Line | Count | Source |
1 | | /* |
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 | | |
13 | | #include <ctype.h> |
14 | | |
15 | | #include <import/sha1.h> |
16 | | |
17 | | #include <haproxy/api.h> |
18 | | #include <haproxy/base64.h> |
19 | | #include <haproxy/cfgparse.h> |
20 | | #include <haproxy/h1.h> |
21 | | #include <haproxy/http-hdr.h> |
22 | | #include <haproxy/tools.h> |
23 | | |
24 | | /* by default, RFC9112#6.1 applies, t-e combined with c-l represents a risk of |
25 | | * smuggling if it crosses another 1.0 agent so we must close at the end of the |
26 | | * transaction. But it may cause difficulties to some very old broken devices. |
27 | | */ |
28 | | int h1_do_not_close_on_insecure_t_e = 0; |
29 | | |
30 | | /* Parse the Transfer-Encoding: header field of an HTTP/1 request, looking for |
31 | | * "chunked" encoding to perform some checks (it must be the last encoding for |
32 | | * the request and must not be performed twice for any message). The |
33 | | * H1_MF_TE_CHUNKED is set if a valid "chunked" encoding is found. The |
34 | | * H1_MF_TE_OTHER flag is set if any other encoding is found. The H1_MF_XFER_ENC |
35 | | * flag is always set. The H1_MF_CHNK is set when "chunked" encoding is the last |
36 | | * one. Note that transfer codings are case-insensitive (cf RFC7230#4). This |
37 | | * function returns -2 for a fatal error, -1 for an error that may be hiidden by |
38 | | * config, 0 if the whole header can be dropped (not used yet), or >0 if the |
39 | | * value can be indexed. |
40 | | */ |
41 | | int h1_parse_xfer_enc_header(struct h1m *h1m, struct ist value) |
42 | 2.60k | { |
43 | 2.60k | char *e, *n; |
44 | 2.60k | struct ist word; |
45 | 2.60k | int ret = 1; |
46 | | |
47 | | /* Reject empty header */ |
48 | 2.60k | if (istptr(value) == istend(value)) { |
49 | 11 | ret = -1; |
50 | 11 | goto end; |
51 | 11 | } |
52 | | |
53 | 2.59k | h1m->flags |= H1_MF_XFER_ENC; |
54 | | |
55 | 2.59k | word.ptr = value.ptr - 1; // -1 for next loop's pre-increment |
56 | 2.59k | e = istend(value); |
57 | | |
58 | 7.17k | while (++word.ptr < e) { |
59 | | /* skip leading delimiter and blanks */ |
60 | 4.58k | if (HTTP_IS_LWS(*word.ptr)) |
61 | 503 | continue; |
62 | | |
63 | 4.08k | n = http_find_hdr_value_end(word.ptr, e); // next comma or end of line |
64 | | |
65 | | /* a comma at the end means the last value is empty */ |
66 | 4.08k | if (n+1 == e) |
67 | 19 | ret = -1; |
68 | 4.08k | word.len = n - word.ptr; |
69 | | |
70 | | /* trim trailing blanks */ |
71 | 4.99k | while (word.len && HTTP_IS_LWS(word.ptr[word.len-1])) |
72 | 911 | word.len--; |
73 | | |
74 | 4.08k | h1m->flags &= ~H1_MF_CHNK; |
75 | | |
76 | | /* empty values are forbidden */ |
77 | 4.08k | if (!word.len) |
78 | 389 | ret = -1; |
79 | 3.69k | else if (isteqi(word, ist("chunked"))) { |
80 | 1.44k | if (h1m->flags & H1_MF_TE_CHUNKED) { |
81 | | /* cf RFC7230#3.3.1 : A sender MUST NOT apply |
82 | | * chunked more than once to a message body |
83 | | * (i.e., chunking an already chunked message is |
84 | | * not allowed) |
85 | | */ |
86 | 291 | ret = -1; |
87 | 291 | } |
88 | 1.44k | h1m->flags |= (H1_MF_TE_CHUNKED|H1_MF_CHNK); |
89 | 1.44k | } |
90 | 2.25k | else { |
91 | 2.25k | if ((h1m->flags & (H1_MF_RESP|H1_MF_TE_CHUNKED)) == H1_MF_TE_CHUNKED) { |
92 | | /* cf RFC7230#3.3.1 : If any transfer coding |
93 | | * other than chunked is applied to a request |
94 | | * payload body, the sender MUST apply chunked |
95 | | * as the final transfer coding to ensure that |
96 | | * the message is properly framed. |
97 | | */ |
98 | 1 | ret = -2; |
99 | 1 | goto end; |
100 | 1 | } |
101 | 2.25k | h1m->flags |= H1_MF_TE_OTHER; |
102 | 2.25k | } |
103 | | |
104 | 4.08k | word.ptr = n; |
105 | 4.08k | } |
106 | | |
107 | 2.60k | end: |
108 | 2.60k | return ret; |
109 | 2.59k | } |
110 | | |
111 | | /* Validate the authority and the host header value for CONNECT method. If there |
112 | | * is hast header, its value is normalized. 0 is returned on success, -1 if the |
113 | | * authority is invalid and -2 if the host is invalid. |
114 | | */ |
115 | | static int h1_validate_connect_authority(struct ist scheme, struct ist authority, struct ist *host_hdr) |
116 | 287 | { |
117 | 287 | struct ist uri_host, uri_port, host, host_port; |
118 | | |
119 | 287 | if (isttest(scheme) || !isttest(authority)) |
120 | 0 | goto invalid_authority; |
121 | 287 | uri_host = authority; |
122 | 287 | uri_port = http_get_host_port(authority); |
123 | 287 | if (!istlen(uri_port)) |
124 | 30 | goto invalid_authority; |
125 | 257 | uri_host.len -= (istlen(uri_port) + 1); |
126 | | |
127 | 257 | if (!host_hdr || !isttest(*host_hdr)) |
128 | 15 | goto end; |
129 | | |
130 | | /* Get the port of the host header value, if any */ |
131 | 242 | host = *host_hdr; |
132 | 242 | host_port = http_get_host_port(*host_hdr); |
133 | 242 | if (isttest(host_port)) |
134 | 132 | host.len -= (istlen(host_port) + 1); |
135 | | |
136 | 242 | if (istlen(host_port)) { |
137 | 121 | if (!isteqi(host, uri_host) || !isteq(host_port, uri_port)) |
138 | 108 | goto invalid_host; |
139 | 13 | if (http_is_default_port(IST_NULL, uri_port)) |
140 | 2 | *host_hdr = host; /* normalize */ |
141 | 13 | } |
142 | 121 | else { |
143 | 121 | if (!http_is_default_port(IST_NULL, uri_port) || !isteqi(host, uri_host)) |
144 | 105 | goto invalid_host; |
145 | 121 | } |
146 | | |
147 | 44 | end: |
148 | 44 | return 0; |
149 | | |
150 | 30 | invalid_authority: |
151 | 30 | return -1; |
152 | | |
153 | 213 | invalid_host: |
154 | 213 | return -2; |
155 | 242 | } |
156 | | |
157 | | |
158 | | /* Validate the authority and the host header value for non-CONNECT method, when |
159 | | * an absolute-URI is detected but when it does not exactly match the host |
160 | | * value. The idea is to detect default port (http or https). authority and host |
161 | | * are defined here. 0 is returned on success, -1 if the host is does not match |
162 | | * the authority. |
163 | | */ |
164 | | static int h1_validate_mismatch_authority(struct ist scheme, struct ist authority, struct ist host_hdr) |
165 | 258 | { |
166 | 258 | struct ist uri_host, uri_port, host, host_port; |
167 | | |
168 | 258 | if (!isttest(scheme)) |
169 | 0 | goto mismatch; |
170 | | |
171 | 258 | uri_host = authority; |
172 | 258 | uri_port = http_get_host_port(authority); |
173 | 258 | if (isttest(uri_port)) |
174 | 128 | uri_host.len -= (istlen(uri_port) + 1); |
175 | | |
176 | 258 | host = host_hdr; |
177 | 258 | host_port = http_get_host_port(host_hdr); |
178 | 258 | if (isttest(host_port)) |
179 | 133 | host.len -= (istlen(host_port) + 1); |
180 | | |
181 | 258 | if (!isttest(uri_port) && !isttest(host_port)) { |
182 | | /* No port on both: we already know the authority does not match |
183 | | * the host value |
184 | | */ |
185 | 14 | goto mismatch; |
186 | 14 | } |
187 | 244 | else if (isttest(uri_port) && !http_is_default_port(scheme, uri_port)) { |
188 | | /* here there is no port for the host value and the port for the |
189 | | * authority is not the default one |
190 | | */ |
191 | 52 | goto mismatch; |
192 | 52 | } |
193 | 192 | else if (isttest(host_port) && !http_is_default_port(scheme, host_port)) { |
194 | | /* here there is no port for the authority and the port for the |
195 | | * host value is not the default one |
196 | | */ |
197 | 38 | goto mismatch; |
198 | 38 | } |
199 | 154 | else { |
200 | | /* the authority or the host value contain a default port and |
201 | | * there is no port on the other value |
202 | | */ |
203 | 154 | if (!isteqi(uri_host, host)) |
204 | 83 | goto mismatch; |
205 | 154 | } |
206 | | |
207 | 71 | return 0; |
208 | | |
209 | 187 | mismatch: |
210 | 187 | return -1; |
211 | 258 | } |
212 | | |
213 | | |
214 | | /* Parse the Connection: header of an HTTP/1 request, looking for "close", |
215 | | * "keep-alive", and "upgrade" values, and updating h1m->flags according to |
216 | | * what was found there. Note that flags are only added, not removed, so the |
217 | | * function is safe for being called multiple times if multiple occurrences |
218 | | * are found. If the flag H1_MF_CLEAN_CONN_HDR, the header value is cleaned |
219 | | * up from "keep-alive" and "close" values. To do so, the header value is |
220 | | * rewritten in place and its length is updated. |
221 | | */ |
222 | | void h1_parse_connection_header(struct h1m *h1m, struct ist *value) |
223 | 2.27k | { |
224 | 2.27k | char *e, *n, *p; |
225 | 2.27k | struct ist word; |
226 | | |
227 | 2.27k | word.ptr = value->ptr - 1; // -1 for next loop's pre-increment |
228 | 2.27k | p = value->ptr; |
229 | 2.27k | e = value->ptr + value->len; |
230 | 2.27k | if (h1m->flags & H1_MF_CLEAN_CONN_HDR) |
231 | 0 | value->len = 0; |
232 | | |
233 | 29.6k | while (++word.ptr < e) { |
234 | | /* skip leading delimiter and blanks */ |
235 | 27.3k | if (HTTP_IS_LWS(*word.ptr)) |
236 | 883 | continue; |
237 | | |
238 | 26.4k | n = http_find_hdr_value_end(word.ptr, e); // next comma or end of line |
239 | 26.4k | word.len = n - word.ptr; |
240 | | |
241 | | /* trim trailing blanks */ |
242 | 27.2k | while (word.len && HTTP_IS_LWS(word.ptr[word.len-1])) |
243 | 714 | word.len--; |
244 | | |
245 | 26.4k | if (isteqi(word, ist("keep-alive"))) { |
246 | 759 | h1m->flags |= H1_MF_CONN_KAL; |
247 | 759 | if (h1m->flags & H1_MF_CLEAN_CONN_HDR) |
248 | 0 | goto skip_val; |
249 | 759 | } |
250 | 25.7k | else if (isteqi(word, ist("close"))) { |
251 | 298 | h1m->flags |= H1_MF_CONN_CLO; |
252 | 298 | if (h1m->flags & H1_MF_CLEAN_CONN_HDR) |
253 | 0 | goto skip_val; |
254 | 298 | } |
255 | 25.4k | else if (isteqi(word, ist("upgrade"))) |
256 | 4.70k | h1m->flags |= H1_MF_CONN_UPG; |
257 | | |
258 | 26.4k | if (h1m->flags & H1_MF_CLEAN_CONN_HDR) { |
259 | 0 | if (value->ptr + value->len == p) { |
260 | | /* no rewrite done till now */ |
261 | 0 | value->len = n - value->ptr; |
262 | 0 | } |
263 | 0 | else { |
264 | 0 | if (value->len) |
265 | 0 | value->ptr[value->len++] = ','; |
266 | 0 | istcat(value, word, e - value->ptr); |
267 | 0 | } |
268 | 0 | } |
269 | | |
270 | 26.4k | skip_val: |
271 | 26.4k | word.ptr = p = n; |
272 | 26.4k | } |
273 | 2.27k | } |
274 | | |
275 | | /* Parse the Upgrade: header of an HTTP/1 request. |
276 | | * If "websocket" is found, set H1_MF_UPG_WEBSOCKET flag |
277 | | * If "h2c" or "h2" found, the value is skipped. |
278 | | */ |
279 | | void h1_parse_upgrade_header(struct h1m *h1m, struct ist *value) |
280 | 1.37k | { |
281 | 1.37k | char *e, *n, *p; |
282 | 1.37k | struct ist word; |
283 | | |
284 | 1.37k | word.ptr = value->ptr - 1; // -1 for next loop's pre-increment |
285 | 1.37k | p = value->ptr; |
286 | 1.37k | e = value->ptr + value->len; |
287 | 1.37k | value->len = 0; |
288 | | |
289 | 24.3k | while (++word.ptr < e) { |
290 | | /* skip leading delimiter and blanks */ |
291 | 22.9k | if (HTTP_IS_LWS(*word.ptr)) |
292 | 439 | continue; |
293 | | |
294 | 22.5k | n = http_find_hdr_value_end(word.ptr, e); // next comma or end of line |
295 | 22.5k | word.len = n - word.ptr; |
296 | | |
297 | | /* trim trailing blanks */ |
298 | 23.2k | while (word.len && HTTP_IS_LWS(word.ptr[word.len-1])) |
299 | 732 | word.len--; |
300 | | |
301 | 22.5k | if (isteqi(word, ist("websocket"))) |
302 | 0 | h1m->flags |= H1_MF_UPG_WEBSOCKET; |
303 | 22.5k | else if (isteqi(word, ist("h2c")) || isteqi(word, ist("h2"))) |
304 | 1.98k | goto skip_val; |
305 | | |
306 | 20.5k | if (value->ptr + value->len == p) { |
307 | | /* no rewrite done till now */ |
308 | 9.69k | value->len = n - value->ptr; |
309 | 9.69k | } |
310 | 10.8k | else { |
311 | 10.8k | if (value->len) |
312 | 10.3k | value->ptr[value->len++] = ','; |
313 | 10.8k | istcat(value, word, e - value->ptr); |
314 | 10.8k | } |
315 | | |
316 | 22.5k | skip_val: |
317 | 22.5k | word.ptr = p = n; |
318 | 22.5k | } |
319 | | |
320 | 1.37k | if (istlen(*value)) |
321 | 1.07k | h1m->flags |= H1_MF_UPG_HDR; |
322 | 1.37k | } |
323 | | |
324 | | /* Macros used in the HTTP/1 parser, to check for the expected presence of |
325 | | * certain bytes (ef: LF) or to skip to next byte and yield in case of failure. |
326 | | */ |
327 | | |
328 | | /* Expects to find an LF at <ptr>. If not, set <state> to <where> and jump to |
329 | | * <bad>. |
330 | | */ |
331 | | #define EXPECT_LF_HERE(ptr, bad, state, where) \ |
332 | 33.5k | do { \ |
333 | 33.5k | if (unlikely(*(ptr) != '\n')) { \ |
334 | 87 | state = (where); \ |
335 | 87 | goto bad; \ |
336 | 87 | } \ |
337 | 33.5k | } while (0) |
338 | | |
339 | | /* Increments pointer <ptr>, continues to label <more> if it's still below |
340 | | * pointer <end>, or goes to <stop> and sets <state> to <where> if the end |
341 | | * of buffer was reached. |
342 | | */ |
343 | | #define EAT_AND_JUMP_OR_RETURN(ptr, end, more, stop, state, where) \ |
344 | 52.2M | do { \ |
345 | 52.2M | if (likely(++(ptr) < (end))) \ |
346 | 52.2M | goto more; \ |
347 | 52.2M | else { \ |
348 | 799 | state = (where); \ |
349 | 799 | goto stop; \ |
350 | 799 | } \ |
351 | 52.2M | } while (0) |
352 | | |
353 | | /* This function parses a contiguous HTTP/1 headers block starting at <start> |
354 | | * and ending before <stop>, at once, and converts it a list of (name,value) |
355 | | * pairs representing header fields into the array <hdr> of size <hdr_num>, |
356 | | * whose last entry will have an empty name and an empty value. If <hdr_num> is |
357 | | * too small to represent the whole message, an error is returned. Some |
358 | | * protocol elements such as content-length and transfer-encoding will be |
359 | | * parsed and stored into h1m as well. <hdr> may be null, in which case only |
360 | | * the parsing state will be updated. This may be used to restart the parsing |
361 | | * where it stopped for example. |
362 | | * |
363 | | * For now it's limited to the response. If the header block is incomplete, |
364 | | * 0 is returned, waiting to be called again with more data to try it again. |
365 | | * The caller is responsible for initializing h1m->state to H1_MSG_RPBEFORE, |
366 | | * and h1m->next to zero on the first call, the parser will do the rest. If |
367 | | * an incomplete message is seen, the caller only needs to present h1m->state |
368 | | * and h1m->next again, with an empty header list so that the parser can start |
369 | | * again. In this case, it will detect that it interrupted a previous session |
370 | | * and will first look for the end of the message before reparsing it again and |
371 | | * indexing it at the same time. This ensures that incomplete messages fed 1 |
372 | | * character at a time are never processed entirely more than exactly twice, |
373 | | * and that there is no need to store all the internal state and pre-parsed |
374 | | * headers or start line between calls. |
375 | | * |
376 | | * A pointer to a start line descriptor may be passed in <slp>, in which case |
377 | | * the parser will fill it with whatever it found. |
378 | | * |
379 | | * The code derived from the main HTTP/1 parser above but was simplified and |
380 | | * optimized to process responses produced or forwarded by haproxy. The caller |
381 | | * is responsible for ensuring that the message doesn't wrap, and should ensure |
382 | | * it is complete to avoid having to retry the operation after a failed |
383 | | * attempt. The message is not supposed to be invalid, which is why a few |
384 | | * properties such as the character set used in the header field names are not |
385 | | * checked. In case of an unparsable response message, a negative value will be |
386 | | * returned with h1m->err_pos and h1m->err_state matching the location and |
387 | | * state where the error was met. Leading blank likes are tolerated but not |
388 | | * recommended. If flag H1_MF_HDRS_ONLY is set in h1m->flags, only headers are |
389 | | * parsed and the start line is skipped. It is not required to set h1m->state |
390 | | * nor h1m->next in this case. |
391 | | * |
392 | | * This function returns : |
393 | | * -1 in case of error. In this case, h1m->err_state is filled (if h1m is |
394 | | * set) with the state the error occurred in and h1m->err_pos with the |
395 | | * the position relative to <start> |
396 | | * -2 if the output is full (hdr_num reached). err_state and err_pos also |
397 | | * indicate where it failed. |
398 | | * 0 in case of missing data. |
399 | | * > 0 on success, it then corresponds to the number of bytes read since |
400 | | * <start> so that the caller can go on with the payload. |
401 | | */ |
402 | | int h1_headers_to_hdr_list(char *start, const char *stop, |
403 | | struct http_hdr *hdr, unsigned int hdr_num, |
404 | | struct h1m *h1m, union h1_sl *slp) |
405 | 5.73k | { |
406 | 5.73k | enum h1m_state state; |
407 | 5.73k | register char *ptr; |
408 | 5.73k | register const char *end; |
409 | 5.73k | unsigned int hdr_count; |
410 | 5.73k | unsigned int skip; /* number of bytes skipped at the beginning */ |
411 | 5.73k | unsigned int sol; /* start of line */ |
412 | 5.73k | unsigned int col; /* position of the colon */ |
413 | 5.73k | unsigned int eol; /* end of line */ |
414 | 5.73k | unsigned int sov; /* start of value */ |
415 | 5.73k | union h1_sl sl; |
416 | 5.73k | int skip_update; |
417 | 5.73k | int restarting; |
418 | 5.73k | int host_idx; |
419 | 5.73k | struct ist n, v; /* header name and value during parsing */ |
420 | | |
421 | 5.73k | skip = 0; // do it only once to keep track of the leading CRLF. |
422 | | |
423 | 5.73k | try_again: |
424 | 5.73k | hdr_count = sol = col = eol = sov = 0; |
425 | 5.73k | sl.st.status = 0; |
426 | 5.73k | skip_update = restarting = 0; |
427 | 5.73k | host_idx = -1; |
428 | | |
429 | 5.73k | if (h1m->flags & H1_MF_HDRS_ONLY) { |
430 | 80 | state = H1_MSG_HDR_FIRST; |
431 | 80 | h1m->next = 0; |
432 | 80 | } |
433 | 5.65k | else { |
434 | 5.65k | state = h1m->state; |
435 | 5.65k | if (h1m->state != H1_MSG_RQBEFORE && h1m->state != H1_MSG_RPBEFORE) |
436 | 0 | restarting = 1; |
437 | 5.65k | } |
438 | | |
439 | 5.73k | ptr = start + h1m->next; |
440 | 5.73k | end = stop; |
441 | | |
442 | 5.73k | if (unlikely(ptr >= end)) |
443 | 0 | goto http_msg_ood; |
444 | | |
445 | | /* don't update output if hdr is NULL or if we're restarting */ |
446 | 5.73k | if (!hdr || restarting) |
447 | 0 | skip_update = 1; |
448 | | |
449 | 5.73k | switch (state) { |
450 | 2.56k | case H1_MSG_RQBEFORE: |
451 | 3.14k | http_msg_rqbefore: |
452 | 3.14k | if (likely(HTTP_IS_TOKEN(*ptr))) { |
453 | | /* we have a start of message, we may have skipped some |
454 | | * heading CRLF. Skip them now. |
455 | | */ |
456 | 2.52k | skip += ptr - start; |
457 | 2.52k | start = ptr; |
458 | | |
459 | 2.52k | sol = 0; |
460 | 2.52k | sl.rq.m.ptr = ptr; |
461 | 2.52k | hdr_count = 0; |
462 | 2.52k | state = H1_MSG_RQMETH; |
463 | 2.52k | goto http_msg_rqmeth; |
464 | 2.52k | } |
465 | | |
466 | 618 | if (unlikely(!HTTP_IS_CRLF(*ptr))) { |
467 | 8 | state = H1_MSG_RQBEFORE; |
468 | 8 | goto http_msg_invalid; |
469 | 8 | } |
470 | | |
471 | 610 | if (unlikely(*ptr == '\n')) |
472 | 198 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore, http_msg_ood, state, H1_MSG_RQBEFORE); |
473 | 412 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore_cr, http_msg_ood, state, H1_MSG_RQBEFORE_CR); |
474 | | /* stop here */ |
475 | | |
476 | 0 | case H1_MSG_RQBEFORE_CR: |
477 | 404 | http_msg_rqbefore_cr: |
478 | 404 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RQBEFORE_CR); |
479 | 390 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore, http_msg_ood, state, H1_MSG_RQBEFORE); |
480 | | /* stop here */ |
481 | | |
482 | 0 | case H1_MSG_RQMETH: |
483 | 11.7M | http_msg_rqmeth: |
484 | 11.7M | if (likely(HTTP_IS_TOKEN(*ptr))) |
485 | 11.7M | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth, http_msg_ood, state, H1_MSG_RQMETH); |
486 | | |
487 | 2.52k | if (likely(HTTP_IS_SPHT(*ptr))) { |
488 | 2.48k | sl.rq.m.len = ptr - sl.rq.m.ptr; |
489 | 2.48k | sl.rq.meth = find_http_meth(start, sl.rq.m.len); |
490 | 2.48k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth_sp, http_msg_ood, state, H1_MSG_RQMETH_SP); |
491 | 2.48k | } |
492 | | |
493 | 39 | if (likely(HTTP_IS_CRLF(*ptr))) { |
494 | | /* HTTP 0.9 request */ |
495 | 36 | sl.rq.m.len = ptr - sl.rq.m.ptr; |
496 | 36 | sl.rq.meth = find_http_meth(sl.rq.m.ptr, sl.rq.m.len); |
497 | 41 | http_msg_req09_uri: |
498 | 41 | sl.rq.u.ptr = ptr; |
499 | 257 | http_msg_req09_uri_e: |
500 | 257 | sl.rq.u.len = ptr - sl.rq.u.ptr; |
501 | 263 | http_msg_req09_ver: |
502 | 263 | sl.rq.v = ist2(ptr, 0); |
503 | 263 | goto http_msg_rqline_eol; |
504 | 257 | } |
505 | 3 | state = H1_MSG_RQMETH; |
506 | 3 | goto http_msg_invalid; |
507 | | |
508 | 0 | case H1_MSG_RQMETH_SP: |
509 | 2.69k | http_msg_rqmeth_sp: |
510 | 2.69k | if (likely(!HTTP_IS_LWS(*ptr))) { |
511 | 2.43k | sl.rq.u.ptr = ptr; |
512 | 2.43k | goto http_msg_rquri; |
513 | 2.43k | } |
514 | 255 | if (likely(HTTP_IS_SPHT(*ptr))) |
515 | 250 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth_sp, http_msg_ood, state, H1_MSG_RQMETH_SP); |
516 | | /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
517 | 5 | goto http_msg_req09_uri; |
518 | | |
519 | 5 | case H1_MSG_RQURI: |
520 | 2.43k | http_msg_rquri: |
521 | 2.43k | #ifdef HA_UNALIGNED_LE |
522 | | /* speedup: skip bytes not between 0x24 and 0x7e inclusive */ |
523 | 3.98M | while (ptr <= end - sizeof(int)) { |
524 | 3.98M | if (is_char4_outside(*(uint *)ptr, 0x24, 0x7e)) |
525 | 2.23k | break; |
526 | | |
527 | 3.98M | ptr += sizeof(int); |
528 | 3.98M | } |
529 | 2.43k | #endif |
530 | 2.43k | if (ptr >= end) { |
531 | 1 | state = H1_MSG_RQURI; |
532 | 1 | goto http_msg_ood; |
533 | 1 | } |
534 | 804k | http_msg_rquri2: |
535 | 804k | if (likely((unsigned char)(*ptr - 33) <= 93)) { /* 33 to 126 included */ |
536 | 802k | if (*ptr == '#') { |
537 | 1 | if (h1m->err_pos < -1) /* PR_O2_REQBUG_OK not set */ |
538 | 1 | goto invalid_char; |
539 | 0 | if (h1m->err_pos == -1) /* PR_O2_REQBUG_OK set: just log */ |
540 | 0 | h1m->err_pos = ptr - start + skip; |
541 | 0 | } |
542 | 802k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri2, http_msg_ood, state, H1_MSG_RQURI); |
543 | 802k | } |
544 | | |
545 | 2.41k | if (likely(HTTP_IS_SPHT(*ptr))) { |
546 | 2.17k | sl.rq.u.len = ptr - sl.rq.u.ptr; |
547 | 2.17k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri_sp, http_msg_ood, state, H1_MSG_RQURI_SP); |
548 | 2.17k | } |
549 | 235 | if (likely((unsigned char)*ptr >= 128)) { |
550 | | /* non-ASCII chars are forbidden unless option |
551 | | * accept-unsafe-violations-in-http-request is enabled in the frontend. |
552 | | * In any case, we capture the faulty char. |
553 | | */ |
554 | 12 | if (h1m->err_pos < -1) |
555 | 12 | goto invalid_char; |
556 | 0 | if (h1m->err_pos == -1) |
557 | 0 | h1m->err_pos = ptr - start + skip; |
558 | 0 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri, http_msg_ood, state, H1_MSG_RQURI); |
559 | 0 | } |
560 | | |
561 | 223 | if (likely(HTTP_IS_CRLF(*ptr))) { |
562 | | /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
563 | 216 | goto http_msg_req09_uri_e; |
564 | 216 | } |
565 | | |
566 | | /* OK forbidden chars, 0..31 or 127 */ |
567 | 20 | invalid_char: |
568 | 20 | state = H1_MSG_RQURI; |
569 | 20 | goto http_msg_invalid; |
570 | | |
571 | 0 | case H1_MSG_RQURI_SP: |
572 | 2.40k | http_msg_rquri_sp: |
573 | 2.40k | if (likely(!HTTP_IS_LWS(*ptr))) { |
574 | 2.16k | sl.rq.v.ptr = ptr; |
575 | 2.16k | goto http_msg_rqver; |
576 | 2.16k | } |
577 | 238 | if (likely(HTTP_IS_SPHT(*ptr))) |
578 | 232 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri_sp, http_msg_ood, state, H1_MSG_RQURI_SP); |
579 | | /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
580 | 6 | goto http_msg_req09_ver; |
581 | | |
582 | | |
583 | 6 | case H1_MSG_RQVER: |
584 | 2.09M | http_msg_rqver: |
585 | 2.09M | if (likely(HTTP_IS_VER_TOKEN(*ptr))) |
586 | 2.08M | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqver, http_msg_ood, state, H1_MSG_RQVER); |
587 | | |
588 | 2.16k | if (likely(HTTP_IS_CRLF(*ptr))) { |
589 | 2.15k | sl.rq.v.len = ptr - sl.rq.v.ptr; |
590 | 2.41k | http_msg_rqline_eol: |
591 | | /* We have seen the end of line. Note that we do not |
592 | | * necessarily have the \n yet, but at least we know that we |
593 | | * have EITHER \r OR \n, otherwise the request would not be |
594 | | * complete. We can then record the request length and return |
595 | | * to the caller which will be able to register it. |
596 | | */ |
597 | | |
598 | 2.41k | if (likely(!skip_update)) { |
599 | 2.41k | if ((sl.rq.v.len == 8) && |
600 | 1.38k | (*(sl.rq.v.ptr + 5) > '1' || |
601 | 903 | (*(sl.rq.v.ptr + 5) == '1' && *(sl.rq.v.ptr + 7) >= '1'))) |
602 | 767 | h1m->flags |= H1_MF_VER_11; |
603 | | |
604 | 2.41k | if (unlikely(hdr_count >= hdr_num)) { |
605 | 0 | state = H1_MSG_RQVER; |
606 | 0 | goto http_output_full; |
607 | 0 | } |
608 | 2.41k | if (!(h1m->flags & H1_MF_NO_PHDR)) |
609 | 2.41k | http_set_hdr(&hdr[hdr_count++], ist(":method"), sl.rq.m); |
610 | | |
611 | 2.41k | if (unlikely(hdr_count >= hdr_num)) { |
612 | 0 | state = H1_MSG_RQVER; |
613 | 0 | goto http_output_full; |
614 | 0 | } |
615 | 2.41k | if (!(h1m->flags & H1_MF_NO_PHDR)) |
616 | 2.41k | http_set_hdr(&hdr[hdr_count++], ist(":path"), sl.rq.u); |
617 | 2.41k | } |
618 | | |
619 | 2.41k | sol = ptr - start; |
620 | 2.41k | if (likely(*ptr == '\r')) |
621 | 315 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqline_end, http_msg_ood, state, H1_MSG_RQLINE_END); |
622 | 2.10k | goto http_msg_rqline_end; |
623 | 2.41k | } |
624 | | |
625 | | /* neither an HTTP_VER token nor a CRLF */ |
626 | 5 | state = H1_MSG_RQVER; |
627 | 5 | goto http_msg_invalid; |
628 | | |
629 | 0 | case H1_MSG_RQLINE_END: |
630 | 2.40k | http_msg_rqline_end: |
631 | | /* check for HTTP/0.9 request : no version information |
632 | | * available. sol must point to the first of CR or LF. However |
633 | | * since we don't save these elements between calls, if we come |
634 | | * here from a restart, we don't necessarily know. Thus in this |
635 | | * case we simply start over. |
636 | | */ |
637 | 2.40k | if (restarting) |
638 | 0 | goto restart; |
639 | | |
640 | 2.40k | if (unlikely(sl.rq.v.len == 0)) |
641 | 255 | goto http_msg_last_lf; |
642 | | |
643 | 2.15k | EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RQLINE_END); |
644 | 2.14k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_first, http_msg_ood, state, H1_MSG_HDR_FIRST); |
645 | | /* stop here */ |
646 | | |
647 | | /* |
648 | | * Common states below |
649 | | */ |
650 | 3.09k | case H1_MSG_RPBEFORE: |
651 | 3.66k | http_msg_rpbefore: |
652 | 3.66k | if (likely(HTTP_IS_TOKEN(*ptr))) { |
653 | | /* we have a start of message, we may have skipped some |
654 | | * heading CRLF. Skip them now. |
655 | | */ |
656 | 3.06k | skip += ptr - start; |
657 | 3.06k | start = ptr; |
658 | | |
659 | 3.06k | sol = 0; |
660 | 3.06k | sl.st.v.ptr = ptr; |
661 | 3.06k | hdr_count = 0; |
662 | 3.06k | state = H1_MSG_RPVER; |
663 | 3.06k | goto http_msg_rpver; |
664 | 3.06k | } |
665 | | |
666 | 607 | if (unlikely(!HTTP_IS_CRLF(*ptr))) { |
667 | 3 | state = H1_MSG_RPBEFORE; |
668 | 3 | goto http_msg_invalid; |
669 | 3 | } |
670 | | |
671 | 604 | if (unlikely(*ptr == '\n')) |
672 | 201 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, H1_MSG_RPBEFORE); |
673 | 403 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore_cr, http_msg_ood, state, H1_MSG_RPBEFORE_CR); |
674 | | /* stop here */ |
675 | | |
676 | 0 | case H1_MSG_RPBEFORE_CR: |
677 | 398 | http_msg_rpbefore_cr: |
678 | 398 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RPBEFORE_CR); |
679 | 383 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, H1_MSG_RPBEFORE); |
680 | | /* stop here */ |
681 | | |
682 | 0 | case H1_MSG_RPVER: |
683 | 1.61M | http_msg_rpver: |
684 | 1.61M | if (likely(HTTP_IS_VER_TOKEN(*ptr))) |
685 | 1.61M | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver, http_msg_ood, state, H1_MSG_RPVER); |
686 | | |
687 | 3.05k | if (likely(HTTP_IS_SPHT(*ptr))) { |
688 | 3.04k | sl.st.v.len = ptr - sl.st.v.ptr; |
689 | | |
690 | 3.04k | if ((sl.st.v.len == 8) && |
691 | 1.57k | (*(sl.st.v.ptr + 5) > '1' || |
692 | 1.01k | (*(sl.st.v.ptr + 5) == '1' && *(sl.st.v.ptr + 7) >= '1'))) |
693 | 890 | h1m->flags |= H1_MF_VER_11; |
694 | | |
695 | 3.04k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, H1_MSG_RPVER_SP); |
696 | 3.04k | } |
697 | 9 | state = H1_MSG_RPVER; |
698 | 9 | goto http_msg_invalid; |
699 | | |
700 | 0 | case H1_MSG_RPVER_SP: |
701 | 3.26k | http_msg_rpver_sp: |
702 | 3.26k | if (likely(!HTTP_IS_LWS(*ptr))) { |
703 | 3.01k | sl.st.status = 0; |
704 | 3.01k | sl.st.c.ptr = ptr; |
705 | 3.01k | goto http_msg_rpcode; |
706 | 3.01k | } |
707 | 245 | if (likely(HTTP_IS_SPHT(*ptr))) |
708 | 242 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, H1_MSG_RPVER_SP); |
709 | | /* so it's a CR/LF, this is invalid */ |
710 | 3 | state = H1_MSG_RPVER_SP; |
711 | 3 | goto http_msg_invalid; |
712 | | |
713 | 0 | case H1_MSG_RPCODE: |
714 | 6.86k | http_msg_rpcode: |
715 | 6.86k | if (likely(HTTP_IS_DIGIT(*ptr))) { |
716 | 3.85k | if (ptr - sl.st.c.ptr >= 3) { |
717 | | /* more than 3 digits */ |
718 | 2 | if (h1m->err_pos == -1) /* only capture the error pointer */ |
719 | 0 | h1m->err_pos = ptr - start + skip; |
720 | 2 | else if (h1m->err_pos < -1 || sl.st.status >= ((uint16_t)~0 - 9) / 10) { |
721 | | /* strict checks or risk of overflow */ |
722 | 2 | state = H1_MSG_RPCODE; |
723 | 2 | goto http_msg_invalid; |
724 | 2 | } |
725 | 2 | } |
726 | 3.85k | sl.st.status = sl.st.status * 10 + *ptr - '0'; |
727 | 3.85k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode, http_msg_ood, state, H1_MSG_RPCODE); |
728 | 3.85k | } |
729 | | |
730 | 3.01k | if (unlikely(!HTTP_IS_LWS(*ptr))) { |
731 | 10 | state = H1_MSG_RPCODE; |
732 | 10 | goto http_msg_invalid; |
733 | 10 | } |
734 | | |
735 | 3.00k | if (likely(HTTP_IS_SPHT(*ptr))) { |
736 | 122 | sl.st.c.len = ptr - sl.st.c.ptr; |
737 | 122 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, H1_MSG_RPCODE_SP); |
738 | 122 | } |
739 | | |
740 | | /* so it's a CR/LF, so there is no reason phrase */ |
741 | 2.88k | sl.st.c.len = ptr - sl.st.c.ptr; |
742 | | |
743 | 2.89k | http_msg_rsp_reason: |
744 | 2.89k | sl.st.r = ist2(ptr, 0); |
745 | 2.89k | goto http_msg_rpline_eol; |
746 | | |
747 | 0 | case H1_MSG_RPCODE_SP: |
748 | 340 | http_msg_rpcode_sp: |
749 | 340 | if (likely(!HTTP_IS_LWS(*ptr))) { |
750 | 101 | sl.st.r.ptr = ptr; |
751 | 101 | goto http_msg_rpreason; |
752 | 101 | } |
753 | 239 | if (likely(HTTP_IS_SPHT(*ptr))) |
754 | 221 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, H1_MSG_RPCODE_SP); |
755 | | /* so it's a CR/LF, so there is no reason phrase */ |
756 | 18 | goto http_msg_rsp_reason; |
757 | | |
758 | 18 | case H1_MSG_RPREASON: |
759 | 196k | http_msg_rpreason: |
760 | 196k | if (likely(!HTTP_IS_CRLF(*ptr))) |
761 | 196k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpreason, http_msg_ood, state, H1_MSG_RPREASON); |
762 | 87 | sl.st.r.len = ptr - sl.st.r.ptr; |
763 | 2.98k | http_msg_rpline_eol: |
764 | | /* We have seen the end of line. Note that we do not |
765 | | * necessarily have the \n yet, but at least we know that we |
766 | | * have EITHER \r OR \n, otherwise the response would not be |
767 | | * complete. We can then record the response length and return |
768 | | * to the caller which will be able to register it. |
769 | | */ |
770 | | |
771 | 2.98k | if (likely(!skip_update)) { |
772 | 2.98k | if (unlikely(hdr_count >= hdr_num)) { |
773 | 0 | state = H1_MSG_RPREASON; |
774 | 0 | goto http_output_full; |
775 | 0 | } |
776 | 2.98k | if (!(h1m->flags & H1_MF_NO_PHDR)) |
777 | 2.98k | http_set_hdr(&hdr[hdr_count++], ist(":status"), sl.st.c); |
778 | 2.98k | } |
779 | | |
780 | 2.98k | sol = ptr - start; |
781 | 2.98k | if (likely(*ptr == '\r')) |
782 | 70 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpline_end, http_msg_ood, state, H1_MSG_RPLINE_END); |
783 | 2.91k | goto http_msg_rpline_end; |
784 | | |
785 | 2.91k | case H1_MSG_RPLINE_END: |
786 | 2.98k | http_msg_rpline_end: |
787 | | /* sol must point to the first of CR or LF. */ |
788 | 2.98k | EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_RPLINE_END); |
789 | 2.97k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_first, http_msg_ood, state, H1_MSG_HDR_FIRST); |
790 | | /* stop here */ |
791 | | |
792 | 80 | case H1_MSG_HDR_FIRST: |
793 | 5.16k | http_msg_hdr_first: |
794 | 5.16k | sol = ptr - start; |
795 | 5.16k | if (likely(!HTTP_IS_CRLF(*ptr))) { |
796 | 4.76k | goto http_msg_hdr_name; |
797 | 4.76k | } |
798 | | |
799 | 403 | if (likely(*ptr == '\r')) |
800 | 9 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, H1_MSG_LAST_LF); |
801 | 394 | goto http_msg_last_lf; |
802 | | |
803 | 394 | case H1_MSG_HDR_NAME: |
804 | 9.06M | http_msg_hdr_name: |
805 | | /* assumes sol points to the first char */ |
806 | 9.06M | if (likely(HTTP_IS_TOKEN(*ptr))) { |
807 | 9.04M | if (!skip_update) { |
808 | | /* turn it to lower case if needed */ |
809 | 9.04M | if (isupper((unsigned char)*ptr) && h1m->flags & H1_MF_TOLOWER) |
810 | 0 | *ptr = tolower((unsigned char)*ptr); |
811 | 9.04M | } |
812 | 9.04M | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, H1_MSG_HDR_NAME); |
813 | 9.04M | } |
814 | | |
815 | 22.8k | if (likely(*ptr == ':')) { |
816 | 22.4k | col = ptr - start; |
817 | 22.4k | if (col <= sol) { |
818 | 19 | state = H1_MSG_HDR_NAME; |
819 | 19 | goto http_msg_invalid; |
820 | 19 | } |
821 | 22.4k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, H1_MSG_HDR_L1_SP); |
822 | 22.4k | } |
823 | | |
824 | 380 | if (likely(h1m->err_pos < -1) || *ptr == '\n') { |
825 | 380 | state = H1_MSG_HDR_NAME; |
826 | 380 | goto http_msg_invalid; |
827 | 380 | } |
828 | | |
829 | 0 | if (h1m->err_pos == -1) /* capture the error pointer */ |
830 | 0 | h1m->err_pos = ptr - start + skip; /* >= 0 now */ |
831 | | |
832 | | /* and we still accept this non-token character */ |
833 | 0 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, H1_MSG_HDR_NAME); |
834 | | |
835 | 0 | case H1_MSG_HDR_L1_SP: |
836 | 26.4k | http_msg_hdr_l1_sp: |
837 | | /* assumes sol points to the first char */ |
838 | 26.4k | if (likely(HTTP_IS_SPHT(*ptr))) |
839 | 3.60k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, H1_MSG_HDR_L1_SP); |
840 | | |
841 | | /* header value can be basically anything except CR/LF */ |
842 | 22.8k | sov = ptr - start; |
843 | | |
844 | 22.8k | if (likely(!HTTP_IS_CRLF(*ptr))) { |
845 | 12.6k | goto http_msg_hdr_val; |
846 | 12.6k | } |
847 | | |
848 | 10.1k | if (likely(*ptr == '\r')) |
849 | 1.50k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lf, http_msg_ood, state, H1_MSG_HDR_L1_LF); |
850 | 8.65k | goto http_msg_hdr_l1_lf; |
851 | | |
852 | 8.65k | case H1_MSG_HDR_L1_LF: |
853 | 10.1k | http_msg_hdr_l1_lf: |
854 | 10.1k | EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_HDR_L1_LF); |
855 | 10.1k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lws, http_msg_ood, state, H1_MSG_HDR_L1_LWS); |
856 | | |
857 | 0 | case H1_MSG_HDR_L1_LWS: |
858 | 10.1k | http_msg_hdr_l1_lws: |
859 | 10.1k | if (likely(HTTP_IS_SPHT(*ptr))) { |
860 | 430 | if (!skip_update) { |
861 | | /* replace HT,CR,LF with spaces */ |
862 | 1.08k | for (; start + sov < ptr; sov++) |
863 | 654 | start[sov] = ' '; |
864 | 430 | } |
865 | 430 | goto http_msg_hdr_l1_sp; |
866 | 430 | } |
867 | | /* we had a header consisting only in spaces ! */ |
868 | 9.70k | eol = sov; |
869 | 9.70k | goto http_msg_complete_header; |
870 | | |
871 | 0 | case H1_MSG_HDR_VAL: |
872 | 13.5k | http_msg_hdr_val: |
873 | | /* assumes sol points to the first char, and sov |
874 | | * points to the first character of the value. |
875 | | */ |
876 | | |
877 | | /* speedup: we'll skip packs of 4 or 8 bytes not containing bytes 0x0D |
878 | | * and lower. In fact since most of the time is spent in the loop, we |
879 | | * also remove the sign bit test so that bytes 0x8e..0x0d break the |
880 | | * loop, but we don't care since they're very rare in header values. |
881 | | */ |
882 | 13.5k | #ifdef HA_UNALIGNED_LE64 |
883 | 2.14M | while (ptr <= end - sizeof(long)) { |
884 | 2.14M | if (is_char8_below_opt(*(ulong *)ptr, 0x0e)) |
885 | 11.4k | goto http_msg_hdr_val2; |
886 | 2.13M | ptr += sizeof(long); |
887 | 2.13M | } |
888 | 2.08k | #endif |
889 | 2.08k | #ifdef HA_UNALIGNED_LE |
890 | 2.62k | while (ptr <= end - sizeof(int)) { |
891 | 1.27k | if (is_char4_below_opt(*(uint *)ptr, 0x0e)) |
892 | 741 | goto http_msg_hdr_val2; |
893 | 537 | ptr += sizeof(int); |
894 | 537 | } |
895 | 1.34k | #endif |
896 | 1.34k | if (ptr >= end) { |
897 | 5 | state = H1_MSG_HDR_VAL; |
898 | 5 | goto http_msg_ood; |
899 | 5 | } |
900 | 26.6M | http_msg_hdr_val2: |
901 | 26.6M | if (likely(!*ptr)) { |
902 | | /* RFC9110 clarified that NUL is explicitly forbidden in header values |
903 | | * (like CR and LF). |
904 | | */ |
905 | 8 | if (h1m->err_pos < -1) { /* PR_O2_REQBUG_OK not set */ |
906 | 8 | state = H1_MSG_HDR_VAL; |
907 | 8 | goto http_msg_invalid; |
908 | 8 | } |
909 | 0 | if (h1m->err_pos == -1) /* PR_O2_REQBUG_OK set: just log */ |
910 | 0 | h1m->err_pos = ptr - start + skip; |
911 | 0 | } |
912 | 26.6M | if (likely(!HTTP_IS_CRLF(*ptr))) |
913 | 26.6M | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_val2, http_msg_ood, state, H1_MSG_HDR_VAL); |
914 | | |
915 | 13.4k | eol = ptr - start; |
916 | | /* Note: we could also copy eol into ->eoh so that we have the |
917 | | * real header end in case it ends with lots of LWS, but is this |
918 | | * really needed ? |
919 | | */ |
920 | 13.4k | if (likely(*ptr == '\r')) |
921 | 2.22k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lf, http_msg_ood, state, H1_MSG_HDR_L2_LF); |
922 | 11.2k | goto http_msg_hdr_l2_lf; |
923 | | |
924 | 11.2k | case H1_MSG_HDR_L2_LF: |
925 | 13.4k | http_msg_hdr_l2_lf: |
926 | 13.4k | EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_HDR_L2_LF); |
927 | 13.4k | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lws, http_msg_ood, state, H1_MSG_HDR_L2_LWS); |
928 | | |
929 | 0 | case H1_MSG_HDR_L2_LWS: |
930 | 13.4k | http_msg_hdr_l2_lws: |
931 | 13.4k | if (unlikely(HTTP_IS_SPHT(*ptr))) { |
932 | 840 | if (!skip_update) { |
933 | | /* LWS: replace HT,CR,LF with spaces */ |
934 | 1.91k | for (; start + eol < ptr; eol++) |
935 | 1.07k | start[eol] = ' '; |
936 | 840 | } |
937 | 840 | goto http_msg_hdr_val; |
938 | 840 | } |
939 | 22.2k | http_msg_complete_header: |
940 | | /* |
941 | | * It was a new header, so the last one is finished. Assumes |
942 | | * <sol> points to the first char of the name, <col> to the |
943 | | * colon, <sov> points to the first character of the value and |
944 | | * <eol> to the first CR or LF so we know how the line ends. We |
945 | | * will trim spaces around the value. It's possible to do it by |
946 | | * adjusting <eol> and <sov> which are no more used after this. |
947 | | * We can add the header field to the list. |
948 | | */ |
949 | 22.2k | if (likely(!skip_update)) { |
950 | 22.2k | while (sov < eol && HTTP_IS_LWS(start[sov])) |
951 | 0 | sov++; |
952 | | |
953 | 23.2k | while (eol - 1 > sov && HTTP_IS_LWS(start[eol - 1])) |
954 | 975 | eol--; |
955 | | |
956 | | |
957 | 22.2k | n = ist2(start + sol, col - sol); |
958 | 22.2k | v = ist2(start + sov, eol - sov); |
959 | | |
960 | 22.2k | do { |
961 | 22.2k | int ret; |
962 | | |
963 | 22.2k | if (unlikely(hdr_count >= hdr_num)) { |
964 | 2 | state = H1_MSG_HDR_L2_LWS; |
965 | 2 | goto http_output_full; |
966 | 2 | } |
967 | | |
968 | | /* Skip headers whose names contain forbidden |
969 | | * chars. When any is detected, h1m->err_pos >= 0, |
970 | | * so we recheck the name only when an error was |
971 | | * detected. |
972 | | */ |
973 | 22.2k | if (unlikely(h1m->err_pos >= 0)) { |
974 | 0 | size_t i = 0; |
975 | 0 | while (i < n.len && HTTP_IS_TOKEN(n.ptr[i])) |
976 | 0 | i++; |
977 | |
|
978 | 0 | if (i < n.len) |
979 | 0 | break; |
980 | 0 | } |
981 | | |
982 | 22.2k | if (isteqi(n, ist("transfer-encoding"))) { |
983 | 2.60k | ret = h1_parse_xfer_enc_header(h1m, v); |
984 | 2.60k | if (ret < 0) { |
985 | | /* For the response only, don't report error if PR_O2_RSPBUG_OK is set |
986 | | * and the error can be hidden */ |
987 | 61 | if (ret == -2 || !(h1m->flags & H1_MF_RESP) || (h1m->err_pos < -1)) { |
988 | 61 | state = H1_MSG_HDR_L2_LWS; |
989 | 61 | ptr = v.ptr; /* Set ptr on the error */ |
990 | 61 | goto http_msg_invalid; |
991 | 61 | } |
992 | 0 | if (h1m->err_pos == -1) |
993 | 0 | h1m->err_pos = ptr - start + skip; |
994 | 0 | } |
995 | 2.54k | else if (ret == 0) { |
996 | | /* skip it */ |
997 | 0 | break; |
998 | 0 | } |
999 | 2.60k | } |
1000 | 19.6k | else if (isteqi(n, ist("content-length"))) { |
1001 | 1.59k | unsigned long long body_len = h1m->body_len; |
1002 | | |
1003 | 1.59k | ret = http_parse_cont_len_header(&v, &body_len, (h1m->flags & H1_MF_CLEN)); |
1004 | 1.59k | if (ret < 0) { |
1005 | 255 | state = H1_MSG_HDR_L2_LWS; |
1006 | 255 | ptr = v.ptr; /* Set ptr on the error */ |
1007 | 255 | goto http_msg_invalid; |
1008 | 255 | } |
1009 | 1.33k | else if (ret == 0) { |
1010 | | /* skip it */ |
1011 | 328 | break; |
1012 | 328 | } |
1013 | 1.01k | h1m->flags |= H1_MF_CLEN; |
1014 | 1.01k | h1m->curr_len = h1m->body_len = body_len; |
1015 | 1.01k | } |
1016 | 18.0k | else if (isteqi(n, ist("connection"))) { |
1017 | 2.27k | h1_parse_connection_header(h1m, &v); |
1018 | 2.27k | if (!v.len) { |
1019 | | /* skip it */ |
1020 | 230 | break; |
1021 | 230 | } |
1022 | 2.27k | } |
1023 | 15.8k | else if (isteqi(n, ist("upgrade"))) { |
1024 | 1.37k | h1_parse_upgrade_header(h1m, &v); |
1025 | 1.37k | if (!v.len) { |
1026 | | /* skip it */ |
1027 | 300 | break; |
1028 | 300 | } |
1029 | 1.37k | } |
1030 | 14.4k | else if (!(h1m->flags & H1_MF_RESP) && isteqi(n, ist("host"))) { |
1031 | 2.60k | if (host_idx == -1) { |
1032 | 780 | host_idx = hdr_count; |
1033 | 780 | if (http_authority_has_forbidden_char(v)) { |
1034 | 29 | state = H1_MSG_HDR_L2_LWS; |
1035 | 29 | ptr = v.ptr; /* Set ptr on the error */ |
1036 | 29 | goto http_msg_invalid; |
1037 | 29 | } |
1038 | 780 | } |
1039 | 1.82k | else { |
1040 | 1.82k | if (!isteqi(v, hdr[host_idx].v)) { |
1041 | 61 | state = H1_MSG_HDR_L2_LWS; |
1042 | 61 | ptr = v.ptr; /* Set ptr on the error */ |
1043 | 61 | goto http_msg_invalid; |
1044 | 61 | } |
1045 | | /* if the same host, skip it */ |
1046 | 1.76k | break; |
1047 | 1.82k | } |
1048 | 2.60k | } |
1049 | | |
1050 | 19.2k | http_set_hdr(&hdr[hdr_count++], n, v); |
1051 | 19.2k | } while (0); |
1052 | 22.2k | } |
1053 | | |
1054 | 21.8k | sol = ptr - start; |
1055 | | |
1056 | 21.8k | if (likely(!HTTP_IS_CRLF(*ptr))) |
1057 | 18.5k | goto http_msg_hdr_name; |
1058 | | |
1059 | 3.36k | if (likely(*ptr == '\r')) |
1060 | 120 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, H1_MSG_LAST_LF); |
1061 | 3.24k | goto http_msg_last_lf; |
1062 | | |
1063 | 3.24k | case H1_MSG_LAST_LF: |
1064 | 3.98k | http_msg_last_lf: |
1065 | 3.98k | EXPECT_LF_HERE(ptr, http_msg_invalid, state, H1_MSG_LAST_LF); |
1066 | 3.96k | ptr++; |
1067 | | /* <ptr> now points to the first byte of payload. If needed sol |
1068 | | * still points to the first of either CR or LF of the empty |
1069 | | * line ending the headers block. |
1070 | | */ |
1071 | 3.96k | if (likely(!skip_update)) { |
1072 | 3.96k | if (unlikely(hdr_count >= hdr_num)) { |
1073 | 1 | state = H1_MSG_LAST_LF; |
1074 | 1 | goto http_output_full; |
1075 | 1 | } |
1076 | 3.96k | http_set_hdr(&hdr[hdr_count++], ist2(start+sol, 0), ist("")); |
1077 | 3.96k | } |
1078 | | |
1079 | | /* reaching here we've parsed the whole message. We may detect |
1080 | | * that we were already continuing an interrupted parsing pass |
1081 | | * so we were silently looking for the end of message not |
1082 | | * updating anything before deciding to parse it fully at once. |
1083 | | * It's guaranteed that we won't match this test twice in a row |
1084 | | * since restarting will turn zero. |
1085 | | */ |
1086 | 3.96k | if (restarting) |
1087 | 0 | goto restart; |
1088 | | |
1089 | | |
1090 | 3.96k | if (!(h1m->flags & (H1_MF_HDRS_ONLY|H1_MF_RESP))) { |
1091 | 2.15k | struct http_uri_parser parser = http_uri_parser_init(sl.rq.u); |
1092 | 2.15k | struct ist scheme, authority = IST_NULL; |
1093 | 2.15k | int ret; |
1094 | | |
1095 | | /* WT: gcc seems to see a path where sl.rq.u.ptr was used |
1096 | | * uninitialized, but it doesn't know that the function is |
1097 | | * called with initial states making this impossible. |
1098 | | */ |
1099 | 2.15k | ALREADY_CHECKED(sl.rq.u.ptr); |
1100 | 2.15k | switch (parser.format) { |
1101 | 164 | case URI_PARSER_FORMAT_ASTERISK: |
1102 | | /* We must take care "PRI * HTTP/2.0" is supported here. check for OTHER methods here is enough */ |
1103 | 164 | if ((sl.rq.meth != HTTP_METH_OTHER && sl.rq.meth != HTTP_METH_OPTIONS) || istlen(sl.rq.u) != 1) { |
1104 | 8 | ptr = sl.rq.u.ptr; /* Set ptr on the error */ |
1105 | 8 | goto http_msg_invalid; |
1106 | 8 | } |
1107 | 156 | break; |
1108 | | |
1109 | 482 | case URI_PARSER_FORMAT_ABSPATH: |
1110 | 482 | if (sl.rq.meth == HTTP_METH_CONNECT) { |
1111 | 1 | ptr = sl.rq.u.ptr; /* Set ptr on the error */ |
1112 | 1 | goto http_msg_invalid; |
1113 | 1 | } |
1114 | 481 | break; |
1115 | | |
1116 | 1.47k | case URI_PARSER_FORMAT_ABSURI_OR_AUTHORITY: |
1117 | 1.47k | scheme = http_parse_scheme(&parser); |
1118 | 1.47k | authority = http_parse_authority(&parser, 1); |
1119 | 1.47k | if (http_authority_has_forbidden_char(authority)) { |
1120 | 50 | if (h1m->err_pos < -1) { |
1121 | 50 | state = H1_MSG_LAST_LF; |
1122 | | /* WT: gcc seems to see a path where sl.rq.u.ptr was used |
1123 | | * uninitialized, but it doesn't know that the function is |
1124 | | * called with initial states making this impossible. |
1125 | | */ |
1126 | 50 | ALREADY_CHECKED(sl.rq.u.ptr); |
1127 | 50 | ptr = sl.rq.u.ptr; /* Set ptr on the error */ |
1128 | 50 | goto http_msg_invalid; |
1129 | 50 | } |
1130 | 0 | if (h1m->err_pos == -1) /* capture the error pointer */ |
1131 | 0 | h1m->err_pos = sl.rq.u.ptr - start + skip; /* >= 0 now */ |
1132 | 0 | } |
1133 | | |
1134 | 1.42k | if (!isttest(scheme)) { /* scheme not found: MUST be an authority */ |
1135 | 431 | struct ist *host = NULL; |
1136 | | |
1137 | 431 | if (sl.rq.meth != HTTP_METH_CONNECT) { |
1138 | 144 | ptr = sl.rq.u.ptr; /* Set ptr on the error */ |
1139 | 144 | goto http_msg_invalid; |
1140 | 144 | } |
1141 | 287 | if (host_idx != -1) |
1142 | 248 | host = &hdr[host_idx].v; |
1143 | 287 | ret = h1_validate_connect_authority(scheme, authority, host); |
1144 | 287 | if (ret < 0) { |
1145 | 243 | if (h1m->err_pos < -1) { |
1146 | 243 | state = H1_MSG_LAST_LF; |
1147 | | /* WT: gcc seems to see a path where sl.rq.u.ptr was used |
1148 | | * uninitialized, but it doesn't know that the function is |
1149 | | * called with initial states making this impossible. |
1150 | | */ |
1151 | 243 | ALREADY_CHECKED(sl.rq.u.ptr); |
1152 | 243 | ptr = ((ret == -1) ? sl.rq.u.ptr : host->ptr); /* Set ptr on the error */ |
1153 | 243 | goto http_msg_invalid; |
1154 | 243 | } |
1155 | 0 | if (h1m->err_pos == -1) /* capture the error pointer */ |
1156 | 0 | h1m->err_pos = ((ret == -1) ? sl.rq.u.ptr : host->ptr) - start + skip; /* >= 0 now */ |
1157 | 0 | } |
1158 | 287 | } |
1159 | 998 | else { /* Scheme found: MUST be an absolute-URI */ |
1160 | 998 | struct ist host = IST_NULL; |
1161 | | |
1162 | 998 | if (sl.rq.meth == HTTP_METH_CONNECT) { |
1163 | 1 | ptr = sl.rq.u.ptr; /* Set ptr on the error */ |
1164 | 1 | goto http_msg_invalid; |
1165 | 1 | } |
1166 | | |
1167 | 997 | if (host_idx != -1) |
1168 | 365 | host = hdr[host_idx].v; |
1169 | | /* For non-CONNECT method, the authority must match the host header value */ |
1170 | 997 | if (isttest(host) && !isteqi(authority, host)) { |
1171 | 258 | ret = h1_validate_mismatch_authority(scheme, authority, host); |
1172 | 258 | if (ret < 0) { |
1173 | 187 | if (h1m->err_pos < -1) { |
1174 | 187 | state = H1_MSG_LAST_LF; |
1175 | 187 | ptr = host.ptr; /* Set ptr on the error */ |
1176 | 187 | goto http_msg_invalid; |
1177 | 187 | } |
1178 | 0 | if (h1m->err_pos == -1) /* capture the error pointer */ |
1179 | 0 | h1m->err_pos = host.ptr - start + skip; /* >= 0 now */ |
1180 | 0 | } |
1181 | 258 | } |
1182 | 997 | } |
1183 | 854 | break; |
1184 | | |
1185 | 854 | default: |
1186 | 26 | ptr = sl.rq.u.ptr; /* Set ptr on the error */ |
1187 | 26 | goto http_msg_invalid; |
1188 | 2.15k | } |
1189 | 2.15k | } |
1190 | | |
1191 | 3.30k | state = H1_MSG_DATA; |
1192 | 3.30k | if (h1m->flags & H1_MF_XFER_ENC) { |
1193 | 1.37k | if (h1m->flags & H1_MF_CLEN) { |
1194 | | /* T-E + C-L: force close and remove C-L */ |
1195 | 95 | if (!h1_do_not_close_on_insecure_t_e) |
1196 | 95 | h1m->flags |= H1_MF_CONN_CLO; |
1197 | | |
1198 | 95 | h1m->flags &= ~H1_MF_CLEN; |
1199 | 95 | h1m->curr_len = h1m->body_len = 0; |
1200 | 95 | hdr_count = http_del_hdr(hdr, ist("content-length")); |
1201 | 95 | } |
1202 | 1.28k | else if (!(h1m->flags & H1_MF_VER_11)) { |
1203 | | /* T-E + HTTP/1.0: force close */ |
1204 | 582 | h1m->flags |= H1_MF_CONN_CLO; |
1205 | 582 | } |
1206 | | |
1207 | 1.37k | if (h1m->flags & H1_MF_CHNK) |
1208 | 1.13k | state = H1_MSG_CHUNK_SIZE; |
1209 | 244 | else if (!(h1m->flags & H1_MF_RESP)) { |
1210 | | /* cf RFC7230#3.3.3 : transfer-encoding in |
1211 | | * request without chunked encoding is invalid. |
1212 | | */ |
1213 | 6 | goto http_msg_invalid; |
1214 | 6 | } |
1215 | 1.37k | } |
1216 | 3.30k | break; |
1217 | | |
1218 | 3.30k | default: |
1219 | | /* impossible states */ |
1220 | 0 | goto http_msg_invalid; |
1221 | 5.73k | } |
1222 | | |
1223 | | /* Now we've left the headers state and are either in H1_MSG_DATA or |
1224 | | * H1_MSG_CHUNK_SIZE. |
1225 | | */ |
1226 | | |
1227 | 3.30k | if (slp && !skip_update) |
1228 | 3.30k | *slp = sl; |
1229 | | |
1230 | 3.30k | h1m->state = state; |
1231 | 3.30k | h1m->next = ptr - start + skip; |
1232 | 3.30k | return h1m->next; |
1233 | | |
1234 | 805 | http_msg_ood: |
1235 | | /* out of data at <ptr> during state <state> */ |
1236 | 805 | if (slp && !skip_update) |
1237 | 805 | *slp = sl; |
1238 | | |
1239 | 805 | h1m->state = state; |
1240 | 805 | h1m->next = ptr - start + skip; |
1241 | 805 | return 0; |
1242 | | |
1243 | 1.62k | http_msg_invalid: |
1244 | | /* invalid message, error at <ptr> */ |
1245 | 1.62k | if (slp && !skip_update) |
1246 | 1.54k | *slp = sl; |
1247 | | |
1248 | 1.62k | h1m->err_state = h1m->state = state; |
1249 | 1.62k | h1m->err_pos = h1m->next = ptr - start + skip; |
1250 | 1.62k | return -1; |
1251 | | |
1252 | 3 | http_output_full: |
1253 | | /* no more room to store the current header, error at <ptr> */ |
1254 | 3 | if (slp && !skip_update) |
1255 | 3 | *slp = sl; |
1256 | | |
1257 | 3 | h1m->err_state = h1m->state = state; |
1258 | 3 | h1m->err_pos = h1m->next = ptr - start + skip; |
1259 | 3 | return -2; |
1260 | | |
1261 | 0 | restart: |
1262 | 0 | h1m->flags &= H1_MF_RESTART_MASK; |
1263 | 0 | h1m->curr_len = h1m->body_len = h1m->next = 0; |
1264 | 0 | if (h1m->flags & H1_MF_RESP) |
1265 | 0 | h1m->state = H1_MSG_RPBEFORE; |
1266 | 0 | else |
1267 | 0 | h1m->state = H1_MSG_RQBEFORE; |
1268 | 0 | goto try_again; |
1269 | 5.73k | } |
1270 | | |
1271 | | /* Generate a random key for a WebSocket Handshake in respect with rfc6455 |
1272 | | * The key is 128-bits long encoded as a base64 string in <key_out> parameter |
1273 | | * (25 bytes long). |
1274 | | */ |
1275 | | void h1_generate_random_ws_input_key(char key_out[25]) |
1276 | 0 | { |
1277 | | /* generate a random websocket key */ |
1278 | 0 | uint64_t rand1, rand2; |
1279 | 0 | char key[16]; |
1280 | |
|
1281 | 0 | ha_random64_pair_hashed(&rand1, &rand2); |
1282 | 0 | memcpy(key, &rand1, 8); |
1283 | 0 | memcpy(&key[8], &rand2, 8); |
1284 | 0 | a2base64(key, 16, key_out, 25); |
1285 | 0 | } |
1286 | | |
1287 | 0 | #define H1_WS_KEY_SUFFIX_GUID "258EAFA5-E914-47DA-95CA-C5AB0DC85B11" |
1288 | | |
1289 | | /* |
1290 | | * Calculate the WebSocket handshake response key from <key_in>. Following the |
1291 | | * rfc6455, <key_in> must be 24 bytes longs. The result is stored in <key_out> |
1292 | | * as a 29 bytes long string. |
1293 | | */ |
1294 | | void h1_calculate_ws_output_key(const char *key, char *result) |
1295 | 0 | { |
1296 | 0 | blk_SHA_CTX sha1_ctx; |
1297 | 0 | char hash_in[60], hash_out[20]; |
1298 | | |
1299 | | /* concatenate the key with a fixed suffix */ |
1300 | 0 | memcpy(hash_in, key, 24); |
1301 | 0 | memcpy(&hash_in[24], H1_WS_KEY_SUFFIX_GUID, 36); |
1302 | | |
1303 | | /* sha1 the result */ |
1304 | 0 | blk_SHA1_Init(&sha1_ctx); |
1305 | 0 | blk_SHA1_Update(&sha1_ctx, hash_in, 60); |
1306 | 0 | blk_SHA1_Final((unsigned char *)hash_out, &sha1_ctx); |
1307 | | |
1308 | | /* encode in base64 the hash */ |
1309 | 0 | a2base64(hash_out, 20, result, 29); |
1310 | 0 | } |
1311 | | |
1312 | | /* config parser for global "h1-do-not-close-on-insecure-transfer-encoding" */ |
1313 | | static int cfg_parse_h1_do_not_close_insecure_t_e(char **args, int section_type, struct proxy *curpx, |
1314 | | const struct proxy *defpx, const char *file, int line, |
1315 | | char **err) |
1316 | 0 | { |
1317 | 0 | if (too_many_args(0, args, err, NULL)) |
1318 | 0 | return -1; |
1319 | | |
1320 | 0 | h1_do_not_close_on_insecure_t_e = 1; |
1321 | 0 | return 0; |
1322 | 0 | } |
1323 | | |
1324 | | /* config keyword parsers */ |
1325 | | static struct cfg_kw_list cfg_kws = {{ }, { |
1326 | | { CFG_GLOBAL, "h1-do-not-close-on-insecure-transfer-encoding", cfg_parse_h1_do_not_close_insecure_t_e }, |
1327 | | { 0, NULL, NULL }, |
1328 | | }}; |
1329 | | |
1330 | | INITCALL1(STG_REGISTER, cfg_register_keywords, &cfg_kws); |