
4937 { "line_sub", acl_parse_str, acl_fetch_line, acl_match_sub },
4938 { "line_dir", acl_parse_str, acl_fetch_line, acl_match_dir },
4939 { "line_dom", acl_parse_str, acl_fetch_line, acl_match_dom },
4940
Willy Tarreau8797c062007-05-07 00:55:35 +02004941 { "cook", acl_parse_str, acl_fetch_cook, acl_match_str },
4942 { "cook_reg", acl_parse_reg, acl_fetch_cook, acl_match_reg },
4943 { "cook_beg", acl_parse_str, acl_fetch_cook, acl_match_beg },
4944 { "cook_end", acl_parse_str, acl_fetch_cook, acl_match_end },
4945 { "cook_sub", acl_parse_str, acl_fetch_cook, acl_match_sub },
4946 { "cook_dir", acl_parse_str, acl_fetch_cook, acl_match_dir },
4947 { "cook_dom", acl_parse_str, acl_fetch_cook, acl_match_dom },
4948 { "cook_pst", acl_parse_none, acl_fetch_cook, acl_match_pst },
4949
4950 { "auth_user", acl_parse_str, acl_fetch_user, acl_match_str },
4951 { "auth_regex", acl_parse_reg, acl_fetch_user, acl_match_reg },
4952 { "auth_clear", acl_parse_str, acl_fetch_auth, acl_match_str },
4953 { "auth_md5", acl_parse_str, acl_fetch_auth, acl_match_md5 },
4954 { NULL, NULL, NULL, NULL },
4955#endif
4956}};
4957
4958
4959__attribute__((constructor))
4960static void __http_protocol_init(void)
4961{
4962 acl_register_keywords(&acl_kws);
4963}
4964
4965
Willy Tarreau58f10d72006-12-04 02:26:12 +01004966/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02004967 * Local variables:
4968 * c-indent-level: 8
4969 * c-basic-offset: 8
4970 * End:
4971 */