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 *, uerr_t *);
78 static char *basic_authentication_encode (const char *, const char *);
79 static bool known_authentication_scheme_p (const char *, const char *);
80 static void ensure_extension (struct http_stat *, const char *, int *);
81 static void load_cookies (void);
84 # define MIN(x, y) ((x) > (y) ? (y) : (x))
88 static bool cookies_loaded_p;
89 static struct cookie_jar *wget_cookie_jar;
91 #define TEXTHTML_S "text/html"
92 #define TEXTXHTML_S "application/xhtml+xml"
93 #define TEXTCSS_S "text/css"
95 /* Some status code validation macros: */
96 #define H_10X(x) (((x) >= 100) && ((x) < 200))
97 #define H_20X(x) (((x) >= 200) && ((x) < 300))
98 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
99 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
100 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
101 || (x) == HTTP_STATUS_SEE_OTHER \
102 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
104 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
105 /* Successful 2xx. */
106 #define HTTP_STATUS_OK 200
107 #define HTTP_STATUS_CREATED 201
108 #define HTTP_STATUS_ACCEPTED 202
109 #define HTTP_STATUS_NO_CONTENT 204
110 #define HTTP_STATUS_PARTIAL_CONTENTS 206
112 /* Redirection 3xx. */
113 #define HTTP_STATUS_MULTIPLE_CHOICES 300
114 #define HTTP_STATUS_MOVED_PERMANENTLY 301
115 #define HTTP_STATUS_MOVED_TEMPORARILY 302
116 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
117 #define HTTP_STATUS_NOT_MODIFIED 304
118 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
120 /* Client error 4xx. */
121 #define HTTP_STATUS_BAD_REQUEST 400
122 #define HTTP_STATUS_UNAUTHORIZED 401
123 #define HTTP_STATUS_FORBIDDEN 403
124 #define HTTP_STATUS_NOT_FOUND 404
125 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
127 /* Server errors 5xx. */
128 #define HTTP_STATUS_INTERNAL 500
129 #define HTTP_STATUS_NOT_IMPLEMENTED 501
130 #define HTTP_STATUS_BAD_GATEWAY 502
131 #define HTTP_STATUS_UNAVAILABLE 503
134 rel_none, rel_name, rel_value, rel_both
141 struct request_header {
143 enum rp release_policy;
145 int hcount, hcapacity;
150 /* Create a new, empty request. Set the request's method and its
151 arguments. METHOD should be a literal string (or it should outlive
152 the request) because it will not be freed. ARG will be freed by
155 static struct request *
156 request_new (const char *method, char *arg)
158 struct request *req = xnew0 (struct request);
160 req->headers = xnew_array (struct request_header, req->hcapacity);
161 req->method = method;
166 /* Return the method string passed with the last call to
167 request_set_method. */
170 request_method (const struct request *req)
175 /* Free one header according to the release policy specified with
176 request_set_header. */
179 release_header (struct request_header *hdr)
181 switch (hdr->release_policy)
198 /* Set the request named NAME to VALUE. Specifically, this means that
199 a "NAME: VALUE\r\n" header line will be used in the request. If a
200 header with the same name previously existed in the request, its
201 value will be replaced by this one. A NULL value means do nothing.
203 RELEASE_POLICY determines whether NAME and VALUE should be released
204 (freed) with request_free. Allowed values are:
206 - rel_none - don't free NAME or VALUE
207 - rel_name - free NAME when done
208 - rel_value - free VALUE when done
209 - rel_both - free both NAME and VALUE when done
211 Setting release policy is useful when arguments come from different
212 sources. For example:
214 // Don't free literal strings!
215 request_set_header (req, "Pragma", "no-cache", rel_none);
217 // Don't free a global variable, we'll need it later.
218 request_set_header (req, "Referer", opt.referer, rel_none);
220 // Value freshly allocated, free it when done.
221 request_set_header (req, "Range",
222 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
227 request_set_header (struct request *req, const char *name, const char *value,
228 enum rp release_policy)
230 struct request_header *hdr;
235 /* A NULL value is a no-op; if freeing the name is requested,
236 free it now to avoid leaks. */
237 if (release_policy == rel_name || release_policy == rel_both)
238 xfree ((void *)name);
242 for (i = 0; i < req->hcount; i++)
244 hdr = &req->headers[i];
245 if (0 == strcasecmp (name, hdr->name))
247 /* Replace existing header. */
248 release_header (hdr);
249 hdr->name = (void *)name;
250 hdr->value = (void *)value;
251 hdr->release_policy = release_policy;
256 /* Install new header. */
258 if (req->hcount >= req->hcapacity)
260 req->hcapacity <<= 1;
261 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
263 hdr = &req->headers[req->hcount++];
264 hdr->name = (void *)name;
265 hdr->value = (void *)value;
266 hdr->release_policy = release_policy;
269 /* Like request_set_header, but sets the whole header line, as
270 provided by the user using the `--header' option. For example,
271 request_set_user_header (req, "Foo: bar") works just like
272 request_set_header (req, "Foo", "bar"). */
275 request_set_user_header (struct request *req, const char *header)
278 const char *p = strchr (header, ':');
281 BOUNDED_TO_ALLOCA (header, p, name);
283 while (c_isspace (*p))
285 request_set_header (req, xstrdup (name), (char *) p, rel_name);
288 /* Remove the header with specified name from REQ. Returns true if
289 the header was actually removed, false otherwise. */
292 request_remove_header (struct request *req, const char *name)
295 for (i = 0; i < req->hcount; i++)
297 struct request_header *hdr = &req->headers[i];
298 if (0 == strcasecmp (name, hdr->name))
300 release_header (hdr);
301 /* Move the remaining headers by one. */
302 if (i < req->hcount - 1)
303 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
311 #define APPEND(p, str) do { \
312 int A_len = strlen (str); \
313 memcpy (p, str, A_len); \
317 /* Construct the request and write it to FD using fd_write.
318 If warc_tmp is set to a file pointer, the request string will
319 also be written to that file. */
322 request_send (const struct request *req, int fd, FILE *warc_tmp)
324 char *request_string, *p;
325 int i, size, write_error;
327 /* Count the request size. */
330 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
331 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
333 for (i = 0; i < req->hcount; i++)
335 struct request_header *hdr = &req->headers[i];
336 /* NAME ": " VALUE "\r\n" */
337 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
343 p = request_string = alloca_array (char, size);
345 /* Generate the request. */
347 APPEND (p, req->method); *p++ = ' ';
348 APPEND (p, req->arg); *p++ = ' ';
349 memcpy (p, "HTTP/1.1\r\n", 10); p += 10;
351 for (i = 0; i < req->hcount; i++)
353 struct request_header *hdr = &req->headers[i];
354 APPEND (p, hdr->name);
355 *p++ = ':', *p++ = ' ';
356 APPEND (p, hdr->value);
357 *p++ = '\r', *p++ = '\n';
360 *p++ = '\r', *p++ = '\n', *p++ = '\0';
361 assert (p - request_string == size);
365 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
367 /* Send the request to the server. */
369 write_error = fd_write (fd, request_string, size - 1, -1);
371 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
373 else if (warc_tmp != NULL)
375 /* Write a copy of the data to the WARC record. */
376 int warc_tmp_written = fwrite (request_string, 1, size - 1, warc_tmp);
377 if (warc_tmp_written != size - 1)
383 /* Release the resources used by REQ. */
386 request_free (struct request *req)
389 xfree_null (req->arg);
390 for (i = 0; i < req->hcount; i++)
391 release_header (&req->headers[i]);
392 xfree_null (req->headers);
396 static struct hash_table *basic_authed_hosts;
398 /* Find out if this host has issued a Basic challenge yet; if so, give
399 * it the username, password. A temporary measure until we can get
400 * proper authentication in place. */
403 maybe_send_basic_creds (const char *hostname, const char *user,
404 const char *passwd, struct request *req)
406 bool do_challenge = false;
408 if (opt.auth_without_challenge)
410 DEBUGP (("Auth-without-challenge set, sending Basic credentials.\n"));
413 else if (basic_authed_hosts
414 && hash_table_contains(basic_authed_hosts, hostname))
416 DEBUGP (("Found %s in basic_authed_hosts.\n", quote (hostname)));
421 DEBUGP (("Host %s has not issued a general basic challenge.\n",
426 request_set_header (req, "Authorization",
427 basic_authentication_encode (user, passwd),
434 register_basic_auth_host (const char *hostname)
436 if (!basic_authed_hosts)
438 basic_authed_hosts = make_nocase_string_hash_table (1);
440 if (!hash_table_contains(basic_authed_hosts, hostname))
442 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
443 DEBUGP (("Inserted %s into basic_authed_hosts\n", quote (hostname)));
448 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
449 PROMISED_SIZE bytes are sent over the wire -- if the file is
450 longer, read only that much; if the file is shorter, report an error.
451 If warc_tmp is set to a file pointer, the post data will
452 also be written to that file. */
455 body_file_send (int sock, const char *file_name, wgint promised_size, FILE *warc_tmp)
457 static char chunk[8192];
462 DEBUGP (("[writing BODY file %s ... ", file_name));
464 fp = fopen (file_name, "rb");
467 while (!feof (fp) && written < promised_size)
470 int length = fread (chunk, 1, sizeof (chunk), fp);
473 towrite = MIN (promised_size - written, length);
474 write_error = fd_write (sock, chunk, towrite, -1);
480 if (warc_tmp != NULL)
482 /* Write a copy of the data to the WARC record. */
483 int warc_tmp_written = fwrite (chunk, 1, towrite, warc_tmp);
484 if (warc_tmp_written != towrite)
494 /* If we've written less than was promised, report a (probably
495 nonsensical) error rather than break the promise. */
496 if (written < promised_size)
502 assert (written == promised_size);
503 DEBUGP (("done]\n"));
507 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
508 If so, return the pointer to the position after the line, otherwise
509 return NULL. This is used as callback to fd_read_hunk. The data
510 between START and PEEKED has been read and cannot be "unread"; the
511 data after PEEKED has only been peeked. */
514 response_head_terminator (const char *start, const char *peeked, int peeklen)
518 /* If at first peek, verify whether HUNK starts with "HTTP". If
519 not, this is a HTTP/0.9 request and we must bail out without
521 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
524 /* Look for "\n[\r]\n", and return the following position if found.
525 Start two chars before the current to cover the possibility that
526 part of the terminator (e.g. "\n\r") arrived in the previous
528 p = peeked - start < 2 ? start : peeked - 2;
529 end = peeked + peeklen;
531 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
532 for (; p < end - 2; p++)
535 if (p[1] == '\r' && p[2] == '\n')
537 else if (p[1] == '\n')
540 /* p==end-2: check for \n\n directly preceding END. */
541 if (p[0] == '\n' && p[1] == '\n')
547 /* The maximum size of a single HTTP response we care to read. Rather
548 than being a limit of the reader implementation, this limit
549 prevents Wget from slurping all available memory upon encountering
550 malicious or buggy server output, thus protecting the user. Define
551 it to 0 to remove the limit. */
553 #define HTTP_RESPONSE_MAX_SIZE 65536
555 /* Read the HTTP request head from FD and return it. The error
556 conditions are the same as with fd_read_hunk.
558 To support HTTP/0.9 responses, this function tries to make sure
559 that the data begins with "HTTP". If this is not the case, no data
560 is read and an empty request is returned, so that the remaining
561 data can be treated as body. */
564 read_http_response_head (int fd)
566 return fd_read_hunk (fd, response_head_terminator, 512,
567 HTTP_RESPONSE_MAX_SIZE);
571 /* The response data. */
574 /* The array of pointers that indicate where each header starts.
575 For example, given this HTTP response:
582 The headers are located like this:
584 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
586 headers[0] headers[1] headers[2] headers[3]
588 I.e. headers[0] points to the beginning of the request,
589 headers[1] points to the end of the first header and the
590 beginning of the second one, etc. */
592 const char **headers;
595 /* Create a new response object from the text of the HTTP response,
596 available in HEAD. That text is automatically split into
597 constituent header lines for fast retrieval using
600 static struct response *
601 resp_new (const char *head)
606 struct response *resp = xnew0 (struct response);
611 /* Empty head means that we're dealing with a headerless
612 (HTTP/0.9) response. In that case, don't set HEADERS at
617 /* Split HEAD into header lines, so that resp_header_* functions
618 don't need to do this over and over again. */
624 DO_REALLOC (resp->headers, size, count + 1, const char *);
625 resp->headers[count++] = hdr;
627 /* Break upon encountering an empty line. */
628 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
631 /* Find the end of HDR, including continuations. */
634 const char *end = strchr (hdr, '\n');
640 while (*hdr == ' ' || *hdr == '\t');
642 DO_REALLOC (resp->headers, size, count + 1, const char *);
643 resp->headers[count] = NULL;
648 /* Locate the header named NAME in the request data, starting with
649 position START. This allows the code to loop through the request
650 data, filtering for all requests of a given name. Returns the
651 found position, or -1 for failure. The code that uses this
652 function typically looks like this:
654 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
655 ... do something with header ...
657 If you only care about one header, use resp_header_get instead of
661 resp_header_locate (const struct response *resp, const char *name, int start,
662 const char **begptr, const char **endptr)
665 const char **headers = resp->headers;
668 if (!headers || !headers[1])
671 name_len = strlen (name);
677 for (; headers[i + 1]; i++)
679 const char *b = headers[i];
680 const char *e = headers[i + 1];
682 && b[name_len] == ':'
683 && 0 == strncasecmp (b, name, name_len))
686 while (b < e && c_isspace (*b))
688 while (b < e && c_isspace (e[-1]))
698 /* Find and retrieve the header named NAME in the request data. If
699 found, set *BEGPTR to its starting, and *ENDPTR to its ending
700 position, and return true. Otherwise return false.
702 This function is used as a building block for resp_header_copy
703 and resp_header_strdup. */
706 resp_header_get (const struct response *resp, const char *name,
707 const char **begptr, const char **endptr)
709 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
713 /* Copy the response header named NAME to buffer BUF, no longer than
714 BUFSIZE (BUFSIZE includes the terminating 0). If the header
715 exists, true is returned, false otherwise. If there should be no
716 limit on the size of the header, use resp_header_strdup instead.
718 If BUFSIZE is 0, no data is copied, but the boolean indication of
719 whether the header is present is still returned. */
722 resp_header_copy (const struct response *resp, const char *name,
723 char *buf, int bufsize)
726 if (!resp_header_get (resp, name, &b, &e))
730 int len = MIN (e - b, bufsize - 1);
731 memcpy (buf, b, len);
737 /* Return the value of header named NAME in RESP, allocated with
738 malloc. If such a header does not exist in RESP, return NULL. */
741 resp_header_strdup (const struct response *resp, const char *name)
744 if (!resp_header_get (resp, name, &b, &e))
746 return strdupdelim (b, e);
749 /* Parse the HTTP status line, which is of format:
751 HTTP-Version SP Status-Code SP Reason-Phrase
753 The function returns the status-code, or -1 if the status line
754 appears malformed. The pointer to "reason-phrase" message is
755 returned in *MESSAGE. */
758 resp_status (const struct response *resp, char **message)
765 /* For a HTTP/0.9 response, assume status 200. */
767 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
771 p = resp->headers[0];
772 end = resp->headers[1];
778 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
782 /* Match the HTTP version. This is optional because Gnutella
783 servers have been reported to not specify HTTP version. */
784 if (p < end && *p == '/')
787 while (p < end && c_isdigit (*p))
789 if (p < end && *p == '.')
791 while (p < end && c_isdigit (*p))
795 while (p < end && c_isspace (*p))
797 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
800 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
805 while (p < end && c_isspace (*p))
807 while (p < end && c_isspace (end[-1]))
809 *message = strdupdelim (p, end);
815 /* Release the resources used by RESP. */
818 resp_free (struct response *resp)
820 xfree_null (resp->headers);
824 /* Print a single line of response, the characters [b, e). We tried
826 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
827 but that failed to escape the non-printable characters and, in fact,
828 caused crashes in UTF-8 locales. */
831 print_response_line(const char *prefix, const char *b, const char *e)
834 BOUNDED_TO_ALLOCA(b, e, copy);
835 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
836 quotearg_style (escape_quoting_style, copy));
839 /* Print the server response, line by line, omitting the trailing CRLF
840 from individual header lines, and prefixed with PREFIX. */
843 print_server_response (const struct response *resp, const char *prefix)
848 for (i = 0; resp->headers[i + 1]; i++)
850 const char *b = resp->headers[i];
851 const char *e = resp->headers[i + 1];
853 if (b < e && e[-1] == '\n')
855 if (b < e && e[-1] == '\r')
857 print_response_line(prefix, b, e);
861 /* Parse the `Content-Range' header and extract the information it
862 contains. Returns true if successful, false otherwise. */
864 parse_content_range (const char *hdr, wgint *first_byte_ptr,
865 wgint *last_byte_ptr, wgint *entity_length_ptr)
869 /* Ancient versions of Netscape proxy server, presumably predating
870 rfc2068, sent out `Content-Range' without the "bytes"
872 if (0 == strncasecmp (hdr, "bytes", 5))
875 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
879 while (c_isspace (*hdr))
884 if (!c_isdigit (*hdr))
886 for (num = 0; c_isdigit (*hdr); hdr++)
887 num = 10 * num + (*hdr - '0');
888 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
890 *first_byte_ptr = num;
892 for (num = 0; c_isdigit (*hdr); hdr++)
893 num = 10 * num + (*hdr - '0');
894 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
896 *last_byte_ptr = num;
901 for (num = 0; c_isdigit (*hdr); hdr++)
902 num = 10 * num + (*hdr - '0');
903 *entity_length_ptr = num;
907 /* Read the body of the request, but don't store it anywhere and don't
908 display a progress gauge. This is useful for reading the bodies of
909 administrative responses to which we will soon issue another
910 request. The response is not useful to the user, but reading it
911 allows us to continue using the same connection to the server.
913 If reading fails, false is returned, true otherwise. In debug
914 mode, the body is displayed for debugging purposes. */
917 skip_short_body (int fd, wgint contlen, bool chunked)
920 SKIP_SIZE = 512, /* size of the download buffer */
921 SKIP_THRESHOLD = 4096 /* the largest size we read */
923 wgint remaining_chunk_size = 0;
924 char dlbuf[SKIP_SIZE + 1];
925 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
927 /* If the body is too large, it makes more sense to simply close the
928 connection than to try to read the body. */
929 if (contlen > SKIP_THRESHOLD)
932 while (contlen > 0 || chunked)
937 if (remaining_chunk_size == 0)
939 char *line = fd_read_line (fd);
944 remaining_chunk_size = strtol (line, &endl, 16);
947 if (remaining_chunk_size == 0)
949 line = fd_read_line (fd);
955 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
958 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
960 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
963 /* Don't normally report the error since this is an
964 optimization that should be invisible to the user. */
965 DEBUGP (("] aborting (%s).\n",
966 ret < 0 ? fd_errstr (fd) : "EOF received"));
973 remaining_chunk_size -= ret;
974 if (remaining_chunk_size == 0)
976 char *line = fd_read_line (fd);
984 /* Safe even if %.*s bogusly expects terminating \0 because
985 we've zero-terminated dlbuf above. */
986 DEBUGP (("%.*s", ret, dlbuf));
989 DEBUGP (("] done.\n"));
993 #define NOT_RFC2231 0
994 #define RFC2231_NOENCODING 1
995 #define RFC2231_ENCODING 2
997 /* extract_param extracts the parameter name into NAME.
998 However, if the parameter name is in RFC2231 format then
999 this function adjusts NAME by stripping of the trailing
1000 characters that are not part of the name but are present to
1001 indicate the presence of encoding information in the value
1002 or a fragment of a long parameter value
1005 modify_param_name(param_token *name)
1007 const char *delim1 = memchr (name->b, '*', name->e - name->b);
1008 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
1014 result = NOT_RFC2231;
1016 else if(delim1 == delim2)
1018 if ((name->e - 1) == delim1)
1020 result = RFC2231_ENCODING;
1024 result = RFC2231_NOENCODING;
1031 result = RFC2231_ENCODING;
1036 /* extract_param extract the paramater value into VALUE.
1037 Like modify_param_name this function modifies VALUE by
1038 stripping off the encoding information from the actual value
1041 modify_param_value (param_token *value, int encoding_type )
1043 if (RFC2231_ENCODING == encoding_type)
1045 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1046 if ( delim != NULL )
1048 value->b = (delim+1);
1053 /* Extract a parameter from the string (typically an HTTP header) at
1054 **SOURCE and advance SOURCE to the next parameter. Return false
1055 when there are no more parameters to extract. The name of the
1056 parameter is returned in NAME, and the value in VALUE. If the
1057 parameter has no value, the token's value is zeroed out.
1059 For example, if *SOURCE points to the string "attachment;
1060 filename=\"foo bar\"", the first call to this function will return
1061 the token named "attachment" and no value, and the second call will
1062 return the token named "filename" and value "foo bar". The third
1063 call will return false, indicating no more valid tokens. */
1066 extract_param (const char **source, param_token *name, param_token *value,
1069 const char *p = *source;
1071 while (c_isspace (*p)) ++p;
1075 return false; /* no error; nothing more to extract */
1080 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1082 if (name->b == name->e)
1083 return false; /* empty name: error */
1084 while (c_isspace (*p)) ++p;
1085 if (*p == separator || !*p) /* no value */
1088 if (*p == separator) ++p;
1093 return false; /* error */
1095 /* *p is '=', extract value */
1097 while (c_isspace (*p)) ++p;
1098 if (*p == '"') /* quoted */
1101 while (*p && *p != '"') ++p;
1105 /* Currently at closing quote; find the end of param. */
1106 while (c_isspace (*p)) ++p;
1107 while (*p && *p != separator) ++p;
1108 if (*p == separator)
1111 /* garbage after closed quote, e.g. foo="bar"baz */
1117 while (*p && *p != separator) ++p;
1119 while (value->e != value->b && c_isspace (value->e[-1]))
1121 if (*p == separator) ++p;
1125 int param_type = modify_param_name(name);
1126 if (NOT_RFC2231 != param_type)
1128 modify_param_value(value, param_type);
1134 #undef RFC2231_NOENCODING
1135 #undef RFC2231_ENCODING
1137 /* Appends the string represented by VALUE to FILENAME */
1140 append_value_to_filename (char **filename, param_token const * const value)
1142 int original_length = strlen(*filename);
1143 int new_length = strlen(*filename) + (value->e - value->b);
1144 *filename = xrealloc (*filename, new_length+1);
1145 memcpy (*filename + original_length, value->b, (value->e - value->b));
1146 (*filename)[new_length] = '\0';
1150 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1152 /* Parse the contents of the `Content-Disposition' header, extracting
1153 the information useful to Wget. Content-Disposition is a header
1154 borrowed from MIME; when used in HTTP, it typically serves for
1155 specifying the desired file name of the resource. For example:
1157 Content-Disposition: attachment; filename="flora.jpg"
1159 Wget will skip the tokens it doesn't care about, such as
1160 "attachment" in the previous example; it will also skip other
1161 unrecognized params. If the header is syntactically correct and
1162 contains a file name, a copy of the file name is stored in
1163 *filename and true is returned. Otherwise, the function returns
1166 The file name is stripped of directory components and must not be
1169 Historically, this function returned filename prefixed with opt.dir_prefix,
1170 now that logic is handled by the caller, new code should pay attention,
1171 changed by crq, Sep 2010.
1175 parse_content_disposition (const char *hdr, char **filename)
1177 param_token name, value;
1179 while (extract_param (&hdr, &name, &value, ';'))
1181 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1182 if ( isFilename && value.b != NULL)
1184 /* Make the file name begin at the last slash or backslash. */
1185 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1186 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1187 if (last_slash && last_bs)
1188 value.b = 1 + MAX (last_slash, last_bs);
1189 else if (last_slash || last_bs)
1190 value.b = 1 + (last_slash ? last_slash : last_bs);
1191 if (value.b == value.e)
1195 append_value_to_filename (filename, &value);
1197 *filename = strdupdelim (value.b, value.e);
1208 /* Persistent connections. Currently, we cache the most recently used
1209 connection as persistent, provided that the HTTP server agrees to
1210 make it such. The persistence data is stored in the variables
1211 below. Ideally, it should be possible to cache an arbitrary fixed
1212 number of these connections. */
1214 /* Whether a persistent connection is active. */
1215 static bool pconn_active;
1218 /* The socket of the connection. */
1221 /* Host and port of the currently active persistent connection. */
1225 /* Whether a ssl handshake has occoured on this connection. */
1228 /* Whether the connection was authorized. This is only done by
1229 NTLM, which authorizes *connections* rather than individual
1230 requests. (That practice is peculiar for HTTP, but it is a
1231 useful optimization.) */
1235 /* NTLM data of the current connection. */
1236 struct ntlmdata ntlm;
1240 /* Mark the persistent connection as invalid and free the resources it
1241 uses. This is used by the CLOSE_* macros after they forcefully
1242 close a registered persistent connection. */
1245 invalidate_persistent (void)
1247 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1248 pconn_active = false;
1249 fd_close (pconn.socket);
1254 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1255 persistent. This will enable someone to use the same connection
1256 later. In the context of HTTP, this must be called only AFTER the
1257 response has been received and the server has promised that the
1258 connection will remain alive.
1260 If a previous connection was persistent, it is closed. */
1263 register_persistent (const char *host, int port, int fd, bool ssl)
1267 if (pconn.socket == fd)
1269 /* The connection FD is already registered. */
1274 /* The old persistent connection is still active; close it
1275 first. This situation arises whenever a persistent
1276 connection exists, but we then connect to a different
1277 host, and try to register a persistent connection to that
1279 invalidate_persistent ();
1283 pconn_active = true;
1285 pconn.host = xstrdup (host);
1288 pconn.authorized = false;
1290 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1293 /* Return true if a persistent connection is available for connecting
1297 persistent_available_p (const char *host, int port, bool ssl,
1298 bool *host_lookup_failed)
1300 /* First, check whether a persistent connection is active at all. */
1304 /* If we want SSL and the last connection wasn't or vice versa,
1305 don't use it. Checking for host and port is not enough because
1306 HTTP and HTTPS can apparently coexist on the same port. */
1307 if (ssl != pconn.ssl)
1310 /* If we're not connecting to the same port, we're not interested. */
1311 if (port != pconn.port)
1314 /* If the host is the same, we're in business. If not, there is
1315 still hope -- read below. */
1316 if (0 != strcasecmp (host, pconn.host))
1318 /* Check if pconn.socket is talking to HOST under another name.
1319 This happens often when both sites are virtual hosts
1320 distinguished only by name and served by the same network
1321 interface, and hence the same web server (possibly set up by
1322 the ISP and serving many different web sites). This
1323 admittedly unconventional optimization does not contradict
1324 HTTP and works well with popular server software. */
1328 struct address_list *al;
1331 /* Don't try to talk to two different SSL sites over the same
1332 secure connection! (Besides, it's not clear that
1333 name-based virtual hosting is even possible with SSL.) */
1336 /* If pconn.socket's peer is one of the IP addresses HOST
1337 resolves to, pconn.socket is for all intents and purposes
1338 already talking to HOST. */
1340 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1342 /* Can't get the peer's address -- something must be very
1343 wrong with the connection. */
1344 invalidate_persistent ();
1347 al = lookup_host (host, 0);
1350 *host_lookup_failed = true;
1354 found = address_list_contains (al, &ip);
1355 address_list_release (al);
1360 /* The persistent connection's peer address was found among the
1361 addresses HOST resolved to; therefore, pconn.sock is in fact
1362 already talking to HOST -- no need to reconnect. */
1365 /* Finally, check whether the connection is still open. This is
1366 important because most servers implement liberal (short) timeout
1367 on persistent connections. Wget can of course always reconnect
1368 if the connection doesn't work out, but it's nicer to know in
1369 advance. This test is a logical followup of the first test, but
1370 is "expensive" and therefore placed at the end of the list.
1372 (Current implementation of test_socket_open has a nice side
1373 effect that it treats sockets with pending data as "closed".
1374 This is exactly what we want: if a broken server sends message
1375 body in response to HEAD, or if it sends more than conent-length
1376 data, we won't reuse the corrupted connection.) */
1378 if (!test_socket_open (pconn.socket))
1380 /* Oops, the socket is no longer open. Now that we know that,
1381 let's invalidate the persistent connection before returning
1383 invalidate_persistent ();
1390 /* The idea behind these two CLOSE macros is to distinguish between
1391 two cases: one when the job we've been doing is finished, and we
1392 want to close the connection and leave, and two when something is
1393 seriously wrong and we're closing the connection as part of
1396 In case of keep_alive, CLOSE_FINISH should leave the connection
1397 open, while CLOSE_INVALIDATE should still close it.
1399 Note that the semantics of the flag `keep_alive' is "this
1400 connection *will* be reused (the server has promised not to close
1401 the connection once we're done)", while the semantics of
1402 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1403 active, registered connection". */
1405 #define CLOSE_FINISH(fd) do { \
1408 if (pconn_active && (fd) == pconn.socket) \
1409 invalidate_persistent (); \
1418 #define CLOSE_INVALIDATE(fd) do { \
1419 if (pconn_active && (fd) == pconn.socket) \
1420 invalidate_persistent (); \
1428 wgint len; /* received length */
1429 wgint contlen; /* expected length */
1430 wgint restval; /* the restart value */
1431 int res; /* the result of last read */
1432 char *rderrmsg; /* error message from read error */
1433 char *newloc; /* new location (redirection) */
1434 char *remote_time; /* remote time-stamp string */
1435 char *error; /* textual HTTP error */
1436 int statcode; /* status code */
1437 char *message; /* status message */
1438 wgint rd_size; /* amount of data read from socket */
1439 double dltime; /* time it took to download the data */
1440 const char *referer; /* value of the referer header. */
1441 char *local_file; /* local file name. */
1442 bool existence_checked; /* true if we already checked for a file's
1443 existence after having begun to download
1444 (needed in gethttp for when connection is
1445 interrupted/restarted. */
1446 bool timestamp_checked; /* true if pre-download time-stamping checks
1447 * have already been performed */
1448 char *orig_file_name; /* name of file to compare for time-stamping
1449 * (might be != local_file if -K is set) */
1450 wgint orig_file_size; /* size of file to compare for time-stamping */
1451 time_t orig_file_tstamp; /* time-stamp of file to compare for
1456 free_hstat (struct http_stat *hs)
1458 xfree_null (hs->newloc);
1459 xfree_null (hs->remote_time);
1460 xfree_null (hs->error);
1461 xfree_null (hs->rderrmsg);
1462 xfree_null (hs->local_file);
1463 xfree_null (hs->orig_file_name);
1464 xfree_null (hs->message);
1466 /* Guard against being called twice. */
1468 hs->remote_time = NULL;
1473 get_file_flags (const char *filename, int *dt)
1475 logprintf (LOG_VERBOSE, _("\
1476 File %s already there; not retrieving.\n\n"), quote (filename));
1477 /* If the file is there, we suppose it's retrieved OK. */
1480 /* #### Bogusness alert. */
1481 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1482 if (has_html_suffix_p (filename))
1486 /* Download the response body from the socket and writes it to
1487 an output file. The headers have already been read from the
1488 socket. If WARC is enabled, the response body will also be
1489 written to a WARC response record.
1491 hs, contlen, contrange, chunked_transfer_encoding and url are
1492 parameters from the gethttp method. fp is a pointer to the
1495 url, warc_timestamp_str, warc_request_uuid, warc_ip, type
1496 and statcode will be saved in the headers of the WARC record.
1497 The head parameter contains the HTTP headers of the response.
1499 If fp is NULL and WARC is enabled, the response body will be
1500 written only to the WARC file. If WARC is disabled and fp
1501 is a file pointer, the data will be written to the file.
1502 If fp is a file pointer and WARC is enabled, the body will
1503 be written to both destinations.
1505 Returns the error code. */
1507 read_response_body (struct http_stat *hs, int sock, FILE *fp, wgint contlen,
1508 wgint contrange, bool chunked_transfer_encoding,
1509 char *url, char *warc_timestamp_str, char *warc_request_uuid,
1510 ip_address *warc_ip, char *type, int statcode, char *head)
1512 int warc_payload_offset = 0;
1513 FILE *warc_tmp = NULL;
1516 if (opt.warc_filename != NULL)
1518 /* Open a temporary file where we can write the response before we
1519 add it to the WARC record. */
1520 warc_tmp = warc_tempfile ();
1521 if (warc_tmp == NULL)
1522 warcerr = WARC_TMP_FOPENERR;
1526 /* We should keep the response headers for the WARC record. */
1527 int head_len = strlen (head);
1528 int warc_tmp_written = fwrite (head, 1, head_len, warc_tmp);
1529 if (warc_tmp_written != head_len)
1530 warcerr = WARC_TMP_FWRITEERR;
1531 warc_payload_offset = head_len;
1536 if (warc_tmp != NULL)
1544 /* This confuses the timestamping code that checks for file size.
1545 #### The timestamping code should be smarter about file size. */
1546 if (opt.save_headers && hs->restval == 0)
1547 fwrite (head, 1, strlen (head), fp);
1550 /* Read the response body. */
1553 /* If content-length is present, read that much; otherwise, read
1554 until EOF. The HTTP spec doesn't require the server to
1555 actually close the connection when it's done sending data. */
1556 flags |= rb_read_exactly;
1557 if (fp != NULL && hs->restval > 0 && contrange == 0)
1558 /* If the server ignored our range request, instruct fd_read_body
1559 to skip the first RESTVAL bytes of body. */
1560 flags |= rb_skip_startpos;
1561 if (chunked_transfer_encoding)
1562 flags |= rb_chunked_transfer_encoding;
1564 hs->len = hs->restval;
1566 /* Download the response body and write it to fp.
1567 If we are working on a WARC file, we simultaneously write the
1568 response body to warc_tmp. */
1569 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1570 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1574 if (warc_tmp != NULL)
1576 /* Create a response record and write it to the WARC file.
1577 Note: per the WARC standard, the request and response should share
1578 the same date header. We re-use the timestamp of the request.
1579 The response record should also refer to the uuid of the request. */
1580 bool r = warc_write_response_record (url, warc_timestamp_str,
1581 warc_request_uuid, warc_ip,
1582 warc_tmp, warc_payload_offset,
1583 type, statcode, hs->newloc);
1585 /* warc_write_response_record has closed warc_tmp. */
1591 return RETRFINISHED;
1594 if (warc_tmp != NULL)
1599 /* Error while writing to fd. */
1602 else if (hs->res == -3)
1604 /* Error while writing to warc_tmp. */
1605 return WARC_TMP_FWRITEERR;
1610 hs->rderrmsg = xstrdup (fd_errstr (sock));
1611 return RETRFINISHED;
1615 #define BEGINS_WITH(line, string_constant) \
1616 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1617 && (c_isspace (line[sizeof (string_constant) - 1]) \
1618 || !line[sizeof (string_constant) - 1]))
1621 #define SET_USER_AGENT(req) do { \
1622 if (!opt.useragent) \
1623 request_set_header (req, "User-Agent", \
1624 aprintf ("Wget/%s (VMS %s %s)", \
1625 version_string, vms_arch(), vms_vers()), \
1627 else if (*opt.useragent) \
1628 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1630 #else /* def __VMS */
1631 #define SET_USER_AGENT(req) do { \
1632 if (!opt.useragent) \
1633 request_set_header (req, "User-Agent", \
1634 aprintf ("Wget/%s (%s)", \
1635 version_string, OS_TYPE), \
1637 else if (*opt.useragent) \
1638 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1640 #endif /* def __VMS [else] */
1642 /* Retrieve a document through HTTP protocol. It recognizes status
1643 code, and correctly handles redirections. It closes the network
1644 socket. If it receives an error from the functions below it, it
1645 will print it if there is enough information to do so (almost
1646 always), returning the error to the caller (i.e. http_loop).
1648 Various HTTP parameters are stored to hs.
1650 If PROXY is non-NULL, the connection will be made to the proxy
1651 server, and u->url will be requested. */
1653 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1654 struct iri *iri, int count)
1656 struct request *req;
1659 char *user, *passwd;
1663 wgint contlen, contrange;
1670 /* Set to 1 when the authorization has already been sent and should
1671 not be tried again. */
1672 bool auth_finished = false;
1674 /* Set to 1 when just globally-set Basic authorization has been sent;
1675 * should prevent further Basic negotiations, but not other
1677 bool basic_auth_finished = false;
1679 /* Whether NTLM authentication is used for this request. */
1680 bool ntlm_seen = false;
1682 /* Whether our connection to the remote host is through SSL. */
1683 bool using_ssl = false;
1685 /* Whether a HEAD request will be issued (as opposed to GET or
1687 bool head_only = !!(*dt & HEAD_ONLY);
1690 struct response *resp;
1694 /* Declare WARC variables. */
1695 bool warc_enabled = (opt.warc_filename != NULL);
1696 FILE *warc_tmp = NULL;
1697 char warc_timestamp_str [21];
1698 char warc_request_uuid [48];
1699 ip_address *warc_ip = NULL;
1700 off_t warc_payload_offset = -1;
1702 /* Whether this connection will be kept alive after the HTTP request
1706 /* Is the server using the chunked transfer encoding? */
1707 bool chunked_transfer_encoding = false;
1709 /* Whether keep-alive should be inhibited. */
1710 bool inhibit_keep_alive =
1711 !opt.http_keep_alive || opt.ignore_length;
1713 /* Headers sent when using POST. */
1714 wgint body_data_size = 0;
1716 bool host_lookup_failed = false;
1719 if (u->scheme == SCHEME_HTTPS)
1721 /* Initialize the SSL context. After this has once been done,
1722 it becomes a no-op. */
1725 scheme_disable (SCHEME_HTTPS);
1726 logprintf (LOG_NOTQUIET,
1727 _("Disabling SSL due to encountered errors.\n"));
1728 return SSLINITFAILED;
1731 #endif /* HAVE_SSL */
1733 /* Initialize certain elements of struct http_stat. */
1737 hs->rderrmsg = NULL;
1739 hs->remote_time = NULL;
1745 /* Prepare the request to send. */
1748 const char *meth = "GET";
1751 else if (opt.method)
1753 /* Use the full path, i.e. one that includes the leading slash and
1754 the query string. E.g. if u->path is "foo/bar" and u->query is
1755 "param=value", full_path will be "/foo/bar?param=value". */
1758 /* When using SSL over proxy, CONNECT establishes a direct
1759 connection to the HTTPS server. Therefore use the same
1760 argument as when talking to the server directly. */
1761 && u->scheme != SCHEME_HTTPS
1764 meth_arg = xstrdup (u->url);
1766 meth_arg = url_full_path (u);
1767 req = request_new (meth, meth_arg);
1770 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1771 if (*dt & SEND_NOCACHE)
1773 /* Cache-Control MUST be obeyed by all HTTP/1.1 caching mechanisms... */
1774 request_set_header (req, "Cache-Control", "no-cache, must-revalidate", rel_none);
1776 /* ... but some HTTP/1.0 caches doesn't implement Cache-Control. */
1777 request_set_header (req, "Pragma", "no-cache", rel_none);
1780 request_set_header (req, "Range",
1781 aprintf ("bytes=%s-",
1782 number_to_static_string (hs->restval)),
1784 SET_USER_AGENT (req);
1785 request_set_header (req, "Accept", "*/*", rel_none);
1787 /* Find the username and password for authentication. */
1790 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1791 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1792 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1794 /* We only do "site-wide" authentication with "global" user/password
1795 * values unless --auth-no-challange has been requested; URL user/password
1796 * info overrides. */
1797 if (user && passwd && (!u->user || opt.auth_without_challenge))
1799 /* If this is a host for which we've already received a Basic
1800 * challenge, we'll go ahead and send Basic authentication creds. */
1801 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1804 /* Generate the Host header, HOST:PORT. Take into account that:
1806 - Broken server-side software often doesn't recognize the PORT
1807 argument, so we must generate "Host: www.server.com" instead of
1808 "Host: www.server.com:80" (and likewise for https port).
1810 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1811 becomes ambiguous and needs to be rewritten as "Host:
1812 [3ffe:8100:200:2::2]:1234". */
1814 /* Formats arranged for hfmt[add_port][add_squares]. */
1815 static const char *hfmt[][2] = {
1816 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1818 int add_port = u->port != scheme_default_port (u->scheme);
1819 int add_squares = strchr (u->host, ':') != NULL;
1820 request_set_header (req, "Host",
1821 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1825 if (inhibit_keep_alive)
1826 request_set_header (req, "Connection", "Close", rel_none);
1830 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1833 request_set_header (req, "Connection", "Close", rel_none);
1834 request_set_header (req, "Proxy-Connection", "Keep-Alive", rel_none);
1841 if (opt.body_data || opt.body_file)
1843 request_set_header (req, "Content-Type",
1844 "application/x-www-form-urlencoded", rel_none);
1847 body_data_size = strlen (opt.body_data);
1850 body_data_size = file_size (opt.body_file);
1851 if (body_data_size == -1)
1853 logprintf (LOG_NOTQUIET, _("BODY data file %s missing: %s\n"),
1854 quote (opt.body_file), strerror (errno));
1858 request_set_header (req, "Content-Length",
1859 xstrdup (number_to_static_string (body_data_size)),
1865 /* We need to come back here when the initial attempt to retrieve
1866 without authorization header fails. (Expected to happen at least
1867 for the Digest authorization scheme.) */
1870 request_set_header (req, "Cookie",
1871 cookie_header (wget_cookie_jar,
1872 u->host, u->port, u->path,
1874 u->scheme == SCHEME_HTTPS
1881 /* Add the user headers. */
1882 if (opt.user_headers)
1885 for (i = 0; opt.user_headers[i]; i++)
1886 request_set_user_header (req, opt.user_headers[i]);
1892 char *proxy_user, *proxy_passwd;
1893 /* For normal username and password, URL components override
1894 command-line/wgetrc parameters. With proxy
1895 authentication, it's the reverse, because proxy URLs are
1896 normally the "permanent" ones, so command-line args
1897 should take precedence. */
1898 if (opt.proxy_user && opt.proxy_passwd)
1900 proxy_user = opt.proxy_user;
1901 proxy_passwd = opt.proxy_passwd;
1905 proxy_user = proxy->user;
1906 proxy_passwd = proxy->passwd;
1908 /* #### This does not appear right. Can't the proxy request,
1909 say, `Digest' authentication? */
1910 if (proxy_user && proxy_passwd)
1911 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1913 /* If we're using a proxy, we will be connecting to the proxy
1917 /* Proxy authorization over SSL is handled below. */
1919 if (u->scheme != SCHEME_HTTPS)
1921 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1926 /* Establish the connection. */
1928 if (inhibit_keep_alive)
1932 /* Look for a persistent connection to target host, unless a
1933 proxy is used. The exception is when SSL is in use, in which
1934 case the proxy is nothing but a passthrough to the target
1935 host, registered as a connection to the latter. */
1936 struct url *relevant = conn;
1938 if (u->scheme == SCHEME_HTTPS)
1942 if (persistent_available_p (relevant->host, relevant->port,
1944 relevant->scheme == SCHEME_HTTPS,
1948 &host_lookup_failed))
1950 int family = socket_family (pconn.socket, ENDPOINT_PEER);
1951 sock = pconn.socket;
1952 using_ssl = pconn.ssl;
1954 if (family == AF_INET6)
1955 logprintf (LOG_VERBOSE, _("Reusing existing connection to [%s]:%d.\n"),
1956 quotearg_style (escape_quoting_style, pconn.host),
1960 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1961 quotearg_style (escape_quoting_style, pconn.host),
1963 DEBUGP (("Reusing fd %d.\n", sock));
1964 if (pconn.authorized)
1965 /* If the connection is already authorized, the "Basic"
1966 authorization added by code above is unnecessary and
1968 request_remove_header (req, "Authorization");
1970 else if (host_lookup_failed)
1973 logprintf(LOG_NOTQUIET,
1974 _("%s: unable to resolve host address %s\n"),
1975 exec_name, quote (relevant->host));
1978 else if (sock != -1)
1986 sock = connect_to_host (conn->host, conn->port);
1995 return (retryable_socket_connect_error (errno)
1996 ? CONERROR : CONIMPOSSIBLE);
2000 if (proxy && u->scheme == SCHEME_HTTPS)
2002 /* When requesting SSL URLs through proxies, use the
2003 CONNECT method to request passthrough. */
2004 struct request *connreq = request_new ("CONNECT",
2005 aprintf ("%s:%d", u->host, u->port));
2006 SET_USER_AGENT (connreq);
2009 request_set_header (connreq, "Proxy-Authorization",
2010 proxyauth, rel_value);
2011 /* Now that PROXYAUTH is part of the CONNECT request,
2012 zero it out so we don't send proxy authorization with
2013 the regular request below. */
2016 request_set_header (connreq, "Host",
2017 aprintf ("%s:%d", u->host, u->port),
2020 write_error = request_send (connreq, sock, 0);
2021 request_free (connreq);
2022 if (write_error < 0)
2024 CLOSE_INVALIDATE (sock);
2029 head = read_http_response_head (sock);
2032 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
2034 CLOSE_INVALIDATE (sock);
2044 DEBUGP (("proxy responded with: [%s]\n", head));
2046 resp = resp_new (head);
2047 statcode = resp_status (resp, &message);
2050 char *tms = datetime_str (time (NULL));
2051 logprintf (LOG_VERBOSE, "%d\n", statcode);
2052 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
2053 quotearg_style (escape_quoting_style,
2054 _("Malformed status line")));
2059 hs->message = xstrdup (message);
2062 if (statcode != 200)
2065 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
2066 message ? quotearg_style (escape_quoting_style, message) : "?");
2067 xfree_null (message);
2071 xfree_null (message);
2073 /* SOCK is now *really* connected to u->host, so update CONN
2074 to reflect this. That way register_persistent will
2075 register SOCK as being connected to u->host:u->port. */
2079 if (conn->scheme == SCHEME_HTTPS)
2081 if (!ssl_connect_wget (sock, u->host))
2087 else if (!ssl_check_certificate (sock, u->host))
2091 return VERIFCERTERR;
2095 #endif /* HAVE_SSL */
2098 /* Open the temporary file where we will write the request. */
2101 warc_tmp = warc_tempfile ();
2102 if (warc_tmp == NULL)
2104 CLOSE_INVALIDATE (sock);
2106 return WARC_TMP_FOPENERR;
2111 warc_ip = (ip_address *) alloca (sizeof (ip_address));
2112 socket_ip_address (sock, warc_ip, ENDPOINT_PEER);
2116 /* Send the request to server. */
2117 write_error = request_send (req, sock, warc_tmp);
2119 if (write_error >= 0)
2123 DEBUGP (("[BODY data: %s]\n", opt.body_data));
2124 write_error = fd_write (sock, opt.body_data, body_data_size, -1);
2125 if (write_error >= 0 && warc_tmp != NULL)
2127 /* Remember end of headers / start of payload. */
2128 warc_payload_offset = ftello (warc_tmp);
2130 /* Write a copy of the data to the WARC record. */
2131 int warc_tmp_written = fwrite (opt.body_data, 1, body_data_size, warc_tmp);
2132 if (warc_tmp_written != body_data_size)
2136 else if (opt.body_file && body_data_size != 0)
2138 if (warc_tmp != NULL)
2139 /* Remember end of headers / start of payload */
2140 warc_payload_offset = ftello (warc_tmp);
2142 write_error = body_file_send (sock, opt.body_file, body_data_size, warc_tmp);
2146 if (write_error < 0)
2148 CLOSE_INVALIDATE (sock);
2151 if (warc_tmp != NULL)
2154 if (write_error == -2)
2155 return WARC_TMP_FWRITEERR;
2159 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
2160 proxy ? "Proxy" : "HTTP");
2169 /* Generate a timestamp and uuid for this request. */
2170 warc_timestamp (warc_timestamp_str);
2171 warc_uuid_str (warc_request_uuid);
2173 /* Create a request record and store it in the WARC file. */
2174 warc_result = warc_write_request_record (u->url, warc_timestamp_str,
2175 warc_request_uuid, warc_ip,
2176 warc_tmp, warc_payload_offset);
2179 CLOSE_INVALIDATE (sock);
2184 /* warc_write_request_record has also closed warc_tmp. */
2189 head = read_http_response_head (sock);
2194 logputs (LOG_NOTQUIET, _("No data received.\n"));
2195 CLOSE_INVALIDATE (sock);
2201 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
2203 CLOSE_INVALIDATE (sock);
2208 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
2210 resp = resp_new (head);
2212 /* Check for status line. */
2214 statcode = resp_status (resp, &message);
2217 char *tms = datetime_str (time (NULL));
2218 logprintf (LOG_VERBOSE, "%d\n", statcode);
2219 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
2220 quotearg_style (escape_quoting_style,
2221 _("Malformed status line")));
2222 CLOSE_INVALIDATE (sock);
2229 if (H_10X (statcode))
2231 DEBUGP (("Ignoring response\n"));
2237 hs->message = xstrdup (message);
2238 if (!opt.server_response)
2239 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
2240 message ? quotearg_style (escape_quoting_style, message) : "");
2243 logprintf (LOG_VERBOSE, "\n");
2244 print_server_response (resp, " ");
2247 if (!opt.ignore_length
2248 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
2252 parsed = str_to_wgint (hdrval, NULL, 10);
2253 if (parsed == WGINT_MAX && errno == ERANGE)
2256 #### If Content-Length is out of range, it most likely
2257 means that the file is larger than 2G and that we're
2258 compiled without LFS. In that case we should probably
2259 refuse to even attempt to download the file. */
2262 else if (parsed < 0)
2264 /* Negative Content-Length; nonsensical, so we can't
2265 assume any information about the content to receive. */
2272 /* Check for keep-alive related responses. */
2273 if (!inhibit_keep_alive && contlen != -1)
2275 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2277 if (0 == strcasecmp (hdrval, "Close"))
2282 chunked_transfer_encoding = false;
2283 if (resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval))
2284 && 0 == strcasecmp (hdrval, "chunked"))
2285 chunked_transfer_encoding = true;
2287 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2291 const char *scbeg, *scend;
2292 /* The jar should have been created by now. */
2293 assert (wget_cookie_jar != NULL);
2295 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2296 &scbeg, &scend)) != -1;
2299 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2300 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2301 u->path, set_cookie);
2306 /* The server has promised that it will not close the connection
2307 when we're done. This means that we can register it. */
2308 register_persistent (conn->host, conn->port, sock, using_ssl);
2310 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2312 /* Authorization is required. */
2314 /* Normally we are not interested in the response body.
2315 But if we are writing a WARC file we are: we like to keep everyting. */
2319 type = resp_header_strdup (resp, "Content-Type");
2320 err = read_response_body (hs, sock, NULL, contlen, 0,
2321 chunked_transfer_encoding,
2322 u->url, warc_timestamp_str,
2323 warc_request_uuid, warc_ip, type,
2327 if (err != RETRFINISHED || hs->res < 0)
2329 CLOSE_INVALIDATE (sock);
2331 xfree_null (message);
2337 CLOSE_FINISH (sock);
2341 /* Since WARC is disabled, we are not interested in the response body. */
2342 if (keep_alive && !head_only
2343 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2344 CLOSE_FINISH (sock);
2346 CLOSE_INVALIDATE (sock);
2349 pconn.authorized = false;
2350 uerr_t auth_err = RETROK;
2351 if (!auth_finished && (user && passwd))
2353 /* IIS sends multiple copies of WWW-Authenticate, one with
2354 the value "negotiate", and other(s) with data. Loop over
2355 all the occurrences and pick the one we recognize. */
2357 const char *wabeg, *waend;
2358 char *www_authenticate = NULL;
2360 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2361 &wabeg, &waend)) != -1;
2363 if (known_authentication_scheme_p (wabeg, waend))
2365 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2369 if (!www_authenticate)
2371 /* If the authentication header is missing or
2372 unrecognized, there's no sense in retrying. */
2373 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2375 else if (!basic_auth_finished
2376 || !BEGINS_WITH (www_authenticate, "Basic"))
2378 char *pth = url_full_path (u);
2381 auth_stat = xmalloc (sizeof (uerr_t));
2382 *auth_stat = RETROK;
2384 value = create_authorization_line (www_authenticate,
2386 request_method (req),
2391 auth_err = *auth_stat;
2392 if (auth_err == RETROK)
2394 request_set_header (req, "Authorization", value, rel_value);
2396 if (BEGINS_WITH (www_authenticate, "NTLM"))
2398 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2400 /* Need to register this host as using basic auth,
2401 * so we automatically send creds next time. */
2402 register_basic_auth_host (u->host);
2406 xfree_null (message);
2410 goto retry_with_auth;
2414 /* Creating the Authorization header went wrong */
2419 /* We already did Basic auth, and it failed. Gotta
2424 xfree_null (message);
2427 if (auth_err == RETROK)
2432 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2434 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2436 pconn.authorized = true;
2439 /* Determine the local filename if needed. Notice that if -O is used
2440 * hstat.local_file is set by http_loop to the argument of -O. */
2441 if (!hs->local_file)
2443 char *local_file = NULL;
2445 /* Honor Content-Disposition whether possible. */
2446 if (!opt.content_disposition
2447 || !resp_header_copy (resp, "Content-Disposition",
2448 hdrval, sizeof (hdrval))
2449 || !parse_content_disposition (hdrval, &local_file))
2451 /* The Content-Disposition header is missing or broken.
2452 * Choose unique file name according to given URL. */
2453 hs->local_file = url_file_name (u, NULL);
2457 DEBUGP (("Parsed filename from Content-Disposition: %s\n",
2459 hs->local_file = url_file_name (u, local_file);
2463 /* TODO: perform this check only once. */
2464 if (!hs->existence_checked && file_exists_p (hs->local_file))
2466 if (opt.noclobber && !opt.output_document)
2468 /* If opt.noclobber is turned on and file already exists, do not
2469 retrieve the file. But if the output_document was given, then this
2470 test was already done and the file didn't exist. Hence the !opt.output_document */
2471 get_file_flags (hs->local_file, dt);
2475 xfree_null (message);
2476 return RETRUNNEEDED;
2478 else if (!ALLOW_CLOBBER)
2480 char *unique = unique_name (hs->local_file, true);
2481 if (unique != hs->local_file)
2482 xfree (hs->local_file);
2483 hs->local_file = unique;
2486 hs->existence_checked = true;
2488 /* Support timestamping */
2489 /* TODO: move this code out of gethttp. */
2490 if (opt.timestamping && !hs->timestamp_checked)
2492 size_t filename_len = strlen (hs->local_file);
2493 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2494 bool local_dot_orig_file_exists = false;
2495 char *local_filename = NULL;
2498 if (opt.backup_converted)
2499 /* If -K is specified, we'll act on the assumption that it was specified
2500 last time these files were downloaded as well, and instead of just
2501 comparing local file X against server file X, we'll compare local
2502 file X.orig (if extant, else X) against server file X. If -K
2503 _wasn't_ specified last time, or the server contains files called
2504 *.orig, -N will be back to not operating correctly with -k. */
2506 /* Would a single s[n]printf() call be faster? --dan
2508 Definitely not. sprintf() is horribly slow. It's a
2509 different question whether the difference between the two
2510 affects a program. Usually I'd say "no", but at one
2511 point I profiled Wget, and found that a measurable and
2512 non-negligible amount of time was lost calling sprintf()
2513 in url.c. Replacing sprintf with inline calls to
2514 strcpy() and number_to_string() made a difference.
2516 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2517 memcpy (filename_plus_orig_suffix + filename_len,
2518 ORIG_SFX, sizeof (ORIG_SFX));
2520 /* Try to stat() the .orig file. */
2521 if (stat (filename_plus_orig_suffix, &st) == 0)
2523 local_dot_orig_file_exists = true;
2524 local_filename = filename_plus_orig_suffix;
2528 if (!local_dot_orig_file_exists)
2529 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2530 if (stat (hs->local_file, &st) == 0)
2531 local_filename = hs->local_file;
2533 if (local_filename != NULL)
2534 /* There was a local file, so we'll check later to see if the version
2535 the server has is the same version we already have, allowing us to
2538 hs->orig_file_name = xstrdup (local_filename);
2539 hs->orig_file_size = st.st_size;
2540 hs->orig_file_tstamp = st.st_mtime;
2542 /* Modification time granularity is 2 seconds for Windows, so
2543 increase local time by 1 second for later comparison. */
2544 ++hs->orig_file_tstamp;
2551 hs->statcode = statcode;
2553 hs->error = xstrdup (_("Malformed status line"));
2555 hs->error = xstrdup (_("(no description)"));
2557 hs->error = xstrdup (message);
2558 xfree_null (message);
2560 type = resp_header_strdup (resp, "Content-Type");
2563 char *tmp = strchr (type, ';');
2566 /* sXXXav: only needed if IRI support is enabled */
2567 char *tmp2 = tmp + 1;
2569 while (tmp > type && c_isspace (tmp[-1]))
2573 /* Try to get remote encoding if needed */
2574 if (opt.enable_iri && !opt.encoding_remote)
2576 tmp = parse_charset (tmp2);
2578 set_content_encoding (iri, tmp);
2582 hs->newloc = resp_header_strdup (resp, "Location");
2583 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2585 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2587 wgint first_byte_pos, last_byte_pos, entity_length;
2588 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2591 contrange = first_byte_pos;
2592 contlen = last_byte_pos - first_byte_pos + 1;
2597 /* 20x responses are counted among successful by default. */
2598 if (H_20X (statcode))
2601 /* Return if redirected. */
2602 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2604 /* RFC2068 says that in case of the 300 (multiple choices)
2605 response, the server can output a preferred URL through
2606 `Location' header; otherwise, the request should be treated
2607 like GET. So, if the location is set, it will be a
2608 redirection; otherwise, just proceed normally. */
2609 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2613 logprintf (LOG_VERBOSE,
2614 _("Location: %s%s\n"),
2615 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2616 hs->newloc ? _(" [following]") : "");
2618 /* In case the caller cares to look... */
2623 /* Normally we are not interested in the response body of a redirect.
2624 But if we are writing a WARC file we are: we like to keep everyting. */
2627 int err = read_response_body (hs, sock, NULL, contlen, 0,
2628 chunked_transfer_encoding,
2629 u->url, warc_timestamp_str,
2630 warc_request_uuid, warc_ip, type,
2633 if (err != RETRFINISHED || hs->res < 0)
2635 CLOSE_INVALIDATE (sock);
2641 CLOSE_FINISH (sock);
2645 /* Since WARC is disabled, we are not interested in the response body. */
2646 if (keep_alive && !head_only
2647 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2648 CLOSE_FINISH (sock);
2650 CLOSE_INVALIDATE (sock);
2655 /* From RFC2616: The status codes 303 and 307 have
2656 been added for servers that wish to make unambiguously
2657 clear which kind of reaction is expected of the client.
2659 A 307 should be redirected using the same method,
2660 in other words, a POST should be preserved and not
2661 converted to a GET in that case.
2663 With strict adherence to RFC2616, POST requests are not
2664 converted to a GET request on 301 Permanent Redirect
2665 or 302 Temporary Redirect.
2667 A switch may be provided later based on the HTTPbis draft
2668 that allows clients to convert POST requests to GET
2669 requests on 301 and 302 response codes. */
2672 case HTTP_STATUS_TEMPORARY_REDIRECT:
2673 return NEWLOCATION_KEEP_POST;
2675 case HTTP_STATUS_MOVED_PERMANENTLY:
2676 if (opt.method && strcasecmp (opt.method, "post") != 0)
2677 return NEWLOCATION_KEEP_POST;
2679 case HTTP_STATUS_MOVED_TEMPORARILY:
2680 if (opt.method && strcasecmp (opt.method, "post") != 0)
2681 return NEWLOCATION_KEEP_POST;
2691 /* If content-type is not given, assume text/html. This is because
2692 of the multitude of broken CGI's that "forget" to generate the
2695 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2696 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2702 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2707 if (opt.adjust_extension)
2710 /* -E / --adjust-extension / adjust_extension = on was specified,
2711 and this is a text/html file. If some case-insensitive
2712 variation on ".htm[l]" isn't already the file's suffix,
2715 ensure_extension (hs, ".html", dt);
2717 else if (*dt & TEXTCSS)
2719 ensure_extension (hs, ".css", dt);
2723 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2724 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK
2725 && contrange == 0 && contlen >= 0 && hs->restval >= contlen))
2727 /* If `-c' is in use and the file has been fully downloaded (or
2728 the remote file has shrunk), Wget effectively requests bytes
2729 after the end of file and the server response with 416
2730 (or 200 with a <= Content-Length. */
2731 logputs (LOG_VERBOSE, _("\
2732 \n The file is already fully retrieved; nothing to do.\n\n"));
2733 /* In case the caller inspects. */
2736 /* Mark as successfully retrieved. */
2739 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2740 might be more bytes in the body. */
2742 return RETRUNNEEDED;
2744 if ((contrange != 0 && contrange != hs->restval)
2745 || (H_PARTIAL (statcode) && !contrange))
2747 /* The Range request was somehow misunderstood by the server.
2750 CLOSE_INVALIDATE (sock);
2757 hs->contlen = contlen + contrange;
2763 /* No need to print this output if the body won't be
2764 downloaded at all, or if the original server response is
2766 logputs (LOG_VERBOSE, _("Length: "));
2769 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2770 if (contlen + contrange >= 1024)
2771 logprintf (LOG_VERBOSE, " (%s)",
2772 human_readable (contlen + contrange));
2775 if (contlen >= 1024)
2776 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2777 number_to_static_string (contlen),
2778 human_readable (contlen));
2780 logprintf (LOG_VERBOSE, _(", %s remaining"),
2781 number_to_static_string (contlen));
2785 logputs (LOG_VERBOSE,
2786 opt.ignore_length ? _("ignored") : _("unspecified"));
2788 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2790 logputs (LOG_VERBOSE, "\n");
2794 /* Return if we have no intention of further downloading. */
2795 if ((!(*dt & RETROKF) && !opt.content_on_error) || head_only)
2797 /* In case the caller cares to look... */
2802 /* Normally we are not interested in the response body of a error responses.
2803 But if we are writing a WARC file we are: we like to keep everyting. */
2806 int err = read_response_body (hs, sock, NULL, contlen, 0,
2807 chunked_transfer_encoding,
2808 u->url, warc_timestamp_str,
2809 warc_request_uuid, warc_ip, type,
2812 if (err != RETRFINISHED || hs->res < 0)
2814 CLOSE_INVALIDATE (sock);
2820 CLOSE_FINISH (sock);
2824 /* Since WARC is disabled, we are not interested in the response body. */
2826 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2827 servers not to send body in response to a HEAD request, and
2828 those that do will likely be caught by test_socket_open.
2829 If not, they can be worked around using
2830 `--no-http-keep-alive'. */
2831 CLOSE_FINISH (sock);
2833 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2834 /* Successfully skipped the body; also keep using the socket. */
2835 CLOSE_FINISH (sock);
2837 CLOSE_INVALIDATE (sock);
2842 return RETRFINISHED;
2846 For VMS, define common fopen() optional arguments.
2849 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2850 # define FOPEN_BIN_FLAG 3
2851 #else /* def __VMS */
2852 # define FOPEN_BIN_FLAG true
2853 #endif /* def __VMS [else] */
2855 /* Open the local file. */
2858 mkalldirs (hs->local_file);
2860 rotate_backups (hs->local_file);
2867 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2868 #else /* def __VMS */
2869 fp = fopen (hs->local_file, "ab");
2870 #endif /* def __VMS [else] */
2872 else if (ALLOW_CLOBBER || count > 0)
2874 if (opt.unlink && file_exists_p (hs->local_file))
2876 int res = unlink (hs->local_file);
2879 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file,
2881 CLOSE_INVALIDATE (sock);
2892 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2893 #else /* def __VMS */
2894 fp = fopen (hs->local_file, "wb");
2895 #endif /* def __VMS [else] */
2899 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2900 if (!fp && errno == EEXIST)
2902 /* We cannot just invent a new name and use it (which is
2903 what functions like unique_create typically do)
2904 because we told the user we'd use this name.
2905 Instead, return and retry the download. */
2906 logprintf (LOG_NOTQUIET,
2907 _("%s has sprung into existence.\n"),
2909 CLOSE_INVALIDATE (sock);
2912 return FOPEN_EXCL_ERR;
2917 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2918 CLOSE_INVALIDATE (sock);
2927 /* Print fetch message, if opt.verbose. */
2930 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2931 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2935 err = read_response_body (hs, sock, fp, contlen, contrange,
2936 chunked_transfer_encoding,
2937 u->url, warc_timestamp_str,
2938 warc_request_uuid, warc_ip, type,
2941 /* Now we no longer need to store the response header. */
2946 CLOSE_FINISH (sock);
2948 CLOSE_INVALIDATE (sock);
2956 /* The genuine HTTP loop! This is the part where the retrieval is
2957 retried, and retried, and retried, and... */
2959 http_loop (struct url *u, struct url *original_url, char **newloc,
2960 char **local_file, const char *referer, int *dt, struct url *proxy,
2964 bool got_head = false; /* used for time-stamping and filename detection */
2965 bool time_came_from_head = false;
2966 bool got_name = false;
2969 uerr_t err, ret = TRYLIMEXC;
2970 time_t tmr = -1; /* remote time-stamp */
2971 struct http_stat hstat; /* HTTP status */
2973 bool send_head_first = true;
2975 bool force_full_retrieve = false;
2978 /* If we are writing to a WARC file: always retrieve the whole file. */
2979 if (opt.warc_filename != NULL)
2980 force_full_retrieve = true;
2983 /* Assert that no value for *LOCAL_FILE was passed. */
2984 assert (local_file == NULL || *local_file == NULL);
2986 /* Set LOCAL_FILE parameter. */
2987 if (local_file && opt.output_document)
2988 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2990 /* Reset NEWLOC parameter. */
2993 /* This used to be done in main(), but it's a better idea to do it
2994 here so that we don't go through the hoops if we're just using
2999 /* Warn on (likely bogus) wildcard usage in HTTP. */
3000 if (opt.ftp_glob && has_wildcards_p (u->path))
3001 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
3003 /* Setup hstat struct. */
3005 hstat.referer = referer;
3007 if (opt.output_document)
3009 hstat.local_file = xstrdup (opt.output_document);
3012 else if (!opt.content_disposition)
3015 url_file_name (opt.trustservernames ? u : original_url, NULL);
3019 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
3021 /* If opt.noclobber is turned on and file already exists, do not
3022 retrieve the file. But if the output_document was given, then this
3023 test was already done and the file didn't exist. Hence the !opt.output_document */
3024 get_file_flags (hstat.local_file, dt);
3029 /* Reset the counter. */
3032 /* Reset the document type. */
3035 /* Skip preliminary HEAD request if we're not in spider mode. */
3037 send_head_first = false;
3039 /* Send preliminary HEAD request if --content-disposition and -c are used
3041 if (opt.content_disposition && opt.always_rest)
3042 send_head_first = true;
3044 /* Send preliminary HEAD request if -N is given and we have an existing
3045 * destination file. */
3046 if (!opt.output_document)
3047 file_name = url_file_name (opt.trustservernames ? u : original_url, NULL);
3049 file_name = xstrdup (opt.output_document);
3050 if (opt.timestamping && (file_exists_p (file_name)
3051 || opt.content_disposition))
3052 send_head_first = true;
3058 /* Increment the pass counter. */
3060 sleep_between_retrievals (count);
3062 /* Get the current time string. */
3063 tms = datetime_str (time (NULL));
3065 if (opt.spider && !got_head)
3066 logprintf (LOG_VERBOSE, _("\
3067 Spider mode enabled. Check if remote file exists.\n"));
3069 /* Print fetch message, if opt.verbose. */
3072 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3077 sprintf (tmp, _("(try:%2d)"), count);
3078 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
3083 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
3088 ws_changetitle (hurl);
3093 /* Default document type is empty. However, if spider mode is
3094 on or time-stamping is employed, HEAD_ONLY commands is
3095 encoded within *dt. */
3096 if (send_head_first && !got_head)
3101 /* Decide whether or not to restart. */
3102 if (force_full_retrieve)
3103 hstat.restval = hstat.len;
3104 else if (opt.always_rest
3106 && stat (hstat.local_file, &st) == 0
3107 && S_ISREG (st.st_mode))
3108 /* When -c is used, continue from on-disk size. (Can't use
3109 hstat.len even if count>1 because we don't want a failed
3110 first attempt to clobber existing data.) */
3111 hstat.restval = st.st_size;
3113 /* otherwise, continue where the previous try left off */
3114 hstat.restval = hstat.len;
3118 /* Decide whether to send the no-cache directive. We send it in
3120 a) we're using a proxy, and we're past our first retrieval.
3121 Some proxies are notorious for caching incomplete data, so
3122 we require a fresh get.
3123 b) caching is explicitly inhibited. */
3124 if ((proxy && count > 1) /* a */
3125 || !opt.allow_cache) /* b */
3126 *dt |= SEND_NOCACHE;
3128 *dt &= ~SEND_NOCACHE;
3130 /* Try fetching the document, or at least its head. */
3131 err = gethttp (u, &hstat, dt, proxy, iri, count);
3134 tms = datetime_str (time (NULL));
3136 /* Get the new location (with or without the redirection). */
3138 *newloc = xstrdup (hstat.newloc);
3142 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
3143 case CONERROR: case READERR: case WRITEFAILED:
3144 case RANGEERR: case FOPEN_EXCL_ERR:
3145 /* Non-fatal errors continue executing the loop, which will
3146 bring them to "while" statement at the end, to judge
3147 whether the number of tries was exceeded. */
3148 printwhat (count, opt.ntry);
3150 case FWRITEERR: case FOPENERR:
3151 /* Another fatal error. */
3152 logputs (LOG_VERBOSE, "\n");
3153 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
3154 quote (hstat.local_file), strerror (errno));
3155 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case SSLINITFAILED:
3156 case CONTNOTSUPPORTED: case VERIFCERTERR: case FILEBADFILE:
3158 /* Fatal errors just return from the function. */
3162 /* A missing attribute in a Header is a fatal Protocol error. */
3163 logputs (LOG_VERBOSE, "\n");
3164 logprintf (LOG_NOTQUIET, _("Required attribute missing from Header received.\n"));
3168 logputs (LOG_VERBOSE, "\n");
3169 logprintf (LOG_NOTQUIET, _("Username/Password Authentication Failed.\n"));
3173 /* A fatal WARC error. */
3174 logputs (LOG_VERBOSE, "\n");
3175 logprintf (LOG_NOTQUIET, _("Cannot write to WARC file.\n"));
3178 case WARC_TMP_FOPENERR: case WARC_TMP_FWRITEERR:
3179 /* A fatal WARC error. */
3180 logputs (LOG_VERBOSE, "\n");
3181 logprintf (LOG_NOTQUIET, _("Cannot write to temporary WARC file.\n"));
3185 /* Another fatal error. */
3186 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
3190 /* Another fatal error. */
3191 logputs (LOG_VERBOSE, "\n");
3192 logprintf (LOG_NOTQUIET, _("Cannot unlink %s (%s).\n"),
3193 quote (hstat.local_file), strerror (errno));
3197 case NEWLOCATION_KEEP_POST:
3198 /* Return the new location to the caller. */
3201 logprintf (LOG_NOTQUIET,
3202 _("ERROR: Redirection (%d) without location.\n"),
3212 /* The file was already fully retrieved. */
3216 /* Deal with you later. */
3219 /* All possibilities should have been exhausted. */
3223 if (!(*dt & RETROKF))
3228 /* #### Ugly ugly ugly! */
3229 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3230 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
3233 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
3235 && (hstat.statcode == 500 || hstat.statcode == 501))
3240 /* Maybe we should always keep track of broken links, not just in
3242 * Don't log error if it was UTF-8 encoded because we will try
3243 * once unencoded. */
3244 else if (opt.spider && !iri->utf8_encode)
3246 /* #### Again: ugly ugly ugly! */
3248 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
3249 nonexisting_url (hurl);
3250 logprintf (LOG_NOTQUIET, _("\
3251 Remote file does not exist -- broken link!!!\n"));
3255 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
3256 tms, hstat.statcode,
3257 quotearg_style (escape_quoting_style, hstat.error));
3259 logputs (LOG_VERBOSE, "\n");
3265 /* Did we get the time-stamp? */
3268 got_head = true; /* no more time-stamping */
3270 if (opt.timestamping && !hstat.remote_time)
3272 logputs (LOG_NOTQUIET, _("\
3273 Last-modified header missing -- time-stamps turned off.\n"));
3275 else if (hstat.remote_time)
3277 /* Convert the date-string into struct tm. */
3278 tmr = http_atotm (hstat.remote_time);
3279 if (tmr == (time_t) (-1))
3280 logputs (LOG_VERBOSE, _("\
3281 Last-modified header invalid -- time-stamp ignored.\n"));
3282 if (*dt & HEAD_ONLY)
3283 time_came_from_head = true;
3286 if (send_head_first)
3288 /* The time-stamping section. */
3289 if (opt.timestamping)
3291 if (hstat.orig_file_name) /* Perform the following
3292 checks only if the file
3294 download already exists. */
3296 if (hstat.remote_time &&
3297 tmr != (time_t) (-1))
3299 /* Now time-stamping can be used validly.
3300 Time-stamping means that if the sizes of
3301 the local and remote file match, and local
3302 file is newer than the remote file, it will
3303 not be retrieved. Otherwise, the normal
3304 download procedure is resumed. */
3305 if (hstat.orig_file_tstamp >= tmr)
3307 if (hstat.contlen == -1
3308 || hstat.orig_file_size == hstat.contlen)
3310 logprintf (LOG_VERBOSE, _("\
3311 Server file no newer than local file %s -- not retrieving.\n\n"),
3312 quote (hstat.orig_file_name));
3318 logprintf (LOG_VERBOSE, _("\
3319 The sizes do not match (local %s) -- retrieving.\n"),
3320 number_to_static_string (hstat.orig_file_size));
3325 force_full_retrieve = true;
3326 logputs (LOG_VERBOSE,
3327 _("Remote file is newer, retrieving.\n"));
3330 logputs (LOG_VERBOSE, "\n");
3334 /* free_hstat (&hstat); */
3335 hstat.timestamp_checked = true;
3340 bool finished = true;
3345 logputs (LOG_VERBOSE, _("\
3346 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
3351 logprintf (LOG_VERBOSE, _("\
3352 Remote file exists but does not contain any link -- not retrieving.\n\n"));
3353 ret = RETROK; /* RETRUNNEEDED is not for caller. */
3360 logprintf (LOG_VERBOSE, _("\
3361 Remote file exists and could contain further links,\n\
3362 but recursion is disabled -- not retrieving.\n\n"));
3366 logprintf (LOG_VERBOSE, _("\
3367 Remote file exists.\n\n"));
3369 ret = RETROK; /* RETRUNNEEDED is not for caller. */
3374 logprintf (LOG_NONVERBOSE,
3375 _("%s URL: %s %2d %s\n"),
3376 tms, u->url, hstat.statcode,
3377 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
3384 count = 0; /* the retrieve count for HEAD is reset */
3386 } /* send_head_first */
3389 if (opt.useservertimestamps
3390 && (tmr != (time_t) (-1))
3391 && ((hstat.len == hstat.contlen) ||
3392 ((hstat.res == 0) && (hstat.contlen == -1))))
3394 const char *fl = NULL;
3395 set_local_file (&fl, hstat.local_file);
3399 /* Reparse time header, in case it's changed. */
3400 if (time_came_from_head
3401 && hstat.remote_time && hstat.remote_time[0])
3403 newtmr = http_atotm (hstat.remote_time);
3404 if (newtmr != (time_t)-1)
3410 /* End of time-stamping section. */
3412 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3413 total_download_time += hstat.dltime;
3415 if (hstat.len == hstat.contlen)
3419 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3421 logprintf (LOG_VERBOSE,
3423 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3424 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3426 write_to_stdout ? "" : quote (hstat.local_file),
3427 number_to_static_string (hstat.len),
3428 number_to_static_string (hstat.contlen));
3429 logprintf (LOG_NONVERBOSE,
3430 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3432 number_to_static_string (hstat.len),
3433 number_to_static_string (hstat.contlen),
3434 hstat.local_file, count);
3437 total_downloaded_bytes += hstat.rd_size;
3439 /* Remember that we downloaded the file for later ".orig" code. */
3440 if (*dt & ADDED_HTML_EXTENSION)
3441 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3443 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3448 else if (hstat.res == 0) /* No read error */
3450 if (hstat.contlen == -1) /* We don't know how much we were supposed
3451 to get, so assume we succeeded. */
3455 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3457 logprintf (LOG_VERBOSE,
3459 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3460 : _("%s (%s) - %s saved [%s]\n\n"),
3462 write_to_stdout ? "" : quote (hstat.local_file),
3463 number_to_static_string (hstat.len));
3464 logprintf (LOG_NONVERBOSE,
3465 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3466 tms, u->url, number_to_static_string (hstat.len),
3467 hstat.local_file, count);
3470 total_downloaded_bytes += hstat.rd_size;
3472 /* Remember that we downloaded the file for later ".orig" code. */
3473 if (*dt & ADDED_HTML_EXTENSION)
3474 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3476 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3481 else if (hstat.len < hstat.contlen) /* meaning we lost the
3482 connection too soon */
3484 logprintf (LOG_VERBOSE,
3485 _("%s (%s) - Connection closed at byte %s. "),
3486 tms, tmrate, number_to_static_string (hstat.len));
3487 printwhat (count, opt.ntry);
3490 else if (hstat.len != hstat.restval)
3491 /* Getting here would mean reading more data than
3492 requested with content-length, which we never do. */
3496 /* Getting here probably means that the content-length was
3497 * _less_ than the original, local size. We should probably
3498 * truncate or re-read, or something. FIXME */
3503 else /* from now on hstat.res can only be -1 */
3505 if (hstat.contlen == -1)
3507 logprintf (LOG_VERBOSE,
3508 _("%s (%s) - Read error at byte %s (%s)."),
3509 tms, tmrate, number_to_static_string (hstat.len),
3511 printwhat (count, opt.ntry);
3514 else /* hstat.res == -1 and contlen is given */
3516 logprintf (LOG_VERBOSE,
3517 _("%s (%s) - Read error at byte %s/%s (%s). "),
3519 number_to_static_string (hstat.len),
3520 number_to_static_string (hstat.contlen),
3522 printwhat (count, opt.ntry);
3528 while (!opt.ntry || (count < opt.ntry));
3531 if (ret == RETROK && local_file)
3532 *local_file = xstrdup (hstat.local_file);
3533 free_hstat (&hstat);
3538 /* Check whether the result of strptime() indicates success.
3539 strptime() returns the pointer to how far it got to in the string.
3540 The processing has been successful if the string is at `GMT' or
3541 `+X', or at the end of the string.
3543 In extended regexp parlance, the function returns 1 if P matches
3544 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3545 can return) is considered a failure and 0 is returned. */
3547 check_end (const char *p)
3551 while (c_isspace (*p))
3554 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3555 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3561 /* Convert the textual specification of time in TIME_STRING to the
3562 number of seconds since the Epoch.
3564 TIME_STRING can be in any of the three formats RFC2616 allows the
3565 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3566 as well as the time format used in the Set-Cookie header.
3567 Timezones are ignored, and should be GMT.
3569 Return the computed time_t representation, or -1 if the conversion
3572 This function uses strptime with various string formats for parsing
3573 TIME_STRING. This results in a parser that is not as lenient in
3574 interpreting TIME_STRING as I would like it to be. Being based on
3575 strptime, it always allows shortened months, one-digit days, etc.,
3576 but due to the multitude of formats in which time can be
3577 represented, an ideal HTTP time parser would be even more
3578 forgiving. It should completely ignore things like week days and
3579 concentrate only on the various forms of representing years,
3580 months, days, hours, minutes, and seconds. For example, it would
3581 be nice if it accepted ISO 8601 out of the box.
3583 I've investigated free and PD code for this purpose, but none was
3584 usable. getdate was big and unwieldy, and had potential copyright
3585 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3586 distributed with phttpd, is excellent, but we cannot use it because
3587 it is not assigned to the FSF. So I stuck it with strptime. */
3590 http_atotm (const char *time_string)
3592 /* NOTE: Solaris strptime man page claims that %n and %t match white
3593 space, but that's not universally available. Instead, we simply
3594 use ` ' to mean "skip all WS", which works under all strptime
3595 implementations I've tested. */
3597 static const char *time_formats[] = {
3598 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3599 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3600 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3601 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3602 (used in Set-Cookie, defined in the
3603 Netscape cookie specification.) */
3605 const char *oldlocale;
3606 char savedlocale[256];
3608 time_t ret = (time_t) -1;
3610 /* Solaris strptime fails to recognize English month names in
3611 non-English locales, which we work around by temporarily setting
3612 locale to C before invoking strptime. */
3613 oldlocale = setlocale (LC_TIME, NULL);
3616 size_t l = strlen (oldlocale) + 1;
3617 if (l >= sizeof savedlocale)
3618 savedlocale[0] = '\0';
3620 memcpy (savedlocale, oldlocale, l);
3622 else savedlocale[0] = '\0';
3624 setlocale (LC_TIME, "C");
3626 for (i = 0; i < countof (time_formats); i++)
3630 /* Some versions of strptime use the existing contents of struct
3631 tm to recalculate the date according to format. Zero it out
3632 to prevent stack garbage from influencing strptime. */
3635 if (check_end (strptime (time_string, time_formats[i], &t)))
3642 /* Restore the previous locale. */
3644 setlocale (LC_TIME, savedlocale);
3649 /* Authorization support: We support three authorization schemes:
3651 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3653 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3654 consisting of answering to the server's challenge with the proper
3657 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3658 Stenberg for libcurl. Like digest, NTLM is based on a
3659 challenge-response mechanism, but unlike digest, it is non-standard
3660 (authenticates TCP connections rather than requests), undocumented
3661 and Microsoft-specific. */
3663 /* Create the authentication header contents for the `Basic' scheme.
3664 This is done by encoding the string "USER:PASS" to base64 and
3665 prepending the string "Basic " in front of it. */
3668 basic_authentication_encode (const char *user, const char *passwd)
3671 int len1 = strlen (user) + 1 + strlen (passwd);
3673 t1 = (char *)alloca (len1 + 1);
3674 sprintf (t1, "%s:%s", user, passwd);
3676 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3677 base64_encode (t1, len1, t2);
3679 return concat_strings ("Basic ", t2, (char *) 0);
3682 #define SKIP_WS(x) do { \
3683 while (c_isspace (*(x))) \
3687 #ifdef ENABLE_DIGEST
3688 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3689 an array of 16 bytes containing the hash keys, and BUF should be a
3690 buffer of 33 writable characters (32 for hex digits plus one for
3691 zero termination). */
3693 dump_hash (char *buf, const unsigned char *hash)
3697 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3699 *buf++ = XNUM_TO_digit (*hash >> 4);
3700 *buf++ = XNUM_TO_digit (*hash & 0xf);
3705 /* Take the line apart to find the challenge, and compose a digest
3706 authorization header. See RFC2069 section 2.1.2. */
3708 digest_authentication_encode (const char *au, const char *user,
3709 const char *passwd, const char *method,
3710 const char *path, uerr_t *auth_err)
3712 static char *realm, *opaque, *nonce, *qop, *algorithm;
3717 { "realm", &realm },
3718 { "opaque", &opaque },
3719 { "nonce", &nonce },
3721 { "algorithm", &algorithm }
3723 char cnonce[16] = "";
3727 param_token name, value;
3730 realm = opaque = nonce = algorithm = qop = NULL;
3732 au += 6; /* skip over `Digest' */
3733 while (extract_param (&au, &name, &value, ','))
3736 size_t namelen = name.e - name.b;
3737 for (i = 0; i < countof (options); i++)
3738 if (namelen == strlen (options[i].name)
3739 && 0 == strncmp (name.b, options[i].name,
3742 *options[i].variable = strdupdelim (value.b, value.e);
3747 if (qop != NULL && strcmp(qop,"auth"))
3749 logprintf (LOG_NOTQUIET, _("Unsupported quality of protection '%s'.\n"), qop);
3750 xfree_null (qop); /* force freeing mem and return */
3753 else if (algorithm != NULL && strcmp (algorithm,"MD5") && strcmp (algorithm,"MD5-sess"))
3755 logprintf (LOG_NOTQUIET, _("Unsupported algorithm '%s'.\n"), algorithm);
3756 xfree_null (qop); /* force freeing mem and return */
3760 if (!realm || !nonce || !user || !passwd || !path || !method || !qop)
3763 xfree_null (opaque);
3766 xfree_null (algorithm);
3768 *auth_err = UNKNOWNATTR;
3770 *auth_err = ATTRMISSING;
3774 /* Calculate the digest value. */
3777 unsigned char hash[MD5_DIGEST_SIZE];
3778 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3779 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3781 /* A1BUF = H(user ":" realm ":" password) */
3782 md5_init_ctx (&ctx);
3783 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3784 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3785 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3786 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3787 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3788 md5_finish_ctx (&ctx, hash);
3790 dump_hash (a1buf, hash);
3792 if (algorithm && !strcmp (algorithm, "MD5-sess"))
3794 /* A1BUF = H( H(user ":" realm ":" password) ":" nonce ":" cnonce ) */
3795 snprintf (cnonce, sizeof (cnonce), "%08x", random_number(INT_MAX));
3797 md5_init_ctx (&ctx);
3798 // md5_process_bytes (hash, MD5_DIGEST_SIZE, &ctx);
3799 md5_process_bytes (a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3800 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3801 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3802 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3803 md5_process_bytes ((unsigned char *)cnonce, strlen (cnonce), &ctx);
3804 md5_finish_ctx (&ctx, hash);
3806 dump_hash (a1buf, hash);
3809 /* A2BUF = H(method ":" path) */
3810 md5_init_ctx (&ctx);
3811 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3812 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3813 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3814 md5_finish_ctx (&ctx, hash);
3815 dump_hash (a2buf, hash);
3817 if (qop && (!strcmp(qop, "auth") || !strcmp (qop, "auth-int")))
3819 /* RFC 2617 Digest Access Authentication */
3820 /* generate random hex string */
3822 snprintf(cnonce, sizeof(cnonce), "%08x", random_number(INT_MAX));
3824 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" noncecount ":" clientnonce ":" qop ": " A2BUF) */
3825 md5_init_ctx (&ctx);
3826 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3827 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3828 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3829 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3830 md5_process_bytes ((unsigned char *)"00000001", 8, &ctx); /* TODO: keep track of server nonce values */
3831 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3832 md5_process_bytes ((unsigned char *)cnonce, strlen(cnonce), &ctx);
3833 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3834 md5_process_bytes ((unsigned char *)qop, strlen(qop), &ctx);
3835 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3836 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3837 md5_finish_ctx (&ctx, hash);
3841 /* RFC 2069 Digest Access Authentication */
3842 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3843 md5_init_ctx (&ctx);
3844 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3845 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3846 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3847 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3848 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3849 md5_finish_ctx (&ctx, hash);
3852 dump_hash (response_digest, hash);
3854 res_size = strlen (user)
3858 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3859 + (opaque ? strlen (opaque) : 0)
3860 + (algorithm ? strlen (algorithm) : 0)
3865 res = xmalloc (res_size);
3867 if (qop && !strcmp (qop, "auth"))
3869 res_len = snprintf (res, res_size, "Digest "\
3870 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\""\
3871 ", qop=auth, nc=00000001, cnonce=\"%s\"",
3872 user, realm, nonce, path, response_digest, cnonce);
3877 res_len = snprintf (res, res_size, "Digest "\
3878 "username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3879 user, realm, nonce, path, response_digest);
3884 res_len += snprintf(res + res_len, res_size - res_len, ", opaque=\"%s\"", opaque);
3889 snprintf(res + res_len, res_size - res_len, ", algorithm=\"%s\"", algorithm);
3894 xfree_null (opaque);
3897 xfree_null (algorithm);
3901 #endif /* ENABLE_DIGEST */
3903 /* Computing the size of a string literal must take into account that
3904 value returned by sizeof includes the terminating \0. */
3905 #define STRSIZE(literal) (sizeof (literal) - 1)
3907 /* Whether chars in [b, e) begin with the literal string provided as
3908 first argument and are followed by whitespace or terminating \0.
3909 The comparison is case-insensitive. */
3910 #define STARTS(literal, b, e) \
3912 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3913 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3914 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3915 || c_isspace (b[STRSIZE (literal)])))
3918 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3920 return STARTS ("Basic", hdrbeg, hdrend)
3921 #ifdef ENABLE_DIGEST
3922 || STARTS ("Digest", hdrbeg, hdrend)
3925 || STARTS ("NTLM", hdrbeg, hdrend)
3932 /* Create the HTTP authorization request header. When the
3933 `WWW-Authenticate' response header is seen, according to the
3934 authorization scheme specified in that header (`Basic' and `Digest'
3935 are supported by the current implementation), produce an
3936 appropriate HTTP authorization request header. */
3938 create_authorization_line (const char *au, const char *user,
3939 const char *passwd, const char *method,
3940 const char *path, bool *finished, uerr_t *auth_err)
3942 /* We are called only with known schemes, so we can dispatch on the
3944 switch (c_toupper (*au))
3946 case 'B': /* Basic */
3948 return basic_authentication_encode (user, passwd);
3949 #ifdef ENABLE_DIGEST
3950 case 'D': /* Digest */
3952 return digest_authentication_encode (au, user, passwd, method, path, auth_err);
3955 case 'N': /* NTLM */
3956 if (!ntlm_input (&pconn.ntlm, au))
3961 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3964 /* We shouldn't get here -- this function should be only called
3965 with values approved by known_authentication_scheme_p. */
3973 if (!wget_cookie_jar)
3974 wget_cookie_jar = cookie_jar_new ();
3975 if (opt.cookies_input && !cookies_loaded_p)
3977 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3978 cookies_loaded_p = true;
3985 if (wget_cookie_jar)
3986 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3992 xfree_null (pconn.host);
3993 if (wget_cookie_jar)
3994 cookie_jar_delete (wget_cookie_jar);
3998 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
4000 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
4002 int len = strlen (ext);
4005 strncpy (shortext, ext, len - 1);
4006 shortext[len - 1] = '\0';
4009 if (last_period_in_local_filename == NULL
4010 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
4011 || 0 == strcasecmp (last_period_in_local_filename, ext)))
4013 int local_filename_len = strlen (hs->local_file);
4014 /* Resize the local file, allowing for ".html" preceded by
4015 optional ".NUMBER". */
4016 hs->local_file = xrealloc (hs->local_file,
4017 local_filename_len + 24 + len);
4018 strcpy (hs->local_file + local_filename_len, ext);
4019 /* If clobbering is not allowed and the file, as named,
4020 exists, tack on ".NUMBER.html" instead. */
4021 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
4025 sprintf (hs->local_file + local_filename_len,
4026 ".%d%s", ext_num++, ext);
4027 while (file_exists_p (hs->local_file));
4029 *dt |= ADDED_HTML_EXTENSION;
4037 test_parse_content_disposition()
4045 { "filename=\"file.ext\"", "file.ext", true },
4046 { "attachment; filename=\"file.ext\"", "file.ext", true },
4047 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
4048 { "attachment", NULL, false },
4049 { "attachement; filename*=UTF-8'en-US'hello.txt", "hello.txt", true },
4050 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", "helloworld.txt", true },
4053 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
4058 res = parse_content_disposition (test_array[i].hdrval, &filename);
4060 mu_assert ("test_parse_content_disposition: wrong result",
4061 res == test_array[i].result
4063 || 0 == strcmp (test_array[i].filename, filename)));
4069 #endif /* TESTING */
4072 * vim: et sts=2 sw=2 cino+={s