Willy Tarreau | 0da5b3b | 2017-09-21 09:30:46 +0200 | [diff] [blame] | 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 | |
Willy Tarreau | 794f9af | 2017-07-26 09:07:47 +0200 | [diff] [blame] | 13 | #include <ctype.h> |
Willy Tarreau | 0da5b3b | 2017-09-21 09:30:46 +0200 | [diff] [blame] | 14 | #include <common/config.h> |
Willy Tarreau | 794f9af | 2017-07-26 09:07:47 +0200 | [diff] [blame] | 15 | #include <common/http-hdr.h> |
Willy Tarreau | 0da5b3b | 2017-09-21 09:30:46 +0200 | [diff] [blame] | 16 | |
| 17 | #include <proto/h1.h> |
Willy Tarreau | 8740c8b | 2017-09-21 10:22:25 +0200 | [diff] [blame] | 18 | #include <proto/hdr_idx.h> |
Willy Tarreau | 0da5b3b | 2017-09-21 09:30:46 +0200 | [diff] [blame] | 19 | |
| 20 | /* It is about twice as fast on recent architectures to lookup a byte in a |
| 21 | * table than to perform a boolean AND or OR between two tests. Refer to |
| 22 | * RFC2616/RFC5234/RFC7230 for those chars. A token is any ASCII char that is |
| 23 | * neither a separator nor a CTL char. An http ver_token is any ASCII which can |
| 24 | * be found in an HTTP version, which includes 'H', 'T', 'P', '/', '.' and any |
| 25 | * digit. Note: please do not overwrite values in assignment since gcc-2.95 |
| 26 | * will not handle them correctly. It's worth noting that chars 128..255 are |
| 27 | * nothing, not even control chars. |
| 28 | */ |
| 29 | const unsigned char h1_char_classes[256] = { |
| 30 | [ 0] = H1_FLG_CTL, |
| 31 | [ 1] = H1_FLG_CTL, |
| 32 | [ 2] = H1_FLG_CTL, |
| 33 | [ 3] = H1_FLG_CTL, |
| 34 | [ 4] = H1_FLG_CTL, |
| 35 | [ 5] = H1_FLG_CTL, |
| 36 | [ 6] = H1_FLG_CTL, |
| 37 | [ 7] = H1_FLG_CTL, |
| 38 | [ 8] = H1_FLG_CTL, |
| 39 | [ 9] = H1_FLG_SPHT | H1_FLG_LWS | H1_FLG_SEP | H1_FLG_CTL, |
| 40 | [ 10] = H1_FLG_CRLF | H1_FLG_LWS | H1_FLG_CTL, |
| 41 | [ 11] = H1_FLG_CTL, |
| 42 | [ 12] = H1_FLG_CTL, |
| 43 | [ 13] = H1_FLG_CRLF | H1_FLG_LWS | H1_FLG_CTL, |
| 44 | [ 14] = H1_FLG_CTL, |
| 45 | [ 15] = H1_FLG_CTL, |
| 46 | [ 16] = H1_FLG_CTL, |
| 47 | [ 17] = H1_FLG_CTL, |
| 48 | [ 18] = H1_FLG_CTL, |
| 49 | [ 19] = H1_FLG_CTL, |
| 50 | [ 20] = H1_FLG_CTL, |
| 51 | [ 21] = H1_FLG_CTL, |
| 52 | [ 22] = H1_FLG_CTL, |
| 53 | [ 23] = H1_FLG_CTL, |
| 54 | [ 24] = H1_FLG_CTL, |
| 55 | [ 25] = H1_FLG_CTL, |
| 56 | [ 26] = H1_FLG_CTL, |
| 57 | [ 27] = H1_FLG_CTL, |
| 58 | [ 28] = H1_FLG_CTL, |
| 59 | [ 29] = H1_FLG_CTL, |
| 60 | [ 30] = H1_FLG_CTL, |
| 61 | [ 31] = H1_FLG_CTL, |
| 62 | [' '] = H1_FLG_SPHT | H1_FLG_LWS | H1_FLG_SEP, |
| 63 | ['!'] = H1_FLG_TOK, |
| 64 | ['"'] = H1_FLG_SEP, |
| 65 | ['#'] = H1_FLG_TOK, |
| 66 | ['$'] = H1_FLG_TOK, |
| 67 | ['%'] = H1_FLG_TOK, |
| 68 | ['&'] = H1_FLG_TOK, |
| 69 | [ 39] = H1_FLG_TOK, |
| 70 | ['('] = H1_FLG_SEP, |
| 71 | [')'] = H1_FLG_SEP, |
| 72 | ['*'] = H1_FLG_TOK, |
| 73 | ['+'] = H1_FLG_TOK, |
| 74 | [','] = H1_FLG_SEP, |
| 75 | ['-'] = H1_FLG_TOK, |
| 76 | ['.'] = H1_FLG_TOK | H1_FLG_VER, |
| 77 | ['/'] = H1_FLG_SEP | H1_FLG_VER, |
| 78 | ['0'] = H1_FLG_TOK | H1_FLG_VER, |
| 79 | ['1'] = H1_FLG_TOK | H1_FLG_VER, |
| 80 | ['2'] = H1_FLG_TOK | H1_FLG_VER, |
| 81 | ['3'] = H1_FLG_TOK | H1_FLG_VER, |
| 82 | ['4'] = H1_FLG_TOK | H1_FLG_VER, |
| 83 | ['5'] = H1_FLG_TOK | H1_FLG_VER, |
| 84 | ['6'] = H1_FLG_TOK | H1_FLG_VER, |
| 85 | ['7'] = H1_FLG_TOK | H1_FLG_VER, |
| 86 | ['8'] = H1_FLG_TOK | H1_FLG_VER, |
| 87 | ['9'] = H1_FLG_TOK | H1_FLG_VER, |
| 88 | [':'] = H1_FLG_SEP, |
| 89 | [';'] = H1_FLG_SEP, |
| 90 | ['<'] = H1_FLG_SEP, |
| 91 | ['='] = H1_FLG_SEP, |
| 92 | ['>'] = H1_FLG_SEP, |
| 93 | ['?'] = H1_FLG_SEP, |
| 94 | ['@'] = H1_FLG_SEP, |
| 95 | ['A'] = H1_FLG_TOK, |
| 96 | ['B'] = H1_FLG_TOK, |
| 97 | ['C'] = H1_FLG_TOK, |
| 98 | ['D'] = H1_FLG_TOK, |
| 99 | ['E'] = H1_FLG_TOK, |
| 100 | ['F'] = H1_FLG_TOK, |
| 101 | ['G'] = H1_FLG_TOK, |
| 102 | ['H'] = H1_FLG_TOK | H1_FLG_VER, |
| 103 | ['I'] = H1_FLG_TOK, |
| 104 | ['J'] = H1_FLG_TOK, |
| 105 | ['K'] = H1_FLG_TOK, |
| 106 | ['L'] = H1_FLG_TOK, |
| 107 | ['M'] = H1_FLG_TOK, |
| 108 | ['N'] = H1_FLG_TOK, |
| 109 | ['O'] = H1_FLG_TOK, |
| 110 | ['P'] = H1_FLG_TOK | H1_FLG_VER, |
| 111 | ['Q'] = H1_FLG_TOK, |
| 112 | ['R'] = H1_FLG_TOK | H1_FLG_VER, |
| 113 | ['S'] = H1_FLG_TOK | H1_FLG_VER, |
| 114 | ['T'] = H1_FLG_TOK | H1_FLG_VER, |
| 115 | ['U'] = H1_FLG_TOK, |
| 116 | ['V'] = H1_FLG_TOK, |
| 117 | ['W'] = H1_FLG_TOK, |
| 118 | ['X'] = H1_FLG_TOK, |
| 119 | ['Y'] = H1_FLG_TOK, |
| 120 | ['Z'] = H1_FLG_TOK, |
| 121 | ['['] = H1_FLG_SEP, |
| 122 | [ 92] = H1_FLG_SEP, |
| 123 | [']'] = H1_FLG_SEP, |
| 124 | ['^'] = H1_FLG_TOK, |
| 125 | ['_'] = H1_FLG_TOK, |
| 126 | ['`'] = H1_FLG_TOK, |
| 127 | ['a'] = H1_FLG_TOK, |
| 128 | ['b'] = H1_FLG_TOK, |
| 129 | ['c'] = H1_FLG_TOK, |
| 130 | ['d'] = H1_FLG_TOK, |
| 131 | ['e'] = H1_FLG_TOK, |
| 132 | ['f'] = H1_FLG_TOK, |
| 133 | ['g'] = H1_FLG_TOK, |
| 134 | ['h'] = H1_FLG_TOK, |
| 135 | ['i'] = H1_FLG_TOK, |
| 136 | ['j'] = H1_FLG_TOK, |
| 137 | ['k'] = H1_FLG_TOK, |
| 138 | ['l'] = H1_FLG_TOK, |
| 139 | ['m'] = H1_FLG_TOK, |
| 140 | ['n'] = H1_FLG_TOK, |
| 141 | ['o'] = H1_FLG_TOK, |
| 142 | ['p'] = H1_FLG_TOK, |
| 143 | ['q'] = H1_FLG_TOK, |
| 144 | ['r'] = H1_FLG_TOK, |
| 145 | ['s'] = H1_FLG_TOK, |
| 146 | ['t'] = H1_FLG_TOK, |
| 147 | ['u'] = H1_FLG_TOK, |
| 148 | ['v'] = H1_FLG_TOK, |
| 149 | ['w'] = H1_FLG_TOK, |
| 150 | ['x'] = H1_FLG_TOK, |
| 151 | ['y'] = H1_FLG_TOK, |
| 152 | ['z'] = H1_FLG_TOK, |
| 153 | ['{'] = H1_FLG_SEP, |
| 154 | ['|'] = H1_FLG_TOK, |
| 155 | ['}'] = H1_FLG_SEP, |
| 156 | ['~'] = H1_FLG_TOK, |
| 157 | [127] = H1_FLG_CTL, |
| 158 | }; |
Willy Tarreau | db4893d | 2017-09-21 08:40:02 +0200 | [diff] [blame] | 159 | |
| 160 | |
Willy Tarreau | 8740c8b | 2017-09-21 10:22:25 +0200 | [diff] [blame] | 161 | /* |
| 162 | * This function parses a status line between <ptr> and <end>, starting with |
| 163 | * parser state <state>. Only states HTTP_MSG_RPVER, HTTP_MSG_RPVER_SP, |
| 164 | * HTTP_MSG_RPCODE, HTTP_MSG_RPCODE_SP and HTTP_MSG_RPREASON are handled. Others |
| 165 | * will give undefined results. |
| 166 | * Note that it is upon the caller's responsibility to ensure that ptr < end, |
| 167 | * and that msg->sol points to the beginning of the response. |
| 168 | * If a complete line is found (which implies that at least one CR or LF is |
| 169 | * found before <end>, the updated <ptr> is returned, otherwise NULL is |
| 170 | * returned indicating an incomplete line (which does not mean that parts have |
| 171 | * not been updated). In the incomplete case, if <ret_ptr> or <ret_state> are |
| 172 | * non-NULL, they are fed with the new <ptr> and <state> values to be passed |
| 173 | * upon next call. |
| 174 | * |
| 175 | * This function was intentionally designed to be called from |
| 176 | * http_msg_analyzer() with the lowest overhead. It should integrate perfectly |
| 177 | * within its state machine and use the same macros, hence the need for same |
| 178 | * labels and variable names. Note that msg->sol is left unchanged. |
| 179 | */ |
| 180 | const char *http_parse_stsline(struct http_msg *msg, |
| 181 | enum h1_state state, const char *ptr, const char *end, |
| 182 | unsigned int *ret_ptr, enum h1_state *ret_state) |
| 183 | { |
| 184 | const char *msg_start = msg->chn->buf->p; |
| 185 | |
| 186 | switch (state) { |
| 187 | case HTTP_MSG_RPVER: |
| 188 | http_msg_rpver: |
| 189 | if (likely(HTTP_IS_VER_TOKEN(*ptr))) |
| 190 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver, http_msg_ood, state, HTTP_MSG_RPVER); |
| 191 | |
| 192 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 193 | msg->sl.st.v_l = ptr - msg_start; |
| 194 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, HTTP_MSG_RPVER_SP); |
| 195 | } |
| 196 | msg->err_state = HTTP_MSG_RPVER; |
| 197 | state = HTTP_MSG_ERROR; |
| 198 | break; |
| 199 | |
| 200 | case HTTP_MSG_RPVER_SP: |
| 201 | http_msg_rpver_sp: |
| 202 | if (likely(!HTTP_IS_LWS(*ptr))) { |
| 203 | msg->sl.st.c = ptr - msg_start; |
| 204 | goto http_msg_rpcode; |
| 205 | } |
| 206 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 207 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, HTTP_MSG_RPVER_SP); |
| 208 | /* so it's a CR/LF, this is invalid */ |
| 209 | msg->err_state = HTTP_MSG_RPVER_SP; |
| 210 | state = HTTP_MSG_ERROR; |
| 211 | break; |
| 212 | |
| 213 | case HTTP_MSG_RPCODE: |
| 214 | http_msg_rpcode: |
| 215 | if (likely(!HTTP_IS_LWS(*ptr))) |
| 216 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode, http_msg_ood, state, HTTP_MSG_RPCODE); |
| 217 | |
| 218 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 219 | msg->sl.st.c_l = ptr - msg_start - msg->sl.st.c; |
| 220 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, HTTP_MSG_RPCODE_SP); |
| 221 | } |
| 222 | |
| 223 | /* so it's a CR/LF, so there is no reason phrase */ |
| 224 | msg->sl.st.c_l = ptr - msg_start - msg->sl.st.c; |
| 225 | http_msg_rsp_reason: |
| 226 | /* FIXME: should we support HTTP responses without any reason phrase ? */ |
| 227 | msg->sl.st.r = ptr - msg_start; |
| 228 | msg->sl.st.r_l = 0; |
| 229 | goto http_msg_rpline_eol; |
| 230 | |
| 231 | case HTTP_MSG_RPCODE_SP: |
| 232 | http_msg_rpcode_sp: |
| 233 | if (likely(!HTTP_IS_LWS(*ptr))) { |
| 234 | msg->sl.st.r = ptr - msg_start; |
| 235 | goto http_msg_rpreason; |
| 236 | } |
| 237 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 238 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, HTTP_MSG_RPCODE_SP); |
| 239 | /* so it's a CR/LF, so there is no reason phrase */ |
| 240 | goto http_msg_rsp_reason; |
| 241 | |
| 242 | case HTTP_MSG_RPREASON: |
| 243 | http_msg_rpreason: |
| 244 | if (likely(!HTTP_IS_CRLF(*ptr))) |
| 245 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpreason, http_msg_ood, state, HTTP_MSG_RPREASON); |
| 246 | msg->sl.st.r_l = ptr - msg_start - msg->sl.st.r; |
| 247 | http_msg_rpline_eol: |
| 248 | /* We have seen the end of line. Note that we do not |
| 249 | * necessarily have the \n yet, but at least we know that we |
| 250 | * have EITHER \r OR \n, otherwise the response would not be |
| 251 | * complete. We can then record the response length and return |
| 252 | * to the caller which will be able to register it. |
| 253 | */ |
| 254 | msg->sl.st.l = ptr - msg_start - msg->sol; |
| 255 | return ptr; |
| 256 | |
| 257 | default: |
| 258 | #ifdef DEBUG_FULL |
| 259 | fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state); |
| 260 | exit(1); |
| 261 | #endif |
| 262 | ; |
| 263 | } |
| 264 | |
| 265 | http_msg_ood: |
| 266 | /* out of valid data */ |
| 267 | if (ret_state) |
| 268 | *ret_state = state; |
| 269 | if (ret_ptr) |
| 270 | *ret_ptr = ptr - msg_start; |
| 271 | return NULL; |
| 272 | } |
| 273 | |
| 274 | /* |
| 275 | * This function parses a request line between <ptr> and <end>, starting with |
| 276 | * parser state <state>. Only states HTTP_MSG_RQMETH, HTTP_MSG_RQMETH_SP, |
| 277 | * HTTP_MSG_RQURI, HTTP_MSG_RQURI_SP and HTTP_MSG_RQVER are handled. Others |
| 278 | * will give undefined results. |
| 279 | * Note that it is upon the caller's responsibility to ensure that ptr < end, |
| 280 | * and that msg->sol points to the beginning of the request. |
| 281 | * If a complete line is found (which implies that at least one CR or LF is |
| 282 | * found before <end>, the updated <ptr> is returned, otherwise NULL is |
| 283 | * returned indicating an incomplete line (which does not mean that parts have |
| 284 | * not been updated). In the incomplete case, if <ret_ptr> or <ret_state> are |
| 285 | * non-NULL, they are fed with the new <ptr> and <state> values to be passed |
| 286 | * upon next call. |
| 287 | * |
| 288 | * This function was intentionally designed to be called from |
| 289 | * http_msg_analyzer() with the lowest overhead. It should integrate perfectly |
| 290 | * within its state machine and use the same macros, hence the need for same |
| 291 | * labels and variable names. Note that msg->sol is left unchanged. |
| 292 | */ |
| 293 | const char *http_parse_reqline(struct http_msg *msg, |
| 294 | enum h1_state state, const char *ptr, const char *end, |
| 295 | unsigned int *ret_ptr, enum h1_state *ret_state) |
| 296 | { |
| 297 | const char *msg_start = msg->chn->buf->p; |
| 298 | |
| 299 | switch (state) { |
| 300 | case HTTP_MSG_RQMETH: |
| 301 | http_msg_rqmeth: |
| 302 | if (likely(HTTP_IS_TOKEN(*ptr))) |
| 303 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth, http_msg_ood, state, HTTP_MSG_RQMETH); |
| 304 | |
| 305 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 306 | msg->sl.rq.m_l = ptr - msg_start; |
| 307 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth_sp, http_msg_ood, state, HTTP_MSG_RQMETH_SP); |
| 308 | } |
| 309 | |
| 310 | if (likely(HTTP_IS_CRLF(*ptr))) { |
| 311 | /* HTTP 0.9 request */ |
| 312 | msg->sl.rq.m_l = ptr - msg_start; |
| 313 | http_msg_req09_uri: |
| 314 | msg->sl.rq.u = ptr - msg_start; |
| 315 | http_msg_req09_uri_e: |
| 316 | msg->sl.rq.u_l = ptr - msg_start - msg->sl.rq.u; |
| 317 | http_msg_req09_ver: |
| 318 | msg->sl.rq.v = ptr - msg_start; |
| 319 | msg->sl.rq.v_l = 0; |
| 320 | goto http_msg_rqline_eol; |
| 321 | } |
| 322 | msg->err_state = HTTP_MSG_RQMETH; |
| 323 | state = HTTP_MSG_ERROR; |
| 324 | break; |
| 325 | |
| 326 | case HTTP_MSG_RQMETH_SP: |
| 327 | http_msg_rqmeth_sp: |
| 328 | if (likely(!HTTP_IS_LWS(*ptr))) { |
| 329 | msg->sl.rq.u = ptr - msg_start; |
| 330 | goto http_msg_rquri; |
| 331 | } |
| 332 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 333 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqmeth_sp, http_msg_ood, state, HTTP_MSG_RQMETH_SP); |
| 334 | /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
| 335 | goto http_msg_req09_uri; |
| 336 | |
| 337 | case HTTP_MSG_RQURI: |
| 338 | http_msg_rquri: |
| 339 | #if defined(__x86_64__) || \ |
| 340 | defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) || \ |
| 341 | defined(__ARM_ARCH_7A__) |
| 342 | /* speedup: skip bytes not between 0x21 and 0x7e inclusive */ |
| 343 | while (ptr <= end - sizeof(int)) { |
| 344 | int x = *(int *)ptr - 0x21212121; |
| 345 | if (x & 0x80808080) |
| 346 | break; |
| 347 | |
| 348 | x -= 0x5e5e5e5e; |
| 349 | if (!(x & 0x80808080)) |
| 350 | break; |
| 351 | |
| 352 | ptr += sizeof(int); |
| 353 | } |
| 354 | #endif |
| 355 | if (ptr >= end) { |
| 356 | state = HTTP_MSG_RQURI; |
| 357 | goto http_msg_ood; |
| 358 | } |
| 359 | http_msg_rquri2: |
| 360 | if (likely((unsigned char)(*ptr - 33) <= 93)) /* 33 to 126 included */ |
| 361 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri2, http_msg_ood, state, HTTP_MSG_RQURI); |
| 362 | |
| 363 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 364 | msg->sl.rq.u_l = ptr - msg_start - msg->sl.rq.u; |
| 365 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri_sp, http_msg_ood, state, HTTP_MSG_RQURI_SP); |
| 366 | } |
| 367 | |
| 368 | if (likely((unsigned char)*ptr >= 128)) { |
| 369 | /* non-ASCII chars are forbidden unless option |
| 370 | * accept-invalid-http-request is enabled in the frontend. |
| 371 | * In any case, we capture the faulty char. |
| 372 | */ |
| 373 | if (msg->err_pos < -1) |
| 374 | goto invalid_char; |
| 375 | if (msg->err_pos == -1) |
| 376 | msg->err_pos = ptr - msg_start; |
| 377 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri, http_msg_ood, state, HTTP_MSG_RQURI); |
| 378 | } |
| 379 | |
| 380 | if (likely(HTTP_IS_CRLF(*ptr))) { |
| 381 | /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
| 382 | goto http_msg_req09_uri_e; |
| 383 | } |
| 384 | |
| 385 | /* OK forbidden chars, 0..31 or 127 */ |
| 386 | invalid_char: |
| 387 | msg->err_pos = ptr - msg_start; |
| 388 | msg->err_state = HTTP_MSG_RQURI; |
| 389 | state = HTTP_MSG_ERROR; |
| 390 | break; |
| 391 | |
| 392 | case HTTP_MSG_RQURI_SP: |
| 393 | http_msg_rquri_sp: |
| 394 | if (likely(!HTTP_IS_LWS(*ptr))) { |
| 395 | msg->sl.rq.v = ptr - msg_start; |
| 396 | goto http_msg_rqver; |
| 397 | } |
| 398 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 399 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rquri_sp, http_msg_ood, state, HTTP_MSG_RQURI_SP); |
| 400 | /* so it's a CR/LF, meaning an HTTP 0.9 request */ |
| 401 | goto http_msg_req09_ver; |
| 402 | |
| 403 | case HTTP_MSG_RQVER: |
| 404 | http_msg_rqver: |
| 405 | if (likely(HTTP_IS_VER_TOKEN(*ptr))) |
| 406 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqver, http_msg_ood, state, HTTP_MSG_RQVER); |
| 407 | |
| 408 | if (likely(HTTP_IS_CRLF(*ptr))) { |
| 409 | msg->sl.rq.v_l = ptr - msg_start - msg->sl.rq.v; |
| 410 | http_msg_rqline_eol: |
| 411 | /* We have seen the end of line. Note that we do not |
| 412 | * necessarily have the \n yet, but at least we know that we |
| 413 | * have EITHER \r OR \n, otherwise the request would not be |
| 414 | * complete. We can then record the request length and return |
| 415 | * to the caller which will be able to register it. |
| 416 | */ |
| 417 | msg->sl.rq.l = ptr - msg_start - msg->sol; |
| 418 | return ptr; |
| 419 | } |
| 420 | |
| 421 | /* neither an HTTP_VER token nor a CRLF */ |
| 422 | msg->err_state = HTTP_MSG_RQVER; |
| 423 | state = HTTP_MSG_ERROR; |
| 424 | break; |
| 425 | |
| 426 | default: |
| 427 | #ifdef DEBUG_FULL |
| 428 | fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state); |
| 429 | exit(1); |
| 430 | #endif |
| 431 | ; |
| 432 | } |
| 433 | |
| 434 | http_msg_ood: |
| 435 | /* out of valid data */ |
| 436 | if (ret_state) |
| 437 | *ret_state = state; |
| 438 | if (ret_ptr) |
| 439 | *ret_ptr = ptr - msg_start; |
| 440 | return NULL; |
| 441 | } |
| 442 | |
| 443 | /* |
| 444 | * This function parses an HTTP message, either a request or a response, |
| 445 | * depending on the initial msg->msg_state. The caller is responsible for |
| 446 | * ensuring that the message does not wrap. The function can be preempted |
| 447 | * everywhere when data are missing and recalled at the exact same location |
| 448 | * with no information loss. The message may even be realigned between two |
| 449 | * calls. The header index is re-initialized when switching from |
| 450 | * MSG_R[PQ]BEFORE to MSG_RPVER|MSG_RQMETH. It modifies msg->sol among other |
| 451 | * fields. Note that msg->sol will be initialized after completing the first |
| 452 | * state, so that none of the msg pointers has to be initialized prior to the |
| 453 | * first call. |
| 454 | */ |
| 455 | void http_msg_analyzer(struct http_msg *msg, struct hdr_idx *idx) |
| 456 | { |
| 457 | enum h1_state state; /* updated only when leaving the FSM */ |
| 458 | register char *ptr, *end; /* request pointers, to avoid dereferences */ |
| 459 | struct buffer *buf; |
| 460 | |
| 461 | state = msg->msg_state; |
| 462 | buf = msg->chn->buf; |
| 463 | ptr = buf->p + msg->next; |
| 464 | end = buf->p + buf->i; |
| 465 | |
| 466 | if (unlikely(ptr >= end)) |
| 467 | goto http_msg_ood; |
| 468 | |
| 469 | switch (state) { |
| 470 | /* |
| 471 | * First, states that are specific to the response only. |
| 472 | * We check them first so that request and headers are |
| 473 | * closer to each other (accessed more often). |
| 474 | */ |
| 475 | case HTTP_MSG_RPBEFORE: |
| 476 | http_msg_rpbefore: |
| 477 | if (likely(HTTP_IS_TOKEN(*ptr))) { |
| 478 | /* we have a start of message, but we have to check |
| 479 | * first if we need to remove some CRLF. We can only |
| 480 | * do this when o=0. |
| 481 | */ |
| 482 | if (unlikely(ptr != buf->p)) { |
| 483 | if (buf->o) |
| 484 | goto http_msg_ood; |
| 485 | /* Remove empty leading lines, as recommended by RFC2616. */ |
| 486 | bi_fast_delete(buf, ptr - buf->p); |
| 487 | } |
| 488 | msg->sol = 0; |
| 489 | msg->sl.st.l = 0; /* used in debug mode */ |
| 490 | hdr_idx_init(idx); |
| 491 | state = HTTP_MSG_RPVER; |
| 492 | goto http_msg_rpver; |
| 493 | } |
| 494 | |
| 495 | if (unlikely(!HTTP_IS_CRLF(*ptr))) { |
| 496 | state = HTTP_MSG_RPBEFORE; |
| 497 | goto http_msg_invalid; |
| 498 | } |
| 499 | |
| 500 | if (unlikely(*ptr == '\n')) |
| 501 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, HTTP_MSG_RPBEFORE); |
| 502 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore_cr, http_msg_ood, state, HTTP_MSG_RPBEFORE_CR); |
| 503 | /* stop here */ |
| 504 | |
| 505 | case HTTP_MSG_RPBEFORE_CR: |
| 506 | http_msg_rpbefore_cr: |
| 507 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_RPBEFORE_CR); |
| 508 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, HTTP_MSG_RPBEFORE); |
| 509 | /* stop here */ |
| 510 | |
| 511 | case HTTP_MSG_RPVER: |
| 512 | http_msg_rpver: |
| 513 | case HTTP_MSG_RPVER_SP: |
| 514 | case HTTP_MSG_RPCODE: |
| 515 | case HTTP_MSG_RPCODE_SP: |
| 516 | case HTTP_MSG_RPREASON: |
| 517 | ptr = (char *)http_parse_stsline(msg, |
| 518 | state, ptr, end, |
| 519 | &msg->next, &msg->msg_state); |
| 520 | if (unlikely(!ptr)) |
| 521 | return; |
| 522 | |
| 523 | /* we have a full response and we know that we have either a CR |
| 524 | * or an LF at <ptr>. |
| 525 | */ |
| 526 | hdr_idx_set_start(idx, msg->sl.st.l, *ptr == '\r'); |
| 527 | |
| 528 | msg->sol = ptr - buf->p; |
| 529 | if (likely(*ptr == '\r')) |
| 530 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpline_end, http_msg_ood, state, HTTP_MSG_RPLINE_END); |
| 531 | goto http_msg_rpline_end; |
| 532 | |
| 533 | case HTTP_MSG_RPLINE_END: |
| 534 | http_msg_rpline_end: |
| 535 | /* msg->sol must point to the first of CR or LF. */ |
| 536 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_RPLINE_END); |
| 537 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_first, http_msg_ood, state, HTTP_MSG_HDR_FIRST); |
| 538 | /* stop here */ |
| 539 | |
| 540 | /* |
| 541 | * Second, states that are specific to the request only |
| 542 | */ |
| 543 | case HTTP_MSG_RQBEFORE: |
| 544 | http_msg_rqbefore: |
| 545 | if (likely(HTTP_IS_TOKEN(*ptr))) { |
| 546 | /* we have a start of message, but we have to check |
| 547 | * first if we need to remove some CRLF. We can only |
| 548 | * do this when o=0. |
| 549 | */ |
| 550 | if (likely(ptr != buf->p)) { |
| 551 | if (buf->o) |
| 552 | goto http_msg_ood; |
| 553 | /* Remove empty leading lines, as recommended by RFC2616. */ |
| 554 | bi_fast_delete(buf, ptr - buf->p); |
| 555 | } |
| 556 | msg->sol = 0; |
| 557 | msg->sl.rq.l = 0; /* used in debug mode */ |
| 558 | state = HTTP_MSG_RQMETH; |
| 559 | goto http_msg_rqmeth; |
| 560 | } |
| 561 | |
| 562 | if (unlikely(!HTTP_IS_CRLF(*ptr))) { |
| 563 | state = HTTP_MSG_RQBEFORE; |
| 564 | goto http_msg_invalid; |
| 565 | } |
| 566 | |
| 567 | if (unlikely(*ptr == '\n')) |
| 568 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore, http_msg_ood, state, HTTP_MSG_RQBEFORE); |
| 569 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore_cr, http_msg_ood, state, HTTP_MSG_RQBEFORE_CR); |
| 570 | /* stop here */ |
| 571 | |
| 572 | case HTTP_MSG_RQBEFORE_CR: |
| 573 | http_msg_rqbefore_cr: |
| 574 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_RQBEFORE_CR); |
| 575 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqbefore, http_msg_ood, state, HTTP_MSG_RQBEFORE); |
| 576 | /* stop here */ |
| 577 | |
| 578 | case HTTP_MSG_RQMETH: |
| 579 | http_msg_rqmeth: |
| 580 | case HTTP_MSG_RQMETH_SP: |
| 581 | case HTTP_MSG_RQURI: |
| 582 | case HTTP_MSG_RQURI_SP: |
| 583 | case HTTP_MSG_RQVER: |
| 584 | ptr = (char *)http_parse_reqline(msg, |
| 585 | state, ptr, end, |
| 586 | &msg->next, &msg->msg_state); |
| 587 | if (unlikely(!ptr)) |
| 588 | return; |
| 589 | |
| 590 | /* we have a full request and we know that we have either a CR |
| 591 | * or an LF at <ptr>. |
| 592 | */ |
| 593 | hdr_idx_set_start(idx, msg->sl.rq.l, *ptr == '\r'); |
| 594 | |
| 595 | msg->sol = ptr - buf->p; |
| 596 | if (likely(*ptr == '\r')) |
| 597 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rqline_end, http_msg_ood, state, HTTP_MSG_RQLINE_END); |
| 598 | goto http_msg_rqline_end; |
| 599 | |
| 600 | case HTTP_MSG_RQLINE_END: |
| 601 | http_msg_rqline_end: |
| 602 | /* check for HTTP/0.9 request : no version information available. |
| 603 | * msg->sol must point to the first of CR or LF. |
| 604 | */ |
| 605 | if (unlikely(msg->sl.rq.v_l == 0)) |
| 606 | goto http_msg_last_lf; |
| 607 | |
| 608 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_RQLINE_END); |
| 609 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_first, http_msg_ood, state, HTTP_MSG_HDR_FIRST); |
| 610 | /* stop here */ |
| 611 | |
| 612 | /* |
| 613 | * Common states below |
| 614 | */ |
| 615 | case HTTP_MSG_HDR_FIRST: |
| 616 | http_msg_hdr_first: |
| 617 | msg->sol = ptr - buf->p; |
| 618 | if (likely(!HTTP_IS_CRLF(*ptr))) { |
| 619 | goto http_msg_hdr_name; |
| 620 | } |
| 621 | |
| 622 | if (likely(*ptr == '\r')) |
| 623 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, HTTP_MSG_LAST_LF); |
| 624 | goto http_msg_last_lf; |
| 625 | |
| 626 | case HTTP_MSG_HDR_NAME: |
| 627 | http_msg_hdr_name: |
| 628 | /* assumes msg->sol points to the first char */ |
| 629 | if (likely(HTTP_IS_TOKEN(*ptr))) |
| 630 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, HTTP_MSG_HDR_NAME); |
| 631 | |
| 632 | if (likely(*ptr == ':')) |
| 633 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, HTTP_MSG_HDR_L1_SP); |
| 634 | |
| 635 | if (likely(msg->err_pos < -1) || *ptr == '\n') { |
| 636 | state = HTTP_MSG_HDR_NAME; |
| 637 | goto http_msg_invalid; |
| 638 | } |
| 639 | |
| 640 | if (msg->err_pos == -1) /* capture error pointer */ |
| 641 | msg->err_pos = ptr - buf->p; /* >= 0 now */ |
| 642 | |
| 643 | /* and we still accept this non-token character */ |
| 644 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, HTTP_MSG_HDR_NAME); |
| 645 | |
| 646 | case HTTP_MSG_HDR_L1_SP: |
| 647 | http_msg_hdr_l1_sp: |
| 648 | /* assumes msg->sol points to the first char */ |
| 649 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 650 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, HTTP_MSG_HDR_L1_SP); |
| 651 | |
| 652 | /* header value can be basically anything except CR/LF */ |
| 653 | msg->sov = ptr - buf->p; |
| 654 | |
| 655 | if (likely(!HTTP_IS_CRLF(*ptr))) { |
| 656 | goto http_msg_hdr_val; |
| 657 | } |
| 658 | |
| 659 | if (likely(*ptr == '\r')) |
| 660 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lf, http_msg_ood, state, HTTP_MSG_HDR_L1_LF); |
| 661 | goto http_msg_hdr_l1_lf; |
| 662 | |
| 663 | case HTTP_MSG_HDR_L1_LF: |
| 664 | http_msg_hdr_l1_lf: |
| 665 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_HDR_L1_LF); |
| 666 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lws, http_msg_ood, state, HTTP_MSG_HDR_L1_LWS); |
| 667 | |
| 668 | case HTTP_MSG_HDR_L1_LWS: |
| 669 | http_msg_hdr_l1_lws: |
| 670 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 671 | /* replace HT,CR,LF with spaces */ |
| 672 | for (; buf->p + msg->sov < ptr; msg->sov++) |
| 673 | buf->p[msg->sov] = ' '; |
| 674 | goto http_msg_hdr_l1_sp; |
| 675 | } |
| 676 | /* we had a header consisting only in spaces ! */ |
| 677 | msg->eol = msg->sov; |
| 678 | goto http_msg_complete_header; |
| 679 | |
| 680 | case HTTP_MSG_HDR_VAL: |
| 681 | http_msg_hdr_val: |
| 682 | /* assumes msg->sol points to the first char, and msg->sov |
| 683 | * points to the first character of the value. |
| 684 | */ |
| 685 | |
| 686 | /* speedup: we'll skip packs of 4 or 8 bytes not containing bytes 0x0D |
| 687 | * and lower. In fact since most of the time is spent in the loop, we |
| 688 | * also remove the sign bit test so that bytes 0x8e..0x0d break the |
| 689 | * loop, but we don't care since they're very rare in header values. |
| 690 | */ |
| 691 | #if defined(__x86_64__) |
| 692 | while (ptr <= end - sizeof(long)) { |
| 693 | if ((*(long *)ptr - 0x0e0e0e0e0e0e0e0eULL) & 0x8080808080808080ULL) |
| 694 | goto http_msg_hdr_val2; |
| 695 | ptr += sizeof(long); |
| 696 | } |
| 697 | #endif |
| 698 | #if defined(__x86_64__) || \ |
| 699 | defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) || \ |
| 700 | defined(__ARM_ARCH_7A__) |
| 701 | while (ptr <= end - sizeof(int)) { |
| 702 | if ((*(int*)ptr - 0x0e0e0e0e) & 0x80808080) |
| 703 | goto http_msg_hdr_val2; |
| 704 | ptr += sizeof(int); |
| 705 | } |
| 706 | #endif |
| 707 | if (ptr >= end) { |
| 708 | state = HTTP_MSG_HDR_VAL; |
| 709 | goto http_msg_ood; |
| 710 | } |
| 711 | http_msg_hdr_val2: |
| 712 | if (likely(!HTTP_IS_CRLF(*ptr))) |
| 713 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_val2, http_msg_ood, state, HTTP_MSG_HDR_VAL); |
| 714 | |
| 715 | msg->eol = ptr - buf->p; |
| 716 | /* Note: we could also copy eol into ->eoh so that we have the |
| 717 | * real header end in case it ends with lots of LWS, but is this |
| 718 | * really needed ? |
| 719 | */ |
| 720 | if (likely(*ptr == '\r')) |
| 721 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lf, http_msg_ood, state, HTTP_MSG_HDR_L2_LF); |
| 722 | goto http_msg_hdr_l2_lf; |
| 723 | |
| 724 | case HTTP_MSG_HDR_L2_LF: |
| 725 | http_msg_hdr_l2_lf: |
| 726 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_HDR_L2_LF); |
| 727 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lws, http_msg_ood, state, HTTP_MSG_HDR_L2_LWS); |
| 728 | |
| 729 | case HTTP_MSG_HDR_L2_LWS: |
| 730 | http_msg_hdr_l2_lws: |
| 731 | if (unlikely(HTTP_IS_SPHT(*ptr))) { |
| 732 | /* LWS: replace HT,CR,LF with spaces */ |
| 733 | for (; buf->p + msg->eol < ptr; msg->eol++) |
| 734 | buf->p[msg->eol] = ' '; |
| 735 | goto http_msg_hdr_val; |
| 736 | } |
| 737 | http_msg_complete_header: |
| 738 | /* |
| 739 | * It was a new header, so the last one is finished. |
| 740 | * Assumes msg->sol points to the first char, msg->sov points |
| 741 | * to the first character of the value and msg->eol to the |
| 742 | * first CR or LF so we know how the line ends. We insert last |
| 743 | * header into the index. |
| 744 | */ |
| 745 | if (unlikely(hdr_idx_add(msg->eol - msg->sol, buf->p[msg->eol] == '\r', |
| 746 | idx, idx->tail) < 0)) { |
| 747 | state = HTTP_MSG_HDR_L2_LWS; |
| 748 | goto http_msg_invalid; |
| 749 | } |
| 750 | |
| 751 | msg->sol = ptr - buf->p; |
| 752 | if (likely(!HTTP_IS_CRLF(*ptr))) { |
| 753 | goto http_msg_hdr_name; |
| 754 | } |
| 755 | |
| 756 | if (likely(*ptr == '\r')) |
| 757 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, HTTP_MSG_LAST_LF); |
| 758 | goto http_msg_last_lf; |
| 759 | |
| 760 | case HTTP_MSG_LAST_LF: |
| 761 | http_msg_last_lf: |
| 762 | /* Assumes msg->sol points to the first of either CR or LF. |
| 763 | * Sets ->sov and ->next to the total header length, ->eoh to |
| 764 | * the last CRLF, and ->eol to the last CRLF length (1 or 2). |
| 765 | */ |
| 766 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_LAST_LF); |
| 767 | ptr++; |
| 768 | msg->sov = msg->next = ptr - buf->p; |
| 769 | msg->eoh = msg->sol; |
| 770 | msg->sol = 0; |
| 771 | msg->eol = msg->sov - msg->eoh; |
| 772 | msg->msg_state = HTTP_MSG_BODY; |
| 773 | return; |
| 774 | |
| 775 | case HTTP_MSG_ERROR: |
| 776 | /* this may only happen if we call http_msg_analyser() twice with an error */ |
| 777 | break; |
| 778 | |
| 779 | default: |
| 780 | #ifdef DEBUG_FULL |
| 781 | fprintf(stderr, "FIXME !!!! impossible state at %s:%d = %d\n", __FILE__, __LINE__, state); |
| 782 | exit(1); |
| 783 | #endif |
| 784 | ; |
| 785 | } |
| 786 | http_msg_ood: |
| 787 | /* out of data */ |
| 788 | msg->msg_state = state; |
| 789 | msg->next = ptr - buf->p; |
| 790 | return; |
| 791 | |
| 792 | http_msg_invalid: |
| 793 | /* invalid message */ |
| 794 | msg->err_state = state; |
| 795 | msg->msg_state = HTTP_MSG_ERROR; |
| 796 | msg->next = ptr - buf->p; |
| 797 | return; |
| 798 | } |
| 799 | |
Willy Tarreau | 794f9af | 2017-07-26 09:07:47 +0200 | [diff] [blame] | 800 | /* This function parses a contiguous HTTP/1 headers block starting at <start> |
| 801 | * and ending before <stop>, at once, and converts it a list of (name,value) |
| 802 | * pairs representing header fields into the array <hdr> of size <hdr_num>, |
| 803 | * whose last entry will have an empty name and an empty value. If <hdr_num> is |
| 804 | * too small to represent the whole message, an error is returned. If <h1m> is |
| 805 | * not NULL, some protocol elements such as content-length and transfer-encoding |
| 806 | * will be parsed and stored there as well. |
| 807 | * |
| 808 | * For now it's limited to the response. If the header block is incomplete, |
| 809 | * 0 is returned, waiting to be called again with more data to try it again. |
| 810 | * |
| 811 | * The code derived from the main HTTP/1 parser above but was simplified and |
| 812 | * optimized to process responses produced or forwarded by haproxy. The caller |
| 813 | * is responsible for ensuring that the message doesn't wrap, and should ensure |
| 814 | * it is complete to avoid having to retry the operation after a failed |
| 815 | * attempt. The message is not supposed to be invalid, which is why a few |
| 816 | * properties such as the character set used in the header field names are not |
| 817 | * checked. In case of an unparsable response message, a negative value will be |
| 818 | * returned with h1m->err_pos and h1m->err_state matching the location and |
| 819 | * state where the error was met. Leading blank likes are tolerated but not |
| 820 | * recommended. |
| 821 | * |
| 822 | * This function returns : |
| 823 | * -1 in case of error. In this case, h1m->err_state is filled (if h1m is |
| 824 | * set) with the state the error occurred in and h2-m>err_pos with the |
| 825 | * the position relative to <start> |
| 826 | * -2 if the output is full (hdr_num reached). err_state and err_pos also |
| 827 | * indicate where it failed. |
| 828 | * 0 in case of missing data. |
| 829 | * > 0 on success, it then corresponds to the number of bytes read since |
| 830 | * <start> so that the caller can go on with the payload. |
| 831 | */ |
| 832 | int h1_headers_to_hdr_list(char *start, const char *stop, |
| 833 | struct http_hdr *hdr, unsigned int hdr_num, |
| 834 | struct h1m *h1m) |
| 835 | { |
| 836 | enum h1_state state = HTTP_MSG_RPBEFORE; |
| 837 | register char *ptr = start; |
| 838 | register const char *end = stop; |
| 839 | unsigned int hdr_count = 0; |
| 840 | unsigned int code = 0; /* status code, ASCII form */ |
| 841 | unsigned int st_c; /* beginning of status code, relative to msg_start */ |
| 842 | unsigned int st_c_l; /* length of status code */ |
| 843 | unsigned int sol = 0; /* start of line */ |
| 844 | unsigned int col = 0; /* position of the colon */ |
| 845 | unsigned int eol = 0; /* end of line */ |
| 846 | unsigned int sov = 0; /* start of value */ |
| 847 | unsigned int skip = 0; /* number of bytes skipped at the beginning */ |
| 848 | struct ist n, v; /* header name and value during parsing */ |
| 849 | |
| 850 | if (unlikely(ptr >= end)) |
| 851 | goto http_msg_ood; |
| 852 | |
| 853 | switch (state) { |
| 854 | case HTTP_MSG_RPBEFORE: |
| 855 | http_msg_rpbefore: |
| 856 | if (likely(HTTP_IS_TOKEN(*ptr))) { |
| 857 | /* we have a start of message, we may have skipped some |
| 858 | * heading CRLF. Skip them now. |
| 859 | */ |
| 860 | skip += ptr - start; |
| 861 | start = ptr; |
| 862 | |
| 863 | sol = 0; |
| 864 | hdr_count = 0; |
| 865 | state = HTTP_MSG_RPVER; |
| 866 | goto http_msg_rpver; |
| 867 | } |
| 868 | |
| 869 | if (unlikely(!HTTP_IS_CRLF(*ptr))) { |
| 870 | state = HTTP_MSG_RPBEFORE; |
| 871 | goto http_msg_invalid; |
| 872 | } |
| 873 | |
| 874 | if (unlikely(*ptr == '\n')) |
| 875 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, HTTP_MSG_RPBEFORE); |
| 876 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore_cr, http_msg_ood, state, HTTP_MSG_RPBEFORE_CR); |
| 877 | /* stop here */ |
| 878 | |
| 879 | case HTTP_MSG_RPBEFORE_CR: |
| 880 | http_msg_rpbefore_cr: |
| 881 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_RPBEFORE_CR); |
| 882 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpbefore, http_msg_ood, state, HTTP_MSG_RPBEFORE); |
| 883 | /* stop here */ |
| 884 | |
| 885 | case HTTP_MSG_RPVER: |
| 886 | http_msg_rpver: |
| 887 | if (likely(HTTP_IS_VER_TOKEN(*ptr))) |
| 888 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver, http_msg_ood, state, HTTP_MSG_RPVER); |
| 889 | |
| 890 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 891 | /* version length = ptr - start */ |
| 892 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, HTTP_MSG_RPVER_SP); |
| 893 | } |
| 894 | state = HTTP_MSG_RPVER; |
| 895 | goto http_msg_invalid; |
| 896 | |
| 897 | case HTTP_MSG_RPVER_SP: |
| 898 | http_msg_rpver_sp: |
| 899 | if (likely(!HTTP_IS_LWS(*ptr))) { |
| 900 | code = 0; |
| 901 | st_c = ptr - start; |
| 902 | goto http_msg_rpcode; |
| 903 | } |
| 904 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 905 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpver_sp, http_msg_ood, state, HTTP_MSG_RPVER_SP); |
| 906 | /* so it's a CR/LF, this is invalid */ |
| 907 | state = HTTP_MSG_RPVER_SP; |
| 908 | goto http_msg_invalid; |
| 909 | |
| 910 | case HTTP_MSG_RPCODE: |
| 911 | http_msg_rpcode: |
| 912 | if (likely(!HTTP_IS_LWS(*ptr))) { |
Willy Tarreau | d22e83a | 2017-10-31 08:02:24 +0100 | [diff] [blame] | 913 | code = code * 10 + *ptr - '0'; |
Willy Tarreau | 794f9af | 2017-07-26 09:07:47 +0200 | [diff] [blame] | 914 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode, http_msg_ood, state, HTTP_MSG_RPCODE); |
| 915 | } |
| 916 | |
| 917 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 918 | st_c_l = ptr - start - st_c; |
| 919 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, HTTP_MSG_RPCODE_SP); |
| 920 | } |
| 921 | |
| 922 | /* so it's a CR/LF, so there is no reason phrase */ |
| 923 | st_c_l = ptr - start - st_c; |
| 924 | |
| 925 | http_msg_rsp_reason: |
| 926 | /* reason = ptr - start; */ |
| 927 | /* reason length = 0 */ |
| 928 | goto http_msg_rpline_eol; |
| 929 | |
| 930 | case HTTP_MSG_RPCODE_SP: |
| 931 | http_msg_rpcode_sp: |
| 932 | if (likely(!HTTP_IS_LWS(*ptr))) { |
| 933 | /* reason = ptr - start */ |
| 934 | goto http_msg_rpreason; |
| 935 | } |
| 936 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 937 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpcode_sp, http_msg_ood, state, HTTP_MSG_RPCODE_SP); |
| 938 | /* so it's a CR/LF, so there is no reason phrase */ |
| 939 | goto http_msg_rsp_reason; |
| 940 | |
| 941 | case HTTP_MSG_RPREASON: |
| 942 | http_msg_rpreason: |
| 943 | if (likely(!HTTP_IS_CRLF(*ptr))) |
| 944 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpreason, http_msg_ood, state, HTTP_MSG_RPREASON); |
| 945 | /* reason length = ptr - start - reason */ |
| 946 | http_msg_rpline_eol: |
| 947 | /* We have seen the end of line. Note that we do not |
| 948 | * necessarily have the \n yet, but at least we know that we |
| 949 | * have EITHER \r OR \n, otherwise the response would not be |
| 950 | * complete. We can then record the response length and return |
| 951 | * to the caller which will be able to register it. |
| 952 | */ |
| 953 | |
| 954 | if (unlikely(hdr_count >= hdr_num)) { |
| 955 | state = HTTP_MSG_RPREASON; |
| 956 | goto http_output_full; |
| 957 | } |
| 958 | http_set_hdr(&hdr[hdr_count++], ist(":status"), ist2(start + st_c, st_c_l)); |
Willy Tarreau | d22e83a | 2017-10-31 08:02:24 +0100 | [diff] [blame] | 959 | if (h1m) |
| 960 | h1m->status = code; |
Willy Tarreau | 794f9af | 2017-07-26 09:07:47 +0200 | [diff] [blame] | 961 | |
| 962 | sol = ptr - start; |
| 963 | if (likely(*ptr == '\r')) |
| 964 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_rpline_end, http_msg_ood, state, HTTP_MSG_RPLINE_END); |
| 965 | goto http_msg_rpline_end; |
| 966 | |
| 967 | case HTTP_MSG_RPLINE_END: |
| 968 | http_msg_rpline_end: |
| 969 | /* sol must point to the first of CR or LF. */ |
| 970 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_RPLINE_END); |
| 971 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_first, http_msg_ood, state, HTTP_MSG_HDR_FIRST); |
| 972 | /* stop here */ |
| 973 | |
| 974 | case HTTP_MSG_HDR_FIRST: |
| 975 | http_msg_hdr_first: |
| 976 | sol = ptr - start; |
| 977 | if (likely(!HTTP_IS_CRLF(*ptr))) { |
| 978 | goto http_msg_hdr_name; |
| 979 | } |
| 980 | |
| 981 | if (likely(*ptr == '\r')) |
| 982 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, HTTP_MSG_LAST_LF); |
| 983 | goto http_msg_last_lf; |
| 984 | |
| 985 | case HTTP_MSG_HDR_NAME: |
| 986 | http_msg_hdr_name: |
| 987 | /* assumes sol points to the first char */ |
| 988 | if (likely(HTTP_IS_TOKEN(*ptr))) { |
| 989 | /* turn it to lower case if needed */ |
| 990 | if (isupper((unsigned char)*ptr)) |
| 991 | *ptr = tolower(*ptr); |
| 992 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, HTTP_MSG_HDR_NAME); |
| 993 | } |
| 994 | |
| 995 | if (likely(*ptr == ':')) { |
| 996 | col = ptr - start; |
| 997 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, HTTP_MSG_HDR_L1_SP); |
| 998 | } |
| 999 | |
| 1000 | if (HTTP_IS_LWS(*ptr)) { |
| 1001 | state = HTTP_MSG_HDR_NAME; |
| 1002 | goto http_msg_invalid; |
| 1003 | } |
| 1004 | |
| 1005 | /* now we have a non-token character in the header field name, |
| 1006 | * it's up to the H1 layer to have decided whether or not it |
| 1007 | * was acceptable. If we find it here, it was considered |
| 1008 | * acceptable due to configuration rules so we obey. |
| 1009 | */ |
| 1010 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_name, http_msg_ood, state, HTTP_MSG_HDR_NAME); |
| 1011 | |
| 1012 | case HTTP_MSG_HDR_L1_SP: |
| 1013 | http_msg_hdr_l1_sp: |
| 1014 | /* assumes sol points to the first char */ |
| 1015 | if (likely(HTTP_IS_SPHT(*ptr))) |
| 1016 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_sp, http_msg_ood, state, HTTP_MSG_HDR_L1_SP); |
| 1017 | |
| 1018 | /* header value can be basically anything except CR/LF */ |
| 1019 | sov = ptr - start; |
| 1020 | |
| 1021 | if (likely(!HTTP_IS_CRLF(*ptr))) { |
| 1022 | goto http_msg_hdr_val; |
| 1023 | } |
| 1024 | |
| 1025 | if (likely(*ptr == '\r')) |
| 1026 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lf, http_msg_ood, state, HTTP_MSG_HDR_L1_LF); |
| 1027 | goto http_msg_hdr_l1_lf; |
| 1028 | |
| 1029 | case HTTP_MSG_HDR_L1_LF: |
| 1030 | http_msg_hdr_l1_lf: |
| 1031 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_HDR_L1_LF); |
| 1032 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l1_lws, http_msg_ood, state, HTTP_MSG_HDR_L1_LWS); |
| 1033 | |
| 1034 | case HTTP_MSG_HDR_L1_LWS: |
| 1035 | http_msg_hdr_l1_lws: |
| 1036 | if (likely(HTTP_IS_SPHT(*ptr))) { |
| 1037 | /* replace HT,CR,LF with spaces */ |
| 1038 | for (; start + sov < ptr; sov++) |
| 1039 | start[sov] = ' '; |
| 1040 | goto http_msg_hdr_l1_sp; |
| 1041 | } |
| 1042 | /* we had a header consisting only in spaces ! */ |
| 1043 | eol = sov; |
| 1044 | goto http_msg_complete_header; |
| 1045 | |
| 1046 | case HTTP_MSG_HDR_VAL: |
| 1047 | http_msg_hdr_val: |
| 1048 | /* assumes sol points to the first char, and sov |
| 1049 | * points to the first character of the value. |
| 1050 | */ |
| 1051 | |
| 1052 | /* speedup: we'll skip packs of 4 or 8 bytes not containing bytes 0x0D |
| 1053 | * and lower. In fact since most of the time is spent in the loop, we |
| 1054 | * also remove the sign bit test so that bytes 0x8e..0x0d break the |
| 1055 | * loop, but we don't care since they're very rare in header values. |
| 1056 | */ |
| 1057 | #if defined(__x86_64__) |
| 1058 | while (ptr <= end - sizeof(long)) { |
| 1059 | if ((*(long *)ptr - 0x0e0e0e0e0e0e0e0eULL) & 0x8080808080808080ULL) |
| 1060 | goto http_msg_hdr_val2; |
| 1061 | ptr += sizeof(long); |
| 1062 | } |
| 1063 | #endif |
| 1064 | #if defined(__x86_64__) || \ |
| 1065 | defined(__i386__) || defined(__i486__) || defined(__i586__) || defined(__i686__) || \ |
| 1066 | defined(__ARM_ARCH_7A__) |
| 1067 | while (ptr <= end - sizeof(int)) { |
| 1068 | if ((*(int*)ptr - 0x0e0e0e0e) & 0x80808080) |
| 1069 | goto http_msg_hdr_val2; |
| 1070 | ptr += sizeof(int); |
| 1071 | } |
| 1072 | #endif |
| 1073 | if (ptr >= end) { |
| 1074 | state = HTTP_MSG_HDR_VAL; |
| 1075 | goto http_msg_ood; |
| 1076 | } |
| 1077 | http_msg_hdr_val2: |
| 1078 | if (likely(!HTTP_IS_CRLF(*ptr))) |
| 1079 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_val2, http_msg_ood, state, HTTP_MSG_HDR_VAL); |
| 1080 | |
| 1081 | eol = ptr - start; |
| 1082 | /* Note: we could also copy eol into ->eoh so that we have the |
| 1083 | * real header end in case it ends with lots of LWS, but is this |
| 1084 | * really needed ? |
| 1085 | */ |
| 1086 | if (likely(*ptr == '\r')) |
| 1087 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lf, http_msg_ood, state, HTTP_MSG_HDR_L2_LF); |
| 1088 | goto http_msg_hdr_l2_lf; |
| 1089 | |
| 1090 | case HTTP_MSG_HDR_L2_LF: |
| 1091 | http_msg_hdr_l2_lf: |
| 1092 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_HDR_L2_LF); |
| 1093 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_hdr_l2_lws, http_msg_ood, state, HTTP_MSG_HDR_L2_LWS); |
| 1094 | |
| 1095 | case HTTP_MSG_HDR_L2_LWS: |
| 1096 | http_msg_hdr_l2_lws: |
| 1097 | if (unlikely(HTTP_IS_SPHT(*ptr))) { |
| 1098 | /* LWS: replace HT,CR,LF with spaces */ |
| 1099 | for (; start + eol < ptr; eol++) |
| 1100 | start[eol] = ' '; |
| 1101 | goto http_msg_hdr_val; |
| 1102 | } |
| 1103 | http_msg_complete_header: |
| 1104 | /* |
| 1105 | * It was a new header, so the last one is finished. Assumes |
| 1106 | * <sol> points to the first char of the name, <col> to the |
| 1107 | * colon, <sov> points to the first character of the value and |
| 1108 | * <eol> to the first CR or LF so we know how the line ends. We |
| 1109 | * will trim spaces around the value. It's possible to do it by |
| 1110 | * adjusting <eol> and <sov> which are no more used after this. |
| 1111 | * We can add the header field to the list. |
| 1112 | */ |
| 1113 | while (sov < eol && HTTP_IS_LWS(start[sov])) |
| 1114 | sov++; |
| 1115 | |
| 1116 | while (eol - 1 > sov && HTTP_IS_LWS(start[eol - 1])) |
| 1117 | eol--; |
| 1118 | |
| 1119 | |
| 1120 | n = ist2(start + sol, col - sol); |
| 1121 | v = ist2(start + sov, eol - sov); |
| 1122 | |
| 1123 | if (unlikely(hdr_count >= hdr_num)) { |
| 1124 | state = HTTP_MSG_HDR_L2_LWS; |
| 1125 | goto http_output_full; |
| 1126 | } |
| 1127 | http_set_hdr(&hdr[hdr_count++], n, v); |
| 1128 | |
| 1129 | if (h1m) { |
| 1130 | long long cl; |
| 1131 | |
Willy Tarreau | d22e83a | 2017-10-31 08:02:24 +0100 | [diff] [blame] | 1132 | if (h1m->status >= 100 && h1m->status < 200) |
| 1133 | h1m->curr_len = h1m->body_len = 0; |
| 1134 | else if (h1m->status == 304 || h1m->status == 204) { |
Willy Tarreau | 8ea0f38 | 2017-10-30 19:31:59 +0100 | [diff] [blame] | 1135 | /* no contents, claim c-len is present and set to zero */ |
| 1136 | h1m->flags |= H1_MF_CLEN; |
| 1137 | h1m->curr_len = h1m->body_len = 0; |
| 1138 | } |
| 1139 | else if (isteq(n, ist("transfer-encoding"))) { |
Willy Tarreau | 794f9af | 2017-07-26 09:07:47 +0200 | [diff] [blame] | 1140 | h1m->flags &= ~H1_MF_CLEN; |
| 1141 | h1m->flags |= H1_MF_CHNK; |
| 1142 | } |
| 1143 | else if (isteq(n, ist("content-length")) && !(h1m->flags & H1_MF_CHNK)) { |
| 1144 | h1m->flags |= H1_MF_CLEN; |
| 1145 | strl2llrc(v.ptr, v.len, &cl); |
| 1146 | h1m->curr_len = h1m->body_len = cl; |
| 1147 | } |
| 1148 | } |
| 1149 | |
| 1150 | sol = ptr - start; |
| 1151 | if (likely(!HTTP_IS_CRLF(*ptr))) |
| 1152 | goto http_msg_hdr_name; |
| 1153 | |
| 1154 | if (likely(*ptr == '\r')) |
| 1155 | EAT_AND_JUMP_OR_RETURN(ptr, end, http_msg_last_lf, http_msg_ood, state, HTTP_MSG_LAST_LF); |
| 1156 | goto http_msg_last_lf; |
| 1157 | |
| 1158 | case HTTP_MSG_LAST_LF: |
| 1159 | http_msg_last_lf: |
| 1160 | EXPECT_LF_HERE(ptr, http_msg_invalid, state, HTTP_MSG_LAST_LF); |
| 1161 | ptr++; |
| 1162 | /* <ptr> now points to the first byte of payload. If needed sol |
| 1163 | * still points to the first of either CR or LF of the empty |
| 1164 | * line ending the headers block. |
| 1165 | */ |
| 1166 | if (unlikely(hdr_count >= hdr_num)) { |
| 1167 | state = HTTP_MSG_LAST_LF; |
| 1168 | goto http_output_full; |
| 1169 | } |
| 1170 | http_set_hdr(&hdr[hdr_count++], ist(""), ist("")); |
| 1171 | state = HTTP_MSG_BODY; |
| 1172 | break; |
| 1173 | |
| 1174 | default: |
| 1175 | /* impossible states */ |
| 1176 | goto http_msg_invalid; |
| 1177 | } |
| 1178 | |
| 1179 | /* reaching here, we've parsed the whole message and the state is |
| 1180 | * HTTP_MSG_BODY. |
| 1181 | */ |
| 1182 | return ptr - start + skip; |
| 1183 | |
| 1184 | http_msg_ood: |
| 1185 | /* out of data at <ptr> during state <state> */ |
| 1186 | return 0; |
| 1187 | |
| 1188 | http_msg_invalid: |
| 1189 | /* invalid message, error at <ptr> */ |
| 1190 | if (h1m) { |
| 1191 | h1m->err_state = state; |
| 1192 | h1m->err_pos = ptr - start + skip; |
| 1193 | } |
| 1194 | return -1; |
| 1195 | |
| 1196 | http_output_full: |
| 1197 | /* no more room to store the current header, error at <ptr> */ |
| 1198 | if (h1m) { |
| 1199 | h1m->err_state = state; |
| 1200 | h1m->err_pos = ptr - start + skip; |
| 1201 | } |
| 1202 | return -2; |
| 1203 | } |
| 1204 | |
Willy Tarreau | db4893d | 2017-09-21 08:40:02 +0200 | [diff] [blame] | 1205 | /* This function skips trailers in the buffer associated with HTTP message |
| 1206 | * <msg>. The first visited position is msg->next. If the end of the trailers is |
| 1207 | * found, the function returns >0. So, the caller can automatically schedul it |
| 1208 | * to be forwarded, and switch msg->msg_state to HTTP_MSG_DONE. If not enough |
| 1209 | * data are available, the function does not change anything except maybe |
| 1210 | * msg->sol if it could parse some lines, and returns zero. If a parse error |
| 1211 | * is encountered, the function returns < 0 and does not change anything except |
| 1212 | * maybe msg->sol. Note that the message must already be in HTTP_MSG_TRAILERS |
| 1213 | * state before calling this function, which implies that all non-trailers data |
| 1214 | * have already been scheduled for forwarding, and that msg->next exactly |
| 1215 | * matches the length of trailers already parsed and not forwarded. It is also |
| 1216 | * important to note that this function is designed to be able to parse wrapped |
| 1217 | * headers at end of buffer. |
| 1218 | */ |
| 1219 | int http_forward_trailers(struct http_msg *msg) |
| 1220 | { |
| 1221 | const struct buffer *buf = msg->chn->buf; |
| 1222 | |
| 1223 | /* we have msg->next which points to next line. Look for CRLF. But |
| 1224 | * first, we reset msg->sol */ |
| 1225 | msg->sol = 0; |
| 1226 | while (1) { |
| 1227 | const char *p1 = NULL, *p2 = NULL; |
| 1228 | const char *start = b_ptr(buf, msg->next + msg->sol); |
| 1229 | const char *stop = bi_end(buf); |
| 1230 | const char *ptr = start; |
| 1231 | int bytes = 0; |
| 1232 | |
| 1233 | /* scan current line and stop at LF or CRLF */ |
| 1234 | while (1) { |
| 1235 | if (ptr == stop) |
| 1236 | return 0; |
| 1237 | |
| 1238 | if (*ptr == '\n') { |
| 1239 | if (!p1) |
| 1240 | p1 = ptr; |
| 1241 | p2 = ptr; |
| 1242 | break; |
| 1243 | } |
| 1244 | |
| 1245 | if (*ptr == '\r') { |
| 1246 | if (p1) { |
| 1247 | msg->err_pos = buffer_count(buf, buf->p, ptr); |
| 1248 | return -1; |
| 1249 | } |
| 1250 | p1 = ptr; |
| 1251 | } |
| 1252 | |
| 1253 | ptr++; |
| 1254 | if (ptr >= buf->data + buf->size) |
| 1255 | ptr = buf->data; |
| 1256 | } |
| 1257 | |
| 1258 | /* after LF; point to beginning of next line */ |
| 1259 | p2++; |
| 1260 | if (p2 >= buf->data + buf->size) |
| 1261 | p2 = buf->data; |
| 1262 | |
| 1263 | bytes = p2 - start; |
| 1264 | if (bytes < 0) |
| 1265 | bytes += buf->size; |
| 1266 | msg->sol += bytes; |
| 1267 | |
| 1268 | /* LF/CRLF at beginning of line => end of trailers at p2. |
| 1269 | * Everything was scheduled for forwarding, there's nothing left |
| 1270 | * from this message. */ |
| 1271 | if (p1 == start) |
| 1272 | return 1; |
| 1273 | |
| 1274 | /* OK, next line then */ |
| 1275 | } |
| 1276 | } |