2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009, 2010, 2011, 2012 Free Software Foundation,
6 This file is part of GNU Wget.
8 GNU Wget is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 GNU Wget is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with Wget. If not, see <http://www.gnu.org/licenses/>.
21 Additional permission under GNU GPL version 3 section 7
23 If you modify this program, or any covered work, by linking or
24 combining it with the OpenSSL project's OpenSSL library (or a
25 modified version of that library), containing parts covered by the
26 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
27 grants you additional permission to convey the resulting work.
28 Corresponding Source for a non-source form of such a combination
29 shall include the source code for the parts of OpenSSL used as well
30 as that of the covered work. */
55 # include "http-ntlm.h"
69 #endif /* def __VMS */
71 extern char *version_string;
75 static char *create_authorization_line (const char *, const char *,
76 const char *, const char *,
77 const char *, bool *);
78 static char *basic_authentication_encode (const char *, const char *);
79 static bool known_authentication_scheme_p (const char *, const char *);
80 static void ensure_extension (struct http_stat *, const char *, int *);
81 static void load_cookies (void);
84 # define MIN(x, y) ((x) > (y) ? (y) : (x))
88 static bool cookies_loaded_p;
89 static struct cookie_jar *wget_cookie_jar;
91 #define TEXTHTML_S "text/html"
92 #define TEXTXHTML_S "application/xhtml+xml"
93 #define TEXTCSS_S "text/css"
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)
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
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 */
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
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
134 rel_none, rel_name, rel_value, rel_both
141 struct request_header {
143 enum rp release_policy;
145 int hcount, hcapacity;
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
155 static struct request *
156 request_new (const char *method, char *arg)
158 struct request *req = xnew0 (struct request);
160 req->headers = xnew_array (struct request_header, req->hcapacity);
161 req->method = method;
166 /* Return the method string passed with the last call to
167 request_set_method. */
170 request_method (const struct request *req)
175 /* Free one header according to the release policy specified with
176 request_set_header. */
179 release_header (struct request_header *hdr)
181 switch (hdr->release_policy)
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.
203 RELEASE_POLICY determines whether NAME and VALUE should be released
204 (freed) with request_free. Allowed values are:
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
211 Setting release policy is useful when arguments come from different
212 sources. For example:
214 // Don't free literal strings!
215 request_set_header (req, "Pragma", "no-cache", rel_none);
217 // Don't free a global variable, we'll need it later.
218 request_set_header (req, "Referer", opt.referer, rel_none);
220 // Value freshly allocated, free it when done.
221 request_set_header (req, "Range",
222 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
227 request_set_header (struct request *req, const char *name, const char *value,
228 enum rp release_policy)
230 struct request_header *hdr;
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);
242 for (i = 0; i < req->hcount; i++)
244 hdr = &req->headers[i];
245 if (0 == strcasecmp (name, hdr->name))
247 /* Replace existing header. */
248 release_header (hdr);
249 hdr->name = (void *)name;
250 hdr->value = (void *)value;
251 hdr->release_policy = release_policy;
256 /* Install new header. */
258 if (req->hcount >= req->hcapacity)
260 req->hcapacity <<= 1;
261 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
263 hdr = &req->headers[req->hcount++];
264 hdr->name = (void *)name;
265 hdr->value = (void *)value;
266 hdr->release_policy = release_policy;
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"). */
275 request_set_user_header (struct request *req, const char *header)
278 const char *p = strchr (header, ':');
281 BOUNDED_TO_ALLOCA (header, p, name);
283 while (c_isspace (*p))
285 request_set_header (req, xstrdup (name), (char *) p, rel_name);
288 /* Remove the header with specified name from REQ. Returns true if
289 the header was actually removed, false otherwise. */
292 request_remove_header (struct request *req, const char *name)
295 for (i = 0; i < req->hcount; i++)
297 struct request_header *hdr = &req->headers[i];
298 if (0 == strcasecmp (name, hdr->name))
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));
311 #define APPEND(p, str) do { \
312 int A_len = strlen (str); \
313 memcpy (p, str, A_len); \
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. */
322 request_send (const struct request *req, int fd, FILE *warc_tmp)
324 char *request_string, *p;
325 int i, size, write_error;
327 /* Count the request size. */
330 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
331 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
333 for (i = 0; i < req->hcount; i++)
335 struct request_header *hdr = &req->headers[i];
336 /* NAME ": " VALUE "\r\n" */
337 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
343 p = request_string = alloca_array (char, size);
345 /* Generate the request. */
347 APPEND (p, req->method); *p++ = ' ';
348 APPEND (p, req->arg); *p++ = ' ';
349 memcpy (p, "HTTP/1.1\r\n", 10); p += 10;
351 for (i = 0; i < req->hcount; i++)
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';
360 *p++ = '\r', *p++ = '\n', *p++ = '\0';
361 assert (p - request_string == size);
365 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
367 /* Send the request to the server. */
369 write_error = fd_write (fd, request_string, size - 1, -1);
371 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
373 else if (warc_tmp != NULL)
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)
383 /* Release the resources used by REQ. */
386 request_free (struct request *req)
389 xfree_null (req->arg);
390 for (i = 0; i < req->hcount; i++)
391 release_header (&req->headers[i]);
392 xfree_null (req->headers);
396 static struct hash_table *basic_authed_hosts;
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. */
403 maybe_send_basic_creds (const char *hostname, const char *user,
404 const char *passwd, struct request *req)
406 bool do_challenge = false;
408 if (opt.auth_without_challenge)
410 DEBUGP (("Auth-without-challenge set, sending Basic credentials.\n"));
413 else if (basic_authed_hosts
414 && hash_table_contains(basic_authed_hosts, hostname))
416 DEBUGP (("Found %s in basic_authed_hosts.\n", quote (hostname)));
421 DEBUGP (("Host %s has not issued a general basic challenge.\n",
426 request_set_header (req, "Authorization",
427 basic_authentication_encode (user, passwd),
434 register_basic_auth_host (const char *hostname)
436 if (!basic_authed_hosts)
438 basic_authed_hosts = make_nocase_string_hash_table (1);
440 if (!hash_table_contains(basic_authed_hosts, hostname))
442 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
443 DEBUGP (("Inserted %s into basic_authed_hosts\n", quote (hostname)));
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. */
455 body_file_send (int sock, const char *file_name, wgint promised_size, FILE *warc_tmp)
457 static char chunk[8192];
462 DEBUGP (("[writing BODY file %s ... ", file_name));
464 fp = fopen (file_name, "rb");
467 while (!feof (fp) && written < promised_size)
470 int length = fread (chunk, 1, sizeof (chunk), fp);
473 towrite = MIN (promised_size - written, length);
474 write_error = fd_write (sock, chunk, towrite, -1);
480 if (warc_tmp != NULL)
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)
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)
502 assert (written == promised_size);
503 DEBUGP (("done]\n"));
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. */
514 response_head_terminator (const char *start, const char *peeked, int peeklen)
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
521 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
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
528 p = peeked - start < 2 ? start : peeked - 2;
529 end = peeked + peeklen;
531 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
532 for (; p < end - 2; p++)
535 if (p[1] == '\r' && p[2] == '\n')
537 else if (p[1] == '\n')
540 /* p==end-2: check for \n\n directly preceding END. */
541 if (p[0] == '\n' && p[1] == '\n')
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. */
553 #define HTTP_RESPONSE_MAX_SIZE 65536
555 /* Read the HTTP request head from FD and return it. The error
556 conditions are the same as with fd_read_hunk.
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. */
564 read_http_response_head (int fd)
566 return fd_read_hunk (fd, response_head_terminator, 512,
567 HTTP_RESPONSE_MAX_SIZE);
571 /* The response data. */
574 /* The array of pointers that indicate where each header starts.
575 For example, given this HTTP response:
582 The headers are located like this:
584 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
586 headers[0] headers[1] headers[2] headers[3]
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. */
592 const char **headers;
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
600 static struct response *
601 resp_new (const char *head)
606 struct response *resp = xnew0 (struct response);
611 /* Empty head means that we're dealing with a headerless
612 (HTTP/0.9) response. In that case, don't set HEADERS at
617 /* Split HEAD into header lines, so that resp_header_* functions
618 don't need to do this over and over again. */
624 DO_REALLOC (resp->headers, size, count + 1, const char *);
625 resp->headers[count++] = hdr;
627 /* Break upon encountering an empty line. */
628 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
631 /* Find the end of HDR, including continuations. */
634 const char *end = strchr (hdr, '\n');
640 while (*hdr == ' ' || *hdr == '\t');
642 DO_REALLOC (resp->headers, size, count + 1, const char *);
643 resp->headers[count] = NULL;
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:
654 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
655 ... do something with header ...
657 If you only care about one header, use resp_header_get instead of
661 resp_header_locate (const struct response *resp, const char *name, int start,
662 const char **begptr, const char **endptr)
665 const char **headers = resp->headers;
668 if (!headers || !headers[1])
671 name_len = strlen (name);
677 for (; headers[i + 1]; i++)
679 const char *b = headers[i];
680 const char *e = headers[i + 1];
682 && b[name_len] == ':'
683 && 0 == strncasecmp (b, name, name_len))
686 while (b < e && c_isspace (*b))
688 while (b < e && c_isspace (e[-1]))
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.
702 This function is used as a building block for resp_header_copy
703 and resp_header_strdup. */
706 resp_header_get (const struct response *resp, const char *name,
707 const char **begptr, const char **endptr)
709 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
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.
718 If BUFSIZE is 0, no data is copied, but the boolean indication of
719 whether the header is present is still returned. */
722 resp_header_copy (const struct response *resp, const char *name,
723 char *buf, int bufsize)
726 if (!resp_header_get (resp, name, &b, &e))
730 int len = MIN (e - b, bufsize - 1);
731 memcpy (buf, b, len);
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. */
741 resp_header_strdup (const struct response *resp, const char *name)
744 if (!resp_header_get (resp, name, &b, &e))
746 return strdupdelim (b, e);
749 /* Parse the HTTP status line, which is of format:
751 HTTP-Version SP Status-Code SP Reason-Phrase
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. */
758 resp_status (const struct response *resp, char **message)
765 /* For a HTTP/0.9 response, assume status 200. */
767 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
771 p = resp->headers[0];
772 end = resp->headers[1];
778 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
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 == '/')
787 while (p < end && c_isdigit (*p))
789 if (p < end && *p == '.')
791 while (p < end && c_isdigit (*p))
795 while (p < end && c_isspace (*p))
797 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
800 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
805 while (p < end && c_isspace (*p))
807 while (p < end && c_isspace (end[-1]))
809 *message = strdupdelim (p, end);
815 /* Release the resources used by RESP. */
818 resp_free (struct response *resp)
820 xfree_null (resp->headers);
824 /* Print a single line of response, the characters [b, e). We tried
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. */
831 print_response_line(const char *prefix, const char *b, const char *e)
834 BOUNDED_TO_ALLOCA(b, e, copy);
835 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
836 quotearg_style (escape_quoting_style, copy));
839 /* Print the server response, line by line, omitting the trailing CRLF
840 from individual header lines, and prefixed with PREFIX. */
843 print_server_response (const struct response *resp, const char *prefix)
848 for (i = 0; resp->headers[i + 1]; i++)
850 const char *b = resp->headers[i];
851 const char *e = resp->headers[i + 1];
853 if (b < e && e[-1] == '\n')
855 if (b < e && e[-1] == '\r')
857 print_response_line(prefix, b, e);
861 /* Parse the `Content-Range' header and extract the information it
862 contains. Returns true if successful, false otherwise. */
864 parse_content_range (const char *hdr, wgint *first_byte_ptr,
865 wgint *last_byte_ptr, wgint *entity_length_ptr)
869 /* Ancient versions of Netscape proxy server, presumably predating
870 rfc2068, sent out `Content-Range' without the "bytes"
872 if (0 == strncasecmp (hdr, "bytes", 5))
875 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
879 while (c_isspace (*hdr))
884 if (!c_isdigit (*hdr))
886 for (num = 0; c_isdigit (*hdr); hdr++)
887 num = 10 * num + (*hdr - '0');
888 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
890 *first_byte_ptr = num;
892 for (num = 0; c_isdigit (*hdr); hdr++)
893 num = 10 * num + (*hdr - '0');
894 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
896 *last_byte_ptr = num;
901 for (num = 0; c_isdigit (*hdr); hdr++)
902 num = 10 * num + (*hdr - '0');
903 *entity_length_ptr = num;
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.
913 If reading fails, false is returned, true otherwise. In debug
914 mode, the body is displayed for debugging purposes. */
917 skip_short_body (int fd, wgint contlen, bool chunked)
920 SKIP_SIZE = 512, /* size of the download buffer */
921 SKIP_THRESHOLD = 4096 /* the largest size we read */
923 wgint remaining_chunk_size = 0;
924 char dlbuf[SKIP_SIZE + 1];
925 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
927 assert (contlen != -1 || contlen);
929 /* If the body is too large, it makes more sense to simply close the
930 connection than to try to read the body. */
931 if (contlen > SKIP_THRESHOLD)
934 while (contlen > 0 || chunked)
939 if (remaining_chunk_size == 0)
941 char *line = fd_read_line (fd);
946 remaining_chunk_size = strtol (line, &endl, 16);
949 if (remaining_chunk_size == 0)
951 line = fd_read_line (fd);
957 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
960 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
962 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
965 /* Don't normally report the error since this is an
966 optimization that should be invisible to the user. */
967 DEBUGP (("] aborting (%s).\n",
968 ret < 0 ? fd_errstr (fd) : "EOF received"));
975 remaining_chunk_size -= ret;
976 if (remaining_chunk_size == 0)
978 char *line = fd_read_line (fd);
986 /* Safe even if %.*s bogusly expects terminating \0 because
987 we've zero-terminated dlbuf above. */
988 DEBUGP (("%.*s", ret, dlbuf));
991 DEBUGP (("] done.\n"));
995 #define NOT_RFC2231 0
996 #define RFC2231_NOENCODING 1
997 #define RFC2231_ENCODING 2
999 /* extract_param extracts the parameter name into NAME.
1000 However, if the parameter name is in RFC2231 format then
1001 this function adjusts NAME by stripping of the trailing
1002 characters that are not part of the name but are present to
1003 indicate the presence of encoding information in the value
1004 or a fragment of a long parameter value
1007 modify_param_name(param_token *name)
1009 const char *delim1 = memchr (name->b, '*', name->e - name->b);
1010 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
1016 result = NOT_RFC2231;
1018 else if(delim1 == delim2)
1020 if ((name->e - 1) == delim1)
1022 result = RFC2231_ENCODING;
1026 result = RFC2231_NOENCODING;
1033 result = RFC2231_ENCODING;
1038 /* extract_param extract the paramater value into VALUE.
1039 Like modify_param_name this function modifies VALUE by
1040 stripping off the encoding information from the actual value
1043 modify_param_value (param_token *value, int encoding_type )
1045 if (RFC2231_ENCODING == encoding_type)
1047 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1048 if ( delim != NULL )
1050 value->b = (delim+1);
1055 /* Extract a parameter from the string (typically an HTTP header) at
1056 **SOURCE and advance SOURCE to the next parameter. Return false
1057 when there are no more parameters to extract. The name of the
1058 parameter is returned in NAME, and the value in VALUE. If the
1059 parameter has no value, the token's value is zeroed out.
1061 For example, if *SOURCE points to the string "attachment;
1062 filename=\"foo bar\"", the first call to this function will return
1063 the token named "attachment" and no value, and the second call will
1064 return the token named "filename" and value "foo bar". The third
1065 call will return false, indicating no more valid tokens. */
1068 extract_param (const char **source, param_token *name, param_token *value,
1071 const char *p = *source;
1073 while (c_isspace (*p)) ++p;
1077 return false; /* no error; nothing more to extract */
1082 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1084 if (name->b == name->e)
1085 return false; /* empty name: error */
1086 while (c_isspace (*p)) ++p;
1087 if (*p == separator || !*p) /* no value */
1090 if (*p == separator) ++p;
1095 return false; /* error */
1097 /* *p is '=', extract value */
1099 while (c_isspace (*p)) ++p;
1100 if (*p == '"') /* quoted */
1103 while (*p && *p != '"') ++p;
1107 /* Currently at closing quote; find the end of param. */
1108 while (c_isspace (*p)) ++p;
1109 while (*p && *p != separator) ++p;
1110 if (*p == separator)
1113 /* garbage after closed quote, e.g. foo="bar"baz */
1119 while (*p && *p != separator) ++p;
1121 while (value->e != value->b && c_isspace (value->e[-1]))
1123 if (*p == separator) ++p;
1127 int param_type = modify_param_name(name);
1128 if (NOT_RFC2231 != param_type)
1130 modify_param_value(value, param_type);
1136 #undef RFC2231_NOENCODING
1137 #undef RFC2231_ENCODING
1139 /* Appends the string represented by VALUE to FILENAME */
1142 append_value_to_filename (char **filename, param_token const * const value)
1144 int original_length = strlen(*filename);
1145 int new_length = strlen(*filename) + (value->e - value->b);
1146 *filename = xrealloc (*filename, new_length+1);
1147 memcpy (*filename + original_length, value->b, (value->e - value->b));
1148 (*filename)[new_length] = '\0';
1152 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1154 /* Parse the contents of the `Content-Disposition' header, extracting
1155 the information useful to Wget. Content-Disposition is a header
1156 borrowed from MIME; when used in HTTP, it typically serves for
1157 specifying the desired file name of the resource. For example:
1159 Content-Disposition: attachment; filename="flora.jpg"
1161 Wget will skip the tokens it doesn't care about, such as
1162 "attachment" in the previous example; it will also skip other
1163 unrecognized params. If the header is syntactically correct and
1164 contains a file name, a copy of the file name is stored in
1165 *filename and true is returned. Otherwise, the function returns
1168 The file name is stripped of directory components and must not be
1171 Historically, this function returned filename prefixed with opt.dir_prefix,
1172 now that logic is handled by the caller, new code should pay attention,
1173 changed by crq, Sep 2010.
1177 parse_content_disposition (const char *hdr, char **filename)
1179 param_token name, value;
1181 while (extract_param (&hdr, &name, &value, ';'))
1183 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1184 if ( isFilename && value.b != NULL)
1186 /* Make the file name begin at the last slash or backslash. */
1187 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1188 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1189 if (last_slash && last_bs)
1190 value.b = 1 + MAX (last_slash, last_bs);
1191 else if (last_slash || last_bs)
1192 value.b = 1 + (last_slash ? last_slash : last_bs);
1193 if (value.b == value.e)
1197 append_value_to_filename (filename, &value);
1199 *filename = strdupdelim (value.b, value.e);
1210 /* Persistent connections. Currently, we cache the most recently used
1211 connection as persistent, provided that the HTTP server agrees to
1212 make it such. The persistence data is stored in the variables
1213 below. Ideally, it should be possible to cache an arbitrary fixed
1214 number of these connections. */
1216 /* Whether a persistent connection is active. */
1217 static bool pconn_active;
1220 /* The socket of the connection. */
1223 /* Host and port of the currently active persistent connection. */
1227 /* Whether a ssl handshake has occoured on this connection. */
1230 /* Whether the connection was authorized. This is only done by
1231 NTLM, which authorizes *connections* rather than individual
1232 requests. (That practice is peculiar for HTTP, but it is a
1233 useful optimization.) */
1237 /* NTLM data of the current connection. */
1238 struct ntlmdata ntlm;
1242 /* Mark the persistent connection as invalid and free the resources it
1243 uses. This is used by the CLOSE_* macros after they forcefully
1244 close a registered persistent connection. */
1247 invalidate_persistent (void)
1249 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1250 pconn_active = false;
1251 fd_close (pconn.socket);
1256 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1257 persistent. This will enable someone to use the same connection
1258 later. In the context of HTTP, this must be called only AFTER the
1259 response has been received and the server has promised that the
1260 connection will remain alive.
1262 If a previous connection was persistent, it is closed. */
1265 register_persistent (const char *host, int port, int fd, bool ssl)
1269 if (pconn.socket == fd)
1271 /* The connection FD is already registered. */
1276 /* The old persistent connection is still active; close it
1277 first. This situation arises whenever a persistent
1278 connection exists, but we then connect to a different
1279 host, and try to register a persistent connection to that
1281 invalidate_persistent ();
1285 pconn_active = true;
1287 pconn.host = xstrdup (host);
1290 pconn.authorized = false;
1292 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1295 /* Return true if a persistent connection is available for connecting
1299 persistent_available_p (const char *host, int port, bool ssl,
1300 bool *host_lookup_failed)
1302 /* First, check whether a persistent connection is active at all. */
1306 /* If we want SSL and the last connection wasn't or vice versa,
1307 don't use it. Checking for host and port is not enough because
1308 HTTP and HTTPS can apparently coexist on the same port. */
1309 if (ssl != pconn.ssl)
1312 /* If we're not connecting to the same port, we're not interested. */
1313 if (port != pconn.port)
1316 /* If the host is the same, we're in business. If not, there is
1317 still hope -- read below. */
1318 if (0 != strcasecmp (host, pconn.host))
1320 /* Check if pconn.socket is talking to HOST under another name.
1321 This happens often when both sites are virtual hosts
1322 distinguished only by name and served by the same network
1323 interface, and hence the same web server (possibly set up by
1324 the ISP and serving many different web sites). This
1325 admittedly unconventional optimization does not contradict
1326 HTTP and works well with popular server software. */
1330 struct address_list *al;
1333 /* Don't try to talk to two different SSL sites over the same
1334 secure connection! (Besides, it's not clear that
1335 name-based virtual hosting is even possible with SSL.) */
1338 /* If pconn.socket's peer is one of the IP addresses HOST
1339 resolves to, pconn.socket is for all intents and purposes
1340 already talking to HOST. */
1342 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1344 /* Can't get the peer's address -- something must be very
1345 wrong with the connection. */
1346 invalidate_persistent ();
1349 al = lookup_host (host, 0);
1352 *host_lookup_failed = true;
1356 found = address_list_contains (al, &ip);
1357 address_list_release (al);
1362 /* The persistent connection's peer address was found among the
1363 addresses HOST resolved to; therefore, pconn.sock is in fact
1364 already talking to HOST -- no need to reconnect. */
1367 /* Finally, check whether the connection is still open. This is
1368 important because most servers implement liberal (short) timeout
1369 on persistent connections. Wget can of course always reconnect
1370 if the connection doesn't work out, but it's nicer to know in
1371 advance. This test is a logical followup of the first test, but
1372 is "expensive" and therefore placed at the end of the list.
1374 (Current implementation of test_socket_open has a nice side
1375 effect that it treats sockets with pending data as "closed".
1376 This is exactly what we want: if a broken server sends message
1377 body in response to HEAD, or if it sends more than conent-length
1378 data, we won't reuse the corrupted connection.) */
1380 if (!test_socket_open (pconn.socket))
1382 /* Oops, the socket is no longer open. Now that we know that,
1383 let's invalidate the persistent connection before returning
1385 invalidate_persistent ();
1392 /* The idea behind these two CLOSE macros is to distinguish between
1393 two cases: one when the job we've been doing is finished, and we
1394 want to close the connection and leave, and two when something is
1395 seriously wrong and we're closing the connection as part of
1398 In case of keep_alive, CLOSE_FINISH should leave the connection
1399 open, while CLOSE_INVALIDATE should still close it.
1401 Note that the semantics of the flag `keep_alive' is "this
1402 connection *will* be reused (the server has promised not to close
1403 the connection once we're done)", while the semantics of
1404 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1405 active, registered connection". */
1407 #define CLOSE_FINISH(fd) do { \
1410 if (pconn_active && (fd) == pconn.socket) \
1411 invalidate_persistent (); \
1420 #define CLOSE_INVALIDATE(fd) do { \
1421 if (pconn_active && (fd) == pconn.socket) \
1422 invalidate_persistent (); \
1430 wgint len; /* received length */
1431 wgint contlen; /* expected length */
1432 wgint restval; /* the restart value */
1433 int res; /* the result of last read */
1434 char *rderrmsg; /* error message from read error */
1435 char *newloc; /* new location (redirection) */
1436 char *remote_time; /* remote time-stamp string */
1437 char *error; /* textual HTTP error */
1438 int statcode; /* status code */
1439 char *message; /* status message */
1440 wgint rd_size; /* amount of data read from socket */
1441 double dltime; /* time it took to download the data */
1442 const char *referer; /* value of the referer header. */
1443 char *local_file; /* local file name. */
1444 bool existence_checked; /* true if we already checked for a file's
1445 existence after having begun to download
1446 (needed in gethttp for when connection is
1447 interrupted/restarted. */
1448 bool timestamp_checked; /* true if pre-download time-stamping checks
1449 * have already been performed */
1450 char *orig_file_name; /* name of file to compare for time-stamping
1451 * (might be != local_file if -K is set) */
1452 wgint orig_file_size; /* size of file to compare for time-stamping */
1453 time_t orig_file_tstamp; /* time-stamp of file to compare for
1458 free_hstat (struct http_stat *hs)
1460 xfree_null (hs->newloc);
1461 xfree_null (hs->remote_time);
1462 xfree_null (hs->error);
1463 xfree_null (hs->rderrmsg);
1464 xfree_null (hs->local_file);
1465 xfree_null (hs->orig_file_name);
1466 xfree_null (hs->message);
1468 /* Guard against being called twice. */
1470 hs->remote_time = NULL;
1475 get_file_flags (const char *filename, int *dt)
1477 logprintf (LOG_VERBOSE, _("\
1478 File %s already there; not retrieving.\n\n"), quote (filename));
1479 /* If the file is there, we suppose it's retrieved OK. */
1482 /* #### Bogusness alert. */
1483 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1484 if (has_html_suffix_p (filename))
1488 /* Download the response body from the socket and writes it to
1489 an output file. The headers have already been read from the
1490 socket. If WARC is enabled, the response body will also be
1491 written to a WARC response record.
1493 hs, contlen, contrange, chunked_transfer_encoding and url are
1494 parameters from the gethttp method. fp is a pointer to the
1497 url, warc_timestamp_str, warc_request_uuid, warc_ip, type
1498 and statcode will be saved in the headers of the WARC record.
1499 The head parameter contains the HTTP headers of the response.
1501 If fp is NULL and WARC is enabled, the response body will be
1502 written only to the WARC file. If WARC is disabled and fp
1503 is a file pointer, the data will be written to the file.
1504 If fp is a file pointer and WARC is enabled, the body will
1505 be written to both destinations.
1507 Returns the error code. */
1509 read_response_body (struct http_stat *hs, int sock, FILE *fp, wgint contlen,
1510 wgint contrange, bool chunked_transfer_encoding,
1511 char *url, char *warc_timestamp_str, char *warc_request_uuid,
1512 ip_address *warc_ip, char *type, int statcode, char *head)
1514 int warc_payload_offset = 0;
1515 FILE *warc_tmp = NULL;
1518 if (opt.warc_filename != NULL)
1520 /* Open a temporary file where we can write the response before we
1521 add it to the WARC record. */
1522 warc_tmp = warc_tempfile ();
1523 if (warc_tmp == NULL)
1524 warcerr = WARC_TMP_FOPENERR;
1528 /* We should keep the response headers for the WARC record. */
1529 int head_len = strlen (head);
1530 int warc_tmp_written = fwrite (head, 1, head_len, warc_tmp);
1531 if (warc_tmp_written != head_len)
1532 warcerr = WARC_TMP_FWRITEERR;
1533 warc_payload_offset = head_len;
1538 if (warc_tmp != NULL)
1546 /* This confuses the timestamping code that checks for file size.
1547 #### The timestamping code should be smarter about file size. */
1548 if (opt.save_headers && hs->restval == 0)
1549 fwrite (head, 1, strlen (head), fp);
1552 /* Read the response body. */
1555 /* If content-length is present, read that much; otherwise, read
1556 until EOF. The HTTP spec doesn't require the server to
1557 actually close the connection when it's done sending data. */
1558 flags |= rb_read_exactly;
1559 if (fp != NULL && hs->restval > 0 && contrange == 0)
1560 /* If the server ignored our range request, instruct fd_read_body
1561 to skip the first RESTVAL bytes of body. */
1562 flags |= rb_skip_startpos;
1563 if (chunked_transfer_encoding)
1564 flags |= rb_chunked_transfer_encoding;
1566 hs->len = hs->restval;
1568 /* Download the response body and write it to fp.
1569 If we are working on a WARC file, we simultaneously write the
1570 response body to warc_tmp. */
1571 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1572 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1576 if (warc_tmp != NULL)
1578 /* Create a response record and write it to the WARC file.
1579 Note: per the WARC standard, the request and response should share
1580 the same date header. We re-use the timestamp of the request.
1581 The response record should also refer to the uuid of the request. */
1582 bool r = warc_write_response_record (url, warc_timestamp_str,
1583 warc_request_uuid, warc_ip,
1584 warc_tmp, warc_payload_offset,
1585 type, statcode, hs->newloc);
1587 /* warc_write_response_record has closed warc_tmp. */
1593 return RETRFINISHED;
1596 if (warc_tmp != NULL)
1601 /* Error while writing to fd. */
1604 else if (hs->res == -3)
1606 /* Error while writing to warc_tmp. */
1607 return WARC_TMP_FWRITEERR;
1612 hs->rderrmsg = xstrdup (fd_errstr (sock));
1613 return RETRFINISHED;
1617 #define BEGINS_WITH(line, string_constant) \
1618 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1619 && (c_isspace (line[sizeof (string_constant) - 1]) \
1620 || !line[sizeof (string_constant) - 1]))
1623 #define SET_USER_AGENT(req) do { \
1624 if (!opt.useragent) \
1625 request_set_header (req, "User-Agent", \
1626 aprintf ("Wget/%s (VMS %s %s)", \
1627 version_string, vms_arch(), vms_vers()), \
1629 else if (*opt.useragent) \
1630 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1632 #else /* def __VMS */
1633 #define SET_USER_AGENT(req) do { \
1634 if (!opt.useragent) \
1635 request_set_header (req, "User-Agent", \
1636 aprintf ("Wget/%s (%s)", \
1637 version_string, OS_TYPE), \
1639 else if (*opt.useragent) \
1640 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1642 #endif /* def __VMS [else] */
1644 /* The flags that allow clobbering the file (opening with "wb").
1645 Defined here to avoid repetition later. #### This will require
1647 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1648 || opt.dirstruct || opt.output_document)
1650 /* Retrieve a document through HTTP protocol. It recognizes status
1651 code, and correctly handles redirections. It closes the network
1652 socket. If it receives an error from the functions below it, it
1653 will print it if there is enough information to do so (almost
1654 always), returning the error to the caller (i.e. http_loop).
1656 Various HTTP parameters are stored to hs.
1658 If PROXY is non-NULL, the connection will be made to the proxy
1659 server, and u->url will be requested. */
1661 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1662 struct iri *iri, int count)
1664 struct request *req;
1667 char *user, *passwd;
1671 wgint contlen, contrange;
1678 /* Set to 1 when the authorization has already been sent and should
1679 not be tried again. */
1680 bool auth_finished = false;
1682 /* Set to 1 when just globally-set Basic authorization has been sent;
1683 * should prevent further Basic negotiations, but not other
1685 bool basic_auth_finished = false;
1687 /* Whether NTLM authentication is used for this request. */
1688 bool ntlm_seen = false;
1690 /* Whether our connection to the remote host is through SSL. */
1691 bool using_ssl = false;
1693 /* Whether a HEAD request will be issued (as opposed to GET or
1695 bool head_only = !!(*dt & HEAD_ONLY);
1698 struct response *resp;
1702 /* Declare WARC variables. */
1703 bool warc_enabled = (opt.warc_filename != NULL);
1704 FILE *warc_tmp = NULL;
1705 char warc_timestamp_str [21];
1706 char warc_request_uuid [48];
1707 ip_address *warc_ip = NULL;
1708 off_t warc_payload_offset = -1;
1710 /* Whether this connection will be kept alive after the HTTP request
1714 /* Is the server using the chunked transfer encoding? */
1715 bool chunked_transfer_encoding = false;
1717 /* Whether keep-alive should be inhibited. */
1718 bool inhibit_keep_alive =
1719 !opt.http_keep_alive || opt.ignore_length;
1721 /* Headers sent when using POST. */
1722 wgint body_data_size = 0;
1724 bool host_lookup_failed = false;
1727 if (u->scheme == SCHEME_HTTPS)
1729 /* Initialize the SSL context. After this has once been done,
1730 it becomes a no-op. */
1733 scheme_disable (SCHEME_HTTPS);
1734 logprintf (LOG_NOTQUIET,
1735 _("Disabling SSL due to encountered errors.\n"));
1736 return SSLINITFAILED;
1739 #endif /* HAVE_SSL */
1741 /* Initialize certain elements of struct http_stat. */
1745 hs->rderrmsg = NULL;
1747 hs->remote_time = NULL;
1753 /* Prepare the request to send. */
1756 const char *meth = "GET";
1759 else if (opt.method)
1761 /* Use the full path, i.e. one that includes the leading slash and
1762 the query string. E.g. if u->path is "foo/bar" and u->query is
1763 "param=value", full_path will be "/foo/bar?param=value". */
1766 /* When using SSL over proxy, CONNECT establishes a direct
1767 connection to the HTTPS server. Therefore use the same
1768 argument as when talking to the server directly. */
1769 && u->scheme != SCHEME_HTTPS
1772 meth_arg = xstrdup (u->url);
1774 meth_arg = url_full_path (u);
1775 req = request_new (meth, meth_arg);
1778 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1779 if (*dt & SEND_NOCACHE)
1781 /* Cache-Control MUST be obeyed by all HTTP/1.1 caching mechanisms... */
1782 request_set_header (req, "Cache-Control", "no-cache, must-revalidate", rel_none);
1784 /* ... but some HTTP/1.0 caches doesn't implement Cache-Control. */
1785 request_set_header (req, "Pragma", "no-cache", rel_none);
1788 request_set_header (req, "Range",
1789 aprintf ("bytes=%s-",
1790 number_to_static_string (hs->restval)),
1792 SET_USER_AGENT (req);
1793 request_set_header (req, "Accept", "*/*", rel_none);
1795 /* Find the username and password for authentication. */
1798 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1799 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1800 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1802 /* We only do "site-wide" authentication with "global" user/password
1803 * values unless --auth-no-challange has been requested; URL user/password
1804 * info overrides. */
1805 if (user && passwd && (!u->user || opt.auth_without_challenge))
1807 /* If this is a host for which we've already received a Basic
1808 * challenge, we'll go ahead and send Basic authentication creds. */
1809 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1812 /* Generate the Host header, HOST:PORT. Take into account that:
1814 - Broken server-side software often doesn't recognize the PORT
1815 argument, so we must generate "Host: www.server.com" instead of
1816 "Host: www.server.com:80" (and likewise for https port).
1818 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1819 becomes ambiguous and needs to be rewritten as "Host:
1820 [3ffe:8100:200:2::2]:1234". */
1822 /* Formats arranged for hfmt[add_port][add_squares]. */
1823 static const char *hfmt[][2] = {
1824 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1826 int add_port = u->port != scheme_default_port (u->scheme);
1827 int add_squares = strchr (u->host, ':') != NULL;
1828 request_set_header (req, "Host",
1829 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1833 if (inhibit_keep_alive)
1834 request_set_header (req, "Connection", "Close", rel_none);
1838 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1841 request_set_header (req, "Connection", "Close", rel_none);
1842 request_set_header (req, "Proxy-Connection", "Keep-Alive", rel_none);
1849 if (opt.body_data || opt.body_file)
1851 request_set_header (req, "Content-Type",
1852 "application/x-www-form-urlencoded", rel_none);
1855 body_data_size = strlen (opt.body_data);
1858 body_data_size = file_size (opt.body_file);
1859 if (body_data_size == -1)
1861 logprintf (LOG_NOTQUIET, _("BODY data file %s missing: %s\n"),
1862 quote (opt.body_file), strerror (errno));
1866 request_set_header (req, "Content-Length",
1867 xstrdup (number_to_static_string (body_data_size)),
1873 /* We need to come back here when the initial attempt to retrieve
1874 without authorization header fails. (Expected to happen at least
1875 for the Digest authorization scheme.) */
1878 request_set_header (req, "Cookie",
1879 cookie_header (wget_cookie_jar,
1880 u->host, u->port, u->path,
1882 u->scheme == SCHEME_HTTPS
1889 /* Add the user headers. */
1890 if (opt.user_headers)
1893 for (i = 0; opt.user_headers[i]; i++)
1894 request_set_user_header (req, opt.user_headers[i]);
1900 char *proxy_user, *proxy_passwd;
1901 /* For normal username and password, URL components override
1902 command-line/wgetrc parameters. With proxy
1903 authentication, it's the reverse, because proxy URLs are
1904 normally the "permanent" ones, so command-line args
1905 should take precedence. */
1906 if (opt.proxy_user && opt.proxy_passwd)
1908 proxy_user = opt.proxy_user;
1909 proxy_passwd = opt.proxy_passwd;
1913 proxy_user = proxy->user;
1914 proxy_passwd = proxy->passwd;
1916 /* #### This does not appear right. Can't the proxy request,
1917 say, `Digest' authentication? */
1918 if (proxy_user && proxy_passwd)
1919 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1921 /* If we're using a proxy, we will be connecting to the proxy
1925 /* Proxy authorization over SSL is handled below. */
1927 if (u->scheme != SCHEME_HTTPS)
1929 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1934 /* Establish the connection. */
1936 if (inhibit_keep_alive)
1940 /* Look for a persistent connection to target host, unless a
1941 proxy is used. The exception is when SSL is in use, in which
1942 case the proxy is nothing but a passthrough to the target
1943 host, registered as a connection to the latter. */
1944 struct url *relevant = conn;
1946 if (u->scheme == SCHEME_HTTPS)
1950 if (persistent_available_p (relevant->host, relevant->port,
1952 relevant->scheme == SCHEME_HTTPS,
1956 &host_lookup_failed))
1958 int family = socket_family (pconn.socket, ENDPOINT_PEER);
1959 sock = pconn.socket;
1960 using_ssl = pconn.ssl;
1962 if (family == AF_INET6)
1963 logprintf (LOG_VERBOSE, _("Reusing existing connection to [%s]:%d.\n"),
1964 quotearg_style (escape_quoting_style, pconn.host),
1968 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1969 quotearg_style (escape_quoting_style, pconn.host),
1971 DEBUGP (("Reusing fd %d.\n", sock));
1972 if (pconn.authorized)
1973 /* If the connection is already authorized, the "Basic"
1974 authorization added by code above is unnecessary and
1976 request_remove_header (req, "Authorization");
1978 else if (host_lookup_failed)
1981 logprintf(LOG_NOTQUIET,
1982 _("%s: unable to resolve host address %s\n"),
1983 exec_name, quote (relevant->host));
1986 else if (sock != -1)
1994 sock = connect_to_host (conn->host, conn->port);
2003 return (retryable_socket_connect_error (errno)
2004 ? CONERROR : CONIMPOSSIBLE);
2008 if (proxy && u->scheme == SCHEME_HTTPS)
2010 /* When requesting SSL URLs through proxies, use the
2011 CONNECT method to request passthrough. */
2012 struct request *connreq = request_new ("CONNECT",
2013 aprintf ("%s:%d", u->host, u->port));
2014 SET_USER_AGENT (connreq);
2017 request_set_header (connreq, "Proxy-Authorization",
2018 proxyauth, rel_value);
2019 /* Now that PROXYAUTH is part of the CONNECT request,
2020 zero it out so we don't send proxy authorization with
2021 the regular request below. */
2024 /* Examples in rfc2817 use the Host header in CONNECT
2025 requests. I don't see how that gains anything, given
2026 that the contents of Host would be exactly the same as
2027 the contents of CONNECT. */
2029 write_error = request_send (connreq, sock, 0);
2030 request_free (connreq);
2031 if (write_error < 0)
2033 CLOSE_INVALIDATE (sock);
2038 head = read_http_response_head (sock);
2041 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
2043 CLOSE_INVALIDATE (sock);
2053 DEBUGP (("proxy responded with: [%s]\n", head));
2055 resp = resp_new (head);
2056 statcode = resp_status (resp, &message);
2059 char *tms = datetime_str (time (NULL));
2060 logprintf (LOG_VERBOSE, "%d\n", statcode);
2061 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
2062 quotearg_style (escape_quoting_style,
2063 _("Malformed status line")));
2068 hs->message = xstrdup (message);
2071 if (statcode != 200)
2074 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
2075 message ? quotearg_style (escape_quoting_style, message) : "?");
2076 xfree_null (message);
2080 xfree_null (message);
2082 /* SOCK is now *really* connected to u->host, so update CONN
2083 to reflect this. That way register_persistent will
2084 register SOCK as being connected to u->host:u->port. */
2088 if (conn->scheme == SCHEME_HTTPS)
2090 if (!ssl_connect_wget (sock, u->host))
2096 else if (!ssl_check_certificate (sock, u->host))
2100 return VERIFCERTERR;
2104 #endif /* HAVE_SSL */
2107 /* Open the temporary file where we will write the request. */
2110 warc_tmp = warc_tempfile ();
2111 if (warc_tmp == NULL)
2113 CLOSE_INVALIDATE (sock);
2115 return WARC_TMP_FOPENERR;
2120 warc_ip = (ip_address *) alloca (sizeof (ip_address));
2121 socket_ip_address (sock, warc_ip, ENDPOINT_PEER);
2125 /* Send the request to server. */
2126 write_error = request_send (req, sock, warc_tmp);
2128 if (write_error >= 0)
2132 DEBUGP (("[BODY data: %s]\n", opt.body_data));
2133 write_error = fd_write (sock, opt.body_data, body_data_size, -1);
2134 if (write_error >= 0 && warc_tmp != NULL)
2136 /* Remember end of headers / start of payload. */
2137 warc_payload_offset = ftello (warc_tmp);
2139 /* Write a copy of the data to the WARC record. */
2140 int warc_tmp_written = fwrite (opt.body_data, 1, body_data_size, warc_tmp);
2141 if (warc_tmp_written != body_data_size)
2145 else if (opt.body_file && body_data_size != 0)
2147 if (warc_tmp != NULL)
2148 /* Remember end of headers / start of payload */
2149 warc_payload_offset = ftello (warc_tmp);
2151 write_error = body_file_send (sock, opt.body_file, body_data_size, warc_tmp);
2155 if (write_error < 0)
2157 CLOSE_INVALIDATE (sock);
2160 if (warc_tmp != NULL)
2163 if (write_error == -2)
2164 return WARC_TMP_FWRITEERR;
2168 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
2169 proxy ? "Proxy" : "HTTP");
2178 /* Generate a timestamp and uuid for this request. */
2179 warc_timestamp (warc_timestamp_str);
2180 warc_uuid_str (warc_request_uuid);
2182 /* Create a request record and store it in the WARC file. */
2183 warc_result = warc_write_request_record (u->url, warc_timestamp_str,
2184 warc_request_uuid, warc_ip,
2185 warc_tmp, warc_payload_offset);
2188 CLOSE_INVALIDATE (sock);
2193 /* warc_write_request_record has also closed warc_tmp. */
2198 head = read_http_response_head (sock);
2203 logputs (LOG_NOTQUIET, _("No data received.\n"));
2204 CLOSE_INVALIDATE (sock);
2210 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
2212 CLOSE_INVALIDATE (sock);
2217 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
2219 resp = resp_new (head);
2221 /* Check for status line. */
2223 statcode = resp_status (resp, &message);
2226 char *tms = datetime_str (time (NULL));
2227 logprintf (LOG_VERBOSE, "%d\n", statcode);
2228 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
2229 quotearg_style (escape_quoting_style,
2230 _("Malformed status line")));
2231 CLOSE_INVALIDATE (sock);
2238 if (H_10X (statcode))
2240 DEBUGP (("Ignoring response\n"));
2246 hs->message = xstrdup (message);
2247 if (!opt.server_response)
2248 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
2249 message ? quotearg_style (escape_quoting_style, message) : "");
2252 logprintf (LOG_VERBOSE, "\n");
2253 print_server_response (resp, " ");
2256 if (!opt.ignore_length
2257 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
2261 parsed = str_to_wgint (hdrval, NULL, 10);
2262 if (parsed == WGINT_MAX && errno == ERANGE)
2265 #### If Content-Length is out of range, it most likely
2266 means that the file is larger than 2G and that we're
2267 compiled without LFS. In that case we should probably
2268 refuse to even attempt to download the file. */
2271 else if (parsed < 0)
2273 /* Negative Content-Length; nonsensical, so we can't
2274 assume any information about the content to receive. */
2281 /* Check for keep-alive related responses. */
2282 if (!inhibit_keep_alive && contlen != -1)
2284 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2286 if (0 == strcasecmp (hdrval, "Close"))
2291 chunked_transfer_encoding = false;
2292 if (resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval))
2293 && 0 == strcasecmp (hdrval, "chunked"))
2294 chunked_transfer_encoding = true;
2296 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2300 const char *scbeg, *scend;
2301 /* The jar should have been created by now. */
2302 assert (wget_cookie_jar != NULL);
2304 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2305 &scbeg, &scend)) != -1;
2308 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2309 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2310 u->path, set_cookie);
2315 /* The server has promised that it will not close the connection
2316 when we're done. This means that we can register it. */
2317 register_persistent (conn->host, conn->port, sock, using_ssl);
2319 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2321 /* Authorization is required. */
2323 /* Normally we are not interested in the response body.
2324 But if we are writing a WARC file we are: we like to keep everyting. */
2328 type = resp_header_strdup (resp, "Content-Type");
2329 err = read_response_body (hs, sock, NULL, contlen, 0,
2330 chunked_transfer_encoding,
2331 u->url, warc_timestamp_str,
2332 warc_request_uuid, warc_ip, type,
2336 if (err != RETRFINISHED || hs->res < 0)
2338 CLOSE_INVALIDATE (sock);
2340 xfree_null (message);
2346 CLOSE_FINISH (sock);
2350 /* Since WARC is disabled, we are not interested in the response body. */
2351 if (keep_alive && !head_only
2352 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2353 CLOSE_FINISH (sock);
2355 CLOSE_INVALIDATE (sock);
2358 pconn.authorized = false;
2359 if (!auth_finished && (user && passwd))
2361 /* IIS sends multiple copies of WWW-Authenticate, one with
2362 the value "negotiate", and other(s) with data. Loop over
2363 all the occurrences and pick the one we recognize. */
2365 const char *wabeg, *waend;
2366 char *www_authenticate = NULL;
2368 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2369 &wabeg, &waend)) != -1;
2371 if (known_authentication_scheme_p (wabeg, waend))
2373 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2377 if (!www_authenticate)
2379 /* If the authentication header is missing or
2380 unrecognized, there's no sense in retrying. */
2381 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2383 else if (!basic_auth_finished
2384 || !BEGINS_WITH (www_authenticate, "Basic"))
2387 pth = url_full_path (u);
2388 request_set_header (req, "Authorization",
2389 create_authorization_line (www_authenticate,
2391 request_method (req),
2395 if (BEGINS_WITH (www_authenticate, "NTLM"))
2397 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2399 /* Need to register this host as using basic auth,
2400 * so we automatically send creds next time. */
2401 register_basic_auth_host (u->host);
2404 xfree_null (message);
2407 goto retry_with_auth;
2411 /* We already did Basic auth, and it failed. Gotta
2415 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2417 xfree_null (message);
2422 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2424 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2426 pconn.authorized = true;
2429 /* Determine the local filename if needed. Notice that if -O is used
2430 * hstat.local_file is set by http_loop to the argument of -O. */
2431 if (!hs->local_file)
2433 char *local_file = NULL;
2435 /* Honor Content-Disposition whether possible. */
2436 if (!opt.content_disposition
2437 || !resp_header_copy (resp, "Content-Disposition",
2438 hdrval, sizeof (hdrval))
2439 || !parse_content_disposition (hdrval, &local_file))
2441 /* The Content-Disposition header is missing or broken.
2442 * Choose unique file name according to given URL. */
2443 hs->local_file = url_file_name (u, NULL);
2447 DEBUGP (("Parsed filename from Content-Disposition: %s\n",
2449 hs->local_file = url_file_name (u, local_file);
2453 /* TODO: perform this check only once. */
2454 if (!hs->existence_checked && file_exists_p (hs->local_file))
2456 if (opt.noclobber && !opt.output_document)
2458 /* If opt.noclobber is turned on and file already exists, do not
2459 retrieve the file. But if the output_document was given, then this
2460 test was already done and the file didn't exist. Hence the !opt.output_document */
2461 get_file_flags (hs->local_file, dt);
2465 xfree_null (message);
2466 return RETRUNNEEDED;
2468 else if (!ALLOW_CLOBBER)
2470 char *unique = unique_name (hs->local_file, true);
2471 if (unique != hs->local_file)
2472 xfree (hs->local_file);
2473 hs->local_file = unique;
2476 hs->existence_checked = true;
2478 /* Support timestamping */
2479 /* TODO: move this code out of gethttp. */
2480 if (opt.timestamping && !hs->timestamp_checked)
2482 size_t filename_len = strlen (hs->local_file);
2483 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2484 bool local_dot_orig_file_exists = false;
2485 char *local_filename = NULL;
2488 if (opt.backup_converted)
2489 /* If -K is specified, we'll act on the assumption that it was specified
2490 last time these files were downloaded as well, and instead of just
2491 comparing local file X against server file X, we'll compare local
2492 file X.orig (if extant, else X) against server file X. If -K
2493 _wasn't_ specified last time, or the server contains files called
2494 *.orig, -N will be back to not operating correctly with -k. */
2496 /* Would a single s[n]printf() call be faster? --dan
2498 Definitely not. sprintf() is horribly slow. It's a
2499 different question whether the difference between the two
2500 affects a program. Usually I'd say "no", but at one
2501 point I profiled Wget, and found that a measurable and
2502 non-negligible amount of time was lost calling sprintf()
2503 in url.c. Replacing sprintf with inline calls to
2504 strcpy() and number_to_string() made a difference.
2506 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2507 memcpy (filename_plus_orig_suffix + filename_len,
2508 ORIG_SFX, sizeof (ORIG_SFX));
2510 /* Try to stat() the .orig file. */
2511 if (stat (filename_plus_orig_suffix, &st) == 0)
2513 local_dot_orig_file_exists = true;
2514 local_filename = filename_plus_orig_suffix;
2518 if (!local_dot_orig_file_exists)
2519 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2520 if (stat (hs->local_file, &st) == 0)
2521 local_filename = hs->local_file;
2523 if (local_filename != NULL)
2524 /* There was a local file, so we'll check later to see if the version
2525 the server has is the same version we already have, allowing us to
2528 hs->orig_file_name = xstrdup (local_filename);
2529 hs->orig_file_size = st.st_size;
2530 hs->orig_file_tstamp = st.st_mtime;
2532 /* Modification time granularity is 2 seconds for Windows, so
2533 increase local time by 1 second for later comparison. */
2534 ++hs->orig_file_tstamp;
2541 hs->statcode = statcode;
2543 hs->error = xstrdup (_("Malformed status line"));
2545 hs->error = xstrdup (_("(no description)"));
2547 hs->error = xstrdup (message);
2548 xfree_null (message);
2550 type = resp_header_strdup (resp, "Content-Type");
2553 char *tmp = strchr (type, ';');
2556 /* sXXXav: only needed if IRI support is enabled */
2557 char *tmp2 = tmp + 1;
2559 while (tmp > type && c_isspace (tmp[-1]))
2563 /* Try to get remote encoding if needed */
2564 if (opt.enable_iri && !opt.encoding_remote)
2566 tmp = parse_charset (tmp2);
2568 set_content_encoding (iri, tmp);
2572 hs->newloc = resp_header_strdup (resp, "Location");
2573 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2575 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2577 wgint first_byte_pos, last_byte_pos, entity_length;
2578 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2581 contrange = first_byte_pos;
2582 contlen = last_byte_pos - first_byte_pos + 1;
2587 /* 20x responses are counted among successful by default. */
2588 if (H_20X (statcode))
2591 /* Return if redirected. */
2592 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2594 /* RFC2068 says that in case of the 300 (multiple choices)
2595 response, the server can output a preferred URL through
2596 `Location' header; otherwise, the request should be treated
2597 like GET. So, if the location is set, it will be a
2598 redirection; otherwise, just proceed normally. */
2599 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2603 logprintf (LOG_VERBOSE,
2604 _("Location: %s%s\n"),
2605 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2606 hs->newloc ? _(" [following]") : "");
2608 /* In case the caller cares to look... */
2613 /* Normally we are not interested in the response body of a redirect.
2614 But if we are writing a WARC file we are: we like to keep everyting. */
2617 int err = read_response_body (hs, sock, NULL, contlen, 0,
2618 chunked_transfer_encoding,
2619 u->url, warc_timestamp_str,
2620 warc_request_uuid, warc_ip, type,
2623 if (err != RETRFINISHED || hs->res < 0)
2625 CLOSE_INVALIDATE (sock);
2631 CLOSE_FINISH (sock);
2635 /* Since WARC is disabled, we are not interested in the response body. */
2636 if (keep_alive && !head_only
2637 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2638 CLOSE_FINISH (sock);
2640 CLOSE_INVALIDATE (sock);
2645 /* From RFC2616: The status codes 303 and 307 have
2646 been added for servers that wish to make unambiguously
2647 clear which kind of reaction is expected of the client.
2649 A 307 should be redirected using the same method,
2650 in other words, a POST should be preserved and not
2651 converted to a GET in that case.
2653 With strict adherence to RFC2616, POST requests are not
2654 converted to a GET request on 301 Permanent Redirect
2655 or 302 Temporary Redirect.
2657 A switch may be provided later based on the HTTPbis draft
2658 that allows clients to convert POST requests to GET
2659 requests on 301 and 302 response codes. */
2662 case HTTP_STATUS_TEMPORARY_REDIRECT:
2663 return NEWLOCATION_KEEP_POST;
2665 case HTTP_STATUS_MOVED_PERMANENTLY:
2666 if (opt.method && strcasecmp (opt.method, "post") != 0)
2667 return NEWLOCATION_KEEP_POST;
2669 case HTTP_STATUS_MOVED_TEMPORARILY:
2670 if (opt.method && strcasecmp (opt.method, "post") != 0)
2671 return NEWLOCATION_KEEP_POST;
2681 /* If content-type is not given, assume text/html. This is because
2682 of the multitude of broken CGI's that "forget" to generate the
2685 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2686 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2692 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2697 if (opt.adjust_extension)
2700 /* -E / --adjust-extension / adjust_extension = on was specified,
2701 and this is a text/html file. If some case-insensitive
2702 variation on ".htm[l]" isn't already the file's suffix,
2705 ensure_extension (hs, ".html", dt);
2707 else if (*dt & TEXTCSS)
2709 ensure_extension (hs, ".css", dt);
2713 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2714 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK
2715 && contrange == 0 && contlen >= 0 && hs->restval >= contlen))
2717 /* If `-c' is in use and the file has been fully downloaded (or
2718 the remote file has shrunk), Wget effectively requests bytes
2719 after the end of file and the server response with 416
2720 (or 200 with a <= Content-Length. */
2721 logputs (LOG_VERBOSE, _("\
2722 \n The file is already fully retrieved; nothing to do.\n\n"));
2723 /* In case the caller inspects. */
2726 /* Mark as successfully retrieved. */
2729 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2730 might be more bytes in the body. */
2732 return RETRUNNEEDED;
2734 if ((contrange != 0 && contrange != hs->restval)
2735 || (H_PARTIAL (statcode) && !contrange))
2737 /* The Range request was somehow misunderstood by the server.
2740 CLOSE_INVALIDATE (sock);
2747 hs->contlen = contlen + contrange;
2753 /* No need to print this output if the body won't be
2754 downloaded at all, or if the original server response is
2756 logputs (LOG_VERBOSE, _("Length: "));
2759 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2760 if (contlen + contrange >= 1024)
2761 logprintf (LOG_VERBOSE, " (%s)",
2762 human_readable (contlen + contrange));
2765 if (contlen >= 1024)
2766 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2767 number_to_static_string (contlen),
2768 human_readable (contlen));
2770 logprintf (LOG_VERBOSE, _(", %s remaining"),
2771 number_to_static_string (contlen));
2775 logputs (LOG_VERBOSE,
2776 opt.ignore_length ? _("ignored") : _("unspecified"));
2778 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2780 logputs (LOG_VERBOSE, "\n");
2784 /* Return if we have no intention of further downloading. */
2785 if ((!(*dt & RETROKF) && !opt.content_on_error) || head_only)
2787 /* In case the caller cares to look... */
2792 /* Normally we are not interested in the response body of a error responses.
2793 But if we are writing a WARC file we are: we like to keep everyting. */
2796 int err = read_response_body (hs, sock, NULL, contlen, 0,
2797 chunked_transfer_encoding,
2798 u->url, warc_timestamp_str,
2799 warc_request_uuid, warc_ip, type,
2802 if (err != RETRFINISHED || hs->res < 0)
2804 CLOSE_INVALIDATE (sock);
2810 CLOSE_FINISH (sock);
2814 /* Since WARC is disabled, we are not interested in the response body. */
2816 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2817 servers not to send body in response to a HEAD request, and
2818 those that do will likely be caught by test_socket_open.
2819 If not, they can be worked around using
2820 `--no-http-keep-alive'. */
2821 CLOSE_FINISH (sock);
2823 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2824 /* Successfully skipped the body; also keep using the socket. */
2825 CLOSE_FINISH (sock);
2827 CLOSE_INVALIDATE (sock);
2832 return RETRFINISHED;
2836 For VMS, define common fopen() optional arguments.
2839 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2840 # define FOPEN_BIN_FLAG 3
2841 #else /* def __VMS */
2842 # define FOPEN_BIN_FLAG true
2843 #endif /* def __VMS [else] */
2845 /* Open the local file. */
2848 mkalldirs (hs->local_file);
2850 rotate_backups (hs->local_file);
2857 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2858 #else /* def __VMS */
2859 fp = fopen (hs->local_file, "ab");
2860 #endif /* def __VMS [else] */
2862 else if (ALLOW_CLOBBER || count > 0)
2864 if (opt.unlink && file_exists_p (hs->local_file))
2866 int res = unlink (hs->local_file);
2869 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file,
2871 CLOSE_INVALIDATE (sock);
2882 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2883 #else /* def __VMS */
2884 fp = fopen (hs->local_file, "wb");
2885 #endif /* def __VMS [else] */
2889 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2890 if (!fp && errno == EEXIST)
2892 /* We cannot just invent a new name and use it (which is
2893 what functions like unique_create typically do)
2894 because we told the user we'd use this name.
2895 Instead, return and retry the download. */
2896 logprintf (LOG_NOTQUIET,
2897 _("%s has sprung into existence.\n"),
2899 CLOSE_INVALIDATE (sock);
2902 return FOPEN_EXCL_ERR;
2907 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2908 CLOSE_INVALIDATE (sock);
2917 /* Print fetch message, if opt.verbose. */
2920 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2921 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2925 err = read_response_body (hs, sock, fp, contlen, contrange,
2926 chunked_transfer_encoding,
2927 u->url, warc_timestamp_str,
2928 warc_request_uuid, warc_ip, type,
2931 /* Now we no longer need to store the response header. */
2936 CLOSE_FINISH (sock);
2938 CLOSE_INVALIDATE (sock);
2946 /* The genuine HTTP loop! This is the part where the retrieval is
2947 retried, and retried, and retried, and... */
2949 http_loop (struct url *u, struct url *original_url, char **newloc,
2950 char **local_file, const char *referer, int *dt, struct url *proxy,
2954 bool got_head = false; /* used for time-stamping and filename detection */
2955 bool time_came_from_head = false;
2956 bool got_name = false;
2959 uerr_t err, ret = TRYLIMEXC;
2960 time_t tmr = -1; /* remote time-stamp */
2961 struct http_stat hstat; /* HTTP status */
2963 bool send_head_first = true;
2965 bool force_full_retrieve = false;
2968 /* If we are writing to a WARC file: always retrieve the whole file. */
2969 if (opt.warc_filename != NULL)
2970 force_full_retrieve = true;
2973 /* Assert that no value for *LOCAL_FILE was passed. */
2974 assert (local_file == NULL || *local_file == NULL);
2976 /* Set LOCAL_FILE parameter. */
2977 if (local_file && opt.output_document)
2978 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2980 /* Reset NEWLOC parameter. */
2983 /* This used to be done in main(), but it's a better idea to do it
2984 here so that we don't go through the hoops if we're just using
2989 /* Warn on (likely bogus) wildcard usage in HTTP. */
2990 if (opt.ftp_glob && has_wildcards_p (u->path))
2991 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2993 /* Setup hstat struct. */
2995 hstat.referer = referer;
2997 if (opt.output_document)
2999 hstat.local_file = xstrdup (opt.output_document);
3002 else if (!opt.content_disposition)
3005 url_file_name (opt.trustservernames ? u : original_url, NULL);
3009 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
3011 /* If opt.noclobber is turned on and file already exists, do not
3012 retrieve the file. But if the output_document was given, then this
3013 test was already done and the file didn't exist. Hence the !opt.output_document */
3014 get_file_flags (hstat.local_file, dt);
3019 /* Reset the counter. */
3022 /* Reset the document type. */
3025 /* Skip preliminary HEAD request if we're not in spider mode. */
3027 send_head_first = false;
3029 /* Send preliminary HEAD request if --content-disposition and -c are used
3031 if (opt.content_disposition && opt.always_rest)
3032 send_head_first = true;
3034 /* Send preliminary HEAD request if -N is given and we have an existing
3035 * destination file. */
3036 file_name = url_file_name (opt.trustservernames ? u : original_url, NULL);
3037 if (opt.timestamping && (file_exists_p (file_name)
3038 || opt.content_disposition))
3039 send_head_first = true;
3045 /* Increment the pass counter. */
3047 sleep_between_retrievals (count);
3049 /* Get the current time string. */
3050 tms = datetime_str (time (NULL));
3052 if (opt.spider && !got_head)
3053 logprintf (LOG_VERBOSE, _("\
3054 Spider mode enabled. Check if remote file exists.\n"));
3056 /* Print fetch message, if opt.verbose. */
3059 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3064 sprintf (tmp, _("(try:%2d)"), count);
3065 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
3070 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
3075 ws_changetitle (hurl);
3080 /* Default document type is empty. However, if spider mode is
3081 on or time-stamping is employed, HEAD_ONLY commands is
3082 encoded within *dt. */
3083 if (send_head_first && !got_head)
3088 /* Decide whether or not to restart. */
3089 if (force_full_retrieve)
3090 hstat.restval = hstat.len;
3091 else if (opt.always_rest
3093 && stat (hstat.local_file, &st) == 0
3094 && S_ISREG (st.st_mode))
3095 /* When -c is used, continue from on-disk size. (Can't use
3096 hstat.len even if count>1 because we don't want a failed
3097 first attempt to clobber existing data.) */
3098 hstat.restval = st.st_size;
3100 /* otherwise, continue where the previous try left off */
3101 hstat.restval = hstat.len;
3105 /* Decide whether to send the no-cache directive. We send it in
3107 a) we're using a proxy, and we're past our first retrieval.
3108 Some proxies are notorious for caching incomplete data, so
3109 we require a fresh get.
3110 b) caching is explicitly inhibited. */
3111 if ((proxy && count > 1) /* a */
3112 || !opt.allow_cache) /* b */
3113 *dt |= SEND_NOCACHE;
3115 *dt &= ~SEND_NOCACHE;
3117 /* Try fetching the document, or at least its head. */
3118 err = gethttp (u, &hstat, dt, proxy, iri, count);
3121 tms = datetime_str (time (NULL));
3123 /* Get the new location (with or without the redirection). */
3125 *newloc = xstrdup (hstat.newloc);
3129 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
3130 case CONERROR: case READERR: case WRITEFAILED:
3131 case RANGEERR: case FOPEN_EXCL_ERR:
3132 /* Non-fatal errors continue executing the loop, which will
3133 bring them to "while" statement at the end, to judge
3134 whether the number of tries was exceeded. */
3135 printwhat (count, opt.ntry);
3137 case FWRITEERR: case FOPENERR:
3138 /* Another fatal error. */
3139 logputs (LOG_VERBOSE, "\n");
3140 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
3141 quote (hstat.local_file), strerror (errno));
3142 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
3143 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
3145 /* Fatal errors just return from the function. */
3149 /* A fatal WARC error. */
3150 logputs (LOG_VERBOSE, "\n");
3151 logprintf (LOG_NOTQUIET, _("Cannot write to WARC file.\n"));
3154 case WARC_TMP_FOPENERR: case WARC_TMP_FWRITEERR:
3155 /* A fatal WARC error. */
3156 logputs (LOG_VERBOSE, "\n");
3157 logprintf (LOG_NOTQUIET, _("Cannot write to temporary WARC file.\n"));
3161 /* Another fatal error. */
3162 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
3166 /* Another fatal error. */
3167 logputs (LOG_VERBOSE, "\n");
3168 logprintf (LOG_NOTQUIET, _("Cannot unlink %s (%s).\n"),
3169 quote (hstat.local_file), strerror (errno));
3173 case NEWLOCATION_KEEP_POST:
3174 /* Return the new location to the caller. */
3177 logprintf (LOG_NOTQUIET,
3178 _("ERROR: Redirection (%d) without location.\n"),
3188 /* The file was already fully retrieved. */
3192 /* Deal with you later. */
3195 /* All possibilities should have been exhausted. */
3199 if (!(*dt & RETROKF))
3204 /* #### Ugly ugly ugly! */
3205 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3206 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
3209 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
3211 && (hstat.statcode == 500 || hstat.statcode == 501))
3216 /* Maybe we should always keep track of broken links, not just in
3218 * Don't log error if it was UTF-8 encoded because we will try
3219 * once unencoded. */
3220 else if (opt.spider && !iri->utf8_encode)
3222 /* #### Again: ugly ugly ugly! */
3224 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3225 nonexisting_url (hurl);
3226 logprintf (LOG_NOTQUIET, _("\
3227 Remote file does not exist -- broken link!!!\n"));
3231 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
3232 tms, hstat.statcode,
3233 quotearg_style (escape_quoting_style, hstat.error));
3235 logputs (LOG_VERBOSE, "\n");
3241 /* Did we get the time-stamp? */
3244 got_head = true; /* no more time-stamping */
3246 if (opt.timestamping && !hstat.remote_time)
3248 logputs (LOG_NOTQUIET, _("\
3249 Last-modified header missing -- time-stamps turned off.\n"));
3251 else if (hstat.remote_time)
3253 /* Convert the date-string into struct tm. */
3254 tmr = http_atotm (hstat.remote_time);
3255 if (tmr == (time_t) (-1))
3256 logputs (LOG_VERBOSE, _("\
3257 Last-modified header invalid -- time-stamp ignored.\n"));
3258 if (*dt & HEAD_ONLY)
3259 time_came_from_head = true;
3262 if (send_head_first)
3264 /* The time-stamping section. */
3265 if (opt.timestamping)
3267 if (hstat.orig_file_name) /* Perform the following
3268 checks only if the file
3270 download already exists. */
3272 if (hstat.remote_time &&
3273 tmr != (time_t) (-1))
3275 /* Now time-stamping can be used validly.
3276 Time-stamping means that if the sizes of
3277 the local and remote file match, and local
3278 file is newer than the remote file, it will
3279 not be retrieved. Otherwise, the normal
3280 download procedure is resumed. */
3281 if (hstat.orig_file_tstamp >= tmr)
3283 if (hstat.contlen == -1
3284 || hstat.orig_file_size == hstat.contlen)
3286 logprintf (LOG_VERBOSE, _("\
3287 Server file no newer than local file %s -- not retrieving.\n\n"),
3288 quote (hstat.orig_file_name));
3294 logprintf (LOG_VERBOSE, _("\
3295 The sizes do not match (local %s) -- retrieving.\n"),
3296 number_to_static_string (hstat.orig_file_size));
3301 force_full_retrieve = true;
3302 logputs (LOG_VERBOSE,
3303 _("Remote file is newer, retrieving.\n"));
3306 logputs (LOG_VERBOSE, "\n");
3310 /* free_hstat (&hstat); */
3311 hstat.timestamp_checked = true;
3316 bool finished = true;
3321 logputs (LOG_VERBOSE, _("\
3322 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
3327 logprintf (LOG_VERBOSE, _("\
3328 Remote file exists but does not contain any link -- not retrieving.\n\n"));
3329 ret = RETROK; /* RETRUNNEEDED is not for caller. */
3336 logprintf (LOG_VERBOSE, _("\
3337 Remote file exists and could contain further links,\n\
3338 but recursion is disabled -- not retrieving.\n\n"));
3342 logprintf (LOG_VERBOSE, _("\
3343 Remote file exists.\n\n"));
3345 ret = RETROK; /* RETRUNNEEDED is not for caller. */
3350 logprintf (LOG_NONVERBOSE,
3351 _("%s URL: %s %2d %s\n"),
3352 tms, u->url, hstat.statcode,
3353 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
3360 count = 0; /* the retrieve count for HEAD is reset */
3362 } /* send_head_first */
3365 if (opt.useservertimestamps
3366 && (tmr != (time_t) (-1))
3367 && ((hstat.len == hstat.contlen) ||
3368 ((hstat.res == 0) && (hstat.contlen == -1))))
3370 const char *fl = NULL;
3371 set_local_file (&fl, hstat.local_file);
3375 /* Reparse time header, in case it's changed. */
3376 if (time_came_from_head
3377 && hstat.remote_time && hstat.remote_time[0])
3379 newtmr = http_atotm (hstat.remote_time);
3380 if (newtmr != (time_t)-1)
3386 /* End of time-stamping section. */
3388 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3389 total_download_time += hstat.dltime;
3391 if (hstat.len == hstat.contlen)
3395 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3397 logprintf (LOG_VERBOSE,
3399 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3400 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3402 write_to_stdout ? "" : quote (hstat.local_file),
3403 number_to_static_string (hstat.len),
3404 number_to_static_string (hstat.contlen));
3405 logprintf (LOG_NONVERBOSE,
3406 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3408 number_to_static_string (hstat.len),
3409 number_to_static_string (hstat.contlen),
3410 hstat.local_file, count);
3413 total_downloaded_bytes += hstat.rd_size;
3415 /* Remember that we downloaded the file for later ".orig" code. */
3416 if (*dt & ADDED_HTML_EXTENSION)
3417 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3419 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3424 else if (hstat.res == 0) /* No read error */
3426 if (hstat.contlen == -1) /* We don't know how much we were supposed
3427 to get, so assume we succeeded. */
3431 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3433 logprintf (LOG_VERBOSE,
3435 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3436 : _("%s (%s) - %s saved [%s]\n\n"),
3438 write_to_stdout ? "" : quote (hstat.local_file),
3439 number_to_static_string (hstat.len));
3440 logprintf (LOG_NONVERBOSE,
3441 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3442 tms, u->url, number_to_static_string (hstat.len),
3443 hstat.local_file, count);
3446 total_downloaded_bytes += hstat.rd_size;
3448 /* Remember that we downloaded the file for later ".orig" code. */
3449 if (*dt & ADDED_HTML_EXTENSION)
3450 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3452 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3457 else if (hstat.len < hstat.contlen) /* meaning we lost the
3458 connection too soon */
3460 logprintf (LOG_VERBOSE,
3461 _("%s (%s) - Connection closed at byte %s. "),
3462 tms, tmrate, number_to_static_string (hstat.len));
3463 printwhat (count, opt.ntry);
3466 else if (hstat.len != hstat.restval)
3467 /* Getting here would mean reading more data than
3468 requested with content-length, which we never do. */
3472 /* Getting here probably means that the content-length was
3473 * _less_ than the original, local size. We should probably
3474 * truncate or re-read, or something. FIXME */
3479 else /* from now on hstat.res can only be -1 */
3481 if (hstat.contlen == -1)
3483 logprintf (LOG_VERBOSE,
3484 _("%s (%s) - Read error at byte %s (%s)."),
3485 tms, tmrate, number_to_static_string (hstat.len),
3487 printwhat (count, opt.ntry);
3490 else /* hstat.res == -1 and contlen is given */
3492 logprintf (LOG_VERBOSE,
3493 _("%s (%s) - Read error at byte %s/%s (%s). "),
3495 number_to_static_string (hstat.len),
3496 number_to_static_string (hstat.contlen),
3498 printwhat (count, opt.ntry);
3504 while (!opt.ntry || (count < opt.ntry));
3507 if (ret == RETROK && local_file)
3508 *local_file = xstrdup (hstat.local_file);
3509 free_hstat (&hstat);
3514 /* Check whether the result of strptime() indicates success.
3515 strptime() returns the pointer to how far it got to in the string.
3516 The processing has been successful if the string is at `GMT' or
3517 `+X', or at the end of the string.
3519 In extended regexp parlance, the function returns 1 if P matches
3520 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3521 can return) is considered a failure and 0 is returned. */
3523 check_end (const char *p)
3527 while (c_isspace (*p))
3530 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3531 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3537 /* Convert the textual specification of time in TIME_STRING to the
3538 number of seconds since the Epoch.
3540 TIME_STRING can be in any of the three formats RFC2616 allows the
3541 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3542 as well as the time format used in the Set-Cookie header.
3543 Timezones are ignored, and should be GMT.
3545 Return the computed time_t representation, or -1 if the conversion
3548 This function uses strptime with various string formats for parsing
3549 TIME_STRING. This results in a parser that is not as lenient in
3550 interpreting TIME_STRING as I would like it to be. Being based on
3551 strptime, it always allows shortened months, one-digit days, etc.,
3552 but due to the multitude of formats in which time can be
3553 represented, an ideal HTTP time parser would be even more
3554 forgiving. It should completely ignore things like week days and
3555 concentrate only on the various forms of representing years,
3556 months, days, hours, minutes, and seconds. For example, it would
3557 be nice if it accepted ISO 8601 out of the box.
3559 I've investigated free and PD code for this purpose, but none was
3560 usable. getdate was big and unwieldy, and had potential copyright
3561 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3562 distributed with phttpd, is excellent, but we cannot use it because
3563 it is not assigned to the FSF. So I stuck it with strptime. */
3566 http_atotm (const char *time_string)
3568 /* NOTE: Solaris strptime man page claims that %n and %t match white
3569 space, but that's not universally available. Instead, we simply
3570 use ` ' to mean "skip all WS", which works under all strptime
3571 implementations I've tested. */
3573 static const char *time_formats[] = {
3574 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3575 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3576 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3577 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3578 (used in Set-Cookie, defined in the
3579 Netscape cookie specification.) */
3581 const char *oldlocale;
3582 char savedlocale[256];
3584 time_t ret = (time_t) -1;
3586 /* Solaris strptime fails to recognize English month names in
3587 non-English locales, which we work around by temporarily setting
3588 locale to C before invoking strptime. */
3589 oldlocale = setlocale (LC_TIME, NULL);
3592 size_t l = strlen (oldlocale) + 1;
3593 if (l >= sizeof savedlocale)
3594 savedlocale[0] = '\0';
3596 memcpy (savedlocale, oldlocale, l);
3598 else savedlocale[0] = '\0';
3600 setlocale (LC_TIME, "C");
3602 for (i = 0; i < countof (time_formats); i++)
3606 /* Some versions of strptime use the existing contents of struct
3607 tm to recalculate the date according to format. Zero it out
3608 to prevent stack garbage from influencing strptime. */
3611 if (check_end (strptime (time_string, time_formats[i], &t)))
3618 /* Restore the previous locale. */
3620 setlocale (LC_TIME, savedlocale);
3625 /* Authorization support: We support three authorization schemes:
3627 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3629 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3630 consisting of answering to the server's challenge with the proper
3633 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3634 Stenberg for libcurl. Like digest, NTLM is based on a
3635 challenge-response mechanism, but unlike digest, it is non-standard
3636 (authenticates TCP connections rather than requests), undocumented
3637 and Microsoft-specific. */
3639 /* Create the authentication header contents for the `Basic' scheme.
3640 This is done by encoding the string "USER:PASS" to base64 and
3641 prepending the string "Basic " in front of it. */
3644 basic_authentication_encode (const char *user, const char *passwd)
3647 int len1 = strlen (user) + 1 + strlen (passwd);
3649 t1 = (char *)alloca (len1 + 1);
3650 sprintf (t1, "%s:%s", user, passwd);
3652 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3653 base64_encode (t1, len1, t2);
3655 return concat_strings ("Basic ", t2, (char *) 0);
3658 #define SKIP_WS(x) do { \
3659 while (c_isspace (*(x))) \
3663 #ifdef ENABLE_DIGEST
3664 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3665 an array of 16 bytes containing the hash keys, and BUF should be a
3666 buffer of 33 writable characters (32 for hex digits plus one for
3667 zero termination). */
3669 dump_hash (char *buf, const unsigned char *hash)
3673 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3675 *buf++ = XNUM_TO_digit (*hash >> 4);
3676 *buf++ = XNUM_TO_digit (*hash & 0xf);
3681 /* Take the line apart to find the challenge, and compose a digest
3682 authorization header. See RFC2069 section 2.1.2. */
3684 digest_authentication_encode (const char *au, const char *user,
3685 const char *passwd, const char *method,
3688 static char *realm, *opaque, *nonce, *qop, *algorithm;
3693 { "realm", &realm },
3694 { "opaque", &opaque },
3695 { "nonce", &nonce },
3697 { "algorithm", &algorithm }
3699 char cnonce[16] = "";
3703 param_token name, value;
3706 realm = opaque = nonce = qop = algorithm = NULL;
3708 au += 6; /* skip over `Digest' */
3709 while (extract_param (&au, &name, &value, ','))
3712 size_t namelen = name.e - name.b;
3713 for (i = 0; i < countof (options); i++)
3714 if (namelen == strlen (options[i].name)
3715 && 0 == strncmp (name.b, options[i].name,
3718 *options[i].variable = strdupdelim (value.b, value.e);
3723 if (qop != NULL && strcmp(qop,"auth"))
3725 logprintf (LOG_NOTQUIET, _("Unsupported quality of protection '%s'.\n"), qop);
3726 user = NULL; /* force freeing mem and return */
3729 if (algorithm != NULL && strcmp (algorithm,"MD5") && strcmp (algorithm,"MD5-sess"))
3731 logprintf (LOG_NOTQUIET, _("Unsupported algorithm '%s'.\n"), algorithm);
3732 user = NULL; /* force freeing mem and return */
3735 if (!realm || !nonce || !user || !passwd || !path || !method)
3738 xfree_null (opaque);
3741 xfree_null (algorithm);
3745 /* Calculate the digest value. */
3748 unsigned char hash[MD5_DIGEST_SIZE];
3749 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3750 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3752 /* A1BUF = H(user ":" realm ":" password) */
3753 md5_init_ctx (&ctx);
3754 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3755 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3756 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3757 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3758 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3759 md5_finish_ctx (&ctx, hash);
3761 dump_hash (a1buf, hash);
3763 if (! strcmp (algorithm, "MD5-sess"))
3765 /* A1BUF = H( H(user ":" realm ":" password) ":" nonce ":" cnonce ) */
3766 snprintf (cnonce, sizeof (cnonce), "%08x", random_number(INT_MAX));
3768 md5_init_ctx (&ctx);
3769 // md5_process_bytes (hash, MD5_DIGEST_SIZE, &ctx);
3770 md5_process_bytes (a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3771 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3772 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3773 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3774 md5_process_bytes ((unsigned char *)cnonce, strlen (cnonce), &ctx);
3775 md5_finish_ctx (&ctx, hash);
3777 dump_hash (a1buf, hash);
3780 /* A2BUF = H(method ":" path) */
3781 md5_init_ctx (&ctx);
3782 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3783 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3784 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3785 md5_finish_ctx (&ctx, hash);
3786 dump_hash (a2buf, hash);
3788 if (!strcmp(qop, "auth") || !strcmp (qop, "auth-int"))
3790 /* RFC 2617 Digest Access Authentication */
3791 /* generate random hex string */
3793 snprintf(cnonce, sizeof(cnonce), "%08x", random_number(INT_MAX));
3795 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" noncecount ":" clientnonce ":" qop ": " A2BUF) */
3796 md5_init_ctx (&ctx);
3797 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3798 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3799 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3800 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3801 md5_process_bytes ((unsigned char *)"00000001", 8, &ctx); /* TODO: keep track of server nonce values */
3802 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3803 md5_process_bytes ((unsigned char *)cnonce, strlen(cnonce), &ctx);
3804 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3805 md5_process_bytes ((unsigned char *)qop, strlen(qop), &ctx);
3806 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3807 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3808 md5_finish_ctx (&ctx, hash);
3812 /* RFC 2069 Digest Access Authentication */
3813 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3814 md5_init_ctx (&ctx);
3815 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3816 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3817 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3818 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3819 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3820 md5_finish_ctx (&ctx, hash);
3823 dump_hash (response_digest, hash);
3825 res_size = strlen (user)
3829 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3830 + (opaque ? strlen (opaque) : 0)
3831 + (algorithm ? strlen (algorithm) : 0)
3836 res = xmalloc (res_size);
3838 if (!strcmp(qop,"auth"))
3840 res_len = snprintf (res, res_size, "Digest "\
3841 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\""\
3842 ", qop=auth, nc=00000001, cnonce=\"%s\"",
3843 user, realm, nonce, path, response_digest, cnonce);
3848 res_len = snprintf (res, res_size, "Digest "\
3849 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3850 user, realm, nonce, path, response_digest);
3855 res_len += snprintf(res + res_len, res_size - res_len, ", opaque=\"%s\"", opaque);
3860 snprintf(res + res_len, res_size - res_len, ", algorithm=\"%s\"", algorithm);
3865 #endif /* ENABLE_DIGEST */
3867 /* Computing the size of a string literal must take into account that
3868 value returned by sizeof includes the terminating \0. */
3869 #define STRSIZE(literal) (sizeof (literal) - 1)
3871 /* Whether chars in [b, e) begin with the literal string provided as
3872 first argument and are followed by whitespace or terminating \0.
3873 The comparison is case-insensitive. */
3874 #define STARTS(literal, b, e) \
3876 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3877 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3878 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3879 || c_isspace (b[STRSIZE (literal)])))
3882 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3884 return STARTS ("Basic", hdrbeg, hdrend)
3885 #ifdef ENABLE_DIGEST
3886 || STARTS ("Digest", hdrbeg, hdrend)
3889 || STARTS ("NTLM", hdrbeg, hdrend)
3896 /* Create the HTTP authorization request header. When the
3897 `WWW-Authenticate' response header is seen, according to the
3898 authorization scheme specified in that header (`Basic' and `Digest'
3899 are supported by the current implementation), produce an
3900 appropriate HTTP authorization request header. */
3902 create_authorization_line (const char *au, const char *user,
3903 const char *passwd, const char *method,
3904 const char *path, bool *finished)
3906 /* We are called only with known schemes, so we can dispatch on the
3908 switch (c_toupper (*au))
3910 case 'B': /* Basic */
3912 return basic_authentication_encode (user, passwd);
3913 #ifdef ENABLE_DIGEST
3914 case 'D': /* Digest */
3916 return digest_authentication_encode (au, user, passwd, method, path);
3919 case 'N': /* NTLM */
3920 if (!ntlm_input (&pconn.ntlm, au))
3925 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3928 /* We shouldn't get here -- this function should be only called
3929 with values approved by known_authentication_scheme_p. */
3937 if (!wget_cookie_jar)
3938 wget_cookie_jar = cookie_jar_new ();
3939 if (opt.cookies_input && !cookies_loaded_p)
3941 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3942 cookies_loaded_p = true;
3949 if (wget_cookie_jar)
3950 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3956 xfree_null (pconn.host);
3957 if (wget_cookie_jar)
3958 cookie_jar_delete (wget_cookie_jar);
3962 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3964 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3966 int len = strlen (ext);
3969 strncpy (shortext, ext, len - 1);
3970 shortext[len - 1] = '\0';
3973 if (last_period_in_local_filename == NULL
3974 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3975 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3977 int local_filename_len = strlen (hs->local_file);
3978 /* Resize the local file, allowing for ".html" preceded by
3979 optional ".NUMBER". */
3980 hs->local_file = xrealloc (hs->local_file,
3981 local_filename_len + 24 + len);
3982 strcpy (hs->local_file + local_filename_len, ext);
3983 /* If clobbering is not allowed and the file, as named,
3984 exists, tack on ".NUMBER.html" instead. */
3985 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3989 sprintf (hs->local_file + local_filename_len,
3990 ".%d%s", ext_num++, ext);
3991 while (file_exists_p (hs->local_file));
3993 *dt |= ADDED_HTML_EXTENSION;
4001 test_parse_content_disposition()
4009 { "filename=\"file.ext\"", "file.ext", true },
4010 { "attachment; filename=\"file.ext\"", "file.ext", true },
4011 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
4012 { "attachment", NULL, false },
4013 { "attachement; filename*=UTF-8'en-US'hello.txt", "hello.txt", true },
4014 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", "helloworld.txt", true },
4017 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
4022 res = parse_content_disposition (test_array[i].hdrval, &filename);
4024 mu_assert ("test_parse_content_disposition: wrong result",
4025 res == test_array[i].result
4027 || 0 == strcmp (test_array[i].filename, filename)));
4033 #endif /* TESTING */
4036 * vim: et sts=2 sw=2 cino+={s