1/* HTTP support. 2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 3 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012, 2014 Free Software Foundation, 4 Inc. 5 6This file is part of GNU Wget. 7 8GNU Wget is free software; you can redistribute it and/or modify 9it under the terms of the GNU General Public License as published by 10the Free Software Foundation; either version 3 of the License, or 11 (at your option) any later version. 12 13GNU Wget is distributed in the hope that it will be useful, 14but WITHOUT ANY WARRANTY; without even the implied warranty of 15MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 16GNU General Public License for more details. 17 18You should have received a copy of the GNU General Public License 19along with Wget. If not, see <http://www.gnu.org/licenses/>. 20 21Additional permission under GNU GPL version 3 section 7 22 23If you modify this program, or any covered work, by linking or 24combining it with the OpenSSL project's OpenSSL library (or a 25modified version of that library), containing parts covered by the 26terms of the OpenSSL or SSLeay licenses, the Free Software Foundation 27grants you additional permission to convey the resulting work. 28Corresponding Source for a non-source form of such a combination 29shall include the source code for the parts of OpenSSL used as well 30as that of the covered work. */ 31 32#include "wget.h" 33 34#include <stdio.h> 35#include <stdlib.h> 36#include <string.h> 37#include <unistd.h> 38#include <assert.h> 39#include <errno.h> 40#include <time.h> 41#include <locale.h> 42 43#include "hash.h" 44#include "http.h" 45#include "utils.h" 46#include "url.h" 47#include "host.h" 48#include "retr.h" 49#include "connect.h" 50#include "netrc.h" 51#ifdef HAVE_SSL 52# include "ssl.h" 53#endif 54#ifdef ENABLE_NTLM 55# include "http-ntlm.h" 56#endif 57#include "cookies.h" 58#include "md5.h" 59#include "convert.h" 60#include "spider.h" 61#include "warc.h" 62 63#ifdef TESTING 64#include "test.h" 65#endif 66 67#ifdef __VMS 68# include "vms.h" 69#endif /* def __VMS */ 70 71extern char *version_string; 72 73/* Forward decls. */ 74struct http_stat; 75static char *create_authorization_line (const char *, const char *, 76 const char *, const char *, 77 const char *, bool *, uerr_t *); 78static char *basic_authentication_encode (const char *, const char *); 79static bool known_authentication_scheme_p (const char *, const char *); 80static void ensure_extension (struct http_stat *, const char *, int *); 81static void load_cookies (void); 82 83#ifndef MIN 84# define MIN(x, y) ((x) > (y) ? (y) : (x)) 85#endif 86 87 88static bool cookies_loaded_p; 89static struct cookie_jar *wget_cookie_jar; 90 91#define TEXTHTML_S "text/html" 92#define TEXTXHTML_S "application/xhtml+xml" 93#define TEXTCSS_S "text/css" 94 95/* Some status code validation macros: */ 96#define H_10X(x) (((x) >= 100) && ((x) < 200)) 97#define H_20X(x) (((x) >= 200) && ((x) < 300)) 98#define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS) 99#define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \ 100 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \ 101 || (x) == HTTP_STATUS_SEE_OTHER \ 102 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT) 103 104/* HTTP/1.0 status codes from RFC1945, provided for reference. */ 105/* Successful 2xx. */ 106#define HTTP_STATUS_OK 200 107#define HTTP_STATUS_CREATED 201 108#define HTTP_STATUS_ACCEPTED 202 109#define HTTP_STATUS_NO_CONTENT 204 110#define HTTP_STATUS_PARTIAL_CONTENTS 206 111 112/* Redirection 3xx. */ 113#define HTTP_STATUS_MULTIPLE_CHOICES 300 114#define HTTP_STATUS_MOVED_PERMANENTLY 301 115#define HTTP_STATUS_MOVED_TEMPORARILY 302 116#define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */ 117#define HTTP_STATUS_NOT_MODIFIED 304 118#define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */ 119 120/* Client error 4xx. */ 121#define HTTP_STATUS_BAD_REQUEST 400 122#define HTTP_STATUS_UNAUTHORIZED 401 123#define HTTP_STATUS_FORBIDDEN 403 124#define HTTP_STATUS_NOT_FOUND 404 125#define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416 126 127/* Server errors 5xx. */ 128#define HTTP_STATUS_INTERNAL 500 129#define HTTP_STATUS_NOT_IMPLEMENTED 501 130#define HTTP_STATUS_BAD_GATEWAY 502 131#define HTTP_STATUS_UNAVAILABLE 503 132 133enum rp { 134 rel_none, rel_name, rel_value, rel_both 135}; 136 137struct request { 138 const char *method; 139 char *arg; 140 141 struct request_header { 142 char *name, *value; 143 enum rp release_policy; 144 } *headers; 145 int hcount, hcapacity; 146}; 147 148extern int numurls; 149 150/* Create a new, empty request. Set the request's method and its 151 arguments. METHOD should be a literal string (or it should outlive 152 the request) because it will not be freed. ARG will be freed by 153 request_free. */ 154 155static struct request * 156request_new (const char *method, char *arg) 157{ 158 struct request *req = xnew0 (struct request); 159 req->hcapacity = 8; 160 req->headers = xnew_array (struct request_header, req->hcapacity); 161 req->method = method; 162 req->arg = arg; 163 return req; 164} 165 166/* Return the method string passed with the last call to 167 request_set_method. */ 168 169static const char * 170request_method (const struct request *req) 171{ 172 return req->method; 173} 174 175/* Free one header according to the release policy specified with 176 request_set_header. */ 177 178static void 179release_header (struct request_header *hdr) 180{ 181 switch (hdr->release_policy) 182 { 183 case rel_none: 184 break; 185 case rel_name: 186 xfree (hdr->name); 187 break; 188 case rel_value: 189 xfree (hdr->value); 190 break; 191 case rel_both: 192 xfree (hdr->name); 193 xfree (hdr->value); 194 break; 195 } 196} 197 198/* Set the request named NAME to VALUE. Specifically, this means that 199 a "NAME: VALUE\r\n" header line will be used in the request. If a 200 header with the same name previously existed in the request, its 201 value will be replaced by this one. A NULL value means do nothing. 202 203 RELEASE_POLICY determines whether NAME and VALUE should be released 204 (freed) with request_free. Allowed values are: 205 206 - rel_none - don't free NAME or VALUE 207 - rel_name - free NAME when done 208 - rel_value - free VALUE when done 209 - rel_both - free both NAME and VALUE when done 210 211 Setting release policy is useful when arguments come from different 212 sources. For example: 213 214 // Don't free literal strings! 215 request_set_header (req, "Pragma", "no-cache", rel_none); 216 217 // Don't free a global variable, we'll need it later. 218 request_set_header (req, "Referer", opt.referer, rel_none); 219 220 // Value freshly allocated, free it when done. 221 request_set_header (req, "Range", 222 aprintf ("bytes=%s-", number_to_static_string (hs->restval)), 223 rel_value); 224 */ 225 226static void 227request_set_header (struct request *req, const char *name, const char *value, 228 enum rp release_policy) 229{ 230 struct request_header *hdr; 231 int i; 232 233 if (!value) 234 { 235 /* A NULL value is a no-op; if freeing the name is requested, 236 free it now to avoid leaks. */ 237 if (release_policy == rel_name || release_policy == rel_both) 238 xfree ((void *)name); 239 return; 240 } 241 242 for (i = 0; i < req->hcount; i++) 243 { 244 hdr = &req->headers[i]; 245 if (0 == strcasecmp (name, hdr->name)) 246 { 247 /* Replace existing header. */ 248 release_header (hdr); 249 hdr->name = (void *)name; 250 hdr->value = (void *)value; 251 hdr->release_policy = release_policy; 252 return; 253 } 254 } 255 256 /* Install new header. */ 257 258 if (req->hcount >= req->hcapacity) 259 { 260 req->hcapacity <<= 1; 261 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr)); 262 } 263 hdr = &req->headers[req->hcount++]; 264 hdr->name = (void *)name; 265 hdr->value = (void *)value; 266 hdr->release_policy = release_policy; 267} 268 269/* Like request_set_header, but sets the whole header line, as 270 provided by the user using the `--header' option. For example, 271 request_set_user_header (req, "Foo: bar") works just like 272 request_set_header (req, "Foo", "bar"). */ 273 274static void 275request_set_user_header (struct request *req, const char *header) 276{ 277 char *name; 278 const char *p = strchr (header, ':'); 279 if (!p) 280 return; 281 BOUNDED_TO_ALLOCA (header, p, name); 282 ++p; 283 while (c_isspace (*p)) 284 ++p; 285 request_set_header (req, xstrdup (name), (char *) p, rel_name); 286} 287 288/* Remove the header with specified name from REQ. Returns true if 289 the header was actually removed, false otherwise. */ 290 291static bool 292request_remove_header (struct request *req, const char *name) 293{ 294 int i; 295 for (i = 0; i < req->hcount; i++) 296 { 297 struct request_header *hdr = &req->headers[i]; 298 if (0 == strcasecmp (name, hdr->name)) 299 { 300 release_header (hdr); 301 /* Move the remaining headers by one. */ 302 if (i < req->hcount - 1) 303 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr)); 304 --req->hcount; 305 return true; 306 } 307 } 308 return false; 309} 310 311#define APPEND(p, str) do { \ 312 int A_len = strlen (str); \ 313 memcpy (p, str, A_len); \ 314 p += A_len; \ 315} while (0) 316 317/* Construct the request and write it to FD using fd_write. 318 If warc_tmp is set to a file pointer, the request string will 319 also be written to that file. */ 320 321static int 322request_send (const struct request *req, int fd, FILE *warc_tmp) 323{ 324 char *request_string, *p; 325 int i, size, write_error; 326 327 /* Count the request size. */ 328 size = 0; 329 330 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */ 331 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2; 332 333 for (i = 0; i < req->hcount; i++) 334 { 335 struct request_header *hdr = &req->headers[i]; 336 /* NAME ": " VALUE "\r\n" */ 337 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2; 338 } 339 340 /* "\r\n\0" */ 341 size += 3; 342 343 p = request_string = alloca_array (char, size); 344 345 /* Generate the request. */ 346 347 APPEND (p, req->method); *p++ = ' '; 348 APPEND (p, req->arg); *p++ = ' '; 349 memcpy (p, "HTTP/1.1\r\n", 10); p += 10; 350 351 for (i = 0; i < req->hcount; i++) 352 { 353 struct request_header *hdr = &req->headers[i]; 354 APPEND (p, hdr->name); 355 *p++ = ':', *p++ = ' '; 356 APPEND (p, hdr->value); 357 *p++ = '\r', *p++ = '\n'; 358 } 359 360 *p++ = '\r', *p++ = '\n', *p++ = '\0'; 361 assert (p - request_string == size); 362 363#undef APPEND 364 365 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string)); 366 367 /* Send the request to the server. */ 368 369 write_error = fd_write (fd, request_string, size - 1, -1); 370 if (write_error < 0) 371 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"), 372 fd_errstr (fd)); 373 else if (warc_tmp != NULL) 374 { 375 /* Write a copy of the data to the WARC record. */ 376 int warc_tmp_written = fwrite (request_string, 1, size - 1, warc_tmp); 377 if (warc_tmp_written != size - 1) 378 return -2; 379 } 380 return write_error; 381} 382 383/* Release the resources used by REQ. */ 384 385static void 386request_free (struct request *req) 387{ 388 int i; 389 xfree_null (req->arg); 390 for (i = 0; i < req->hcount; i++) 391 release_header (&req->headers[i]); 392 xfree_null (req->headers); 393 xfree (req); 394} 395 396static struct hash_table *basic_authed_hosts; 397 398/* Find out if this host has issued a Basic challenge yet; if so, give 399 * it the username, password. A temporary measure until we can get 400 * proper authentication in place. */ 401 402static bool 403maybe_send_basic_creds (const char *hostname, const char *user, 404 const char *passwd, struct request *req) 405{ 406 bool do_challenge = false; 407 408 if (opt.auth_without_challenge) 409 { 410 DEBUGP (("Auth-without-challenge set, sending Basic credentials.\n")); 411 do_challenge = true; 412 } 413 else if (basic_authed_hosts 414 && hash_table_contains(basic_authed_hosts, hostname)) 415 { 416 DEBUGP (("Found %s in basic_authed_hosts.\n", quote (hostname))); 417 do_challenge = true; 418 } 419 else 420 { 421 DEBUGP (("Host %s has not issued a general basic challenge.\n", 422 quote (hostname))); 423 } 424 if (do_challenge) 425 { 426 request_set_header (req, "Authorization", 427 basic_authentication_encode (user, passwd), 428 rel_value); 429 } 430 return do_challenge; 431} 432 433static void 434register_basic_auth_host (const char *hostname) 435{ 436 if (!basic_authed_hosts) 437 { 438 basic_authed_hosts = make_nocase_string_hash_table (1); 439 } 440 if (!hash_table_contains(basic_authed_hosts, hostname)) 441 { 442 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL); 443 DEBUGP (("Inserted %s into basic_authed_hosts\n", quote (hostname))); 444 } 445} 446 447 448/* Send the contents of FILE_NAME to SOCK. Make sure that exactly 449 PROMISED_SIZE bytes are sent over the wire -- if the file is 450 longer, read only that much; if the file is shorter, report an error. 451 If warc_tmp is set to a file pointer, the post data will 452 also be written to that file. */ 453 454static int 455body_file_send (int sock, const char *file_name, wgint promised_size, FILE *warc_tmp) 456{ 457 static char chunk[8192]; 458 wgint written = 0; 459 int write_error; 460 FILE *fp; 461 462 DEBUGP (("[writing BODY file %s ... ", file_name)); 463 464 fp = fopen (file_name, "rb"); 465 if (!fp) 466 return -1; 467 while (!feof (fp) && written < promised_size) 468 { 469 int towrite; 470 int length = fread (chunk, 1, sizeof (chunk), fp); 471 if (length == 0) 472 break; 473 towrite = MIN (promised_size - written, length); 474 write_error = fd_write (sock, chunk, towrite, -1); 475 if (write_error < 0) 476 { 477 fclose (fp); 478 return -1; 479 } 480 if (warc_tmp != NULL) 481 { 482 /* Write a copy of the data to the WARC record. */ 483 int warc_tmp_written = fwrite (chunk, 1, towrite, warc_tmp); 484 if (warc_tmp_written != towrite) 485 { 486 fclose (fp); 487 return -2; 488 } 489 } 490 written += towrite; 491 } 492 fclose (fp); 493 494 /* If we've written less than was promised, report a (probably 495 nonsensical) error rather than break the promise. */ 496 if (written < promised_size) 497 { 498 errno = EINVAL; 499 return -1; 500 } 501 502 assert (written == promised_size); 503 DEBUGP (("done]\n")); 504 return 0; 505} 506 507/* Determine whether [START, PEEKED + PEEKLEN) contains an empty line. 508 If so, return the pointer to the position after the line, otherwise 509 return NULL. This is used as callback to fd_read_hunk. The data 510 between START and PEEKED has been read and cannot be "unread"; the 511 data after PEEKED has only been peeked. */ 512 513static const char * 514response_head_terminator (const char *start, const char *peeked, int peeklen) 515{ 516 const char *p, *end; 517 518 /* If at first peek, verify whether HUNK starts with "HTTP". If 519 not, this is a HTTP/0.9 request and we must bail out without 520 reading anything. */ 521 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4))) 522 return start; 523 524 /* Look for "\n[\r]\n", and return the following position if found. 525 Start two chars before the current to cover the possibility that 526 part of the terminator (e.g. "\n\r") arrived in the previous 527 batch. */ 528 p = peeked - start < 2 ? start : peeked - 2; 529 end = peeked + peeklen; 530 531 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */ 532 for (; p < end - 2; p++) 533 if (*p == '\n') 534 { 535 if (p[1] == '\r' && p[2] == '\n') 536 return p + 3; 537 else if (p[1] == '\n') 538 return p + 2; 539 } 540 /* p==end-2: check for \n\n directly preceding END. */ 541 if (p[0] == '\n' && p[1] == '\n') 542 return p + 2; 543 544 return NULL; 545} 546 547/* The maximum size of a single HTTP response we care to read. Rather 548 than being a limit of the reader implementation, this limit 549 prevents Wget from slurping all available memory upon encountering 550 malicious or buggy server output, thus protecting the user. Define 551 it to 0 to remove the limit. */ 552 553#define HTTP_RESPONSE_MAX_SIZE 65536 554 555/* Read the HTTP request head from FD and return it. The error 556 conditions are the same as with fd_read_hunk. 557 558 To support HTTP/0.9 responses, this function tries to make sure 559 that the data begins with "HTTP". If this is not the case, no data 560 is read and an empty request is returned, so that the remaining 561 data can be treated as body. */ 562 563static char * 564read_http_response_head (int fd) 565{ 566 return fd_read_hunk (fd, response_head_terminator, 512, 567 HTTP_RESPONSE_MAX_SIZE); 568} 569 570struct response { 571 /* The response data. */ 572 const char *data; 573 574 /* The array of pointers that indicate where each header starts. 575 For example, given this HTTP response: 576 577 HTTP/1.0 200 Ok 578 Description: some 579 text 580 Etag: x 581 582 The headers are located like this: 583 584 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n" 585 ^ ^ ^ ^ 586 headers[0] headers[1] headers[2] headers[3] 587 588 I.e. headers[0] points to the beginning of the request, 589 headers[1] points to the end of the first header and the 590 beginning of the second one, etc. */ 591 592 const char **headers; 593}; 594 595/* Create a new response object from the text of the HTTP response, 596 available in HEAD. That text is automatically split into 597 constituent header lines for fast retrieval using 598 resp_header_*. */ 599 600static struct response * 601resp_new (const char *head) 602{ 603 const char *hdr; 604 int count, size; 605 606 struct response *resp = xnew0 (struct response); 607 resp->data = head; 608 609 if (*head == '\0') 610 { 611 /* Empty head means that we're dealing with a headerless 612 (HTTP/0.9) response. In that case, don't set HEADERS at 613 all. */ 614 return resp; 615 } 616 617 /* Split HEAD into header lines, so that resp_header_* functions 618 don't need to do this over and over again. */ 619 620 size = count = 0; 621 hdr = head; 622 while (1) 623 { 624 DO_REALLOC (resp->headers, size, count + 1, const char *); 625 resp->headers[count++] = hdr; 626 627 /* Break upon encountering an empty line. */ 628 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n') 629 break; 630 631 /* Find the end of HDR, including continuations. */ 632 do 633 { 634 const char *end = strchr (hdr, '\n'); 635 if (end) 636 hdr = end + 1; 637 else 638 hdr += strlen (hdr); 639 } 640 while (*hdr == ' ' || *hdr == '\t'); 641 } 642 DO_REALLOC (resp->headers, size, count + 1, const char *); 643 resp->headers[count] = NULL; 644 645 return resp; 646} 647 648/* Locate the header named NAME in the request data, starting with 649 position START. This allows the code to loop through the request 650 data, filtering for all requests of a given name. Returns the 651 found position, or -1 for failure. The code that uses this 652 function typically looks like this: 653 654 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++) 655 ... do something with header ... 656 657 If you only care about one header, use resp_header_get instead of 658 this function. */ 659 660static int 661resp_header_locate (const struct response *resp, const char *name, int start, 662 const char **begptr, const char **endptr) 663{ 664 int i; 665 const char **headers = resp->headers; 666 int name_len; 667 668 if (!headers || !headers[1]) 669 return -1; 670 671 name_len = strlen (name); 672 if (start > 0) 673 i = start; 674 else 675 i = 1; 676 677 for (; headers[i + 1]; i++) 678 { 679 const char *b = headers[i]; 680 const char *e = headers[i + 1]; 681 if (e - b > name_len 682 && b[name_len] == ':' 683 && 0 == strncasecmp (b, name, name_len)) 684 { 685 b += name_len + 1; 686 while (b < e && c_isspace (*b)) 687 ++b; 688 while (b < e && c_isspace (e[-1])) 689 --e; 690 *begptr = b; 691 *endptr = e; 692 return i; 693 } 694 } 695 return -1; 696} 697 698/* Find and retrieve the header named NAME in the request data. If 699 found, set *BEGPTR to its starting, and *ENDPTR to its ending 700 position, and return true. Otherwise return false. 701 702 This function is used as a building block for resp_header_copy 703 and resp_header_strdup. */ 704 705static bool 706resp_header_get (const struct response *resp, const char *name, 707 const char **begptr, const char **endptr) 708{ 709 int pos = resp_header_locate (resp, name, 0, begptr, endptr); 710 return pos != -1; 711} 712 713/* Copy the response header named NAME to buffer BUF, no longer than 714 BUFSIZE (BUFSIZE includes the terminating 0). If the header 715 exists, true is returned, false otherwise. If there should be no 716 limit on the size of the header, use resp_header_strdup instead. 717 718 If BUFSIZE is 0, no data is copied, but the boolean indication of 719 whether the header is present is still returned. */ 720 721static bool 722resp_header_copy (const struct response *resp, const char *name, 723 char *buf, int bufsize) 724{ 725 const char *b, *e; 726 if (!resp_header_get (resp, name, &b, &e)) 727 return false; 728 if (bufsize) 729 { 730 int len = MIN (e - b, bufsize - 1); 731 memcpy (buf, b, len); 732 buf[len] = '\0'; 733 } 734 return true; 735} 736 737/* Return the value of header named NAME in RESP, allocated with 738 malloc. If such a header does not exist in RESP, return NULL. */ 739 740static char * 741resp_header_strdup (const struct response *resp, const char *name) 742{ 743 const char *b, *e; 744 if (!resp_header_get (resp, name, &b, &e)) 745 return NULL; 746 return strdupdelim (b, e); 747} 748 749/* Parse the HTTP status line, which is of format: 750 751 HTTP-Version SP Status-Code SP Reason-Phrase 752 753 The function returns the status-code, or -1 if the status line 754 appears malformed. The pointer to "reason-phrase" message is 755 returned in *MESSAGE. */ 756 757static int 758resp_status (const struct response *resp, char **message) 759{ 760 int status; 761 const char *p, *end; 762 763 if (!resp->headers) 764 { 765 /* For a HTTP/0.9 response, assume status 200. */ 766 if (message) 767 *message = xstrdup (_("No headers, assuming HTTP/0.9")); 768 return 200; 769 } 770 771 p = resp->headers[0]; 772 end = resp->headers[1]; 773 774 if (!end) 775 return -1; 776 777 /* "HTTP" */ 778 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4)) 779 return -1; 780 p += 4; 781 782 /* Match the HTTP version. This is optional because Gnutella 783 servers have been reported to not specify HTTP version. */ 784 if (p < end && *p == '/') 785 { 786 ++p; 787 while (p < end && c_isdigit (*p)) 788 ++p; 789 if (p < end && *p == '.') 790 ++p; 791 while (p < end && c_isdigit (*p)) 792 ++p; 793 } 794 795 while (p < end && c_isspace (*p)) 796 ++p; 797 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2])) 798 return -1; 799 800 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0'); 801 p += 3; 802 803 if (message) 804 { 805 while (p < end && c_isspace (*p)) 806 ++p; 807 while (p < end && c_isspace (end[-1])) 808 --end; 809 *message = strdupdelim (p, end); 810 } 811 812 return status; 813} 814 815/* Release the resources used by RESP. */ 816 817static void 818resp_free (struct response *resp) 819{ 820 xfree_null (resp->headers); 821 xfree (resp); 822} 823 824/* Print a single line of response, the characters [b, e). We tried 825 getting away with 826 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b); 827 but that failed to escape the non-printable characters and, in fact, 828 caused crashes in UTF-8 locales. */ 829 830static void 831print_response_line(const char *prefix, const char *b, const char *e) 832{ 833 char *copy; 834 BOUNDED_TO_ALLOCA(b, e, copy); 835 logprintf (LOG_ALWAYS, "%s%s\n", prefix, 836 quotearg_style (escape_quoting_style, copy)); 837} 838 839/* Print the server response, line by line, omitting the trailing CRLF 840 from individual header lines, and prefixed with PREFIX. */ 841 842static void 843print_server_response (const struct response *resp, const char *prefix) 844{ 845 int i; 846 if (!resp->headers) 847 return; 848 for (i = 0; resp->headers[i + 1]; i++) 849 { 850 const char *b = resp->headers[i]; 851 const char *e = resp->headers[i + 1]; 852 /* Skip CRLF */ 853 if (b < e && e[-1] == '\n') 854 --e; 855 if (b < e && e[-1] == '\r') 856 --e; 857 print_response_line(prefix, b, e); 858 } 859} 860 861/* Parse the `Content-Range' header and extract the information it 862 contains. Returns true if successful, false otherwise. */ 863static bool 864parse_content_range (const char *hdr, wgint *first_byte_ptr, 865 wgint *last_byte_ptr, wgint *entity_length_ptr) 866{ 867 wgint num; 868 869 /* Ancient versions of Netscape proxy server, presumably predating 870 rfc2068, sent out `Content-Range' without the "bytes" 871 specifier. */ 872 if (0 == strncasecmp (hdr, "bytes", 5)) 873 { 874 hdr += 5; 875 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the 876 HTTP spec. */ 877 if (*hdr == ':') 878 ++hdr; 879 while (c_isspace (*hdr)) 880 ++hdr; 881 if (!*hdr) 882 return false; 883 } 884 if (!c_isdigit (*hdr)) 885 return false; 886 for (num = 0; c_isdigit (*hdr); hdr++) 887 num = 10 * num + (*hdr - '0'); 888 if (*hdr != '-' || !c_isdigit (*(hdr + 1))) 889 return false; 890 *first_byte_ptr = num; 891 ++hdr; 892 for (num = 0; c_isdigit (*hdr); hdr++) 893 num = 10 * num + (*hdr - '0'); 894 if (*hdr != '/' || !c_isdigit (*(hdr + 1))) 895 return false; 896 *last_byte_ptr = num; 897 ++hdr; 898 if (*hdr == '*') 899 num = -1; 900 else 901 for (num = 0; c_isdigit (*hdr); hdr++) 902 num = 10 * num + (*hdr - '0'); 903 *entity_length_ptr = num; 904 return true; 905} 906 907/* Read the body of the request, but don't store it anywhere and don't 908 display a progress gauge. This is useful for reading the bodies of 909 administrative responses to which we will soon issue another 910 request. The response is not useful to the user, but reading it 911 allows us to continue using the same connection to the server. 912 913 If reading fails, false is returned, true otherwise. In debug 914 mode, the body is displayed for debugging purposes. */ 915 916static bool 917skip_short_body (int fd, wgint contlen, bool chunked) 918{ 919 enum { 920 SKIP_SIZE = 512, /* size of the download buffer */ 921 SKIP_THRESHOLD = 4096 /* the largest size we read */ 922 }; 923 wgint remaining_chunk_size = 0; 924 char dlbuf[SKIP_SIZE + 1]; 925 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */ 926 927 /* If the body is too large, it makes more sense to simply close the 928 connection than to try to read the body. */ 929 if (contlen > SKIP_THRESHOLD) 930 return false; 931 932 while (contlen > 0 || chunked) 933 { 934 int ret; 935 if (chunked) 936 { 937 if (remaining_chunk_size == 0) 938 { 939 char *line = fd_read_line (fd); 940 char *endl; 941 if (line == NULL) 942 break; 943 944 remaining_chunk_size = strtol (line, &endl, 16); 945 xfree (line); 946 947 if (remaining_chunk_size == 0) 948 { 949 line = fd_read_line (fd); 950 xfree_null (line); 951 break; 952 } 953 } 954 955 contlen = MIN (remaining_chunk_size, SKIP_SIZE); 956 } 957 958 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen))); 959 960 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1); 961 if (ret <= 0) 962 { 963 /* Don't normally report the error since this is an 964 optimization that should be invisible to the user. */ 965 DEBUGP (("] aborting (%s).\n", 966 ret < 0 ? fd_errstr (fd) : "EOF received")); 967 return false; 968 } 969 contlen -= ret; 970 971 if (chunked) 972 { 973 remaining_chunk_size -= ret; 974 if (remaining_chunk_size == 0) 975 { 976 char *line = fd_read_line (fd); 977 if (line == NULL) 978 return false; 979 else 980 xfree (line); 981 } 982 } 983 984 /* Safe even if %.*s bogusly expects terminating \0 because 985 we've zero-terminated dlbuf above. */ 986 DEBUGP (("%.*s", ret, dlbuf)); 987 } 988 989 DEBUGP (("] done.\n")); 990 return true; 991} 992 993#define NOT_RFC2231 0 994#define RFC2231_NOENCODING 1 995#define RFC2231_ENCODING 2 996 997/* extract_param extracts the parameter name into NAME. 998 However, if the parameter name is in RFC2231 format then 999 this function adjusts NAME by stripping of the trailing 1000 characters that are not part of the name but are present to 1001 indicate the presence of encoding information in the value 1002 or a fragment of a long parameter value 1003*/ 1004static int 1005modify_param_name(param_token *name) 1006{ 1007 const char *delim1 = memchr (name->b, '*', name->e - name->b); 1008 const char *delim2 = memrchr (name->b, '*', name->e - name->b); 1009 1010 int result; 1011 1012 if(delim1 == NULL) 1013 { 1014 result = NOT_RFC2231; 1015 } 1016 else if(delim1 == delim2) 1017 { 1018 if ((name->e - 1) == delim1) 1019 { 1020 result = RFC2231_ENCODING; 1021 } 1022 else 1023 { 1024 result = RFC2231_NOENCODING; 1025 } 1026 name->e = delim1; 1027 } 1028 else 1029 { 1030 name->e = delim1; 1031 result = RFC2231_ENCODING; 1032 } 1033 return result; 1034} 1035 1036/* extract_param extract the paramater value into VALUE. 1037 Like modify_param_name this function modifies VALUE by 1038 stripping off the encoding information from the actual value 1039*/ 1040static void 1041modify_param_value (param_token *value, int encoding_type ) 1042{ 1043 if (encoding_type == RFC2231_ENCODING) 1044 { 1045 const char *delim = memrchr (value->b, '\'', value->e - value->b); 1046 if (delim != NULL) 1047 { 1048 value->b = (delim+1); 1049 } 1050 } 1051} 1052 1053/* Extract a parameter from the string (typically an HTTP header) at 1054 **SOURCE and advance SOURCE to the next parameter. Return false 1055 when there are no more parameters to extract. The name of the 1056 parameter is returned in NAME, and the value in VALUE. If the 1057 parameter has no value, the token's value is zeroed out. 1058 1059 For example, if *SOURCE points to the string "attachment; 1060 filename=\"foo bar\"", the first call to this function will return 1061 the token named "attachment" and no value, and the second call will 1062 return the token named "filename" and value "foo bar". The third 1063 call will return false, indicating no more valid tokens. 1064 1065 is_url_encoded is an out parameter. If not NULL, a boolean value will be 1066 stored into it, letting the caller know whether or not the extracted value is 1067 URL-encoded. The caller can then decode it with url_unescape(), which however 1068 performs decoding in-place. URL-encoding is used by RFC 2231 to support 1069 non-US-ASCII characters in HTTP header values. */ 1070 1071bool 1072extract_param (const char **source, param_token *name, param_token *value, 1073 char separator, bool *is_url_encoded) 1074{ 1075 const char *p = *source; 1076 int param_type; 1077 if (is_url_encoded) 1078 *is_url_encoded = false; /* initializing the out parameter */ 1079 1080 while (c_isspace (*p)) ++p; 1081 if (!*p) 1082 { 1083 *source = p; 1084 return false; /* no error; nothing more to extract */ 1085 } 1086 1087 /* Extract name. */ 1088 name->b = p; 1089 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p; 1090 name->e = p; 1091 if (name->b == name->e) 1092 return false; /* empty name: error */ 1093 while (c_isspace (*p)) ++p; 1094 if (*p == separator || !*p) /* no value */ 1095 { 1096 xzero (*value); 1097 if (*p == separator) ++p; 1098 *source = p; 1099 return true; 1100 } 1101 if (*p != '=') 1102 return false; /* error */ 1103 1104 /* *p is '=', extract value */ 1105 ++p; 1106 while (c_isspace (*p)) ++p; 1107 if (*p == '"') /* quoted */ 1108 { 1109 value->b = ++p; 1110 while (*p && *p != '"') ++p; 1111 if (!*p) 1112 return false; 1113 value->e = p++; 1114 /* Currently at closing quote; find the end of param. */ 1115 while (c_isspace (*p)) ++p; 1116 while (*p && *p != separator) ++p; 1117 if (*p == separator) 1118 ++p; 1119 else if (*p) 1120 /* garbage after closed quote, e.g. foo="bar"baz */ 1121 return false; 1122 } 1123 else /* unquoted */ 1124 { 1125 value->b = p; 1126 while (*p && *p != separator) ++p; 1127 value->e = p; 1128 while (value->e != value->b && c_isspace (value->e[-1])) 1129 --value->e; 1130 if (*p == separator) ++p; 1131 } 1132 *source = p; 1133 1134 param_type = modify_param_name(name); 1135 if (param_type != NOT_RFC2231) 1136 { 1137 if (param_type == RFC2231_ENCODING && is_url_encoded) 1138 *is_url_encoded = true; 1139 modify_param_value(value, param_type); 1140 } 1141 return true; 1142} 1143 1144#undef NOT_RFC2231 1145#undef RFC2231_NOENCODING 1146#undef RFC2231_ENCODING 1147 1148/* Appends the string represented by VALUE to FILENAME */ 1149 1150static void 1151append_value_to_filename (char **filename, param_token const * const value, 1152 bool is_url_encoded) 1153{ 1154 int original_length = strlen(*filename); 1155 int new_length = strlen(*filename) + (value->e - value->b); 1156 *filename = xrealloc (*filename, new_length+1); 1157 memcpy (*filename + original_length, value->b, (value->e - value->b)); 1158 (*filename)[new_length] = '\0'; 1159 if (is_url_encoded) 1160 url_unescape (*filename + original_length); 1161} 1162 1163#undef MAX 1164#define MAX(p, q) ((p) > (q) ? (p) : (q)) 1165 1166/* Parse the contents of the `Content-Disposition' header, extracting 1167 the information useful to Wget. Content-Disposition is a header 1168 borrowed from MIME; when used in HTTP, it typically serves for 1169 specifying the desired file name of the resource. For example: 1170 1171 Content-Disposition: attachment; filename="flora.jpg" 1172 1173 Wget will skip the tokens it doesn't care about, such as 1174 "attachment" in the previous example; it will also skip other 1175 unrecognized params. If the header is syntactically correct and 1176 contains a file name, a copy of the file name is stored in 1177 *filename and true is returned. Otherwise, the function returns 1178 false. 1179 1180 The file name is stripped of directory components and must not be 1181 empty. 1182 1183 Historically, this function returned filename prefixed with opt.dir_prefix, 1184 now that logic is handled by the caller, new code should pay attention, 1185 changed by crq, Sep 2010. 1186 1187*/ 1188static bool 1189parse_content_disposition (const char *hdr, char **filename) 1190{ 1191 param_token name, value; 1192 *filename = NULL; 1193 bool is_url_encoded = false; 1194 for ( ; extract_param (&hdr, &name, &value, ';', &is_url_encoded); 1195 is_url_encoded = false) 1196 { 1197 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" ); 1198 if ( isFilename && value.b != NULL) 1199 { 1200 /* Make the file name begin at the last slash or backslash. */ 1201 const char *last_slash = memrchr (value.b, '/', value.e - value.b); 1202 const char *last_bs = memrchr (value.b, '\\', value.e - value.b); 1203 if (last_slash && last_bs) 1204 value.b = 1 + MAX (last_slash, last_bs); 1205 else if (last_slash || last_bs) 1206 value.b = 1 + (last_slash ? last_slash : last_bs); 1207 if (value.b == value.e) 1208 continue; 1209 1210 if (*filename) 1211 append_value_to_filename (filename, &value, is_url_encoded); 1212 else 1213 { 1214 *filename = strdupdelim (value.b, value.e); 1215 if (is_url_encoded) 1216 url_unescape (*filename); 1217 } 1218 } 1219 } 1220 1221 if (*filename) 1222 return true; 1223 else 1224 return false; 1225} 1226 1227 1228/* Persistent connections. Currently, we cache the most recently used 1229 connection as persistent, provided that the HTTP server agrees to 1230 make it such. The persistence data is stored in the variables 1231 below. Ideally, it should be possible to cache an arbitrary fixed 1232 number of these connections. */ 1233 1234/* Whether a persistent connection is active. */ 1235static bool pconn_active; 1236 1237static struct { 1238 /* The socket of the connection. */ 1239 int socket; 1240 1241 /* Host and port of the currently active persistent connection. */ 1242 char *host; 1243 int port; 1244 1245 /* Whether a ssl handshake has occoured on this connection. */ 1246 bool ssl; 1247 1248 /* Whether the connection was authorized. This is only done by 1249 NTLM, which authorizes *connections* rather than individual 1250 requests. (That practice is peculiar for HTTP, but it is a 1251 useful optimization.) */ 1252 bool authorized; 1253 1254#ifdef ENABLE_NTLM 1255 /* NTLM data of the current connection. */ 1256 struct ntlmdata ntlm; 1257#endif 1258} pconn; 1259 1260/* Mark the persistent connection as invalid and free the resources it 1261 uses. This is used by the CLOSE_* macros after they forcefully 1262 close a registered persistent connection. */ 1263 1264static void 1265invalidate_persistent (void) 1266{ 1267 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket)); 1268 pconn_active = false; 1269 fd_close (pconn.socket); 1270 xfree (pconn.host); 1271 xzero (pconn); 1272} 1273 1274/* Register FD, which should be a TCP/IP connection to HOST:PORT, as 1275 persistent. This will enable someone to use the same connection 1276 later. In the context of HTTP, this must be called only AFTER the 1277 response has been received and the server has promised that the 1278 connection will remain alive. 1279 1280 If a previous connection was persistent, it is closed. */ 1281 1282static void 1283register_persistent (const char *host, int port, int fd, bool ssl) 1284{ 1285 if (pconn_active) 1286 { 1287 if (pconn.socket == fd) 1288 { 1289 /* The connection FD is already registered. */ 1290 return; 1291 } 1292 else 1293 { 1294 /* The old persistent connection is still active; close it 1295 first. This situation arises whenever a persistent 1296 connection exists, but we then connect to a different 1297 host, and try to register a persistent connection to that 1298 one. */ 1299 invalidate_persistent (); 1300 } 1301 } 1302 1303 pconn_active = true; 1304 pconn.socket = fd; 1305 pconn.host = xstrdup (host); 1306 pconn.port = port; 1307 pconn.ssl = ssl; 1308 pconn.authorized = false; 1309 1310 DEBUGP (("Registered socket %d for persistent reuse.\n", fd)); 1311} 1312 1313/* Return true if a persistent connection is available for connecting 1314 to HOST:PORT. */ 1315 1316static bool 1317persistent_available_p (const char *host, int port, bool ssl, 1318 bool *host_lookup_failed) 1319{ 1320 /* First, check whether a persistent connection is active at all. */ 1321 if (!pconn_active) 1322 return false; 1323 1324 /* If we want SSL and the last connection wasn't or vice versa, 1325 don't use it. Checking for host and port is not enough because 1326 HTTP and HTTPS can apparently coexist on the same port. */ 1327 if (ssl != pconn.ssl) 1328 return false; 1329 1330 /* If we're not connecting to the same port, we're not interested. */ 1331 if (port != pconn.port) 1332 return false; 1333 1334 /* If the host is the same, we're in business. If not, there is 1335 still hope -- read below. */ 1336 if (0 != strcasecmp (host, pconn.host)) 1337 { 1338 /* Check if pconn.socket is talking to HOST under another name. 1339 This happens often when both sites are virtual hosts 1340 distinguished only by name and served by the same network 1341 interface, and hence the same web server (possibly set up by 1342 the ISP and serving many different web sites). This 1343 admittedly unconventional optimization does not contradict 1344 HTTP and works well with popular server software. */ 1345 1346 bool found; 1347 ip_address ip; 1348 struct address_list *al; 1349 1350 if (ssl) 1351 /* Don't try to talk to two different SSL sites over the same 1352 secure connection! (Besides, it's not clear that 1353 name-based virtual hosting is even possible with SSL.) */ 1354 return false; 1355 1356 /* If pconn.socket's peer is one of the IP addresses HOST 1357 resolves to, pconn.socket is for all intents and purposes 1358 already talking to HOST. */ 1359 1360 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER)) 1361 { 1362 /* Can't get the peer's address -- something must be very 1363 wrong with the connection. */ 1364 invalidate_persistent (); 1365 return false; 1366 } 1367 al = lookup_host (host, 0); 1368 if (!al) 1369 { 1370 *host_lookup_failed = true; 1371 return false; 1372 } 1373 1374 found = address_list_contains (al, &ip); 1375 address_list_release (al); 1376 1377 if (!found) 1378 return false; 1379 1380 /* The persistent connection's peer address was found among the 1381 addresses HOST resolved to; therefore, pconn.sock is in fact 1382 already talking to HOST -- no need to reconnect. */ 1383 } 1384 1385 /* Finally, check whether the connection is still open. This is 1386 important because most servers implement liberal (short) timeout 1387 on persistent connections. Wget can of course always reconnect 1388 if the connection doesn't work out, but it's nicer to know in 1389 advance. This test is a logical followup of the first test, but 1390 is "expensive" and therefore placed at the end of the list. 1391 1392 (Current implementation of test_socket_open has a nice side 1393 effect that it treats sockets with pending data as "closed". 1394 This is exactly what we want: if a broken server sends message 1395 body in response to HEAD, or if it sends more than conent-length 1396 data, we won't reuse the corrupted connection.) */ 1397 1398 if (!test_socket_open (pconn.socket)) 1399 { 1400 /* Oops, the socket is no longer open. Now that we know that, 1401 let's invalidate the persistent connection before returning 1402 0. */ 1403 invalidate_persistent (); 1404 return false; 1405 } 1406 1407 return true; 1408} 1409 1410/* The idea behind these two CLOSE macros is to distinguish between 1411 two cases: one when the job we've been doing is finished, and we 1412 want to close the connection and leave, and two when something is 1413 seriously wrong and we're closing the connection as part of 1414 cleanup. 1415 1416 In case of keep_alive, CLOSE_FINISH should leave the connection 1417 open, while CLOSE_INVALIDATE should still close it. 1418 1419 Note that the semantics of the flag `keep_alive' is "this 1420 connection *will* be reused (the server has promised not to close 1421 the connection once we're done)", while the semantics of 1422 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an 1423 active, registered connection". */ 1424 1425#define CLOSE_FINISH(fd) do { \ 1426 if (!keep_alive) \ 1427 { \ 1428 if (pconn_active && (fd) == pconn.socket) \ 1429 invalidate_persistent (); \ 1430 else \ 1431 { \ 1432 fd_close (fd); \ 1433 fd = -1; \ 1434 } \ 1435 } \ 1436} while (0) 1437 1438#define CLOSE_INVALIDATE(fd) do { \ 1439 if (pconn_active && (fd) == pconn.socket) \ 1440 invalidate_persistent (); \ 1441 else \ 1442 fd_close (fd); \ 1443 fd = -1; \ 1444} while (0) 1445 1446struct http_stat 1447{ 1448 wgint len; /* received length */ 1449 wgint contlen; /* expected length */ 1450 wgint restval; /* the restart value */ 1451 int res; /* the result of last read */ 1452 char *rderrmsg; /* error message from read error */ 1453 char *newloc; /* new location (redirection) */ 1454 char *remote_time; /* remote time-stamp string */ 1455 char *error; /* textual HTTP error */ 1456 int statcode; /* status code */ 1457 char *message; /* status message */ 1458 wgint rd_size; /* amount of data read from socket */ 1459 double dltime; /* time it took to download the data */ 1460 const char *referer; /* value of the referer header. */ 1461 char *local_file; /* local file name. */ 1462 bool existence_checked; /* true if we already checked for a file's 1463 existence after having begun to download 1464 (needed in gethttp for when connection is 1465 interrupted/restarted. */ 1466 bool timestamp_checked; /* true if pre-download time-stamping checks 1467 * have already been performed */ 1468 char *orig_file_name; /* name of file to compare for time-stamping 1469 * (might be != local_file if -K is set) */ 1470 wgint orig_file_size; /* size of file to compare for time-stamping */ 1471 time_t orig_file_tstamp; /* time-stamp of file to compare for 1472 * time-stamping */ 1473}; 1474 1475static void 1476free_hstat (struct http_stat *hs) 1477{ 1478 xfree_null (hs->newloc); 1479 xfree_null (hs->remote_time); 1480 xfree_null (hs->error); 1481 xfree_null (hs->rderrmsg); 1482 xfree_null (hs->local_file); 1483 xfree_null (hs->orig_file_name); 1484 xfree_null (hs->message); 1485 1486 /* Guard against being called twice. */ 1487 hs->newloc = NULL; 1488 hs->remote_time = NULL; 1489 hs->error = NULL; 1490} 1491 1492static void 1493get_file_flags (const char *filename, int *dt) 1494{ 1495 logprintf (LOG_VERBOSE, _("\ 1496File %s already there; not retrieving.\n\n"), quote (filename)); 1497 /* If the file is there, we suppose it's retrieved OK. */ 1498 *dt |= RETROKF; 1499 1500 /* #### Bogusness alert. */ 1501 /* If its suffix is "html" or "htm" or similar, assume text/html. */ 1502 if (has_html_suffix_p (filename)) 1503 *dt |= TEXTHTML; 1504} 1505 1506/* Download the response body from the socket and writes it to 1507 an output file. The headers have already been read from the 1508 socket. If WARC is enabled, the response body will also be 1509 written to a WARC response record. 1510 1511 hs, contlen, contrange, chunked_transfer_encoding and url are 1512 parameters from the gethttp method. fp is a pointer to the 1513 output file. 1514 1515 url, warc_timestamp_str, warc_request_uuid, warc_ip, type 1516 and statcode will be saved in the headers of the WARC record. 1517 The head parameter contains the HTTP headers of the response. 1518 1519 If fp is NULL and WARC is enabled, the response body will be 1520 written only to the WARC file. If WARC is disabled and fp 1521 is a file pointer, the data will be written to the file. 1522 If fp is a file pointer and WARC is enabled, the body will 1523 be written to both destinations. 1524 1525 Returns the error code. */ 1526static int 1527read_response_body (struct http_stat *hs, int sock, FILE *fp, wgint contlen, 1528 wgint contrange, bool chunked_transfer_encoding, 1529 char *url, char *warc_timestamp_str, char *warc_request_uuid, 1530 ip_address *warc_ip, char *type, int statcode, char *head) 1531{ 1532 int warc_payload_offset = 0; 1533 FILE *warc_tmp = NULL; 1534 int warcerr = 0; 1535 1536 if (opt.warc_filename != NULL) 1537 { 1538 /* Open a temporary file where we can write the response before we 1539 add it to the WARC record. */ 1540 warc_tmp = warc_tempfile (); 1541 if (warc_tmp == NULL) 1542 warcerr = WARC_TMP_FOPENERR; 1543 1544 if (warcerr == 0) 1545 { 1546 /* We should keep the response headers for the WARC record. */ 1547 int head_len = strlen (head); 1548 int warc_tmp_written = fwrite (head, 1, head_len, warc_tmp); 1549 if (warc_tmp_written != head_len) 1550 warcerr = WARC_TMP_FWRITEERR; 1551 warc_payload_offset = head_len; 1552 } 1553 1554 if (warcerr != 0) 1555 { 1556 if (warc_tmp != NULL) 1557 fclose (warc_tmp); 1558 return warcerr; 1559 } 1560 } 1561 1562 if (fp != NULL) 1563 { 1564 /* This confuses the timestamping code that checks for file size. 1565 #### The timestamping code should be smarter about file size. */ 1566 if (opt.save_headers && hs->restval == 0) 1567 fwrite (head, 1, strlen (head), fp); 1568 } 1569 1570 /* Read the response body. */ 1571 int flags = 0; 1572 if (contlen != -1) 1573 /* If content-length is present, read that much; otherwise, read 1574 until EOF. The HTTP spec doesn't require the server to 1575 actually close the connection when it's done sending data. */ 1576 flags |= rb_read_exactly; 1577 if (fp != NULL && hs->restval > 0 && contrange == 0) 1578 /* If the server ignored our range request, instruct fd_read_body 1579 to skip the first RESTVAL bytes of body. */ 1580 flags |= rb_skip_startpos; 1581 if (chunked_transfer_encoding) 1582 flags |= rb_chunked_transfer_encoding; 1583 1584 hs->len = hs->restval; 1585 hs->rd_size = 0; 1586 /* Download the response body and write it to fp. 1587 If we are working on a WARC file, we simultaneously write the 1588 response body to warc_tmp. */ 1589 hs->res = fd_read_body (hs->local_file, sock, fp, contlen != -1 ? contlen : 0, 1590 hs->restval, &hs->rd_size, &hs->len, &hs->dltime, 1591 flags, warc_tmp); 1592 if (hs->res >= 0) 1593 { 1594 if (warc_tmp != NULL) 1595 { 1596 /* Create a response record and write it to the WARC file. 1597 Note: per the WARC standard, the request and response should share 1598 the same date header. We re-use the timestamp of the request. 1599 The response record should also refer to the uuid of the request. */ 1600 bool r = warc_write_response_record (url, warc_timestamp_str, 1601 warc_request_uuid, warc_ip, 1602 warc_tmp, warc_payload_offset, 1603 type, statcode, hs->newloc); 1604 1605 /* warc_write_response_record has closed warc_tmp. */ 1606 1607 if (! r) 1608 return WARC_ERR; 1609 } 1610 1611 return RETRFINISHED; 1612 } 1613 1614 if (warc_tmp != NULL) 1615 fclose (warc_tmp); 1616 1617 if (hs->res == -2) 1618 { 1619 /* Error while writing to fd. */ 1620 return FWRITEERR; 1621 } 1622 else if (hs->res == -3) 1623 { 1624 /* Error while writing to warc_tmp. */ 1625 return WARC_TMP_FWRITEERR; 1626 } 1627 else 1628 { 1629 /* A read error! */ 1630 hs->rderrmsg = xstrdup (fd_errstr (sock)); 1631 return RETRFINISHED; 1632 } 1633} 1634 1635#define BEGINS_WITH(line, string_constant) \ 1636 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \ 1637 && (c_isspace (line[sizeof (string_constant) - 1]) \ 1638 || !line[sizeof (string_constant) - 1])) 1639 1640#ifdef __VMS 1641#define SET_USER_AGENT(req) do { \ 1642 if (!opt.useragent) \ 1643 request_set_header (req, "User-Agent", \ 1644 aprintf ("Wget/%s (VMS %s %s)", \ 1645 version_string, vms_arch(), vms_vers()), \ 1646 rel_value); \ 1647 else if (*opt.useragent) \ 1648 request_set_header (req, "User-Agent", opt.useragent, rel_none); \ 1649} while (0) 1650#else /* def __VMS */ 1651#define SET_USER_AGENT(req) do { \ 1652 if (!opt.useragent) \ 1653 request_set_header (req, "User-Agent", \ 1654 aprintf ("Wget/%s (%s)", \ 1655 version_string, OS_TYPE), \ 1656 rel_value); \ 1657 else if (*opt.useragent) \ 1658 request_set_header (req, "User-Agent", opt.useragent, rel_none); \ 1659} while (0) 1660#endif /* def __VMS [else] */ 1661 1662/* Retrieve a document through HTTP protocol. It recognizes status 1663 code, and correctly handles redirections. It closes the network 1664 socket. If it receives an error from the functions below it, it 1665 will print it if there is enough information to do so (almost 1666 always), returning the error to the caller (i.e. http_loop). 1667 1668 Various HTTP parameters are stored to hs. 1669 1670 If PROXY is non-NULL, the connection will be made to the proxy 1671 server, and u->url will be requested. */ 1672static uerr_t 1673gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy, 1674 struct iri *iri, int count) 1675{ 1676 struct request *req; 1677 1678 char *type; 1679 char *user, *passwd; 1680 char *proxyauth; 1681 int statcode; 1682 int write_error; 1683 wgint contlen, contrange; 1684 struct url *conn; 1685 FILE *fp; 1686 int err; 1687 1688 int sock = -1; 1689 1690 /* Set to 1 when the authorization has already been sent and should 1691 not be tried again. */ 1692 bool auth_finished = false; 1693 1694 /* Set to 1 when just globally-set Basic authorization has been sent; 1695 * should prevent further Basic negotiations, but not other 1696 * mechanisms. */ 1697 bool basic_auth_finished = false; 1698 1699 /* Whether NTLM authentication is used for this request. */ 1700 bool ntlm_seen = false; 1701 1702 /* Whether our connection to the remote host is through SSL. */ 1703 bool using_ssl = false; 1704 1705 /* Whether a HEAD request will be issued (as opposed to GET or 1706 POST). */ 1707 bool head_only = !!(*dt & HEAD_ONLY); 1708 1709 char *head; 1710 struct response *resp; 1711 char hdrval[512]; 1712 char *message; 1713 1714 /* Declare WARC variables. */ 1715 bool warc_enabled = (opt.warc_filename != NULL); 1716 FILE *warc_tmp = NULL; 1717 char warc_timestamp_str [21]; 1718 char warc_request_uuid [48]; 1719 ip_address *warc_ip = NULL; 1720 off_t warc_payload_offset = -1; 1721 1722 /* Whether this connection will be kept alive after the HTTP request 1723 is done. */ 1724 bool keep_alive; 1725 1726 /* Is the server using the chunked transfer encoding? */ 1727 bool chunked_transfer_encoding = false; 1728 1729 /* Whether keep-alive should be inhibited. */ 1730 bool inhibit_keep_alive = 1731 !opt.http_keep_alive || opt.ignore_length; 1732 1733 /* Headers sent when using POST. */ 1734 wgint body_data_size = 0; 1735 1736 bool host_lookup_failed = false; 1737 1738#ifdef HAVE_SSL 1739 if (u->scheme == SCHEME_HTTPS) 1740 { 1741 /* Initialize the SSL context. After this has once been done, 1742 it becomes a no-op. */ 1743 if (!ssl_init ()) 1744 { 1745 scheme_disable (SCHEME_HTTPS); 1746 logprintf (LOG_NOTQUIET, 1747 _("Disabling SSL due to encountered errors.\n")); 1748 return SSLINITFAILED; 1749 } 1750 } 1751#endif /* HAVE_SSL */ 1752 1753 /* Initialize certain elements of struct http_stat. */ 1754 hs->len = 0; 1755 hs->contlen = -1; 1756 hs->res = -1; 1757 hs->rderrmsg = NULL; 1758 hs->newloc = NULL; 1759 hs->remote_time = NULL; 1760 hs->error = NULL; 1761 hs->message = NULL; 1762 1763 conn = u; 1764 1765 /* Prepare the request to send. */ 1766 { 1767 char *meth_arg; 1768 const char *meth = "GET"; 1769 if (head_only) 1770 meth = "HEAD"; 1771 else if (opt.method) 1772 meth = opt.method; 1773 /* Use the full path, i.e. one that includes the leading slash and 1774 the query string. E.g. if u->path is "foo/bar" and u->query is 1775 "param=value", full_path will be "/foo/bar?param=value". */ 1776 if (proxy 1777#ifdef HAVE_SSL 1778 /* When using SSL over proxy, CONNECT establishes a direct 1779 connection to the HTTPS server. Therefore use the same 1780 argument as when talking to the server directly. */ 1781 && u->scheme != SCHEME_HTTPS 1782#endif 1783 ) 1784 meth_arg = xstrdup (u->url); 1785 else 1786 meth_arg = url_full_path (u); 1787 req = request_new (meth, meth_arg); 1788 } 1789 1790 request_set_header (req, "Referer", (char *) hs->referer, rel_none); 1791 if (*dt & SEND_NOCACHE) 1792 { 1793 /* Cache-Control MUST be obeyed by all HTTP/1.1 caching mechanisms... */ 1794 request_set_header (req, "Cache-Control", "no-cache, must-revalidate", rel_none); 1795 1796 /* ... but some HTTP/1.0 caches doesn't implement Cache-Control. */ 1797 request_set_header (req, "Pragma", "no-cache", rel_none); 1798 } 1799 if (hs->restval) 1800 request_set_header (req, "Range", 1801 aprintf ("bytes=%s-", 1802 number_to_static_string (hs->restval)), 1803 rel_value); 1804 SET_USER_AGENT (req); 1805 request_set_header (req, "Accept", "*/*", rel_none); 1806 1807 /* Find the username and password for authentication. */ 1808 user = u->user; 1809 passwd = u->passwd; 1810 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0); 1811 user = user ? user : (opt.http_user ? opt.http_user : opt.user); 1812 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd); 1813 1814 /* We only do "site-wide" authentication with "global" user/password 1815 * values unless --auth-no-challange has been requested; URL user/password 1816 * info overrides. */ 1817 if (user && passwd && (!u->user || opt.auth_without_challenge)) 1818 { 1819 /* If this is a host for which we've already received a Basic 1820 * challenge, we'll go ahead and send Basic authentication creds. */ 1821 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req); 1822 } 1823 1824 /* Generate the Host header, HOST:PORT. Take into account that: 1825 1826 - Broken server-side software often doesn't recognize the PORT 1827 argument, so we must generate "Host: www.server.com" instead of 1828 "Host: www.server.com:80" (and likewise for https port). 1829 1830 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234" 1831 becomes ambiguous and needs to be rewritten as "Host: 1832 [3ffe:8100:200:2::2]:1234". */ 1833 { 1834 /* Formats arranged for hfmt[add_port][add_squares]. */ 1835 static const char *hfmt[][2] = { 1836 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" } 1837 }; 1838 int add_port = u->port != scheme_default_port (u->scheme); 1839 int add_squares = strchr (u->host, ':') != NULL; 1840 request_set_header (req, "Host", 1841 aprintf (hfmt[add_port][add_squares], u->host, u->port), 1842 rel_value); 1843 } 1844 1845 if (inhibit_keep_alive) 1846 request_set_header (req, "Connection", "Close", rel_none); 1847 else 1848 { 1849 request_set_header (req, "Connection", "Keep-Alive", rel_none); 1850 if (proxy) 1851 request_set_header (req, "Proxy-Connection", "Keep-Alive", rel_none); 1852 } 1853 1854 if (opt.method) 1855 { 1856 1857 if (opt.body_data || opt.body_file) 1858 { 1859 request_set_header (req, "Content-Type", 1860 "application/x-www-form-urlencoded", rel_none); 1861 1862 if (opt.body_data) 1863 body_data_size = strlen (opt.body_data); 1864 else 1865 { 1866 body_data_size = file_size (opt.body_file); 1867 if (body_data_size == -1) 1868 { 1869 logprintf (LOG_NOTQUIET, _("BODY data file %s missing: %s\n"), 1870 quote (opt.body_file), strerror (errno)); 1871 return FILEBADFILE; 1872 } 1873 } 1874 request_set_header (req, "Content-Length", 1875 xstrdup (number_to_static_string (body_data_size)), 1876 rel_value); 1877 } 1878 } 1879 1880 retry_with_auth: 1881 /* We need to come back here when the initial attempt to retrieve 1882 without authorization header fails. (Expected to happen at least 1883 for the Digest authorization scheme.) */ 1884 1885 if (opt.cookies) 1886 request_set_header (req, "Cookie", 1887 cookie_header (wget_cookie_jar, 1888 u->host, u->port, u->path, 1889#ifdef HAVE_SSL 1890 u->scheme == SCHEME_HTTPS 1891#else 1892 0 1893#endif 1894 ), 1895 rel_value); 1896 1897 /* Add the user headers. */ 1898 if (opt.user_headers) 1899 { 1900 int i; 1901 for (i = 0; opt.user_headers[i]; i++) 1902 request_set_user_header (req, opt.user_headers[i]); 1903 } 1904 1905 proxyauth = NULL; 1906 if (proxy) 1907 { 1908 char *proxy_user, *proxy_passwd; 1909 /* For normal username and password, URL components override 1910 command-line/wgetrc parameters. With proxy 1911 authentication, it's the reverse, because proxy URLs are 1912 normally the "permanent" ones, so command-line args 1913 should take precedence. */ 1914 if (opt.proxy_user && opt.proxy_passwd) 1915 { 1916 proxy_user = opt.proxy_user; 1917 proxy_passwd = opt.proxy_passwd; 1918 } 1919 else 1920 { 1921 proxy_user = proxy->user; 1922 proxy_passwd = proxy->passwd; 1923 } 1924 /* #### This does not appear right. Can't the proxy request, 1925 say, `Digest' authentication? */ 1926 if (proxy_user && proxy_passwd) 1927 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd); 1928 1929 /* If we're using a proxy, we will be connecting to the proxy 1930 server. */ 1931 conn = proxy; 1932 1933 /* Proxy authorization over SSL is handled below. */ 1934#ifdef HAVE_SSL 1935 if (u->scheme != SCHEME_HTTPS) 1936#endif 1937 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value); 1938 } 1939 1940 keep_alive = true; 1941 1942 /* Establish the connection. */ 1943 1944 if (inhibit_keep_alive) 1945 keep_alive = false; 1946 else 1947 { 1948 /* Look for a persistent connection to target host, unless a 1949 proxy is used. The exception is when SSL is in use, in which 1950 case the proxy is nothing but a passthrough to the target 1951 host, registered as a connection to the latter. */ 1952 struct url *relevant = conn; 1953#ifdef HAVE_SSL 1954 if (u->scheme == SCHEME_HTTPS) 1955 relevant = u; 1956#endif 1957 1958 if (persistent_available_p (relevant->host, relevant->port, 1959#ifdef HAVE_SSL 1960 relevant->scheme == SCHEME_HTTPS, 1961#else 1962 0, 1963#endif 1964 &host_lookup_failed)) 1965 { 1966 int family = socket_family (pconn.socket, ENDPOINT_PEER); 1967 sock = pconn.socket; 1968 using_ssl = pconn.ssl; 1969#if ENABLE_IPV6 1970 if (family == AF_INET6) 1971 logprintf (LOG_VERBOSE, _("Reusing existing connection to [%s]:%d.\n"), 1972 quotearg_style (escape_quoting_style, pconn.host), 1973 pconn.port); 1974 else 1975#endif 1976 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"), 1977 quotearg_style (escape_quoting_style, pconn.host), 1978 pconn.port); 1979 DEBUGP (("Reusing fd %d.\n", sock)); 1980 if (pconn.authorized) 1981 /* If the connection is already authorized, the "Basic" 1982 authorization added by code above is unnecessary and 1983 only hurts us. */ 1984 request_remove_header (req, "Authorization"); 1985 } 1986 else if (host_lookup_failed) 1987 { 1988 request_free (req); 1989 logprintf(LOG_NOTQUIET, 1990 _("%s: unable to resolve host address %s\n"), 1991 exec_name, quote (relevant->host)); 1992 return HOSTERR; 1993 } 1994 else if (sock != -1) 1995 { 1996 sock = -1; 1997 } 1998 } 1999 2000 if (sock < 0) 2001 { 2002 sock = connect_to_host (conn->host, conn->port); 2003 if (sock == E_HOST) 2004 { 2005 request_free (req); 2006 return HOSTERR; 2007 } 2008 else if (sock < 0) 2009 { 2010 request_free (req); 2011 return (retryable_socket_connect_error (errno) 2012 ? CONERROR : CONIMPOSSIBLE); 2013 } 2014 2015#ifdef HAVE_SSL 2016 if (proxy && u->scheme == SCHEME_HTTPS) 2017 { 2018 /* When requesting SSL URLs through proxies, use the 2019 CONNECT method to request passthrough. */ 2020 struct request *connreq = request_new ("CONNECT", 2021 aprintf ("%s:%d", u->host, u->port)); 2022 SET_USER_AGENT (connreq); 2023 if (proxyauth) 2024 { 2025 request_set_header (connreq, "Proxy-Authorization", 2026 proxyauth, rel_value); 2027 /* Now that PROXYAUTH is part of the CONNECT request, 2028 zero it out so we don't send proxy authorization with 2029 the regular request below. */ 2030 proxyauth = NULL; 2031 } 2032 request_set_header (connreq, "Host", 2033 aprintf ("%s:%d", u->host, u->port), 2034 rel_value); 2035 2036 write_error = request_send (connreq, sock, 0); 2037 request_free (connreq); 2038 if (write_error < 0) 2039 { 2040 CLOSE_INVALIDATE (sock); 2041 request_free (req); 2042 return WRITEFAILED; 2043 } 2044 2045 head = read_http_response_head (sock); 2046 if (!head) 2047 { 2048 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"), 2049 fd_errstr (sock)); 2050 CLOSE_INVALIDATE (sock); 2051 request_free (req); 2052 return HERR; 2053 } 2054 message = NULL; 2055 if (!*head) 2056 { 2057 xfree (head); 2058 goto failed_tunnel; 2059 } 2060 DEBUGP (("proxy responded with: [%s]\n", head)); 2061 2062 resp = resp_new (head); 2063 statcode = resp_status (resp, &message); 2064 if (statcode < 0) 2065 { 2066 char *tms = datetime_str (time (NULL)); 2067 logprintf (LOG_VERBOSE, "%d\n", statcode); 2068 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode, 2069 quotearg_style (escape_quoting_style, 2070 _("Malformed status line"))); 2071 xfree (head); 2072 request_free (req); 2073 return HERR; 2074 } 2075 hs->message = xstrdup (message); 2076 resp_free (resp); 2077 xfree (head); 2078 if (statcode != 200) 2079 { 2080 failed_tunnel: 2081 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"), 2082 message ? quotearg_style (escape_quoting_style, message) : "?"); 2083 xfree_null (message); 2084 request_free (req); 2085 return CONSSLERR; 2086 } 2087 xfree_null (message); 2088 2089 /* SOCK is now *really* connected to u->host, so update CONN 2090 to reflect this. That way register_persistent will 2091 register SOCK as being connected to u->host:u->port. */ 2092 conn = u; 2093 } 2094 2095 if (conn->scheme == SCHEME_HTTPS) 2096 { 2097 if (!ssl_connect_wget (sock, u->host)) 2098 { 2099 fd_close (sock); 2100 request_free (req); 2101 return CONSSLERR; 2102 } 2103 else if (!ssl_check_certificate (sock, u->host)) 2104 { 2105 fd_close (sock); 2106 request_free (req); 2107 return VERIFCERTERR; 2108 } 2109 using_ssl = true; 2110 } 2111#endif /* HAVE_SSL */ 2112 } 2113 2114 /* Open the temporary file where we will write the request. */ 2115 if (warc_enabled) 2116 { 2117 warc_tmp = warc_tempfile (); 2118 if (warc_tmp == NULL) 2119 { 2120 CLOSE_INVALIDATE (sock); 2121 request_free (req); 2122 return WARC_TMP_FOPENERR; 2123 } 2124 2125 if (! proxy) 2126 { 2127 warc_ip = (ip_address *) alloca (sizeof (ip_address)); 2128 socket_ip_address (sock, warc_ip, ENDPOINT_PEER); 2129 } 2130 } 2131 2132 /* Send the request to server. */ 2133 write_error = request_send (req, sock, warc_tmp); 2134 2135 if (write_error >= 0) 2136 { 2137 if (opt.body_data) 2138 { 2139 DEBUGP (("[BODY data: %s]\n", opt.body_data)); 2140 write_error = fd_write (sock, opt.body_data, body_data_size, -1); 2141 if (write_error >= 0 && warc_tmp != NULL) 2142 { 2143 /* Remember end of headers / start of payload. */ 2144 warc_payload_offset = ftello (warc_tmp); 2145 2146 /* Write a copy of the data to the WARC record. */ 2147 int warc_tmp_written = fwrite (opt.body_data, 1, body_data_size, warc_tmp); 2148 if (warc_tmp_written != body_data_size) 2149 write_error = -2; 2150 } 2151 } 2152 else if (opt.body_file && body_data_size != 0) 2153 { 2154 if (warc_tmp != NULL) 2155 /* Remember end of headers / start of payload */ 2156 warc_payload_offset = ftello (warc_tmp); 2157 2158 write_error = body_file_send (sock, opt.body_file, body_data_size, warc_tmp); 2159 } 2160 } 2161 2162 if (write_error < 0) 2163 { 2164 CLOSE_INVALIDATE (sock); 2165 request_free (req); 2166 2167 if (warc_tmp != NULL) 2168 fclose (warc_tmp); 2169 2170 if (write_error == -2) 2171 return WARC_TMP_FWRITEERR; 2172 else 2173 return WRITEFAILED; 2174 } 2175 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "), 2176 proxy ? "Proxy" : "HTTP"); 2177 contlen = -1; 2178 contrange = 0; 2179 *dt &= ~RETROKF; 2180 2181 2182 if (warc_enabled) 2183 { 2184 bool warc_result; 2185 /* Generate a timestamp and uuid for this request. */ 2186 warc_timestamp (warc_timestamp_str); 2187 warc_uuid_str (warc_request_uuid); 2188 2189 /* Create a request record and store it in the WARC file. */ 2190 warc_result = warc_write_request_record (u->url, warc_timestamp_str, 2191 warc_request_uuid, warc_ip, 2192 warc_tmp, warc_payload_offset); 2193 if (! warc_result) 2194 { 2195 CLOSE_INVALIDATE (sock); 2196 request_free (req); 2197 return WARC_ERR; 2198 } 2199 2200 /* warc_write_request_record has also closed warc_tmp. */ 2201 } 2202 2203 2204read_header: 2205 head = read_http_response_head (sock); 2206 if (!head) 2207 { 2208 if (errno == 0) 2209 { 2210 logputs (LOG_NOTQUIET, _("No data received.\n")); 2211 CLOSE_INVALIDATE (sock); 2212 request_free (req); 2213 return HEOF; 2214 } 2215 else 2216 { 2217 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"), 2218 fd_errstr (sock)); 2219 CLOSE_INVALIDATE (sock); 2220 request_free (req); 2221 return HERR; 2222 } 2223 } 2224 DEBUGP (("\n---response begin---\n%s---response end---\n", head)); 2225 2226 resp = resp_new (head); 2227 2228 /* Check for status line. */ 2229 message = NULL; 2230 statcode = resp_status (resp, &message); 2231 if (statcode < 0) 2232 { 2233 char *tms = datetime_str (time (NULL)); 2234 logprintf (LOG_VERBOSE, "%d\n", statcode); 2235 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode, 2236 quotearg_style (escape_quoting_style, 2237 _("Malformed status line"))); 2238 CLOSE_INVALIDATE (sock); 2239 resp_free (resp); 2240 request_free (req); 2241 xfree (head); 2242 return HERR; 2243 } 2244 2245 if (H_10X (statcode)) 2246 { 2247 DEBUGP (("Ignoring response\n")); 2248 resp_free (resp); 2249 xfree (head); 2250 goto read_header; 2251 } 2252 2253 hs->message = xstrdup (message); 2254 if (!opt.server_response) 2255 logprintf (LOG_VERBOSE, "%2d %s\n", statcode, 2256 message ? quotearg_style (escape_quoting_style, message) : ""); 2257 else 2258 { 2259 logprintf (LOG_VERBOSE, "\n"); 2260 print_server_response (resp, " "); 2261 } 2262 2263 if (!opt.ignore_length 2264 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval))) 2265 { 2266 wgint parsed; 2267 errno = 0; 2268 parsed = str_to_wgint (hdrval, NULL, 10); 2269 if (parsed == WGINT_MAX && errno == ERANGE) 2270 { 2271 /* Out of range. 2272 #### If Content-Length is out of range, it most likely 2273 means that the file is larger than 2G and that we're 2274 compiled without LFS. In that case we should probably 2275 refuse to even attempt to download the file. */ 2276 contlen = -1; 2277 } 2278 else if (parsed < 0) 2279 { 2280 /* Negative Content-Length; nonsensical, so we can't 2281 assume any information about the content to receive. */ 2282 contlen = -1; 2283 } 2284 else 2285 contlen = parsed; 2286 } 2287 2288 /* Check for keep-alive related responses. */ 2289 if (!inhibit_keep_alive && contlen != -1) 2290 { 2291 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval))) 2292 { 2293 if (0 == strcasecmp (hdrval, "Close")) 2294 keep_alive = false; 2295 } 2296 } 2297 2298 chunked_transfer_encoding = false; 2299 if (resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval)) 2300 && 0 == strcasecmp (hdrval, "chunked")) 2301 chunked_transfer_encoding = true; 2302 2303 /* Handle (possibly multiple instances of) the Set-Cookie header. */ 2304 if (opt.cookies) 2305 { 2306 int scpos; 2307 const char *scbeg, *scend; 2308 /* The jar should have been created by now. */ 2309 assert (wget_cookie_jar != NULL); 2310 for (scpos = 0; 2311 (scpos = resp_header_locate (resp, "Set-Cookie", scpos, 2312 &scbeg, &scend)) != -1; 2313 ++scpos) 2314 { 2315 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie); 2316 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, 2317 u->path, set_cookie); 2318 } 2319 } 2320 2321 if (keep_alive) 2322 /* The server has promised that it will not close the connection 2323 when we're done. This means that we can register it. */ 2324 register_persistent (conn->host, conn->port, sock, using_ssl); 2325 2326 if (statcode == HTTP_STATUS_UNAUTHORIZED) 2327 { 2328 /* Authorization is required. */ 2329 2330 /* Normally we are not interested in the response body. 2331 But if we are writing a WARC file we are: we like to keep everyting. */ 2332 if (warc_enabled) 2333 { 2334 int _err; 2335 type = resp_header_strdup (resp, "Content-Type"); 2336 _err = read_response_body (hs, sock, NULL, contlen, 0, 2337 chunked_transfer_encoding, 2338 u->url, warc_timestamp_str, 2339 warc_request_uuid, warc_ip, type, 2340 statcode, head); 2341 xfree_null (type); 2342 2343 if (_err != RETRFINISHED || hs->res < 0) 2344 { 2345 CLOSE_INVALIDATE (sock); 2346 request_free (req); 2347 xfree_null (message); 2348 resp_free (resp); 2349 xfree (head); 2350 return _err; 2351 } 2352 else 2353 CLOSE_FINISH (sock); 2354 } 2355 else 2356 { 2357 /* Since WARC is disabled, we are not interested in the response body. */ 2358 if (keep_alive && !head_only 2359 && skip_short_body (sock, contlen, chunked_transfer_encoding)) 2360 CLOSE_FINISH (sock); 2361 else 2362 CLOSE_INVALIDATE (sock); 2363 } 2364 2365 pconn.authorized = false; 2366 uerr_t auth_err = RETROK; 2367 if (!auth_finished && (user && passwd)) 2368 { 2369 /* IIS sends multiple copies of WWW-Authenticate, one with 2370 the value "negotiate", and other(s) with data. Loop over 2371 all the occurrences and pick the one we recognize. */ 2372 int wapos; 2373 const char *wabeg, *waend; 2374 char *www_authenticate = NULL; 2375 for (wapos = 0; 2376 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos, 2377 &wabeg, &waend)) != -1; 2378 ++wapos) 2379 if (known_authentication_scheme_p (wabeg, waend)) 2380 { 2381 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate); 2382 break; 2383 } 2384 2385 if (!www_authenticate) 2386 { 2387 /* If the authentication header is missing or 2388 unrecognized, there's no sense in retrying. */ 2389 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n")); 2390 } 2391 else if (!basic_auth_finished 2392 || !BEGINS_WITH (www_authenticate, "Basic")) 2393 { 2394 char *pth = url_full_path (u); 2395 const char *value; 2396 uerr_t *auth_stat; 2397 auth_stat = xmalloc (sizeof (uerr_t)); 2398 *auth_stat = RETROK; 2399 2400 value = create_authorization_line (www_authenticate, 2401 user, passwd, 2402 request_method (req), 2403 pth, 2404 &auth_finished, 2405 auth_stat); 2406 2407 auth_err = *auth_stat; 2408 if (auth_err == RETROK) 2409 { 2410 request_set_header (req, "Authorization", value, rel_value); 2411 2412 if (BEGINS_WITH (www_authenticate, "NTLM")) 2413 ntlm_seen = true; 2414 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic")) 2415 { 2416 /* Need to register this host as using basic auth, 2417 * so we automatically send creds next time. */ 2418 register_basic_auth_host (u->host); 2419 } 2420 2421 xfree (pth); 2422 xfree_null (message); 2423 resp_free (resp); 2424 xfree (head); 2425 xfree (auth_stat); 2426 xfree (hs->message); 2427 goto retry_with_auth; 2428 } 2429 else 2430 { 2431 /* Creating the Authorization header went wrong */ 2432 } 2433 } 2434 else 2435 { 2436 /* We already did Basic auth, and it failed. Gotta 2437 * give up. */ 2438 } 2439 } 2440 request_free (req); 2441 xfree_null (message); 2442 resp_free (resp); 2443 xfree (head); 2444 if (auth_err == RETROK) 2445 return AUTHFAILED; 2446 else 2447 return auth_err; 2448 } 2449 else /* statcode != HTTP_STATUS_UNAUTHORIZED */ 2450 { 2451 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */ 2452 if (ntlm_seen) 2453 pconn.authorized = true; 2454 } 2455 2456 /* Determine the local filename if needed. Notice that if -O is used 2457 * hstat.local_file is set by http_loop to the argument of -O. */ 2458 if (!hs->local_file) 2459 { 2460 char *local_file = NULL; 2461 2462 /* Honor Content-Disposition whether possible. */ 2463 if (!opt.content_disposition 2464 || !resp_header_copy (resp, "Content-Disposition", 2465 hdrval, sizeof (hdrval)) 2466 || !parse_content_disposition (hdrval, &local_file)) 2467 { 2468 /* The Content-Disposition header is missing or broken. 2469 * Choose unique file name according to given URL. */ 2470 hs->local_file = url_file_name (u, NULL); 2471 } 2472 else 2473 { 2474 DEBUGP (("Parsed filename from Content-Disposition: %s\n", 2475 local_file)); 2476 hs->local_file = url_file_name (u, local_file); 2477 } 2478 2479 xfree_null (local_file); 2480 } 2481 2482 /* TODO: perform this check only once. */ 2483 if (!hs->existence_checked && file_exists_p (hs->local_file)) 2484 { 2485 if (opt.noclobber && !opt.output_document) 2486 { 2487 /* If opt.noclobber is turned on and file already exists, do not 2488 retrieve the file. But if the output_document was given, then this 2489 test was already done and the file didn't exist. Hence the !opt.output_document */ 2490 get_file_flags (hs->local_file, dt); 2491 request_free (req); 2492 resp_free (resp); 2493 xfree (head); 2494 xfree_null (message); 2495 return RETRUNNEEDED; 2496 } 2497 else if (!ALLOW_CLOBBER) 2498 { 2499 char *unique = unique_name (hs->local_file, true); 2500 if (unique != hs->local_file) 2501 xfree (hs->local_file); 2502 hs->local_file = unique; 2503 } 2504 } 2505 hs->existence_checked = true; 2506 2507 /* Support timestamping */ 2508 /* TODO: move this code out of gethttp. */ 2509 if (opt.timestamping && !hs->timestamp_checked) 2510 { 2511 size_t filename_len = strlen (hs->local_file); 2512 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX)); 2513 bool local_dot_orig_file_exists = false; 2514 char *local_filename = NULL; 2515 struct_stat st; 2516 2517 if (opt.backup_converted) 2518 /* If -K is specified, we'll act on the assumption that it was specified 2519 last time these files were downloaded as well, and instead of just 2520 comparing local file X against server file X, we'll compare local 2521 file X.orig (if extant, else X) against server file X. If -K 2522 _wasn't_ specified last time, or the server contains files called 2523 *.orig, -N will be back to not operating correctly with -k. */ 2524 { 2525 /* Would a single s[n]printf() call be faster? --dan 2526 2527 Definitely not. sprintf() is horribly slow. It's a 2528 different question whether the difference between the two 2529 affects a program. Usually I'd say "no", but at one 2530 point I profiled Wget, and found that a measurable and 2531 non-negligible amount of time was lost calling sprintf() 2532 in url.c. Replacing sprintf with inline calls to 2533 strcpy() and number_to_string() made a difference. 2534 --hniksic */ 2535 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len); 2536 memcpy (filename_plus_orig_suffix + filename_len, 2537 ORIG_SFX, sizeof (ORIG_SFX)); 2538 2539 /* Try to stat() the .orig file. */ 2540 if (stat (filename_plus_orig_suffix, &st) == 0) 2541 { 2542 local_dot_orig_file_exists = true; 2543 local_filename = filename_plus_orig_suffix; 2544 } 2545 } 2546 2547 if (!local_dot_orig_file_exists) 2548 /* Couldn't stat() <file>.orig, so try to stat() <file>. */ 2549 if (stat (hs->local_file, &st) == 0) 2550 local_filename = hs->local_file; 2551 2552 if (local_filename != NULL) 2553 /* There was a local file, so we'll check later to see if the version 2554 the server has is the same version we already have, allowing us to 2555 skip a download. */ 2556 { 2557 hs->orig_file_name = xstrdup (local_filename); 2558 hs->orig_file_size = st.st_size; 2559 hs->orig_file_tstamp = st.st_mtime; 2560#ifdef WINDOWS 2561 /* Modification time granularity is 2 seconds for Windows, so 2562 increase local time by 1 second for later comparison. */ 2563 ++hs->orig_file_tstamp; 2564#endif 2565 } 2566 } 2567 2568 request_free (req); 2569 2570 hs->statcode = statcode; 2571 if (statcode == -1) 2572 hs->error = xstrdup (_("Malformed status line")); 2573 else if (!*message) 2574 hs->error = xstrdup (_("(no description)")); 2575 else 2576 hs->error = xstrdup (message); 2577 xfree_null (message); 2578 2579 type = resp_header_strdup (resp, "Content-Type"); 2580 if (type) 2581 { 2582 char *tmp = strchr (type, ';'); 2583 if (tmp) 2584 { 2585 /* sXXXav: only needed if IRI support is enabled */ 2586 char *tmp2 = tmp + 1; 2587 2588 while (tmp > type && c_isspace (tmp[-1])) 2589 --tmp; 2590 *tmp = '\0'; 2591 2592 /* Try to get remote encoding if needed */ 2593 if (opt.enable_iri && !opt.encoding_remote) 2594 { 2595 tmp = parse_charset (tmp2); 2596 if (tmp) 2597 set_content_encoding (iri, tmp); 2598 xfree_null(tmp); 2599 } 2600 } 2601 } 2602 hs->newloc = resp_header_strdup (resp, "Location"); 2603 hs->remote_time = resp_header_strdup (resp, "Last-Modified"); 2604 2605 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval))) 2606 { 2607 wgint first_byte_pos, last_byte_pos, entity_length; 2608 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos, 2609 &entity_length)) 2610 { 2611 contrange = first_byte_pos; 2612 contlen = last_byte_pos - first_byte_pos + 1; 2613 } 2614 } 2615 resp_free (resp); 2616 2617 /* 20x responses are counted among successful by default. */ 2618 if (H_20X (statcode)) 2619 *dt |= RETROKF; 2620 2621 if (statcode == HTTP_STATUS_NO_CONTENT) 2622 { 2623 /* 204 response has no body (RFC 2616, 4.3) */ 2624 2625 /* In case the caller cares to look... */ 2626 hs->len = 0; 2627 hs->res = 0; 2628 hs->restval = 0; 2629 2630 CLOSE_FINISH (sock); 2631 xfree_null (type); 2632 xfree (head); 2633 2634 return RETRFINISHED; 2635 } 2636 2637 /* Return if redirected. */ 2638 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES) 2639 { 2640 /* RFC2068 says that in case of the 300 (multiple choices) 2641 response, the server can output a preferred URL through 2642 `Location' header; otherwise, the request should be treated 2643 like GET. So, if the location is set, it will be a 2644 redirection; otherwise, just proceed normally. */ 2645 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc) 2646 *dt |= RETROKF; 2647 else 2648 { 2649 logprintf (LOG_VERBOSE, 2650 _("Location: %s%s\n"), 2651 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"), 2652 hs->newloc ? _(" [following]") : ""); 2653 2654 /* In case the caller cares to look... */ 2655 hs->len = 0; 2656 hs->res = 0; 2657 hs->restval = 0; 2658 2659 /* Normally we are not interested in the response body of a redirect. 2660 But if we are writing a WARC file we are: we like to keep everyting. */ 2661 if (warc_enabled) 2662 { 2663 int _err = read_response_body (hs, sock, NULL, contlen, 0, 2664 chunked_transfer_encoding, 2665 u->url, warc_timestamp_str, 2666 warc_request_uuid, warc_ip, type, 2667 statcode, head); 2668 2669 if (_err != RETRFINISHED || hs->res < 0) 2670 { 2671 CLOSE_INVALIDATE (sock); 2672 xfree_null (type); 2673 xfree (head); 2674 return _err; 2675 } 2676 else 2677 CLOSE_FINISH (sock); 2678 } 2679 else 2680 { 2681 /* Since WARC is disabled, we are not interested in the response body. */ 2682 if (keep_alive && !head_only 2683 && skip_short_body (sock, contlen, chunked_transfer_encoding)) 2684 CLOSE_FINISH (sock); 2685 else 2686 CLOSE_INVALIDATE (sock); 2687 } 2688 2689 xfree_null (type); 2690 xfree (head); 2691 /* From RFC2616: The status codes 303 and 307 have 2692 been added for servers that wish to make unambiguously 2693 clear which kind of reaction is expected of the client. 2694 2695 A 307 should be redirected using the same method, 2696 in other words, a POST should be preserved and not 2697 converted to a GET in that case. 2698 2699 With strict adherence to RFC2616, POST requests are not 2700 converted to a GET request on 301 Permanent Redirect 2701 or 302 Temporary Redirect. 2702 2703 A switch may be provided later based on the HTTPbis draft 2704 that allows clients to convert POST requests to GET 2705 requests on 301 and 302 response codes. */ 2706 switch (statcode) 2707 { 2708 case HTTP_STATUS_TEMPORARY_REDIRECT: 2709 return NEWLOCATION_KEEP_POST; 2710 case HTTP_STATUS_MOVED_PERMANENTLY: 2711 if (opt.method && strcasecmp (opt.method, "post") != 0) 2712 return NEWLOCATION_KEEP_POST; 2713 break; 2714 case HTTP_STATUS_MOVED_TEMPORARILY: 2715 if (opt.method && strcasecmp (opt.method, "post") != 0) 2716 return NEWLOCATION_KEEP_POST; 2717 break; 2718 default: 2719 return NEWLOCATION; 2720 } 2721 return NEWLOCATION; 2722 } 2723 } 2724 2725 /* If content-type is not given, assume text/html. This is because 2726 of the multitude of broken CGI's that "forget" to generate the 2727 content-type. */ 2728 if (!type || 2729 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) || 2730 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S))) 2731 *dt |= TEXTHTML; 2732 else 2733 *dt &= ~TEXTHTML; 2734 2735 if (type && 2736 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S))) 2737 *dt |= TEXTCSS; 2738 else 2739 *dt &= ~TEXTCSS; 2740 2741 if (opt.adjust_extension) 2742 { 2743 if (*dt & TEXTHTML) 2744 /* -E / --adjust-extension / adjust_extension = on was specified, 2745 and this is a text/html file. If some case-insensitive 2746 variation on ".htm[l]" isn't already the file's suffix, 2747 tack on ".html". */ 2748 { 2749 ensure_extension (hs, ".html", dt); 2750 } 2751 else if (*dt & TEXTCSS) 2752 { 2753 ensure_extension (hs, ".css", dt); 2754 } 2755 } 2756 2757 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE 2758 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK 2759 && contrange == 0 && contlen >= 0 && hs->restval >= contlen)) 2760 { 2761 /* If `-c' is in use and the file has been fully downloaded (or 2762 the remote file has shrunk), Wget effectively requests bytes 2763 after the end of file and the server response with 416 2764 (or 200 with a <= Content-Length. */ 2765 logputs (LOG_VERBOSE, _("\ 2766\n The file is already fully retrieved; nothing to do.\n\n")); 2767 /* In case the caller inspects. */ 2768 hs->len = contlen; 2769 hs->res = 0; 2770 /* Mark as successfully retrieved. */ 2771 *dt |= RETROKF; 2772 xfree_null (type); 2773 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there 2774 might be more bytes in the body. */ 2775 xfree (head); 2776 return RETRUNNEEDED; 2777 } 2778 if ((contrange != 0 && contrange != hs->restval) 2779 || (H_PARTIAL (statcode) && !contrange)) 2780 { 2781 /* The Range request was somehow misunderstood by the server. 2782 Bail out. */ 2783 xfree_null (type); 2784 CLOSE_INVALIDATE (sock); 2785 xfree (head); 2786 return RANGEERR; 2787 } 2788 if (contlen == -1) 2789 hs->contlen = -1; 2790 else 2791 hs->contlen = contlen + contrange; 2792 2793 if (opt.verbose) 2794 { 2795 if (*dt & RETROKF) 2796 { 2797 /* No need to print this output if the body won't be 2798 downloaded at all, or if the original server response is 2799 printed. */ 2800 logputs (LOG_VERBOSE, _("Length: ")); 2801 if (contlen != -1) 2802 { 2803 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange)); 2804 if (contlen + contrange >= 1024) 2805 logprintf (LOG_VERBOSE, " (%s)", 2806 human_readable (contlen + contrange, 10, 1)); 2807 if (contrange) 2808 { 2809 if (contlen >= 1024) 2810 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"), 2811 number_to_static_string (contlen), 2812 human_readable (contlen, 10, 1)); 2813 else 2814 logprintf (LOG_VERBOSE, _(", %s remaining"), 2815 number_to_static_string (contlen)); 2816 } 2817 } 2818 else 2819 logputs (LOG_VERBOSE, 2820 opt.ignore_length ? _("ignored") : _("unspecified")); 2821 if (type) 2822 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type)); 2823 else 2824 logputs (LOG_VERBOSE, "\n"); 2825 } 2826 } 2827 2828 /* Return if we have no intention of further downloading. */ 2829 if ((!(*dt & RETROKF) && !opt.content_on_error) || head_only) 2830 { 2831 /* In case the caller cares to look... */ 2832 hs->len = 0; 2833 hs->res = 0; 2834 hs->restval = 0; 2835 2836 /* Normally we are not interested in the response body of a error responses. 2837 But if we are writing a WARC file we are: we like to keep everyting. */ 2838 if (warc_enabled) 2839 { 2840 int _err = read_response_body (hs, sock, NULL, contlen, 0, 2841 chunked_transfer_encoding, 2842 u->url, warc_timestamp_str, 2843 warc_request_uuid, warc_ip, type, 2844 statcode, head); 2845 2846 if (_err != RETRFINISHED || hs->res < 0) 2847 { 2848 CLOSE_INVALIDATE (sock); 2849 xfree (head); 2850 xfree_null (type); 2851 return _err; 2852 } 2853 else 2854 CLOSE_FINISH (sock); 2855 } 2856 else 2857 { 2858 /* Since WARC is disabled, we are not interested in the response body. */ 2859 if (head_only) 2860 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the 2861 servers not to send body in response to a HEAD request, and 2862 those that do will likely be caught by test_socket_open. 2863 If not, they can be worked around using 2864 `--no-http-keep-alive'. */ 2865 CLOSE_FINISH (sock); 2866 else if (keep_alive 2867 && skip_short_body (sock, contlen, chunked_transfer_encoding)) 2868 /* Successfully skipped the body; also keep using the socket. */ 2869 CLOSE_FINISH (sock); 2870 else 2871 CLOSE_INVALIDATE (sock); 2872 } 2873 2874 xfree (head); 2875 xfree_null (type); 2876 return RETRFINISHED; 2877 } 2878 2879/* 2005-06-17 SMS. 2880 For VMS, define common fopen() optional arguments. 2881*/ 2882#ifdef __VMS 2883# define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id 2884# define FOPEN_BIN_FLAG 3 2885#else /* def __VMS */ 2886# define FOPEN_BIN_FLAG true 2887#endif /* def __VMS [else] */ 2888 2889 /* Open the local file. */ 2890 if (!output_stream) 2891 { 2892 mkalldirs (hs->local_file); 2893 if (opt.backups) 2894 rotate_backups (hs->local_file); 2895 if (hs->restval) 2896 { 2897#ifdef __VMS 2898 int open_id; 2899 2900 open_id = 21; 2901 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS); 2902#else /* def __VMS */ 2903 fp = fopen (hs->local_file, "ab"); 2904#endif /* def __VMS [else] */ 2905 } 2906 else if (ALLOW_CLOBBER || count > 0) 2907 { 2908 if (opt.unlink && file_exists_p (hs->local_file)) 2909 { 2910 if (unlink (hs->local_file) < 0) 2911 { 2912 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, 2913 strerror (errno)); 2914 CLOSE_INVALIDATE (sock); 2915 xfree (head); 2916 xfree_null (type); 2917 return UNLINKERR; 2918 } 2919 } 2920 2921#ifdef __VMS 2922 int open_id; 2923 2924 open_id = 22; 2925 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS); 2926#else /* def __VMS */ 2927 fp = fopen (hs->local_file, "wb"); 2928#endif /* def __VMS [else] */ 2929 } 2930 else 2931 { 2932 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG); 2933 if (!fp && errno == EEXIST) 2934 { 2935 /* We cannot just invent a new name and use it (which is 2936 what functions like unique_create typically do) 2937 because we told the user we'd use this name. 2938 Instead, return and retry the download. */ 2939 logprintf (LOG_NOTQUIET, 2940 _("%s has sprung into existence.\n"), 2941 hs->local_file); 2942 CLOSE_INVALIDATE (sock); 2943 xfree (head); 2944 xfree_null (type); 2945 return FOPEN_EXCL_ERR; 2946 } 2947 } 2948 if (!fp) 2949 { 2950 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno)); 2951 CLOSE_INVALIDATE (sock); 2952 xfree (head); 2953 xfree_null (type); 2954 return FOPENERR; 2955 } 2956 } 2957 else 2958 fp = output_stream; 2959 2960 /* Print fetch message, if opt.verbose. */ 2961 logprintf (LOG_VERBOSE, _("Saving to: %s\n"), 2962 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file)); 2963 2964 2965 err = read_response_body (hs, sock, fp, contlen, contrange, 2966 chunked_transfer_encoding, 2967 u->url, warc_timestamp_str, 2968 warc_request_uuid, warc_ip, type, 2969 statcode, head); 2970 2971 /* Now we no longer need to store the response header. */ 2972 xfree (head); 2973 xfree_null (type); 2974 2975 if (hs->res >= 0) 2976 CLOSE_FINISH (sock); 2977 else 2978 CLOSE_INVALIDATE (sock); 2979 2980 if (!output_stream) 2981 fclose (fp); 2982 2983 return err; 2984} 2985 2986/* The genuine HTTP loop! This is the part where the retrieval is 2987 retried, and retried, and retried, and... */ 2988uerr_t 2989http_loop (struct url *u, struct url *original_url, char **newloc, 2990 char **local_file, const char *referer, int *dt, struct url *proxy, 2991 struct iri *iri) 2992{ 2993 int count; 2994 bool got_head = false; /* used for time-stamping and filename detection */ 2995 bool time_came_from_head = false; 2996 bool got_name = false; 2997 char *tms; 2998 const char *tmrate; 2999 uerr_t err, ret = TRYLIMEXC; 3000 time_t tmr = -1; /* remote time-stamp */ 3001 struct http_stat hstat; /* HTTP status */ 3002 struct_stat st; 3003 bool send_head_first = true; 3004 char *file_name; 3005 bool force_full_retrieve = false; 3006 3007 3008 /* If we are writing to a WARC file: always retrieve the whole file. */ 3009 if (opt.warc_filename != NULL) 3010 force_full_retrieve = true; 3011 3012 3013 /* Assert that no value for *LOCAL_FILE was passed. */ 3014 assert (local_file == NULL || *local_file == NULL); 3015 3016 /* Set LOCAL_FILE parameter. */ 3017 if (local_file && opt.output_document) 3018 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document); 3019 3020 /* Reset NEWLOC parameter. */ 3021 *newloc = NULL; 3022 3023 /* This used to be done in main, but it's a better idea to do it 3024 here so that we don't go through the hoops if we're just using 3025 FTP or whatever. */ 3026 if (opt.cookies) 3027 load_cookies (); 3028 3029 /* Warn on (likely bogus) wildcard usage in HTTP. */ 3030 if (opt.ftp_glob && has_wildcards_p (u->path)) 3031 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n")); 3032 3033 /* Setup hstat struct. */ 3034 xzero (hstat); 3035 hstat.referer = referer; 3036 3037 if (opt.output_document) 3038 { 3039 hstat.local_file = xstrdup (opt.output_document); 3040 got_name = true; 3041 } 3042 else if (!opt.content_disposition) 3043 { 3044 hstat.local_file = 3045 url_file_name (opt.trustservernames ? u : original_url, NULL); 3046 got_name = true; 3047 } 3048 3049 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document) 3050 { 3051 /* If opt.noclobber is turned on and file already exists, do not 3052 retrieve the file. But if the output_document was given, then this 3053 test was already done and the file didn't exist. Hence the !opt.output_document */ 3054 get_file_flags (hstat.local_file, dt); 3055 ret = RETROK; 3056 goto exit; 3057 } 3058 3059 /* Reset the counter. */ 3060 count = 0; 3061 3062 /* Reset the document type. */ 3063 *dt = 0; 3064 3065 /* Skip preliminary HEAD request if we're not in spider mode. */ 3066 if (!opt.spider) 3067 send_head_first = false; 3068 3069 /* Send preliminary HEAD request if --content-disposition and -c are used 3070 together. */ 3071 if (opt.content_disposition && opt.always_rest) 3072 send_head_first = true; 3073 3074 /* Send preliminary HEAD request if -N is given and we have an existing 3075 * destination file. */ 3076 if (!opt.output_document) 3077 file_name = url_file_name (opt.trustservernames ? u : original_url, NULL); 3078 else 3079 file_name = xstrdup (opt.output_document); 3080 if (opt.timestamping && (file_exists_p (file_name) 3081 || opt.content_disposition)) 3082 send_head_first = true; 3083 xfree (file_name); 3084 3085 /* THE loop */ 3086 do 3087 { 3088 /* Increment the pass counter. */ 3089 ++count; 3090 sleep_between_retrievals (count); 3091 3092 /* Get the current time string. */ 3093 tms = datetime_str (time (NULL)); 3094 3095 if (opt.spider && !got_head) 3096 logprintf (LOG_VERBOSE, _("\ 3097Spider mode enabled. Check if remote file exists.\n")); 3098 3099 /* Print fetch message, if opt.verbose. */ 3100 if (opt.verbose) 3101 { 3102 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD); 3103 3104 if (count > 1) 3105 { 3106 char tmp[256]; 3107 sprintf (tmp, _("(try:%2d)"), count); 3108 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n", 3109 tms, tmp, hurl); 3110 } 3111 else 3112 { 3113 logprintf (LOG_NOTQUIET, "--%s-- %s\n", 3114 tms, hurl); 3115 } 3116 3117#ifdef WINDOWS 3118 ws_changetitle (hurl); 3119#endif 3120 xfree (hurl); 3121 } 3122 3123 /* Default document type is empty. However, if spider mode is 3124 on or time-stamping is employed, HEAD_ONLY commands is 3125 encoded within *dt. */ 3126 if (send_head_first && !got_head) 3127 *dt |= HEAD_ONLY; 3128 else 3129 *dt &= ~HEAD_ONLY; 3130 3131 /* Decide whether or not to restart. */ 3132 if (force_full_retrieve) 3133 hstat.restval = hstat.len; 3134 else if (opt.start_pos >= 0) 3135 hstat.restval = opt.start_pos; 3136 else if (opt.always_rest 3137 && got_name 3138 && stat (hstat.local_file, &st) == 0 3139 && S_ISREG (st.st_mode)) 3140 /* When -c is used, continue from on-disk size. (Can't use 3141 hstat.len even if count>1 because we don't want a failed 3142 first attempt to clobber existing data.) */ 3143 hstat.restval = st.st_size; 3144 else if (count > 1) 3145 /* otherwise, continue where the previous try left off */ 3146 hstat.restval = hstat.len; 3147 else 3148 hstat.restval = 0; 3149 3150 /* Decide whether to send the no-cache directive. We send it in 3151 two cases: 3152 a) we're using a proxy, and we're past our first retrieval. 3153 Some proxies are notorious for caching incomplete data, so 3154 we require a fresh get. 3155 b) caching is explicitly inhibited. */ 3156 if ((proxy && count > 1) /* a */ 3157 || !opt.allow_cache) /* b */ 3158 *dt |= SEND_NOCACHE; 3159 else 3160 *dt &= ~SEND_NOCACHE; 3161 3162 /* Try fetching the document, or at least its head. */ 3163 err = gethttp (u, &hstat, dt, proxy, iri, count); 3164 3165 /* Time? */ 3166 tms = datetime_str (time (NULL)); 3167 3168 /* Get the new location (with or without the redirection). */ 3169 if (hstat.newloc) 3170 *newloc = xstrdup (hstat.newloc); 3171 3172 switch (err) 3173 { 3174 case HERR: case HEOF: case CONSOCKERR: 3175 case CONERROR: case READERR: case WRITEFAILED: 3176 case RANGEERR: case FOPEN_EXCL_ERR: 3177 /* Non-fatal errors continue executing the loop, which will 3178 bring them to "while" statement at the end, to judge 3179 whether the number of tries was exceeded. */ 3180 printwhat (count, opt.ntry); 3181 continue; 3182 case FWRITEERR: case FOPENERR: 3183 /* Another fatal error. */ 3184 logputs (LOG_VERBOSE, "\n"); 3185 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"), 3186 quote (hstat.local_file), strerror (errno)); 3187 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case SSLINITFAILED: 3188 case CONTNOTSUPPORTED: case VERIFCERTERR: case FILEBADFILE: 3189 case UNKNOWNATTR: 3190 /* Fatal errors just return from the function. */ 3191 ret = err; 3192 goto exit; 3193 case ATTRMISSING: 3194 /* A missing attribute in a Header is a fatal Protocol error. */ 3195 logputs (LOG_VERBOSE, "\n"); 3196 logprintf (LOG_NOTQUIET, _("Required attribute missing from Header received.\n")); 3197 ret = err; 3198 goto exit; 3199 case AUTHFAILED: 3200 logputs (LOG_VERBOSE, "\n"); 3201 logprintf (LOG_NOTQUIET, _("Username/Password Authentication Failed.\n")); 3202 ret = err; 3203 goto exit; 3204 case WARC_ERR: 3205 /* A fatal WARC error. */ 3206 logputs (LOG_VERBOSE, "\n"); 3207 logprintf (LOG_NOTQUIET, _("Cannot write to WARC file.\n")); 3208 ret = err; 3209 goto exit; 3210 case WARC_TMP_FOPENERR: case WARC_TMP_FWRITEERR: 3211 /* A fatal WARC error. */ 3212 logputs (LOG_VERBOSE, "\n"); 3213 logprintf (LOG_NOTQUIET, _("Cannot write to temporary WARC file.\n")); 3214 ret = err; 3215 goto exit; 3216 case CONSSLERR: 3217 /* Another fatal error. */ 3218 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n")); 3219 ret = err; 3220 goto exit; 3221 case UNLINKERR: 3222 /* Another fatal error. */ 3223 logputs (LOG_VERBOSE, "\n"); 3224 logprintf (LOG_NOTQUIET, _("Cannot unlink %s (%s).\n"), 3225 quote (hstat.local_file), strerror (errno)); 3226 ret = err; 3227 goto exit; 3228 case NEWLOCATION: 3229 case NEWLOCATION_KEEP_POST: 3230 /* Return the new location to the caller. */ 3231 if (!*newloc) 3232 { 3233 logprintf (LOG_NOTQUIET, 3234 _("ERROR: Redirection (%d) without location.\n"), 3235 hstat.statcode); 3236 ret = WRONGCODE; 3237 } 3238 else 3239 { 3240 ret = err; 3241 } 3242 goto exit; 3243 case RETRUNNEEDED: 3244 /* The file was already fully retrieved. */ 3245 ret = RETROK; 3246 goto exit; 3247 case RETRFINISHED: 3248 /* Deal with you later. */ 3249 break; 3250 default: 3251 /* All possibilities should have been exhausted. */ 3252 abort (); 3253 } 3254 3255 if (!(*dt & RETROKF)) 3256 { 3257 char *hurl = NULL; 3258 if (!opt.verbose) 3259 { 3260 /* #### Ugly ugly ugly! */ 3261 hurl = url_string (u, URL_AUTH_HIDE_PASSWD); 3262 logprintf (LOG_NONVERBOSE, "%s:\n", hurl); 3263 } 3264 3265 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */ 3266 if (*dt & HEAD_ONLY 3267 && (hstat.statcode == 500 || hstat.statcode == 501)) 3268 { 3269 got_head = true; 3270 continue; 3271 } 3272 /* Maybe we should always keep track of broken links, not just in 3273 * spider mode. 3274 * Don't log error if it was UTF-8 encoded because we will try 3275 * once unencoded. */ 3276 else if (opt.spider && !iri->utf8_encode) 3277 { 3278 /* #### Again: ugly ugly ugly! */ 3279 if (!hurl) 3280 hurl = url_string (u, URL_AUTH_HIDE_PASSWD); 3281 nonexisting_url (hurl); 3282 logprintf (LOG_NOTQUIET, _("\ 3283Remote file does not exist -- broken link!!!\n")); 3284 } 3285 else 3286 { 3287 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), 3288 tms, hstat.statcode, 3289 quotearg_style (escape_quoting_style, hstat.error)); 3290 } 3291 logputs (LOG_VERBOSE, "\n"); 3292 ret = WRONGCODE; 3293 xfree_null (hurl); 3294 goto exit; 3295 } 3296 3297 /* Did we get the time-stamp? */ 3298 if (!got_head) 3299 { 3300 got_head = true; /* no more time-stamping */ 3301 3302 if (opt.timestamping && !hstat.remote_time) 3303 { 3304 logputs (LOG_NOTQUIET, _("\ 3305Last-modified header missing -- time-stamps turned off.\n")); 3306 } 3307 else if (hstat.remote_time) 3308 { 3309 /* Convert the date-string into struct tm. */ 3310 tmr = http_atotm (hstat.remote_time); 3311 if (tmr == (time_t) (-1)) 3312 logputs (LOG_VERBOSE, _("\ 3313Last-modified header invalid -- time-stamp ignored.\n")); 3314 if (*dt & HEAD_ONLY) 3315 time_came_from_head = true; 3316 } 3317 3318 if (send_head_first) 3319 { 3320 /* The time-stamping section. */ 3321 if (opt.timestamping) 3322 { 3323 if (hstat.orig_file_name) /* Perform the following 3324 checks only if the file 3325 we're supposed to 3326 download already exists. */ 3327 { 3328 if (hstat.remote_time && 3329 tmr != (time_t) (-1)) 3330 { 3331 /* Now time-stamping can be used validly. 3332 Time-stamping means that if the sizes of 3333 the local and remote file match, and local 3334 file is newer than the remote file, it will 3335 not be retrieved. Otherwise, the normal 3336 download procedure is resumed. */ 3337 if (hstat.orig_file_tstamp >= tmr) 3338 { 3339 if (hstat.contlen == -1 3340 || hstat.orig_file_size == hstat.contlen) 3341 { 3342 logprintf (LOG_VERBOSE, _("\ 3343Server file no newer than local file %s -- not retrieving.\n\n"), 3344 quote (hstat.orig_file_name)); 3345 ret = RETROK; 3346 goto exit; 3347 } 3348 else 3349 { 3350 logprintf (LOG_VERBOSE, _("\ 3351The sizes do not match (local %s) -- retrieving.\n"), 3352 number_to_static_string (hstat.orig_file_size)); 3353 } 3354 } 3355 else 3356 { 3357 force_full_retrieve = true; 3358 logputs (LOG_VERBOSE, 3359 _("Remote file is newer, retrieving.\n")); 3360 } 3361 3362 logputs (LOG_VERBOSE, "\n"); 3363 } 3364 } 3365 3366 /* free_hstat (&hstat); */ 3367 hstat.timestamp_checked = true; 3368 } 3369 3370 if (opt.spider) 3371 { 3372 bool finished = true; 3373 if (opt.recursive) 3374 { 3375 if (*dt & TEXTHTML) 3376 { 3377 logputs (LOG_VERBOSE, _("\ 3378Remote file exists and could contain links to other resources -- retrieving.\n\n")); 3379 finished = false; 3380 } 3381 else 3382 { 3383 logprintf (LOG_VERBOSE, _("\ 3384Remote file exists but does not contain any link -- not retrieving.\n\n")); 3385 ret = RETROK; /* RETRUNNEEDED is not for caller. */ 3386 } 3387 } 3388 else 3389 { 3390 if (*dt & TEXTHTML) 3391 { 3392 logprintf (LOG_VERBOSE, _("\ 3393Remote file exists and could contain further links,\n\ 3394but recursion is disabled -- not retrieving.\n\n")); 3395 } 3396 else 3397 { 3398 logprintf (LOG_VERBOSE, _("\ 3399Remote file exists.\n\n")); 3400 } 3401 ret = RETROK; /* RETRUNNEEDED is not for caller. */ 3402 } 3403 3404 if (finished) 3405 { 3406 logprintf (LOG_NONVERBOSE, 3407 _("%s URL: %s %2d %s\n"), 3408 tms, u->url, hstat.statcode, 3409 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : ""); 3410 goto exit; 3411 } 3412 } 3413 3414 got_name = true; 3415 *dt &= ~HEAD_ONLY; 3416 count = 0; /* the retrieve count for HEAD is reset */ 3417 xfree_null (hstat.message); 3418 xfree_null (hstat.error); 3419 continue; 3420 } /* send_head_first */ 3421 } /* !got_head */ 3422 3423 if (opt.useservertimestamps 3424 && (tmr != (time_t) (-1)) 3425 && ((hstat.len == hstat.contlen) || 3426 ((hstat.res == 0) && (hstat.contlen == -1)))) 3427 { 3428 const char *fl = NULL; 3429 set_local_file (&fl, hstat.local_file); 3430 if (fl) 3431 { 3432 time_t newtmr = -1; 3433 /* Reparse time header, in case it's changed. */ 3434 if (time_came_from_head 3435 && hstat.remote_time && hstat.remote_time[0]) 3436 { 3437 newtmr = http_atotm (hstat.remote_time); 3438 if (newtmr != (time_t)-1) 3439 tmr = newtmr; 3440 } 3441 touch (fl, tmr); 3442 } 3443 } 3444 /* End of time-stamping section. */ 3445 3446 tmrate = retr_rate (hstat.rd_size, hstat.dltime); 3447 total_download_time += hstat.dltime; 3448 3449 if (hstat.len == hstat.contlen) 3450 { 3451 if (*dt & RETROKF) 3452 { 3453 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document)); 3454 3455 logprintf (LOG_VERBOSE, 3456 write_to_stdout 3457 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n") 3458 : _("%s (%s) - %s saved [%s/%s]\n\n"), 3459 tms, tmrate, 3460 write_to_stdout ? "" : quote (hstat.local_file), 3461 number_to_static_string (hstat.len), 3462 number_to_static_string (hstat.contlen)); 3463 logprintf (LOG_NONVERBOSE, 3464 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n", 3465 tms, u->url, 3466 number_to_static_string (hstat.len), 3467 number_to_static_string (hstat.contlen), 3468 hstat.local_file, count); 3469 } 3470 ++numurls; 3471 total_downloaded_bytes += hstat.rd_size; 3472 3473 /* Remember that we downloaded the file for later ".orig" code. */ 3474 if (*dt & ADDED_HTML_EXTENSION) 3475 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file); 3476 else 3477 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file); 3478 3479 ret = RETROK; 3480 goto exit; 3481 } 3482 else if (hstat.res == 0) /* No read error */ 3483 { 3484 if (hstat.contlen == -1) /* We don't know how much we were supposed 3485 to get, so assume we succeeded. */ 3486 { 3487 if (*dt & RETROKF) 3488 { 3489 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document)); 3490 3491 logprintf (LOG_VERBOSE, 3492 write_to_stdout 3493 ? _("%s (%s) - written to stdout %s[%s]\n\n") 3494 : _("%s (%s) - %s saved [%s]\n\n"), 3495 tms, tmrate, 3496 write_to_stdout ? "" : quote (hstat.local_file), 3497 number_to_static_string (hstat.len)); 3498 logprintf (LOG_NONVERBOSE, 3499 "%s URL:%s [%s] -> \"%s\" [%d]\n", 3500 tms, u->url, number_to_static_string (hstat.len), 3501 hstat.local_file, count); 3502 } 3503 ++numurls; 3504 total_downloaded_bytes += hstat.rd_size; 3505 3506 /* Remember that we downloaded the file for later ".orig" code. */ 3507 if (*dt & ADDED_HTML_EXTENSION) 3508 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file); 3509 else 3510 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file); 3511 3512 ret = RETROK; 3513 goto exit; 3514 } 3515 else if (hstat.len < hstat.contlen) /* meaning we lost the 3516 connection too soon */ 3517 { 3518 logprintf (LOG_VERBOSE, 3519 _("%s (%s) - Connection closed at byte %s. "), 3520 tms, tmrate, number_to_static_string (hstat.len)); 3521 printwhat (count, opt.ntry); 3522 continue; 3523 } 3524 else if (hstat.len != hstat.restval) 3525 /* Getting here would mean reading more data than 3526 requested with content-length, which we never do. */ 3527 abort (); 3528 else 3529 { 3530 /* Getting here probably means that the content-length was 3531 * _less_ than the original, local size. We should probably 3532 * truncate or re-read, or something. FIXME */ 3533 ret = RETROK; 3534 goto exit; 3535 } 3536 } 3537 else /* from now on hstat.res can only be -1 */ 3538 { 3539 if (hstat.contlen == -1) 3540 { 3541 logprintf (LOG_VERBOSE, 3542 _("%s (%s) - Read error at byte %s (%s)."), 3543 tms, tmrate, number_to_static_string (hstat.len), 3544 hstat.rderrmsg); 3545 printwhat (count, opt.ntry); 3546 continue; 3547 } 3548 else /* hstat.res == -1 and contlen is given */ 3549 { 3550 logprintf (LOG_VERBOSE, 3551 _("%s (%s) - Read error at byte %s/%s (%s). "), 3552 tms, tmrate, 3553 number_to_static_string (hstat.len), 3554 number_to_static_string (hstat.contlen), 3555 hstat.rderrmsg); 3556 printwhat (count, opt.ntry); 3557 continue; 3558 } 3559 } 3560 /* not reached */ 3561 } 3562 while (!opt.ntry || (count < opt.ntry)); 3563 3564exit: 3565 if (ret == RETROK && local_file) 3566 { 3567 xfree_null (*local_file); 3568 *local_file = xstrdup (hstat.local_file); 3569 } 3570 free_hstat (&hstat); 3571 3572 return ret; 3573} 3574 3575/* Check whether the result of strptime() indicates success. 3576 strptime() returns the pointer to how far it got to in the string. 3577 The processing has been successful if the string is at `GMT' or 3578 `+X', or at the end of the string. 3579 3580 In extended regexp parlance, the function returns 1 if P matches 3581 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime 3582 can return) is considered a failure and 0 is returned. */ 3583static bool 3584check_end (const char *p) 3585{ 3586 if (!p) 3587 return false; 3588 while (c_isspace (*p)) 3589 ++p; 3590 if (!*p 3591 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T') 3592 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1]))) 3593 return true; 3594 else 3595 return false; 3596} 3597 3598/* Convert the textual specification of time in TIME_STRING to the 3599 number of seconds since the Epoch. 3600 3601 TIME_STRING can be in any of the three formats RFC2616 allows the 3602 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date, 3603 as well as the time format used in the Set-Cookie header. 3604 Timezones are ignored, and should be GMT. 3605 3606 Return the computed time_t representation, or -1 if the conversion 3607 fails. 3608 3609 This function uses strptime with various string formats for parsing 3610 TIME_STRING. This results in a parser that is not as lenient in 3611 interpreting TIME_STRING as I would like it to be. Being based on 3612 strptime, it always allows shortened months, one-digit days, etc., 3613 but due to the multitude of formats in which time can be 3614 represented, an ideal HTTP time parser would be even more 3615 forgiving. It should completely ignore things like week days and 3616 concentrate only on the various forms of representing years, 3617 months, days, hours, minutes, and seconds. For example, it would 3618 be nice if it accepted ISO 8601 out of the box. 3619 3620 I've investigated free and PD code for this purpose, but none was 3621 usable. getdate was big and unwieldy, and had potential copyright 3622 issues, or so I was informed. Dr. Marcus Hennecke's atotm(), 3623 distributed with phttpd, is excellent, but we cannot use it because 3624 it is not assigned to the FSF. So I stuck it with strptime. */ 3625 3626time_t 3627http_atotm (const char *time_string) 3628{ 3629 /* NOTE: Solaris strptime man page claims that %n and %t match white 3630 space, but that's not universally available. Instead, we simply 3631 use ` ' to mean "skip all WS", which works under all strptime 3632 implementations I've tested. */ 3633 3634 static const char *time_formats[] = { 3635 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */ 3636 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */ 3637 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */ 3638 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57 3639 (used in Set-Cookie, defined in the 3640 Netscape cookie specification.) */ 3641 }; 3642 const char *oldlocale; 3643 char savedlocale[256]; 3644 size_t i; 3645 time_t ret = (time_t) -1; 3646 3647 /* Solaris strptime fails to recognize English month names in 3648 non-English locales, which we work around by temporarily setting 3649 locale to C before invoking strptime. */ 3650 oldlocale = setlocale (LC_TIME, NULL); 3651 if (oldlocale) 3652 { 3653 size_t l = strlen (oldlocale) + 1; 3654 if (l >= sizeof savedlocale) 3655 savedlocale[0] = '\0'; 3656 else 3657 memcpy (savedlocale, oldlocale, l); 3658 } 3659 else savedlocale[0] = '\0'; 3660 3661 setlocale (LC_TIME, "C"); 3662 3663 for (i = 0; i < countof (time_formats); i++) 3664 { 3665 struct tm t; 3666 3667 /* Some versions of strptime use the existing contents of struct 3668 tm to recalculate the date according to format. Zero it out 3669 to prevent stack garbage from influencing strptime. */ 3670 xzero (t); 3671 3672 if (check_end (strptime (time_string, time_formats[i], &t))) 3673 { 3674 ret = timegm (&t); 3675 break; 3676 } 3677 } 3678 3679 /* Restore the previous locale. */ 3680 if (savedlocale[0]) 3681 setlocale (LC_TIME, savedlocale); 3682 3683 return ret; 3684} 3685 3686/* Authorization support: We support three authorization schemes: 3687 3688 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string; 3689 3690 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>, 3691 consisting of answering to the server's challenge with the proper 3692 MD5 digests. 3693 3694 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel 3695 Stenberg for libcurl. Like digest, NTLM is based on a 3696 challenge-response mechanism, but unlike digest, it is non-standard 3697 (authenticates TCP connections rather than requests), undocumented 3698 and Microsoft-specific. */ 3699 3700/* Create the authentication header contents for the `Basic' scheme. 3701 This is done by encoding the string "USER:PASS" to base64 and 3702 prepending the string "Basic " in front of it. */ 3703 3704static char * 3705basic_authentication_encode (const char *user, const char *passwd) 3706{ 3707 char *t1, *t2; 3708 int len1 = strlen (user) + 1 + strlen (passwd); 3709 3710 t1 = (char *)alloca (len1 + 1); 3711 sprintf (t1, "%s:%s", user, passwd); 3712 3713 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1); 3714 base64_encode (t1, len1, t2); 3715 3716 return concat_strings ("Basic ", t2, (char *) 0); 3717} 3718 3719#define SKIP_WS(x) do { \ 3720 while (c_isspace (*(x))) \ 3721 ++(x); \ 3722} while (0) 3723 3724#ifdef ENABLE_DIGEST 3725/* Dump the hexadecimal representation of HASH to BUF. HASH should be 3726 an array of 16 bytes containing the hash keys, and BUF should be a 3727 buffer of 33 writable characters (32 for hex digits plus one for 3728 zero termination). */ 3729static void 3730dump_hash (char *buf, const unsigned char *hash) 3731{ 3732 int i; 3733 3734 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++) 3735 { 3736 *buf++ = XNUM_TO_digit (*hash >> 4); 3737 *buf++ = XNUM_TO_digit (*hash & 0xf); 3738 } 3739 *buf = '\0'; 3740} 3741 3742/* Take the line apart to find the challenge, and compose a digest 3743 authorization header. See RFC2069 section 2.1.2. */ 3744static char * 3745digest_authentication_encode (const char *au, const char *user, 3746 const char *passwd, const char *method, 3747 const char *path, uerr_t *auth_err) 3748{ 3749 static char *realm, *opaque, *nonce, *qop, *algorithm; 3750 static struct { 3751 const char *name; 3752 char **variable; 3753 } options[] = { 3754 { "realm", &realm }, 3755 { "opaque", &opaque }, 3756 { "nonce", &nonce }, 3757 { "qop", &qop }, 3758 { "algorithm", &algorithm } 3759 }; 3760 char cnonce[16] = ""; 3761 char *res; 3762 int res_len; 3763 size_t res_size; 3764 param_token name, value; 3765 3766 3767 realm = opaque = nonce = algorithm = qop = NULL; 3768 3769 au += 6; /* skip over `Digest' */ 3770 while (extract_param (&au, &name, &value, ',', NULL)) 3771 { 3772 size_t i; 3773 size_t namelen = name.e - name.b; 3774 for (i = 0; i < countof (options); i++) 3775 if (namelen == strlen (options[i].name) 3776 && 0 == strncmp (name.b, options[i].name, 3777 namelen)) 3778 { 3779 *options[i].variable = strdupdelim (value.b, value.e); 3780 break; 3781 } 3782 } 3783 3784 if (qop != NULL && strcmp(qop,"auth")) 3785 { 3786 logprintf (LOG_NOTQUIET, _("Unsupported quality of protection '%s'.\n"), qop); 3787 xfree_null (qop); /* force freeing mem and return */ 3788 qop = NULL; 3789 } 3790 else if (algorithm != NULL && strcmp (algorithm,"MD5") && strcmp (algorithm,"MD5-sess")) 3791 { 3792 logprintf (LOG_NOTQUIET, _("Unsupported algorithm '%s'.\n"), algorithm); 3793 xfree_null (qop); /* force freeing mem and return */ 3794 qop = NULL; 3795 } 3796 3797 if (!realm || !nonce || !user || !passwd || !path || !method || !qop) 3798 { 3799 xfree_null (realm); 3800 xfree_null (opaque); 3801 xfree_null (nonce); 3802 xfree_null (qop); 3803 xfree_null (algorithm); 3804 if (!qop) 3805 *auth_err = UNKNOWNATTR; 3806 else 3807 *auth_err = ATTRMISSING; 3808 return NULL; 3809 } 3810 3811 /* Calculate the digest value. */ 3812 { 3813 struct md5_ctx ctx; 3814 unsigned char hash[MD5_DIGEST_SIZE]; 3815 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1]; 3816 char response_digest[MD5_DIGEST_SIZE * 2 + 1]; 3817 3818 /* A1BUF = H(user ":" realm ":" password) */ 3819 md5_init_ctx (&ctx); 3820 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx); 3821 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3822 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx); 3823 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3824 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx); 3825 md5_finish_ctx (&ctx, hash); 3826 3827 dump_hash (a1buf, hash); 3828 3829 if (algorithm && !strcmp (algorithm, "MD5-sess")) 3830 { 3831 /* A1BUF = H( H(user ":" realm ":" password) ":" nonce ":" cnonce ) */ 3832 snprintf (cnonce, sizeof (cnonce), "%08x", random_number(INT_MAX)); 3833 3834 md5_init_ctx (&ctx); 3835 // md5_process_bytes (hash, MD5_DIGEST_SIZE, &ctx); 3836 md5_process_bytes (a1buf, MD5_DIGEST_SIZE * 2, &ctx); 3837 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3838 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx); 3839 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3840 md5_process_bytes ((unsigned char *)cnonce, strlen (cnonce), &ctx); 3841 md5_finish_ctx (&ctx, hash); 3842 3843 dump_hash (a1buf, hash); 3844 } 3845 3846 /* A2BUF = H(method ":" path) */ 3847 md5_init_ctx (&ctx); 3848 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx); 3849 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3850 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx); 3851 md5_finish_ctx (&ctx, hash); 3852 dump_hash (a2buf, hash); 3853 3854 if (qop && (!strcmp(qop, "auth") || !strcmp (qop, "auth-int"))) 3855 { 3856 /* RFC 2617 Digest Access Authentication */ 3857 /* generate random hex string */ 3858 if (!*cnonce) 3859 snprintf(cnonce, sizeof(cnonce), "%08x", random_number(INT_MAX)); 3860 3861 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" noncecount ":" clientnonce ":" qop ": " A2BUF) */ 3862 md5_init_ctx (&ctx); 3863 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx); 3864 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3865 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx); 3866 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3867 md5_process_bytes ((unsigned char *)"00000001", 8, &ctx); /* TODO: keep track of server nonce values */ 3868 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3869 md5_process_bytes ((unsigned char *)cnonce, strlen(cnonce), &ctx); 3870 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3871 md5_process_bytes ((unsigned char *)qop, strlen(qop), &ctx); 3872 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3873 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx); 3874 md5_finish_ctx (&ctx, hash); 3875 } 3876 else 3877 { 3878 /* RFC 2069 Digest Access Authentication */ 3879 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */ 3880 md5_init_ctx (&ctx); 3881 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx); 3882 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3883 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx); 3884 md5_process_bytes ((unsigned char *)":", 1, &ctx); 3885 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx); 3886 md5_finish_ctx (&ctx, hash); 3887 } 3888 3889 dump_hash (response_digest, hash); 3890 3891 res_size = strlen (user) 3892 + strlen (realm) 3893 + strlen (nonce) 3894 + strlen (path) 3895 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/ 3896 + (opaque ? strlen (opaque) : 0) 3897 + (algorithm ? strlen (algorithm) : 0) 3898 + (qop ? 128: 0) 3899 + strlen (cnonce) 3900 + 128; 3901 3902 res = xmalloc (res_size); 3903 3904 if (qop && !strcmp (qop, "auth")) 3905 { 3906 res_len = snprintf (res, res_size, "Digest "\ 3907 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\""\ 3908 ", qop=auth, nc=00000001, cnonce=\"%s\"", 3909 user, realm, nonce, path, response_digest, cnonce); 3910 3911 } 3912 else 3913 { 3914 res_len = snprintf (res, res_size, "Digest "\ 3915 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"", 3916 user, realm, nonce, path, response_digest); 3917 } 3918 3919 if (opaque) 3920 { 3921 res_len += snprintf(res + res_len, res_size - res_len, ", opaque=\"%s\"", opaque); 3922 } 3923 3924 if (algorithm) 3925 { 3926 snprintf(res + res_len, res_size - res_len, ", algorithm=\"%s\"", algorithm); 3927 } 3928 } 3929 3930 xfree_null (realm); 3931 xfree_null (opaque); 3932 xfree_null (nonce); 3933 xfree_null (qop); 3934 xfree_null (algorithm); 3935 3936 return res; 3937} 3938#endif /* ENABLE_DIGEST */ 3939 3940/* Computing the size of a string literal must take into account that 3941 value returned by sizeof includes the terminating \0. */ 3942#define STRSIZE(literal) (sizeof (literal) - 1) 3943 3944/* Whether chars in [b, e) begin with the literal string provided as 3945 first argument and are followed by whitespace or terminating \0. 3946 The comparison is case-insensitive. */ 3947#define STARTS(literal, b, e) \ 3948 ((e > b) \ 3949 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \ 3950 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \ 3951 && ((size_t) ((e) - (b)) == STRSIZE (literal) \ 3952 || c_isspace (b[STRSIZE (literal)]))) 3953 3954static bool 3955known_authentication_scheme_p (const char *hdrbeg, const char *hdrend) 3956{ 3957 return STARTS ("Basic", hdrbeg, hdrend) 3958#ifdef ENABLE_DIGEST 3959 || STARTS ("Digest", hdrbeg, hdrend) 3960#endif 3961#ifdef ENABLE_NTLM 3962 || STARTS ("NTLM", hdrbeg, hdrend) 3963#endif 3964 ; 3965} 3966 3967#undef STARTS 3968 3969/* Create the HTTP authorization request header. When the 3970 `WWW-Authenticate' response header is seen, according to the 3971 authorization scheme specified in that header (`Basic' and `Digest' 3972 are supported by the current implementation), produce an 3973 appropriate HTTP authorization request header. */ 3974static char * 3975create_authorization_line (const char *au, const char *user, 3976 const char *passwd, const char *method, 3977 const char *path, bool *finished, uerr_t *auth_err) 3978{ 3979 /* We are called only with known schemes, so we can dispatch on the 3980 first letter. */ 3981 switch (c_toupper (*au)) 3982 { 3983 case 'B': /* Basic */ 3984 *finished = true; 3985 return basic_authentication_encode (user, passwd); 3986#ifdef ENABLE_DIGEST 3987 case 'D': /* Digest */ 3988 *finished = true; 3989 return digest_authentication_encode (au, user, passwd, method, path, auth_err); 3990#endif 3991#ifdef ENABLE_NTLM 3992 case 'N': /* NTLM */ 3993 if (!ntlm_input (&pconn.ntlm, au)) 3994 { 3995 *finished = true; 3996 return NULL; 3997 } 3998 return ntlm_output (&pconn.ntlm, user, passwd, finished); 3999#endif 4000 default: 4001 /* We shouldn't get here -- this function should be only called 4002 with values approved by known_authentication_scheme_p. */ 4003 abort (); 4004 } 4005} 4006 4007static void 4008load_cookies (void) 4009{ 4010 if (!wget_cookie_jar) 4011 wget_cookie_jar = cookie_jar_new (); 4012 if (opt.cookies_input && !cookies_loaded_p) 4013 { 4014 cookie_jar_load (wget_cookie_jar, opt.cookies_input); 4015 cookies_loaded_p = true; 4016 } 4017} 4018 4019void 4020save_cookies (void) 4021{ 4022 if (wget_cookie_jar) 4023 cookie_jar_save (wget_cookie_jar, opt.cookies_output); 4024} 4025 4026void 4027http_cleanup (void) 4028{ 4029 xfree_null (pconn.host); 4030 if (wget_cookie_jar) 4031 cookie_jar_delete (wget_cookie_jar); 4032} 4033 4034void 4035ensure_extension (struct http_stat *hs, const char *ext, int *dt) 4036{ 4037 char *last_period_in_local_filename = strrchr (hs->local_file, '.'); 4038 char shortext[8]; 4039 int len; 4040 shortext[0] = '\0'; 4041 len = strlen (ext); 4042 if (len == 5) 4043 { 4044 memcpy (shortext, ext, len - 1); 4045 shortext[len - 1] = '\0'; 4046 } 4047 4048 if (last_period_in_local_filename == NULL 4049 || !(0 == strcasecmp (last_period_in_local_filename, shortext) 4050 || 0 == strcasecmp (last_period_in_local_filename, ext))) 4051 { 4052 int local_filename_len = strlen (hs->local_file); 4053 /* Resize the local file, allowing for ".html" preceded by 4054 optional ".NUMBER". */ 4055 hs->local_file = xrealloc (hs->local_file, 4056 local_filename_len + 24 + len); 4057 strcpy (hs->local_file + local_filename_len, ext); 4058 /* If clobbering is not allowed and the file, as named, 4059 exists, tack on ".NUMBER.html" instead. */ 4060 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file)) 4061 { 4062 int ext_num = 1; 4063 do 4064 sprintf (hs->local_file + local_filename_len, 4065 ".%d%s", ext_num++, ext); 4066 while (file_exists_p (hs->local_file)); 4067 } 4068 *dt |= ADDED_HTML_EXTENSION; 4069 } 4070} 4071 4072 4073#ifdef TESTING 4074 4075const char * 4076test_parse_content_disposition(void) 4077{ 4078 unsigned i; 4079 static const struct { 4080 const char *hdrval; 4081 const char *filename; 4082 bool result; 4083 } test_array[] = { 4084 { "filename=\"file.ext\"", "file.ext", true }, 4085 { "attachment; filename=\"file.ext\"", "file.ext", true }, 4086 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true }, 4087 { "attachment", NULL, false }, 4088 { "attachement; filename*=UTF-8'en-US'hello.txt", "hello.txt", true }, 4089 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", "helloworld.txt", true }, 4090 }; 4091 4092 for (i = 0; i < countof(test_array); ++i) 4093 { 4094 char *filename; 4095 bool res; 4096 4097 res = parse_content_disposition (test_array[i].hdrval, &filename); 4098 4099 mu_assert ("test_parse_content_disposition: wrong result", 4100 res == test_array[i].result 4101 && (res == false 4102 || 0 == strcmp (test_array[i].filename, filename))); 4103 } 4104 4105 return NULL; 4106} 4107 4108#endif /* TESTING */ 4109 4110/* 4111 * vim: et sts=2 sw=2 cino+={s 4112 */ 4113