blob: 359488bea56bcb9fae243132ba7d56639718463e [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 Tarreau8d5d7f22007-01-21 19:16:41 +01001378 goto http_msg_invalid;
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001379
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001380 http_msg_hdr_l1_sp:
1381 case HTTP_MSG_HDR_L1_SP:
1382 /* assumes msg->sol points to the first char and msg->col to the colon */
1383 if (likely(HTTP_IS_SPHT(*ptr)))
1384 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_sp, HTTP_MSG_HDR_L1_SP);
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001385
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001386 /* header value can be basically anything except CR/LF */
1387 msg->sov = ptr - buf->data;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001388
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001389 if (likely(!HTTP_IS_CRLF(*ptr))) {
1390 goto http_msg_hdr_val;
1391 }
1392
1393 if (likely(*ptr == '\r'))
1394 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_lf, HTTP_MSG_HDR_L1_LF);
1395 goto http_msg_hdr_l1_lf;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001396
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001397 http_msg_hdr_l1_lf:
1398 case HTTP_MSG_HDR_L1_LF:
1399 EXPECT_LF_HERE(ptr, http_msg_invalid);
1400 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l1_lws, HTTP_MSG_HDR_L1_LWS);
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001401
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001402 http_msg_hdr_l1_lws:
1403 case HTTP_MSG_HDR_L1_LWS:
1404 if (likely(HTTP_IS_SPHT(*ptr))) {
1405 /* replace HT,CR,LF with spaces */
1406 for (; buf->data+msg->sov < ptr; msg->sov++)
1407 buf->data[msg->sov] = ' ';
1408 goto http_msg_hdr_l1_sp;
1409 }
Willy Tarreauaa9dce32007-03-18 23:50:16 +01001410 /* we had a header consisting only in spaces ! */
1411 msg->eol = buf->data + msg->sov;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001412 goto http_msg_complete_header;
1413
1414 http_msg_hdr_val:
1415 case HTTP_MSG_HDR_VAL:
1416 /* assumes msg->sol points to the first char, msg->col to the
1417 * colon, and msg->sov points to the first character of the
1418 * value.
1419 */
1420 if (likely(!HTTP_IS_CRLF(*ptr)))
1421 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_val, HTTP_MSG_HDR_VAL);
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001422
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001423 msg->eol = ptr;
1424 /* Note: we could also copy eol into ->eoh so that we have the
1425 * real header end in case it ends with lots of LWS, but is this
1426 * really needed ?
1427 */
1428 if (likely(*ptr == '\r'))
1429 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l2_lf, HTTP_MSG_HDR_L2_LF);
1430 goto http_msg_hdr_l2_lf;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001431
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001432 http_msg_hdr_l2_lf:
1433 case HTTP_MSG_HDR_L2_LF:
1434 EXPECT_LF_HERE(ptr, http_msg_invalid);
1435 EAT_AND_JUMP_OR_RETURN(http_msg_hdr_l2_lws, HTTP_MSG_HDR_L2_LWS);
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001436
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001437 http_msg_hdr_l2_lws:
1438 case HTTP_MSG_HDR_L2_LWS:
1439 if (unlikely(HTTP_IS_SPHT(*ptr))) {
1440 /* LWS: replace HT,CR,LF with spaces */
1441 for (; msg->eol < ptr; msg->eol++)
1442 *msg->eol = ' ';
1443 goto http_msg_hdr_val;
1444 }
1445 http_msg_complete_header:
1446 /*
1447 * It was a new header, so the last one is finished.
1448 * Assumes msg->sol points to the first char, msg->col to the
1449 * colon, msg->sov points to the first character of the value
1450 * and msg->eol to the first CR or LF so we know how the line
1451 * ends. We insert last header into the index.
1452 */
1453 /*
1454 fprintf(stderr,"registering %-2d bytes : ", msg->eol - msg->sol);
1455 write(2, msg->sol, msg->eol-msg->sol);
1456 fprintf(stderr,"\n");
1457 */
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001458
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001459 if (unlikely(hdr_idx_add(msg->eol - msg->sol, *msg->eol == '\r',
1460 idx, idx->tail) < 0))
1461 goto http_msg_invalid;
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001462
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001463 msg->sol = ptr;
1464 if (likely(!HTTP_IS_CRLF(*ptr))) {
1465 goto http_msg_hdr_name;
1466 }
1467
1468 if (likely(*ptr == '\r'))
1469 EAT_AND_JUMP_OR_RETURN(http_msg_last_lf, HTTP_MSG_LAST_LF);
1470 goto http_msg_last_lf;
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001471
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001472 http_msg_last_lf:
1473 case HTTP_MSG_LAST_LF:
1474 /* Assumes msg->sol points to the first of either CR or LF */
1475 EXPECT_LF_HERE(ptr, http_msg_invalid);
1476 ptr++;
1477 buf->lr = ptr;
1478 msg->eoh = msg->sol - buf->data;
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001479 msg->msg_state = HTTP_MSG_BODY;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001480 return;
1481#ifdef DEBUG_FULL
1482 default:
1483 fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state);
1484 exit(1);
Willy Tarreau230fd0b2006-12-17 12:05:00 +01001485#endif
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001486 }
1487 http_msg_ood:
1488 /* out of data */
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001489 msg->msg_state = state;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001490 buf->lr = ptr;
1491 return;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001492
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001493 http_msg_invalid:
1494 /* invalid message */
Willy Tarreaub326fcc2007-03-03 13:54:32 +01001495 msg->msg_state = HTTP_MSG_ERROR;
Willy Tarreau7552c032009-03-01 11:10:40 +01001496 buf->lr = ptr;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001497 return;
1498}
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001499
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02001500/* This function performs all the processing enabled for the current request.
Willy Tarreauedcf6682008-11-30 23:15:34 +01001501 * It returns 1 if the processing can continue on next analysers, or zero if it
1502 * needs more data, encounters an error, or wants to immediately abort the
1503 * request. It relies on buffers flags, and updates s->req->analysers. Its
1504 * behaviour is rather simple:
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001505 * - all enabled analysers are called in turn from the lower to the higher
1506 * bit.
Willy Tarreauedcf6682008-11-30 23:15:34 +01001507 * - the analyser must check for errors and timeouts, and react as expected.
1508 * It does not have to close anything upon error, the caller will.
1509 * - if the analyser does not have enough data, it must return 0without calling
1510 * other ones. It should also probably do a buffer_write_dis() to ensure
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001511 * that unprocessed data will not be forwarded. But that probably depends on
Willy Tarreauedcf6682008-11-30 23:15:34 +01001512 * the protocol.
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001513 * - if an analyser has enough data, it just has to pass on to the next
Willy Tarreauedcf6682008-11-30 23:15:34 +01001514 * analyser without using buffer_write_dis() (enabled by default).
Willy Tarreau41f40ed2008-08-21 10:05:00 +02001515 * - if an analyser thinks it has no added value anymore staying here, it must
1516 * reset its bit from the analysers flags in order not to be called anymore.
1517 *
1518 * In the future, analysers should be able to indicate that they want to be
1519 * called after XXX bytes have been received (or transfered), and the min of
1520 * all's wishes will be used to ring back (unless a special condition occurs).
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001521 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001522int http_process_request(struct session *s, struct buffer *req)
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001523{
Willy Tarreau59234e92008-11-30 23:51:27 +01001524 /*
1525 * We will parse the partial (or complete) lines.
1526 * We will check the request syntax, and also join multi-line
1527 * headers. An index of all the lines will be elaborated while
1528 * parsing.
1529 *
1530 * For the parsing, we use a 28 states FSM.
1531 *
1532 * Here is the information we currently have :
Willy Tarreauf073a832009-03-01 23:21:47 +01001533 * req->data + msg->som = beginning of request
Willy Tarreau59234e92008-11-30 23:51:27 +01001534 * req->data + req->eoh = end of processed headers / start of current one
1535 * req->data + req->eol = end of current header or line (LF or CRLF)
1536 * req->lr = first non-visited byte
1537 * req->r = end of data
1538 */
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001539
Willy Tarreau59234e92008-11-30 23:51:27 +01001540 int cur_idx;
1541 struct http_txn *txn = &s->txn;
1542 struct http_msg *msg = &txn->req;
1543 struct proxy *cur_proxy;
Willy Tarreau976f1ee2006-12-17 10:06:03 +01001544
Willy Tarreau6bf17362009-02-24 10:48:35 +01001545 DPRINTF(stderr,"[%u] %s: session=%p b=%p, exp(r,w)=%u,%u bf=%08x bl=%d analysers=%02x\n",
1546 now_ms, __FUNCTION__,
1547 s,
1548 req,
1549 req->rex, req->wex,
1550 req->flags,
1551 req->l,
1552 req->analysers);
1553
Willy Tarreau59234e92008-11-30 23:51:27 +01001554 if (likely(req->lr < req->r))
1555 http_msg_analyzer(req, msg, &txn->hdr_idx);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001556
Willy Tarreau59234e92008-11-30 23:51:27 +01001557 /* 1: we might have to print this header in debug mode */
1558 if (unlikely((global.mode & MODE_DEBUG) &&
1559 (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) &&
1560 (msg->msg_state == HTTP_MSG_BODY || msg->msg_state == HTTP_MSG_ERROR))) {
1561 char *eol, *sol;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001562
Willy Tarreau59234e92008-11-30 23:51:27 +01001563 sol = req->data + msg->som;
1564 eol = sol + msg->sl.rq.l;
1565 debug_hdr("clireq", s, sol, eol);
Willy Tarreau45e73e32006-12-17 00:05:15 +01001566
Willy Tarreau59234e92008-11-30 23:51:27 +01001567 sol += hdr_idx_first_pos(&txn->hdr_idx);
1568 cur_idx = hdr_idx_first_idx(&txn->hdr_idx);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001569
Willy Tarreau59234e92008-11-30 23:51:27 +01001570 while (cur_idx) {
1571 eol = sol + txn->hdr_idx.v[cur_idx].len;
1572 debug_hdr("clihdr", s, sol, eol);
1573 sol = eol + txn->hdr_idx.v[cur_idx].cr + 1;
1574 cur_idx = txn->hdr_idx.v[cur_idx].next;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001575 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001576 }
1577
Willy Tarreau58f10d72006-12-04 02:26:12 +01001578
Willy Tarreau59234e92008-11-30 23:51:27 +01001579 /*
1580 * Now we quickly check if we have found a full valid request.
1581 * If not so, we check the FD and buffer states before leaving.
1582 * A full request is indicated by the fact that we have seen
1583 * the double LF/CRLF, so the state is HTTP_MSG_BODY. Invalid
1584 * requests are checked first.
1585 *
1586 */
Willy Tarreau58f10d72006-12-04 02:26:12 +01001587
Willy Tarreau59234e92008-11-30 23:51:27 +01001588 if (unlikely(msg->msg_state != HTTP_MSG_BODY)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01001589 /*
Willy Tarreau59234e92008-11-30 23:51:27 +01001590 * First, let's catch bad requests.
Willy Tarreau58f10d72006-12-04 02:26:12 +01001591 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001592 if (unlikely(msg->msg_state == HTTP_MSG_ERROR))
1593 goto return_bad_req;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001594
Willy Tarreau59234e92008-11-30 23:51:27 +01001595 /* 1: Since we are in header mode, if there's no space
1596 * left for headers, we won't be able to free more
1597 * later, so the session will never terminate. We
1598 * must terminate it now.
1599 */
1600 if (unlikely(req->flags & BF_FULL)) {
1601 /* FIXME: check if URI is set and return Status
1602 * 414 Request URI too long instead.
Willy Tarreau58f10d72006-12-04 02:26:12 +01001603 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001604 goto return_bad_req;
1605 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001606
Willy Tarreau59234e92008-11-30 23:51:27 +01001607 /* 2: have we encountered a read error ? */
1608 else if (req->flags & BF_READ_ERROR) {
1609 /* we cannot return any message on error */
1610 msg->msg_state = HTTP_MSG_ERROR;
1611 req->analysers = 0;
1612 s->fe->failed_req++;
1613 if (!(s->flags & SN_ERR_MASK))
1614 s->flags |= SN_ERR_CLICL;
1615 if (!(s->flags & SN_FINST_MASK))
1616 s->flags |= SN_FINST_R;
1617 return 0;
1618 }
Willy Tarreauf9839bd2008-08-27 23:57:16 +02001619
Willy Tarreau59234e92008-11-30 23:51:27 +01001620 /* 3: has the read timeout expired ? */
1621 else if (req->flags & BF_READ_TIMEOUT || tick_is_expired(req->analyse_exp, now_ms)) {
1622 /* read timeout : give up with an error message. */
1623 txn->status = 408;
1624 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_408));
1625 msg->msg_state = HTTP_MSG_ERROR;
1626 req->analysers = 0;
1627 s->fe->failed_req++;
1628 if (!(s->flags & SN_ERR_MASK))
1629 s->flags |= SN_ERR_CLITO;
1630 if (!(s->flags & SN_FINST_MASK))
1631 s->flags |= SN_FINST_R;
1632 return 0;
1633 }
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02001634
Willy Tarreau59234e92008-11-30 23:51:27 +01001635 /* 4: have we encountered a close ? */
1636 else if (req->flags & BF_SHUTR) {
1637 txn->status = 400;
1638 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_400));
1639 msg->msg_state = HTTP_MSG_ERROR;
1640 req->analysers = 0;
1641 s->fe->failed_req++;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02001642
Willy Tarreau59234e92008-11-30 23:51:27 +01001643 if (!(s->flags & SN_ERR_MASK))
1644 s->flags |= SN_ERR_CLICL;
1645 if (!(s->flags & SN_FINST_MASK))
1646 s->flags |= SN_FINST_R;
Willy Tarreaudafde432008-08-17 01:00:46 +02001647 return 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001648 }
1649
Willy Tarreau59234e92008-11-30 23:51:27 +01001650 buffer_write_dis(req);
Willy Tarreau1b194fe2009-03-21 21:10:04 +01001651 req->flags |= BF_READ_DONTWAIT; /* try to get back here ASAP */
1652
Willy Tarreau59234e92008-11-30 23:51:27 +01001653 /* just set the request timeout once at the beginning of the request */
1654 if (!tick_isset(req->analyse_exp))
1655 req->analyse_exp = tick_add_ifset(now_ms, s->fe->timeout.httpreq);
Willy Tarreau58f10d72006-12-04 02:26:12 +01001656
Willy Tarreau59234e92008-11-30 23:51:27 +01001657 /* we're not ready yet */
1658 return 0;
1659 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001660
Willy Tarreau67f0eea2008-08-10 22:55:22 +02001661
Willy Tarreau59234e92008-11-30 23:51:27 +01001662 /****************************************************************
1663 * More interesting part now : we know that we have a complete *
1664 * request which at least looks like HTTP. We have an indicator *
1665 * of each header's length, so we can parse them quickly. *
1666 ****************************************************************/
Willy Tarreau9cdde232007-05-02 20:58:19 +02001667
Willy Tarreau59234e92008-11-30 23:51:27 +01001668 req->analysers &= ~AN_REQ_HTTP_HDR;
1669 req->analyse_exp = TICK_ETERNITY;
1670
1671 /* ensure we keep this pointer to the beginning of the message */
1672 msg->sol = req->data + msg->som;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001673
Willy Tarreau59234e92008-11-30 23:51:27 +01001674 /*
1675 * 1: identify the method
1676 */
1677 txn->meth = find_http_meth(&req->data[msg->som], msg->sl.rq.m_l);
1678
1679 /* we can make use of server redirect on GET and HEAD */
1680 if (txn->meth == HTTP_METH_GET || txn->meth == HTTP_METH_HEAD)
1681 s->flags |= SN_REDIRECTABLE;
Willy Tarreaufa7e1022008-10-19 07:30:41 +02001682
Willy Tarreau59234e92008-11-30 23:51:27 +01001683 /*
1684 * 2: check if the URI matches the monitor_uri.
1685 * We have to do this for every request which gets in, because
1686 * the monitor-uri is defined by the frontend.
1687 */
1688 if (unlikely((s->fe->monitor_uri_len != 0) &&
1689 (s->fe->monitor_uri_len == msg->sl.rq.u_l) &&
1690 !memcmp(&req->data[msg->sl.rq.u],
1691 s->fe->monitor_uri,
1692 s->fe->monitor_uri_len))) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01001693 /*
Willy Tarreau59234e92008-11-30 23:51:27 +01001694 * We have found the monitor URI
Willy Tarreau58f10d72006-12-04 02:26:12 +01001695 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001696 struct acl_cond *cond;
1697 cur_proxy = s->fe;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001698
Willy Tarreau59234e92008-11-30 23:51:27 +01001699 s->flags |= SN_MONITOR;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001700
Willy Tarreau59234e92008-11-30 23:51:27 +01001701 /* Check if we want to fail this monitor request or not */
1702 list_for_each_entry(cond, &cur_proxy->mon_fail_cond, list) {
1703 int ret = acl_exec_cond(cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreau11382812008-07-09 16:18:21 +02001704
Willy Tarreau59234e92008-11-30 23:51:27 +01001705 ret = acl_pass(ret);
1706 if (cond->pol == ACL_COND_UNLESS)
1707 ret = !ret;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001708
Willy Tarreau59234e92008-11-30 23:51:27 +01001709 if (ret) {
1710 /* we fail this request, let's return 503 service unavail */
1711 txn->status = 503;
1712 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_503));
1713 goto return_prx_cond;
Willy Tarreaub80c2302007-11-30 20:51:32 +01001714 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001715 }
Willy Tarreaua5555ec2008-11-30 19:02:32 +01001716
Willy Tarreau59234e92008-11-30 23:51:27 +01001717 /* nothing to fail, let's reply normaly */
1718 txn->status = 200;
1719 stream_int_retnclose(req->prod, &http_200_chunk);
1720 goto return_prx_cond;
1721 }
1722
1723 /*
1724 * 3: Maybe we have to copy the original REQURI for the logs ?
1725 * Note: we cannot log anymore if the request has been
1726 * classified as invalid.
1727 */
1728 if (unlikely(s->logs.logwait & LW_REQ)) {
1729 /* we have a complete HTTP request that we must log */
1730 if ((txn->uri = pool_alloc2(pool2_requri)) != NULL) {
1731 int urilen = msg->sl.rq.l;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001732
Willy Tarreau59234e92008-11-30 23:51:27 +01001733 if (urilen >= REQURI_LEN)
1734 urilen = REQURI_LEN - 1;
1735 memcpy(txn->uri, &req->data[msg->som], urilen);
1736 txn->uri[urilen] = 0;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001737
Willy Tarreau59234e92008-11-30 23:51:27 +01001738 if (!(s->logs.logwait &= ~LW_REQ))
1739 s->do_log(s);
1740 } else {
1741 Alert("HTTP logging : out of memory.\n");
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001742 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001743 }
Willy Tarreau06619262006-12-17 08:37:22 +01001744
Willy Tarreau59234e92008-11-30 23:51:27 +01001745 /* 4. We may have to convert HTTP/0.9 requests to HTTP/1.0 */
1746 if (unlikely(msg->sl.rq.v_l == 0)) {
1747 int delta;
1748 char *cur_end;
1749 msg->sol = req->data + msg->som;
1750 cur_end = msg->sol + msg->sl.rq.l;
1751 delta = 0;
Willy Tarreau06619262006-12-17 08:37:22 +01001752
Willy Tarreau59234e92008-11-30 23:51:27 +01001753 if (msg->sl.rq.u_l == 0) {
1754 /* if no URI was set, add "/" */
1755 delta = buffer_replace2(req, cur_end, cur_end, " /", 2);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001756 cur_end += delta;
Willy Tarreau59234e92008-11-30 23:51:27 +01001757 msg->eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01001758 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001759 /* add HTTP version */
1760 delta = buffer_replace2(req, cur_end, cur_end, " HTTP/1.0\r\n", 11);
1761 msg->eoh += delta;
1762 cur_end += delta;
1763 cur_end = (char *)http_parse_reqline(msg, req->data,
1764 HTTP_MSG_RQMETH,
1765 msg->sol, cur_end + 1,
1766 NULL, NULL);
1767 if (unlikely(!cur_end))
1768 goto return_bad_req;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001769
Willy Tarreau59234e92008-11-30 23:51:27 +01001770 /* we have a full HTTP/1.0 request now and we know that
1771 * we have either a CR or an LF at <ptr>.
Willy Tarreau58f10d72006-12-04 02:26:12 +01001772 */
Willy Tarreau59234e92008-11-30 23:51:27 +01001773 hdr_idx_set_start(&txn->hdr_idx, msg->sl.rq.l, *cur_end == '\r');
1774 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01001775
1776
Willy Tarreau59234e92008-11-30 23:51:27 +01001777 /* 5: we may need to capture headers */
1778 if (unlikely((s->logs.logwait & LW_REQHDR) && s->fe->req_cap))
1779 capture_headers(req->data + msg->som, &txn->hdr_idx,
1780 txn->req.cap, s->fe->req_cap);
Willy Tarreau11382812008-07-09 16:18:21 +02001781
Willy Tarreau59234e92008-11-30 23:51:27 +01001782 /*
1783 * 6: we will have to evaluate the filters.
1784 * As opposed to version 1.2, now they will be evaluated in the
1785 * filters order and not in the header order. This means that
1786 * each filter has to be validated among all headers.
1787 *
1788 * We can now check whether we want to switch to another
1789 * backend, in which case we will re-check the backend's
1790 * filters and various options. In order to support 3-level
1791 * switching, here's how we should proceed :
1792 *
1793 * a) run be.
1794 * if (switch) then switch ->be to the new backend.
1795 * b) run be if (be != fe).
1796 * There cannot be any switch from there, so ->be cannot be
1797 * changed anymore.
1798 *
1799 * => filters always apply to ->be, then ->be may change.
1800 *
1801 * The response path will be able to apply either ->be, or
1802 * ->be then ->fe filters in order to match the reverse of
1803 * the forward sequence.
1804 */
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001805
Willy Tarreau59234e92008-11-30 23:51:27 +01001806 do {
1807 struct acl_cond *cond;
1808 struct redirect_rule *rule;
1809 struct proxy *rule_set = s->be;
1810 cur_proxy = s->be;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001811
Willy Tarreau59234e92008-11-30 23:51:27 +01001812 /* first check whether we have some ACLs set to redirect this request */
1813 list_for_each_entry(rule, &cur_proxy->redirect_rules, list) {
1814 int ret = acl_exec_cond(rule->cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001815
Willy Tarreau59234e92008-11-30 23:51:27 +01001816 ret = acl_pass(ret);
1817 if (rule->cond->pol == ACL_COND_UNLESS)
1818 ret = !ret;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001819
Willy Tarreau59234e92008-11-30 23:51:27 +01001820 if (ret) {
1821 struct chunk rdr = { trash, 0 };
1822 const char *msg_fmt;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001823
Willy Tarreau59234e92008-11-30 23:51:27 +01001824 /* build redirect message */
1825 switch(rule->code) {
1826 case 303:
1827 rdr.len = strlen(HTTP_303);
1828 msg_fmt = HTTP_303;
1829 break;
1830 case 301:
1831 rdr.len = strlen(HTTP_301);
1832 msg_fmt = HTTP_301;
1833 break;
1834 case 302:
1835 default:
1836 rdr.len = strlen(HTTP_302);
1837 msg_fmt = HTTP_302;
1838 break;
1839 }
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001840
Willy Tarreau59234e92008-11-30 23:51:27 +01001841 if (unlikely(rdr.len > sizeof(trash)))
1842 goto return_bad_req;
1843 memcpy(rdr.str, msg_fmt, rdr.len);
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001844
Willy Tarreau59234e92008-11-30 23:51:27 +01001845 switch(rule->type) {
1846 case REDIRECT_TYPE_PREFIX: {
1847 const char *path;
1848 int pathlen;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001849
Willy Tarreau59234e92008-11-30 23:51:27 +01001850 path = http_get_path(txn);
1851 /* build message using path */
1852 if (path) {
1853 pathlen = txn->req.sl.rq.u_l + (txn->req.sol+txn->req.sl.rq.u) - path;
Willy Tarreau79da4692008-11-19 20:03:04 +01001854 if (rule->flags & REDIRECT_FLAG_DROP_QS) {
1855 int qs = 0;
1856 while (qs < pathlen) {
1857 if (path[qs] == '?') {
1858 pathlen = qs;
1859 break;
1860 }
1861 qs++;
1862 }
1863 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001864 } else {
1865 path = "/";
1866 pathlen = 1;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001867 }
1868
Willy Tarreau59234e92008-11-30 23:51:27 +01001869 if (rdr.len + rule->rdr_len + pathlen > sizeof(trash) - 4)
1870 goto return_bad_req;
1871
Willy Tarreaufe651a52008-11-19 21:15:17 +01001872 /* add prefix. Note that if prefix == "/", we don't want to
1873 * add anything, otherwise it makes it hard for the user to
1874 * configure a self-redirection.
1875 */
1876 if (rule->rdr_len != 1 || *rule->rdr_str != '/') {
1877 memcpy(rdr.str + rdr.len, rule->rdr_str, rule->rdr_len);
1878 rdr.len += rule->rdr_len;
1879 }
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001880
Willy Tarreau59234e92008-11-30 23:51:27 +01001881 /* add path */
1882 memcpy(rdr.str + rdr.len, path, pathlen);
1883 rdr.len += pathlen;
1884 break;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001885 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001886 case REDIRECT_TYPE_LOCATION:
1887 default:
1888 if (rdr.len + rule->rdr_len > sizeof(trash) - 4)
1889 goto return_bad_req;
Willy Tarreaub463dfb2008-06-07 23:08:56 +02001890
Willy Tarreau59234e92008-11-30 23:51:27 +01001891 /* add location */
1892 memcpy(rdr.str + rdr.len, rule->rdr_str, rule->rdr_len);
1893 rdr.len += rule->rdr_len;
1894 break;
1895 }
Willy Tarreau11382812008-07-09 16:18:21 +02001896
Willy Tarreau0140f252008-11-19 21:07:09 +01001897 if (rule->cookie_len) {
1898 memcpy(rdr.str + rdr.len, "\r\nSet-Cookie: ", 14);
1899 rdr.len += 14;
1900 memcpy(rdr.str + rdr.len, rule->cookie_str, rule->cookie_len);
1901 rdr.len += rule->cookie_len;
1902 memcpy(rdr.str + rdr.len, "\r\n", 2);
1903 rdr.len += 2;
1904 }
1905
Willy Tarreau59234e92008-11-30 23:51:27 +01001906 /* add end of headers */
1907 memcpy(rdr.str + rdr.len, "\r\n\r\n", 4);
1908 rdr.len += 4;
Willy Tarreau5c8e3e02007-05-07 00:58:25 +02001909
Willy Tarreau59234e92008-11-30 23:51:27 +01001910 txn->status = rule->code;
1911 /* let's log the request time */
1912 s->logs.tv_request = now;
1913 stream_int_retnclose(req->prod, &rdr);
1914 goto return_prx_cond;
Willy Tarreau5c8e3e02007-05-07 00:58:25 +02001915 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001916 }
Willy Tarreau5c8e3e02007-05-07 00:58:25 +02001917
Willy Tarreau59234e92008-11-30 23:51:27 +01001918 /* first check whether we have some ACLs set to block this request */
1919 list_for_each_entry(cond, &cur_proxy->block_cond, list) {
1920 int ret = acl_exec_cond(cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreau53b6c742006-12-17 13:37:46 +01001921
Willy Tarreau59234e92008-11-30 23:51:27 +01001922 ret = acl_pass(ret);
1923 if (cond->pol == ACL_COND_UNLESS)
1924 ret = !ret;
Willy Tarreauf1221aa2006-12-17 22:14:12 +01001925
Willy Tarreau59234e92008-11-30 23:51:27 +01001926 if (ret) {
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01001927 txn->status = 403;
Willy Tarreau06619262006-12-17 08:37:22 +01001928 /* let's log the request time */
Willy Tarreau59234e92008-11-30 23:51:27 +01001929 s->logs.tv_request = now;
1930 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_403));
Willy Tarreau06619262006-12-17 08:37:22 +01001931 goto return_prx_cond;
1932 }
Willy Tarreau59234e92008-11-30 23:51:27 +01001933 }
1934
1935 /* try headers filters */
1936 if (rule_set->req_exp != NULL) {
1937 if (apply_filters_to_request(s, req, rule_set->req_exp) < 0)
1938 goto return_bad_req;
1939 }
1940
1941 if (!(s->flags & SN_BE_ASSIGNED) && (s->be != cur_proxy)) {
1942 /* to ensure correct connection accounting on
1943 * the backend, we count the connection for the
1944 * one managing the queue.
1945 */
1946 s->be->beconn++;
1947 if (s->be->beconn > s->be->beconn_max)
1948 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01001949 proxy_inc_be_ctr(s->be);
Willy Tarreau59234e92008-11-30 23:51:27 +01001950 s->flags |= SN_BE_ASSIGNED;
1951 }
Willy Tarreau06619262006-12-17 08:37:22 +01001952
Willy Tarreau59234e92008-11-30 23:51:27 +01001953 /* has the request been denied ? */
1954 if (txn->flags & TX_CLDENY) {
1955 /* no need to go further */
1956 txn->status = 403;
1957 /* let's log the request time */
1958 s->logs.tv_request = now;
1959 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_403));
1960 goto return_prx_cond;
1961 }
Willy Tarreau06619262006-12-17 08:37:22 +01001962
Willy Tarreau59234e92008-11-30 23:51:27 +01001963 /* We might have to check for "Connection:" */
1964 if (((s->fe->options | s->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO)) &&
1965 !(s->flags & SN_CONN_CLOSED)) {
1966 char *cur_ptr, *cur_end, *cur_next;
1967 int cur_idx, old_idx, delta, val;
1968 struct hdr_idx_elem *cur_hdr;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001969
Willy Tarreau59234e92008-11-30 23:51:27 +01001970 cur_next = req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
1971 old_idx = 0;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01001972
Willy Tarreau59234e92008-11-30 23:51:27 +01001973 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
1974 cur_hdr = &txn->hdr_idx.v[cur_idx];
1975 cur_ptr = cur_next;
1976 cur_end = cur_ptr + cur_hdr->len;
1977 cur_next = cur_end + cur_hdr->cr + 1;
1978
1979 val = http_header_match2(cur_ptr, cur_end, "Connection", 10);
1980 if (val) {
1981 /* 3 possibilities :
1982 * - we have already set Connection: close,
1983 * so we remove this line.
1984 * - we have not yet set Connection: close,
1985 * but this line indicates close. We leave
1986 * it untouched and set the flag.
1987 * - we have not yet set Connection: close,
1988 * and this line indicates non-close. We
1989 * replace it.
1990 */
1991 if (s->flags & SN_CONN_CLOSED) {
1992 delta = buffer_replace2(req, cur_ptr, cur_next, NULL, 0);
1993 txn->req.eoh += delta;
1994 cur_next += delta;
1995 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
1996 txn->hdr_idx.used--;
1997 cur_hdr->len = 0;
1998 } else {
1999 if (strncasecmp(cur_ptr + val, "close", 5) != 0) {
2000 delta = buffer_replace2(req, cur_ptr + val, cur_end,
2001 "close", 5);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002002 cur_next += delta;
Willy Tarreau59234e92008-11-30 23:51:27 +01002003 cur_hdr->len += delta;
2004 txn->req.eoh += delta;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002005 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002006 s->flags |= SN_CONN_CLOSED;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002007 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002008 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002009 old_idx = cur_idx;
Willy Tarreau06619262006-12-17 08:37:22 +01002010 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002011 }
2012 /* add request headers from the rule sets in the same order */
2013 for (cur_idx = 0; cur_idx < rule_set->nb_reqadd; cur_idx++) {
2014 if (unlikely(http_header_add_tail(req,
2015 &txn->req,
2016 &txn->hdr_idx,
2017 rule_set->req_add[cur_idx])) < 0)
2018 goto return_bad_req;
2019 }
Willy Tarreaub2513902006-12-17 14:52:38 +01002020
Willy Tarreau59234e92008-11-30 23:51:27 +01002021 /* check if stats URI was requested, and if an auth is needed */
2022 if (rule_set->uri_auth != NULL &&
2023 (txn->meth == HTTP_METH_GET || txn->meth == HTTP_METH_HEAD)) {
2024 /* we have to check the URI and auth for this request.
2025 * FIXME!!! that one is rather dangerous, we want to
2026 * make it follow standard rules (eg: clear req->analysers).
2027 */
2028 if (stats_check_uri_auth(s, rule_set)) {
2029 req->analysers = 0;
2030 return 0;
Willy Tarreaub2513902006-12-17 14:52:38 +01002031 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002032 }
Willy Tarreaub2513902006-12-17 14:52:38 +01002033
Willy Tarreau59234e92008-11-30 23:51:27 +01002034 /* now check whether we have some switching rules for this request */
2035 if (!(s->flags & SN_BE_ASSIGNED)) {
2036 struct switching_rule *rule;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002037
Willy Tarreau59234e92008-11-30 23:51:27 +01002038 list_for_each_entry(rule, &cur_proxy->switching_rules, list) {
2039 int ret;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002040
Willy Tarreau59234e92008-11-30 23:51:27 +01002041 ret = acl_exec_cond(rule->cond, cur_proxy, s, txn, ACL_DIR_REQ);
Willy Tarreau11382812008-07-09 16:18:21 +02002042
Willy Tarreau59234e92008-11-30 23:51:27 +01002043 ret = acl_pass(ret);
2044 if (rule->cond->pol == ACL_COND_UNLESS)
2045 ret = !ret;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002046
Willy Tarreau59234e92008-11-30 23:51:27 +01002047 if (ret) {
2048 s->be = rule->be.backend;
2049 s->be->beconn++;
2050 if (s->be->beconn > s->be->beconn_max)
2051 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01002052 proxy_inc_be_ctr(s->be);
Willy Tarreau6e4261e2007-09-18 18:36:05 +02002053
Willy Tarreau59234e92008-11-30 23:51:27 +01002054 /* assign new parameters to the session from the new backend */
2055 s->rep->rto = s->req->wto = s->be->timeout.server;
2056 s->req->cto = s->be->timeout.connect;
2057 s->conn_retries = s->be->conn_retries;
2058 s->flags |= SN_BE_ASSIGNED;
2059 break;
Willy Tarreau55ea7572007-06-17 19:56:27 +02002060 }
2061 }
Willy Tarreauf1221aa2006-12-17 22:14:12 +01002062 }
2063
Willy Tarreau59234e92008-11-30 23:51:27 +01002064 if (!(s->flags & SN_BE_ASSIGNED) && cur_proxy->defbe.be) {
2065 /* No backend was set, but there was a default
2066 * backend set in the frontend, so we use it and
2067 * loop again.
2068 */
2069 s->be = cur_proxy->defbe.be;
2070 s->be->beconn++;
2071 if (s->be->beconn > s->be->beconn_max)
2072 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01002073 proxy_inc_be_ctr(s->be);
Willy Tarreau58f10d72006-12-04 02:26:12 +01002074
Willy Tarreau59234e92008-11-30 23:51:27 +01002075 /* assign new parameters to the session from the new backend */
2076 s->rep->rto = s->req->wto = s->be->timeout.server;
2077 s->req->cto = s->be->timeout.connect;
2078 s->conn_retries = s->be->conn_retries;
2079 s->flags |= SN_BE_ASSIGNED;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01002080 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002081 } while (s->be != cur_proxy); /* we loop only if s->be has changed */
Willy Tarreau58f10d72006-12-04 02:26:12 +01002082
Willy Tarreau59234e92008-11-30 23:51:27 +01002083 if (!(s->flags & SN_BE_ASSIGNED)) {
2084 /* To ensure correct connection accounting on
2085 * the backend, we count the connection for the
2086 * one managing the queue.
Willy Tarreau06619262006-12-17 08:37:22 +01002087 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002088 s->be->beconn++;
2089 if (s->be->beconn > s->be->beconn_max)
2090 s->be->beconn_max = s->be->beconn;
Willy Tarreau7f062c42009-03-05 18:43:00 +01002091 proxy_inc_be_ctr(s->be);
Willy Tarreau59234e92008-11-30 23:51:27 +01002092 s->flags |= SN_BE_ASSIGNED;
2093 }
Willy Tarreau06619262006-12-17 08:37:22 +01002094
Willy Tarreau59234e92008-11-30 23:51:27 +01002095 /*
2096 * Right now, we know that we have processed the entire headers
2097 * and that unwanted requests have been filtered out. We can do
2098 * whatever we want with the remaining request. Also, now we
2099 * may have separate values for ->fe, ->be.
2100 */
Willy Tarreau06619262006-12-17 08:37:22 +01002101
Willy Tarreau59234e92008-11-30 23:51:27 +01002102 /*
2103 * If HTTP PROXY is set we simply get remote server address
2104 * parsing incoming request.
2105 */
2106 if ((s->be->options & PR_O_HTTP_PROXY) && !(s->flags & SN_ADDR_SET)) {
2107 url2sa(req->data + msg->sl.rq.u, msg->sl.rq.u_l, &s->srv_addr);
2108 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01002109
Willy Tarreau59234e92008-11-30 23:51:27 +01002110 /*
2111 * 7: the appsession cookie was looked up very early in 1.2,
2112 * so let's do the same now.
2113 */
Willy Tarreau58f10d72006-12-04 02:26:12 +01002114
Willy Tarreau59234e92008-11-30 23:51:27 +01002115 /* It needs to look into the URI */
2116 if (s->be->appsession_name) {
2117 get_srv_from_appsession(s, &req->data[msg->som], msg->sl.rq.l);
2118 }
Willy Tarreau45e73e32006-12-17 00:05:15 +01002119
Willy Tarreau59234e92008-11-30 23:51:27 +01002120 /*
2121 * 8: Now we can work with the cookies.
2122 * Note that doing so might move headers in the request, but
2123 * the fields will stay coherent and the URI will not move.
2124 * This should only be performed in the backend.
2125 */
Willy Tarreaufd39dda2008-10-17 12:01:58 +02002126 if ((s->be->cookie_name || s->be->appsession_name || s->fe->capture_name)
Willy Tarreau59234e92008-11-30 23:51:27 +01002127 && !(txn->flags & (TX_CLDENY|TX_CLTARPIT)))
2128 manage_client_side_cookies(s, req);
Willy Tarreau7ac51f62007-03-25 16:00:04 +02002129
Willy Tarreau59234e92008-11-30 23:51:27 +01002130 /*
2131 * 9: add X-Forwarded-For if either the frontend or the backend
2132 * asks for it.
2133 */
2134 if ((s->fe->options | s->be->options) & PR_O_FWDFOR) {
2135 if (s->cli_addr.ss_family == AF_INET) {
2136 /* Add an X-Forwarded-For header unless the source IP is
2137 * in the 'except' network range.
2138 */
2139 if ((!s->fe->except_mask.s_addr ||
2140 (((struct sockaddr_in *)&s->cli_addr)->sin_addr.s_addr & s->fe->except_mask.s_addr)
2141 != s->fe->except_net.s_addr) &&
2142 (!s->be->except_mask.s_addr ||
2143 (((struct sockaddr_in *)&s->cli_addr)->sin_addr.s_addr & s->be->except_mask.s_addr)
2144 != s->be->except_net.s_addr)) {
Willy Tarreau2a324282006-12-05 00:05:46 +01002145 int len;
Willy Tarreau59234e92008-11-30 23:51:27 +01002146 unsigned char *pn;
2147 pn = (unsigned char *)&((struct sockaddr_in *)&s->cli_addr)->sin_addr;
Ross Westaf72a1d2008-08-03 10:51:45 +02002148
2149 /* Note: we rely on the backend to get the header name to be used for
2150 * x-forwarded-for, because the header is really meant for the backends.
2151 * However, if the backend did not specify any option, we have to rely
2152 * on the frontend's header name.
2153 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002154 if (s->be->fwdfor_hdr_len) {
2155 len = s->be->fwdfor_hdr_len;
2156 memcpy(trash, s->be->fwdfor_hdr_name, len);
Ross Westaf72a1d2008-08-03 10:51:45 +02002157 } else {
Willy Tarreau59234e92008-11-30 23:51:27 +01002158 len = s->fe->fwdfor_hdr_len;
2159 memcpy(trash, s->fe->fwdfor_hdr_name, len);
2160 }
2161 len += sprintf(trash + len, ": %d.%d.%d.%d", pn[0], pn[1], pn[2], pn[3]);
Willy Tarreauedcf6682008-11-30 23:15:34 +01002162
Willy Tarreau4af6f3a2007-03-18 22:36:26 +01002163 if (unlikely(http_header_add_tail2(req, &txn->req,
2164 &txn->hdr_idx, trash, len)) < 0)
Willy Tarreau06619262006-12-17 08:37:22 +01002165 goto return_bad_req;
Willy Tarreau2a324282006-12-05 00:05:46 +01002166 }
2167 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002168 else if (s->cli_addr.ss_family == AF_INET6) {
2169 /* FIXME: for the sake of completeness, we should also support
2170 * 'except' here, although it is mostly useless in this case.
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002171 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002172 int len;
2173 char pn[INET6_ADDRSTRLEN];
2174 inet_ntop(AF_INET6,
2175 (const void *)&((struct sockaddr_in6 *)(&s->cli_addr))->sin6_addr,
2176 pn, sizeof(pn));
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002177
Willy Tarreau59234e92008-11-30 23:51:27 +01002178 /* Note: we rely on the backend to get the header name to be used for
2179 * x-forwarded-for, because the header is really meant for the backends.
2180 * However, if the backend did not specify any option, we have to rely
2181 * on the frontend's header name.
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002182 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002183 if (s->be->fwdfor_hdr_len) {
2184 len = s->be->fwdfor_hdr_len;
2185 memcpy(trash, s->be->fwdfor_hdr_name, len);
2186 } else {
2187 len = s->fe->fwdfor_hdr_len;
2188 memcpy(trash, s->fe->fwdfor_hdr_name, len);
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002189 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002190 len += sprintf(trash + len, ": %s", pn);
Willy Tarreauadfb8562008-08-11 15:24:42 +02002191
Willy Tarreau59234e92008-11-30 23:51:27 +01002192 if (unlikely(http_header_add_tail2(req, &txn->req,
2193 &txn->hdr_idx, trash, len)) < 0)
2194 goto return_bad_req;
2195 }
2196 }
2197
2198 /*
2199 * 10: add "Connection: close" if needed and not yet set.
2200 * Note that we do not need to add it in case of HTTP/1.0.
2201 */
2202 if (!(s->flags & SN_CONN_CLOSED) &&
2203 ((s->fe->options | s->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO))) {
2204 if ((unlikely(msg->sl.rq.v_l != 8) ||
2205 unlikely(req->data[msg->som + msg->sl.rq.v + 7] != '0')) &&
2206 unlikely(http_header_add_tail2(req, &txn->req, &txn->hdr_idx,
2207 "Connection: close", 17)) < 0)
2208 goto return_bad_req;
2209 s->flags |= SN_CONN_CLOSED;
2210 }
2211 /* Before we switch to data, was assignment set in manage_client_side_cookie?
2212 * If not assigned, perhaps we are balancing on url_param, but this is a
2213 * POST; and the parameters are in the body, maybe scan there to find our server.
2214 * (unless headers overflowed the buffer?)
2215 */
2216 if (!(s->flags & (SN_ASSIGNED|SN_DIRECT)) &&
2217 s->txn.meth == HTTP_METH_POST && s->be->url_param_name != NULL &&
2218 s->be->url_param_post_limit != 0 && !(req->flags & BF_FULL) &&
2219 memchr(msg->sol + msg->sl.rq.u, '?', msg->sl.rq.u_l) == NULL) {
2220 /* are there enough bytes here? total == l || r || rlim ?
2221 * len is unsigned, but eoh is int,
2222 * how many bytes of body have we received?
2223 * eoh is the first empty line of the header
2224 */
2225 /* already established CRLF or LF at eoh, move to start of message, find message length in buffer */
2226 unsigned long len = req->l - (msg->sol[msg->eoh] == '\r' ? msg->eoh + 2 : msg->eoh + 1);
2227
2228 /* If we have HTTP/1.1 and Expect: 100-continue, then abort.
2229 * We can't assume responsibility for the server's decision,
2230 * on this URI and header set. See rfc2616: 14.20, 8.2.3,
2231 * We also can't change our mind later, about which server to choose, so round robin.
2232 */
2233 if ((likely(msg->sl.rq.v_l == 8) && req->data[msg->som + msg->sl.rq.v + 7] == '1')) {
2234 struct hdr_ctx ctx;
2235 ctx.idx = 0;
2236 /* Expect is allowed in 1.1, look for it */
2237 http_find_header2("Expect", 6, msg->sol, &txn->hdr_idx, &ctx);
2238 if (ctx.idx != 0 &&
2239 unlikely(ctx.vlen == 12 && strncasecmp(ctx.line+ctx.val, "100-continue", 12) == 0))
2240 /* We can't reliablly stall and wait for data, because of
2241 * .NET clients that don't conform to rfc2616; so, no need for
2242 * the next block to check length expectations.
2243 * We could send 100 status back to the client, but then we need to
2244 * re-write headers, and send the message. And this isn't the right
2245 * place for that action.
2246 * TODO: support Expect elsewhere and delete this block.
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002247 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002248 goto end_check_maybe_wait_for_body;
2249 }
2250
2251 if (likely(len > s->be->url_param_post_limit)) {
2252 /* nothing to do, we got enough */
2253 } else {
2254 /* limit implies we are supposed to need this many bytes
2255 * to find the parameter. Let's see how many bytes we can wait for.
2256 */
2257 long long hint = len;
2258 struct hdr_ctx ctx;
2259 ctx.idx = 0;
2260 http_find_header2("Transfer-Encoding", 17, msg->sol, &txn->hdr_idx, &ctx);
2261 if (ctx.idx && ctx.vlen >= 7 && strncasecmp(ctx.line+ctx.val, "chunked", 7) == 0) {
2262 buffer_write_dis(req);
2263 req->analysers |= AN_REQ_HTTP_BODY;
2264 }
2265 else {
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002266 ctx.idx = 0;
Willy Tarreau59234e92008-11-30 23:51:27 +01002267 http_find_header2("Content-Length", 14, msg->sol, &txn->hdr_idx, &ctx);
2268 /* now if we have a length, we'll take the hint */
2269 if (ctx.idx) {
2270 /* We have Content-Length */
2271 if (strl2llrc(ctx.line+ctx.val,ctx.vlen, &hint))
2272 hint = 0; /* parse failure, untrusted client */
2273 else {
2274 if (hint > 0)
2275 msg->hdr_content_len = hint;
2276 else
2277 hint = 0; /* bad client, sent negative length */
2278 }
2279 }
2280 /* but limited to what we care about, maybe we don't expect any entity data (hint == 0) */
2281 if (s->be->url_param_post_limit < hint)
2282 hint = s->be->url_param_post_limit;
2283 /* now do we really need to buffer more data? */
2284 if (len < hint) {
Willy Tarreau3da77c52008-08-29 09:58:42 +02002285 buffer_write_dis(req);
Willy Tarreau2df28e82008-08-17 15:20:19 +02002286 req->analysers |= AN_REQ_HTTP_BODY;
Willy Tarreau41f40ed2008-08-21 10:05:00 +02002287 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002288 /* else... There are no body bytes to wait for */
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02002289 }
2290 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002291 }
2292 end_check_maybe_wait_for_body:
Willy Tarreaubaaee002006-06-26 02:48:02 +02002293
Willy Tarreau59234e92008-11-30 23:51:27 +01002294 /*************************************************************
2295 * OK, that's finished for the headers. We have done what we *
2296 * could. Let's switch to the DATA state. *
2297 ************************************************************/
Willy Tarreaubaaee002006-06-26 02:48:02 +02002298
Willy Tarreau59234e92008-11-30 23:51:27 +01002299 buffer_set_rlim(req, BUFSIZE); /* no more rewrite needed */
2300 s->logs.tv_request = now;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002301
Willy Tarreau59234e92008-11-30 23:51:27 +01002302 /* When a connection is tarpitted, we use the tarpit timeout,
2303 * which may be the same as the connect timeout if unspecified.
2304 * If unset, then set it to zero because we really want it to
2305 * eventually expire. We build the tarpit as an analyser.
2306 */
2307 if (txn->flags & TX_CLTARPIT) {
Willy Tarreau6f0aa472009-03-08 20:33:29 +01002308 buffer_erase(s->req);
2309 /* wipe the request out so that we can drop the connection early
Willy Tarreau59234e92008-11-30 23:51:27 +01002310 * if the client closes first.
Willy Tarreau2a324282006-12-05 00:05:46 +01002311 */
Willy Tarreau59234e92008-11-30 23:51:27 +01002312 buffer_write_dis(req);
2313 req->analysers |= AN_REQ_HTTP_TARPIT;
2314 req->analyse_exp = tick_add_ifset(now_ms, s->be->timeout.tarpit);
2315 if (!req->analyse_exp)
Willy Tarreau2ab85e62009-03-29 10:24:15 +02002316 req->analyse_exp = tick_add(now_ms, 0);
Willy Tarreau59234e92008-11-30 23:51:27 +01002317 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002318
Willy Tarreau59234e92008-11-30 23:51:27 +01002319 /* OK let's go on with the BODY now */
2320 return 1;
Willy Tarreau06619262006-12-17 08:37:22 +01002321
Willy Tarreau59234e92008-11-30 23:51:27 +01002322 return_bad_req: /* let's centralize all bad requests */
Willy Tarreauf073a832009-03-01 23:21:47 +01002323 if (unlikely(msg->msg_state == HTTP_MSG_ERROR)) {
2324 /* we detected a parsing error. We want to archive this request
2325 * in the dedicated proxy area for later troubleshooting.
2326 */
2327 struct error_snapshot *es = &s->fe->invalid_req;
2328 int maxlen = MIN(req->r - req->data + msg->som, sizeof(es->buf));
2329 memcpy(es->buf, req->data + msg->som, maxlen);
2330 es->pos = req->lr - req->data + msg->som;
2331 es->len = req->r - req->data + msg->som;
Willy Tarreaudefc52d2009-03-04 20:53:44 +01002332 es->when = date; // user-visible date
Willy Tarreauf073a832009-03-01 23:21:47 +01002333 es->sid = s->uniq_id;
2334 es->srv = s->srv;
2335 es->oe = s->be;
2336 es->src = s->cli_addr;
2337 }
Willy Tarreau59234e92008-11-30 23:51:27 +01002338 txn->req.msg_state = HTTP_MSG_ERROR;
2339 txn->status = 400;
2340 req->analysers = 0;
2341 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_400));
2342 s->fe->failed_req++;
Willy Tarreauadfb8562008-08-11 15:24:42 +02002343
Willy Tarreau59234e92008-11-30 23:51:27 +01002344 return_prx_cond:
2345 if (!(s->flags & SN_ERR_MASK))
2346 s->flags |= SN_ERR_PRXCOND;
2347 if (!(s->flags & SN_FINST_MASK))
2348 s->flags |= SN_FINST_R;
Willy Tarreaudafde432008-08-17 01:00:46 +02002349 return 0;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002350}
Willy Tarreauadfb8562008-08-11 15:24:42 +02002351
Willy Tarreau60b85b02008-11-30 23:28:40 +01002352/* This function is an analyser which processes the HTTP tarpit. It always
2353 * returns zero, at the beginning because it prevents any other processing
2354 * from occurring, and at the end because it terminates the request.
2355 */
2356int http_process_tarpit(struct session *s, struct buffer *req)
2357{
2358 struct http_txn *txn = &s->txn;
2359
2360 /* This connection is being tarpitted. The CLIENT side has
2361 * already set the connect expiration date to the right
2362 * timeout. We just have to check that the client is still
2363 * there and that the timeout has not expired.
2364 */
2365 if ((req->flags & (BF_SHUTR|BF_READ_ERROR)) == 0 &&
2366 !tick_is_expired(req->analyse_exp, now_ms))
2367 return 0;
2368
2369 /* We will set the queue timer to the time spent, just for
2370 * logging purposes. We fake a 500 server error, so that the
2371 * attacker will not suspect his connection has been tarpitted.
2372 * It will not cause trouble to the logs because we can exclude
2373 * the tarpitted connections by filtering on the 'PT' status flags.
2374 */
Willy Tarreau60b85b02008-11-30 23:28:40 +01002375 s->logs.t_queue = tv_ms_elapsed(&s->logs.tv_accept, &now);
2376
2377 txn->status = 500;
2378 if (req->flags != BF_READ_ERROR)
2379 stream_int_retnclose(req->prod, error_message(s, HTTP_ERR_500));
2380
2381 req->analysers = 0;
2382 req->analyse_exp = TICK_ETERNITY;
2383
2384 s->fe->failed_req++;
2385 if (!(s->flags & SN_ERR_MASK))
2386 s->flags |= SN_ERR_PRXCOND;
2387 if (!(s->flags & SN_FINST_MASK))
2388 s->flags |= SN_FINST_T;
2389 return 0;
2390}
2391
Willy Tarreaud34af782008-11-30 23:36:37 +01002392/* This function is an analyser which processes the HTTP request body. It looks
2393 * for parameters to be used for the load balancing algorithm (url_param). It
2394 * must only be called after the standard HTTP request processing has occurred,
2395 * because it expects the request to be parsed. It returns zero if it needs to
2396 * read more data, or 1 once it has completed its analysis.
2397 */
2398int http_process_request_body(struct session *s, struct buffer *req)
2399{
2400 struct http_msg *msg = &s->txn.req;
2401 unsigned long body = msg->sol[msg->eoh] == '\r' ? msg->eoh + 2 : msg->eoh + 1;
2402 long long limit = s->be->url_param_post_limit;
2403 struct hdr_ctx ctx;
2404
2405 /* We have to parse the HTTP request body to find any required data.
2406 * "balance url_param check_post" should have been the only way to get
2407 * into this. We were brought here after HTTP header analysis, so all
2408 * related structures are ready.
2409 */
2410
2411 ctx.idx = 0;
2412
2413 /* now if we have a length, we'll take the hint */
2414 http_find_header2("Transfer-Encoding", 17, msg->sol, &s->txn.hdr_idx, &ctx);
2415 if (ctx.idx && ctx.vlen >= 7 && strncasecmp(ctx.line+ctx.val, "chunked", 7) == 0) {
2416 unsigned int chunk = 0;
2417 while (body < req->l && !HTTP_IS_CRLF(msg->sol[body])) {
2418 char c = msg->sol[body];
2419 if (ishex(c)) {
2420 unsigned int hex = toupper(c) - '0';
2421 if (hex > 9)
2422 hex -= 'A' - '9' - 1;
2423 chunk = (chunk << 4) | hex;
2424 } else
2425 break;
2426 body++;
2427 }
2428 if (body + 2 >= req->l) /* we want CRLF too */
2429 goto http_body_end; /* end of buffer? data missing! */
2430
2431 if (memcmp(msg->sol+body, "\r\n", 2) != 0)
2432 goto http_body_end; /* chunked encoding len ends with CRLF, and we don't have it yet */
2433
2434 body += 2; // skip CRLF
2435
2436 /* if we support more then one chunk here, we have to do it again when assigning server
2437 * 1. how much entity data do we have? new var
2438 * 2. should save entity_start, entity_cursor, elen & rlen in req; so we don't repeat scanning here
2439 * 3. test if elen > limit, or set new limit to elen if 0 (end of entity found)
2440 */
2441
2442 if (chunk < limit)
2443 limit = chunk; /* only reading one chunk */
2444 } else {
2445 if (msg->hdr_content_len < limit)
2446 limit = msg->hdr_content_len;
2447 }
2448
2449 http_body_end:
2450 /* we leave once we know we have nothing left to do. This means that we have
2451 * enough bytes, or that we know we'll not get any more (buffer full, read
2452 * buffer closed).
2453 */
2454 if (req->l - body >= limit || /* enough bytes! */
2455 req->flags & (BF_FULL | BF_READ_ERROR | BF_SHUTR | BF_READ_TIMEOUT) ||
2456 tick_is_expired(req->analyse_exp, now_ms)) {
2457 /* The situation will not evolve, so let's give up on the analysis. */
2458 s->logs.tv_request = now; /* update the request timer to reflect full request */
2459 req->analysers &= ~AN_REQ_HTTP_BODY;
2460 req->analyse_exp = TICK_ETERNITY;
2461 return 1;
2462 }
2463 else {
2464 /* Not enough data. We'll re-use the http-request
2465 * timeout here. Ideally, we should set the timeout
2466 * relative to the accept() date. We just set the
2467 * request timeout once at the beginning of the
2468 * request.
2469 */
2470 buffer_write_dis(req);
2471 if (!tick_isset(req->analyse_exp))
2472 req->analyse_exp = tick_add_ifset(now_ms, s->fe->timeout.httpreq);
2473 return 0;
2474 }
2475}
2476
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002477/* This function performs all the processing enabled for the current response.
Willy Tarreaudafde432008-08-17 01:00:46 +02002478 * It normally returns zero, but may return 1 if it absolutely needs to be
2479 * called again after other functions. It relies on buffers flags, and updates
Willy Tarreau2df28e82008-08-17 15:20:19 +02002480 * t->rep->analysers. It might make sense to explode it into several other
Willy Tarreau41f40ed2008-08-21 10:05:00 +02002481 * functions. It works like process_request (see indications above).
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002482 */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002483int process_response(struct session *t)
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002484{
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002485 struct http_txn *txn = &t->txn;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002486 struct buffer *req = t->req;
2487 struct buffer *rep = t->rep;
Willy Tarreauadfb8562008-08-11 15:24:42 +02002488
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02002489 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 +02002490 now_ms, __FUNCTION__,
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02002491 t,
2492 rep,
2493 rep->rex, rep->wex,
2494 rep->flags,
2495 rep->l,
2496 rep->analysers);
Willy Tarreau67f0eea2008-08-10 22:55:22 +02002497
Willy Tarreau2df28e82008-08-17 15:20:19 +02002498 if (rep->analysers & AN_RTR_HTTP_HDR) { /* receiving server headers */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002499 /*
2500 * Now parse the partial (or complete) lines.
2501 * We will check the response syntax, and also join multi-line
2502 * headers. An index of all the lines will be elaborated while
2503 * parsing.
2504 *
2505 * For the parsing, we use a 28 states FSM.
2506 *
2507 * Here is the information we currently have :
Willy Tarreaue393fe22008-08-16 22:18:07 +02002508 * rep->data + rep->som = beginning of response
2509 * rep->data + rep->eoh = end of processed headers / start of current one
2510 * rep->data + rep->eol = end of current header or line (LF or CRLF)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002511 * rep->lr = first non-visited byte
2512 * rep->r = end of data
2513 */
Willy Tarreau7f875f62008-08-11 17:35:01 +02002514
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002515 int cur_idx;
2516 struct http_msg *msg = &txn->rsp;
2517 struct proxy *cur_proxy;
2518
2519 if (likely(rep->lr < rep->r))
2520 http_msg_analyzer(rep, msg, &txn->hdr_idx);
2521
2522 /* 1: we might have to print this header in debug mode */
2523 if (unlikely((global.mode & MODE_DEBUG) &&
2524 (!(global.mode & MODE_QUIET) || (global.mode & MODE_VERBOSE)) &&
2525 (msg->msg_state == HTTP_MSG_BODY || msg->msg_state == HTTP_MSG_ERROR))) {
2526 char *eol, *sol;
2527
2528 sol = rep->data + msg->som;
2529 eol = sol + msg->sl.rq.l;
2530 debug_hdr("srvrep", t, sol, eol);
2531
2532 sol += hdr_idx_first_pos(&txn->hdr_idx);
2533 cur_idx = hdr_idx_first_idx(&txn->hdr_idx);
2534
2535 while (cur_idx) {
2536 eol = sol + txn->hdr_idx.v[cur_idx].len;
2537 debug_hdr("srvhdr", t, sol, eol);
2538 sol = eol + txn->hdr_idx.v[cur_idx].cr + 1;
2539 cur_idx = txn->hdr_idx.v[cur_idx].next;
Willy Tarreauc65a3ba2008-08-11 23:42:50 +02002540 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002541 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002542
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002543 /*
2544 * Now we quickly check if we have found a full valid response.
2545 * If not so, we check the FD and buffer states before leaving.
2546 * A full response is indicated by the fact that we have seen
2547 * the double LF/CRLF, so the state is HTTP_MSG_BODY. Invalid
2548 * responses are checked first.
2549 *
2550 * Depending on whether the client is still there or not, we
2551 * may send an error response back or not. Note that normally
2552 * we should only check for HTTP status there, and check I/O
2553 * errors somewhere else.
2554 */
2555
2556 if (unlikely(msg->msg_state != HTTP_MSG_BODY)) {
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002557 /* Invalid response */
2558 if (unlikely(msg->msg_state == HTTP_MSG_ERROR)) {
Willy Tarreauf073a832009-03-01 23:21:47 +01002559 /* we detected a parsing error. We want to archive this response
2560 * in the dedicated proxy area for later troubleshooting.
2561 */
2562 struct error_snapshot *es = &t->be->invalid_rep;
2563 int maxlen = MIN(rep->r - rep->data + msg->som, sizeof(es->buf));
2564 memcpy(es->buf, rep->data + msg->som, maxlen);
2565 es->pos = rep->lr - rep->data + msg->som;
2566 es->len = rep->r - rep->data + msg->som;
Willy Tarreaudefc52d2009-03-04 20:53:44 +01002567 es->when = date; // user-visible date
Willy Tarreauf073a832009-03-01 23:21:47 +01002568 es->sid = t->uniq_id;
2569 es->srv = t->srv;
2570 es->oe = t->fe;
2571 es->src = t->cli_addr;
2572
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002573 hdr_response_bad:
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002574 buffer_shutr_now(rep);
2575 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002576 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002577 t->srv->failed_resp++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002578 t->be->failed_resp++;
Willy Tarreau2df28e82008-08-17 15:20:19 +02002579 rep->analysers = 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002580 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002581 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002582 if (!(t->flags & SN_ERR_MASK))
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002583 t->flags |= SN_ERR_PRXCOND;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002584 if (!(t->flags & SN_FINST_MASK))
2585 t->flags |= SN_FINST_H;
Willy Tarreau461f6622008-08-15 23:43:19 +02002586
Willy Tarreaudafde432008-08-17 01:00:46 +02002587 return 0;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002588 }
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002589 /* too large response does not fit in buffer. */
2590 else if (rep->flags & BF_FULL) {
2591 goto hdr_response_bad;
2592 }
2593 /* read error */
2594 else if (rep->flags & BF_READ_ERROR) {
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002595 buffer_shutr_now(rep);
2596 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002597 if (t->srv)
2598 t->srv->failed_resp++;
2599 t->be->failed_resp++;
Willy Tarreau2df28e82008-08-17 15:20:19 +02002600 rep->analysers = 0;
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002601 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002602 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002603 if (!(t->flags & SN_ERR_MASK))
2604 t->flags |= SN_ERR_SRVCL;
2605 if (!(t->flags & SN_FINST_MASK))
2606 t->flags |= SN_FINST_H;
Willy Tarreaudafde432008-08-17 01:00:46 +02002607 return 0;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002608 }
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002609 /* read timeout : return a 504 to the client. */
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002610 else if (rep->flags & BF_READ_TIMEOUT) {
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002611 buffer_shutr_now(rep);
2612 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002613 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002614 t->srv->failed_resp++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002615 t->be->failed_resp++;
Willy Tarreau2df28e82008-08-17 15:20:19 +02002616 rep->analysers = 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002617 txn->status = 504;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002618 stream_int_return(rep->cons, error_message(t, HTTP_ERR_504));
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002619 if (!(t->flags & SN_ERR_MASK))
2620 t->flags |= SN_ERR_SRVTO;
2621 if (!(t->flags & SN_FINST_MASK))
2622 t->flags |= SN_FINST_H;
Willy Tarreaudafde432008-08-17 01:00:46 +02002623 return 0;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002624 }
Willy Tarreau8365f932009-03-15 23:11:49 +01002625 /* close from server */
2626 else if (rep->flags & BF_SHUTR) {
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002627 buffer_shutw_now(req);
Willy Tarreau8365f932009-03-15 23:11:49 +01002628 if (t->srv)
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002629 t->srv->failed_resp++;
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002630 t->be->failed_resp++;
2631 rep->analysers = 0;
2632 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002633 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002634 if (!(t->flags & SN_ERR_MASK))
2635 t->flags |= SN_ERR_SRVCL;
2636 if (!(t->flags & SN_FINST_MASK))
2637 t->flags |= SN_FINST_H;
Willy Tarreauf9839bd2008-08-27 23:57:16 +02002638 return 0;
2639 }
Willy Tarreau8365f932009-03-15 23:11:49 +01002640 /* write error to client (we don't send any message then) */
2641 else if (rep->flags & BF_WRITE_ERROR) {
2642 buffer_shutr_now(rep);
2643 t->be->failed_resp++;
2644 rep->analysers = 0;
2645 if (!(t->flags & SN_ERR_MASK))
2646 t->flags |= SN_ERR_CLICL;
2647 if (!(t->flags & SN_FINST_MASK))
2648 t->flags |= SN_FINST_H;
2649 return 0;
2650 }
Willy Tarreau3da77c52008-08-29 09:58:42 +02002651 buffer_write_dis(rep);
Willy Tarreaucebf57e2008-08-15 18:16:37 +02002652 return 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002653 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002654
Willy Tarreau21d2af32008-02-14 20:25:24 +01002655
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002656 /*****************************************************************
2657 * More interesting part now : we know that we have a complete *
2658 * response which at least looks like HTTP. We have an indicator *
2659 * of each header's length, so we can parse them quickly. *
2660 ****************************************************************/
Willy Tarreau21d2af32008-02-14 20:25:24 +01002661
Willy Tarreau2df28e82008-08-17 15:20:19 +02002662 rep->analysers &= ~AN_RTR_HTTP_HDR;
Willy Tarreaua7c52762008-08-16 18:40:18 +02002663
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002664 /* ensure we keep this pointer to the beginning of the message */
2665 msg->sol = rep->data + msg->som;
Willy Tarreau21d2af32008-02-14 20:25:24 +01002666
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002667 /*
2668 * 1: get the status code and check for cacheability.
2669 */
Willy Tarreau21d2af32008-02-14 20:25:24 +01002670
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002671 t->logs.logwait &= ~LW_RESP;
2672 txn->status = strl2ui(rep->data + msg->sl.st.c, msg->sl.st.c_l);
Willy Tarreau21d2af32008-02-14 20:25:24 +01002673
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002674 switch (txn->status) {
2675 case 200:
2676 case 203:
2677 case 206:
2678 case 300:
2679 case 301:
2680 case 410:
2681 /* RFC2616 @13.4:
2682 * "A response received with a status code of
2683 * 200, 203, 206, 300, 301 or 410 MAY be stored
2684 * by a cache (...) unless a cache-control
2685 * directive prohibits caching."
2686 *
2687 * RFC2616 @9.5: POST method :
2688 * "Responses to this method are not cacheable,
2689 * unless the response includes appropriate
2690 * Cache-Control or Expires header fields."
2691 */
2692 if (likely(txn->meth != HTTP_METH_POST) &&
2693 (t->be->options & (PR_O_CHK_CACHE|PR_O_COOK_NOC)))
2694 txn->flags |= TX_CACHEABLE | TX_CACHE_COOK;
2695 break;
2696 default:
2697 break;
2698 }
Willy Tarreau21d2af32008-02-14 20:25:24 +01002699
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002700 /*
2701 * 2: we may need to capture headers
2702 */
2703 if (unlikely((t->logs.logwait & LW_RSPHDR) && t->fe->rsp_cap))
2704 capture_headers(rep->data + msg->som, &txn->hdr_idx,
2705 txn->rsp.cap, t->fe->rsp_cap);
2706
2707 /*
2708 * 3: we will have to evaluate the filters.
2709 * As opposed to version 1.2, now they will be evaluated in the
2710 * filters order and not in the header order. This means that
2711 * each filter has to be validated among all headers.
2712 *
2713 * Filters are tried with ->be first, then with ->fe if it is
2714 * different from ->be.
2715 */
2716
2717 t->flags &= ~SN_CONN_CLOSED; /* prepare for inspection */
2718
2719 cur_proxy = t->be;
2720 while (1) {
2721 struct proxy *rule_set = cur_proxy;
2722
2723 /* try headers filters */
2724 if (rule_set->rsp_exp != NULL) {
2725 if (apply_filters_to_response(t, rep, rule_set->rsp_exp) < 0) {
2726 return_bad_resp:
Willy Tarreau8365f932009-03-15 23:11:49 +01002727 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002728 t->srv->failed_resp++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002729 cur_proxy->failed_resp++;
2730 return_srv_prx_502:
Willy Tarreaufa7e1022008-10-19 07:30:41 +02002731 buffer_shutr_now(rep);
2732 buffer_shutw_now(req);
Willy Tarreau2df28e82008-08-17 15:20:19 +02002733 rep->analysers = 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002734 txn->status = 502;
Willy Tarreau81acfab2008-11-30 19:22:53 +01002735 stream_int_return(rep->cons, error_message(t, HTTP_ERR_502));
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002736 if (!(t->flags & SN_ERR_MASK))
2737 t->flags |= SN_ERR_PRXCOND;
2738 if (!(t->flags & SN_FINST_MASK))
2739 t->flags |= SN_FINST_H;
Willy Tarreaudafde432008-08-17 01:00:46 +02002740 return 0;
Willy Tarreau21d2af32008-02-14 20:25:24 +01002741 }
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002742 }
Willy Tarreau21d2af32008-02-14 20:25:24 +01002743
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002744 /* has the response been denied ? */
2745 if (txn->flags & TX_SVDENY) {
Willy Tarreau8365f932009-03-15 23:11:49 +01002746 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002747 t->srv->failed_secu++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002748 cur_proxy->denied_resp++;
2749 goto return_srv_prx_502;
Willy Tarreau51406232008-03-10 22:04:20 +01002750 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002751
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002752 /* We might have to check for "Connection:" */
2753 if (((t->fe->options | t->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO)) &&
2754 !(t->flags & SN_CONN_CLOSED)) {
2755 char *cur_ptr, *cur_end, *cur_next;
2756 int cur_idx, old_idx, delta, val;
2757 struct hdr_idx_elem *cur_hdr;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002758
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002759 cur_next = rep->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
2760 old_idx = 0;
Willy Tarreau541b5c22008-01-06 23:34:21 +01002761
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002762 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
2763 cur_hdr = &txn->hdr_idx.v[cur_idx];
2764 cur_ptr = cur_next;
2765 cur_end = cur_ptr + cur_hdr->len;
2766 cur_next = cur_end + cur_hdr->cr + 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002767
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002768 val = http_header_match2(cur_ptr, cur_end, "Connection", 10);
2769 if (val) {
2770 /* 3 possibilities :
2771 * - we have already set Connection: close,
2772 * so we remove this line.
2773 * - we have not yet set Connection: close,
2774 * but this line indicates close. We leave
2775 * it untouched and set the flag.
2776 * - we have not yet set Connection: close,
2777 * and this line indicates non-close. We
2778 * replace it.
2779 */
2780 if (t->flags & SN_CONN_CLOSED) {
2781 delta = buffer_replace2(rep, cur_ptr, cur_next, NULL, 0);
2782 txn->rsp.eoh += delta;
2783 cur_next += delta;
2784 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
2785 txn->hdr_idx.used--;
2786 cur_hdr->len = 0;
2787 } else {
2788 if (strncasecmp(cur_ptr + val, "close", 5) != 0) {
2789 delta = buffer_replace2(rep, cur_ptr + val, cur_end,
2790 "close", 5);
2791 cur_next += delta;
2792 cur_hdr->len += delta;
2793 txn->rsp.eoh += delta;
2794 }
2795 t->flags |= SN_CONN_CLOSED;
2796 }
2797 }
2798 old_idx = cur_idx;
Willy Tarreau541b5c22008-01-06 23:34:21 +01002799 }
2800 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002801
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002802 /* add response headers from the rule sets in the same order */
2803 for (cur_idx = 0; cur_idx < rule_set->nb_rspadd; cur_idx++) {
2804 if (unlikely(http_header_add_tail(rep, &txn->rsp, &txn->hdr_idx,
2805 rule_set->rsp_add[cur_idx])) < 0)
2806 goto return_bad_resp;
Willy Tarreau0bbc3cf2006-10-15 14:26:02 +02002807 }
2808
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002809 /* check whether we're already working on the frontend */
2810 if (cur_proxy == t->fe)
2811 break;
2812 cur_proxy = t->fe;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002813 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002814
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002815 /*
2816 * 4: check for server cookie.
2817 */
Willy Tarreaufd39dda2008-10-17 12:01:58 +02002818 if (t->be->cookie_name || t->be->appsession_name || t->fe->capture_name
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002819 || (t->be->options & PR_O_CHK_CACHE))
2820 manage_server_side_cookies(t, rep);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002821
Willy Tarreaubaaee002006-06-26 02:48:02 +02002822
Willy Tarreaua15645d2007-03-18 16:22:39 +01002823 /*
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002824 * 5: check for cache-control or pragma headers if required.
Willy Tarreaua15645d2007-03-18 16:22:39 +01002825 */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002826 if ((t->be->options & (PR_O_COOK_NOC | PR_O_CHK_CACHE)) != 0)
2827 check_response_for_cacheability(t, rep);
Willy Tarreaua15645d2007-03-18 16:22:39 +01002828
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002829 /*
2830 * 6: add server cookie in the response if needed
2831 */
2832 if ((t->srv) && !(t->flags & SN_DIRECT) && (t->be->options & PR_O_COOK_INS) &&
2833 (!(t->be->options & PR_O_COOK_POST) || (txn->meth == HTTP_METH_POST))) {
2834 int len;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002835
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002836 /* the server is known, it's not the one the client requested, we have to
2837 * insert a set-cookie here, except if we want to insert only on POST
2838 * requests and this one isn't. Note that servers which don't have cookies
2839 * (eg: some backup servers) will return a full cookie removal request.
2840 */
2841 len = sprintf(trash, "Set-Cookie: %s=%s; path=/",
2842 t->be->cookie_name,
2843 t->srv->cookie ? t->srv->cookie : "; Expires=Thu, 01-Jan-1970 00:00:01 GMT");
Willy Tarreaubaaee002006-06-26 02:48:02 +02002844
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002845 if (t->be->cookie_domain)
2846 len += sprintf(trash+len, "; domain=%s", t->be->cookie_domain);
Willy Tarreaubaaee002006-06-26 02:48:02 +02002847
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002848 if (unlikely(http_header_add_tail2(rep, &txn->rsp, &txn->hdr_idx,
2849 trash, len)) < 0)
2850 goto return_bad_resp;
2851 txn->flags |= TX_SCK_INSERTED;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002852
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002853 /* Here, we will tell an eventual cache on the client side that we don't
2854 * want it to cache this reply because HTTP/1.0 caches also cache cookies !
2855 * Some caches understand the correct form: 'no-cache="set-cookie"', but
2856 * others don't (eg: apache <= 1.3.26). So we use 'private' instead.
2857 */
2858 if ((t->be->options & PR_O_COOK_NOC) && (txn->flags & TX_CACHEABLE)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02002859
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002860 txn->flags &= ~TX_CACHEABLE & ~TX_CACHE_COOK;
2861
2862 if (unlikely(http_header_add_tail2(rep, &txn->rsp, &txn->hdr_idx,
2863 "Cache-control: private", 22)) < 0)
2864 goto return_bad_resp;
Willy Tarreaua15645d2007-03-18 16:22:39 +01002865 }
2866 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002867
Willy Tarreaubaaee002006-06-26 02:48:02 +02002868
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002869 /*
2870 * 7: check if result will be cacheable with a cookie.
2871 * We'll block the response if security checks have caught
2872 * nasty things such as a cacheable cookie.
2873 */
2874 if (((txn->flags & (TX_CACHEABLE | TX_CACHE_COOK | TX_SCK_ANY)) ==
2875 (TX_CACHEABLE | TX_CACHE_COOK | TX_SCK_ANY)) &&
2876 (t->be->options & PR_O_CHK_CACHE)) {
2877
2878 /* we're in presence of a cacheable response containing
2879 * a set-cookie header. We'll block it as requested by
2880 * the 'checkcache' option, and send an alert.
Willy Tarreaua15645d2007-03-18 16:22:39 +01002881 */
Willy Tarreau8365f932009-03-15 23:11:49 +01002882 if (t->srv)
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002883 t->srv->failed_secu++;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002884 t->be->denied_resp++;
Willy Tarreaua15645d2007-03-18 16:22:39 +01002885
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002886 Alert("Blocking cacheable cookie in response from instance %s, server %s.\n",
2887 t->be->id, t->srv?t->srv->id:"<dispatch>");
2888 send_log(t->be, LOG_ALERT,
2889 "Blocking cacheable cookie in response from instance %s, server %s.\n",
2890 t->be->id, t->srv?t->srv->id:"<dispatch>");
2891 goto return_srv_prx_502;
2892 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01002893
2894 /*
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002895 * 8: add "Connection: close" if needed and not yet set.
2896 * Note that we do not need to add it in case of HTTP/1.0.
Willy Tarreaua15645d2007-03-18 16:22:39 +01002897 */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002898 if (!(t->flags & SN_CONN_CLOSED) &&
2899 ((t->fe->options | t->be->options) & (PR_O_HTTP_CLOSE|PR_O_FORCE_CLO))) {
2900 if ((unlikely(msg->sl.st.v_l != 8) ||
2901 unlikely(req->data[msg->som + 7] != '0')) &&
2902 unlikely(http_header_add_tail2(rep, &txn->rsp, &txn->hdr_idx,
2903 "Connection: close", 17)) < 0)
2904 goto return_bad_resp;
2905 t->flags |= SN_CONN_CLOSED;
2906 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01002907
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002908 /*************************************************************
2909 * OK, that's finished for the headers. We have done what we *
2910 * could. Let's switch to the DATA state. *
2911 ************************************************************/
Willy Tarreaubaaee002006-06-26 02:48:02 +02002912
Willy Tarreaue393fe22008-08-16 22:18:07 +02002913 buffer_set_rlim(rep, BUFSIZE); /* no more rewrite needed */
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002914 t->logs.t_data = tv_ms_elapsed(&t->logs.tv_accept, &now);
Willy Tarreaua15645d2007-03-18 16:22:39 +01002915
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002916#ifdef CONFIG_HAP_TCPSPLICE
2917 if ((t->fe->options & t->be->options) & PR_O_TCPSPLICE) {
2918 /* TCP splicing supported by both FE and BE */
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02002919 tcp_splice_splicefd(rep->cons->fd, rep->prod->fd, 0);
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002920 }
2921#endif
2922 /* if the user wants to log as soon as possible, without counting
2923 * bytes from the server, then this is the right moment. We have
2924 * to temporarily assign bytes_out to log what we currently have.
2925 */
2926 if (t->fe->to_log && !(t->logs.logwait & LW_BYTES)) {
2927 t->logs.t_close = t->logs.t_data; /* to get a valid end date */
2928 t->logs.bytes_out = txn->rsp.eoh;
Willy Tarreaua5555ec2008-11-30 19:02:32 +01002929 t->do_log(t);
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002930 t->logs.bytes_out = 0;
2931 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01002932
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002933 /* Note: we must not try to cheat by jumping directly to DATA,
2934 * otherwise we would not let the client side wake up.
2935 */
Willy Tarreaua15645d2007-03-18 16:22:39 +01002936
Willy Tarreaudafde432008-08-17 01:00:46 +02002937 return 0;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002938 }
Willy Tarreaudafde432008-08-17 01:00:46 +02002939
Willy Tarreau2df28e82008-08-17 15:20:19 +02002940 /* Note: eventhough nobody should set an unknown flag, clearing them right now will
2941 * probably reduce one day's debugging session.
2942 */
2943#ifdef DEBUG_DEV
2944 if (rep->analysers & ~(AN_RTR_HTTP_HDR)) {
2945 fprintf(stderr, "FIXME !!!! unknown analysers flags %s:%d = 0x%08X\n",
2946 __FILE__, __LINE__, rep->analysers);
2947 ABORT_NOW();
2948 }
2949#endif
2950 rep->analysers &= AN_RTR_HTTP_HDR;
Willy Tarreauf5483bf2008-08-14 18:35:40 +02002951 return 0;
2952}
Willy Tarreaua15645d2007-03-18 16:22:39 +01002953
Willy Tarreaubaaee002006-06-26 02:48:02 +02002954/*
2955 * Produces data for the session <s> depending on its source. Expects to be
Willy Tarreau1ae3a052008-08-16 10:56:30 +02002956 * called with client socket shut down on input. Right now, only statistics can
Willy Tarreau72b179a2008-08-28 16:01:32 +02002957 * be produced. It stops by itself by unsetting the BF_HIJACK flag from the
2958 * buffer, which it uses to keep on being called when there is free space in
Willy Tarreau01bf8672008-12-07 18:03:29 +01002959 * the buffer, or simply by letting an empty buffer upon return.
Willy Tarreaubaaee002006-06-26 02:48:02 +02002960 */
Willy Tarreau01bf8672008-12-07 18:03:29 +01002961void produce_content(struct session *s, struct buffer *rep)
Willy Tarreaubaaee002006-06-26 02:48:02 +02002962{
Willy Tarreaubaaee002006-06-26 02:48:02 +02002963 if (s->data_source == DATA_SRC_NONE) {
Willy Tarreau01bf8672008-12-07 18:03:29 +01002964 buffer_stop_hijack(rep);
2965 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002966 }
2967 else if (s->data_source == DATA_SRC_STATS) {
Willy Tarreauc0dde7a2007-01-01 21:38:07 +01002968 /* dump server statistics */
Willy Tarreau0a464892008-12-07 18:30:00 +01002969 int ret = stats_dump_http(s, rep, s->be->uri_auth);
Willy Tarreau91861262007-10-17 17:06:05 +02002970 if (ret >= 0)
Willy Tarreau01bf8672008-12-07 18:03:29 +01002971 return;
Willy Tarreau91861262007-10-17 17:06:05 +02002972 /* -1 indicates an error */
Willy Tarreauc0dde7a2007-01-01 21:38:07 +01002973 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02002974
Willy Tarreau91861262007-10-17 17:06:05 +02002975 /* unknown data source or internal error */
2976 s->txn.status = 500;
Willy Tarreau01bf8672008-12-07 18:03:29 +01002977 stream_int_retnclose(rep->cons, error_message(s, HTTP_ERR_500));
Willy Tarreau91861262007-10-17 17:06:05 +02002978 if (!(s->flags & SN_ERR_MASK))
2979 s->flags |= SN_ERR_PRXCOND;
2980 if (!(s->flags & SN_FINST_MASK))
2981 s->flags |= SN_FINST_R;
Willy Tarreau01bf8672008-12-07 18:03:29 +01002982 buffer_stop_hijack(rep);
2983 return;
Willy Tarreaubaaee002006-06-26 02:48:02 +02002984}
2985
2986
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002987/* Iterate the same filter through all request headers.
2988 * Returns 1 if this filter can be stopped upon return, otherwise 0.
Willy Tarreaua15645d2007-03-18 16:22:39 +01002989 * Since it can manage the switch to another backend, it updates the per-proxy
2990 * DENY stats.
Willy Tarreau58f10d72006-12-04 02:26:12 +01002991 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002992int apply_filter_to_req_headers(struct session *t, struct buffer *req, struct hdr_exp *exp)
Willy Tarreau58f10d72006-12-04 02:26:12 +01002993{
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002994 char term;
2995 char *cur_ptr, *cur_end, *cur_next;
2996 int cur_idx, old_idx, last_hdr;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01002997 struct http_txn *txn = &t->txn;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01002998 struct hdr_idx_elem *cur_hdr;
2999 int len, delta;
Willy Tarreau0f7562b2007-01-07 15:46:13 +01003000
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003001 last_hdr = 0;
3002
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003003 cur_next = req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003004 old_idx = 0;
3005
3006 while (!last_hdr) {
Willy Tarreau3d300592007-03-18 18:34:41 +01003007 if (unlikely(txn->flags & (TX_CLDENY | TX_CLTARPIT)))
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003008 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003009 else if (unlikely(txn->flags & TX_CLALLOW) &&
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003010 (exp->action == ACT_ALLOW ||
3011 exp->action == ACT_DENY ||
3012 exp->action == ACT_TARPIT))
3013 return 0;
3014
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003015 cur_idx = txn->hdr_idx.v[old_idx].next;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003016 if (!cur_idx)
3017 break;
3018
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003019 cur_hdr = &txn->hdr_idx.v[cur_idx];
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003020 cur_ptr = cur_next;
3021 cur_end = cur_ptr + cur_hdr->len;
3022 cur_next = cur_end + cur_hdr->cr + 1;
3023
3024 /* Now we have one header between cur_ptr and cur_end,
3025 * and the next header starts at cur_next.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003026 */
3027
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003028 /* The annoying part is that pattern matching needs
3029 * that we modify the contents to null-terminate all
3030 * strings before testing them.
3031 */
3032
3033 term = *cur_end;
3034 *cur_end = '\0';
3035
3036 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3037 switch (exp->action) {
3038 case ACT_SETBE:
3039 /* It is not possible to jump a second time.
3040 * FIXME: should we return an HTTP/500 here so that
3041 * the admin knows there's a problem ?
3042 */
3043 if (t->be != t->fe)
3044 break;
3045
3046 /* Swithing Proxy */
3047 t->be = (struct proxy *) exp->replace;
3048
matt.farnsworth@nokia.com1c2ab962008-04-14 20:47:37 +02003049 /* right now, the backend switch is not overly complicated
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003050 * because we have associated req_cap and rsp_cap to the
3051 * frontend, and the beconn will be updated later.
3052 */
3053
Willy Tarreaud7c30f92007-12-03 01:38:36 +01003054 t->rep->rto = t->req->wto = t->be->timeout.server;
3055 t->req->cto = t->be->timeout.connect;
Willy Tarreau6e4261e2007-09-18 18:36:05 +02003056 t->conn_retries = t->be->conn_retries;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003057 last_hdr = 1;
3058 break;
3059
3060 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003061 txn->flags |= TX_CLALLOW;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003062 last_hdr = 1;
3063 break;
3064
3065 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003066 txn->flags |= TX_CLDENY;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003067 last_hdr = 1;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003068 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003069 break;
3070
3071 case ACT_TARPIT:
Willy Tarreau3d300592007-03-18 18:34:41 +01003072 txn->flags |= TX_CLTARPIT;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003073 last_hdr = 1;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003074 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003075 break;
3076
3077 case ACT_REPLACE:
3078 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3079 delta = buffer_replace2(req, cur_ptr, cur_end, trash, len);
3080 /* FIXME: if the user adds a newline in the replacement, the
3081 * index will not be recalculated for now, and the new line
3082 * will not be counted as a new header.
3083 */
3084
3085 cur_end += delta;
3086 cur_next += delta;
3087 cur_hdr->len += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003088 txn->req.eoh += delta;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003089 break;
3090
3091 case ACT_REMOVE:
3092 delta = buffer_replace2(req, cur_ptr, cur_next, NULL, 0);
3093 cur_next += delta;
3094
3095 /* FIXME: this should be a separate function */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003096 txn->req.eoh += delta;
3097 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3098 txn->hdr_idx.used--;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003099 cur_hdr->len = 0;
3100 cur_end = NULL; /* null-term has been rewritten */
3101 break;
3102
3103 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01003104 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003105 if (cur_end)
3106 *cur_end = term; /* restore the string terminator */
Willy Tarreau58f10d72006-12-04 02:26:12 +01003107
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003108 /* keep the link from this header to next one in case of later
3109 * removal of next header.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003110 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003111 old_idx = cur_idx;
3112 }
3113 return 0;
3114}
3115
3116
3117/* Apply the filter to the request line.
3118 * Returns 0 if nothing has been done, 1 if the filter has been applied,
3119 * or -1 if a replacement resulted in an invalid request line.
Willy Tarreaua15645d2007-03-18 16:22:39 +01003120 * Since it can manage the switch to another backend, it updates the per-proxy
3121 * DENY stats.
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003122 */
3123int apply_filter_to_req_line(struct session *t, struct buffer *req, struct hdr_exp *exp)
3124{
3125 char term;
3126 char *cur_ptr, *cur_end;
3127 int done;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003128 struct http_txn *txn = &t->txn;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003129 int len, delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003130
Willy Tarreau58f10d72006-12-04 02:26:12 +01003131
Willy Tarreau3d300592007-03-18 18:34:41 +01003132 if (unlikely(txn->flags & (TX_CLDENY | TX_CLTARPIT)))
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003133 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003134 else if (unlikely(txn->flags & TX_CLALLOW) &&
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003135 (exp->action == ACT_ALLOW ||
3136 exp->action == ACT_DENY ||
3137 exp->action == ACT_TARPIT))
3138 return 0;
3139 else if (exp->action == ACT_REMOVE)
3140 return 0;
3141
3142 done = 0;
3143
Willy Tarreau9cdde232007-05-02 20:58:19 +02003144 cur_ptr = req->data + txn->req.som; /* should be equal to txn->sol */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003145 cur_end = cur_ptr + txn->req.sl.rq.l;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003146
3147 /* Now we have the request line between cur_ptr and cur_end */
3148
3149 /* The annoying part is that pattern matching needs
3150 * that we modify the contents to null-terminate all
3151 * strings before testing them.
3152 */
3153
3154 term = *cur_end;
3155 *cur_end = '\0';
3156
3157 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3158 switch (exp->action) {
3159 case ACT_SETBE:
3160 /* It is not possible to jump a second time.
3161 * FIXME: should we return an HTTP/500 here so that
3162 * the admin knows there's a problem ?
Willy Tarreau58f10d72006-12-04 02:26:12 +01003163 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003164 if (t->be != t->fe)
3165 break;
3166
3167 /* Swithing Proxy */
3168 t->be = (struct proxy *) exp->replace;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003169
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003170 /* right now, the backend switch is not too much complicated
3171 * because we have associated req_cap and rsp_cap to the
3172 * frontend, and the beconn will be updated later.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003173 */
3174
Willy Tarreaud7c30f92007-12-03 01:38:36 +01003175 t->rep->rto = t->req->wto = t->be->timeout.server;
3176 t->req->cto = t->be->timeout.connect;
Willy Tarreau6e4261e2007-09-18 18:36:05 +02003177 t->conn_retries = t->be->conn_retries;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003178 done = 1;
3179 break;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003180
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003181 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003182 txn->flags |= TX_CLALLOW;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003183 done = 1;
3184 break;
Willy Tarreaua496b602006-12-17 23:15:24 +01003185
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003186 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003187 txn->flags |= TX_CLDENY;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003188 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003189 done = 1;
3190 break;
Willy Tarreaua496b602006-12-17 23:15:24 +01003191
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003192 case ACT_TARPIT:
Willy Tarreau3d300592007-03-18 18:34:41 +01003193 txn->flags |= TX_CLTARPIT;
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003194 t->be->denied_req++;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003195 done = 1;
3196 break;
Willy Tarreaua496b602006-12-17 23:15:24 +01003197
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003198 case ACT_REPLACE:
3199 *cur_end = term; /* restore the string terminator */
3200 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3201 delta = buffer_replace2(req, cur_ptr, cur_end, trash, len);
3202 /* FIXME: if the user adds a newline in the replacement, the
3203 * index will not be recalculated for now, and the new line
3204 * will not be counted as a new header.
3205 */
Willy Tarreaua496b602006-12-17 23:15:24 +01003206
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003207 txn->req.eoh += delta;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003208 cur_end += delta;
Willy Tarreaua496b602006-12-17 23:15:24 +01003209
Willy Tarreau9cdde232007-05-02 20:58:19 +02003210 txn->req.sol = req->data + txn->req.som; /* should be equal to txn->sol */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003211 cur_end = (char *)http_parse_reqline(&txn->req, req->data,
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003212 HTTP_MSG_RQMETH,
3213 cur_ptr, cur_end + 1,
3214 NULL, NULL);
3215 if (unlikely(!cur_end))
3216 return -1;
Willy Tarreaua496b602006-12-17 23:15:24 +01003217
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003218 /* we have a full request and we know that we have either a CR
3219 * or an LF at <ptr>.
3220 */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003221 txn->meth = find_http_meth(cur_ptr, txn->req.sl.rq.m_l);
3222 hdr_idx_set_start(&txn->hdr_idx, txn->req.sl.rq.l, *cur_end == '\r');
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003223 /* there is no point trying this regex on headers */
3224 return 1;
3225 }
3226 }
3227 *cur_end = term; /* restore the string terminator */
3228 return done;
3229}
Willy Tarreau97de6242006-12-27 17:18:38 +01003230
Willy Tarreau58f10d72006-12-04 02:26:12 +01003231
Willy Tarreau58f10d72006-12-04 02:26:12 +01003232
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003233/*
3234 * Apply all the req filters <exp> to all headers in buffer <req> of session <t>.
3235 * Returns 0 if everything is alright, or -1 in case a replacement lead to an
Willy Tarreaua15645d2007-03-18 16:22:39 +01003236 * unparsable request. Since it can manage the switch to another backend, it
3237 * updates the per-proxy DENY stats.
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003238 */
3239int apply_filters_to_request(struct session *t, struct buffer *req, struct hdr_exp *exp)
3240{
Willy Tarreau3d300592007-03-18 18:34:41 +01003241 struct http_txn *txn = &t->txn;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003242 /* iterate through the filters in the outer loop */
Willy Tarreau3d300592007-03-18 18:34:41 +01003243 while (exp && !(txn->flags & (TX_CLDENY|TX_CLTARPIT))) {
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003244 int ret;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003245
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003246 /*
3247 * The interleaving of transformations and verdicts
3248 * makes it difficult to decide to continue or stop
3249 * the evaluation.
3250 */
3251
Willy Tarreau3d300592007-03-18 18:34:41 +01003252 if ((txn->flags & TX_CLALLOW) &&
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003253 (exp->action == ACT_ALLOW || exp->action == ACT_DENY ||
3254 exp->action == ACT_TARPIT || exp->action == ACT_PASS)) {
3255 exp = exp->next;
3256 continue;
3257 }
3258
3259 /* Apply the filter to the request line. */
3260 ret = apply_filter_to_req_line(t, req, exp);
3261 if (unlikely(ret < 0))
3262 return -1;
3263
3264 if (likely(ret == 0)) {
3265 /* The filter did not match the request, it can be
3266 * iterated through all headers.
3267 */
3268 apply_filter_to_req_headers(t, req, exp);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003269 }
3270 exp = exp->next;
3271 }
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003272 return 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003273}
3274
3275
Willy Tarreaua15645d2007-03-18 16:22:39 +01003276
Willy Tarreau58f10d72006-12-04 02:26:12 +01003277/*
Willy Tarreau396d2c62007-11-04 19:30:00 +01003278 * Manage client-side cookie. It can impact performance by about 2% so it is
3279 * desirable to call it only when needed.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003280 */
3281void manage_client_side_cookies(struct session *t, struct buffer *req)
3282{
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003283 struct http_txn *txn = &t->txn;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003284 char *p1, *p2, *p3, *p4;
3285 char *del_colon, *del_cookie, *colon;
3286 int app_cookies;
3287
3288 appsess *asession_temp = NULL;
3289 appsess local_asession;
3290
3291 char *cur_ptr, *cur_end, *cur_next;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01003292 int cur_idx, old_idx;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003293
Willy Tarreau2a324282006-12-05 00:05:46 +01003294 /* Iterate through the headers.
Willy Tarreau58f10d72006-12-04 02:26:12 +01003295 * we start with the start line.
3296 */
Willy Tarreau83969f42007-01-22 08:55:47 +01003297 old_idx = 0;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003298 cur_next = req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003299
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003300 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003301 struct hdr_idx_elem *cur_hdr;
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003302 int val;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003303
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003304 cur_hdr = &txn->hdr_idx.v[cur_idx];
Willy Tarreau58f10d72006-12-04 02:26:12 +01003305 cur_ptr = cur_next;
3306 cur_end = cur_ptr + cur_hdr->len;
3307 cur_next = cur_end + cur_hdr->cr + 1;
3308
3309 /* We have one full header between cur_ptr and cur_end, and the
3310 * next header starts at cur_next. We're only interested in
3311 * "Cookie:" headers.
3312 */
3313
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003314 val = http_header_match2(cur_ptr, cur_end, "Cookie", 6);
3315 if (!val) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003316 old_idx = cur_idx;
3317 continue;
3318 }
3319
3320 /* Now look for cookies. Conforming to RFC2109, we have to support
3321 * attributes whose name begin with a '$', and associate them with
3322 * the right cookie, if we want to delete this cookie.
3323 * So there are 3 cases for each cookie read :
3324 * 1) it's a special attribute, beginning with a '$' : ignore it.
3325 * 2) it's a server id cookie that we *MAY* want to delete : save
3326 * some pointers on it (last semi-colon, beginning of cookie...)
3327 * 3) it's an application cookie : we *MAY* have to delete a previous
3328 * "special" cookie.
3329 * At the end of loop, if a "special" cookie remains, we may have to
3330 * remove it. If no application cookie persists in the header, we
3331 * *MUST* delete it
3332 */
3333
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003334 colon = p1 = cur_ptr + val; /* first non-space char after 'Cookie:' */
Willy Tarreau58f10d72006-12-04 02:26:12 +01003335
Willy Tarreau58f10d72006-12-04 02:26:12 +01003336 /* del_cookie == NULL => nothing to be deleted */
3337 del_colon = del_cookie = NULL;
3338 app_cookies = 0;
3339
3340 while (p1 < cur_end) {
3341 /* skip spaces and colons, but keep an eye on these ones */
3342 while (p1 < cur_end) {
3343 if (*p1 == ';' || *p1 == ',')
3344 colon = p1;
Willy Tarreau8f8e6452007-06-17 21:51:38 +02003345 else if (!isspace((unsigned char)*p1))
Willy Tarreau58f10d72006-12-04 02:26:12 +01003346 break;
3347 p1++;
3348 }
3349
3350 if (p1 == cur_end)
3351 break;
3352
3353 /* p1 is at the beginning of the cookie name */
3354 p2 = p1;
3355 while (p2 < cur_end && *p2 != '=')
3356 p2++;
3357
3358 if (p2 == cur_end)
3359 break;
3360
3361 p3 = p2 + 1; /* skips the '=' sign */
3362 if (p3 == cur_end)
3363 break;
3364
3365 p4 = p3;
Willy Tarreau8f8e6452007-06-17 21:51:38 +02003366 while (p4 < cur_end && !isspace((unsigned char)*p4) && *p4 != ';' && *p4 != ',')
Willy Tarreau58f10d72006-12-04 02:26:12 +01003367 p4++;
3368
3369 /* here, we have the cookie name between p1 and p2,
3370 * and its value between p3 and p4.
3371 * we can process it :
3372 *
3373 * Cookie: NAME=VALUE;
3374 * | || || |
3375 * | || || +--> p4
3376 * | || |+-------> p3
3377 * | || +--------> p2
3378 * | |+------------> p1
3379 * | +-------------> colon
3380 * +--------------------> cur_ptr
3381 */
3382
3383 if (*p1 == '$') {
3384 /* skip this one */
3385 }
3386 else {
3387 /* first, let's see if we want to capture it */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003388 if (t->fe->capture_name != NULL &&
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003389 txn->cli_cookie == NULL &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003390 (p4 - p1 >= t->fe->capture_namelen) &&
3391 memcmp(p1, t->fe->capture_name, t->fe->capture_namelen) == 0) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003392 int log_len = p4 - p1;
3393
Willy Tarreau086b3b42007-05-13 21:45:51 +02003394 if ((txn->cli_cookie = pool_alloc2(pool2_capture)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003395 Alert("HTTP logging : out of memory.\n");
3396 } else {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003397 if (log_len > t->fe->capture_len)
3398 log_len = t->fe->capture_len;
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003399 memcpy(txn->cli_cookie, p1, log_len);
3400 txn->cli_cookie[log_len] = 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003401 }
3402 }
3403
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003404 if ((p2 - p1 == t->be->cookie_len) && (t->be->cookie_name != NULL) &&
3405 (memcmp(p1, t->be->cookie_name, p2 - p1) == 0)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003406 /* Cool... it's the right one */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003407 struct server *srv = t->be->srv;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003408 char *delim;
3409
3410 /* if we're in cookie prefix mode, we'll search the delimitor so that we
3411 * have the server ID betweek p3 and delim, and the original cookie between
3412 * delim+1 and p4. Otherwise, delim==p4 :
3413 *
3414 * Cookie: NAME=SRV~VALUE;
3415 * | || || | |
3416 * | || || | +--> p4
3417 * | || || +--------> delim
3418 * | || |+-----------> p3
3419 * | || +------------> p2
3420 * | |+----------------> p1
3421 * | +-----------------> colon
3422 * +------------------------> cur_ptr
3423 */
3424
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003425 if (t->be->options & PR_O_COOK_PFX) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003426 for (delim = p3; delim < p4; delim++)
3427 if (*delim == COOKIE_DELIM)
3428 break;
3429 }
3430 else
3431 delim = p4;
3432
3433
3434 /* Here, we'll look for the first running server which supports the cookie.
3435 * This allows to share a same cookie between several servers, for example
3436 * to dedicate backup servers to specific servers only.
3437 * However, to prevent clients from sticking to cookie-less backup server
3438 * when they have incidentely learned an empty cookie, we simply ignore
3439 * empty cookies and mark them as invalid.
3440 */
3441 if (delim == p3)
3442 srv = NULL;
3443
3444 while (srv) {
Willy Tarreau92f2ab12007-02-02 22:14:47 +01003445 if (srv->cookie && (srv->cklen == delim - p3) &&
3446 !memcmp(p3, srv->cookie, delim - p3)) {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003447 if (srv->state & SRV_RUNNING || t->be->options & PR_O_PERSIST) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003448 /* we found the server and it's usable */
Willy Tarreau3d300592007-03-18 18:34:41 +01003449 txn->flags &= ~TX_CK_MASK;
3450 txn->flags |= TX_CK_VALID;
3451 t->flags |= SN_DIRECT | SN_ASSIGNED;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003452 t->srv = srv;
3453 break;
3454 } else {
3455 /* we found a server, but it's down */
Willy Tarreau3d300592007-03-18 18:34:41 +01003456 txn->flags &= ~TX_CK_MASK;
3457 txn->flags |= TX_CK_DOWN;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003458 }
3459 }
3460 srv = srv->next;
3461 }
3462
Willy Tarreau3d300592007-03-18 18:34:41 +01003463 if (!srv && !(txn->flags & TX_CK_DOWN)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003464 /* no server matched this cookie */
Willy Tarreau3d300592007-03-18 18:34:41 +01003465 txn->flags &= ~TX_CK_MASK;
3466 txn->flags |= TX_CK_INVALID;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003467 }
3468
3469 /* depending on the cookie mode, we may have to either :
3470 * - delete the complete cookie if we're in insert+indirect mode, so that
3471 * the server never sees it ;
3472 * - remove the server id from the cookie value, and tag the cookie as an
3473 * application cookie so that it does not get accidentely removed later,
3474 * if we're in cookie prefix mode
3475 */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003476 if ((t->be->options & PR_O_COOK_PFX) && (delim != p4)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003477 int delta; /* negative */
3478
3479 delta = buffer_replace2(req, p3, delim + 1, NULL, 0);
3480 p4 += delta;
3481 cur_end += delta;
3482 cur_next += delta;
3483 cur_hdr->len += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003484 txn->req.eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003485
3486 del_cookie = del_colon = NULL;
3487 app_cookies++; /* protect the header from deletion */
3488 }
3489 else if (del_cookie == NULL &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003490 (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 +01003491 del_cookie = p1;
3492 del_colon = colon;
3493 }
3494 } else {
3495 /* now we know that we must keep this cookie since it's
3496 * not ours. But if we wanted to delete our cookie
3497 * earlier, we cannot remove the complete header, but we
3498 * can remove the previous block itself.
3499 */
3500 app_cookies++;
3501
3502 if (del_cookie != NULL) {
3503 int delta; /* negative */
3504
3505 delta = buffer_replace2(req, del_cookie, p1, NULL, 0);
3506 p4 += delta;
3507 cur_end += delta;
3508 cur_next += delta;
3509 cur_hdr->len += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003510 txn->req.eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003511 del_cookie = del_colon = NULL;
3512 }
3513 }
3514
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003515 if ((t->be->appsession_name != NULL) &&
3516 (memcmp(p1, t->be->appsession_name, p2 - p1) == 0)) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003517 /* first, let's see if the cookie is our appcookie*/
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003518
Willy Tarreau58f10d72006-12-04 02:26:12 +01003519 /* Cool... it's the right one */
3520
3521 asession_temp = &local_asession;
3522
Willy Tarreau63963c62007-05-13 21:29:55 +02003523 if ((asession_temp->sessid = pool_alloc2(apools.sessid)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003524 Alert("Not enough memory process_cli():asession->sessid:malloc().\n");
3525 send_log(t->be, LOG_ALERT, "Not enough memory process_cli():asession->sessid:malloc().\n");
3526 return;
3527 }
3528
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003529 memcpy(asession_temp->sessid, p3, t->be->appsession_len);
3530 asession_temp->sessid[t->be->appsession_len] = 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003531 asession_temp->serverid = NULL;
Willy Tarreau51041c72007-09-09 21:56:53 +02003532
Willy Tarreau58f10d72006-12-04 02:26:12 +01003533 /* only do insert, if lookup fails */
Willy Tarreau51041c72007-09-09 21:56:53 +02003534 asession_temp = appsession_hash_lookup(&(t->be->htbl_proxy), asession_temp->sessid);
3535 if (asession_temp == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02003536 if ((asession_temp = pool_alloc2(pool2_appsess)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003537 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02003538 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003539 Alert("Not enough memory process_cli():asession:calloc().\n");
3540 send_log(t->be, LOG_ALERT, "Not enough memory process_cli():asession:calloc().\n");
3541 return;
3542 }
3543
3544 asession_temp->sessid = local_asession.sessid;
3545 asession_temp->serverid = local_asession.serverid;
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003546 asession_temp->request_count = 0;
Willy Tarreau51041c72007-09-09 21:56:53 +02003547 appsession_hash_insert(&(t->be->htbl_proxy), asession_temp);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003548 } else {
3549 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02003550 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003551 }
Willy Tarreau58f10d72006-12-04 02:26:12 +01003552 if (asession_temp->serverid == NULL) {
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003553 /* TODO redispatch request */
Willy Tarreau58f10d72006-12-04 02:26:12 +01003554 Alert("Found Application Session without matching server.\n");
3555 } else {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003556 struct server *srv = t->be->srv;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003557 while (srv) {
3558 if (strcmp(srv->id, asession_temp->serverid) == 0) {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003559 if (srv->state & SRV_RUNNING || t->be->options & PR_O_PERSIST) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01003560 /* we found the server and it's usable */
Willy Tarreau3d300592007-03-18 18:34:41 +01003561 txn->flags &= ~TX_CK_MASK;
3562 txn->flags |= TX_CK_VALID;
3563 t->flags |= SN_DIRECT | SN_ASSIGNED;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003564 t->srv = srv;
3565 break;
3566 } else {
Willy Tarreau3d300592007-03-18 18:34:41 +01003567 txn->flags &= ~TX_CK_MASK;
3568 txn->flags |= TX_CK_DOWN;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003569 }
3570 }
3571 srv = srv->next;
3572 }/* end while(srv) */
3573 }/* end else if server == NULL */
3574
Willy Tarreau0c303ee2008-07-07 00:09:58 +02003575 asession_temp->expire = tick_add_ifset(now_ms, t->be->timeout.appsession);
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02003576 asession_temp->request_count++;
3577#if defined(DEBUG_HASH)
3578 Alert("manage_client_side_cookies\n");
3579 appsession_hash_dump(&(t->be->htbl_proxy));
3580#endif
Willy Tarreau58f10d72006-12-04 02:26:12 +01003581 }/* end if ((t->proxy->appsession_name != NULL) ... */
3582 }
3583
3584 /* we'll have to look for another cookie ... */
3585 p1 = p4;
3586 } /* while (p1 < cur_end) */
3587
3588 /* There's no more cookie on this line.
3589 * We may have marked the last one(s) for deletion.
3590 * We must do this now in two ways :
3591 * - if there is no app cookie, we simply delete the header ;
3592 * - if there are app cookies, we must delete the end of the
3593 * string properly, including the colon/semi-colon before
3594 * the cookie name.
3595 */
3596 if (del_cookie != NULL) {
3597 int delta;
3598 if (app_cookies) {
3599 delta = buffer_replace2(req, del_colon, cur_end, NULL, 0);
3600 cur_end = del_colon;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003601 cur_hdr->len += delta;
3602 } else {
3603 delta = buffer_replace2(req, cur_ptr, cur_next, NULL, 0);
Willy Tarreau58f10d72006-12-04 02:26:12 +01003604
3605 /* FIXME: this should be a separate function */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003606 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3607 txn->hdr_idx.used--;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003608 cur_hdr->len = 0;
3609 }
Willy Tarreau45e73e32006-12-17 00:05:15 +01003610 cur_next += delta;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01003611 txn->req.eoh += delta;
Willy Tarreau58f10d72006-12-04 02:26:12 +01003612 }
3613
3614 /* keep the link from this header to next one */
3615 old_idx = cur_idx;
3616 } /* end of cookie processing on this header */
3617}
3618
3619
Willy Tarreaua15645d2007-03-18 16:22:39 +01003620/* Iterate the same filter through all response headers contained in <rtr>.
3621 * Returns 1 if this filter can be stopped upon return, otherwise 0.
3622 */
3623int apply_filter_to_resp_headers(struct session *t, struct buffer *rtr, struct hdr_exp *exp)
3624{
3625 char term;
3626 char *cur_ptr, *cur_end, *cur_next;
3627 int cur_idx, old_idx, last_hdr;
3628 struct http_txn *txn = &t->txn;
3629 struct hdr_idx_elem *cur_hdr;
3630 int len, delta;
3631
3632 last_hdr = 0;
3633
3634 cur_next = rtr->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
3635 old_idx = 0;
3636
3637 while (!last_hdr) {
Willy Tarreau3d300592007-03-18 18:34:41 +01003638 if (unlikely(txn->flags & TX_SVDENY))
Willy Tarreaua15645d2007-03-18 16:22:39 +01003639 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003640 else if (unlikely(txn->flags & TX_SVALLOW) &&
Willy Tarreaua15645d2007-03-18 16:22:39 +01003641 (exp->action == ACT_ALLOW ||
3642 exp->action == ACT_DENY))
3643 return 0;
3644
3645 cur_idx = txn->hdr_idx.v[old_idx].next;
3646 if (!cur_idx)
3647 break;
3648
3649 cur_hdr = &txn->hdr_idx.v[cur_idx];
3650 cur_ptr = cur_next;
3651 cur_end = cur_ptr + cur_hdr->len;
3652 cur_next = cur_end + cur_hdr->cr + 1;
3653
3654 /* Now we have one header between cur_ptr and cur_end,
3655 * and the next header starts at cur_next.
3656 */
3657
3658 /* The annoying part is that pattern matching needs
3659 * that we modify the contents to null-terminate all
3660 * strings before testing them.
3661 */
3662
3663 term = *cur_end;
3664 *cur_end = '\0';
3665
3666 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3667 switch (exp->action) {
3668 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003669 txn->flags |= TX_SVALLOW;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003670 last_hdr = 1;
3671 break;
3672
3673 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003674 txn->flags |= TX_SVDENY;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003675 last_hdr = 1;
3676 break;
3677
3678 case ACT_REPLACE:
3679 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3680 delta = buffer_replace2(rtr, cur_ptr, cur_end, trash, len);
3681 /* FIXME: if the user adds a newline in the replacement, the
3682 * index will not be recalculated for now, and the new line
3683 * will not be counted as a new header.
3684 */
3685
3686 cur_end += delta;
3687 cur_next += delta;
3688 cur_hdr->len += delta;
3689 txn->rsp.eoh += delta;
3690 break;
3691
3692 case ACT_REMOVE:
3693 delta = buffer_replace2(rtr, cur_ptr, cur_next, NULL, 0);
3694 cur_next += delta;
3695
3696 /* FIXME: this should be a separate function */
3697 txn->rsp.eoh += delta;
3698 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3699 txn->hdr_idx.used--;
3700 cur_hdr->len = 0;
3701 cur_end = NULL; /* null-term has been rewritten */
3702 break;
3703
3704 }
3705 }
3706 if (cur_end)
3707 *cur_end = term; /* restore the string terminator */
3708
3709 /* keep the link from this header to next one in case of later
3710 * removal of next header.
3711 */
3712 old_idx = cur_idx;
3713 }
3714 return 0;
3715}
3716
3717
3718/* Apply the filter to the status line in the response buffer <rtr>.
3719 * Returns 0 if nothing has been done, 1 if the filter has been applied,
3720 * or -1 if a replacement resulted in an invalid status line.
3721 */
3722int apply_filter_to_sts_line(struct session *t, struct buffer *rtr, struct hdr_exp *exp)
3723{
3724 char term;
3725 char *cur_ptr, *cur_end;
3726 int done;
3727 struct http_txn *txn = &t->txn;
3728 int len, delta;
3729
3730
Willy Tarreau3d300592007-03-18 18:34:41 +01003731 if (unlikely(txn->flags & TX_SVDENY))
Willy Tarreaua15645d2007-03-18 16:22:39 +01003732 return 1;
Willy Tarreau3d300592007-03-18 18:34:41 +01003733 else if (unlikely(txn->flags & TX_SVALLOW) &&
Willy Tarreaua15645d2007-03-18 16:22:39 +01003734 (exp->action == ACT_ALLOW ||
3735 exp->action == ACT_DENY))
3736 return 0;
3737 else if (exp->action == ACT_REMOVE)
3738 return 0;
3739
3740 done = 0;
3741
Willy Tarreau9cdde232007-05-02 20:58:19 +02003742 cur_ptr = rtr->data + txn->rsp.som; /* should be equal to txn->sol */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003743 cur_end = cur_ptr + txn->rsp.sl.rq.l;
3744
3745 /* Now we have the status line between cur_ptr and cur_end */
3746
3747 /* The annoying part is that pattern matching needs
3748 * that we modify the contents to null-terminate all
3749 * strings before testing them.
3750 */
3751
3752 term = *cur_end;
3753 *cur_end = '\0';
3754
3755 if (regexec(exp->preg, cur_ptr, MAX_MATCH, pmatch, 0) == 0) {
3756 switch (exp->action) {
3757 case ACT_ALLOW:
Willy Tarreau3d300592007-03-18 18:34:41 +01003758 txn->flags |= TX_SVALLOW;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003759 done = 1;
3760 break;
3761
3762 case ACT_DENY:
Willy Tarreau3d300592007-03-18 18:34:41 +01003763 txn->flags |= TX_SVDENY;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003764 done = 1;
3765 break;
3766
3767 case ACT_REPLACE:
3768 *cur_end = term; /* restore the string terminator */
3769 len = exp_replace(trash, cur_ptr, exp->replace, pmatch);
3770 delta = buffer_replace2(rtr, cur_ptr, cur_end, trash, len);
3771 /* FIXME: if the user adds a newline in the replacement, the
3772 * index will not be recalculated for now, and the new line
3773 * will not be counted as a new header.
3774 */
3775
3776 txn->rsp.eoh += delta;
3777 cur_end += delta;
3778
Willy Tarreau9cdde232007-05-02 20:58:19 +02003779 txn->rsp.sol = rtr->data + txn->rsp.som; /* should be equal to txn->sol */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003780 cur_end = (char *)http_parse_stsline(&txn->rsp, rtr->data,
Willy Tarreau02785762007-04-03 14:45:44 +02003781 HTTP_MSG_RPVER,
Willy Tarreaua15645d2007-03-18 16:22:39 +01003782 cur_ptr, cur_end + 1,
3783 NULL, NULL);
3784 if (unlikely(!cur_end))
3785 return -1;
3786
3787 /* we have a full respnse and we know that we have either a CR
3788 * or an LF at <ptr>.
3789 */
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003790 txn->status = strl2ui(rtr->data + txn->rsp.sl.st.c, txn->rsp.sl.st.c_l);
Willy Tarreaua15645d2007-03-18 16:22:39 +01003791 hdr_idx_set_start(&txn->hdr_idx, txn->rsp.sl.rq.l, *cur_end == '\r');
3792 /* there is no point trying this regex on headers */
3793 return 1;
3794 }
3795 }
3796 *cur_end = term; /* restore the string terminator */
3797 return done;
3798}
3799
3800
3801
3802/*
3803 * Apply all the resp filters <exp> to all headers in buffer <rtr> of session <t>.
3804 * Returns 0 if everything is alright, or -1 in case a replacement lead to an
3805 * unparsable response.
3806 */
3807int apply_filters_to_response(struct session *t, struct buffer *rtr, struct hdr_exp *exp)
3808{
Willy Tarreau3d300592007-03-18 18:34:41 +01003809 struct http_txn *txn = &t->txn;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003810 /* iterate through the filters in the outer loop */
Willy Tarreau3d300592007-03-18 18:34:41 +01003811 while (exp && !(txn->flags & TX_SVDENY)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003812 int ret;
3813
3814 /*
3815 * The interleaving of transformations and verdicts
3816 * makes it difficult to decide to continue or stop
3817 * the evaluation.
3818 */
3819
Willy Tarreau3d300592007-03-18 18:34:41 +01003820 if ((txn->flags & TX_SVALLOW) &&
Willy Tarreaua15645d2007-03-18 16:22:39 +01003821 (exp->action == ACT_ALLOW || exp->action == ACT_DENY ||
3822 exp->action == ACT_PASS)) {
3823 exp = exp->next;
3824 continue;
3825 }
3826
3827 /* Apply the filter to the status line. */
3828 ret = apply_filter_to_sts_line(t, rtr, exp);
3829 if (unlikely(ret < 0))
3830 return -1;
3831
3832 if (likely(ret == 0)) {
3833 /* The filter did not match the response, it can be
3834 * iterated through all headers.
3835 */
3836 apply_filter_to_resp_headers(t, rtr, exp);
3837 }
3838 exp = exp->next;
3839 }
3840 return 0;
3841}
3842
3843
3844
3845/*
Willy Tarreau396d2c62007-11-04 19:30:00 +01003846 * Manage server-side cookies. It can impact performance by about 2% so it is
3847 * desirable to call it only when needed.
Willy Tarreaua15645d2007-03-18 16:22:39 +01003848 */
3849void manage_server_side_cookies(struct session *t, struct buffer *rtr)
3850{
3851 struct http_txn *txn = &t->txn;
3852 char *p1, *p2, *p3, *p4;
3853
3854 appsess *asession_temp = NULL;
3855 appsess local_asession;
3856
3857 char *cur_ptr, *cur_end, *cur_next;
3858 int cur_idx, old_idx, delta;
3859
Willy Tarreaua15645d2007-03-18 16:22:39 +01003860 /* Iterate through the headers.
3861 * we start with the start line.
3862 */
3863 old_idx = 0;
3864 cur_next = rtr->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
3865
3866 while ((cur_idx = txn->hdr_idx.v[old_idx].next)) {
3867 struct hdr_idx_elem *cur_hdr;
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003868 int val;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003869
3870 cur_hdr = &txn->hdr_idx.v[cur_idx];
3871 cur_ptr = cur_next;
3872 cur_end = cur_ptr + cur_hdr->len;
3873 cur_next = cur_end + cur_hdr->cr + 1;
3874
3875 /* We have one full header between cur_ptr and cur_end, and the
3876 * next header starts at cur_next. We're only interested in
3877 * "Cookie:" headers.
3878 */
3879
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003880 val = http_header_match2(cur_ptr, cur_end, "Set-Cookie", 10);
3881 if (!val) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003882 old_idx = cur_idx;
3883 continue;
3884 }
3885
3886 /* OK, right now we know we have a set-cookie at cur_ptr */
Willy Tarreau3d300592007-03-18 18:34:41 +01003887 txn->flags |= TX_SCK_ANY;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003888
3889
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003890 /* maybe we only wanted to see if there was a set-cookie. Note that
3891 * the cookie capture is declared in the fronend.
3892 */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003893 if (t->be->cookie_name == NULL &&
3894 t->be->appsession_name == NULL &&
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003895 t->fe->capture_name == NULL)
Willy Tarreaua15645d2007-03-18 16:22:39 +01003896 return;
3897
Willy Tarreauaa9dce32007-03-18 23:50:16 +01003898 p1 = cur_ptr + val; /* first non-space char after 'Set-Cookie:' */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003899
3900 while (p1 < cur_end) { /* in fact, we'll break after the first cookie */
Willy Tarreaua15645d2007-03-18 16:22:39 +01003901 if (p1 == cur_end || *p1 == ';') /* end of cookie */
3902 break;
3903
3904 /* p1 is at the beginning of the cookie name */
3905 p2 = p1;
3906
3907 while (p2 < cur_end && *p2 != '=' && *p2 != ';')
3908 p2++;
3909
3910 if (p2 == cur_end || *p2 == ';') /* next cookie */
3911 break;
3912
3913 p3 = p2 + 1; /* skip the '=' sign */
3914 if (p3 == cur_end)
3915 break;
3916
3917 p4 = p3;
Willy Tarreau8f8e6452007-06-17 21:51:38 +02003918 while (p4 < cur_end && !isspace((unsigned char)*p4) && *p4 != ';')
Willy Tarreaua15645d2007-03-18 16:22:39 +01003919 p4++;
3920
3921 /* here, we have the cookie name between p1 and p2,
3922 * and its value between p3 and p4.
3923 * we can process it.
3924 */
3925
3926 /* first, let's see if we want to capture it */
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003927 if (t->fe->capture_name != NULL &&
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003928 txn->srv_cookie == NULL &&
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003929 (p4 - p1 >= t->fe->capture_namelen) &&
3930 memcmp(p1, t->fe->capture_name, t->fe->capture_namelen) == 0) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003931 int log_len = p4 - p1;
3932
Willy Tarreau086b3b42007-05-13 21:45:51 +02003933 if ((txn->srv_cookie = pool_alloc2(pool2_capture)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003934 Alert("HTTP logging : out of memory.\n");
3935 }
3936
Willy Tarreaufd39dda2008-10-17 12:01:58 +02003937 if (log_len > t->fe->capture_len)
3938 log_len = t->fe->capture_len;
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01003939 memcpy(txn->srv_cookie, p1, log_len);
3940 txn->srv_cookie[log_len] = 0;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003941 }
3942
3943 /* now check if we need to process it for persistence */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003944 if ((p2 - p1 == t->be->cookie_len) && (t->be->cookie_name != NULL) &&
3945 (memcmp(p1, t->be->cookie_name, p2 - p1) == 0)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003946 /* Cool... it's the right one */
Willy Tarreau3d300592007-03-18 18:34:41 +01003947 txn->flags |= TX_SCK_SEEN;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003948
3949 /* If the cookie is in insert mode on a known server, we'll delete
3950 * this occurrence because we'll insert another one later.
3951 * We'll delete it too if the "indirect" option is set and we're in
3952 * a direct access. */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003953 if (((t->srv) && (t->be->options & PR_O_COOK_INS)) ||
3954 ((t->flags & SN_DIRECT) && (t->be->options & PR_O_COOK_IND))) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003955 /* this header must be deleted */
3956 delta = buffer_replace2(rtr, cur_ptr, cur_next, NULL, 0);
3957 txn->hdr_idx.v[old_idx].next = cur_hdr->next;
3958 txn->hdr_idx.used--;
3959 cur_hdr->len = 0;
3960 cur_next += delta;
3961 txn->rsp.eoh += delta;
3962
Willy Tarreau3d300592007-03-18 18:34:41 +01003963 txn->flags |= TX_SCK_DELETED;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003964 }
3965 else if ((t->srv) && (t->srv->cookie) &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003966 (t->be->options & PR_O_COOK_RW)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003967 /* replace bytes p3->p4 with the cookie name associated
3968 * with this server since we know it.
3969 */
3970 delta = buffer_replace2(rtr, p3, p4, t->srv->cookie, t->srv->cklen);
3971 cur_hdr->len += delta;
3972 cur_next += delta;
3973 txn->rsp.eoh += delta;
3974
Willy Tarreau3d300592007-03-18 18:34:41 +01003975 txn->flags |= TX_SCK_INSERTED | TX_SCK_DELETED;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003976 }
3977 else if ((t->srv) && (t->srv->cookie) &&
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003978 (t->be->options & PR_O_COOK_PFX)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003979 /* insert the cookie name associated with this server
3980 * before existing cookie, and insert a delimitor between them..
3981 */
3982 delta = buffer_replace2(rtr, p3, p3, t->srv->cookie, t->srv->cklen + 1);
3983 cur_hdr->len += delta;
3984 cur_next += delta;
3985 txn->rsp.eoh += delta;
3986
3987 p3[t->srv->cklen] = COOKIE_DELIM;
Willy Tarreau3d300592007-03-18 18:34:41 +01003988 txn->flags |= TX_SCK_INSERTED | TX_SCK_DELETED;
Willy Tarreaua15645d2007-03-18 16:22:39 +01003989 }
3990 }
3991 /* next, let's see if the cookie is our appcookie */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02003992 else if ((t->be->appsession_name != NULL) &&
3993 (memcmp(p1, t->be->appsession_name, p2 - p1) == 0)) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01003994
3995 /* Cool... it's the right one */
3996
3997 size_t server_id_len = strlen(t->srv->id) + 1;
3998 asession_temp = &local_asession;
3999
Willy Tarreau63963c62007-05-13 21:29:55 +02004000 if ((asession_temp->sessid = pool_alloc2(apools.sessid)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004001 Alert("Not enough Memory process_srv():asession->sessid:malloc().\n");
4002 send_log(t->be, LOG_ALERT, "Not enough Memory process_srv():asession->sessid:malloc().\n");
4003 return;
4004 }
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004005 memcpy(asession_temp->sessid, p3, t->be->appsession_len);
4006 asession_temp->sessid[t->be->appsession_len] = 0;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004007 asession_temp->serverid = NULL;
4008
4009 /* only do insert, if lookup fails */
Ryan Warnick6d0b1fa2008-02-17 11:24:35 +01004010 asession_temp = appsession_hash_lookup(&(t->be->htbl_proxy), asession_temp->sessid);
4011 if (asession_temp == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02004012 if ((asession_temp = pool_alloc2(pool2_appsess)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004013 Alert("Not enough Memory process_srv():asession:calloc().\n");
4014 send_log(t->be, LOG_ALERT, "Not enough Memory process_srv():asession:calloc().\n");
4015 return;
4016 }
4017 asession_temp->sessid = local_asession.sessid;
4018 asession_temp->serverid = local_asession.serverid;
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004019 asession_temp->request_count = 0;
Willy Tarreau51041c72007-09-09 21:56:53 +02004020 appsession_hash_insert(&(t->be->htbl_proxy), asession_temp);
4021 } else {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004022 /* free wasted memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02004023 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau51041c72007-09-09 21:56:53 +02004024 }
4025
Willy Tarreaua15645d2007-03-18 16:22:39 +01004026 if (asession_temp->serverid == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02004027 if ((asession_temp->serverid = pool_alloc2(apools.serverid)) == NULL) {
Willy Tarreaua15645d2007-03-18 16:22:39 +01004028 Alert("Not enough Memory process_srv():asession->sessid:malloc().\n");
4029 send_log(t->be, LOG_ALERT, "Not enough Memory process_srv():asession->sessid:malloc().\n");
4030 return;
4031 }
4032 asession_temp->serverid[0] = '\0';
4033 }
4034
4035 if (asession_temp->serverid[0] == '\0')
4036 memcpy(asession_temp->serverid, t->srv->id, server_id_len);
4037
Willy Tarreau0c303ee2008-07-07 00:09:58 +02004038 asession_temp->expire = tick_add_ifset(now_ms, t->be->timeout.appsession);
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004039 asession_temp->request_count++;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004040#if defined(DEBUG_HASH)
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004041 Alert("manage_server_side_cookies\n");
Willy Tarreau51041c72007-09-09 21:56:53 +02004042 appsession_hash_dump(&(t->be->htbl_proxy));
Willy Tarreaua15645d2007-03-18 16:22:39 +01004043#endif
4044 }/* end if ((t->proxy->appsession_name != NULL) ... */
4045 break; /* we don't want to loop again since there cannot be another cookie on the same line */
4046 } /* we're now at the end of the cookie value */
4047
4048 /* keep the link from this header to next one */
4049 old_idx = cur_idx;
4050 } /* end of cookie processing on this header */
4051}
4052
4053
4054
4055/*
4056 * Check if response is cacheable or not. Updates t->flags.
4057 */
4058void check_response_for_cacheability(struct session *t, struct buffer *rtr)
4059{
4060 struct http_txn *txn = &t->txn;
4061 char *p1, *p2;
4062
4063 char *cur_ptr, *cur_end, *cur_next;
4064 int cur_idx;
4065
Willy Tarreau5df51872007-11-25 16:20:08 +01004066 if (!(txn->flags & TX_CACHEABLE))
Willy Tarreaua15645d2007-03-18 16:22:39 +01004067 return;
4068
4069 /* Iterate through the headers.
4070 * we start with the start line.
4071 */
4072 cur_idx = 0;
4073 cur_next = rtr->data + txn->rsp.som + hdr_idx_first_pos(&txn->hdr_idx);
4074
4075 while ((cur_idx = txn->hdr_idx.v[cur_idx].next)) {
4076 struct hdr_idx_elem *cur_hdr;
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004077 int val;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004078
4079 cur_hdr = &txn->hdr_idx.v[cur_idx];
4080 cur_ptr = cur_next;
4081 cur_end = cur_ptr + cur_hdr->len;
4082 cur_next = cur_end + cur_hdr->cr + 1;
4083
4084 /* We have one full header between cur_ptr and cur_end, and the
4085 * next header starts at cur_next. We're only interested in
4086 * "Cookie:" headers.
4087 */
4088
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004089 val = http_header_match2(cur_ptr, cur_end, "Pragma", 6);
4090 if (val) {
4091 if ((cur_end - (cur_ptr + val) >= 8) &&
4092 strncasecmp(cur_ptr + val, "no-cache", 8) == 0) {
4093 txn->flags &= ~TX_CACHEABLE & ~TX_CACHE_COOK;
4094 return;
4095 }
Willy Tarreaua15645d2007-03-18 16:22:39 +01004096 }
4097
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004098 val = http_header_match2(cur_ptr, cur_end, "Cache-control", 13);
4099 if (!val)
Willy Tarreaua15645d2007-03-18 16:22:39 +01004100 continue;
4101
4102 /* OK, right now we know we have a cache-control header at cur_ptr */
4103
Willy Tarreauaa9dce32007-03-18 23:50:16 +01004104 p1 = cur_ptr + val; /* first non-space char after 'cache-control:' */
Willy Tarreaua15645d2007-03-18 16:22:39 +01004105
4106 if (p1 >= cur_end) /* no more info */
4107 continue;
4108
4109 /* p1 is at the beginning of the value */
4110 p2 = p1;
4111
Willy Tarreau8f8e6452007-06-17 21:51:38 +02004112 while (p2 < cur_end && *p2 != '=' && *p2 != ',' && !isspace((unsigned char)*p2))
Willy Tarreaua15645d2007-03-18 16:22:39 +01004113 p2++;
4114
4115 /* we have a complete value between p1 and p2 */
4116 if (p2 < cur_end && *p2 == '=') {
4117 /* we have something of the form no-cache="set-cookie" */
4118 if ((cur_end - p1 >= 21) &&
4119 strncasecmp(p1, "no-cache=\"set-cookie", 20) == 0
4120 && (p1[20] == '"' || p1[20] == ','))
Willy Tarreau3d300592007-03-18 18:34:41 +01004121 txn->flags &= ~TX_CACHE_COOK;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004122 continue;
4123 }
4124
4125 /* OK, so we know that either p2 points to the end of string or to a comma */
4126 if (((p2 - p1 == 7) && strncasecmp(p1, "private", 7) == 0) ||
4127 ((p2 - p1 == 8) && strncasecmp(p1, "no-store", 8) == 0) ||
4128 ((p2 - p1 == 9) && strncasecmp(p1, "max-age=0", 9) == 0) ||
4129 ((p2 - p1 == 10) && strncasecmp(p1, "s-maxage=0", 10) == 0)) {
Willy Tarreau3d300592007-03-18 18:34:41 +01004130 txn->flags &= ~TX_CACHEABLE & ~TX_CACHE_COOK;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004131 return;
4132 }
4133
4134 if ((p2 - p1 == 6) && strncasecmp(p1, "public", 6) == 0) {
Willy Tarreau3d300592007-03-18 18:34:41 +01004135 txn->flags |= TX_CACHEABLE | TX_CACHE_COOK;
Willy Tarreaua15645d2007-03-18 16:22:39 +01004136 continue;
4137 }
4138 }
4139}
4140
4141
Willy Tarreau58f10d72006-12-04 02:26:12 +01004142/*
4143 * Try to retrieve a known appsession in the URI, then the associated server.
4144 * If the server is found, it's assigned to the session.
4145 */
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004146void get_srv_from_appsession(struct session *t, const char *begin, int len)
Willy Tarreau58f10d72006-12-04 02:26:12 +01004147{
Willy Tarreau3d300592007-03-18 18:34:41 +01004148 struct http_txn *txn = &t->txn;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004149 appsess *asession_temp = NULL;
4150 appsess local_asession;
4151 char *request_line;
4152
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004153 if (t->be->appsession_name == NULL ||
Willy Tarreaub326fcc2007-03-03 13:54:32 +01004154 (t->txn.meth != HTTP_METH_GET && t->txn.meth != HTTP_METH_POST) ||
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004155 (request_line = memchr(begin, ';', len)) == NULL ||
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004156 ((1 + t->be->appsession_name_len + 1 + t->be->appsession_len) > (begin + len - request_line)))
Willy Tarreau58f10d72006-12-04 02:26:12 +01004157 return;
4158
4159 /* skip ';' */
4160 request_line++;
4161
4162 /* look if we have a jsessionid */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004163 if (strncasecmp(request_line, t->be->appsession_name, t->be->appsession_name_len) != 0)
Willy Tarreau58f10d72006-12-04 02:26:12 +01004164 return;
4165
4166 /* skip jsessionid= */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004167 request_line += t->be->appsession_name_len + 1;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004168
4169 /* First try if we already have an appsession */
4170 asession_temp = &local_asession;
4171
Willy Tarreau63963c62007-05-13 21:29:55 +02004172 if ((asession_temp->sessid = pool_alloc2(apools.sessid)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01004173 Alert("Not enough memory process_cli():asession_temp->sessid:calloc().\n");
4174 send_log(t->be, LOG_ALERT, "Not enough Memory process_cli():asession_temp->sessid:calloc().\n");
4175 return;
4176 }
4177
4178 /* Copy the sessionid */
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004179 memcpy(asession_temp->sessid, request_line, t->be->appsession_len);
4180 asession_temp->sessid[t->be->appsession_len] = 0;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004181 asession_temp->serverid = NULL;
4182
4183 /* only do insert, if lookup fails */
Ryan Warnick6d0b1fa2008-02-17 11:24:35 +01004184 asession_temp = appsession_hash_lookup(&(t->be->htbl_proxy), asession_temp->sessid);
4185 if (asession_temp == NULL) {
Willy Tarreau63963c62007-05-13 21:29:55 +02004186 if ((asession_temp = pool_alloc2(pool2_appsess)) == NULL) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01004187 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02004188 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004189 Alert("Not enough memory process_cli():asession:calloc().\n");
4190 send_log(t->be, LOG_ALERT, "Not enough memory process_cli():asession:calloc().\n");
4191 return;
4192 }
4193 asession_temp->sessid = local_asession.sessid;
4194 asession_temp->serverid = local_asession.serverid;
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004195 asession_temp->request_count=0;
Willy Tarreau51041c72007-09-09 21:56:53 +02004196 appsession_hash_insert(&(t->be->htbl_proxy), asession_temp);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004197 }
4198 else {
4199 /* free previously allocated memory */
Willy Tarreau63963c62007-05-13 21:29:55 +02004200 pool_free2(apools.sessid, local_asession.sessid);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004201 }
Willy Tarreau51041c72007-09-09 21:56:53 +02004202
Willy Tarreau0c303ee2008-07-07 00:09:58 +02004203 asession_temp->expire = tick_add_ifset(now_ms, t->be->timeout.appsession);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004204 asession_temp->request_count++;
Willy Tarreau51041c72007-09-09 21:56:53 +02004205
Willy Tarreau58f10d72006-12-04 02:26:12 +01004206#if defined(DEBUG_HASH)
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004207 Alert("get_srv_from_appsession\n");
Willy Tarreau51041c72007-09-09 21:56:53 +02004208 appsession_hash_dump(&(t->be->htbl_proxy));
Willy Tarreau58f10d72006-12-04 02:26:12 +01004209#endif
4210 if (asession_temp->serverid == NULL) {
Aleksandar Lazic697bbb02008-08-13 19:57:02 +02004211 /* TODO redispatch request */
Willy Tarreau58f10d72006-12-04 02:26:12 +01004212 Alert("Found Application Session without matching server.\n");
4213 } else {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004214 struct server *srv = t->be->srv;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004215 while (srv) {
4216 if (strcmp(srv->id, asession_temp->serverid) == 0) {
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004217 if (srv->state & SRV_RUNNING || t->be->options & PR_O_PERSIST) {
Willy Tarreau58f10d72006-12-04 02:26:12 +01004218 /* we found the server and it's usable */
Willy Tarreau3d300592007-03-18 18:34:41 +01004219 txn->flags &= ~TX_CK_MASK;
4220 txn->flags |= TX_CK_VALID;
4221 t->flags |= SN_DIRECT | SN_ASSIGNED;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004222 t->srv = srv;
4223 break;
4224 } else {
Willy Tarreau3d300592007-03-18 18:34:41 +01004225 txn->flags &= ~TX_CK_MASK;
4226 txn->flags |= TX_CK_DOWN;
Willy Tarreau58f10d72006-12-04 02:26:12 +01004227 }
4228 }
4229 srv = srv->next;
4230 }
4231 }
4232}
4233
4234
Willy Tarreaub2513902006-12-17 14:52:38 +01004235/*
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004236 * In a GET or HEAD request, check if the requested URI matches the stats uri
4237 * for the current backend, and if an authorization has been passed and is valid.
Willy Tarreaub2513902006-12-17 14:52:38 +01004238 *
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004239 * It is assumed that the request is either a HEAD or GET and that the
Willy Tarreaue2e27a52007-04-01 00:01:37 +02004240 * t->be->uri_auth field is valid. An HTTP/401 response may be sent, or
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004241 * produce_content() can be called to start sending data.
Willy Tarreaub2513902006-12-17 14:52:38 +01004242 *
4243 * Returns 1 if the session's state changes, otherwise 0.
4244 */
4245int stats_check_uri_auth(struct session *t, struct proxy *backend)
4246{
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004247 struct http_txn *txn = &t->txn;
Willy Tarreaub2513902006-12-17 14:52:38 +01004248 struct uri_auth *uri_auth = backend->uri_auth;
4249 struct user_auth *user;
4250 int authenticated, cur_idx;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004251 char *h;
Willy Tarreaub2513902006-12-17 14:52:38 +01004252
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004253 memset(&t->data_ctx.stats, 0, sizeof(t->data_ctx.stats));
4254
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004255 /* check URI size */
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004256 if (uri_auth->uri_len > txn->req.sl.rq.u_l)
Willy Tarreaub2513902006-12-17 14:52:38 +01004257 return 0;
4258
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004259 h = t->req->data + txn->req.sl.rq.u;
Willy Tarreau8d5d7f22007-01-21 19:16:41 +01004260
Willy Tarreau0214c3a2007-01-07 13:47:30 +01004261 /* the URI is in h */
4262 if (memcmp(h, uri_auth->uri_prefix, uri_auth->uri_len) != 0)
Willy Tarreaub2513902006-12-17 14:52:38 +01004263 return 0;
4264
Willy Tarreaue7150cd2007-07-25 14:43:32 +02004265 h += uri_auth->uri_len;
4266 while (h <= t->req->data + txn->req.sl.rq.u + txn->req.sl.rq.u_l - 3) {
4267 if (memcmp(h, ";up", 3) == 0) {
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004268 t->data_ctx.stats.flags |= STAT_HIDE_DOWN;
Willy Tarreaue7150cd2007-07-25 14:43:32 +02004269 break;
4270 }
4271 h++;
4272 }
4273
4274 if (uri_auth->refresh) {
4275 h = t->req->data + txn->req.sl.rq.u + uri_auth->uri_len;
4276 while (h <= t->req->data + txn->req.sl.rq.u + txn->req.sl.rq.u_l - 10) {
4277 if (memcmp(h, ";norefresh", 10) == 0) {
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004278 t->data_ctx.stats.flags |= STAT_NO_REFRESH;
Willy Tarreaue7150cd2007-07-25 14:43:32 +02004279 break;
4280 }
4281 h++;
4282 }
4283 }
4284
Willy Tarreau55bb8452007-10-17 18:44:57 +02004285 h = t->req->data + txn->req.sl.rq.u + uri_auth->uri_len;
4286 while (h <= t->req->data + txn->req.sl.rq.u + txn->req.sl.rq.u_l - 4) {
4287 if (memcmp(h, ";csv", 4) == 0) {
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004288 t->data_ctx.stats.flags |= STAT_FMT_CSV;
Willy Tarreau55bb8452007-10-17 18:44:57 +02004289 break;
4290 }
4291 h++;
4292 }
4293
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004294 t->data_ctx.stats.flags |= STAT_SHOW_STAT | STAT_SHOW_INFO;
4295
Willy Tarreaub2513902006-12-17 14:52:38 +01004296 /* we are in front of a interceptable URI. Let's check
4297 * if there's an authentication and if it's valid.
4298 */
4299 user = uri_auth->users;
4300 if (!user) {
4301 /* no user auth required, it's OK */
4302 authenticated = 1;
4303 } else {
4304 authenticated = 0;
4305
4306 /* a user list is defined, we have to check.
4307 * skip 21 chars for "Authorization: Basic ".
4308 */
4309
4310 /* FIXME: this should move to an earlier place */
4311 cur_idx = 0;
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004312 h = t->req->data + txn->req.som + hdr_idx_first_pos(&txn->hdr_idx);
4313 while ((cur_idx = txn->hdr_idx.v[cur_idx].next)) {
4314 int len = txn->hdr_idx.v[cur_idx].len;
Willy Tarreaub2513902006-12-17 14:52:38 +01004315 if (len > 14 &&
4316 !strncasecmp("Authorization:", h, 14)) {
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004317 txn->auth_hdr.str = h;
4318 txn->auth_hdr.len = len;
Willy Tarreaub2513902006-12-17 14:52:38 +01004319 break;
4320 }
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004321 h += len + txn->hdr_idx.v[cur_idx].cr + 1;
Willy Tarreaub2513902006-12-17 14:52:38 +01004322 }
4323
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004324 if (txn->auth_hdr.len < 21 ||
4325 memcmp(txn->auth_hdr.str + 14, " Basic ", 7))
Willy Tarreaub2513902006-12-17 14:52:38 +01004326 user = NULL;
4327
4328 while (user) {
Willy Tarreau4dbc4a22007-03-03 16:23:22 +01004329 if ((txn->auth_hdr.len == user->user_len + 14 + 7)
4330 && !memcmp(txn->auth_hdr.str + 14 + 7,
Willy Tarreaub2513902006-12-17 14:52:38 +01004331 user->user_pwd, user->user_len)) {
4332 authenticated = 1;
4333 break;
4334 }
4335 user = user->next;
4336 }
4337 }
4338
4339 if (!authenticated) {
Willy Tarreau0f772532006-12-23 20:51:41 +01004340 struct chunk msg;
Willy Tarreaub2513902006-12-17 14:52:38 +01004341
4342 /* no need to go further */
Willy Tarreau0f772532006-12-23 20:51:41 +01004343 msg.str = trash;
4344 msg.len = sprintf(trash, HTTP_401_fmt, uri_auth->auth_realm);
Willy Tarreau3bac9ff2007-03-18 17:31:28 +01004345 txn->status = 401;
Willy Tarreaudded32d2008-11-30 19:48:07 +01004346 stream_int_retnclose(t->req->prod, &msg);
Willy Tarreau2df28e82008-08-17 15:20:19 +02004347 t->req->analysers = 0;
Willy Tarreaub2513902006-12-17 14:52:38 +01004348 if (!(t->flags & SN_ERR_MASK))
4349 t->flags |= SN_ERR_PRXCOND;
4350 if (!(t->flags & SN_FINST_MASK))
4351 t->flags |= SN_FINST_R;
4352 return 1;
4353 }
4354
Willy Tarreau39f7e6d2008-03-17 21:38:24 +01004355 /* The request is valid, the user is authenticated. Let's start sending
Willy Tarreaub2513902006-12-17 14:52:38 +01004356 * data.
4357 */
Willy Tarreauefb453c2008-10-26 20:49:47 +01004358 buffer_write_dis(t->req);
Willy Tarreau72b179a2008-08-28 16:01:32 +02004359 buffer_shutw_now(t->req);
4360 buffer_shutr_now(t->rep);
Willy Tarreau70089872008-06-13 21:12:51 +02004361 t->logs.tv_request = now;
Willy Tarreaub2513902006-12-17 14:52:38 +01004362 t->data_source = DATA_SRC_STATS;
4363 t->data_state = DATA_ST_INIT;
Willy Tarreau91e99932008-06-30 07:51:00 +02004364 t->task->nice = -32; /* small boost for HTTP statistics */
Willy Tarreau01bf8672008-12-07 18:03:29 +01004365 buffer_install_hijacker(t, t->rep, produce_content);
Willy Tarreaub2513902006-12-17 14:52:38 +01004366 return 1;
4367}
4368
4369
Willy Tarreaubaaee002006-06-26 02:48:02 +02004370/*
Willy Tarreau58f10d72006-12-04 02:26:12 +01004371 * Print a debug line with a header
4372 */
4373void debug_hdr(const char *dir, struct session *t, const char *start, const char *end)
4374{
4375 int len, max;
4376 len = sprintf(trash, "%08x:%s.%s[%04x:%04x]: ", t->uniq_id, t->be->id,
Willy Tarreau3a16b2c2008-08-28 08:54:27 +02004377 dir, (unsigned short)t->req->prod->fd, (unsigned short)t->req->cons->fd);
Willy Tarreau58f10d72006-12-04 02:26:12 +01004378 max = end - start;
4379 UBOUND(max, sizeof(trash) - len - 1);
4380 len += strlcpy2(trash + len, start, max + 1);
4381 trash[len++] = '\n';
4382 write(1, trash, len);
4383}
4384
4385
Willy Tarreau8797c062007-05-07 00:55:35 +02004386/************************************************************************/
4387/* The code below is dedicated to ACL parsing and matching */
4388/************************************************************************/
4389
4390
4391
4392
4393/* 1. Check on METHOD
4394 * We use the pre-parsed method if it is known, and store its number as an
4395 * integer. If it is unknown, we use the pointer and the length.
4396 */
Willy Tarreauae8b7962007-06-09 23:10:04 +02004397static int acl_parse_meth(const char **text, struct acl_pattern *pattern, int *opaque)
Willy Tarreau8797c062007-05-07 00:55:35 +02004398{
4399 int len, meth;
4400
Willy Tarreauae8b7962007-06-09 23:10:04 +02004401 len = strlen(*text);
4402 meth = find_http_meth(*text, len);
Willy Tarreau8797c062007-05-07 00:55:35 +02004403
4404 pattern->val.i = meth;
4405 if (meth == HTTP_METH_OTHER) {
Willy Tarreauae8b7962007-06-09 23:10:04 +02004406 pattern->ptr.str = strdup(*text);
Willy Tarreau8797c062007-05-07 00:55:35 +02004407 if (!pattern->ptr.str)
4408 return 0;
4409 pattern->len = len;
4410 }
4411 return 1;
4412}
4413
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004414static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004415acl_fetch_meth(struct proxy *px, struct session *l4, void *l7, int dir,
4416 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004417{
4418 int meth;
4419 struct http_txn *txn = l7;
4420
Willy Tarreaub6866442008-07-14 23:54:42 +02004421 if (!txn)
4422 return 0;
4423
Willy Tarreauc11416f2007-06-17 16:58:38 +02004424 if (txn->req.msg_state != HTTP_MSG_BODY)
4425 return 0;
4426
Willy Tarreau8797c062007-05-07 00:55:35 +02004427 meth = txn->meth;
4428 test->i = meth;
4429 if (meth == HTTP_METH_OTHER) {
Willy Tarreauc11416f2007-06-17 16:58:38 +02004430 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4431 /* ensure the indexes are not affected */
4432 return 0;
Willy Tarreau8797c062007-05-07 00:55:35 +02004433 test->len = txn->req.sl.rq.m_l;
4434 test->ptr = txn->req.sol;
4435 }
4436 test->flags = ACL_TEST_F_READ_ONLY | ACL_TEST_F_VOL_1ST;
4437 return 1;
4438}
4439
4440static int acl_match_meth(struct acl_test *test, struct acl_pattern *pattern)
4441{
Willy Tarreauc8d7c962007-06-17 08:20:33 +02004442 int icase;
4443
Willy Tarreau8797c062007-05-07 00:55:35 +02004444 if (test->i != pattern->val.i)
Willy Tarreau11382812008-07-09 16:18:21 +02004445 return ACL_PAT_FAIL;
Willy Tarreau8797c062007-05-07 00:55:35 +02004446
4447 if (test->i != HTTP_METH_OTHER)
Willy Tarreau11382812008-07-09 16:18:21 +02004448 return ACL_PAT_PASS;
Willy Tarreau8797c062007-05-07 00:55:35 +02004449
4450 /* Other method, we must compare the strings */
4451 if (pattern->len != test->len)
Willy Tarreau11382812008-07-09 16:18:21 +02004452 return ACL_PAT_FAIL;
Willy Tarreauc8d7c962007-06-17 08:20:33 +02004453
4454 icase = pattern->flags & ACL_PAT_F_IGNORE_CASE;
4455 if ((icase && strncasecmp(pattern->ptr.str, test->ptr, test->len) != 0) ||
4456 (!icase && strncmp(pattern->ptr.str, test->ptr, test->len) != 0))
Willy Tarreau11382812008-07-09 16:18:21 +02004457 return ACL_PAT_FAIL;
4458 return ACL_PAT_PASS;
Willy Tarreau8797c062007-05-07 00:55:35 +02004459}
4460
4461/* 2. Check on Request/Status Version
4462 * We simply compare strings here.
4463 */
Willy Tarreauae8b7962007-06-09 23:10:04 +02004464static int acl_parse_ver(const char **text, struct acl_pattern *pattern, int *opaque)
Willy Tarreau8797c062007-05-07 00:55:35 +02004465{
Willy Tarreauae8b7962007-06-09 23:10:04 +02004466 pattern->ptr.str = strdup(*text);
Willy Tarreau8797c062007-05-07 00:55:35 +02004467 if (!pattern->ptr.str)
4468 return 0;
Willy Tarreauae8b7962007-06-09 23:10:04 +02004469 pattern->len = strlen(*text);
Willy Tarreau8797c062007-05-07 00:55:35 +02004470 return 1;
4471}
4472
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004473static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004474acl_fetch_rqver(struct proxy *px, struct session *l4, void *l7, int dir,
4475 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004476{
4477 struct http_txn *txn = l7;
4478 char *ptr;
4479 int len;
4480
Willy Tarreaub6866442008-07-14 23:54:42 +02004481 if (!txn)
4482 return 0;
4483
Willy Tarreauc11416f2007-06-17 16:58:38 +02004484 if (txn->req.msg_state != HTTP_MSG_BODY)
4485 return 0;
4486
Willy Tarreau8797c062007-05-07 00:55:35 +02004487 len = txn->req.sl.rq.v_l;
4488 ptr = txn->req.sol + txn->req.sl.rq.v - txn->req.som;
4489
4490 while ((len-- > 0) && (*ptr++ != '/'));
4491 if (len <= 0)
4492 return 0;
4493
4494 test->ptr = ptr;
4495 test->len = len;
4496
4497 test->flags = ACL_TEST_F_READ_ONLY | ACL_TEST_F_VOL_1ST;
4498 return 1;
4499}
4500
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004501static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004502acl_fetch_stver(struct proxy *px, struct session *l4, void *l7, int dir,
4503 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004504{
4505 struct http_txn *txn = l7;
4506 char *ptr;
4507 int len;
4508
Willy Tarreaub6866442008-07-14 23:54:42 +02004509 if (!txn)
4510 return 0;
4511
Willy Tarreauc11416f2007-06-17 16:58:38 +02004512 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4513 return 0;
4514
Willy Tarreau8797c062007-05-07 00:55:35 +02004515 len = txn->rsp.sl.st.v_l;
4516 ptr = txn->rsp.sol;
4517
4518 while ((len-- > 0) && (*ptr++ != '/'));
4519 if (len <= 0)
4520 return 0;
4521
4522 test->ptr = ptr;
4523 test->len = len;
4524
4525 test->flags = ACL_TEST_F_READ_ONLY | ACL_TEST_F_VOL_1ST;
4526 return 1;
4527}
4528
4529/* 3. Check on Status Code. We manipulate integers here. */
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004530static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004531acl_fetch_stcode(struct proxy *px, struct session *l4, void *l7, int dir,
4532 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004533{
4534 struct http_txn *txn = l7;
4535 char *ptr;
4536 int len;
4537
Willy Tarreaub6866442008-07-14 23:54:42 +02004538 if (!txn)
4539 return 0;
4540
Willy Tarreauc11416f2007-06-17 16:58:38 +02004541 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4542 return 0;
4543
Willy Tarreau8797c062007-05-07 00:55:35 +02004544 len = txn->rsp.sl.st.c_l;
4545 ptr = txn->rsp.sol + txn->rsp.sl.st.c - txn->rsp.som;
4546
4547 test->i = __strl2ui(ptr, len);
4548 test->flags = ACL_TEST_F_VOL_1ST;
4549 return 1;
4550}
4551
4552/* 4. Check on URL/URI. A pointer to the URI is stored. */
Willy Tarreaud41f8d82007-06-10 10:06:18 +02004553static int
Willy Tarreau97be1452007-06-10 11:47:14 +02004554acl_fetch_url(struct proxy *px, struct session *l4, void *l7, int dir,
4555 struct acl_expr *expr, struct acl_test *test)
Willy Tarreau8797c062007-05-07 00:55:35 +02004556{
4557 struct http_txn *txn = l7;
4558
Willy Tarreaub6866442008-07-14 23:54:42 +02004559 if (!txn)
4560 return 0;
4561
Willy Tarreauc11416f2007-06-17 16:58:38 +02004562 if (txn->req.msg_state != HTTP_MSG_BODY)
4563 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004564
Willy Tarreauc11416f2007-06-17 16:58:38 +02004565 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4566 /* ensure the indexes are not affected */
4567 return 0;
4568
Willy Tarreau8797c062007-05-07 00:55:35 +02004569 test->len = txn->req.sl.rq.u_l;
4570 test->ptr = txn->req.sol + txn->req.sl.rq.u;
4571
Willy Tarreauf3d25982007-05-08 22:45:09 +02004572 /* we do not need to set READ_ONLY because the data is in a buffer */
4573 test->flags = ACL_TEST_F_VOL_1ST;
Willy Tarreau8797c062007-05-07 00:55:35 +02004574 return 1;
4575}
4576
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004577static int
4578acl_fetch_url_ip(struct proxy *px, struct session *l4, void *l7, int dir,
4579 struct acl_expr *expr, struct acl_test *test)
4580{
4581 struct http_txn *txn = l7;
4582
Willy Tarreaub6866442008-07-14 23:54:42 +02004583 if (!txn)
4584 return 0;
4585
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004586 if (txn->req.msg_state != HTTP_MSG_BODY)
4587 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004588
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004589 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4590 /* ensure the indexes are not affected */
4591 return 0;
4592
4593 /* Parse HTTP request */
4594 url2sa(txn->req.sol + txn->req.sl.rq.u, txn->req.sl.rq.u_l, &l4->srv_addr);
4595 test->ptr = (void *)&((struct sockaddr_in *)&l4->srv_addr)->sin_addr;
4596 test->i = AF_INET;
4597
4598 /*
4599 * If we are parsing url in frontend space, we prepare backend stage
4600 * to not parse again the same url ! optimization lazyness...
4601 */
4602 if (px->options & PR_O_HTTP_PROXY)
4603 l4->flags |= SN_ADDR_SET;
4604
4605 test->flags = ACL_TEST_F_READ_ONLY;
4606 return 1;
4607}
4608
4609static int
4610acl_fetch_url_port(struct proxy *px, struct session *l4, void *l7, int dir,
4611 struct acl_expr *expr, struct acl_test *test)
4612{
4613 struct http_txn *txn = l7;
4614
Willy Tarreaub6866442008-07-14 23:54:42 +02004615 if (!txn)
4616 return 0;
4617
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004618 if (txn->req.msg_state != HTTP_MSG_BODY)
4619 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004620
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01004621 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4622 /* ensure the indexes are not affected */
4623 return 0;
4624
4625 /* Same optimization as url_ip */
4626 url2sa(txn->req.sol + txn->req.sl.rq.u, txn->req.sl.rq.u_l, &l4->srv_addr);
4627 test->i = ntohs(((struct sockaddr_in *)&l4->srv_addr)->sin_port);
4628
4629 if (px->options & PR_O_HTTP_PROXY)
4630 l4->flags |= SN_ADDR_SET;
4631
4632 test->flags = ACL_TEST_F_READ_ONLY;
4633 return 1;
4634}
4635
Willy Tarreauc11416f2007-06-17 16:58:38 +02004636/* 5. Check on HTTP header. A pointer to the beginning of the value is returned.
4637 * This generic function is used by both acl_fetch_chdr() and acl_fetch_shdr().
4638 */
Willy Tarreau33a7e692007-06-10 19:45:56 +02004639static int
Willy Tarreauc11416f2007-06-17 16:58:38 +02004640acl_fetch_hdr(struct proxy *px, struct session *l4, void *l7, char *sol,
Willy Tarreau33a7e692007-06-10 19:45:56 +02004641 struct acl_expr *expr, struct acl_test *test)
4642{
4643 struct http_txn *txn = l7;
4644 struct hdr_idx *idx = &txn->hdr_idx;
4645 struct hdr_ctx *ctx = (struct hdr_ctx *)test->ctx.a;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004646
Willy Tarreaub6866442008-07-14 23:54:42 +02004647 if (!txn)
4648 return 0;
4649
Willy Tarreau33a7e692007-06-10 19:45:56 +02004650 if (!(test->flags & ACL_TEST_F_FETCH_MORE))
4651 /* search for header from the beginning */
4652 ctx->idx = 0;
4653
Willy Tarreau33a7e692007-06-10 19:45:56 +02004654 if (http_find_header2(expr->arg.str, expr->arg_len, sol, idx, ctx)) {
4655 test->flags |= ACL_TEST_F_FETCH_MORE;
4656 test->flags |= ACL_TEST_F_VOL_HDR;
4657 test->len = ctx->vlen;
4658 test->ptr = (char *)ctx->line + ctx->val;
4659 return 1;
4660 }
4661
4662 test->flags &= ~ACL_TEST_F_FETCH_MORE;
4663 test->flags |= ACL_TEST_F_VOL_HDR;
4664 return 0;
4665}
4666
Willy Tarreau33a7e692007-06-10 19:45:56 +02004667static int
Willy Tarreauc11416f2007-06-17 16:58:38 +02004668acl_fetch_chdr(struct proxy *px, struct session *l4, void *l7, int dir,
4669 struct acl_expr *expr, struct acl_test *test)
4670{
4671 struct http_txn *txn = l7;
4672
Willy Tarreaub6866442008-07-14 23:54:42 +02004673 if (!txn)
4674 return 0;
4675
Willy Tarreauc11416f2007-06-17 16:58:38 +02004676 if (txn->req.msg_state != HTTP_MSG_BODY)
4677 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004678
Willy Tarreauc11416f2007-06-17 16:58:38 +02004679 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4680 /* ensure the indexes are not affected */
4681 return 0;
4682
4683 return acl_fetch_hdr(px, l4, txn, txn->req.sol, expr, test);
4684}
4685
4686static int
4687acl_fetch_shdr(struct proxy *px, struct session *l4, void *l7, int dir,
4688 struct acl_expr *expr, struct acl_test *test)
4689{
4690 struct http_txn *txn = l7;
4691
Willy Tarreaub6866442008-07-14 23:54:42 +02004692 if (!txn)
4693 return 0;
4694
Willy Tarreauc11416f2007-06-17 16:58:38 +02004695 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4696 return 0;
4697
4698 return acl_fetch_hdr(px, l4, txn, txn->rsp.sol, expr, test);
4699}
4700
4701/* 6. Check on HTTP header count. The number of occurrences is returned.
4702 * This generic function is used by both acl_fetch_chdr* and acl_fetch_shdr*.
4703 */
4704static int
4705acl_fetch_hdr_cnt(struct proxy *px, struct session *l4, void *l7, char *sol,
Willy Tarreau33a7e692007-06-10 19:45:56 +02004706 struct acl_expr *expr, struct acl_test *test)
4707{
4708 struct http_txn *txn = l7;
4709 struct hdr_idx *idx = &txn->hdr_idx;
4710 struct hdr_ctx ctx;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004711 int cnt;
Willy Tarreau8797c062007-05-07 00:55:35 +02004712
Willy Tarreaub6866442008-07-14 23:54:42 +02004713 if (!txn)
4714 return 0;
4715
Willy Tarreau33a7e692007-06-10 19:45:56 +02004716 ctx.idx = 0;
4717 cnt = 0;
4718 while (http_find_header2(expr->arg.str, expr->arg_len, sol, idx, &ctx))
4719 cnt++;
4720
4721 test->i = cnt;
4722 test->flags = ACL_TEST_F_VOL_HDR;
4723 return 1;
4724}
4725
Willy Tarreauc11416f2007-06-17 16:58:38 +02004726static int
4727acl_fetch_chdr_cnt(struct proxy *px, struct session *l4, void *l7, int dir,
4728 struct acl_expr *expr, struct acl_test *test)
4729{
4730 struct http_txn *txn = l7;
4731
Willy Tarreaub6866442008-07-14 23:54:42 +02004732 if (!txn)
4733 return 0;
4734
Willy Tarreauc11416f2007-06-17 16:58:38 +02004735 if (txn->req.msg_state != HTTP_MSG_BODY)
4736 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004737
Willy Tarreauc11416f2007-06-17 16:58:38 +02004738 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4739 /* ensure the indexes are not affected */
4740 return 0;
4741
4742 return acl_fetch_hdr_cnt(px, l4, txn, txn->req.sol, expr, test);
4743}
4744
4745static int
4746acl_fetch_shdr_cnt(struct proxy *px, struct session *l4, void *l7, int dir,
4747 struct acl_expr *expr, struct acl_test *test)
4748{
4749 struct http_txn *txn = l7;
4750
Willy Tarreaub6866442008-07-14 23:54:42 +02004751 if (!txn)
4752 return 0;
4753
Willy Tarreauc11416f2007-06-17 16:58:38 +02004754 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4755 return 0;
4756
4757 return acl_fetch_hdr_cnt(px, l4, txn, txn->rsp.sol, expr, test);
4758}
4759
Willy Tarreau33a7e692007-06-10 19:45:56 +02004760/* 7. Check on HTTP header's integer value. The integer value is returned.
4761 * FIXME: the type is 'int', it may not be appropriate for everything.
Willy Tarreauc11416f2007-06-17 16:58:38 +02004762 * This generic function is used by both acl_fetch_chdr* and acl_fetch_shdr*.
Willy Tarreau33a7e692007-06-10 19:45:56 +02004763 */
4764static int
Willy Tarreauc11416f2007-06-17 16:58:38 +02004765acl_fetch_hdr_val(struct proxy *px, struct session *l4, void *l7, char *sol,
Willy Tarreau33a7e692007-06-10 19:45:56 +02004766 struct acl_expr *expr, struct acl_test *test)
4767{
4768 struct http_txn *txn = l7;
4769 struct hdr_idx *idx = &txn->hdr_idx;
4770 struct hdr_ctx *ctx = (struct hdr_ctx *)test->ctx.a;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004771
Willy Tarreaub6866442008-07-14 23:54:42 +02004772 if (!txn)
4773 return 0;
4774
Willy Tarreau33a7e692007-06-10 19:45:56 +02004775 if (!(test->flags & ACL_TEST_F_FETCH_MORE))
4776 /* search for header from the beginning */
4777 ctx->idx = 0;
4778
Willy Tarreau33a7e692007-06-10 19:45:56 +02004779 if (http_find_header2(expr->arg.str, expr->arg_len, sol, idx, ctx)) {
4780 test->flags |= ACL_TEST_F_FETCH_MORE;
4781 test->flags |= ACL_TEST_F_VOL_HDR;
4782 test->i = strl2ic((char *)ctx->line + ctx->val, ctx->vlen);
4783 return 1;
4784 }
4785
4786 test->flags &= ~ACL_TEST_F_FETCH_MORE;
4787 test->flags |= ACL_TEST_F_VOL_HDR;
4788 return 0;
4789}
4790
Willy Tarreauc11416f2007-06-17 16:58:38 +02004791static int
4792acl_fetch_chdr_val(struct proxy *px, struct session *l4, void *l7, int dir,
4793 struct acl_expr *expr, struct acl_test *test)
4794{
4795 struct http_txn *txn = l7;
4796
Willy Tarreaub6866442008-07-14 23:54:42 +02004797 if (!txn)
4798 return 0;
4799
Willy Tarreauc11416f2007-06-17 16:58:38 +02004800 if (txn->req.msg_state != HTTP_MSG_BODY)
4801 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004802
Willy Tarreauc11416f2007-06-17 16:58:38 +02004803 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4804 /* ensure the indexes are not affected */
4805 return 0;
4806
4807 return acl_fetch_hdr_val(px, l4, txn, txn->req.sol, expr, test);
4808}
4809
4810static int
4811acl_fetch_shdr_val(struct proxy *px, struct session *l4, void *l7, int dir,
4812 struct acl_expr *expr, struct acl_test *test)
4813{
4814 struct http_txn *txn = l7;
4815
Willy Tarreaub6866442008-07-14 23:54:42 +02004816 if (!txn)
4817 return 0;
4818
Willy Tarreauc11416f2007-06-17 16:58:38 +02004819 if (txn->rsp.msg_state != HTTP_MSG_BODY)
4820 return 0;
4821
4822 return acl_fetch_hdr_val(px, l4, txn, txn->rsp.sol, expr, test);
4823}
4824
Willy Tarreau737b0c12007-06-10 21:28:46 +02004825/* 8. Check on URI PATH. A pointer to the PATH is stored. The path starts at
4826 * the first '/' after the possible hostname, and ends before the possible '?'.
4827 */
4828static int
4829acl_fetch_path(struct proxy *px, struct session *l4, void *l7, int dir,
4830 struct acl_expr *expr, struct acl_test *test)
4831{
4832 struct http_txn *txn = l7;
4833 char *ptr, *end;
Willy Tarreau33a7e692007-06-10 19:45:56 +02004834
Willy Tarreaub6866442008-07-14 23:54:42 +02004835 if (!txn)
4836 return 0;
4837
Willy Tarreauc11416f2007-06-17 16:58:38 +02004838 if (txn->req.msg_state != HTTP_MSG_BODY)
4839 return 0;
Willy Tarreaub6866442008-07-14 23:54:42 +02004840
Willy Tarreauc11416f2007-06-17 16:58:38 +02004841 if (txn->rsp.msg_state != HTTP_MSG_RPBEFORE)
4842 /* ensure the indexes are not affected */
4843 return 0;
4844
Willy Tarreau21d2af32008-02-14 20:25:24 +01004845 end = txn->req.sol + txn->req.sl.rq.u + txn->req.sl.rq.u_l;
4846 ptr = http_get_path(txn);
4847 if (!ptr)
Willy Tarreau737b0c12007-06-10 21:28:46 +02004848 return 0;
4849
4850 /* OK, we got the '/' ! */
4851 test->ptr = ptr;
4852
4853 while (ptr < end && *ptr != '?')
4854 ptr++;
4855
4856 test->len = ptr - test->ptr;
4857
4858 /* we do not need to set READ_ONLY because the data is in a buffer */
4859 test->flags = ACL_TEST_F_VOL_1ST;
4860 return 1;
4861}
4862
4863
Willy Tarreau8797c062007-05-07 00:55:35 +02004864
4865/************************************************************************/
4866/* All supported keywords must be declared here. */
4867/************************************************************************/
4868
4869/* Note: must not be declared <const> as its list will be overwritten */
4870static struct acl_kw_list acl_kws = {{ },{
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004871 { "method", acl_parse_meth, acl_fetch_meth, acl_match_meth, ACL_USE_L7REQ_PERMANENT },
4872 { "req_ver", acl_parse_ver, acl_fetch_rqver, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4873 { "resp_ver", acl_parse_ver, acl_fetch_stver, acl_match_str, ACL_USE_L7RTR_VOLATILE },
4874 { "status", acl_parse_int, acl_fetch_stcode, acl_match_int, ACL_USE_L7RTR_PERMANENT },
Willy Tarreau8797c062007-05-07 00:55:35 +02004875
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004876 { "url", acl_parse_str, acl_fetch_url, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4877 { "url_beg", acl_parse_str, acl_fetch_url, acl_match_beg, ACL_USE_L7REQ_VOLATILE },
4878 { "url_end", acl_parse_str, acl_fetch_url, acl_match_end, ACL_USE_L7REQ_VOLATILE },
4879 { "url_sub", acl_parse_str, acl_fetch_url, acl_match_sub, ACL_USE_L7REQ_VOLATILE },
4880 { "url_dir", acl_parse_str, acl_fetch_url, acl_match_dir, ACL_USE_L7REQ_VOLATILE },
4881 { "url_dom", acl_parse_str, acl_fetch_url, acl_match_dom, ACL_USE_L7REQ_VOLATILE },
4882 { "url_reg", acl_parse_reg, acl_fetch_url, acl_match_reg, ACL_USE_L7REQ_VOLATILE },
4883 { "url_ip", acl_parse_ip, acl_fetch_url_ip, acl_match_ip, ACL_USE_L7REQ_VOLATILE },
4884 { "url_port", acl_parse_int, acl_fetch_url_port, acl_match_int, ACL_USE_L7REQ_VOLATILE },
Willy Tarreau8797c062007-05-07 00:55:35 +02004885
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004886 /* note: we should set hdr* to use ACL_USE_HDR_VOLATILE, and chdr* to use L7REQ_VOLATILE */
4887 { "hdr", acl_parse_str, acl_fetch_chdr, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4888 { "hdr_reg", acl_parse_reg, acl_fetch_chdr, acl_match_reg, ACL_USE_L7REQ_VOLATILE },
4889 { "hdr_beg", acl_parse_str, acl_fetch_chdr, acl_match_beg, ACL_USE_L7REQ_VOLATILE },
4890 { "hdr_end", acl_parse_str, acl_fetch_chdr, acl_match_end, ACL_USE_L7REQ_VOLATILE },
4891 { "hdr_sub", acl_parse_str, acl_fetch_chdr, acl_match_sub, ACL_USE_L7REQ_VOLATILE },
4892 { "hdr_dir", acl_parse_str, acl_fetch_chdr, acl_match_dir, ACL_USE_L7REQ_VOLATILE },
4893 { "hdr_dom", acl_parse_str, acl_fetch_chdr, acl_match_dom, ACL_USE_L7REQ_VOLATILE },
4894 { "hdr_cnt", acl_parse_int, acl_fetch_chdr_cnt,acl_match_int, ACL_USE_L7REQ_VOLATILE },
4895 { "hdr_val", acl_parse_int, acl_fetch_chdr_val,acl_match_int, ACL_USE_L7REQ_VOLATILE },
Willy Tarreauc11416f2007-06-17 16:58:38 +02004896
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004897 { "shdr", acl_parse_str, acl_fetch_shdr, acl_match_str, ACL_USE_L7RTR_VOLATILE },
4898 { "shdr_reg", acl_parse_reg, acl_fetch_shdr, acl_match_reg, ACL_USE_L7RTR_VOLATILE },
4899 { "shdr_beg", acl_parse_str, acl_fetch_shdr, acl_match_beg, ACL_USE_L7RTR_VOLATILE },
4900 { "shdr_end", acl_parse_str, acl_fetch_shdr, acl_match_end, ACL_USE_L7RTR_VOLATILE },
4901 { "shdr_sub", acl_parse_str, acl_fetch_shdr, acl_match_sub, ACL_USE_L7RTR_VOLATILE },
4902 { "shdr_dir", acl_parse_str, acl_fetch_shdr, acl_match_dir, ACL_USE_L7RTR_VOLATILE },
4903 { "shdr_dom", acl_parse_str, acl_fetch_shdr, acl_match_dom, ACL_USE_L7RTR_VOLATILE },
4904 { "shdr_cnt", acl_parse_int, acl_fetch_shdr_cnt,acl_match_int, ACL_USE_L7RTR_VOLATILE },
4905 { "shdr_val", acl_parse_int, acl_fetch_shdr_val,acl_match_int, ACL_USE_L7RTR_VOLATILE },
Willy Tarreau737b0c12007-06-10 21:28:46 +02004906
Willy Tarreau0ceba5a2008-07-25 19:31:03 +02004907 { "path", acl_parse_str, acl_fetch_path, acl_match_str, ACL_USE_L7REQ_VOLATILE },
4908 { "path_reg", acl_parse_reg, acl_fetch_path, acl_match_reg, ACL_USE_L7REQ_VOLATILE },
4909 { "path_beg", acl_parse_str, acl_fetch_path, acl_match_beg, ACL_USE_L7REQ_VOLATILE },
4910 { "path_end", acl_parse_str, acl_fetch_path, acl_match_end, ACL_USE_L7REQ_VOLATILE },
4911 { "path_sub", acl_parse_str, acl_fetch_path, acl_match_sub, ACL_USE_L7REQ_VOLATILE },
4912 { "path_dir", acl_parse_str, acl_fetch_path, acl_match_dir, ACL_USE_L7REQ_VOLATILE },
4913 { "path_dom", acl_parse_str, acl_fetch_path, acl_match_dom, ACL_USE_L7REQ_VOLATILE },
Willy Tarreau737b0c12007-06-10 21:28:46 +02004914
Willy Tarreauf3d25982007-05-08 22:45:09 +02004915 { NULL, NULL, NULL, NULL },
4916
4917#if 0
Willy Tarreau8797c062007-05-07 00:55:35 +02004918 { "line", acl_parse_str, acl_fetch_line, acl_match_str },
4919 { "line_reg", acl_parse_reg, acl_fetch_line, acl_match_reg },
4920 { "line_beg", acl_parse_str, acl_fetch_line, acl_match_beg },
4921 { "line_end", acl_parse_str, acl_fetch_line, acl_match_end },
4922 { "line_sub", acl_parse_str, acl_fetch_line, acl_match_sub },
4923 { "line_dir", acl_parse_str, acl_fetch_line, acl_match_dir },
4924 { "line_dom", acl_parse_str, acl_fetch_line, acl_match_dom },
4925
Willy Tarreau8797c062007-05-07 00:55:35 +02004926 { "cook", acl_parse_str, acl_fetch_cook, acl_match_str },
4927 { "cook_reg", acl_parse_reg, acl_fetch_cook, acl_match_reg },
4928 { "cook_beg", acl_parse_str, acl_fetch_cook, acl_match_beg },
4929 { "cook_end", acl_parse_str, acl_fetch_cook, acl_match_end },
4930 { "cook_sub", acl_parse_str, acl_fetch_cook, acl_match_sub },
4931 { "cook_dir", acl_parse_str, acl_fetch_cook, acl_match_dir },
4932 { "cook_dom", acl_parse_str, acl_fetch_cook, acl_match_dom },
4933 { "cook_pst", acl_parse_none, acl_fetch_cook, acl_match_pst },
4934
4935 { "auth_user", acl_parse_str, acl_fetch_user, acl_match_str },
4936 { "auth_regex", acl_parse_reg, acl_fetch_user, acl_match_reg },
4937 { "auth_clear", acl_parse_str, acl_fetch_auth, acl_match_str },
4938 { "auth_md5", acl_parse_str, acl_fetch_auth, acl_match_md5 },
4939 { NULL, NULL, NULL, NULL },
4940#endif
4941}};
4942
4943
4944__attribute__((constructor))
4945static void __http_protocol_init(void)
4946{
4947 acl_register_keywords(&acl_kws);
4948}
4949
4950
Willy Tarreau58f10d72006-12-04 02:26:12 +01004951/*
Willy Tarreaubaaee002006-06-26 02:48:02 +02004952 * Local variables:
4953 * c-indent-level: 8
4954 * c-basic-offset: 8
4955 * End:
4956 */