blob: 15827db56430f9fc0857a6cefe8310b19cd8b68f [file] [log] [blame]
Willy Tarreau0da5b3b2017-09-21 09:30:46 +02001/*
2 * HTTP/1 protocol analyzer
3 *
4 * Copyright 2000-2017 Willy Tarreau <w@1wt.eu>
5 *
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
Willy Tarreau794f9af2017-07-26 09:07:47 +020013#include <ctype.h>
Willy Tarreau0da5b3b2017-09-21 09:30:46 +020014#include <common/config.h>
Willy Tarreauafba57a2018-12-11 13:44:24 +010015#include <common/h1.h>
Willy Tarreau794f9af2017-07-26 09:07:47 +020016#include <common/http-hdr.h>
Willy Tarreau0da5b3b2017-09-21 09:30:46 +020017
Willy Tarreau188e2302018-06-15 11:11:53 +020018#include <proto/channel.h>
Willy Tarreau0da5b3b2017-09-21 09:30:46 +020019
Willy Tarreau73373ab2018-09-14 17:11:33 +020020/* Parse the Content-Length header field of an HTTP/1 request. The function
21 * checks all possible occurrences of a comma-delimited value, and verifies
22 * if any of them doesn't match a previous value. It returns <0 if a value
23 * differs, 0 if the whole header can be dropped (i.e. already known), or >0
24 * if the value can be indexed (first one). In the last case, the value might
25 * be adjusted and the caller must only add the updated value.
26 */
27int h1_parse_cont_len_header(struct h1m *h1m, struct ist *value)
28{
29 char *e, *n;
30 long long cl;
31 int not_first = !!(h1m->flags & H1_MF_CLEN);
32 struct ist word;
33
34 word.ptr = value->ptr - 1; // -1 for next loop's pre-increment
35 e = value->ptr + value->len;
36
37 while (++word.ptr < e) {
38 /* skip leading delimitor and blanks */
39 if (unlikely(HTTP_IS_LWS(*word.ptr)))
40 continue;
41
42 /* digits only now */
43 for (cl = 0, n = word.ptr; n < e; n++) {
44 unsigned int c = *n - '0';
45 if (unlikely(c > 9)) {
46 /* non-digit */
47 if (unlikely(n == word.ptr)) // spaces only
48 goto fail;
49 break;
50 }
51 if (unlikely(cl > ULLONG_MAX / 10ULL))
52 goto fail; /* multiply overflow */
53 cl = cl * 10ULL;
54 if (unlikely(cl + c < cl))
55 goto fail; /* addition overflow */
56 cl = cl + c;
57 }
58
59 /* keep a copy of the exact cleaned value */
60 word.len = n - word.ptr;
61
62 /* skip trailing LWS till next comma or EOL */
63 for (; n < e; n++) {
64 if (!HTTP_IS_LWS(*n)) {
65 if (unlikely(*n != ','))
66 goto fail;
67 break;
68 }
69 }
70
71 /* if duplicate, must be equal */
72 if (h1m->flags & H1_MF_CLEN && cl != h1m->body_len)
73 goto fail;
74
75 /* OK, store this result as the one to be indexed */
76 h1m->flags |= H1_MF_CLEN;
77 h1m->curr_len = h1m->body_len = cl;
78 *value = word;
79 word.ptr = n;
80 }
81 /* here we've reached the end with a single value or a series of
82 * identical values, all matching previous series if any. The last
83 * parsed value was sent back into <value>. We just have to decide
84 * if this occurrence has to be indexed (it's the first one) or
85 * silently skipped (it's not the first one)
86 */
87 return !not_first;
88 fail:
89 return -1;
90}
91
Willy Tarreau2557f6a2018-09-14 16:34:47 +020092/* Parse the Transfer-Encoding: header field of an HTTP/1 request, looking for
93 * "chunked" being the last value, and setting H1_MF_CHNK in h1m->flags only in
94 * this case. Any other token found or any empty header field found will reset
95 * this flag, so that it accurately represents the token's presence at the last
96 * position. The H1_MF_XFER_ENC flag is always set. Note that transfer codings
97 * are case-insensitive (cf RFC7230#4).
98 */
99void h1_parse_xfer_enc_header(struct h1m *h1m, struct ist value)
100{
101 char *e, *n;
102 struct ist word;
103
104 h1m->flags |= H1_MF_XFER_ENC;
105 h1m->flags &= ~H1_MF_CHNK;
106
107 word.ptr = value.ptr - 1; // -1 for next loop's pre-increment
108 e = value.ptr + value.len;
109
110 while (++word.ptr < e) {
111 /* skip leading delimitor and blanks */
112 if (HTTP_IS_LWS(*word.ptr))
113 continue;
114
115 n = http_find_hdr_value_end(word.ptr, e); // next comma or end of line
116 word.len = n - word.ptr;
117
118 /* trim trailing blanks */
119 while (word.len && HTTP_IS_LWS(word.ptr[word.len-1]))
120 word.len--;
121
122 h1m->flags &= ~H1_MF_CHNK;
123 if (isteqi(word, ist("chunked")))
124 h1m->flags |= H1_MF_CHNK;
125
126 word.ptr = n;
127 }
128}
129
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200130/* Parse the Connection: header of an HTTP/1 request, looking for "close",
131 * "keep-alive", and "upgrade" values, and updating h1m->flags according to
132 * what was found there. Note that flags are only added, not removed, so the
133 * function is safe for being called multiple times if multiple occurrences
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100134 * are found. If the flag H1_MF_CLEAN_CONN_HDR, the header value is cleaned
135 * up from "keep-alive" and "close" values. To do so, the header value is
136 * rewritten in place and its length is updated.
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200137 */
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100138void h1_parse_connection_header(struct h1m *h1m, struct ist *value)
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200139{
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100140 char *e, *n, *p;
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200141 struct ist word;
142
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100143 word.ptr = value->ptr - 1; // -1 for next loop's pre-increment
144 p = value->ptr;
145 e = value->ptr + value->len;
146 if (h1m->flags & H1_MF_CLEAN_CONN_HDR)
147 value->len = 0;
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200148
149 while (++word.ptr < e) {
150 /* skip leading delimitor and blanks */
151 if (HTTP_IS_LWS(*word.ptr))
152 continue;
153
154 n = http_find_hdr_value_end(word.ptr, e); // next comma or end of line
155 word.len = n - word.ptr;
156
157 /* trim trailing blanks */
158 while (word.len && HTTP_IS_LWS(word.ptr[word.len-1]))
159 word.len--;
160
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100161 if (isteqi(word, ist("keep-alive"))) {
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200162 h1m->flags |= H1_MF_CONN_KAL;
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100163 if (h1m->flags & H1_MF_CLEAN_CONN_HDR)
164 goto skip_val;
165 }
166 else if (isteqi(word, ist("close"))) {
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200167 h1m->flags |= H1_MF_CONN_CLO;
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100168 if (h1m->flags & H1_MF_CLEAN_CONN_HDR)
169 goto skip_val;
170 }
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200171 else if (isteqi(word, ist("upgrade")))
172 h1m->flags |= H1_MF_CONN_UPG;
173
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100174 if (h1m->flags & H1_MF_CLEAN_CONN_HDR) {
175 if (value->ptr + value->len == p) {
176 /* no rewrite done till now */
177 value->len = n - value->ptr;
178 }
179 else {
180 if (value->len)
181 value->ptr[value->len++] = ',';
182 istcat(value, word, e - value->ptr);
183 }
184 }
185
186 skip_val:
187 word.ptr = p = n;
Willy Tarreau98f5cf72018-09-13 14:15:58 +0200188 }
189}
190
Willy Tarreau538746a2018-12-11 10:59:20 +0100191/* Macros used in the HTTP/1 parser, to check for the expected presence of
192 * certain bytes (ef: LF) or to skip to next byte and yield in case of failure.
193 */
194
195/* Expects to find an LF at <ptr>. If not, set <state> to <where> and jump to
196 * <bad>.
197 */
198#define EXPECT_LF_HERE(ptr, bad, state, where) \
199 do { \
200 if (unlikely(*(ptr) != '\n')) { \
201 state = (where); \
202 goto bad; \
203 } \
204 } while (0)
205
206/* Increments pointer <ptr>, continues to label <more> if it's still below
207 * pointer <end>, or goes to <stop> and sets <state> to <where> if the end
208 * of buffer was reached.
209 */
210#define EAT_AND_JUMP_OR_RETURN(ptr, end, more, stop, state, where) \
211 do { \
212 if (likely(++(ptr) < (end))) \
213 goto more; \
214 else { \
215 state = (where); \
216 goto stop; \
217 } \
218 } while (0)
219
Willy Tarreau794f9af2017-07-26 09:07:47 +0200220/* This function parses a contiguous HTTP/1 headers block starting at <start>
221 * and ending before <stop>, at once, and converts it a list of (name,value)
222 * pairs representing header fields into the array <hdr> of size <hdr_num>,
223 * whose last entry will have an empty name and an empty value. If <hdr_num> is
Willy Tarreau4433c082018-09-11 15:33:32 +0200224 * too small to represent the whole message, an error is returned. Some
225 * protocol elements such as content-length and transfer-encoding will be
Willy Tarreau5384aac2018-09-11 16:04:48 +0200226 * parsed and stored into h1m as well. <hdr> may be null, in which case only
227 * the parsing state will be updated. This may be used to restart the parsing
228 * where it stopped for example.
Willy Tarreau794f9af2017-07-26 09:07:47 +0200229 *
230 * For now it's limited to the response. If the header block is incomplete,
231 * 0 is returned, waiting to be called again with more data to try it again.
Willy Tarreau4433c082018-09-11 15:33:32 +0200232 * The caller is responsible for initializing h1m->state to H1_MSG_RPBEFORE,
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200233 * and h1m->next to zero on the first call, the parser will do the rest. If
234 * an incomplete message is seen, the caller only needs to present h1m->state
235 * and h1m->next again, with an empty header list so that the parser can start
236 * again. In this case, it will detect that it interrupted a previous session
237 * and will first look for the end of the message before reparsing it again and
238 * indexing it at the same time. This ensures that incomplete messages fed 1
239 * character at a time are never processed entirely more than exactly twice,
240 * and that there is no need to store all the internal state and pre-parsed
241 * headers or start line between calls.
Willy Tarreau794f9af2017-07-26 09:07:47 +0200242 *
Willy Tarreaua41393f2018-09-11 15:34:50 +0200243 * A pointer to a start line descriptor may be passed in <slp>, in which case
244 * the parser will fill it with whatever it found.
245 *
Willy Tarreau794f9af2017-07-26 09:07:47 +0200246 * The code derived from the main HTTP/1 parser above but was simplified and
247 * optimized to process responses produced or forwarded by haproxy. The caller
248 * is responsible for ensuring that the message doesn't wrap, and should ensure
249 * it is complete to avoid having to retry the operation after a failed
250 * attempt. The message is not supposed to be invalid, which is why a few
251 * properties such as the character set used in the header field names are not
252 * checked. In case of an unparsable response message, a negative value will be
253 * returned with h1m->err_pos and h1m->err_state matching the location and
254 * state where the error was met. Leading blank likes are tolerated but not
Willy Tarreau0f8fb6b2019-01-04 10:48:03 +0100255 * recommended. If flag H1_MF_HDRS_ONLY is set in h1m->flags, only headers are
256 * parsed and the start line is skipped. It is not required to set h1m->state
257 * nor h1m->next in this case.
Willy Tarreau794f9af2017-07-26 09:07:47 +0200258 *
259 * This function returns :
260 * -1 in case of error. In this case, h1m->err_state is filled (if h1m is
Willy Tarreau801250e2018-09-11 11:45:04 +0200261 * set) with the state the error occurred in and h1m->err_pos with the
Willy Tarreau794f9af2017-07-26 09:07:47 +0200262 * the position relative to <start>
263 * -2 if the output is full (hdr_num reached). err_state and err_pos also
264 * indicate where it failed.
265 * 0 in case of missing data.
266 * > 0 on success, it then corresponds to the number of bytes read since
267 * <start> so that the caller can go on with the payload.
268 */
269int h1_headers_to_hdr_list(char *start, const char *stop,
270 struct http_hdr *hdr, unsigned int hdr_num,
Willy Tarreaua41393f2018-09-11 15:34:50 +0200271 struct h1m *h1m, union h1_sl *slp)
Willy Tarreau794f9af2017-07-26 09:07:47 +0200272{
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200273 enum h1m_state state;
274 register char *ptr;
275 register const char *end;
276 unsigned int hdr_count;
277 unsigned int skip; /* number of bytes skipped at the beginning */
278 unsigned int sol; /* start of line */
279 unsigned int col; /* position of the colon */
280 unsigned int eol; /* end of line */
281 unsigned int sov; /* start of value */
Willy Tarreaua41393f2018-09-11 15:34:50 +0200282 union h1_sl sl;
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200283 int skip_update;
284 int restarting;
Christopher Faulet497ab4f2019-10-11 09:01:44 +0200285 int host_idx;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200286 struct ist n, v; /* header name and value during parsing */
287
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200288 skip = 0; // do it only once to keep track of the leading CRLF.
289
290 try_again:
291 hdr_count = sol = col = eol = sov = 0;
Willy Tarreaua41393f2018-09-11 15:34:50 +0200292 sl.st.status = 0;
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200293 skip_update = restarting = 0;
Christopher Faulet497ab4f2019-10-11 09:01:44 +0200294 host_idx = -1;
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200295
Willy Tarreau0f8fb6b2019-01-04 10:48:03 +0100296 if (h1m->flags & H1_MF_HDRS_ONLY) {
297 state = H1_MSG_HDR_FIRST;
298 h1m->next = 0;
299 }
Christopher Faulet68b1bbd2019-01-04 16:06:48 +0100300 else {
Willy Tarreau0f8fb6b2019-01-04 10:48:03 +0100301 state = h1m->state;
Christopher Faulet68b1bbd2019-01-04 16:06:48 +0100302 if (h1m->state != H1_MSG_RQBEFORE && h1m->state != H1_MSG_RPBEFORE)
303 restarting = 1;
304 }
Willy Tarreau0f8fb6b2019-01-04 10:48:03 +0100305
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200306 ptr = start + h1m->next;
307 end = stop;
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200308
Willy Tarreau794f9af2017-07-26 09:07:47 +0200309 if (unlikely(ptr >= end))
310 goto http_msg_ood;
311
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200312 /* don't update output if hdr is NULL or if we're restarting */
313 if (!hdr || restarting)
Willy Tarreau5384aac2018-09-11 16:04:48 +0200314 skip_update = 1;
315
Willy Tarreau794f9af2017-07-26 09:07:47 +0200316 switch (state) {
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200317 case H1_MSG_RQBEFORE:
318 http_msg_rqbefore:
319 if (likely(HTTP_IS_TOKEN(*ptr))) {
320 /* we have a start of message, we may have skipped some
321 * heading CRLF. Skip them now.
322 */
323 skip += ptr - start;
324 start = ptr;
325
326 sol = 0;
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200327 sl.rq.m.ptr = ptr;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200328 hdr_count = 0;
329 state = H1_MSG_RQMETH;
330 goto http_msg_rqmeth;
331 }
332
333 if (unlikely(!HTTP_IS_CRLF(*ptr))) {
334 state = H1_MSG_RQBEFORE;
335 goto http_msg_invalid;
336 }
337
338 if (unlikely(*ptr == '\n'))
339 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore, http_msg_ood, state, H1_MSG_RQBEFORE);
340 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore_cr, http_msg_ood, state, H1_MSG_RQBEFORE_CR);
341 /* stop here */
342
343 case H1_MSG_RQBEFORE_CR:
344 http_msg_rqbefore_cr:
345 EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RQBEFORE_CR);
346 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore, http_msg_ood, state, H1_MSG_RQBEFORE);
347 /* stop here */
348
349 case H1_MSG_RQMETH:
350 http_msg_rqmeth:
351 if (likely(HTTP_IS_TOKEN(*ptr)))
352 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth, http_msg_ood, state, H1_MSG_RQMETH);
353
354 if (likely(HTTP_IS_SPHT(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200355 sl.rq.m.len = ptr - sl.rq.m.ptr;
356 sl.rq.meth = find_http_meth(start, sl.rq.m.len);
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200357 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth_sp, http_msg_ood, state, H1_MSG_RQMETH_SP);
358 }
359
360 if (likely(HTTP_IS_CRLF(*ptr))) {
361 /* HTTP 0.9 request */
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200362 sl.rq.m.len = ptr - sl.rq.m.ptr;
363 sl.rq.meth = find_http_meth(sl.rq.m.ptr, sl.rq.m.len);
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200364 http_msg_req09_uri:
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200365 sl.rq.u.ptr = ptr;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200366 http_msg_req09_uri_e:
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200367 sl.rq.u.len = ptr - sl.rq.u.ptr;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200368 http_msg_req09_ver:
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200369 sl.rq.v.ptr = ptr;
370 sl.rq.v.len = 0;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200371 goto http_msg_rqline_eol;
372 }
373 state = H1_MSG_RQMETH;
374 goto http_msg_invalid;
375
376 case H1_MSG_RQMETH_SP:
377 http_msg_rqmeth_sp:
378 if (likely(!HTTP_IS_LWS(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200379 sl.rq.u.ptr = ptr;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200380 goto http_msg_rquri;
381 }
382 if (likely(HTTP_IS_SPHT(*ptr)))
383 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth_sp, http_msg_ood, state, H1_MSG_RQMETH_SP);
384 /* so it's a CR/LF, meaning an HTTP 0.9 request */
385 goto http_msg_req09_uri;
386
387 case H1_MSG_RQURI:
388 http_msg_rquri:
389#if defined(__x86_64__) || \
390 defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) || \
391 defined(__ARM_ARCH_7A__)
392 /* speedup: skip bytes not between 0x21 and 0x7e inclusive */
393 while (ptr <= end - sizeof(int)) {
394 int x = *(int *)ptr - 0x21212121;
395 if (x & 0x80808080)
396 break;
397
398 x -= 0x5e5e5e5e;
399 if (!(x & 0x80808080))
400 break;
401
402 ptr += sizeof(int);
403 }
404#endif
405 if (ptr >= end) {
406 state = H1_MSG_RQURI;
407 goto http_msg_ood;
408 }
409 http_msg_rquri2:
410 if (likely((unsigned char)(*ptr - 33) <= 93)) /* 33 to 126 included */
411 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri2, http_msg_ood, state, H1_MSG_RQURI);
412
413 if (likely(HTTP_IS_SPHT(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200414 sl.rq.u.len = ptr - sl.rq.u.ptr;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200415 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri_sp, http_msg_ood, state, H1_MSG_RQURI_SP);
416 }
417 if (likely((unsigned char)*ptr >= 128)) {
418 /* non-ASCII chars are forbidden unless option
419 * accept-invalid-http-request is enabled in the frontend.
420 * In any case, we capture the faulty char.
421 */
422 if (h1m->err_pos < -1)
423 goto invalid_char;
424 if (h1m->err_pos == -1)
425 h1m->err_pos = ptr - start + skip;
426 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri, http_msg_ood, state, H1_MSG_RQURI);
427 }
428
429 if (likely(HTTP_IS_CRLF(*ptr))) {
430 /* so it's a CR/LF, meaning an HTTP 0.9 request */
431 goto http_msg_req09_uri_e;
432 }
433
434 /* OK forbidden chars, 0..31 or 127 */
435 invalid_char:
436 state = H1_MSG_RQURI;
437 goto http_msg_invalid;
438
439 case H1_MSG_RQURI_SP:
440 http_msg_rquri_sp:
441 if (likely(!HTTP_IS_LWS(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200442 sl.rq.v.ptr = ptr;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200443 goto http_msg_rqver;
444 }
445 if (likely(HTTP_IS_SPHT(*ptr)))
446 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri_sp, http_msg_ood, state, H1_MSG_RQURI_SP);
447 /* so it's a CR/LF, meaning an HTTP 0.9 request */
448 goto http_msg_req09_ver;
449
450
451 case H1_MSG_RQVER:
452 http_msg_rqver:
453 if (likely(HTTP_IS_VER_TOKEN(*ptr)))
454 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqver, http_msg_ood, state, H1_MSG_RQVER);
455
456 if (likely(HTTP_IS_CRLF(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200457 sl.rq.v.len = ptr - sl.rq.v.ptr;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200458 http_msg_rqline_eol:
459 /* We have seen the end of line. Note that we do not
460 * necessarily have the \n yet, but at least we know that we
461 * have EITHER \r OR \n, otherwise the request would not be
462 * complete. We can then record the request length and return
463 * to the caller which will be able to register it.
464 */
465
466 if (likely(!skip_update)) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200467 if ((sl.rq.v.len == 8) &&
468 (*(sl.rq.v.ptr + 5) > '1' ||
469 (*(sl.rq.v.ptr + 5) == '1' && *(sl.rq.v.ptr + 7) >= '1')))
Willy Tarreauba5fbca2018-09-13 11:32:51 +0200470 h1m->flags |= H1_MF_VER_11;
471
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200472 if (unlikely(hdr_count >= hdr_num)) {
473 state = H1_MSG_RQVER;
474 goto http_output_full;
475 }
Christopher Faulet25da9e32018-10-08 15:50:15 +0200476 if (!(h1m->flags & H1_MF_NO_PHDR))
477 http_set_hdr(&hdr[hdr_count++], ist(":method"), sl.rq.m);
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200478
479 if (unlikely(hdr_count >= hdr_num)) {
480 state = H1_MSG_RQVER;
481 goto http_output_full;
482 }
Christopher Faulet25da9e32018-10-08 15:50:15 +0200483 if (!(h1m->flags & H1_MF_NO_PHDR))
484 http_set_hdr(&hdr[hdr_count++], ist(":path"), sl.rq.u);
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200485 }
486
487 sol = ptr - start;
488 if (likely(*ptr == '\r'))
489 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqline_end, http_msg_ood, state, H1_MSG_RQLINE_END);
490 goto http_msg_rqline_end;
491 }
492
493 /* neither an HTTP_VER token nor a CRLF */
494 state = H1_MSG_RQVER;
495 goto http_msg_invalid;
496
497 case H1_MSG_RQLINE_END:
498 http_msg_rqline_end:
499 /* check for HTTP/0.9 request : no version information
500 * available. sol must point to the first of CR or LF. However
501 * since we don't save these elements between calls, if we come
502 * here from a restart, we don't necessarily know. Thus in this
503 * case we simply start over.
504 */
505 if (restarting)
506 goto restart;
507
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200508 if (unlikely(sl.rq.v.len == 0))
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200509 goto http_msg_last_lf;
510
511 EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RQLINE_END);
512 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_first, http_msg_ood, state, H1_MSG_HDR_FIRST);
513 /* stop here */
514
515 /*
516 * Common states below
517 */
Willy Tarreau801250e2018-09-11 11:45:04 +0200518 case H1_MSG_RPBEFORE:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200519 http_msg_rpbefore:
520 if (likely(HTTP_IS_TOKEN(*ptr))) {
521 /* we have a start of message, we may have skipped some
522 * heading CRLF. Skip them now.
523 */
524 skip += ptr - start;
525 start = ptr;
526
527 sol = 0;
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200528 sl.st.v.ptr = ptr;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200529 hdr_count = 0;
Willy Tarreau801250e2018-09-11 11:45:04 +0200530 state = H1_MSG_RPVER;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200531 goto http_msg_rpver;
532 }
533
534 if (unlikely(!HTTP_IS_CRLF(*ptr))) {
Willy Tarreau801250e2018-09-11 11:45:04 +0200535 state = H1_MSG_RPBEFORE;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200536 goto http_msg_invalid;
537 }
538
539 if (unlikely(*ptr == '\n'))
Willy Tarreau801250e2018-09-11 11:45:04 +0200540 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, H1_MSG_RPBEFORE);
541 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore_cr, http_msg_ood, state, H1_MSG_RPBEFORE_CR);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200542 /* stop here */
543
Willy Tarreau801250e2018-09-11 11:45:04 +0200544 case H1_MSG_RPBEFORE_CR:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200545 http_msg_rpbefore_cr:
Willy Tarreau801250e2018-09-11 11:45:04 +0200546 EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RPBEFORE_CR);
547 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, H1_MSG_RPBEFORE);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200548 /* stop here */
549
Willy Tarreau801250e2018-09-11 11:45:04 +0200550 case H1_MSG_RPVER:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200551 http_msg_rpver:
552 if (likely(HTTP_IS_VER_TOKEN(*ptr)))
Willy Tarreau801250e2018-09-11 11:45:04 +0200553 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver, http_msg_ood, state, H1_MSG_RPVER);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200554
555 if (likely(HTTP_IS_SPHT(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200556 sl.st.v.len = ptr - sl.st.v.ptr;
Willy Tarreauba5fbca2018-09-13 11:32:51 +0200557
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200558 if ((sl.st.v.len == 8) &&
559 (*(sl.st.v.ptr + 5) > '1' ||
560 (*(sl.st.v.ptr + 5) == '1' && *(sl.st.v.ptr + 7) >= '1')))
Willy Tarreauba5fbca2018-09-13 11:32:51 +0200561 h1m->flags |= H1_MF_VER_11;
562
Willy Tarreau801250e2018-09-11 11:45:04 +0200563 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, H1_MSG_RPVER_SP);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200564 }
Willy Tarreau801250e2018-09-11 11:45:04 +0200565 state = H1_MSG_RPVER;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200566 goto http_msg_invalid;
567
Willy Tarreau801250e2018-09-11 11:45:04 +0200568 case H1_MSG_RPVER_SP:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200569 http_msg_rpver_sp:
570 if (likely(!HTTP_IS_LWS(*ptr))) {
Willy Tarreaua41393f2018-09-11 15:34:50 +0200571 sl.st.status = 0;
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200572 sl.st.c.ptr = ptr;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200573 goto http_msg_rpcode;
574 }
575 if (likely(HTTP_IS_SPHT(*ptr)))
Willy Tarreau801250e2018-09-11 11:45:04 +0200576 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, H1_MSG_RPVER_SP);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200577 /* so it's a CR/LF, this is invalid */
Willy Tarreau801250e2018-09-11 11:45:04 +0200578 state = H1_MSG_RPVER_SP;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200579 goto http_msg_invalid;
580
Willy Tarreau801250e2018-09-11 11:45:04 +0200581 case H1_MSG_RPCODE:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200582 http_msg_rpcode:
Willy Tarreau1b4cf9b2017-11-09 11:15:45 +0100583 if (likely(HTTP_IS_DIGIT(*ptr))) {
Willy Tarreaua41393f2018-09-11 15:34:50 +0200584 sl.st.status = sl.st.status * 10 + *ptr - '0';
Willy Tarreau801250e2018-09-11 11:45:04 +0200585 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode, http_msg_ood, state, H1_MSG_RPCODE);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200586 }
587
Willy Tarreau1b4cf9b2017-11-09 11:15:45 +0100588 if (unlikely(!HTTP_IS_LWS(*ptr))) {
Willy Tarreau801250e2018-09-11 11:45:04 +0200589 state = H1_MSG_RPCODE;
Willy Tarreau1b4cf9b2017-11-09 11:15:45 +0100590 goto http_msg_invalid;
591 }
592
Willy Tarreau794f9af2017-07-26 09:07:47 +0200593 if (likely(HTTP_IS_SPHT(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200594 sl.st.c.len = ptr - sl.st.c.ptr;
Willy Tarreau801250e2018-09-11 11:45:04 +0200595 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, H1_MSG_RPCODE_SP);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200596 }
597
598 /* so it's a CR/LF, so there is no reason phrase */
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200599 sl.st.c.len = ptr - sl.st.c.ptr;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200600
601 http_msg_rsp_reason:
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200602 sl.st.r.ptr = ptr;
603 sl.st.r.len = 0;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200604 goto http_msg_rpline_eol;
605
Willy Tarreau801250e2018-09-11 11:45:04 +0200606 case H1_MSG_RPCODE_SP:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200607 http_msg_rpcode_sp:
608 if (likely(!HTTP_IS_LWS(*ptr))) {
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200609 sl.st.r.ptr = ptr;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200610 goto http_msg_rpreason;
611 }
612 if (likely(HTTP_IS_SPHT(*ptr)))
Willy Tarreau801250e2018-09-11 11:45:04 +0200613 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, H1_MSG_RPCODE_SP);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200614 /* so it's a CR/LF, so there is no reason phrase */
615 goto http_msg_rsp_reason;
616
Willy Tarreau801250e2018-09-11 11:45:04 +0200617 case H1_MSG_RPREASON:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200618 http_msg_rpreason:
619 if (likely(!HTTP_IS_CRLF(*ptr)))
Willy Tarreau801250e2018-09-11 11:45:04 +0200620 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpreason, http_msg_ood, state, H1_MSG_RPREASON);
Christopher Faulet1dc2b492018-10-08 15:34:02 +0200621 sl.st.r.len = ptr - sl.st.r.ptr;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200622 http_msg_rpline_eol:
623 /* We have seen the end of line. Note that we do not
624 * necessarily have the \n yet, but at least we know that we
625 * have EITHER \r OR \n, otherwise the response would not be
626 * complete. We can then record the response length and return
627 * to the caller which will be able to register it.
628 */
629
Willy Tarreau5384aac2018-09-11 16:04:48 +0200630 if (likely(!skip_update)) {
631 if (unlikely(hdr_count >= hdr_num)) {
632 state = H1_MSG_RPREASON;
633 goto http_output_full;
634 }
Christopher Faulet25da9e32018-10-08 15:50:15 +0200635 if (!(h1m->flags & H1_MF_NO_PHDR))
636 http_set_hdr(&hdr[hdr_count++], ist(":status"), sl.st.c);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200637 }
Willy Tarreau794f9af2017-07-26 09:07:47 +0200638
639 sol = ptr - start;
640 if (likely(*ptr == '\r'))
Willy Tarreau801250e2018-09-11 11:45:04 +0200641 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpline_end, http_msg_ood, state, H1_MSG_RPLINE_END);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200642 goto http_msg_rpline_end;
643
Willy Tarreau801250e2018-09-11 11:45:04 +0200644 case H1_MSG_RPLINE_END:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200645 http_msg_rpline_end:
646 /* sol must point to the first of CR or LF. */
Willy Tarreau801250e2018-09-11 11:45:04 +0200647 EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RPLINE_END);
648 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_first, http_msg_ood, state, H1_MSG_HDR_FIRST);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200649 /* stop here */
650
Willy Tarreau801250e2018-09-11 11:45:04 +0200651 case H1_MSG_HDR_FIRST:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200652 http_msg_hdr_first:
653 sol = ptr - start;
654 if (likely(!HTTP_IS_CRLF(*ptr))) {
655 goto http_msg_hdr_name;
656 }
657
658 if (likely(*ptr == '\r'))
Willy Tarreau801250e2018-09-11 11:45:04 +0200659 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, H1_MSG_LAST_LF);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200660 goto http_msg_last_lf;
661
Willy Tarreau801250e2018-09-11 11:45:04 +0200662 case H1_MSG_HDR_NAME:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200663 http_msg_hdr_name:
664 /* assumes sol points to the first char */
665 if (likely(HTTP_IS_TOKEN(*ptr))) {
Christopher Faulet2912f872018-09-19 14:01:04 +0200666 if (!skip_update) {
667 /* turn it to lower case if needed */
668 if (isupper((unsigned char)*ptr) && h1m->flags & H1_MF_TOLOWER)
669 *ptr = tolower(*ptr);
670 }
Willy Tarreau801250e2018-09-11 11:45:04 +0200671 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, H1_MSG_HDR_NAME);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200672 }
673
674 if (likely(*ptr == ':')) {
675 col = ptr - start;
Willy Tarreau801250e2018-09-11 11:45:04 +0200676 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, H1_MSG_HDR_L1_SP);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200677 }
678
Willy Tarreau9aec3052018-09-12 09:20:40 +0200679 if (likely(h1m->err_pos < -1) || *ptr == '\n') {
Willy Tarreau801250e2018-09-11 11:45:04 +0200680 state = H1_MSG_HDR_NAME;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200681 goto http_msg_invalid;
682 }
683
Willy Tarreau9aec3052018-09-12 09:20:40 +0200684 if (h1m->err_pos == -1) /* capture the error pointer */
685 h1m->err_pos = ptr - start + skip; /* >= 0 now */
686
687 /* and we still accept this non-token character */
Willy Tarreau801250e2018-09-11 11:45:04 +0200688 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, H1_MSG_HDR_NAME);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200689
Willy Tarreau801250e2018-09-11 11:45:04 +0200690 case H1_MSG_HDR_L1_SP:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200691 http_msg_hdr_l1_sp:
692 /* assumes sol points to the first char */
693 if (likely(HTTP_IS_SPHT(*ptr)))
Willy Tarreau801250e2018-09-11 11:45:04 +0200694 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, H1_MSG_HDR_L1_SP);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200695
696 /* header value can be basically anything except CR/LF */
697 sov = ptr - start;
698
699 if (likely(!HTTP_IS_CRLF(*ptr))) {
700 goto http_msg_hdr_val;
701 }
702
703 if (likely(*ptr == '\r'))
Willy Tarreau801250e2018-09-11 11:45:04 +0200704 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lf, http_msg_ood, state, H1_MSG_HDR_L1_LF);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200705 goto http_msg_hdr_l1_lf;
706
Willy Tarreau801250e2018-09-11 11:45:04 +0200707 case H1_MSG_HDR_L1_LF:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200708 http_msg_hdr_l1_lf:
Willy Tarreau801250e2018-09-11 11:45:04 +0200709 EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_HDR_L1_LF);
710 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lws, http_msg_ood, state, H1_MSG_HDR_L1_LWS);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200711
Willy Tarreau801250e2018-09-11 11:45:04 +0200712 case H1_MSG_HDR_L1_LWS:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200713 http_msg_hdr_l1_lws:
714 if (likely(HTTP_IS_SPHT(*ptr))) {
Christopher Faulet2912f872018-09-19 14:01:04 +0200715 if (!skip_update) {
716 /* replace HT,CR,LF with spaces */
717 for (; start + sov < ptr; sov++)
718 start[sov] = ' ';
719 }
Willy Tarreau794f9af2017-07-26 09:07:47 +0200720 goto http_msg_hdr_l1_sp;
721 }
722 /* we had a header consisting only in spaces ! */
723 eol = sov;
724 goto http_msg_complete_header;
725
Willy Tarreau801250e2018-09-11 11:45:04 +0200726 case H1_MSG_HDR_VAL:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200727 http_msg_hdr_val:
728 /* assumes sol points to the first char, and sov
729 * points to the first character of the value.
730 */
731
732 /* speedup: we'll skip packs of 4 or 8 bytes not containing bytes 0x0D
733 * and lower. In fact since most of the time is spent in the loop, we
734 * also remove the sign bit test so that bytes 0x8e..0x0d break the
735 * loop, but we don't care since they're very rare in header values.
736 */
737#if defined(__x86_64__)
738 while (ptr <= end - sizeof(long)) {
739 if ((*(long *)ptr - 0x0e0e0e0e0e0e0e0eULL) & 0x8080808080808080ULL)
740 goto http_msg_hdr_val2;
741 ptr += sizeof(long);
742 }
743#endif
744#if defined(__x86_64__) || \
745 defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) || \
746 defined(__ARM_ARCH_7A__)
747 while (ptr <= end - sizeof(int)) {
748 if ((*(int*)ptr - 0x0e0e0e0e) & 0x80808080)
749 goto http_msg_hdr_val2;
750 ptr += sizeof(int);
751 }
752#endif
753 if (ptr >= end) {
Willy Tarreau801250e2018-09-11 11:45:04 +0200754 state = H1_MSG_HDR_VAL;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200755 goto http_msg_ood;
756 }
757 http_msg_hdr_val2:
758 if (likely(!HTTP_IS_CRLF(*ptr)))
Willy Tarreau801250e2018-09-11 11:45:04 +0200759 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_val2, http_msg_ood, state, H1_MSG_HDR_VAL);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200760
761 eol = ptr - start;
762 /* Note: we could also copy eol into ->eoh so that we have the
763 * real header end in case it ends with lots of LWS, but is this
764 * really needed ?
765 */
766 if (likely(*ptr == '\r'))
Willy Tarreau801250e2018-09-11 11:45:04 +0200767 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lf, http_msg_ood, state, H1_MSG_HDR_L2_LF);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200768 goto http_msg_hdr_l2_lf;
769
Willy Tarreau801250e2018-09-11 11:45:04 +0200770 case H1_MSG_HDR_L2_LF:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200771 http_msg_hdr_l2_lf:
Willy Tarreau801250e2018-09-11 11:45:04 +0200772 EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_HDR_L2_LF);
773 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lws, http_msg_ood, state, H1_MSG_HDR_L2_LWS);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200774
Willy Tarreau801250e2018-09-11 11:45:04 +0200775 case H1_MSG_HDR_L2_LWS:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200776 http_msg_hdr_l2_lws:
777 if (unlikely(HTTP_IS_SPHT(*ptr))) {
Christopher Faulet2912f872018-09-19 14:01:04 +0200778 if (!skip_update) {
779 /* LWS: replace HT,CR,LF with spaces */
780 for (; start + eol < ptr; eol++)
781 start[eol] = ' ';
782 }
Willy Tarreau794f9af2017-07-26 09:07:47 +0200783 goto http_msg_hdr_val;
784 }
785 http_msg_complete_header:
786 /*
787 * It was a new header, so the last one is finished. Assumes
788 * <sol> points to the first char of the name, <col> to the
789 * colon, <sov> points to the first character of the value and
790 * <eol> to the first CR or LF so we know how the line ends. We
791 * will trim spaces around the value. It's possible to do it by
792 * adjusting <eol> and <sov> which are no more used after this.
793 * We can add the header field to the list.
794 */
Christopher Faulet2912f872018-09-19 14:01:04 +0200795 if (likely(!skip_update)) {
796 while (sov < eol && HTTP_IS_LWS(start[sov]))
797 sov++;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200798
Christopher Faulet2912f872018-09-19 14:01:04 +0200799 while (eol - 1 > sov && HTTP_IS_LWS(start[eol - 1]))
800 eol--;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200801
802
Christopher Faulet2912f872018-09-19 14:01:04 +0200803 n = ist2(start + sol, col - sol);
804 v = ist2(start + sov, eol - sov);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200805
Christopher Faulet2912f872018-09-19 14:01:04 +0200806 do {
807 int ret;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200808
Christopher Faulet2912f872018-09-19 14:01:04 +0200809 if (unlikely(hdr_count >= hdr_num)) {
810 state = H1_MSG_HDR_L2_LWS;
811 goto http_output_full;
812 }
Willy Tarreau5384aac2018-09-11 16:04:48 +0200813
Christopher Faulet2912f872018-09-19 14:01:04 +0200814 if (isteqi(n, ist("transfer-encoding"))) {
815 h1_parse_xfer_enc_header(h1m, v);
816 }
817 else if (isteqi(n, ist("content-length"))) {
818 ret = h1_parse_cont_len_header(h1m, &v);
Willy Tarreau73373ab2018-09-14 17:11:33 +0200819
Christopher Faulet2912f872018-09-19 14:01:04 +0200820 if (ret < 0) {
821 state = H1_MSG_HDR_L2_LWS;
822 goto http_msg_invalid;
823 }
824 else if (ret == 0) {
825 /* skip it */
826 break;
827 }
Willy Tarreau73373ab2018-09-14 17:11:33 +0200828 }
Christopher Faulet2912f872018-09-19 14:01:04 +0200829 else if (isteqi(n, ist("connection"))) {
Christopher Fauleta51ebb72019-03-29 15:03:13 +0100830 h1_parse_connection_header(h1m, &v);
831 if (!v.len) {
832 /* skip it */
833 break;
834 }
Willy Tarreau73373ab2018-09-14 17:11:33 +0200835 }
Christopher Fauletbc7c03e2019-11-27 14:00:51 +0100836 else if (!(h1m->flags & H1_MF_RESP) && isteqi(n, ist("host"))) {
Christopher Faulet531b83e2019-10-11 13:34:22 +0200837 if (host_idx == -1) {
838 struct ist authority;
839
840 authority = http_get_authority(sl.rq.u, 1);
841 if (authority.len && !isteqi(v, authority)) {
842 if (h1m->err_pos < -1) {
843 state = H1_MSG_HDR_L2_LWS;
844 goto http_msg_invalid;
845 }
846 if (h1m->err_pos == -1) /* capture the error pointer */
847 h1m->err_pos = ptr - start + skip; /* >= 0 now */
848 }
Christopher Faulet497ab4f2019-10-11 09:01:44 +0200849 host_idx = hdr_count;
Christopher Faulet531b83e2019-10-11 13:34:22 +0200850 }
Christopher Faulet497ab4f2019-10-11 09:01:44 +0200851 else {
852 if (!isteqi(v, hdr[host_idx].v)) {
853 state = H1_MSG_HDR_L2_LWS;
854 goto http_msg_invalid;
855 }
856 /* if the same host, skip it */
857 break;
858 }
859 }
Willy Tarreau2ea6bb52018-09-14 16:28:15 +0200860
Christopher Faulet2912f872018-09-19 14:01:04 +0200861 http_set_hdr(&hdr[hdr_count++], n, v);
862 } while (0);
863 }
Willy Tarreau794f9af2017-07-26 09:07:47 +0200864
865 sol = ptr - start;
Christopher Faulet2912f872018-09-19 14:01:04 +0200866
Willy Tarreau794f9af2017-07-26 09:07:47 +0200867 if (likely(!HTTP_IS_CRLF(*ptr)))
868 goto http_msg_hdr_name;
869
870 if (likely(*ptr == '\r'))
Willy Tarreau801250e2018-09-11 11:45:04 +0200871 EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, H1_MSG_LAST_LF);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200872 goto http_msg_last_lf;
873
Willy Tarreau801250e2018-09-11 11:45:04 +0200874 case H1_MSG_LAST_LF:
Willy Tarreau794f9af2017-07-26 09:07:47 +0200875 http_msg_last_lf:
Willy Tarreau801250e2018-09-11 11:45:04 +0200876 EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_LAST_LF);
Willy Tarreau794f9af2017-07-26 09:07:47 +0200877 ptr++;
878 /* <ptr> now points to the first byte of payload. If needed sol
879 * still points to the first of either CR or LF of the empty
880 * line ending the headers block.
881 */
Willy Tarreau5384aac2018-09-11 16:04:48 +0200882 if (likely(!skip_update)) {
883 if (unlikely(hdr_count >= hdr_num)) {
884 state = H1_MSG_LAST_LF;
885 goto http_output_full;
886 }
Christopher Fauletff08a922018-09-25 13:59:46 +0200887 http_set_hdr(&hdr[hdr_count++], ist2(start+sol, 0), ist(""));
Willy Tarreau794f9af2017-07-26 09:07:47 +0200888 }
Willy Tarreau001823c2018-09-12 17:25:32 +0200889
890 /* reaching here we've parsed the whole message. We may detect
891 * that we were already continuing an interrupted parsing pass
892 * so we were silently looking for the end of message not
893 * updating anything before deciding to parse it fully at once.
894 * It's guaranteed that we won't match this test twice in a row
895 * since restarting will turn zero.
896 */
897 if (restarting)
898 goto restart;
899
Willy Tarreau2557f6a2018-09-14 16:34:47 +0200900 state = H1_MSG_DATA;
901 if (h1m->flags & H1_MF_XFER_ENC) {
902 if (h1m->flags & H1_MF_CLEN) {
903 h1m->flags &= ~H1_MF_CLEN;
904 hdr_count = http_del_hdr(hdr, ist("content-length"));
905 }
906
907 if (h1m->flags & H1_MF_CHNK)
908 state = H1_MSG_CHUNK_SIZE;
909 else if (!(h1m->flags & H1_MF_RESP)) {
910 /* cf RFC7230#3.3.3 : transfer-encoding in
911 * request without chunked encoding is invalid.
912 */
913 goto http_msg_invalid;
914 }
915 }
916
Willy Tarreau794f9af2017-07-26 09:07:47 +0200917 break;
918
919 default:
920 /* impossible states */
921 goto http_msg_invalid;
922 }
923
Willy Tarreau001823c2018-09-12 17:25:32 +0200924 /* Now we've left the headers state and are either in H1_MSG_DATA or
925 * H1_MSG_CHUNK_SIZE.
Willy Tarreau794f9af2017-07-26 09:07:47 +0200926 */
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200927
Willy Tarreau5384aac2018-09-11 16:04:48 +0200928 if (slp && !skip_update)
Willy Tarreaua41393f2018-09-11 15:34:50 +0200929 *slp = sl;
930
Willy Tarreau4433c082018-09-11 15:33:32 +0200931 h1m->state = state;
932 h1m->next = ptr - start + skip;
933 return h1m->next;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200934
935 http_msg_ood:
936 /* out of data at <ptr> during state <state> */
Willy Tarreau5384aac2018-09-11 16:04:48 +0200937 if (slp && !skip_update)
Willy Tarreaua41393f2018-09-11 15:34:50 +0200938 *slp = sl;
939
Willy Tarreau4433c082018-09-11 15:33:32 +0200940 h1m->state = state;
941 h1m->next = ptr - start + skip;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200942 return 0;
943
944 http_msg_invalid:
945 /* invalid message, error at <ptr> */
Willy Tarreau5384aac2018-09-11 16:04:48 +0200946 if (slp && !skip_update)
Willy Tarreaua41393f2018-09-11 15:34:50 +0200947 *slp = sl;
948
Willy Tarreau4433c082018-09-11 15:33:32 +0200949 h1m->err_state = h1m->state = state;
950 h1m->err_pos = h1m->next = ptr - start + skip;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200951 return -1;
952
953 http_output_full:
954 /* no more room to store the current header, error at <ptr> */
Willy Tarreau5384aac2018-09-11 16:04:48 +0200955 if (slp && !skip_update)
Willy Tarreaua41393f2018-09-11 15:34:50 +0200956 *slp = sl;
957
Willy Tarreau4433c082018-09-11 15:33:32 +0200958 h1m->err_state = h1m->state = state;
959 h1m->err_pos = h1m->next = ptr - start + skip;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200960 return -2;
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200961
962 restart:
Christopher Faulet84f06532019-09-03 16:05:31 +0200963 h1m->flags &= ~(H1_MF_VER_11|H1_MF_CLEN|H1_MF_XFER_ENC|H1_MF_CHNK|H1_MF_CONN_KAL|H1_MF_CONN_CLO|H1_MF_CONN_UPG);
964 h1m->curr_len = h1m->body_len = h1m->next = 0;
Willy Tarreauc2ab9f52018-09-11 17:57:05 +0200965 if (h1m->flags & H1_MF_RESP)
966 h1m->state = H1_MSG_RPBEFORE;
967 else
968 h1m->state = H1_MSG_RQBEFORE;
Willy Tarreau4c34c0e2018-09-11 16:20:30 +0200969 goto try_again;
Willy Tarreau794f9af2017-07-26 09:07:47 +0200970}
971
Willy Tarreau2510f702017-10-31 17:14:16 +0100972/* This function performs a very minimal parsing of the trailers block present
Willy Tarreauf40e6822018-06-14 16:52:02 +0200973 * at offset <ofs> in <buf> for up to <max> bytes, and returns the number of
Willy Tarreau7314be82018-06-14 13:32:50 +0200974 * bytes to delete to skip the trailers. It may return 0 if it's missing some
975 * input data, or < 0 in case of parse error (in which case the caller may have
976 * to decide how to proceed, possibly eating everything).
Willy Tarreau2510f702017-10-31 17:14:16 +0100977 */
Willy Tarreauf40e6822018-06-14 16:52:02 +0200978int h1_measure_trailers(const struct buffer *buf, unsigned int ofs, unsigned int max)
Willy Tarreau2510f702017-10-31 17:14:16 +0100979{
Willy Tarreauf40e6822018-06-14 16:52:02 +0200980 const char *stop = b_peek(buf, ofs + max);
981 int count = ofs;
Willy Tarreau2510f702017-10-31 17:14:16 +0100982
983 while (1) {
984 const char *p1 = NULL, *p2 = NULL;
Willy Tarreau7314be82018-06-14 13:32:50 +0200985 const char *start = b_peek(buf, count);
Willy Tarreau2510f702017-10-31 17:14:16 +0100986 const char *ptr = start;
Willy Tarreau2510f702017-10-31 17:14:16 +0100987
988 /* scan current line and stop at LF or CRLF */
989 while (1) {
990 if (ptr == stop)
991 return 0;
992
993 if (*ptr == '\n') {
994 if (!p1)
995 p1 = ptr;
996 p2 = ptr;
997 break;
998 }
999
1000 if (*ptr == '\r') {
1001 if (p1)
1002 return -1;
1003 p1 = ptr;
1004 }
1005
Willy Tarreau7314be82018-06-14 13:32:50 +02001006 ptr = b_next(buf, ptr);
Willy Tarreau2510f702017-10-31 17:14:16 +01001007 }
1008
1009 /* after LF; point to beginning of next line */
Willy Tarreau7314be82018-06-14 13:32:50 +02001010 p2 = b_next(buf, p2);
1011 count += b_dist(buf, start, p2);
Willy Tarreau2510f702017-10-31 17:14:16 +01001012
1013 /* LF/CRLF at beginning of line => end of trailers at p2.
1014 * Everything was scheduled for forwarding, there's nothing left
1015 * from this message. */
1016 if (p1 == start)
1017 break;
1018 /* OK, next line then */
1019 }
Willy Tarreauf40e6822018-06-14 16:52:02 +02001020 return count - ofs;
Willy Tarreau2510f702017-10-31 17:14:16 +01001021}