1 /* 2 * request - HTTP request processing 3 * 4 * Fully-rewritten from original EXCEPT for request_check_hostname() 5 * Copyright(c) 2018 Glenn Strauss gstrauss()gluelogic.com All rights reserved 6 * License: BSD 3-clause (same as lighttpd) 7 */ 8 #include "first.h" 9 10 #include "request.h" 11 #include "burl.h" 12 #include "http_header.h" 13 #include "http_kv.h" 14 #include "log.h" 15 #include "sock_addr.h" 16 17 #include <limits.h> 18 #include <stdlib.h> 19 #include <string.h> 20 21 static int request_check_hostname(buffer * const host) { 22 enum { DOMAINLABEL, TOPLABEL } stage = TOPLABEL; 23 size_t i; 24 int label_len = 0; 25 size_t host_len, hostport_len; 26 char *colon; 27 int is_ip = -1; /* -1 don't know yet, 0 no, 1 yes */ 28 int level = 0; 29 30 /* 31 * hostport = host [ ":" port ] 32 * host = hostname | IPv4address | IPv6address 33 * hostname = *( domainlabel "." ) toplabel [ "." ] 34 * domainlabel = alphanum | alphanum *( alphanum | "-" ) alphanum 35 * toplabel = alpha | alpha *( alphanum | "-" ) alphanum 36 * IPv4address = 1*digit "." 1*digit "." 1*digit "." 1*digit 37 * IPv6address = "[" ... "]" 38 * port = *digit 39 */ 40 41 /* IPv6 address */ 42 if (host->ptr[0] == '[') { 43 char *c = host->ptr + 1; 44 int colon_cnt = 0; 45 46 /* check the address inside [...] */ 47 for (; *c && *c != ']'; c++) { 48 if (*c == ':') { 49 if (++colon_cnt > 7) { 50 return -1; 51 } 52 } else if (!light_isxdigit(*c) && '.' != *c) { 53 return -1; 54 } 55 } 56 57 /* missing ] */ 58 if (!*c) { 59 return -1; 60 } 61 62 /* check port */ 63 if (*(c+1) == ':') { 64 for (c += 2; *c; c++) { 65 if (!light_isdigit(*c)) { 66 return -1; 67 } 68 } 69 } 70 else if ('\0' != *(c+1)) { 71 /* only a port is allowed to follow [...] */ 72 return -1; 73 } 74 return 0; 75 } 76 77 hostport_len = host_len = buffer_string_length(host); 78 79 if (NULL != (colon = memchr(host->ptr, ':', host_len))) { 80 char *c = colon + 1; 81 82 /* check portnumber */ 83 for (; *c; c++) { 84 if (!light_isdigit(*c)) return -1; 85 } 86 87 /* remove the port from the host-len */ 88 host_len = colon - host->ptr; 89 } 90 91 /* Host is empty */ 92 if (host_len == 0) return -1; 93 94 /* if the hostname ends in a "." strip it */ 95 if (host->ptr[host_len-1] == '.') { 96 /* shift port info one left */ 97 if (NULL != colon) memmove(colon-1, colon, hostport_len - host_len); 98 buffer_string_set_length(host, --hostport_len); 99 if (--host_len == 0) return -1; 100 } 101 102 103 /* scan from the right and skip the \0 */ 104 for (i = host_len; i-- > 0; ) { 105 const char c = host->ptr[i]; 106 107 switch (stage) { 108 case TOPLABEL: 109 if (c == '.') { 110 /* only switch stage, if this is not the last character */ 111 if (i != host_len - 1) { 112 if (label_len == 0) { 113 return -1; 114 } 115 116 /* check the first character at right of the dot */ 117 if (is_ip == 0) { 118 if (!light_isalnum(host->ptr[i+1])) { 119 return -1; 120 } 121 } else if (!light_isdigit(host->ptr[i+1])) { 122 is_ip = 0; 123 } else if ('-' == host->ptr[i+1]) { 124 return -1; 125 } else { 126 /* just digits */ 127 is_ip = 1; 128 } 129 130 stage = DOMAINLABEL; 131 132 label_len = 0; 133 level++; 134 } else if (i == 0) { 135 /* just a dot and nothing else is evil */ 136 return -1; 137 } 138 } else if (i == 0) { 139 /* the first character of the hostname */ 140 if (!light_isalnum(c)) { 141 return -1; 142 } 143 label_len++; 144 } else { 145 if (c != '-' && !light_isalnum(c)) { 146 return -1; 147 } 148 if (is_ip == -1) { 149 if (!light_isdigit(c)) is_ip = 0; 150 } 151 label_len++; 152 } 153 154 break; 155 case DOMAINLABEL: 156 if (is_ip == 1) { 157 if (c == '.') { 158 if (label_len == 0) { 159 return -1; 160 } 161 162 label_len = 0; 163 level++; 164 } else if (!light_isdigit(c)) { 165 return -1; 166 } else { 167 label_len++; 168 } 169 } else { 170 if (c == '.') { 171 if (label_len == 0) { 172 return -1; 173 } 174 175 /* c is either - or alphanum here */ 176 if ('-' == host->ptr[i+1]) { 177 return -1; 178 } 179 180 label_len = 0; 181 level++; 182 } else if (i == 0) { 183 if (!light_isalnum(c)) { 184 return -1; 185 } 186 label_len++; 187 } else { 188 if (c != '-' && !light_isalnum(c)) { 189 return -1; 190 } 191 label_len++; 192 } 193 } 194 195 break; 196 } 197 } 198 199 /* a IP has to consist of 4 parts */ 200 if (is_ip == 1 && level != 3) { 201 return -1; 202 } 203 204 if (label_len == 0) { 205 return -1; 206 } 207 208 return 0; 209 } 210 211 int http_request_host_normalize(buffer * const b, const int scheme_port) { 212 /* 213 * check for and canonicalize numeric IP address and portnum (optional) 214 * (IP address may be followed by ":portnum" (optional)) 215 * - IPv6: "[...]" 216 * - IPv4: "x.x.x.x" 217 * - IPv4: 12345678 (32-bit decimal number) 218 * - IPv4: 012345678 (32-bit octal number) 219 * - IPv4: 0x12345678 (32-bit hex number) 220 * 221 * allow any chars (except ':' and '\0' and stray '[' or ']') 222 * (other code may check chars more strictly or more pedantically) 223 * ':' delimits (optional) port at end of string 224 * "[]" wraps IPv6 address literal 225 * '\0' should have been rejected earlier were it present 226 * 227 * any chars includes, but is not limited to: 228 * - allow '-' any where, even at beginning of word 229 * (security caution: might be confused for cmd flag if passed to shell) 230 * - allow all-digit TLDs 231 * (might be mistaken for IPv4 addr by inet_aton() 232 * unless non-digits appear in subdomain) 233 */ 234 235 /* Note: not using getaddrinfo() since it does not support "[]" around IPv6 236 * and is not as lenient as inet_aton() and inet_addr() for IPv4 strings. 237 * Not using inet_pton() (when available) on IPv4 for similar reasons. */ 238 239 const char * const p = b->ptr; 240 const size_t blen = buffer_string_length(b); 241 long port = 0; 242 243 if (*p != '[') { 244 char * const colon = (char *)memchr(p, ':', blen); 245 if (colon) { 246 if (*p == ':') return -1; /*(empty host then port, or naked IPv6)*/ 247 if (colon[1] != '\0') { 248 char *e; 249 port = strtol(colon+1, &e, 0); /*(allow decimal, octal, hex)*/ 250 if (0 < port && port <= USHRT_MAX && *e == '\0') { 251 /* valid port */ 252 } else { 253 return -1; 254 } 255 } /*(else ignore stray colon at string end)*/ 256 buffer_string_set_length(b, (size_t)(colon - p)); /*(remove port str)*/ 257 } 258 259 if (light_isdigit(*p)) do { 260 /* (IPv4 address literal or domain starting w/ digit (e.g. 3com))*/ 261 /* (check one-element cache of normalized IPv4 address string) */ 262 static struct { char s[INET_ADDRSTRLEN]; size_t n; } laddr; 263 size_t n = colon ? (size_t)(colon - p) : blen; 264 sock_addr addr; 265 if (n == laddr.n && 0 == memcmp(p, laddr.s, n)) break; 266 if (1 == sock_addr_inet_pton(&addr, p, AF_INET, 0)) { 267 sock_addr_inet_ntop_copy_buffer(b, &addr); 268 n = buffer_string_length(b); 269 if (n < sizeof(laddr.s)) memcpy(laddr.s, b->ptr, (laddr.n = n)); 270 } 271 } while (0); 272 } else do { /* IPv6 addr */ 273 #if defined(HAVE_IPV6) && defined(HAVE_INET_PTON) 274 275 /* (check one-element cache of normalized IPv4 address string) */ 276 static struct { char s[INET6_ADDRSTRLEN]; size_t n; } laddr; 277 sock_addr addr; 278 char *bracket = b->ptr+blen-1; 279 char *percent = strchr(b->ptr+1, '%'); 280 size_t len; 281 int rc; 282 char buf[INET6_ADDRSTRLEN+16]; /*(+16 for potential %interface name)*/ 283 if (blen <= 2) return -1; /*(invalid "[]")*/ 284 if (*bracket != ']') { 285 bracket = (char *)memchr(b->ptr+1, ']', blen-1); 286 if (NULL == bracket || bracket[1] != ':' || bracket - b->ptr == 1){ 287 return -1; 288 } 289 if (bracket[2] != '\0') { /*(ignore stray colon at string end)*/ 290 char *e; 291 port = strtol(bracket+2, &e, 0); /*(allow decimal, octal, hex)*/ 292 if (0 < port && port <= USHRT_MAX && *e == '\0') { 293 /* valid port */ 294 } else { 295 return -1; 296 } 297 } 298 } 299 300 len = (size_t)((percent ? percent : bracket) - (b->ptr+1)); 301 if (laddr.n == len && 0 == memcmp(laddr.s, b->ptr+1, len)) { 302 /* truncate after ']' and re-add normalized port, if needed */ 303 buffer_string_set_length(b, (size_t)(bracket - b->ptr + 1)); 304 break; 305 } 306 307 *bracket = '\0';/*(terminate IPv6 string)*/ 308 if (percent) *percent = '\0'; /*(remove %interface from address)*/ 309 rc = sock_addr_inet_pton(&addr, b->ptr+1, AF_INET6, 0); 310 if (percent) *percent = '%'; /*(restore %interface)*/ 311 *bracket = ']'; /*(restore bracket)*/ 312 if (1 != rc) return -1; 313 314 sock_addr_inet_ntop(&addr, buf, sizeof(buf)); 315 len = strlen(buf); 316 if (percent) { 317 if (percent > bracket) return -1; 318 if (len + (size_t)(bracket - percent) >= sizeof(buf)) return -1; 319 if (len < sizeof(laddr.s)) memcpy(laddr.s, buf, (laddr.n = len)); 320 memcpy(buf+len, percent, (size_t)(bracket - percent)); 321 len += (size_t)(bracket - percent); 322 } 323 buffer_string_set_length(b, 1); /* truncate after '[' */ 324 buffer_append_string_len(b, buf, len); 325 buffer_append_string_len(b, CONST_STR_LEN("]")); 326 327 #else 328 329 return -1; 330 331 #endif 332 } while (0); 333 334 if (0 != port && port != scheme_port) { 335 buffer_append_string_len(b, CONST_STR_LEN(":")); 336 buffer_append_int(b, (int)port); 337 } 338 339 return 0; 340 } 341 342 int http_request_host_policy (buffer * const b, const unsigned int http_parseopts, const int scheme_port) { 343 return (((http_parseopts & HTTP_PARSEOPT_HOST_STRICT) 344 && 0 != request_check_hostname(b)) 345 || ((http_parseopts & HTTP_PARSEOPT_HOST_NORMALIZE) 346 && 0 != http_request_host_normalize(b, scheme_port))); 347 } 348 349 __attribute_pure__ /*(could be even more strict and use __attribute_const__)*/ 350 static int request_uri_is_valid_char(const unsigned char c) { 351 return (c > 32 && c != 127 && c != 255); 352 } 353 354 __attribute_cold__ 355 __attribute_noinline__ 356 static int http_request_header_line_invalid(request_st * const restrict r, const int status, const char * const restrict msg) { 357 if (r->conf.log_request_header_on_error) { 358 if (msg) log_error(r->conf.errh, __FILE__, __LINE__, "%s", msg); 359 } 360 return status; 361 } 362 363 __attribute_cold__ 364 __attribute_noinline__ 365 static int http_request_header_char_invalid(request_st * const restrict r, const char ch, const char * const restrict msg) { 366 if (r->conf.log_request_header_on_error) { 367 if ((unsigned char)ch > 32 && ch != 127) { 368 log_error(r->conf.errh, __FILE__, __LINE__, "%s ('%c')", msg, ch); 369 } 370 else { 371 log_error(r->conf.errh, __FILE__, __LINE__, "%s (0x%x)", msg, ch); 372 } 373 } 374 return 400; 375 } 376 377 /* add header to list of headers 378 * certain headers are also parsed 379 * might drop a header if deemed unnecessary/broken 380 * 381 * returns 0 on success, HTTP status on error 382 */ 383 static int http_request_parse_single_header(request_st * const restrict r, const enum http_header_e id, const char * const restrict k, const size_t klen, const char * const restrict v, const size_t vlen) { 384 buffer **saveb = NULL; 385 386 /* 387 * Note: k might not be '\0'-terminated 388 * Note: v is not '\0'-terminated, and ends with whitespace 389 * (one of '\r' '\n' ' ' '\t') 390 */ 391 392 switch (id) { 393 /*case HTTP_HEADER_OTHER:*/ 394 default: 395 break; 396 case HTTP_HEADER_HOST: 397 if (!(r->rqst_htags & HTTP_HEADER_HOST)) { 398 saveb = &r->http_host; 399 if (vlen >= 1024) { /*(expecting < 256)*/ 400 return http_request_header_line_invalid(r, 400, "uri-authority too long -> 400"); 401 } 402 } 403 else if (NULL != r->http_host 404 && buffer_is_equal_string(r->http_host, v, vlen)) { 405 /* ignore all Host: headers if match authority in request line */ 406 return 0; /* ignore header */ 407 } 408 else { 409 return http_request_header_line_invalid(r, 400, "duplicate Host header -> 400"); 410 } 411 break; 412 case HTTP_HEADER_CONNECTION: 413 /* "Connection: close" is common case if header is present */ 414 if ((vlen == 5 && buffer_eq_icase_ssn(v, CONST_STR_LEN("close"))) 415 || http_header_str_contains_token(v,vlen,CONST_STR_LEN("close"))) { 416 r->keep_alive = 0; 417 break; 418 } 419 if (http_header_str_contains_token(v,vlen,CONST_STR_LEN("keep-alive"))){ 420 r->keep_alive = 1; 421 break; 422 } 423 break; 424 case HTTP_HEADER_CONTENT_TYPE: 425 if (r->rqst_htags & HTTP_HEADER_CONTENT_TYPE) { 426 return http_request_header_line_invalid(r, 400, "duplicate Content-Type header -> 400"); 427 } 428 break; 429 case HTTP_HEADER_IF_NONE_MATCH: 430 /* if dup, only the first one will survive */ 431 if (r->rqst_htags & HTTP_HEADER_IF_NONE_MATCH) { 432 return 0; /* ignore header */ 433 } 434 break; 435 case HTTP_HEADER_CONTENT_LENGTH: 436 if (!(r->rqst_htags & HTTP_HEADER_CONTENT_LENGTH)) { 437 /*(trailing whitespace was removed from vlen)*/ 438 char *err; 439 off_t clen = strtoll(v, &err, 10); 440 if (clen >= 0 && err == v+vlen && light_isdigit(v[0])) { 441 /* (set only if not set to -1 by Transfer-Encoding: chunked) */ 442 if (0 == r->reqbody_length) r->reqbody_length = clen; 443 } 444 else { 445 return http_request_header_line_invalid(r, 400, "invalid Content-Length header -> 400"); 446 } 447 } 448 else { 449 return http_request_header_line_invalid(r, 400, "duplicate Content-Length header -> 400"); 450 } 451 break; 452 case HTTP_HEADER_IF_MODIFIED_SINCE: 453 if (r->rqst_htags & HTTP_HEADER_IF_MODIFIED_SINCE) { 454 /* Proxies sometimes send dup headers 455 * if they are the same we ignore the second 456 * if not, we raise an error */ 457 const buffer *vb = 458 http_header_request_get(r, HTTP_HEADER_IF_MODIFIED_SINCE, 459 CONST_STR_LEN("If-Modified-Since")); 460 if (vb && buffer_is_equal_caseless_string(vb, v, vlen)) { 461 /* ignore it if they are the same */ 462 return 0; /* ignore header */ 463 } 464 else { 465 return http_request_header_line_invalid(r, 400, "duplicate If-Modified-Since header -> 400"); 466 } 467 } 468 break; 469 case HTTP_HEADER_TRANSFER_ENCODING: 470 if (HTTP_VERSION_1_0 == r->http_version) { 471 return http_request_header_line_invalid(r, 400, "HTTP/1.0 with Transfer-Encoding (bad HTTP/1.0 proxy?) -> 400"); 472 } 473 474 if (!buffer_eq_icase_ss(v, vlen, CONST_STR_LEN("chunked"))) { 475 /* Transfer-Encoding might contain additional encodings, 476 * which are not currently supported by lighttpd */ 477 return http_request_header_line_invalid(r, 501, NULL); /* Not Implemented */ 478 } 479 r->reqbody_length = -1; 480 481 /* Transfer-Encoding is a hop-by-hop header, 482 * which must not be blindly forwarded to backends */ 483 return 0; /* skip header */ 484 } 485 486 http_header_request_append(r, id, k, klen, v, vlen); 487 488 if (saveb) { 489 *saveb = http_header_request_get(r, id, k, klen); 490 } 491 492 return 0; 493 } 494 495 __attribute_cold__ 496 static int http_request_parse_proto_loose(request_st * const restrict r, const char * const restrict ptr, const size_t len, const unsigned int http_parseopts) { 497 const char * proto = memchr(ptr, ' ', len); 498 if (NULL == proto) 499 return http_request_header_line_invalid(r, 400, "incomplete request line -> 400"); 500 proto = memchr(proto+1, ' ', len - (proto+1 - ptr)); 501 if (NULL == proto) 502 return http_request_header_line_invalid(r, 400, "incomplete request line -> 400"); 503 ++proto; 504 505 if (proto[0]=='H' && proto[1]=='T' && proto[2]=='T' && proto[3]=='P' && proto[4] == '/') { 506 if (proto[5] == '1' && proto[6] == '.' && (proto[7] == '1' || proto[7] == '0')) { 507 /* length already checked before calling this routine */ 508 /* (len != (size_t)(proto - ptr + 8)) */ 509 if (http_parseopts & HTTP_PARSEOPT_HEADER_STRICT) /*(http_header_strict)*/ 510 return http_request_header_line_invalid(r, 400, "incomplete request line -> 400"); 511 r->http_version = (proto[7] == '1') ? HTTP_VERSION_1_1 : HTTP_VERSION_1_0; 512 } 513 else 514 return http_request_header_line_invalid(r, 505, "unknown HTTP version -> 505"); 515 } 516 else 517 return http_request_header_line_invalid(r, 400, "unknown protocol -> 400"); 518 519 /* keep-alive default: HTTP/1.1 -> true; HTTP/1.0 -> false */ 520 r->keep_alive = (HTTP_VERSION_1_0 != r->http_version); 521 522 return 0; 523 } 524 525 __attribute_cold__ 526 static const char * http_request_parse_reqline_uri(request_st * const restrict r, const char * const restrict uri, const size_t len, const unsigned int http_parseopts) { 527 const char *nuri; 528 if ((len > 7 && buffer_eq_icase_ssn(uri, "http://", 7) 529 && NULL != (nuri = memchr(uri + 7, '/', len-7))) 530 || 531 (len > 8 && buffer_eq_icase_ssn(uri, "https://", 8) 532 && NULL != (nuri = memchr(uri + 8, '/', len-8)))) { 533 const char * const host = uri + (uri[4] == ':' ? 7 : 8); 534 const size_t hostlen = nuri - host; 535 if (0 == hostlen || hostlen >= 1024) { /*(expecting < 256)*/ 536 http_request_header_line_invalid(r, 400, "uri-authority empty or too long -> 400"); 537 return NULL; 538 } 539 /* Insert as host header */ 540 http_header_request_set(r, HTTP_HEADER_HOST, CONST_STR_LEN("Host"), host, hostlen); 541 r->http_host = http_header_request_get(r, HTTP_HEADER_HOST, CONST_STR_LEN("Host")); 542 return nuri; 543 } else if (!(http_parseopts & HTTP_PARSEOPT_HEADER_STRICT) /*(!http_header_strict)*/ 544 || (HTTP_METHOD_CONNECT == r->http_method && (uri[0] == ':' || light_isdigit(uri[0]))) 545 || (HTTP_METHOD_OPTIONS == r->http_method && uri[0] == '*' && 1 == len)) { 546 /* (permitted) */ 547 return uri; 548 } else { 549 http_request_header_line_invalid(r, 400, "request-URI parse error -> 400"); 550 return NULL; 551 } 552 } 553 554 static int http_request_parse_reqline(request_st * const restrict r, const char * const restrict ptr, const unsigned short * const restrict hoff, const unsigned int http_parseopts) { 555 size_t len = hoff[2]; 556 557 /* parse the first line of the request 558 * <method> <uri> <protocol>\r\n 559 * */ 560 if (len < 13) /* minimum len with (!http_header_strict): "x x HTTP/1.0\n" */ 561 return http_request_header_line_invalid(r, 400, "invalid request line (too short) -> 400"); 562 if (ptr[len-2] == '\r') 563 len-=2; 564 else if (!(http_parseopts & HTTP_PARSEOPT_HEADER_STRICT)) /*(!http_header_strict)*/ 565 len-=1; 566 else 567 return http_request_header_line_invalid(r, 400, "missing CR before LF in header -> 400"); 568 569 /* 570 * RFC7230: 571 * HTTP-version = HTTP-name "/" DIGIT "." DIGIT 572 * HTTP-name = %x48.54.54.50 ; "HTTP", case-sensitive 573 */ 574 575 /* protocol is expected to be " HTTP/1.1" or " HTTP/1.0" at end of line */ 576 union proto_un { 577 char c[8]; 578 uint64_t u; 579 }; 580 static const union proto_un http_1_1 = {{'H','T','T','P','/','1','.','1'}}; 581 static const union proto_un http_1_0 = {{'H','T','T','P','/','1','.','0'}}; 582 const char *p = ptr + len - 8; 583 union proto_un proto8; 584 proto8.c[0]=p[0]; proto8.c[1]=p[1]; proto8.c[2]=p[2]; proto8.c[3]=p[3]; 585 proto8.c[4]=p[4]; proto8.c[5]=p[5]; proto8.c[6]=p[6]; proto8.c[7]=p[7]; 586 if (p[-1] == ' ' && http_1_1.u == proto8.u) { 587 r->http_version = HTTP_VERSION_1_1; 588 r->keep_alive = 1; /* keep-alive default: HTTP/1.1 -> true */ 589 } 590 else if (p[-1] == ' ' && http_1_0.u == proto8.u) { 591 r->http_version = HTTP_VERSION_1_0; 592 r->keep_alive = 0; /* keep-alive default: HTTP/1.0 -> false */ 593 } 594 else { 595 int status = http_request_parse_proto_loose(r,ptr,len,http_parseopts); 596 if (0 != status) return status; 597 /*(space char must exist if http_request_parse_proto_loose() succeeds)*/ 598 for (p = ptr + len - 9; p[-1] != ' '; --p) ; 599 } 600 601 /* method is expected to be a short string in the general case */ 602 size_t i = 0; 603 while (ptr[i] != ' ') ++i; 604 #if 0 /*(space must exist if protocol was parsed successfully)*/ 605 while (i < len && ptr[i] != ' ') ++i; 606 if (ptr[i] != ' ') 607 return http_request_header_line_invalid(r, 400, "incomplete request line -> 400"); 608 #endif 609 610 r->http_method = get_http_method_key(ptr, i); 611 if (HTTP_METHOD_UNSET == r->http_method) 612 return http_request_header_line_invalid(r, 501, "unknown http-method -> 501"); 613 614 const char *uri = ptr + i + 1; 615 616 if (uri == p) 617 return http_request_header_line_invalid(r, 400, "no uri specified -> 400"); 618 len = (size_t)(p - uri - 1); 619 620 if (*uri != '/') { /* (common case: (*uri == '/')) */ 621 uri = http_request_parse_reqline_uri(r, uri, len, http_parseopts); 622 if (NULL == uri) return 400; 623 len = (size_t)(p - uri - 1); 624 } 625 626 if (0 == len) 627 return http_request_header_line_invalid(r, 400, "no uri specified -> 400"); 628 629 /* check uri for invalid characters */ 630 if (http_parseopts & HTTP_PARSEOPT_HEADER_STRICT) { /* http_header_strict */ 631 if ((http_parseopts & HTTP_PARSEOPT_URL_NORMALIZE_CTRLS_REJECT)) { 632 /* URI will be checked in http_response_prepare() */ 633 } 634 else { 635 for (i = 0; i < len; ++i) { 636 if (!request_uri_is_valid_char(uri[i])) 637 return http_request_header_char_invalid(r, uri[i], "invalid character in URI -> 400"); 638 } 639 } 640 } 641 else { 642 /* check entire set of request headers for '\0' */ 643 if (NULL != memchr(ptr, '\0', hoff[hoff[0]])) 644 return http_request_header_char_invalid(r, '\0', "invalid character in header -> 400"); 645 } 646 647 buffer_copy_string_len(&r->target, uri, len); 648 buffer_copy_string_len(&r->target_orig, uri, len); 649 return 0; 650 } 651 652 int http_request_parse_target(request_st * const r, int scheme_port) { 653 /* URI is parsed into components at start of request and may 654 * also be re-parsed upon HANDLER_COMEBACK during the request 655 * r->target is expected to be a "/url-part?query-part" 656 * (and *not* a fully-qualified URI starting https://...) 657 * r->uri.authority is expected to be parsed elsewhere into r->http_host 658 */ 659 660 /** 661 * prepare strings 662 * 663 * - uri.path 664 * - uri.query 665 * 666 */ 667 668 /** 669 * Name according to RFC 2396 670 * 671 * - scheme 672 * - authority 673 * - path 674 * - query 675 * 676 * (scheme)://(authority)(path)?(query)#fragment 677 * 678 */ 679 680 /* take initial scheme value from connection-level state 681 * (request r->uri.scheme can be overwritten for later, 682 * for example by mod_extforward or mod_magnet) */ 683 if (scheme_port == 443) 684 buffer_copy_string_len(&r->uri.scheme, CONST_STR_LEN("https")); 685 else 686 buffer_copy_string_len(&r->uri.scheme, CONST_STR_LEN("http")); 687 688 buffer * const target = &r->target; 689 if (r->http_method == HTTP_METHOD_CONNECT 690 || (r->http_method == HTTP_METHOD_OPTIONS 691 && target->ptr[0] == '*' 692 && target->ptr[1] == '\0')) { 693 /* CONNECT ... (or) OPTIONS * ... */ 694 buffer_copy_buffer(&r->uri.path, target); 695 buffer_clear(&r->uri.query); 696 return 0; 697 } 698 699 char *qstr; 700 if (r->conf.http_parseopts & HTTP_PARSEOPT_URL_NORMALIZE) { 701 /*uint32_t len = (uint32_t)buffer_string_length(target);*/ 702 int qs = burl_normalize(target, r->tmp_buf, r->conf.http_parseopts); 703 if (-2 == qs) { 704 log_error(r->conf.errh, __FILE__, __LINE__, 705 "invalid character in URI -> 400 %s", 706 target->ptr); 707 return 400; /* Bad Request */ 708 } 709 qstr = (-1 == qs) ? NULL : target->ptr+qs; 710 #if 0 /* future: might enable here, or below for all requests */ 711 /* (Note: total header size not recalculated on HANDLER_COMEBACK 712 * even if other request headers changed during processing) 713 * (If (0 != r->loops_per_request), then the generated 714 * request is too large. Should a different error be returned?) */ 715 r->rqst_header_len -= len; 716 len = buffer_string_length(target); 717 r->rqst_header_len += len; 718 if (len > MAX_HTTP_REQUEST_URI) { 719 return 414; /* 414 Request-URI Too Long */ 720 } 721 if (r->rqst_header_len > MAX_HTTP_REQUEST_HEADER) { 722 log_error(r->conf.errh, __FILE__, __LINE__, 723 "request header fields too large: %u -> 431", 724 r->rqst_header_len); 725 return 431; /* Request Header Fields Too Large */ 726 } 727 #endif 728 } 729 else { 730 size_t rlen = buffer_string_length(target); 731 qstr = memchr(target->ptr, '#', rlen);/* discard fragment */ 732 if (qstr) { 733 rlen = (size_t)(qstr - target->ptr); 734 buffer_string_set_length(target, rlen); 735 } 736 qstr = memchr(target->ptr, '?', rlen); 737 } 738 739 /** extract query string from target */ 740 const char * const pstr = target->ptr; 741 const uint32_t rlen = buffer_string_length(target); 742 uint32_t plen; 743 if (NULL != qstr) { 744 plen = (uint32_t)(qstr - pstr); 745 buffer_copy_string_len(&r->uri.query, qstr + 1, rlen - plen - 1); 746 } 747 else { 748 plen = rlen; 749 buffer_clear(&r->uri.query); 750 } 751 buffer_copy_string_len(&r->uri.path, pstr, plen); 752 753 /* decode url to path 754 * 755 * - decode url-encodings (e.g. %20 -> ' ') 756 * - remove path-modifiers (e.g. /../) 757 */ 758 759 buffer_urldecode_path(&r->uri.path); 760 buffer_path_simplify(&r->uri.path, &r->uri.path); 761 if (r->uri.path.ptr[0] != '/') { 762 log_error(r->conf.errh, __FILE__, __LINE__, 763 "uri-path does not begin with '/': %s -> 400", r->uri.path.ptr); 764 return 400; /* Bad Request */ 765 } 766 767 return 0; 768 } 769 770 __attribute_cold__ 771 __attribute_noinline__ 772 static int http_request_parse_header_other(request_st * const restrict r, const char * const restrict k, const int klen, const unsigned int http_header_strict) { 773 for (int i = 0; i < klen; ++i) { 774 if (light_isalpha(k[i]) || k[i] == '-') continue; /*(common cases)*/ 775 /** 776 * 1*<any CHAR except CTLs or separators> 777 * CTLs == 0-31 + 127, CHAR = 7-bit ascii (0..127) 778 * 779 */ 780 switch(k[i]) { 781 case ' ': 782 case '\t': 783 return http_request_header_line_invalid(r, 400, "WS character in key -> 400"); 784 case '(': 785 case ')': 786 case '<': 787 case '>': 788 case '@': 789 case ',': 790 case ';': 791 case '\\': 792 case '\"': 793 case '/': 794 case '[': 795 case ']': 796 case '?': 797 case '=': 798 case '{': 799 case '}': 800 return http_request_header_char_invalid(r, k[i], "invalid character in header key -> 400"); 801 default: 802 if (http_header_strict ? (k[i] < 32 || ((unsigned char *)k)[i] >= 127) : k[i] == '\0') 803 return http_request_header_char_invalid(r, k[i], "invalid character in header key -> 400"); 804 break; /* ok */ 805 } 806 } 807 return 0; 808 } 809 810 static int http_request_parse_headers(request_st * const restrict r, char * const restrict ptr, const unsigned short * const restrict hoff, const unsigned int http_parseopts) { 811 const unsigned int http_header_strict = (http_parseopts & HTTP_PARSEOPT_HEADER_STRICT); 812 813 #if 0 /*(not checked here; will later result in invalid label for HTTP header)*/ 814 int i = hoff[2]; 815 816 if (ptr[i] == ' ' || ptr[i] == '\t') { 817 return http_request_header_line_invalid(r, 400, "WS at the start of first line -> 400"); 818 } 819 #endif 820 821 for (int i = 2; i < hoff[0]; ++i) { 822 const char *k = ptr + hoff[i]; 823 /* one past last line hoff[hoff[0]] is to final "\r\n" */ 824 char *end = ptr + hoff[i+1]; 825 826 const char *colon = memchr(k, ':', end - k); 827 if (NULL == colon) 828 return http_request_header_line_invalid(r, 400, "invalid header missing ':' -> 400"); 829 830 const char *v = colon + 1; 831 832 /* RFC7230 Hypertext Transfer Protocol (HTTP/1.1): Message Syntax and Routing 833 * 3.2.4. Field Parsing 834 * [...] 835 * No whitespace is allowed between the header field-name and colon. In 836 * the past, differences in the handling of such whitespace have led to 837 * security vulnerabilities in request routing and response handling. A 838 * server MUST reject any received request message that contains 839 * whitespace between a header field-name and colon with a response code 840 * of 400 (Bad Request). A proxy MUST remove any such whitespace from a 841 * response message before forwarding the message downstream. 842 */ 843 /* (line k[-1] is always preceded by a '\n', 844 * including first header after request-line, 845 * so no need to check colon != k) */ 846 if (colon[-1] == ' ' || colon[-1] == '\t') { 847 if (http_header_strict) { 848 return http_request_header_line_invalid(r, 400, "invalid whitespace between field-name and colon -> 400"); 849 } 850 else { 851 /* remove trailing whitespace from key(if !http_header_strict)*/ 852 do { --colon; } while (colon[-1] == ' ' || colon[-1] == '\t'); 853 } 854 } 855 856 const int klen = (int)(colon - k); 857 if (0 == klen) 858 return http_request_header_line_invalid(r, 400, "invalid header key -> 400"); 859 const enum http_header_e id = http_header_hkey_get(k, klen); 860 861 if (id == HTTP_HEADER_OTHER) { 862 for (int j = 0; j < klen; ++j) { 863 if (light_isalpha(k[j]) || k[j] == '-') continue; /*(common cases)*/ 864 if (0 != http_request_parse_header_other(r, k+j, klen-j, http_header_strict)) 865 return 400; 866 break; 867 } 868 } 869 870 /* remove leading whitespace from value */ 871 while (*v == ' ' || *v == '\t') ++v; 872 873 for (; i+1 <= hoff[0]; ++i) { 874 end = ptr + hoff[i+1]; 875 if (end[0] != ' ' && end[0] != '\t') break; 876 877 /* line folding */ 878 #ifdef __COVERITY__ 879 force_assert(end - k >= 2); 880 #endif 881 if (end[-2] == '\r') 882 end[-2] = ' '; 883 else if (http_header_strict) 884 return http_request_header_line_invalid(r, 400, "missing CR before LF in header -> 400"); 885 end[-1] = ' '; 886 } 887 #ifdef __COVERITY__ 888 /*(buf holding k has non-zero request-line, so end[-2] valid)*/ 889 force_assert(end >= k + 2); 890 #endif 891 if (end[-2] == '\r') 892 --end; 893 else if (http_header_strict) 894 return http_request_header_line_invalid(r, 400, "missing CR before LF in header -> 400"); 895 /* remove trailing whitespace from value (+ remove '\r\n') */ 896 /* (line k[-1] is always preceded by a '\n', 897 * including first header after request-line, 898 * so no need to check (end != k)) */ 899 do { --end; } while (end[-1] == ' ' || end[-1] == '\t'); 900 901 const int vlen = (int)(end - v); 902 /* empty header-fields are not allowed by HTTP-RFC, we just ignore them */ 903 if (vlen <= 0) continue; /* ignore header */ 904 905 if (http_header_strict) { 906 for (int j = 0; j < vlen; ++j) { 907 if ((((unsigned char *)v)[j] < 32 && v[j] != '\t') || v[j]==127) 908 return http_request_header_char_invalid(r, v[j], "invalid character in header -> 400"); 909 } 910 } /* else URI already checked in http_request_parse_reqline() for any '\0' */ 911 912 int status = http_request_parse_single_header(r, id, k, (size_t)klen, v, (size_t)vlen); 913 if (0 != status) return status; 914 } 915 916 return 0; 917 } 918 919 int http_request_parse(request_st * const restrict r, char * const restrict hdrs, const unsigned short * const restrict hoff, const int scheme_port) { 920 /* 921 * Request: "^(GET|POST|HEAD|...) ([^ ]+(\\?[^ ]+|)) (HTTP/1\\.[01])$" 922 * Header : "^([-a-zA-Z]+): (.+)$" 923 * End : "^$" 924 */ 925 926 int status; 927 const unsigned int http_parseopts = r->conf.http_parseopts; 928 929 status = http_request_parse_reqline(r, hdrs, hoff, http_parseopts); 930 if (0 != status) return status; 931 932 status = http_request_parse_target(r, scheme_port); 933 if (0 != status) return status; 934 935 status = http_request_parse_headers(r, hdrs, hoff, http_parseopts); 936 if (0 != status) return status; 937 938 /*(r->http_host might not be set until after parsing request headers)*/ 939 buffer_copy_buffer(&r->uri.authority, r->http_host);/*(copy even if empty)*/ 940 buffer_to_lower(&r->uri.authority); 941 942 /* post-processing */ 943 944 /* check hostname field if it is set */ 945 if (r->http_host) { 946 if (0 != http_request_host_policy(r->http_host, 947 http_parseopts, scheme_port)) 948 return http_request_header_line_invalid(r, 400, "Invalid Hostname -> 400"); 949 } 950 else { 951 if (r->http_version == HTTP_VERSION_1_1) 952 return http_request_header_line_invalid(r, 400, "HTTP/1.1 but Host missing -> 400"); 953 } 954 955 if (0 == r->reqbody_length) { 956 /* POST requires Content-Length (or Transfer-Encoding) 957 * (-1 == r->reqbody_length when Transfer-Encoding: chunked)*/ 958 if (HTTP_METHOD_POST == r->http_method 959 && !(r->rqst_htags & HTTP_HEADER_CONTENT_LENGTH)) { 960 return http_request_header_line_invalid(r, 411, "POST-request, but content-length missing -> 411"); 961 } 962 } 963 else { 964 /* (-1 == r->reqbody_length when Transfer-Encoding: chunked)*/ 965 if (-1 == r->reqbody_length 966 && (r->rqst_htags & HTTP_HEADER_CONTENT_LENGTH)) { 967 /* RFC7230 Hypertext Transfer Protocol (HTTP/1.1): Message Syntax and Routing 968 * 3.3.3. Message Body Length 969 * [...] 970 * If a message is received with both a Transfer-Encoding and a 971 * Content-Length header field, the Transfer-Encoding overrides the 972 * Content-Length. Such a message might indicate an attempt to 973 * perform request smuggling (Section 9.5) or response splitting 974 * (Section 9.4) and ought to be handled as an error. A sender MUST 975 * remove the received Content-Length field prior to forwarding such 976 * a message downstream. 977 */ 978 const unsigned int http_header_strict = 979 (http_parseopts & HTTP_PARSEOPT_HEADER_STRICT); 980 if (http_header_strict) { 981 return http_request_header_line_invalid(r, 400, "invalid Transfer-Encoding + Content-Length -> 400"); 982 } 983 else { 984 /* ignore Content-Length */ 985 http_header_request_unset(r, HTTP_HEADER_CONTENT_LENGTH, CONST_STR_LEN("Content-Length")); 986 } 987 } 988 if (http_method_get_or_head(r->http_method) 989 && !(http_parseopts & HTTP_PARSEOPT_METHOD_GET_BODY)) { 990 return http_request_header_line_invalid(r, 400, "GET/HEAD with content-length -> 400"); 991 } 992 } 993 994 return 0; 995 } 996