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. At least request_set_method must be
151 called before the request can be used. */
153 static struct request *
156 struct request *req = xnew0 (struct request);
158 req->headers = xnew_array (struct request_header, req->hcapacity);
162 /* Set the request's method and its arguments. METH should be a
163 literal string (or it should outlive the request) because it will
164 not be freed. ARG will be freed by request_free. */
167 request_set_method (struct request *req, const char *meth, char *arg)
173 /* Return the method string passed with the last call to
174 request_set_method. */
177 request_method (const struct request *req)
182 /* Free one header according to the release policy specified with
183 request_set_header. */
186 release_header (struct request_header *hdr)
188 switch (hdr->release_policy)
205 /* Set the request named NAME to VALUE. Specifically, this means that
206 a "NAME: VALUE\r\n" header line will be used in the request. If a
207 header with the same name previously existed in the request, its
208 value will be replaced by this one. A NULL value means do nothing.
210 RELEASE_POLICY determines whether NAME and VALUE should be released
211 (freed) with request_free. Allowed values are:
213 - rel_none - don't free NAME or VALUE
214 - rel_name - free NAME when done
215 - rel_value - free VALUE when done
216 - rel_both - free both NAME and VALUE when done
218 Setting release policy is useful when arguments come from different
219 sources. For example:
221 // Don't free literal strings!
222 request_set_header (req, "Pragma", "no-cache", rel_none);
224 // Don't free a global variable, we'll need it later.
225 request_set_header (req, "Referer", opt.referer, rel_none);
227 // Value freshly allocated, free it when done.
228 request_set_header (req, "Range",
229 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
234 request_set_header (struct request *req, const char *name, const char *value,
235 enum rp release_policy)
237 struct request_header *hdr;
242 /* A NULL value is a no-op; if freeing the name is requested,
243 free it now to avoid leaks. */
244 if (release_policy == rel_name || release_policy == rel_both)
245 xfree ((void *)name);
249 for (i = 0; i < req->hcount; i++)
251 hdr = &req->headers[i];
252 if (0 == strcasecmp (name, hdr->name))
254 /* Replace existing header. */
255 release_header (hdr);
256 hdr->name = (void *)name;
257 hdr->value = (void *)value;
258 hdr->release_policy = release_policy;
263 /* Install new header. */
265 if (req->hcount >= req->hcapacity)
267 req->hcapacity <<= 1;
268 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
270 hdr = &req->headers[req->hcount++];
271 hdr->name = (void *)name;
272 hdr->value = (void *)value;
273 hdr->release_policy = release_policy;
276 /* Like request_set_header, but sets the whole header line, as
277 provided by the user using the `--header' option. For example,
278 request_set_user_header (req, "Foo: bar") works just like
279 request_set_header (req, "Foo", "bar"). */
282 request_set_user_header (struct request *req, const char *header)
285 const char *p = strchr (header, ':');
288 BOUNDED_TO_ALLOCA (header, p, name);
290 while (c_isspace (*p))
292 request_set_header (req, xstrdup (name), (char *) p, rel_name);
295 /* Remove the header with specified name from REQ. Returns true if
296 the header was actually removed, false otherwise. */
299 request_remove_header (struct request *req, const char *name)
302 for (i = 0; i < req->hcount; i++)
304 struct request_header *hdr = &req->headers[i];
305 if (0 == strcasecmp (name, hdr->name))
307 release_header (hdr);
308 /* Move the remaining headers by one. */
309 if (i < req->hcount - 1)
310 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
318 #define APPEND(p, str) do { \
319 int A_len = strlen (str); \
320 memcpy (p, str, A_len); \
324 /* Construct the request and write it to FD using fd_write.
325 If warc_tmp is set to a file pointer, the request string will
326 also be written to that file. */
329 request_send (const struct request *req, int fd, FILE *warc_tmp)
331 char *request_string, *p;
332 int i, size, write_error;
334 /* Count the request size. */
337 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
338 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
340 for (i = 0; i < req->hcount; i++)
342 struct request_header *hdr = &req->headers[i];
343 /* NAME ": " VALUE "\r\n" */
344 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
350 p = request_string = alloca_array (char, size);
352 /* Generate the request. */
354 APPEND (p, req->method); *p++ = ' ';
355 APPEND (p, req->arg); *p++ = ' ';
356 memcpy (p, "HTTP/1.1\r\n", 10); p += 10;
358 for (i = 0; i < req->hcount; i++)
360 struct request_header *hdr = &req->headers[i];
361 APPEND (p, hdr->name);
362 *p++ = ':', *p++ = ' ';
363 APPEND (p, hdr->value);
364 *p++ = '\r', *p++ = '\n';
367 *p++ = '\r', *p++ = '\n', *p++ = '\0';
368 assert (p - request_string == size);
372 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
374 /* Send the request to the server. */
376 write_error = fd_write (fd, request_string, size - 1, -1);
378 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
380 else if (warc_tmp != NULL)
382 /* Write a copy of the data to the WARC record. */
383 int warc_tmp_written = fwrite (request_string, 1, size - 1, warc_tmp);
384 if (warc_tmp_written != size - 1)
390 /* Release the resources used by REQ. */
393 request_free (struct request *req)
396 xfree_null (req->arg);
397 for (i = 0; i < req->hcount; i++)
398 release_header (&req->headers[i]);
399 xfree_null (req->headers);
403 static struct hash_table *basic_authed_hosts;
405 /* Find out if this host has issued a Basic challenge yet; if so, give
406 * it the username, password. A temporary measure until we can get
407 * proper authentication in place. */
410 maybe_send_basic_creds (const char *hostname, const char *user,
411 const char *passwd, struct request *req)
413 bool do_challenge = false;
415 if (opt.auth_without_challenge)
417 DEBUGP (("Auth-without-challenge set, sending Basic credentials.\n"));
420 else if (basic_authed_hosts
421 && hash_table_contains(basic_authed_hosts, hostname))
423 DEBUGP (("Found %s in basic_authed_hosts.\n", quote (hostname)));
428 DEBUGP (("Host %s has not issued a general basic challenge.\n",
433 request_set_header (req, "Authorization",
434 basic_authentication_encode (user, passwd),
441 register_basic_auth_host (const char *hostname)
443 if (!basic_authed_hosts)
445 basic_authed_hosts = make_nocase_string_hash_table (1);
447 if (!hash_table_contains(basic_authed_hosts, hostname))
449 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
450 DEBUGP (("Inserted %s into basic_authed_hosts\n", quote (hostname)));
455 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
456 PROMISED_SIZE bytes are sent over the wire -- if the file is
457 longer, read only that much; if the file is shorter, report an error.
458 If warc_tmp is set to a file pointer, the post data will
459 also be written to that file. */
462 body_file_send (int sock, const char *file_name, wgint promised_size, FILE *warc_tmp)
464 static char chunk[8192];
469 DEBUGP (("[writing BODY file %s ... ", file_name));
471 fp = fopen (file_name, "rb");
474 while (!feof (fp) && written < promised_size)
477 int length = fread (chunk, 1, sizeof (chunk), fp);
480 towrite = MIN (promised_size - written, length);
481 write_error = fd_write (sock, chunk, towrite, -1);
487 if (warc_tmp != NULL)
489 /* Write a copy of the data to the WARC record. */
490 int warc_tmp_written = fwrite (chunk, 1, towrite, warc_tmp);
491 if (warc_tmp_written != towrite)
501 /* If we've written less than was promised, report a (probably
502 nonsensical) error rather than break the promise. */
503 if (written < promised_size)
509 assert (written == promised_size);
510 DEBUGP (("done]\n"));
514 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
515 If so, return the pointer to the position after the line, otherwise
516 return NULL. This is used as callback to fd_read_hunk. The data
517 between START and PEEKED has been read and cannot be "unread"; the
518 data after PEEKED has only been peeked. */
521 response_head_terminator (const char *start, const char *peeked, int peeklen)
525 /* If at first peek, verify whether HUNK starts with "HTTP". If
526 not, this is a HTTP/0.9 request and we must bail out without
528 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
531 /* Look for "\n[\r]\n", and return the following position if found.
532 Start two chars before the current to cover the possibility that
533 part of the terminator (e.g. "\n\r") arrived in the previous
535 p = peeked - start < 2 ? start : peeked - 2;
536 end = peeked + peeklen;
538 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
539 for (; p < end - 2; p++)
542 if (p[1] == '\r' && p[2] == '\n')
544 else if (p[1] == '\n')
547 /* p==end-2: check for \n\n directly preceding END. */
548 if (p[0] == '\n' && p[1] == '\n')
554 /* The maximum size of a single HTTP response we care to read. Rather
555 than being a limit of the reader implementation, this limit
556 prevents Wget from slurping all available memory upon encountering
557 malicious or buggy server output, thus protecting the user. Define
558 it to 0 to remove the limit. */
560 #define HTTP_RESPONSE_MAX_SIZE 65536
562 /* Read the HTTP request head from FD and return it. The error
563 conditions are the same as with fd_read_hunk.
565 To support HTTP/0.9 responses, this function tries to make sure
566 that the data begins with "HTTP". If this is not the case, no data
567 is read and an empty request is returned, so that the remaining
568 data can be treated as body. */
571 read_http_response_head (int fd)
573 return fd_read_hunk (fd, response_head_terminator, 512,
574 HTTP_RESPONSE_MAX_SIZE);
578 /* The response data. */
581 /* The array of pointers that indicate where each header starts.
582 For example, given this HTTP response:
589 The headers are located like this:
591 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
593 headers[0] headers[1] headers[2] headers[3]
595 I.e. headers[0] points to the beginning of the request,
596 headers[1] points to the end of the first header and the
597 beginning of the second one, etc. */
599 const char **headers;
602 /* Create a new response object from the text of the HTTP response,
603 available in HEAD. That text is automatically split into
604 constituent header lines for fast retrieval using
607 static struct response *
608 resp_new (const char *head)
613 struct response *resp = xnew0 (struct response);
618 /* Empty head means that we're dealing with a headerless
619 (HTTP/0.9) response. In that case, don't set HEADERS at
624 /* Split HEAD into header lines, so that resp_header_* functions
625 don't need to do this over and over again. */
631 DO_REALLOC (resp->headers, size, count + 1, const char *);
632 resp->headers[count++] = hdr;
634 /* Break upon encountering an empty line. */
635 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
638 /* Find the end of HDR, including continuations. */
641 const char *end = strchr (hdr, '\n');
647 while (*hdr == ' ' || *hdr == '\t');
649 DO_REALLOC (resp->headers, size, count + 1, const char *);
650 resp->headers[count] = NULL;
655 /* Locate the header named NAME in the request data, starting with
656 position START. This allows the code to loop through the request
657 data, filtering for all requests of a given name. Returns the
658 found position, or -1 for failure. The code that uses this
659 function typically looks like this:
661 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
662 ... do something with header ...
664 If you only care about one header, use resp_header_get instead of
668 resp_header_locate (const struct response *resp, const char *name, int start,
669 const char **begptr, const char **endptr)
672 const char **headers = resp->headers;
675 if (!headers || !headers[1])
678 name_len = strlen (name);
684 for (; headers[i + 1]; i++)
686 const char *b = headers[i];
687 const char *e = headers[i + 1];
689 && b[name_len] == ':'
690 && 0 == strncasecmp (b, name, name_len))
693 while (b < e && c_isspace (*b))
695 while (b < e && c_isspace (e[-1]))
705 /* Find and retrieve the header named NAME in the request data. If
706 found, set *BEGPTR to its starting, and *ENDPTR to its ending
707 position, and return true. Otherwise return false.
709 This function is used as a building block for resp_header_copy
710 and resp_header_strdup. */
713 resp_header_get (const struct response *resp, const char *name,
714 const char **begptr, const char **endptr)
716 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
720 /* Copy the response header named NAME to buffer BUF, no longer than
721 BUFSIZE (BUFSIZE includes the terminating 0). If the header
722 exists, true is returned, false otherwise. If there should be no
723 limit on the size of the header, use resp_header_strdup instead.
725 If BUFSIZE is 0, no data is copied, but the boolean indication of
726 whether the header is present is still returned. */
729 resp_header_copy (const struct response *resp, const char *name,
730 char *buf, int bufsize)
733 if (!resp_header_get (resp, name, &b, &e))
737 int len = MIN (e - b, bufsize - 1);
738 memcpy (buf, b, len);
744 /* Return the value of header named NAME in RESP, allocated with
745 malloc. If such a header does not exist in RESP, return NULL. */
748 resp_header_strdup (const struct response *resp, const char *name)
751 if (!resp_header_get (resp, name, &b, &e))
753 return strdupdelim (b, e);
756 /* Parse the HTTP status line, which is of format:
758 HTTP-Version SP Status-Code SP Reason-Phrase
760 The function returns the status-code, or -1 if the status line
761 appears malformed. The pointer to "reason-phrase" message is
762 returned in *MESSAGE. */
765 resp_status (const struct response *resp, char **message)
772 /* For a HTTP/0.9 response, assume status 200. */
774 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
778 p = resp->headers[0];
779 end = resp->headers[1];
785 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
789 /* Match the HTTP version. This is optional because Gnutella
790 servers have been reported to not specify HTTP version. */
791 if (p < end && *p == '/')
794 while (p < end && c_isdigit (*p))
796 if (p < end && *p == '.')
798 while (p < end && c_isdigit (*p))
802 while (p < end && c_isspace (*p))
804 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
807 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
812 while (p < end && c_isspace (*p))
814 while (p < end && c_isspace (end[-1]))
816 *message = strdupdelim (p, end);
822 /* Release the resources used by RESP. */
825 resp_free (struct response *resp)
827 xfree_null (resp->headers);
831 /* Print a single line of response, the characters [b, e). We tried
833 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
834 but that failed to escape the non-printable characters and, in fact,
835 caused crashes in UTF-8 locales. */
838 print_response_line(const char *prefix, const char *b, const char *e)
841 BOUNDED_TO_ALLOCA(b, e, copy);
842 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
843 quotearg_style (escape_quoting_style, copy));
846 /* Print the server response, line by line, omitting the trailing CRLF
847 from individual header lines, and prefixed with PREFIX. */
850 print_server_response (const struct response *resp, const char *prefix)
855 for (i = 0; resp->headers[i + 1]; i++)
857 const char *b = resp->headers[i];
858 const char *e = resp->headers[i + 1];
860 if (b < e && e[-1] == '\n')
862 if (b < e && e[-1] == '\r')
864 print_response_line(prefix, b, e);
868 /* Parse the `Content-Range' header and extract the information it
869 contains. Returns true if successful, false otherwise. */
871 parse_content_range (const char *hdr, wgint *first_byte_ptr,
872 wgint *last_byte_ptr, wgint *entity_length_ptr)
876 /* Ancient versions of Netscape proxy server, presumably predating
877 rfc2068, sent out `Content-Range' without the "bytes"
879 if (0 == strncasecmp (hdr, "bytes", 5))
882 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
886 while (c_isspace (*hdr))
891 if (!c_isdigit (*hdr))
893 for (num = 0; c_isdigit (*hdr); hdr++)
894 num = 10 * num + (*hdr - '0');
895 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
897 *first_byte_ptr = num;
899 for (num = 0; c_isdigit (*hdr); hdr++)
900 num = 10 * num + (*hdr - '0');
901 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
903 *last_byte_ptr = num;
908 for (num = 0; c_isdigit (*hdr); hdr++)
909 num = 10 * num + (*hdr - '0');
910 *entity_length_ptr = num;
914 /* Read the body of the request, but don't store it anywhere and don't
915 display a progress gauge. This is useful for reading the bodies of
916 administrative responses to which we will soon issue another
917 request. The response is not useful to the user, but reading it
918 allows us to continue using the same connection to the server.
920 If reading fails, false is returned, true otherwise. In debug
921 mode, the body is displayed for debugging purposes. */
924 skip_short_body (int fd, wgint contlen, bool chunked)
927 SKIP_SIZE = 512, /* size of the download buffer */
928 SKIP_THRESHOLD = 4096 /* the largest size we read */
930 wgint remaining_chunk_size = 0;
931 char dlbuf[SKIP_SIZE + 1];
932 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
934 assert (contlen != -1 || contlen);
936 /* If the body is too large, it makes more sense to simply close the
937 connection than to try to read the body. */
938 if (contlen > SKIP_THRESHOLD)
941 while (contlen > 0 || chunked)
946 if (remaining_chunk_size == 0)
948 char *line = fd_read_line (fd);
953 remaining_chunk_size = strtol (line, &endl, 16);
956 if (remaining_chunk_size == 0)
958 line = fd_read_line (fd);
964 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
967 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
969 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
972 /* Don't normally report the error since this is an
973 optimization that should be invisible to the user. */
974 DEBUGP (("] aborting (%s).\n",
975 ret < 0 ? fd_errstr (fd) : "EOF received"));
982 remaining_chunk_size -= ret;
983 if (remaining_chunk_size == 0)
985 char *line = fd_read_line (fd);
993 /* Safe even if %.*s bogusly expects terminating \0 because
994 we've zero-terminated dlbuf above. */
995 DEBUGP (("%.*s", ret, dlbuf));
998 DEBUGP (("] done.\n"));
1002 #define NOT_RFC2231 0
1003 #define RFC2231_NOENCODING 1
1004 #define RFC2231_ENCODING 2
1006 /* extract_param extracts the parameter name into NAME.
1007 However, if the parameter name is in RFC2231 format then
1008 this function adjusts NAME by stripping of the trailing
1009 characters that are not part of the name but are present to
1010 indicate the presence of encoding information in the value
1011 or a fragment of a long parameter value
1014 modify_param_name(param_token *name)
1016 const char *delim1 = memchr (name->b, '*', name->e - name->b);
1017 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
1023 result = NOT_RFC2231;
1025 else if(delim1 == delim2)
1027 if ((name->e - 1) == delim1)
1029 result = RFC2231_ENCODING;
1033 result = RFC2231_NOENCODING;
1040 result = RFC2231_ENCODING;
1045 /* extract_param extract the paramater value into VALUE.
1046 Like modify_param_name this function modifies VALUE by
1047 stripping off the encoding information from the actual value
1050 modify_param_value (param_token *value, int encoding_type )
1052 if (RFC2231_ENCODING == encoding_type)
1054 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1055 if ( delim != NULL )
1057 value->b = (delim+1);
1062 /* Extract a parameter from the string (typically an HTTP header) at
1063 **SOURCE and advance SOURCE to the next parameter. Return false
1064 when there are no more parameters to extract. The name of the
1065 parameter is returned in NAME, and the value in VALUE. If the
1066 parameter has no value, the token's value is zeroed out.
1068 For example, if *SOURCE points to the string "attachment;
1069 filename=\"foo bar\"", the first call to this function will return
1070 the token named "attachment" and no value, and the second call will
1071 return the token named "filename" and value "foo bar". The third
1072 call will return false, indicating no more valid tokens. */
1075 extract_param (const char **source, param_token *name, param_token *value,
1078 const char *p = *source;
1080 while (c_isspace (*p)) ++p;
1084 return false; /* no error; nothing more to extract */
1089 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1091 if (name->b == name->e)
1092 return false; /* empty name: error */
1093 while (c_isspace (*p)) ++p;
1094 if (*p == separator || !*p) /* no value */
1097 if (*p == separator) ++p;
1102 return false; /* error */
1104 /* *p is '=', extract value */
1106 while (c_isspace (*p)) ++p;
1107 if (*p == '"') /* quoted */
1110 while (*p && *p != '"') ++p;
1114 /* Currently at closing quote; find the end of param. */
1115 while (c_isspace (*p)) ++p;
1116 while (*p && *p != separator) ++p;
1117 if (*p == separator)
1120 /* garbage after closed quote, e.g. foo="bar"baz */
1126 while (*p && *p != separator) ++p;
1128 while (value->e != value->b && c_isspace (value->e[-1]))
1130 if (*p == separator) ++p;
1134 int param_type = modify_param_name(name);
1135 if (NOT_RFC2231 != param_type)
1137 modify_param_value(value, param_type);
1143 #undef RFC2231_NOENCODING
1144 #undef RFC2231_ENCODING
1146 /* Appends the string represented by VALUE to FILENAME */
1149 append_value_to_filename (char **filename, param_token const * const value)
1151 int original_length = strlen(*filename);
1152 int new_length = strlen(*filename) + (value->e - value->b);
1153 *filename = xrealloc (*filename, new_length+1);
1154 memcpy (*filename + original_length, value->b, (value->e - value->b));
1155 (*filename)[new_length] = '\0';
1159 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1161 /* Parse the contents of the `Content-Disposition' header, extracting
1162 the information useful to Wget. Content-Disposition is a header
1163 borrowed from MIME; when used in HTTP, it typically serves for
1164 specifying the desired file name of the resource. For example:
1166 Content-Disposition: attachment; filename="flora.jpg"
1168 Wget will skip the tokens it doesn't care about, such as
1169 "attachment" in the previous example; it will also skip other
1170 unrecognized params. If the header is syntactically correct and
1171 contains a file name, a copy of the file name is stored in
1172 *filename and true is returned. Otherwise, the function returns
1175 The file name is stripped of directory components and must not be
1178 Historically, this function returned filename prefixed with opt.dir_prefix,
1179 now that logic is handled by the caller, new code should pay attention,
1180 changed by crq, Sep 2010.
1184 parse_content_disposition (const char *hdr, char **filename)
1186 param_token name, value;
1188 while (extract_param (&hdr, &name, &value, ';'))
1190 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1191 if ( isFilename && value.b != NULL)
1193 /* Make the file name begin at the last slash or backslash. */
1194 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1195 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1196 if (last_slash && last_bs)
1197 value.b = 1 + MAX (last_slash, last_bs);
1198 else if (last_slash || last_bs)
1199 value.b = 1 + (last_slash ? last_slash : last_bs);
1200 if (value.b == value.e)
1204 append_value_to_filename (filename, &value);
1206 *filename = strdupdelim (value.b, value.e);
1217 /* Persistent connections. Currently, we cache the most recently used
1218 connection as persistent, provided that the HTTP server agrees to
1219 make it such. The persistence data is stored in the variables
1220 below. Ideally, it should be possible to cache an arbitrary fixed
1221 number of these connections. */
1223 /* Whether a persistent connection is active. */
1224 static bool pconn_active;
1227 /* The socket of the connection. */
1230 /* Host and port of the currently active persistent connection. */
1234 /* Whether a ssl handshake has occoured on this connection. */
1237 /* Whether the connection was authorized. This is only done by
1238 NTLM, which authorizes *connections* rather than individual
1239 requests. (That practice is peculiar for HTTP, but it is a
1240 useful optimization.) */
1244 /* NTLM data of the current connection. */
1245 struct ntlmdata ntlm;
1249 /* Mark the persistent connection as invalid and free the resources it
1250 uses. This is used by the CLOSE_* macros after they forcefully
1251 close a registered persistent connection. */
1254 invalidate_persistent (void)
1256 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1257 pconn_active = false;
1258 fd_close (pconn.socket);
1263 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1264 persistent. This will enable someone to use the same connection
1265 later. In the context of HTTP, this must be called only AFTER the
1266 response has been received and the server has promised that the
1267 connection will remain alive.
1269 If a previous connection was persistent, it is closed. */
1272 register_persistent (const char *host, int port, int fd, bool ssl)
1276 if (pconn.socket == fd)
1278 /* The connection FD is already registered. */
1283 /* The old persistent connection is still active; close it
1284 first. This situation arises whenever a persistent
1285 connection exists, but we then connect to a different
1286 host, and try to register a persistent connection to that
1288 invalidate_persistent ();
1292 pconn_active = true;
1294 pconn.host = xstrdup (host);
1297 pconn.authorized = false;
1299 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1302 /* Return true if a persistent connection is available for connecting
1306 persistent_available_p (const char *host, int port, bool ssl,
1307 bool *host_lookup_failed)
1309 /* First, check whether a persistent connection is active at all. */
1313 /* If we want SSL and the last connection wasn't or vice versa,
1314 don't use it. Checking for host and port is not enough because
1315 HTTP and HTTPS can apparently coexist on the same port. */
1316 if (ssl != pconn.ssl)
1319 /* If we're not connecting to the same port, we're not interested. */
1320 if (port != pconn.port)
1323 /* If the host is the same, we're in business. If not, there is
1324 still hope -- read below. */
1325 if (0 != strcasecmp (host, pconn.host))
1327 /* Check if pconn.socket is talking to HOST under another name.
1328 This happens often when both sites are virtual hosts
1329 distinguished only by name and served by the same network
1330 interface, and hence the same web server (possibly set up by
1331 the ISP and serving many different web sites). This
1332 admittedly unconventional optimization does not contradict
1333 HTTP and works well with popular server software. */
1337 struct address_list *al;
1340 /* Don't try to talk to two different SSL sites over the same
1341 secure connection! (Besides, it's not clear that
1342 name-based virtual hosting is even possible with SSL.) */
1345 /* If pconn.socket's peer is one of the IP addresses HOST
1346 resolves to, pconn.socket is for all intents and purposes
1347 already talking to HOST. */
1349 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1351 /* Can't get the peer's address -- something must be very
1352 wrong with the connection. */
1353 invalidate_persistent ();
1356 al = lookup_host (host, 0);
1359 *host_lookup_failed = true;
1363 found = address_list_contains (al, &ip);
1364 address_list_release (al);
1369 /* The persistent connection's peer address was found among the
1370 addresses HOST resolved to; therefore, pconn.sock is in fact
1371 already talking to HOST -- no need to reconnect. */
1374 /* Finally, check whether the connection is still open. This is
1375 important because most servers implement liberal (short) timeout
1376 on persistent connections. Wget can of course always reconnect
1377 if the connection doesn't work out, but it's nicer to know in
1378 advance. This test is a logical followup of the first test, but
1379 is "expensive" and therefore placed at the end of the list.
1381 (Current implementation of test_socket_open has a nice side
1382 effect that it treats sockets with pending data as "closed".
1383 This is exactly what we want: if a broken server sends message
1384 body in response to HEAD, or if it sends more than conent-length
1385 data, we won't reuse the corrupted connection.) */
1387 if (!test_socket_open (pconn.socket))
1389 /* Oops, the socket is no longer open. Now that we know that,
1390 let's invalidate the persistent connection before returning
1392 invalidate_persistent ();
1399 /* The idea behind these two CLOSE macros is to distinguish between
1400 two cases: one when the job we've been doing is finished, and we
1401 want to close the connection and leave, and two when something is
1402 seriously wrong and we're closing the connection as part of
1405 In case of keep_alive, CLOSE_FINISH should leave the connection
1406 open, while CLOSE_INVALIDATE should still close it.
1408 Note that the semantics of the flag `keep_alive' is "this
1409 connection *will* be reused (the server has promised not to close
1410 the connection once we're done)", while the semantics of
1411 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1412 active, registered connection". */
1414 #define CLOSE_FINISH(fd) do { \
1417 if (pconn_active && (fd) == pconn.socket) \
1418 invalidate_persistent (); \
1427 #define CLOSE_INVALIDATE(fd) do { \
1428 if (pconn_active && (fd) == pconn.socket) \
1429 invalidate_persistent (); \
1437 wgint len; /* received length */
1438 wgint contlen; /* expected length */
1439 wgint restval; /* the restart value */
1440 int res; /* the result of last read */
1441 char *rderrmsg; /* error message from read error */
1442 char *newloc; /* new location (redirection) */
1443 char *remote_time; /* remote time-stamp string */
1444 char *error; /* textual HTTP error */
1445 int statcode; /* status code */
1446 char *message; /* status message */
1447 wgint rd_size; /* amount of data read from socket */
1448 double dltime; /* time it took to download the data */
1449 const char *referer; /* value of the referer header. */
1450 char *local_file; /* local file name. */
1451 bool existence_checked; /* true if we already checked for a file's
1452 existence after having begun to download
1453 (needed in gethttp for when connection is
1454 interrupted/restarted. */
1455 bool timestamp_checked; /* true if pre-download time-stamping checks
1456 * have already been performed */
1457 char *orig_file_name; /* name of file to compare for time-stamping
1458 * (might be != local_file if -K is set) */
1459 wgint orig_file_size; /* size of file to compare for time-stamping */
1460 time_t orig_file_tstamp; /* time-stamp of file to compare for
1465 free_hstat (struct http_stat *hs)
1467 xfree_null (hs->newloc);
1468 xfree_null (hs->remote_time);
1469 xfree_null (hs->error);
1470 xfree_null (hs->rderrmsg);
1471 xfree_null (hs->local_file);
1472 xfree_null (hs->orig_file_name);
1473 xfree_null (hs->message);
1475 /* Guard against being called twice. */
1477 hs->remote_time = NULL;
1482 get_file_flags (const char *filename, int *dt)
1484 logprintf (LOG_VERBOSE, _("\
1485 File %s already there; not retrieving.\n\n"), quote (filename));
1486 /* If the file is there, we suppose it's retrieved OK. */
1489 /* #### Bogusness alert. */
1490 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1491 if (has_html_suffix_p (filename))
1495 /* Download the response body from the socket and writes it to
1496 an output file. The headers have already been read from the
1497 socket. If WARC is enabled, the response body will also be
1498 written to a WARC response record.
1500 hs, contlen, contrange, chunked_transfer_encoding and url are
1501 parameters from the gethttp method. fp is a pointer to the
1504 url, warc_timestamp_str, warc_request_uuid, warc_ip, type
1505 and statcode will be saved in the headers of the WARC record.
1506 The head parameter contains the HTTP headers of the response.
1508 If fp is NULL and WARC is enabled, the response body will be
1509 written only to the WARC file. If WARC is disabled and fp
1510 is a file pointer, the data will be written to the file.
1511 If fp is a file pointer and WARC is enabled, the body will
1512 be written to both destinations.
1514 Returns the error code. */
1516 read_response_body (struct http_stat *hs, int sock, FILE *fp, wgint contlen,
1517 wgint contrange, bool chunked_transfer_encoding,
1518 char *url, char *warc_timestamp_str, char *warc_request_uuid,
1519 ip_address *warc_ip, char *type, int statcode, char *head)
1521 int warc_payload_offset = 0;
1522 FILE *warc_tmp = NULL;
1525 if (opt.warc_filename != NULL)
1527 /* Open a temporary file where we can write the response before we
1528 add it to the WARC record. */
1529 warc_tmp = warc_tempfile ();
1530 if (warc_tmp == NULL)
1531 warcerr = WARC_TMP_FOPENERR;
1535 /* We should keep the response headers for the WARC record. */
1536 int head_len = strlen (head);
1537 int warc_tmp_written = fwrite (head, 1, head_len, warc_tmp);
1538 if (warc_tmp_written != head_len)
1539 warcerr = WARC_TMP_FWRITEERR;
1540 warc_payload_offset = head_len;
1545 if (warc_tmp != NULL)
1553 /* This confuses the timestamping code that checks for file size.
1554 #### The timestamping code should be smarter about file size. */
1555 if (opt.save_headers && hs->restval == 0)
1556 fwrite (head, 1, strlen (head), fp);
1559 /* Read the response body. */
1562 /* If content-length is present, read that much; otherwise, read
1563 until EOF. The HTTP spec doesn't require the server to
1564 actually close the connection when it's done sending data. */
1565 flags |= rb_read_exactly;
1566 if (fp != NULL && hs->restval > 0 && contrange == 0)
1567 /* If the server ignored our range request, instruct fd_read_body
1568 to skip the first RESTVAL bytes of body. */
1569 flags |= rb_skip_startpos;
1570 if (chunked_transfer_encoding)
1571 flags |= rb_chunked_transfer_encoding;
1573 hs->len = hs->restval;
1575 /* Download the response body and write it to fp.
1576 If we are working on a WARC file, we simultaneously write the
1577 response body to warc_tmp. */
1578 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1579 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1583 if (warc_tmp != NULL)
1585 /* Create a response record and write it to the WARC file.
1586 Note: per the WARC standard, the request and response should share
1587 the same date header. We re-use the timestamp of the request.
1588 The response record should also refer to the uuid of the request. */
1589 bool r = warc_write_response_record (url, warc_timestamp_str,
1590 warc_request_uuid, warc_ip,
1591 warc_tmp, warc_payload_offset,
1592 type, statcode, hs->newloc);
1594 /* warc_write_response_record has closed warc_tmp. */
1600 return RETRFINISHED;
1603 if (warc_tmp != NULL)
1608 /* Error while writing to fd. */
1611 else if (hs->res == -3)
1613 /* Error while writing to warc_tmp. */
1614 return WARC_TMP_FWRITEERR;
1619 hs->rderrmsg = xstrdup (fd_errstr (sock));
1620 return RETRFINISHED;
1624 #define BEGINS_WITH(line, string_constant) \
1625 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1626 && (c_isspace (line[sizeof (string_constant) - 1]) \
1627 || !line[sizeof (string_constant) - 1]))
1630 #define SET_USER_AGENT(req) do { \
1631 if (!opt.useragent) \
1632 request_set_header (req, "User-Agent", \
1633 aprintf ("Wget/%s (VMS %s %s)", \
1634 version_string, vms_arch(), vms_vers()), \
1636 else if (*opt.useragent) \
1637 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1639 #else /* def __VMS */
1640 #define SET_USER_AGENT(req) do { \
1641 if (!opt.useragent) \
1642 request_set_header (req, "User-Agent", \
1643 aprintf ("Wget/%s (%s)", \
1644 version_string, OS_TYPE), \
1646 else if (*opt.useragent) \
1647 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1649 #endif /* def __VMS [else] */
1651 /* The flags that allow clobbering the file (opening with "wb").
1652 Defined here to avoid repetition later. #### This will require
1654 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1655 || opt.dirstruct || opt.output_document)
1657 /* Retrieve a document through HTTP protocol. It recognizes status
1658 code, and correctly handles redirections. It closes the network
1659 socket. If it receives an error from the functions below it, it
1660 will print it if there is enough information to do so (almost
1661 always), returning the error to the caller (i.e. http_loop).
1663 Various HTTP parameters are stored to hs.
1665 If PROXY is non-NULL, the connection will be made to the proxy
1666 server, and u->url will be requested. */
1668 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1669 struct iri *iri, int count)
1671 struct request *req;
1674 char *user, *passwd;
1678 wgint contlen, contrange;
1685 /* Set to 1 when the authorization has already been sent and should
1686 not be tried again. */
1687 bool auth_finished = false;
1689 /* Set to 1 when just globally-set Basic authorization has been sent;
1690 * should prevent further Basic negotiations, but not other
1692 bool basic_auth_finished = false;
1694 /* Whether NTLM authentication is used for this request. */
1695 bool ntlm_seen = false;
1697 /* Whether our connection to the remote host is through SSL. */
1698 bool using_ssl = false;
1700 /* Whether a HEAD request will be issued (as opposed to GET or
1702 bool head_only = !!(*dt & HEAD_ONLY);
1705 struct response *resp;
1709 /* Declare WARC variables. */
1710 bool warc_enabled = (opt.warc_filename != NULL);
1711 FILE *warc_tmp = NULL;
1712 char warc_timestamp_str [21];
1713 char warc_request_uuid [48];
1714 ip_address *warc_ip = NULL;
1715 off_t warc_payload_offset = -1;
1717 /* Whether this connection will be kept alive after the HTTP request
1721 /* Is the server using the chunked transfer encoding? */
1722 bool chunked_transfer_encoding = false;
1724 /* Whether keep-alive should be inhibited. */
1725 bool inhibit_keep_alive =
1726 !opt.http_keep_alive || opt.ignore_length;
1728 /* Headers sent when using POST. */
1729 wgint body_data_size = 0;
1731 bool host_lookup_failed = false;
1734 if (u->scheme == SCHEME_HTTPS)
1736 /* Initialize the SSL context. After this has once been done,
1737 it becomes a no-op. */
1740 scheme_disable (SCHEME_HTTPS);
1741 logprintf (LOG_NOTQUIET,
1742 _("Disabling SSL due to encountered errors.\n"));
1743 return SSLINITFAILED;
1746 #endif /* HAVE_SSL */
1748 /* Initialize certain elements of struct http_stat. */
1752 hs->rderrmsg = NULL;
1754 hs->remote_time = NULL;
1760 /* Prepare the request to send. */
1762 req = request_new ();
1765 const char *meth = "GET";
1768 else if (opt.method)
1770 /* Use the full path, i.e. one that includes the leading slash and
1771 the query string. E.g. if u->path is "foo/bar" and u->query is
1772 "param=value", full_path will be "/foo/bar?param=value". */
1775 /* When using SSL over proxy, CONNECT establishes a direct
1776 connection to the HTTPS server. Therefore use the same
1777 argument as when talking to the server directly. */
1778 && u->scheme != SCHEME_HTTPS
1781 meth_arg = xstrdup (u->url);
1783 meth_arg = url_full_path (u);
1784 request_set_method (req, meth, meth_arg);
1787 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1788 if (*dt & SEND_NOCACHE)
1790 /* Cache-Control MUST be obeyed by all HTTP/1.1 caching mechanisms... */
1791 request_set_header (req, "Cache-Control", "no-cache, must-revalidate", rel_none);
1793 /* ... but some HTTP/1.0 caches doesn't implement Cache-Control. */
1794 request_set_header (req, "Pragma", "no-cache", rel_none);
1797 request_set_header (req, "Range",
1798 aprintf ("bytes=%s-",
1799 number_to_static_string (hs->restval)),
1801 SET_USER_AGENT (req);
1802 request_set_header (req, "Accept", "*/*", rel_none);
1804 /* Find the username and password for authentication. */
1807 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1808 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1809 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1811 /* We only do "site-wide" authentication with "global" user/password
1812 * values unless --auth-no-challange has been requested; URL user/password
1813 * info overrides. */
1814 if (user && passwd && (!u->user || opt.auth_without_challenge))
1816 /* If this is a host for which we've already received a Basic
1817 * challenge, we'll go ahead and send Basic authentication creds. */
1818 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1821 /* Generate the Host header, HOST:PORT. Take into account that:
1823 - Broken server-side software often doesn't recognize the PORT
1824 argument, so we must generate "Host: www.server.com" instead of
1825 "Host: www.server.com:80" (and likewise for https port).
1827 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1828 becomes ambiguous and needs to be rewritten as "Host:
1829 [3ffe:8100:200:2::2]:1234". */
1831 /* Formats arranged for hfmt[add_port][add_squares]. */
1832 static const char *hfmt[][2] = {
1833 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1835 int add_port = u->port != scheme_default_port (u->scheme);
1836 int add_squares = strchr (u->host, ':') != NULL;
1837 request_set_header (req, "Host",
1838 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1842 if (inhibit_keep_alive)
1843 request_set_header (req, "Connection", "Close", rel_none);
1847 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1850 request_set_header (req, "Connection", "Close", rel_none);
1851 request_set_header (req, "Proxy-Connection", "Keep-Alive", rel_none);
1858 if (opt.body_data || opt.body_file)
1860 request_set_header (req, "Content-Type",
1861 "application/x-www-form-urlencoded", rel_none);
1864 body_data_size = strlen (opt.body_data);
1867 body_data_size = file_size (opt.body_file);
1868 if (body_data_size == -1)
1870 logprintf (LOG_NOTQUIET, _("BODY data file %s missing: %s\n"),
1871 quote (opt.body_file), strerror (errno));
1875 request_set_header (req, "Content-Length",
1876 xstrdup (number_to_static_string (body_data_size)),
1882 /* We need to come back here when the initial attempt to retrieve
1883 without authorization header fails. (Expected to happen at least
1884 for the Digest authorization scheme.) */
1887 request_set_header (req, "Cookie",
1888 cookie_header (wget_cookie_jar,
1889 u->host, u->port, u->path,
1891 u->scheme == SCHEME_HTTPS
1898 /* Add the user headers. */
1899 if (opt.user_headers)
1902 for (i = 0; opt.user_headers[i]; i++)
1903 request_set_user_header (req, opt.user_headers[i]);
1909 char *proxy_user, *proxy_passwd;
1910 /* For normal username and password, URL components override
1911 command-line/wgetrc parameters. With proxy
1912 authentication, it's the reverse, because proxy URLs are
1913 normally the "permanent" ones, so command-line args
1914 should take precedence. */
1915 if (opt.proxy_user && opt.proxy_passwd)
1917 proxy_user = opt.proxy_user;
1918 proxy_passwd = opt.proxy_passwd;
1922 proxy_user = proxy->user;
1923 proxy_passwd = proxy->passwd;
1925 /* #### This does not appear right. Can't the proxy request,
1926 say, `Digest' authentication? */
1927 if (proxy_user && proxy_passwd)
1928 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1930 /* If we're using a proxy, we will be connecting to the proxy
1934 /* Proxy authorization over SSL is handled below. */
1936 if (u->scheme != SCHEME_HTTPS)
1938 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1943 /* Establish the connection. */
1945 if (inhibit_keep_alive)
1949 /* Look for a persistent connection to target host, unless a
1950 proxy is used. The exception is when SSL is in use, in which
1951 case the proxy is nothing but a passthrough to the target
1952 host, registered as a connection to the latter. */
1953 struct url *relevant = conn;
1955 if (u->scheme == SCHEME_HTTPS)
1959 if (persistent_available_p (relevant->host, relevant->port,
1961 relevant->scheme == SCHEME_HTTPS,
1965 &host_lookup_failed))
1967 int family = socket_family (pconn.socket, ENDPOINT_PEER);
1968 sock = pconn.socket;
1969 using_ssl = pconn.ssl;
1971 if (family == AF_INET6)
1972 logprintf (LOG_VERBOSE, _("Reusing existing connection to [%s]:%d.\n"),
1973 quotearg_style (escape_quoting_style, pconn.host),
1977 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1978 quotearg_style (escape_quoting_style, pconn.host),
1980 DEBUGP (("Reusing fd %d.\n", sock));
1981 if (pconn.authorized)
1982 /* If the connection is already authorized, the "Basic"
1983 authorization added by code above is unnecessary and
1985 request_remove_header (req, "Authorization");
1987 else if (host_lookup_failed)
1990 logprintf(LOG_NOTQUIET,
1991 _("%s: unable to resolve host address %s\n"),
1992 exec_name, quote (relevant->host));
1999 sock = connect_to_host (conn->host, conn->port);
2008 return (retryable_socket_connect_error (errno)
2009 ? CONERROR : CONIMPOSSIBLE);
2013 if (proxy && u->scheme == SCHEME_HTTPS)
2015 /* When requesting SSL URLs through proxies, use the
2016 CONNECT method to request passthrough. */
2017 struct request *connreq = request_new ();
2018 request_set_method (connreq, "CONNECT",
2019 aprintf ("%s:%d", u->host, u->port));
2020 SET_USER_AGENT (connreq);
2023 request_set_header (connreq, "Proxy-Authorization",
2024 proxyauth, rel_value);
2025 /* Now that PROXYAUTH is part of the CONNECT request,
2026 zero it out so we don't send proxy authorization with
2027 the regular request below. */
2030 /* Examples in rfc2817 use the Host header in CONNECT
2031 requests. I don't see how that gains anything, given
2032 that the contents of Host would be exactly the same as
2033 the contents of CONNECT. */
2035 write_error = request_send (connreq, sock, 0);
2036 request_free (connreq);
2037 if (write_error < 0)
2039 CLOSE_INVALIDATE (sock);
2044 head = read_http_response_head (sock);
2047 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
2049 CLOSE_INVALIDATE (sock);
2059 DEBUGP (("proxy responded with: [%s]\n", head));
2061 resp = resp_new (head);
2062 statcode = resp_status (resp, &message);
2065 char *tms = datetime_str (time (NULL));
2066 logprintf (LOG_VERBOSE, "%d\n", statcode);
2067 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
2068 quotearg_style (escape_quoting_style,
2069 _("Malformed status line")));
2074 hs->message = xstrdup (message);
2077 if (statcode != 200)
2080 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
2081 message ? quotearg_style (escape_quoting_style, message) : "?");
2082 xfree_null (message);
2086 xfree_null (message);
2088 /* SOCK is now *really* connected to u->host, so update CONN
2089 to reflect this. That way register_persistent will
2090 register SOCK as being connected to u->host:u->port. */
2094 if (conn->scheme == SCHEME_HTTPS)
2096 if (!ssl_connect_wget (sock, u->host))
2102 else if (!ssl_check_certificate (sock, u->host))
2106 return VERIFCERTERR;
2110 #endif /* HAVE_SSL */
2113 /* Open the temporary file where we will write the request. */
2116 warc_tmp = warc_tempfile ();
2117 if (warc_tmp == NULL)
2119 CLOSE_INVALIDATE (sock);
2121 return WARC_TMP_FOPENERR;
2126 warc_ip = (ip_address *) alloca (sizeof (ip_address));
2127 socket_ip_address (sock, warc_ip, ENDPOINT_PEER);
2131 /* Send the request to server. */
2132 write_error = request_send (req, sock, warc_tmp);
2134 if (write_error >= 0)
2138 DEBUGP (("[BODY data: %s]\n", opt.body_data));
2139 write_error = fd_write (sock, opt.body_data, body_data_size, -1);
2140 if (write_error >= 0 && warc_tmp != NULL)
2142 /* Remember end of headers / start of payload. */
2143 warc_payload_offset = ftello (warc_tmp);
2145 /* Write a copy of the data to the WARC record. */
2146 int warc_tmp_written = fwrite (opt.body_data, 1, body_data_size, warc_tmp);
2147 if (warc_tmp_written != body_data_size)
2151 else if (opt.body_file && body_data_size != 0)
2153 if (warc_tmp != NULL)
2154 /* Remember end of headers / start of payload */
2155 warc_payload_offset = ftello (warc_tmp);
2157 write_error = body_file_send (sock, opt.body_file, body_data_size, warc_tmp);
2161 if (write_error < 0)
2163 CLOSE_INVALIDATE (sock);
2166 if (warc_tmp != NULL)
2169 if (write_error == -2)
2170 return WARC_TMP_FWRITEERR;
2174 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
2175 proxy ? "Proxy" : "HTTP");
2184 /* Generate a timestamp and uuid for this request. */
2185 warc_timestamp (warc_timestamp_str);
2186 warc_uuid_str (warc_request_uuid);
2188 /* Create a request record and store it in the WARC file. */
2189 warc_result = warc_write_request_record (u->url, warc_timestamp_str,
2190 warc_request_uuid, warc_ip,
2191 warc_tmp, warc_payload_offset);
2194 CLOSE_INVALIDATE (sock);
2199 /* warc_write_request_record has also closed warc_tmp. */
2204 head = read_http_response_head (sock);
2209 logputs (LOG_NOTQUIET, _("No data received.\n"));
2210 CLOSE_INVALIDATE (sock);
2216 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
2218 CLOSE_INVALIDATE (sock);
2223 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
2225 resp = resp_new (head);
2227 /* Check for status line. */
2229 statcode = resp_status (resp, &message);
2232 char *tms = datetime_str (time (NULL));
2233 logprintf (LOG_VERBOSE, "%d\n", statcode);
2234 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
2235 quotearg_style (escape_quoting_style,
2236 _("Malformed status line")));
2237 CLOSE_INVALIDATE (sock);
2244 if (H_10X (statcode))
2246 DEBUGP (("Ignoring response\n"));
2252 hs->message = xstrdup (message);
2253 if (!opt.server_response)
2254 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
2255 message ? quotearg_style (escape_quoting_style, message) : "");
2258 logprintf (LOG_VERBOSE, "\n");
2259 print_server_response (resp, " ");
2262 if (!opt.ignore_length
2263 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
2267 parsed = str_to_wgint (hdrval, NULL, 10);
2268 if (parsed == WGINT_MAX && errno == ERANGE)
2271 #### If Content-Length is out of range, it most likely
2272 means that the file is larger than 2G and that we're
2273 compiled without LFS. In that case we should probably
2274 refuse to even attempt to download the file. */
2277 else if (parsed < 0)
2279 /* Negative Content-Length; nonsensical, so we can't
2280 assume any information about the content to receive. */
2287 /* Check for keep-alive related responses. */
2288 if (!inhibit_keep_alive && contlen != -1)
2290 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2292 if (0 == strcasecmp (hdrval, "Close"))
2297 chunked_transfer_encoding = false;
2298 if (resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval))
2299 && 0 == strcasecmp (hdrval, "chunked"))
2300 chunked_transfer_encoding = true;
2302 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2306 const char *scbeg, *scend;
2307 /* The jar should have been created by now. */
2308 assert (wget_cookie_jar != NULL);
2310 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2311 &scbeg, &scend)) != -1;
2314 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2315 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2316 u->path, set_cookie);
2321 /* The server has promised that it will not close the connection
2322 when we're done. This means that we can register it. */
2323 register_persistent (conn->host, conn->port, sock, using_ssl);
2325 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2327 /* Authorization is required. */
2329 /* Normally we are not interested in the response body.
2330 But if we are writing a WARC file we are: we like to keep everyting. */
2334 type = resp_header_strdup (resp, "Content-Type");
2335 err = read_response_body (hs, sock, NULL, contlen, 0,
2336 chunked_transfer_encoding,
2337 u->url, warc_timestamp_str,
2338 warc_request_uuid, warc_ip, type,
2342 if (err != RETRFINISHED || hs->res < 0)
2344 CLOSE_INVALIDATE (sock);
2346 xfree_null (message);
2352 CLOSE_FINISH (sock);
2356 /* Since WARC is disabled, we are not interested in the response body. */
2357 if (keep_alive && !head_only
2358 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2359 CLOSE_FINISH (sock);
2361 CLOSE_INVALIDATE (sock);
2364 pconn.authorized = false;
2365 if (!auth_finished && (user && passwd))
2367 /* IIS sends multiple copies of WWW-Authenticate, one with
2368 the value "negotiate", and other(s) with data. Loop over
2369 all the occurrences and pick the one we recognize. */
2371 const char *wabeg, *waend;
2372 char *www_authenticate = NULL;
2374 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2375 &wabeg, &waend)) != -1;
2377 if (known_authentication_scheme_p (wabeg, waend))
2379 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2383 if (!www_authenticate)
2385 /* If the authentication header is missing or
2386 unrecognized, there's no sense in retrying. */
2387 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2389 else if (!basic_auth_finished
2390 || !BEGINS_WITH (www_authenticate, "Basic"))
2393 pth = url_full_path (u);
2394 request_set_header (req, "Authorization",
2395 create_authorization_line (www_authenticate,
2397 request_method (req),
2401 if (BEGINS_WITH (www_authenticate, "NTLM"))
2403 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2405 /* Need to register this host as using basic auth,
2406 * so we automatically send creds next time. */
2407 register_basic_auth_host (u->host);
2410 xfree_null (message);
2413 goto retry_with_auth;
2417 /* We already did Basic auth, and it failed. Gotta
2421 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2423 xfree_null (message);
2428 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2430 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2432 pconn.authorized = true;
2435 /* Determine the local filename if needed. Notice that if -O is used
2436 * hstat.local_file is set by http_loop to the argument of -O. */
2437 if (!hs->local_file)
2439 char *local_file = NULL;
2441 /* Honor Content-Disposition whether possible. */
2442 if (!opt.content_disposition
2443 || !resp_header_copy (resp, "Content-Disposition",
2444 hdrval, sizeof (hdrval))
2445 || !parse_content_disposition (hdrval, &local_file))
2447 /* The Content-Disposition header is missing or broken.
2448 * Choose unique file name according to given URL. */
2449 hs->local_file = url_file_name (u, NULL);
2453 DEBUGP (("Parsed filename from Content-Disposition: %s\n",
2455 hs->local_file = url_file_name (u, local_file);
2459 /* TODO: perform this check only once. */
2460 if (!hs->existence_checked && file_exists_p (hs->local_file))
2462 if (opt.noclobber && !opt.output_document)
2464 /* If opt.noclobber is turned on and file already exists, do not
2465 retrieve the file. But if the output_document was given, then this
2466 test was already done and the file didn't exist. Hence the !opt.output_document */
2467 get_file_flags (hs->local_file, dt);
2471 xfree_null (message);
2472 return RETRUNNEEDED;
2474 else if (!ALLOW_CLOBBER)
2476 char *unique = unique_name (hs->local_file, true);
2477 if (unique != hs->local_file)
2478 xfree (hs->local_file);
2479 hs->local_file = unique;
2482 hs->existence_checked = true;
2484 /* Support timestamping */
2485 /* TODO: move this code out of gethttp. */
2486 if (opt.timestamping && !hs->timestamp_checked)
2488 size_t filename_len = strlen (hs->local_file);
2489 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2490 bool local_dot_orig_file_exists = false;
2491 char *local_filename = NULL;
2494 if (opt.backup_converted)
2495 /* If -K is specified, we'll act on the assumption that it was specified
2496 last time these files were downloaded as well, and instead of just
2497 comparing local file X against server file X, we'll compare local
2498 file X.orig (if extant, else X) against server file X. If -K
2499 _wasn't_ specified last time, or the server contains files called
2500 *.orig, -N will be back to not operating correctly with -k. */
2502 /* Would a single s[n]printf() call be faster? --dan
2504 Definitely not. sprintf() is horribly slow. It's a
2505 different question whether the difference between the two
2506 affects a program. Usually I'd say "no", but at one
2507 point I profiled Wget, and found that a measurable and
2508 non-negligible amount of time was lost calling sprintf()
2509 in url.c. Replacing sprintf with inline calls to
2510 strcpy() and number_to_string() made a difference.
2512 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2513 memcpy (filename_plus_orig_suffix + filename_len,
2514 ORIG_SFX, sizeof (ORIG_SFX));
2516 /* Try to stat() the .orig file. */
2517 if (stat (filename_plus_orig_suffix, &st) == 0)
2519 local_dot_orig_file_exists = true;
2520 local_filename = filename_plus_orig_suffix;
2524 if (!local_dot_orig_file_exists)
2525 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2526 if (stat (hs->local_file, &st) == 0)
2527 local_filename = hs->local_file;
2529 if (local_filename != NULL)
2530 /* There was a local file, so we'll check later to see if the version
2531 the server has is the same version we already have, allowing us to
2534 hs->orig_file_name = xstrdup (local_filename);
2535 hs->orig_file_size = st.st_size;
2536 hs->orig_file_tstamp = st.st_mtime;
2538 /* Modification time granularity is 2 seconds for Windows, so
2539 increase local time by 1 second for later comparison. */
2540 ++hs->orig_file_tstamp;
2547 hs->statcode = statcode;
2549 hs->error = xstrdup (_("Malformed status line"));
2551 hs->error = xstrdup (_("(no description)"));
2553 hs->error = xstrdup (message);
2554 xfree_null (message);
2556 type = resp_header_strdup (resp, "Content-Type");
2559 char *tmp = strchr (type, ';');
2562 /* sXXXav: only needed if IRI support is enabled */
2563 char *tmp2 = tmp + 1;
2565 while (tmp > type && c_isspace (tmp[-1]))
2569 /* Try to get remote encoding if needed */
2570 if (opt.enable_iri && !opt.encoding_remote)
2572 tmp = parse_charset (tmp2);
2574 set_content_encoding (iri, tmp);
2578 hs->newloc = resp_header_strdup (resp, "Location");
2579 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2581 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2583 wgint first_byte_pos, last_byte_pos, entity_length;
2584 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2587 contrange = first_byte_pos;
2588 contlen = last_byte_pos - first_byte_pos + 1;
2593 /* 20x responses are counted among successful by default. */
2594 if (H_20X (statcode))
2597 /* Return if redirected. */
2598 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2600 /* RFC2068 says that in case of the 300 (multiple choices)
2601 response, the server can output a preferred URL through
2602 `Location' header; otherwise, the request should be treated
2603 like GET. So, if the location is set, it will be a
2604 redirection; otherwise, just proceed normally. */
2605 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2609 logprintf (LOG_VERBOSE,
2610 _("Location: %s%s\n"),
2611 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2612 hs->newloc ? _(" [following]") : "");
2614 /* In case the caller cares to look... */
2619 /* Normally we are not interested in the response body of a redirect.
2620 But if we are writing a WARC file we are: we like to keep everyting. */
2623 int err = read_response_body (hs, sock, NULL, contlen, 0,
2624 chunked_transfer_encoding,
2625 u->url, warc_timestamp_str,
2626 warc_request_uuid, warc_ip, type,
2629 if (err != RETRFINISHED || hs->res < 0)
2631 CLOSE_INVALIDATE (sock);
2637 CLOSE_FINISH (sock);
2641 /* Since WARC is disabled, we are not interested in the response body. */
2642 if (keep_alive && !head_only
2643 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2644 CLOSE_FINISH (sock);
2646 CLOSE_INVALIDATE (sock);
2651 /* From RFC2616: The status codes 303 and 307 have
2652 been added for servers that wish to make unambiguously
2653 clear which kind of reaction is expected of the client.
2655 A 307 should be redirected using the same method,
2656 in other words, a POST should be preserved and not
2657 converted to a GET in that case. */
2658 if (statcode == HTTP_STATUS_TEMPORARY_REDIRECT)
2659 return NEWLOCATION_KEEP_POST;
2664 /* If content-type is not given, assume text/html. This is because
2665 of the multitude of broken CGI's that "forget" to generate the
2668 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2669 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2675 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2680 if (opt.adjust_extension)
2683 /* -E / --adjust-extension / adjust_extension = on was specified,
2684 and this is a text/html file. If some case-insensitive
2685 variation on ".htm[l]" isn't already the file's suffix,
2688 ensure_extension (hs, ".html", dt);
2690 else if (*dt & TEXTCSS)
2692 ensure_extension (hs, ".css", dt);
2696 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2697 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK
2698 && contrange == 0 && contlen >= 0 && hs->restval >= contlen))
2700 /* If `-c' is in use and the file has been fully downloaded (or
2701 the remote file has shrunk), Wget effectively requests bytes
2702 after the end of file and the server response with 416
2703 (or 200 with a <= Content-Length. */
2704 logputs (LOG_VERBOSE, _("\
2705 \n The file is already fully retrieved; nothing to do.\n\n"));
2706 /* In case the caller inspects. */
2709 /* Mark as successfully retrieved. */
2712 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2713 might be more bytes in the body. */
2715 return RETRUNNEEDED;
2717 if ((contrange != 0 && contrange != hs->restval)
2718 || (H_PARTIAL (statcode) && !contrange))
2720 /* The Range request was somehow misunderstood by the server.
2723 CLOSE_INVALIDATE (sock);
2730 hs->contlen = contlen + contrange;
2736 /* No need to print this output if the body won't be
2737 downloaded at all, or if the original server response is
2739 logputs (LOG_VERBOSE, _("Length: "));
2742 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2743 if (contlen + contrange >= 1024)
2744 logprintf (LOG_VERBOSE, " (%s)",
2745 human_readable (contlen + contrange));
2748 if (contlen >= 1024)
2749 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2750 number_to_static_string (contlen),
2751 human_readable (contlen));
2753 logprintf (LOG_VERBOSE, _(", %s remaining"),
2754 number_to_static_string (contlen));
2758 logputs (LOG_VERBOSE,
2759 opt.ignore_length ? _("ignored") : _("unspecified"));
2761 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2763 logputs (LOG_VERBOSE, "\n");
2767 /* Return if we have no intention of further downloading. */
2768 if ((!(*dt & RETROKF) && !opt.content_on_error) || head_only)
2770 /* In case the caller cares to look... */
2775 /* Normally we are not interested in the response body of a error responses.
2776 But if we are writing a WARC file we are: we like to keep everyting. */
2779 int err = read_response_body (hs, sock, NULL, contlen, 0,
2780 chunked_transfer_encoding,
2781 u->url, warc_timestamp_str,
2782 warc_request_uuid, warc_ip, type,
2785 if (err != RETRFINISHED || hs->res < 0)
2787 CLOSE_INVALIDATE (sock);
2793 CLOSE_FINISH (sock);
2797 /* Since WARC is disabled, we are not interested in the response body. */
2799 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2800 servers not to send body in response to a HEAD request, and
2801 those that do will likely be caught by test_socket_open.
2802 If not, they can be worked around using
2803 `--no-http-keep-alive'. */
2804 CLOSE_FINISH (sock);
2806 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2807 /* Successfully skipped the body; also keep using the socket. */
2808 CLOSE_FINISH (sock);
2810 CLOSE_INVALIDATE (sock);
2815 return RETRFINISHED;
2819 For VMS, define common fopen() optional arguments.
2822 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2823 # define FOPEN_BIN_FLAG 3
2824 #else /* def __VMS */
2825 # define FOPEN_BIN_FLAG true
2826 #endif /* def __VMS [else] */
2828 /* Open the local file. */
2831 mkalldirs (hs->local_file);
2833 rotate_backups (hs->local_file);
2840 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2841 #else /* def __VMS */
2842 fp = fopen (hs->local_file, "ab");
2843 #endif /* def __VMS [else] */
2845 else if (ALLOW_CLOBBER || count > 0)
2847 if (opt.unlink && file_exists_p (hs->local_file))
2849 int res = unlink (hs->local_file);
2852 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file,
2854 CLOSE_INVALIDATE (sock);
2865 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2866 #else /* def __VMS */
2867 fp = fopen (hs->local_file, "wb");
2868 #endif /* def __VMS [else] */
2872 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2873 if (!fp && errno == EEXIST)
2875 /* We cannot just invent a new name and use it (which is
2876 what functions like unique_create typically do)
2877 because we told the user we'd use this name.
2878 Instead, return and retry the download. */
2879 logprintf (LOG_NOTQUIET,
2880 _("%s has sprung into existence.\n"),
2882 CLOSE_INVALIDATE (sock);
2885 return FOPEN_EXCL_ERR;
2890 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2891 CLOSE_INVALIDATE (sock);
2900 /* Print fetch message, if opt.verbose. */
2903 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2904 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2908 err = read_response_body (hs, sock, fp, contlen, contrange,
2909 chunked_transfer_encoding,
2910 u->url, warc_timestamp_str,
2911 warc_request_uuid, warc_ip, type,
2914 /* Now we no longer need to store the response header. */
2919 CLOSE_FINISH (sock);
2921 CLOSE_INVALIDATE (sock);
2929 /* The genuine HTTP loop! This is the part where the retrieval is
2930 retried, and retried, and retried, and... */
2932 http_loop (struct url *u, struct url *original_url, char **newloc,
2933 char **local_file, const char *referer, int *dt, struct url *proxy,
2937 bool got_head = false; /* used for time-stamping and filename detection */
2938 bool time_came_from_head = false;
2939 bool got_name = false;
2942 uerr_t err, ret = TRYLIMEXC;
2943 time_t tmr = -1; /* remote time-stamp */
2944 struct http_stat hstat; /* HTTP status */
2946 bool send_head_first = true;
2948 bool force_full_retrieve = false;
2951 /* If we are writing to a WARC file: always retrieve the whole file. */
2952 if (opt.warc_filename != NULL)
2953 force_full_retrieve = true;
2956 /* Assert that no value for *LOCAL_FILE was passed. */
2957 assert (local_file == NULL || *local_file == NULL);
2959 /* Set LOCAL_FILE parameter. */
2960 if (local_file && opt.output_document)
2961 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2963 /* Reset NEWLOC parameter. */
2966 /* This used to be done in main(), but it's a better idea to do it
2967 here so that we don't go through the hoops if we're just using
2972 /* Warn on (likely bogus) wildcard usage in HTTP. */
2973 if (opt.ftp_glob && has_wildcards_p (u->path))
2974 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2976 /* Setup hstat struct. */
2978 hstat.referer = referer;
2980 if (opt.output_document)
2982 hstat.local_file = xstrdup (opt.output_document);
2985 else if (!opt.content_disposition)
2988 url_file_name (opt.trustservernames ? u : original_url, NULL);
2992 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2994 /* If opt.noclobber is turned on and file already exists, do not
2995 retrieve the file. But if the output_document was given, then this
2996 test was already done and the file didn't exist. Hence the !opt.output_document */
2997 get_file_flags (hstat.local_file, dt);
3002 /* Reset the counter. */
3005 /* Reset the document type. */
3008 /* Skip preliminary HEAD request if we're not in spider mode. */
3010 send_head_first = false;
3012 /* Send preliminary HEAD request if --content-disposition and -c are used
3014 if (opt.content_disposition && opt.always_rest)
3015 send_head_first = true;
3017 /* Send preliminary HEAD request if -N is given and we have an existing
3018 * destination file. */
3019 file_name = url_file_name (opt.trustservernames ? u : original_url, NULL);
3020 if (opt.timestamping && (file_exists_p (file_name)
3021 || opt.content_disposition))
3022 send_head_first = true;
3028 /* Increment the pass counter. */
3030 sleep_between_retrievals (count);
3032 /* Get the current time string. */
3033 tms = datetime_str (time (NULL));
3035 if (opt.spider && !got_head)
3036 logprintf (LOG_VERBOSE, _("\
3037 Spider mode enabled. Check if remote file exists.\n"));
3039 /* Print fetch message, if opt.verbose. */
3042 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3047 sprintf (tmp, _("(try:%2d)"), count);
3048 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
3053 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
3058 ws_changetitle (hurl);
3063 /* Default document type is empty. However, if spider mode is
3064 on or time-stamping is employed, HEAD_ONLY commands is
3065 encoded within *dt. */
3066 if (send_head_first && !got_head)
3071 /* Decide whether or not to restart. */
3072 if (force_full_retrieve)
3073 hstat.restval = hstat.len;
3074 else if (opt.always_rest
3076 && stat (hstat.local_file, &st) == 0
3077 && S_ISREG (st.st_mode))
3078 /* When -c is used, continue from on-disk size. (Can't use
3079 hstat.len even if count>1 because we don't want a failed
3080 first attempt to clobber existing data.) */
3081 hstat.restval = st.st_size;
3083 /* otherwise, continue where the previous try left off */
3084 hstat.restval = hstat.len;
3088 /* Decide whether to send the no-cache directive. We send it in
3090 a) we're using a proxy, and we're past our first retrieval.
3091 Some proxies are notorious for caching incomplete data, so
3092 we require a fresh get.
3093 b) caching is explicitly inhibited. */
3094 if ((proxy && count > 1) /* a */
3095 || !opt.allow_cache) /* b */
3096 *dt |= SEND_NOCACHE;
3098 *dt &= ~SEND_NOCACHE;
3100 /* Try fetching the document, or at least its head. */
3101 err = gethttp (u, &hstat, dt, proxy, iri, count);
3104 tms = datetime_str (time (NULL));
3106 /* Get the new location (with or without the redirection). */
3108 *newloc = xstrdup (hstat.newloc);
3112 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
3113 case CONERROR: case READERR: case WRITEFAILED:
3114 case RANGEERR: case FOPEN_EXCL_ERR:
3115 /* Non-fatal errors continue executing the loop, which will
3116 bring them to "while" statement at the end, to judge
3117 whether the number of tries was exceeded. */
3118 printwhat (count, opt.ntry);
3120 case FWRITEERR: case FOPENERR:
3121 /* Another fatal error. */
3122 logputs (LOG_VERBOSE, "\n");
3123 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
3124 quote (hstat.local_file), strerror (errno));
3125 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
3126 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
3128 /* Fatal errors just return from the function. */
3132 /* A fatal WARC error. */
3133 logputs (LOG_VERBOSE, "\n");
3134 logprintf (LOG_NOTQUIET, _("Cannot write to WARC file.\n"));
3137 case WARC_TMP_FOPENERR: case WARC_TMP_FWRITEERR:
3138 /* A fatal WARC error. */
3139 logputs (LOG_VERBOSE, "\n");
3140 logprintf (LOG_NOTQUIET, _("Cannot write to temporary WARC file.\n"));
3144 /* Another fatal error. */
3145 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
3149 /* Another fatal error. */
3150 logputs (LOG_VERBOSE, "\n");
3151 logprintf (LOG_NOTQUIET, _("Cannot unlink %s (%s).\n"),
3152 quote (hstat.local_file), strerror (errno));
3156 case NEWLOCATION_KEEP_POST:
3157 /* Return the new location to the caller. */
3160 logprintf (LOG_NOTQUIET,
3161 _("ERROR: Redirection (%d) without location.\n"),
3171 /* The file was already fully retrieved. */
3175 /* Deal with you later. */
3178 /* All possibilities should have been exhausted. */
3182 if (!(*dt & RETROKF))
3187 /* #### Ugly ugly ugly! */
3188 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3189 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
3192 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
3194 && (hstat.statcode == 500 || hstat.statcode == 501))
3199 /* Maybe we should always keep track of broken links, not just in
3201 * Don't log error if it was UTF-8 encoded because we will try
3202 * once unencoded. */
3203 else if (opt.spider && !iri->utf8_encode)
3205 /* #### Again: ugly ugly ugly! */
3207 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3208 nonexisting_url (hurl);
3209 logprintf (LOG_NOTQUIET, _("\
3210 Remote file does not exist -- broken link!!!\n"));
3214 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
3215 tms, hstat.statcode,
3216 quotearg_style (escape_quoting_style, hstat.error));
3218 logputs (LOG_VERBOSE, "\n");
3224 /* Did we get the time-stamp? */
3227 got_head = true; /* no more time-stamping */
3229 if (opt.timestamping && !hstat.remote_time)
3231 logputs (LOG_NOTQUIET, _("\
3232 Last-modified header missing -- time-stamps turned off.\n"));
3234 else if (hstat.remote_time)
3236 /* Convert the date-string into struct tm. */
3237 tmr = http_atotm (hstat.remote_time);
3238 if (tmr == (time_t) (-1))
3239 logputs (LOG_VERBOSE, _("\
3240 Last-modified header invalid -- time-stamp ignored.\n"));
3241 if (*dt & HEAD_ONLY)
3242 time_came_from_head = true;
3245 if (send_head_first)
3247 /* The time-stamping section. */
3248 if (opt.timestamping)
3250 if (hstat.orig_file_name) /* Perform the following
3251 checks only if the file
3253 download already exists. */
3255 if (hstat.remote_time &&
3256 tmr != (time_t) (-1))
3258 /* Now time-stamping can be used validly.
3259 Time-stamping means that if the sizes of
3260 the local and remote file match, and local
3261 file is newer than the remote file, it will
3262 not be retrieved. Otherwise, the normal
3263 download procedure is resumed. */
3264 if (hstat.orig_file_tstamp >= tmr)
3266 if (hstat.contlen == -1
3267 || hstat.orig_file_size == hstat.contlen)
3269 logprintf (LOG_VERBOSE, _("\
3270 Server file no newer than local file %s -- not retrieving.\n\n"),
3271 quote (hstat.orig_file_name));
3277 logprintf (LOG_VERBOSE, _("\
3278 The sizes do not match (local %s) -- retrieving.\n"),
3279 number_to_static_string (hstat.orig_file_size));
3284 force_full_retrieve = true;
3285 logputs (LOG_VERBOSE,
3286 _("Remote file is newer, retrieving.\n"));
3289 logputs (LOG_VERBOSE, "\n");
3293 /* free_hstat (&hstat); */
3294 hstat.timestamp_checked = true;
3299 bool finished = true;
3304 logputs (LOG_VERBOSE, _("\
3305 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
3310 logprintf (LOG_VERBOSE, _("\
3311 Remote file exists but does not contain any link -- not retrieving.\n\n"));
3312 ret = RETROK; /* RETRUNNEEDED is not for caller. */
3319 logprintf (LOG_VERBOSE, _("\
3320 Remote file exists and could contain further links,\n\
3321 but recursion is disabled -- not retrieving.\n\n"));
3325 logprintf (LOG_VERBOSE, _("\
3326 Remote file exists.\n\n"));
3328 ret = RETROK; /* RETRUNNEEDED is not for caller. */
3333 logprintf (LOG_NONVERBOSE,
3334 _("%s URL: %s %2d %s\n"),
3335 tms, u->url, hstat.statcode,
3336 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
3343 count = 0; /* the retrieve count for HEAD is reset */
3345 } /* send_head_first */
3348 if (opt.useservertimestamps
3349 && (tmr != (time_t) (-1))
3350 && ((hstat.len == hstat.contlen) ||
3351 ((hstat.res == 0) && (hstat.contlen == -1))))
3353 const char *fl = NULL;
3354 set_local_file (&fl, hstat.local_file);
3358 /* Reparse time header, in case it's changed. */
3359 if (time_came_from_head
3360 && hstat.remote_time && hstat.remote_time[0])
3362 newtmr = http_atotm (hstat.remote_time);
3363 if (newtmr != (time_t)-1)
3369 /* End of time-stamping section. */
3371 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3372 total_download_time += hstat.dltime;
3374 if (hstat.len == hstat.contlen)
3378 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3380 logprintf (LOG_VERBOSE,
3382 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3383 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3385 write_to_stdout ? "" : quote (hstat.local_file),
3386 number_to_static_string (hstat.len),
3387 number_to_static_string (hstat.contlen));
3388 logprintf (LOG_NONVERBOSE,
3389 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3391 number_to_static_string (hstat.len),
3392 number_to_static_string (hstat.contlen),
3393 hstat.local_file, count);
3396 total_downloaded_bytes += hstat.rd_size;
3398 /* Remember that we downloaded the file for later ".orig" code. */
3399 if (*dt & ADDED_HTML_EXTENSION)
3400 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3402 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3407 else if (hstat.res == 0) /* No read error */
3409 if (hstat.contlen == -1) /* We don't know how much we were supposed
3410 to get, so assume we succeeded. */
3414 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3416 logprintf (LOG_VERBOSE,
3418 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3419 : _("%s (%s) - %s saved [%s]\n\n"),
3421 write_to_stdout ? "" : quote (hstat.local_file),
3422 number_to_static_string (hstat.len));
3423 logprintf (LOG_NONVERBOSE,
3424 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3425 tms, u->url, number_to_static_string (hstat.len),
3426 hstat.local_file, count);
3429 total_downloaded_bytes += hstat.rd_size;
3431 /* Remember that we downloaded the file for later ".orig" code. */
3432 if (*dt & ADDED_HTML_EXTENSION)
3433 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3435 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3440 else if (hstat.len < hstat.contlen) /* meaning we lost the
3441 connection too soon */
3443 logprintf (LOG_VERBOSE,
3444 _("%s (%s) - Connection closed at byte %s. "),
3445 tms, tmrate, number_to_static_string (hstat.len));
3446 printwhat (count, opt.ntry);
3449 else if (hstat.len != hstat.restval)
3450 /* Getting here would mean reading more data than
3451 requested with content-length, which we never do. */
3455 /* Getting here probably means that the content-length was
3456 * _less_ than the original, local size. We should probably
3457 * truncate or re-read, or something. FIXME */
3462 else /* from now on hstat.res can only be -1 */
3464 if (hstat.contlen == -1)
3466 logprintf (LOG_VERBOSE,
3467 _("%s (%s) - Read error at byte %s (%s)."),
3468 tms, tmrate, number_to_static_string (hstat.len),
3470 printwhat (count, opt.ntry);
3473 else /* hstat.res == -1 and contlen is given */
3475 logprintf (LOG_VERBOSE,
3476 _("%s (%s) - Read error at byte %s/%s (%s). "),
3478 number_to_static_string (hstat.len),
3479 number_to_static_string (hstat.contlen),
3481 printwhat (count, opt.ntry);
3487 while (!opt.ntry || (count < opt.ntry));
3490 if (ret == RETROK && local_file)
3491 *local_file = xstrdup (hstat.local_file);
3492 free_hstat (&hstat);
3497 /* Check whether the result of strptime() indicates success.
3498 strptime() returns the pointer to how far it got to in the string.
3499 The processing has been successful if the string is at `GMT' or
3500 `+X', or at the end of the string.
3502 In extended regexp parlance, the function returns 1 if P matches
3503 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3504 can return) is considered a failure and 0 is returned. */
3506 check_end (const char *p)
3510 while (c_isspace (*p))
3513 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3514 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3520 /* Convert the textual specification of time in TIME_STRING to the
3521 number of seconds since the Epoch.
3523 TIME_STRING can be in any of the three formats RFC2616 allows the
3524 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3525 as well as the time format used in the Set-Cookie header.
3526 Timezones are ignored, and should be GMT.
3528 Return the computed time_t representation, or -1 if the conversion
3531 This function uses strptime with various string formats for parsing
3532 TIME_STRING. This results in a parser that is not as lenient in
3533 interpreting TIME_STRING as I would like it to be. Being based on
3534 strptime, it always allows shortened months, one-digit days, etc.,
3535 but due to the multitude of formats in which time can be
3536 represented, an ideal HTTP time parser would be even more
3537 forgiving. It should completely ignore things like week days and
3538 concentrate only on the various forms of representing years,
3539 months, days, hours, minutes, and seconds. For example, it would
3540 be nice if it accepted ISO 8601 out of the box.
3542 I've investigated free and PD code for this purpose, but none was
3543 usable. getdate was big and unwieldy, and had potential copyright
3544 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3545 distributed with phttpd, is excellent, but we cannot use it because
3546 it is not assigned to the FSF. So I stuck it with strptime. */
3549 http_atotm (const char *time_string)
3551 /* NOTE: Solaris strptime man page claims that %n and %t match white
3552 space, but that's not universally available. Instead, we simply
3553 use ` ' to mean "skip all WS", which works under all strptime
3554 implementations I've tested. */
3556 static const char *time_formats[] = {
3557 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3558 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3559 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3560 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3561 (used in Set-Cookie, defined in the
3562 Netscape cookie specification.) */
3564 const char *oldlocale;
3565 char savedlocale[256];
3567 time_t ret = (time_t) -1;
3569 /* Solaris strptime fails to recognize English month names in
3570 non-English locales, which we work around by temporarily setting
3571 locale to C before invoking strptime. */
3572 oldlocale = setlocale (LC_TIME, NULL);
3575 size_t l = strlen (oldlocale) + 1;
3576 if (l >= sizeof savedlocale)
3577 savedlocale[0] = '\0';
3579 memcpy (savedlocale, oldlocale, l);
3581 else savedlocale[0] = '\0';
3583 setlocale (LC_TIME, "C");
3585 for (i = 0; i < countof (time_formats); i++)
3589 /* Some versions of strptime use the existing contents of struct
3590 tm to recalculate the date according to format. Zero it out
3591 to prevent stack garbage from influencing strptime. */
3594 if (check_end (strptime (time_string, time_formats[i], &t)))
3601 /* Restore the previous locale. */
3603 setlocale (LC_TIME, savedlocale);
3608 /* Authorization support: We support three authorization schemes:
3610 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3612 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3613 consisting of answering to the server's challenge with the proper
3616 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3617 Stenberg for libcurl. Like digest, NTLM is based on a
3618 challenge-response mechanism, but unlike digest, it is non-standard
3619 (authenticates TCP connections rather than requests), undocumented
3620 and Microsoft-specific. */
3622 /* Create the authentication header contents for the `Basic' scheme.
3623 This is done by encoding the string "USER:PASS" to base64 and
3624 prepending the string "Basic " in front of it. */
3627 basic_authentication_encode (const char *user, const char *passwd)
3630 int len1 = strlen (user) + 1 + strlen (passwd);
3632 t1 = (char *)alloca (len1 + 1);
3633 sprintf (t1, "%s:%s", user, passwd);
3635 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3636 base64_encode (t1, len1, t2);
3638 return concat_strings ("Basic ", t2, (char *) 0);
3641 #define SKIP_WS(x) do { \
3642 while (c_isspace (*(x))) \
3646 #ifdef ENABLE_DIGEST
3647 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3648 an array of 16 bytes containing the hash keys, and BUF should be a
3649 buffer of 33 writable characters (32 for hex digits plus one for
3650 zero termination). */
3652 dump_hash (char *buf, const unsigned char *hash)
3656 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3658 *buf++ = XNUM_TO_digit (*hash >> 4);
3659 *buf++ = XNUM_TO_digit (*hash & 0xf);
3664 /* Take the line apart to find the challenge, and compose a digest
3665 authorization header. See RFC2069 section 2.1.2. */
3667 digest_authentication_encode (const char *au, const char *user,
3668 const char *passwd, const char *method,
3671 static char *realm, *opaque, *nonce, *qop, *algorithm;
3676 { "realm", &realm },
3677 { "opaque", &opaque },
3678 { "nonce", &nonce },
3680 { "algorithm", &algorithm }
3682 char cnonce[16] = "";
3686 param_token name, value;
3689 realm = opaque = nonce = qop = algorithm = NULL;
3691 au += 6; /* skip over `Digest' */
3692 while (extract_param (&au, &name, &value, ','))
3695 size_t namelen = name.e - name.b;
3696 for (i = 0; i < countof (options); i++)
3697 if (namelen == strlen (options[i].name)
3698 && 0 == strncmp (name.b, options[i].name,
3701 *options[i].variable = strdupdelim (value.b, value.e);
3706 if (qop != NULL && strcmp(qop,"auth"))
3708 logprintf (LOG_NOTQUIET, _("Unsupported quality of protection '%s'.\n"), qop);
3709 user = NULL; /* force freeing mem and return */
3712 if (algorithm != NULL && strcmp (algorithm,"MD5") && strcmp (algorithm,"MD5-sess"))
3714 logprintf (LOG_NOTQUIET, _("Unsupported algorithm '%s'.\n"), algorithm);
3715 user = NULL; /* force freeing mem and return */
3718 if (!realm || !nonce || !user || !passwd || !path || !method)
3721 xfree_null (opaque);
3724 xfree_null (algorithm);
3728 /* Calculate the digest value. */
3731 unsigned char hash[MD5_DIGEST_SIZE];
3732 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3733 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3735 /* A1BUF = H(user ":" realm ":" password) */
3736 md5_init_ctx (&ctx);
3737 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3738 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3739 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3740 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3741 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3742 md5_finish_ctx (&ctx, hash);
3744 dump_hash (a1buf, hash);
3746 if (! strcmp (algorithm, "MD5-sess"))
3748 /* A1BUF = H( H(user ":" realm ":" password) ":" nonce ":" cnonce ) */
3749 snprintf (cnonce, sizeof (cnonce), "%08x", random_number(INT_MAX));
3751 md5_init_ctx (&ctx);
3752 // md5_process_bytes (hash, MD5_DIGEST_SIZE, &ctx);
3753 md5_process_bytes (a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3754 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3755 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3756 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3757 md5_process_bytes ((unsigned char *)cnonce, strlen (cnonce), &ctx);
3758 md5_finish_ctx (&ctx, hash);
3760 dump_hash (a1buf, hash);
3763 /* A2BUF = H(method ":" path) */
3764 md5_init_ctx (&ctx);
3765 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3766 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3767 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3768 md5_finish_ctx (&ctx, hash);
3769 dump_hash (a2buf, hash);
3771 if (!strcmp(qop, "auth") || !strcmp (qop, "auth-int"))
3773 /* RFC 2617 Digest Access Authentication */
3774 /* generate random hex string */
3776 snprintf(cnonce, sizeof(cnonce), "%08x", random_number(INT_MAX));
3778 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" noncecount ":" clientnonce ":" qop ": " A2BUF) */
3779 md5_init_ctx (&ctx);
3780 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3781 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3782 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3783 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3784 md5_process_bytes ((unsigned char *)"00000001", 8, &ctx); /* TODO: keep track of server nonce values */
3785 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3786 md5_process_bytes ((unsigned char *)cnonce, strlen(cnonce), &ctx);
3787 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3788 md5_process_bytes ((unsigned char *)qop, strlen(qop), &ctx);
3789 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3790 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3791 md5_finish_ctx (&ctx, hash);
3795 /* RFC 2069 Digest Access Authentication */
3796 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3797 md5_init_ctx (&ctx);
3798 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3799 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3800 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3801 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3802 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3803 md5_finish_ctx (&ctx, hash);
3806 dump_hash (response_digest, hash);
3808 res_size = strlen (user)
3812 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3813 + (opaque ? strlen (opaque) : 0)
3814 + (algorithm ? strlen (algorithm) : 0)
3819 res = xmalloc (res_size);
3821 if (!strcmp(qop,"auth"))
3823 res_len = snprintf (res, res_size, "Digest "\
3824 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\""\
3825 ", qop=auth, nc=00000001, cnonce=\"%s\"",
3826 user, realm, nonce, path, response_digest, cnonce);
3831 res_len = snprintf (res, res_size, "Digest "\
3832 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3833 user, realm, nonce, path, response_digest);
3838 res_len += snprintf(res + res_len, res_size - res_len, ", opaque=\"%s\"", opaque);
3843 snprintf(res + res_len, res_size - res_len, ", algorithm=\"%s\"", algorithm);
3848 #endif /* ENABLE_DIGEST */
3850 /* Computing the size of a string literal must take into account that
3851 value returned by sizeof includes the terminating \0. */
3852 #define STRSIZE(literal) (sizeof (literal) - 1)
3854 /* Whether chars in [b, e) begin with the literal string provided as
3855 first argument and are followed by whitespace or terminating \0.
3856 The comparison is case-insensitive. */
3857 #define STARTS(literal, b, e) \
3859 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3860 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3861 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3862 || c_isspace (b[STRSIZE (literal)])))
3865 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3867 return STARTS ("Basic", hdrbeg, hdrend)
3868 #ifdef ENABLE_DIGEST
3869 || STARTS ("Digest", hdrbeg, hdrend)
3872 || STARTS ("NTLM", hdrbeg, hdrend)
3879 /* Create the HTTP authorization request header. When the
3880 `WWW-Authenticate' response header is seen, according to the
3881 authorization scheme specified in that header (`Basic' and `Digest'
3882 are supported by the current implementation), produce an
3883 appropriate HTTP authorization request header. */
3885 create_authorization_line (const char *au, const char *user,
3886 const char *passwd, const char *method,
3887 const char *path, bool *finished)
3889 /* We are called only with known schemes, so we can dispatch on the
3891 switch (c_toupper (*au))
3893 case 'B': /* Basic */
3895 return basic_authentication_encode (user, passwd);
3896 #ifdef ENABLE_DIGEST
3897 case 'D': /* Digest */
3899 return digest_authentication_encode (au, user, passwd, method, path);
3902 case 'N': /* NTLM */
3903 if (!ntlm_input (&pconn.ntlm, au))
3908 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3911 /* We shouldn't get here -- this function should be only called
3912 with values approved by known_authentication_scheme_p. */
3920 if (!wget_cookie_jar)
3921 wget_cookie_jar = cookie_jar_new ();
3922 if (opt.cookies_input && !cookies_loaded_p)
3924 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3925 cookies_loaded_p = true;
3932 if (wget_cookie_jar)
3933 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3939 xfree_null (pconn.host);
3940 if (wget_cookie_jar)
3941 cookie_jar_delete (wget_cookie_jar);
3945 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3947 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3949 int len = strlen (ext);
3952 strncpy (shortext, ext, len - 1);
3953 shortext[len - 1] = '\0';
3956 if (last_period_in_local_filename == NULL
3957 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3958 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3960 int local_filename_len = strlen (hs->local_file);
3961 /* Resize the local file, allowing for ".html" preceded by
3962 optional ".NUMBER". */
3963 hs->local_file = xrealloc (hs->local_file,
3964 local_filename_len + 24 + len);
3965 strcpy (hs->local_file + local_filename_len, ext);
3966 /* If clobbering is not allowed and the file, as named,
3967 exists, tack on ".NUMBER.html" instead. */
3968 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3972 sprintf (hs->local_file + local_filename_len,
3973 ".%d%s", ext_num++, ext);
3974 while (file_exists_p (hs->local_file));
3976 *dt |= ADDED_HTML_EXTENSION;
3984 test_parse_content_disposition()
3992 { "filename=\"file.ext\"", "file.ext", true },
3993 { "attachment; filename=\"file.ext\"", "file.ext", true },
3994 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
3995 { "attachment", NULL, false },
3996 { "attachement; filename*=UTF-8'en-US'hello.txt", "hello.txt", true },
3997 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", "helloworld.txt", true },
4000 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
4005 res = parse_content_disposition (test_array[i].hdrval, &filename);
4007 mu_assert ("test_parse_content_disposition: wrong result",
4008 res == test_array[i].result
4010 || 0 == strcmp (test_array[i].filename, filename)));
4016 #endif /* TESTING */
4019 * vim: et sts=2 sw=2 cino+={s