blob: b62c1304f13bc94909ad5df4e17ae14e91c1ac2c [file] [log] [blame]
Willy Tarreauf24ea8e2017-11-21 19:55:27 +01001/*
2 * HTTP/2 protocol processing
3 *
4 * Copyright 2017 Willy Tarreau <w@1wt.eu>
5 * Copyright (C) 2017 HAProxy Technologies
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining
8 * a copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sublicense, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be
16 * included in all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
20 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
21 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
22 * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
23 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
25 * OTHER DEALINGS IN THE SOFTWARE.
26 */
27
Willy Tarreaua1bd1fa2019-03-29 17:26:33 +010028#include <inttypes.h>
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020029#include <haproxy/api.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020030#include <haproxy/global.h>
Willy Tarreaubf073142020-06-03 12:04:01 +020031#include <haproxy/h2.h>
Willy Tarreau0017be02020-06-02 19:25:28 +020032#include <haproxy/http-hdr-t.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020033#include <haproxy/http.h>
Amaury Denoyelle03534102021-07-07 10:49:28 +020034#include <haproxy/http_htx.h>
Willy Tarreau16f958c2020-06-03 08:44:35 +020035#include <haproxy/htx.h>
Willy Tarreaueb6f7012020-05-27 16:21:26 +020036#include <import/ist.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020037
Willy Tarreauf24ea8e2017-11-21 19:55:27 +010038
Willy Tarreau9c84d822019-01-30 15:09:21 +010039struct h2_frame_definition h2_frame_definition[H2_FT_ENTRIES] = {
40 [H2_FT_DATA ] = { .dir = 3, .min_id = 1, .max_id = H2_MAX_STREAM_ID, .min_len = 0, .max_len = H2_MAX_FRAME_LEN, },
41 [H2_FT_HEADERS ] = { .dir = 3, .min_id = 1, .max_id = H2_MAX_STREAM_ID, .min_len = 1, .max_len = H2_MAX_FRAME_LEN, },
42 [H2_FT_PRIORITY ] = { .dir = 3, .min_id = 1, .max_id = H2_MAX_STREAM_ID, .min_len = 5, .max_len = 5, },
43 [H2_FT_RST_STREAM ] = { .dir = 3, .min_id = 1, .max_id = H2_MAX_STREAM_ID, .min_len = 4, .max_len = 4, },
44 [H2_FT_SETTINGS ] = { .dir = 3, .min_id = 0, .max_id = 0, .min_len = 0, .max_len = H2_MAX_FRAME_LEN, },
45 [H2_FT_PUSH_PROMISE ] = { .dir = 0, .min_id = 1, .max_id = H2_MAX_STREAM_ID, .min_len = 4, .max_len = H2_MAX_FRAME_LEN, },
46 [H2_FT_PING ] = { .dir = 3, .min_id = 0, .max_id = 0, .min_len = 8, .max_len = 8, },
47 [H2_FT_GOAWAY ] = { .dir = 3, .min_id = 0, .max_id = 0, .min_len = 8, .max_len = H2_MAX_FRAME_LEN, },
48 [H2_FT_WINDOW_UPDATE] = { .dir = 3, .min_id = 0, .max_id = H2_MAX_STREAM_ID, .min_len = 4, .max_len = 4, },
49 [H2_FT_CONTINUATION ] = { .dir = 3, .min_id = 1, .max_id = H2_MAX_STREAM_ID, .min_len = 0, .max_len = H2_MAX_FRAME_LEN, },
50};
Willy Tarreauf24ea8e2017-11-21 19:55:27 +010051
Willy Tarreau54f53ef2019-11-22 16:02:43 +010052/* Looks into <ist> for forbidden characters for header values (0x00, 0x0A,
53 * 0x0D), starting at pointer <start> which must be within <ist>. Returns
54 * non-zero if such a character is found, 0 otherwise. When run on unlikely
55 * header match, it's recommended to first check for the presence of control
56 * chars using ist_find_ctl().
57 */
58static int has_forbidden_char(const struct ist ist, const char *start)
59{
60 do {
61 if ((uint8_t)*start <= 0x0d &&
62 (1U << (uint8_t)*start) & ((1<<13) | (1<<10) | (1<<0)))
63 return 1;
64 start++;
65 } while (start < ist.ptr + ist.len);
66 return 0;
67}
68
Willy Tarreaubeefaee2018-12-19 13:08:08 +010069/* Parse the Content-Length header field of an HTTP/2 request. The function
70 * checks all possible occurrences of a comma-delimited value, and verifies
71 * if any of them doesn't match a previous value. It returns <0 if a value
72 * differs, 0 if the whole header can be dropped (i.e. already known), or >0
73 * if the value can be indexed (first one). In the last case, the value might
74 * be adjusted and the caller must only add the updated value.
75 */
76int h2_parse_cont_len_header(unsigned int *msgf, struct ist *value, unsigned long long *body_len)
77{
78 char *e, *n;
79 unsigned long long cl;
80 int not_first = !!(*msgf & H2_MSGF_BODY_CL);
81 struct ist word;
82
83 word.ptr = value->ptr - 1; // -1 for next loop's pre-increment
84 e = value->ptr + value->len;
85
86 while (++word.ptr < e) {
Ilya Shipitsin6fb0f212020-04-02 15:25:26 +050087 /* skip leading delimiter and blanks */
Willy Tarreaubeefaee2018-12-19 13:08:08 +010088 if (unlikely(HTTP_IS_LWS(*word.ptr)))
89 continue;
90
91 /* digits only now */
92 for (cl = 0, n = word.ptr; n < e; n++) {
93 unsigned int c = *n - '0';
94 if (unlikely(c > 9)) {
95 /* non-digit */
96 if (unlikely(n == word.ptr)) // spaces only
97 goto fail;
98 break;
99 }
100 if (unlikely(cl > ULLONG_MAX / 10ULL))
101 goto fail; /* multiply overflow */
102 cl = cl * 10ULL;
103 if (unlikely(cl + c < cl))
104 goto fail; /* addition overflow */
105 cl = cl + c;
106 }
107
108 /* keep a copy of the exact cleaned value */
109 word.len = n - word.ptr;
110
111 /* skip trailing LWS till next comma or EOL */
112 for (; n < e; n++) {
113 if (!HTTP_IS_LWS(*n)) {
114 if (unlikely(*n != ','))
115 goto fail;
116 break;
117 }
118 }
119
120 /* if duplicate, must be equal */
121 if (*msgf & H2_MSGF_BODY_CL && cl != *body_len)
122 goto fail;
123
124 /* OK, store this result as the one to be indexed */
125 *msgf |= H2_MSGF_BODY_CL;
126 *body_len = cl;
127 *value = word;
128 word.ptr = n;
129 }
130 /* here we've reached the end with a single value or a series of
131 * identical values, all matching previous series if any. The last
132 * parsed value was sent back into <value>. We just have to decide
133 * if this occurrence has to be indexed (it's the first one) or
134 * silently skipped (it's not the first one)
135 */
136 return !not_first;
137 fail:
138 return -1;
139}
140
Willy Tarreau6deb4122018-11-27 15:34:18 +0100141/* Prepare the request line into <htx> from pseudo headers stored in <phdr[]>.
142 * <fields> indicates what was found so far. This should be called once at the
143 * detection of the first general header field or at the end of the request if
144 * no general header field was found yet. Returns the created start line on
145 * success, or NULL on failure. Upon success, <msgf> is updated with a few
146 * H2_MSGF_* flags indicating what was found while parsing.
Willy Tarreau2be362c2019-10-08 11:59:37 +0200147 *
148 * The rules below deserve a bit of explanation. There tends to be some
149 * confusion regarding H2's authority vs the Host header. They are different
150 * though may sometimes be exchanged. In H2, the request line is broken into :
151 * - :method
152 * - :scheme
153 * - :authority
154 * - :path
155 *
156 * An equivalent HTTP/1.x absolute-form request would then look like :
157 * <:method> <:scheme>://<:authority><:path> HTTP/x.y
158 *
159 * Except for CONNECT which doesn't have scheme nor path and looks like :
160 * <:method> <:authority> HTTP/x.y
161 *
162 * It's worth noting that H2 still supports an encoding to map H1 origin-form
163 * and asterisk-form requests. These ones do not specify the authority. However
164 * in H2 they must still specify the scheme, which is not present in H1. Also,
165 * when encoding an absolute-form H1 request without a path, the path
166 * automatically becomes "/" except for the OPTIONS method where it
167 * becomes "*".
168 *
169 * As such it is explicitly permitted for an H2 client to send a request
170 * featuring a Host header and no :authority, though it's not the recommended
171 * way to use H2 for a client. It is however the only permitted way to encode
172 * an origin-form H1 request over H2. Thus we need to respect such differences
173 * as much as possible when re-encoding the H2 request into HTX.
Willy Tarreau6deb4122018-11-27 15:34:18 +0100174 */
175static struct htx_sl *h2_prepare_htx_reqline(uint32_t fields, struct ist *phdr, struct htx *htx, unsigned int *msgf)
176{
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100177 struct ist uri, meth_sl;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100178 unsigned int flags = HTX_SL_F_NONE;
179 struct htx_sl *sl;
Willy Tarreau9255e7e2019-03-05 10:47:37 +0100180 size_t i;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100181
182 if ((fields & H2_PHDR_FND_METH) && isteq(phdr[H2_PHDR_IDX_METH], ist("CONNECT"))) {
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100183 if (fields & H2_PHDR_FND_PROT) {
184 /* rfc 8441 Extended Connect Protocol
185 * #4 :scheme and :path must be present, as well as
186 * :authority like all h2 requests
187 */
188 if (!(fields & H2_PHDR_FND_SCHM)) {
189 /* missing scheme */
190 goto fail;
191 }
192 else if (!(fields & H2_PHDR_FND_PATH)) {
193 /* missing path */
194 goto fail;
195 }
196 else if (!(fields & H2_PHDR_FND_AUTH)) {
197 /* missing authority */
198 goto fail;
199 }
200
201 flags |= HTX_SL_F_HAS_SCHM;
202 if (isteqi(phdr[H2_PHDR_IDX_SCHM], ist("http")))
203 flags |= HTX_SL_F_SCHM_HTTP;
204 else if (isteqi(phdr[H2_PHDR_IDX_SCHM], ist("https")))
205 flags |= HTX_SL_F_SCHM_HTTPS;
Willy Tarreau9e0c2b52021-08-10 15:37:34 +0200206 else if (!http_validate_scheme(phdr[H2_PHDR_IDX_SCHM]))
207 htx->flags |= HTX_FL_PARSING_ERROR;
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100208
209 meth_sl = ist("GET");
210
211 *msgf |= H2_MSGF_EXT_CONNECT;
212 /* no ES on the HEADERS frame but no body either for
213 * Extended CONNECT */
214 *msgf &= ~H2_MSGF_BODY;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100215 }
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100216 else {
217 /* RFC 7540 #8.2.6 regarding CONNECT: ":scheme" and ":path"
218 * MUST be omitted ; ":authority" contains the host and port
219 * to connect to.
220 */
221 if (fields & H2_PHDR_FND_SCHM) {
222 /* scheme not allowed */
223 goto fail;
224 }
225 else if (fields & H2_PHDR_FND_PATH) {
226 /* path not allowed */
227 goto fail;
228 }
229 else if (!(fields & H2_PHDR_FND_AUTH)) {
230 /* missing authority */
231 goto fail;
232 }
233
234 meth_sl = phdr[H2_PHDR_IDX_METH];
Willy Tarreau6deb4122018-11-27 15:34:18 +0100235 }
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100236
Willy Tarreau6deb4122018-11-27 15:34:18 +0100237 *msgf |= H2_MSGF_BODY_TUNNEL;
238 }
239 else if ((fields & (H2_PHDR_FND_METH|H2_PHDR_FND_SCHM|H2_PHDR_FND_PATH)) !=
240 (H2_PHDR_FND_METH|H2_PHDR_FND_SCHM|H2_PHDR_FND_PATH)) {
241 /* RFC 7540 #8.1.2.3 : all requests MUST include exactly one
242 * valid value for the ":method", ":scheme" and ":path" phdr
243 * unless it is a CONNECT request.
244 */
245 if (!(fields & H2_PHDR_FND_METH)) {
246 /* missing method */
247 goto fail;
248 }
249 else if (!(fields & H2_PHDR_FND_SCHM)) {
250 /* missing scheme */
251 goto fail;
252 }
253 else {
254 /* missing path */
255 goto fail;
256 }
257 }
Willy Tarreau2be362c2019-10-08 11:59:37 +0200258 else { /* regular methods */
Willy Tarreau92919f72019-10-08 16:53:07 +0200259 /* RFC3986#6.2.2.1: scheme is case-insensitive. We need to
260 * classify the scheme as "present/http", "present/https",
261 * "present/other", "absent" so as to decide whether or not
262 * we're facing a normalized URI that will have to be encoded
263 * in origin or absolute form. Indeed, 7540#8.1.2.3 says that
264 * clients should use the absolute form, thus we cannot infer
265 * whether or not the client wanted to use a proxy here.
266 */
267 flags |= HTX_SL_F_HAS_SCHM;
268 if (isteqi(phdr[H2_PHDR_IDX_SCHM], ist("http")))
269 flags |= HTX_SL_F_SCHM_HTTP;
270 else if (isteqi(phdr[H2_PHDR_IDX_SCHM], ist("https")))
271 flags |= HTX_SL_F_SCHM_HTTPS;
Willy Tarreau9e0c2b52021-08-10 15:37:34 +0200272 else if (!http_validate_scheme(phdr[H2_PHDR_IDX_SCHM]))
273 htx->flags |= HTX_FL_PARSING_ERROR;
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100274
275 meth_sl = phdr[H2_PHDR_IDX_METH];
Willy Tarreau92919f72019-10-08 16:53:07 +0200276 }
277
278 if (!(flags & HTX_SL_F_HAS_SCHM)) {
279 /* no scheme, use authority only (CONNECT) */
280 uri = phdr[H2_PHDR_IDX_AUTH];
Willy Tarreau1440fe82019-10-08 17:34:50 +0200281 flags |= HTX_SL_F_HAS_AUTHORITY;
Willy Tarreau92919f72019-10-08 16:53:07 +0200282 }
Willy Tarreau30ee1ef2019-10-08 18:33:19 +0200283 else if (fields & H2_PHDR_FND_AUTH) {
284 /* authority is present, let's use the absolute form. We simply
285 * use the trash to concatenate them since all of them MUST fit
286 * in a bufsize since it's where they come from.
Willy Tarreau92919f72019-10-08 16:53:07 +0200287 */
Willy Tarreaufd2658c2020-02-26 13:51:38 +0100288 if (unlikely(!phdr[H2_PHDR_IDX_PATH].len))
289 goto fail; // 7540#8.1.2.3: :path must not be empty
290
Willy Tarreau92919f72019-10-08 16:53:07 +0200291 uri = ist2bin(trash.area, phdr[H2_PHDR_IDX_SCHM]);
292 istcat(&uri, ist("://"), trash.size);
293 istcat(&uri, phdr[H2_PHDR_IDX_AUTH], trash.size);
294 if (!isteq(phdr[H2_PHDR_IDX_PATH], ist("*")))
295 istcat(&uri, phdr[H2_PHDR_IDX_PATH], trash.size);
Willy Tarreau1440fe82019-10-08 17:34:50 +0200296 flags |= HTX_SL_F_HAS_AUTHORITY;
Willy Tarreau30ee1ef2019-10-08 18:33:19 +0200297
298 if (flags & (HTX_SL_F_SCHM_HTTP|HTX_SL_F_SCHM_HTTPS)) {
299 /* we don't know if it was originally an absolute or a
300 * relative request because newer versions of HTTP use
301 * the absolute URI format by default, which we call
302 * the normalized URI format internally. This is the
303 * strongly recommended way of sending a request for
304 * a regular client, so we cannot distinguish this
305 * from a request intended for a proxy. For other
306 * schemes however there is no doubt.
307 */
308 flags |= HTX_SL_F_NORMALIZED_URI;
309 }
Willy Tarreau92919f72019-10-08 16:53:07 +0200310 }
311 else {
312 /* usual schemes with or without authority, use origin form */
313 uri = phdr[H2_PHDR_IDX_PATH];
Willy Tarreau1440fe82019-10-08 17:34:50 +0200314 if (fields & H2_PHDR_FND_AUTH)
315 flags |= HTX_SL_F_HAS_AUTHORITY;
Willy Tarreau2be362c2019-10-08 11:59:37 +0200316 }
Willy Tarreau6deb4122018-11-27 15:34:18 +0100317
Willy Tarreau2be362c2019-10-08 11:59:37 +0200318 /* make sure the final URI isn't empty. Note that 7540#8.1.2.3 states
319 * that :path must not be empty.
320 */
Willy Tarreau92919f72019-10-08 16:53:07 +0200321 if (!uri.len)
Willy Tarreau6deb4122018-11-27 15:34:18 +0100322 goto fail;
323
Willy Tarreau2be362c2019-10-08 11:59:37 +0200324 /* The final URI must not contain LWS nor CTL characters */
Willy Tarreau92919f72019-10-08 16:53:07 +0200325 for (i = 0; i < uri.len; i++) {
326 unsigned char c = uri.ptr[i];
Willy Tarreau9255e7e2019-03-05 10:47:37 +0100327 if (HTTP_IS_LWS(c) || HTTP_IS_CTL(c))
328 htx->flags |= HTX_FL_PARSING_ERROR;
329 }
330
Willy Tarreau6deb4122018-11-27 15:34:18 +0100331 /* Set HTX start-line flags */
332 flags |= HTX_SL_F_VER_11; // V2 in fact
333 flags |= HTX_SL_F_XFER_LEN; // xfer len always known with H2
334
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100335 sl = htx_add_stline(htx, HTX_BLK_REQ_SL, flags, meth_sl, uri, ist("HTTP/2.0"));
Willy Tarreau6deb4122018-11-27 15:34:18 +0100336 if (!sl)
337 goto fail;
338
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100339 sl->info.req.meth = find_http_meth(meth_sl.ptr, meth_sl.len);
Christopher Faulet7d247f02020-12-02 14:26:36 +0100340 if (sl->info.req.meth == HTTP_METH_HEAD)
341 *msgf |= H2_MSGF_BODYLESS_RSP;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100342 return sl;
343 fail:
344 return NULL;
345}
346
347/* Takes an H2 request present in the headers list <list> terminated by a name
348 * being <NULL,0> and emits the equivalent HTX request according to the rules
349 * documented in RFC7540 #8.1.2. The output contents are emitted in <htx>, and
350 * non-zero is returned if some bytes were emitted. In case of error, a
351 * negative error code is returned.
352 *
353 * Upon success, <msgf> is filled with a few H2_MSGF_* flags indicating what
354 * was found while parsing. The caller must set it to zero in or H2_MSGF_BODY
355 * if a body is detected (!ES).
356 *
357 * The headers list <list> must be composed of :
358 * - n.name != NULL, n.len > 0 : literal header name
359 * - n.name == NULL, n.len > 0 : indexed pseudo header name number <n.len>
360 * among H2_PHDR_IDX_*
361 * - n.name ignored, n.len == 0 : end of list
362 * - in all cases except the end of list, v.name and v.len must designate a
363 * valid value.
364 *
365 * The Cookie header will be reassembled at the end, and for this, the <list>
366 * will be used to create a linked list, so its contents may be destroyed.
367 */
Willy Tarreau4790f7c2019-01-24 11:33:02 +0100368int h2_make_htx_request(struct http_hdr *list, struct htx *htx, unsigned int *msgf, unsigned long long *body_len)
Willy Tarreau6deb4122018-11-27 15:34:18 +0100369{
370 struct ist phdr_val[H2_PHDR_NUM_ENTRIES];
371 uint32_t fields; /* bit mask of H2_PHDR_FND_* */
372 uint32_t idx;
373 int ck, lck; /* cookie index and last cookie index */
374 int phdr;
375 int ret;
376 int i;
377 struct htx_sl *sl = NULL;
378 unsigned int sl_flags = 0;
Willy Tarreau54f53ef2019-11-22 16:02:43 +0100379 const char *ctl;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100380
381 lck = ck = -1; // no cookie for now
382 fields = 0;
383 for (idx = 0; list[idx].n.len != 0; idx++) {
384 if (!list[idx].n.ptr) {
385 /* this is an indexed pseudo-header */
386 phdr = list[idx].n.len;
387 }
388 else {
389 /* this can be any type of header */
Willy Tarreau146f53a2019-11-24 10:34:39 +0100390 /* RFC7540#8.1.2: upper case not allowed in header field names.
391 * #10.3: header names must be valid (i.e. match a token).
392 * For pseudo-headers we check from 2nd char and for other ones
393 * from the first char, because HTTP_IS_TOKEN() also excludes
394 * the colon.
395 */
Willy Tarreau6deb4122018-11-27 15:34:18 +0100396 phdr = h2_str_to_phdr(list[idx].n);
Willy Tarreau146f53a2019-11-24 10:34:39 +0100397
398 for (i = !!phdr; i < list[idx].n.len; i++)
399 if ((uint8_t)(list[idx].n.ptr[i] - 'A') < 'Z' - 'A' || !HTTP_IS_TOKEN(list[idx].n.ptr[i]))
400 goto fail;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100401 }
402
Willy Tarreau54f53ef2019-11-22 16:02:43 +0100403 /* RFC7540#10.3: intermediaries forwarding to HTTP/1 must take care of
404 * rejecting NUL, CR and LF characters.
405 */
406 ctl = ist_find_ctl(list[idx].v);
407 if (unlikely(ctl) && has_forbidden_char(list[idx].v, ctl))
408 goto fail;
409
Willy Tarreau6deb4122018-11-27 15:34:18 +0100410 if (phdr > 0 && phdr < H2_PHDR_NUM_ENTRIES) {
411 /* insert a pseudo header by its index (in phdr) and value (in value) */
412 if (fields & ((1 << phdr) | H2_PHDR_FND_NONE)) {
413 if (fields & H2_PHDR_FND_NONE) {
414 /* pseudo header field after regular headers */
415 goto fail;
416 }
417 else {
418 /* repeated pseudo header field */
419 goto fail;
420 }
421 }
422 fields |= 1 << phdr;
423 phdr_val[phdr] = list[idx].v;
424 continue;
425 }
426 else if (phdr != 0) {
427 /* invalid pseudo header -- should never happen here */
428 goto fail;
429 }
430
431 /* regular header field in (name,value) */
432 if (unlikely(!(fields & H2_PHDR_FND_NONE))) {
433 /* no more pseudo-headers, time to build the request line */
434 sl = h2_prepare_htx_reqline(fields, phdr_val, htx, msgf);
435 if (!sl)
436 goto fail;
437 fields |= H2_PHDR_FND_NONE;
438 }
439
440 if (isteq(list[idx].n, ist("host")))
441 fields |= H2_PHDR_FND_HOST;
442
Willy Tarreaubeefaee2018-12-19 13:08:08 +0100443 if (isteq(list[idx].n, ist("content-length"))) {
Willy Tarreau4790f7c2019-01-24 11:33:02 +0100444 ret = h2_parse_cont_len_header(msgf, &list[idx].v, body_len);
Willy Tarreaubeefaee2018-12-19 13:08:08 +0100445 if (ret < 0)
446 goto fail;
447
Willy Tarreau6deb4122018-11-27 15:34:18 +0100448 sl_flags |= HTX_SL_F_CLEN;
Willy Tarreaubeefaee2018-12-19 13:08:08 +0100449 if (ret == 0)
450 continue; // skip this duplicate
Willy Tarreau6deb4122018-11-27 15:34:18 +0100451 }
452
453 /* these ones are forbidden in requests (RFC7540#8.1.2.2) */
454 if (isteq(list[idx].n, ist("connection")) ||
455 isteq(list[idx].n, ist("proxy-connection")) ||
456 isteq(list[idx].n, ist("keep-alive")) ||
457 isteq(list[idx].n, ist("upgrade")) ||
458 isteq(list[idx].n, ist("transfer-encoding")))
459 goto fail;
460
461 if (isteq(list[idx].n, ist("te")) && !isteq(list[idx].v, ist("trailers")))
462 goto fail;
463
464 /* cookie requires special processing at the end */
465 if (isteq(list[idx].n, ist("cookie"))) {
466 list[idx].n.len = -1;
467
468 if (ck < 0)
469 ck = idx;
470 else
471 list[lck].n.len = idx;
472
473 lck = idx;
474 continue;
475 }
476
477 if (!htx_add_header(htx, list[idx].n, list[idx].v))
478 goto fail;
479 }
480
481 /* RFC7540#8.1.2.1 mandates to reject response pseudo-headers (:status) */
482 if (fields & H2_PHDR_FND_STAT)
483 goto fail;
484
485 /* Let's dump the request now if not yet emitted. */
486 if (!(fields & H2_PHDR_FND_NONE)) {
487 sl = h2_prepare_htx_reqline(fields, phdr_val, htx, msgf);
488 if (!sl)
489 goto fail;
490 }
491
Christopher Fauletd0db4232021-01-22 11:46:30 +0100492 if (*msgf & H2_MSGF_BODY_TUNNEL)
493 *msgf &= ~(H2_MSGF_BODY|H2_MSGF_BODY_CL);
494
Christopher Fauletd1ac2b92020-12-02 19:12:22 +0100495 if (!(*msgf & H2_MSGF_BODY) || ((*msgf & H2_MSGF_BODY_CL) && *body_len == 0) ||
496 (*msgf & H2_MSGF_BODY_TUNNEL)) {
497 /* Request without body or tunnel requested */
Christopher Faulet44af3cf2019-02-18 10:12:56 +0100498 sl_flags |= HTX_SL_F_BODYLESS;
Christopher Fauletd1ac2b92020-12-02 19:12:22 +0100499 htx->flags |= HTX_FL_EOM;
500 }
Christopher Faulet44af3cf2019-02-18 10:12:56 +0100501
Amaury Denoyellec9a0afc2020-12-11 17:53:09 +0100502 if (*msgf & H2_MSGF_EXT_CONNECT) {
503 if (!htx_add_header(htx, ist("upgrade"), phdr_val[H2_PHDR_IDX_PROT]))
504 goto fail;
505 if (!htx_add_header(htx, ist("connection"), ist("upgrade")))
506 goto fail;
507 sl_flags |= HTX_SL_F_CONN_UPG;
508 }
509
Willy Tarreau6deb4122018-11-27 15:34:18 +0100510 /* update the start line with last detected header info */
511 sl->flags |= sl_flags;
512
513 /* complete with missing Host if needed */
514 if ((fields & (H2_PHDR_FND_HOST|H2_PHDR_FND_AUTH)) == H2_PHDR_FND_AUTH) {
515 /* missing Host field, use :authority instead */
516 if (!htx_add_header(htx, ist("host"), phdr_val[H2_PHDR_IDX_AUTH]))
517 goto fail;
518 }
519
520 /* now we may have to build a cookie list. We'll dump the values of all
521 * visited headers.
522 */
523 if (ck >= 0) {
524 uint32_t fs; // free space
525 uint32_t bs; // block size
526 uint32_t vl; // value len
Willy Tarreau164e0612018-12-18 11:00:41 +0100527 uint32_t tl; // total length
Willy Tarreau6deb4122018-11-27 15:34:18 +0100528 struct htx_blk *blk;
529
530 blk = htx_add_header(htx, ist("cookie"), list[ck].v);
531 if (!blk)
532 goto fail;
533
Willy Tarreau164e0612018-12-18 11:00:41 +0100534 tl = list[ck].v.len;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100535 fs = htx_free_data_space(htx);
536 bs = htx_get_blksz(blk);
537
538 /* for each extra cookie, we'll extend the cookie's value and
539 * insert "; " before the new value.
540 */
Willy Tarreau164e0612018-12-18 11:00:41 +0100541 fs += tl; // first one is already counted
Tim Duesterhus15683552021-03-04 23:50:13 +0100542 while ((ck = list[ck].n.len) >= 0) {
Willy Tarreau6deb4122018-11-27 15:34:18 +0100543 vl = list[ck].v.len;
Willy Tarreau164e0612018-12-18 11:00:41 +0100544 tl += vl + 2;
545 if (tl > fs)
Willy Tarreau6deb4122018-11-27 15:34:18 +0100546 goto fail;
547
Christopher Faulet3e2638e2019-06-18 09:49:16 +0200548 htx_change_blk_value_len(htx, blk, tl);
Willy Tarreau6deb4122018-11-27 15:34:18 +0100549 *(char *)(htx_get_blk_ptr(htx, blk) + bs + 0) = ';';
550 *(char *)(htx_get_blk_ptr(htx, blk) + bs + 1) = ' ';
551 memcpy(htx_get_blk_ptr(htx, blk) + bs + 2, list[ck].v.ptr, vl);
552 bs += vl + 2;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100553 }
554
555 }
556
557 /* now send the end of headers marker */
Christopher Faulet5be651d2021-01-22 15:28:03 +0100558 if (!htx_add_endof(htx, HTX_BLK_EOH))
559 goto fail;
Willy Tarreau6deb4122018-11-27 15:34:18 +0100560
Amaury Denoyelle03534102021-07-07 10:49:28 +0200561 /* proceed to scheme-based normalization on target-URI */
562 if (fields & H2_PHDR_FND_SCHM)
563 http_scheme_based_normalize(htx);
564
Willy Tarreau6deb4122018-11-27 15:34:18 +0100565 ret = 1;
566 return ret;
567
568 fail:
569 return -1;
570}
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200571
572/* Prepare the status line into <htx> from pseudo headers stored in <phdr[]>.
573 * <fields> indicates what was found so far. This should be called once at the
574 * detection of the first general header field or at the end of the message if
575 * no general header field was found yet. Returns the created start line on
576 * success, or NULL on failure. Upon success, <msgf> is updated with a few
577 * H2_MSGF_* flags indicating what was found while parsing.
578 */
579static struct htx_sl *h2_prepare_htx_stsline(uint32_t fields, struct ist *phdr, struct htx *htx, unsigned int *msgf)
580{
Christopher Faulet89899422020-12-07 18:24:43 +0100581 unsigned int status, flags = HTX_SL_F_NONE;
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200582 struct htx_sl *sl;
Amaury Denoyelle74162742020-12-11 17:53:05 +0100583 struct ist stat;
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200584
585 /* only :status is allowed as a pseudo header */
586 if (!(fields & H2_PHDR_FND_STAT))
587 goto fail;
588
589 if (phdr[H2_PHDR_IDX_STAT].len != 3)
590 goto fail;
591
Amaury Denoyelle74162742020-12-11 17:53:05 +0100592 /* if Extended CONNECT is used, convert status code from 200 to htx 101
593 * following rfc 8441 */
594 if (unlikely(*msgf & H2_MSGF_EXT_CONNECT) &&
595 isteq(phdr[H2_PHDR_IDX_STAT], ist("200"))) {
596 stat = ist("101");
597 status = 101;
598 }
599 else {
600 unsigned char h, t, u;
601
602 stat = phdr[H2_PHDR_IDX_STAT];
603
604 h = stat.ptr[0] - '0';
605 t = stat.ptr[1] - '0';
606 u = stat.ptr[2] - '0';
607 if (h > 9 || t > 9 || u > 9)
608 goto fail;
609 status = h * 100 + t * 10 + u;
610 }
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200611
Christopher Faulet89899422020-12-07 18:24:43 +0100612 /* 101 responses are not supported in H2, so return a error.
613 * On 1xx responses there is no ES on the HEADERS frame but there is no
614 * body. So remove the flag H2_MSGF_BODY and add H2_MSGF_RSP_1XX to
615 * notify the decoder another HEADERS frame is expected.
Ilya Shipitsinacf84592021-02-06 22:29:08 +0500616 * 204/304 response have no body by definition. So remove the flag
Christopher Faulet7d247f02020-12-02 14:26:36 +0100617 * H2_MSGF_BODY and set H2_MSGF_BODYLESS_RSP.
Amaury Denoyelle74162742020-12-11 17:53:05 +0100618 *
619 * Note however that there is a special condition for Extended CONNECT.
620 * In this case, we explicitly convert it to HTX 101 to mimic
621 * Get+Upgrade HTTP/1.1 mechanism
Christopher Faulet0b465482019-02-19 15:14:23 +0100622 */
Amaury Denoyelle74162742020-12-11 17:53:05 +0100623 if (status == 101) {
624 if (!(*msgf & H2_MSGF_EXT_CONNECT))
625 goto fail;
626 }
Christopher Faulet89899422020-12-07 18:24:43 +0100627 else if (status < 200) {
Christopher Faulet0b465482019-02-19 15:14:23 +0100628 *msgf |= H2_MSGF_RSP_1XX;
629 *msgf &= ~H2_MSGF_BODY;
630 }
Amaury Denoyelle74162742020-12-11 17:53:05 +0100631 else if (status == 204 || status == 304) {
Christopher Faulet7d247f02020-12-02 14:26:36 +0100632 *msgf &= ~H2_MSGF_BODY;
633 *msgf |= H2_MSGF_BODYLESS_RSP;
634 }
Christopher Faulet0b465482019-02-19 15:14:23 +0100635
Christopher Faulet89899422020-12-07 18:24:43 +0100636 /* Set HTX start-line flags */
637 flags |= HTX_SL_F_VER_11; // V2 in fact
638 flags |= HTX_SL_F_XFER_LEN; // xfer len always known with H2
639
Amaury Denoyelle74162742020-12-11 17:53:05 +0100640 sl = htx_add_stline(htx, HTX_BLK_RES_SL, flags, ist("HTTP/2.0"), stat, ist(""));
Christopher Faulet89899422020-12-07 18:24:43 +0100641 if (!sl)
642 goto fail;
643 sl->info.res.status = status;
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200644 return sl;
645 fail:
646 return NULL;
647}
648
649/* Takes an H2 response present in the headers list <list> terminated by a name
650 * being <NULL,0> and emits the equivalent HTX response according to the rules
651 * documented in RFC7540 #8.1.2. The output contents are emitted in <htx>, and
652 * a positive value is returned if some bytes were emitted. In case of error, a
653 * negative error code is returned.
654 *
655 * Upon success, <msgf> is filled with a few H2_MSGF_* flags indicating what
656 * was found while parsing. The caller must set it to zero in or H2_MSGF_BODY
657 * if a body is detected (!ES).
658 *
659 * The headers list <list> must be composed of :
660 * - n.name != NULL, n.len > 0 : literal header name
661 * - n.name == NULL, n.len > 0 : indexed pseudo header name number <n.len>
662 * among H2_PHDR_IDX_*
663 * - n.name ignored, n.len == 0 : end of list
664 * - in all cases except the end of list, v.name and v.len must designate a
665 * valid value.
Amaury Denoyelle74162742020-12-11 17:53:05 +0100666 *
667 * <upgrade_protocol> is only used if the htx status code is 101 indicating a
668 * response to an upgrade or h2-equivalent request.
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200669 */
Amaury Denoyelle74162742020-12-11 17:53:05 +0100670int h2_make_htx_response(struct http_hdr *list, struct htx *htx, unsigned int *msgf, unsigned long long *body_len, char *upgrade_protocol)
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200671{
672 struct ist phdr_val[H2_PHDR_NUM_ENTRIES];
673 uint32_t fields; /* bit mask of H2_PHDR_FND_* */
674 uint32_t idx;
675 int phdr;
676 int ret;
677 int i;
678 struct htx_sl *sl = NULL;
679 unsigned int sl_flags = 0;
Willy Tarreau54f53ef2019-11-22 16:02:43 +0100680 const char *ctl;
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200681
682 fields = 0;
683 for (idx = 0; list[idx].n.len != 0; idx++) {
684 if (!list[idx].n.ptr) {
685 /* this is an indexed pseudo-header */
686 phdr = list[idx].n.len;
687 }
688 else {
689 /* this can be any type of header */
Willy Tarreau146f53a2019-11-24 10:34:39 +0100690 /* RFC7540#8.1.2: upper case not allowed in header field names.
691 * #10.3: header names must be valid (i.e. match a token).
692 * For pseudo-headers we check from 2nd char and for other ones
693 * from the first char, because HTTP_IS_TOKEN() also excludes
694 * the colon.
695 */
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200696 phdr = h2_str_to_phdr(list[idx].n);
Willy Tarreau146f53a2019-11-24 10:34:39 +0100697
698 for (i = !!phdr; i < list[idx].n.len; i++)
699 if ((uint8_t)(list[idx].n.ptr[i] - 'A') < 'Z' - 'A' || !HTTP_IS_TOKEN(list[idx].n.ptr[i]))
700 goto fail;
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200701 }
702
Willy Tarreau54f53ef2019-11-22 16:02:43 +0100703 /* RFC7540#10.3: intermediaries forwarding to HTTP/1 must take care of
704 * rejecting NUL, CR and LF characters.
705 */
706 ctl = ist_find_ctl(list[idx].v);
707 if (unlikely(ctl) && has_forbidden_char(list[idx].v, ctl))
708 goto fail;
709
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200710 if (phdr > 0 && phdr < H2_PHDR_NUM_ENTRIES) {
711 /* insert a pseudo header by its index (in phdr) and value (in value) */
712 if (fields & ((1 << phdr) | H2_PHDR_FND_NONE)) {
713 if (fields & H2_PHDR_FND_NONE) {
714 /* pseudo header field after regular headers */
715 goto fail;
716 }
717 else {
718 /* repeated pseudo header field */
719 goto fail;
720 }
721 }
722 fields |= 1 << phdr;
723 phdr_val[phdr] = list[idx].v;
724 continue;
725 }
726 else if (phdr != 0) {
727 /* invalid pseudo header -- should never happen here */
728 goto fail;
729 }
730
731 /* regular header field in (name,value) */
732 if (!(fields & H2_PHDR_FND_NONE)) {
733 /* no more pseudo-headers, time to build the status line */
734 sl = h2_prepare_htx_stsline(fields, phdr_val, htx, msgf);
735 if (!sl)
736 goto fail;
737 fields |= H2_PHDR_FND_NONE;
738 }
739
Willy Tarreaubeefaee2018-12-19 13:08:08 +0100740 if (isteq(list[idx].n, ist("content-length"))) {
Willy Tarreau4790f7c2019-01-24 11:33:02 +0100741 ret = h2_parse_cont_len_header(msgf, &list[idx].v, body_len);
Willy Tarreaubeefaee2018-12-19 13:08:08 +0100742 if (ret < 0)
743 goto fail;
744
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200745 sl_flags |= HTX_SL_F_CLEN;
Willy Tarreaubeefaee2018-12-19 13:08:08 +0100746 if (ret == 0)
747 continue; // skip this duplicate
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200748 }
749
750 /* these ones are forbidden in responses (RFC7540#8.1.2.2) */
751 if (isteq(list[idx].n, ist("connection")) ||
752 isteq(list[idx].n, ist("proxy-connection")) ||
753 isteq(list[idx].n, ist("keep-alive")) ||
754 isteq(list[idx].n, ist("upgrade")) ||
755 isteq(list[idx].n, ist("transfer-encoding")))
756 goto fail;
757
758 if (!htx_add_header(htx, list[idx].n, list[idx].v))
759 goto fail;
760 }
761
762 /* RFC7540#8.1.2.1 mandates to reject request pseudo-headers */
763 if (fields & (H2_PHDR_FND_AUTH|H2_PHDR_FND_METH|H2_PHDR_FND_PATH|H2_PHDR_FND_SCHM))
764 goto fail;
765
766 /* Let's dump the request now if not yet emitted. */
767 if (!(fields & H2_PHDR_FND_NONE)) {
768 sl = h2_prepare_htx_stsline(fields, phdr_val, htx, msgf);
769 if (!sl)
770 goto fail;
Amaury Denoyelle74162742020-12-11 17:53:05 +0100771 }
772
773 if (sl->info.res.status == 101 && upgrade_protocol) {
774 if (!htx_add_header(htx, ist("connection"), ist("upgrade")))
775 goto fail;
776 if (!htx_add_header(htx, ist("upgrade"), ist(upgrade_protocol)))
777 goto fail;
778 sl_flags |= HTX_SL_F_CONN_UPG;
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200779 }
780
Amaury Denoyelle74162742020-12-11 17:53:05 +0100781 if ((*msgf & H2_MSGF_BODY_TUNNEL) &&
782 ((sl->info.res.status >= 200 && sl->info.res.status < 300) || sl->info.res.status == 101))
Christopher Fauletd0db4232021-01-22 11:46:30 +0100783 *msgf &= ~(H2_MSGF_BODY|H2_MSGF_BODY_CL);
784 else
785 *msgf &= ~H2_MSGF_BODY_TUNNEL;
786
Christopher Fauletd1ac2b92020-12-02 19:12:22 +0100787 if (!(*msgf & H2_MSGF_BODY) || ((*msgf & H2_MSGF_BODY_CL) && *body_len == 0) ||
788 (*msgf & H2_MSGF_BODY_TUNNEL)) {
Ilya Shipitsinacf84592021-02-06 22:29:08 +0500789 /* Response without body or tunnel successfully established */
Christopher Faulet44af3cf2019-02-18 10:12:56 +0100790 sl_flags |= HTX_SL_F_BODYLESS;
Christopher Fauletd1ac2b92020-12-02 19:12:22 +0100791 htx->flags |= HTX_FL_EOM;
792 }
Christopher Faulet44af3cf2019-02-18 10:12:56 +0100793
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200794 /* update the start line with last detected header info */
795 sl->flags |= sl_flags;
796
797 if ((*msgf & (H2_MSGF_BODY|H2_MSGF_BODY_TUNNEL|H2_MSGF_BODY_CL)) == H2_MSGF_BODY) {
798 /* FIXME: Do we need to signal anything when we have a body and
799 * no content-length, to have the equivalent of H1's chunked
800 * encoding?
801 */
802 }
803
804 /* now send the end of headers marker */
Christopher Faulet5be651d2021-01-22 15:28:03 +0100805 if (!htx_add_endof(htx, HTX_BLK_EOH))
806 goto fail;
Willy Tarreau1329b5b2018-10-08 14:49:20 +0200807
808 ret = 1;
809 return ret;
810
811 fail:
812 return -1;
813}
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100814
Christopher Faulet2d7c5392019-06-03 10:41:26 +0200815/* Takes an H2 headers list <list> terminated by a name being <NULL,0> and emits
816 * the equivalent HTX trailers blocks. The output contents are emitted in <htx>,
817 * and a positive value is returned if some bytes were emitted. In case of
818 * error, a negative error code is returned. The caller must have verified that
819 * the message in the buffer is compatible with receipt of trailers.
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100820 *
821 * The headers list <list> must be composed of :
822 * - n.name != NULL, n.len > 0 : literal header name
823 * - n.name == NULL, n.len > 0 : indexed pseudo header name number <n.len>
824 * among H2_PHDR_IDX_* (illegal here)
825 * - n.name ignored, n.len == 0 : end of list
826 * - in all cases except the end of list, v.name and v.len must designate a
827 * valid value.
828 */
829int h2_make_htx_trailers(struct http_hdr *list, struct htx *htx)
830{
Willy Tarreau54f53ef2019-11-22 16:02:43 +0100831 const char *ctl;
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100832 uint32_t idx;
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100833 int i;
834
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100835 for (idx = 0; list[idx].n.len != 0; idx++) {
836 if (!list[idx].n.ptr) {
837 /* This is an indexed pseudo-header (RFC7540#8.1.2.1) */
838 goto fail;
839 }
840
Willy Tarreau146f53a2019-11-24 10:34:39 +0100841 /* RFC7540#8.1.2: upper case not allowed in header field names.
842 * #10.3: header names must be valid (i.e. match a token). This
843 * also catches pseudo-headers which are forbidden in trailers.
844 */
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100845 for (i = 0; i < list[idx].n.len; i++)
Willy Tarreau146f53a2019-11-24 10:34:39 +0100846 if ((uint8_t)(list[idx].n.ptr[i] - 'A') < 'Z' - 'A' || !HTTP_IS_TOKEN(list[idx].n.ptr[i]))
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100847 goto fail;
848
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100849 /* these ones are forbidden in trailers (RFC7540#8.1.2.2) */
850 if (isteq(list[idx].n, ist("host")) ||
851 isteq(list[idx].n, ist("content-length")) ||
852 isteq(list[idx].n, ist("connection")) ||
853 isteq(list[idx].n, ist("proxy-connection")) ||
854 isteq(list[idx].n, ist("keep-alive")) ||
855 isteq(list[idx].n, ist("upgrade")) ||
856 isteq(list[idx].n, ist("te")) ||
857 isteq(list[idx].n, ist("transfer-encoding")))
858 goto fail;
859
Willy Tarreau54f53ef2019-11-22 16:02:43 +0100860 /* RFC7540#10.3: intermediaries forwarding to HTTP/1 must take care of
861 * rejecting NUL, CR and LF characters.
862 */
863 ctl = ist_find_ctl(list[idx].v);
864 if (unlikely(ctl) && has_forbidden_char(list[idx].v, ctl))
865 goto fail;
866
Christopher Faulet2d7c5392019-06-03 10:41:26 +0200867 if (!htx_add_trailer(htx, list[idx].n, list[idx].v))
868 goto fail;
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100869 }
870
Christopher Faulet2d7c5392019-06-03 10:41:26 +0200871 if (!htx_add_endof(htx, HTX_BLK_EOT))
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100872 goto fail;
873
Willy Tarreau1e1f27c2019-01-03 18:39:54 +0100874 return 1;
875
876 fail:
877 return -1;
878}