2 Copyright (C) 2003 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
34 #include <sys/types.h>
45 #if TIME_WITH_SYS_TIME
46 # include <sys/time.h>
50 # include <sys/time.h>
67 # include "gen_sslfunc.h"
75 extern char *version_string;
76 extern LARGE_INT total_downloaded_bytes;
78 extern FILE *output_stream;
79 extern int output_stream_regular;
82 # define MIN(x, y) ((x) > (y) ? (y) : (x))
86 static int cookies_loaded_p;
87 struct cookie_jar *wget_cookie_jar;
89 #define TEXTHTML_S "text/html"
90 #define TEXTXHTML_S "application/xhtml+xml"
92 /* Some status code validation macros: */
93 #define H_20X(x) (((x) >= 200) && ((x) < 300))
94 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
95 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
96 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
97 || (x) == HTTP_STATUS_SEE_OTHER \
98 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
100 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
101 /* Successful 2xx. */
102 #define HTTP_STATUS_OK 200
103 #define HTTP_STATUS_CREATED 201
104 #define HTTP_STATUS_ACCEPTED 202
105 #define HTTP_STATUS_NO_CONTENT 204
106 #define HTTP_STATUS_PARTIAL_CONTENTS 206
108 /* Redirection 3xx. */
109 #define HTTP_STATUS_MULTIPLE_CHOICES 300
110 #define HTTP_STATUS_MOVED_PERMANENTLY 301
111 #define HTTP_STATUS_MOVED_TEMPORARILY 302
112 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
113 #define HTTP_STATUS_NOT_MODIFIED 304
114 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
116 /* Client error 4xx. */
117 #define HTTP_STATUS_BAD_REQUEST 400
118 #define HTTP_STATUS_UNAUTHORIZED 401
119 #define HTTP_STATUS_FORBIDDEN 403
120 #define HTTP_STATUS_NOT_FOUND 404
121 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
123 /* Server errors 5xx. */
124 #define HTTP_STATUS_INTERNAL 500
125 #define HTTP_STATUS_NOT_IMPLEMENTED 501
126 #define HTTP_STATUS_BAD_GATEWAY 502
127 #define HTTP_STATUS_UNAVAILABLE 503
130 rel_none, rel_name, rel_value, rel_both
137 struct request_header {
139 enum rp release_policy;
141 int hcount, hcapacity;
144 /* Create a new, empty request. At least request_set_method must be
145 called before the request can be used. */
147 static struct request *
150 struct request *req = xnew0 (struct request);
152 req->headers = xnew_array (struct request_header, req->hcapacity);
156 /* Set the request's method and its arguments. METH should be a
157 literal string (or it should outlive the request) because it will
158 not be freed. ARG will be freed by request_free. */
161 request_set_method (struct request *req, const char *meth, char *arg)
167 /* Return the method string passed with the last call to
168 request_set_method. */
171 request_method (const struct request *req)
176 /* Free one header according to the release policy specified with
177 request_set_header. */
180 release_header (struct request_header *hdr)
182 switch (hdr->release_policy)
199 /* Set the request named NAME to VALUE. Specifically, this means that
200 a "NAME: VALUE\r\n" header line will be used in the request. If a
201 header with the same name previously existed in the request, its
202 value will be replaced by this one.
204 RELEASE_POLICY determines whether NAME and VALUE should be released
205 (freed) with request_free. Allowed values are:
207 - rel_none - don't free NAME or VALUE
208 - rel_name - free NAME when done
209 - rel_value - free VALUE when done
210 - rel_both - free both NAME and VALUE when done
212 Setting release policy is useful when arguments come from different
213 sources. For example:
215 // Don't free literal strings!
216 request_set_header (req, "Pragma", "no-cache", rel_none);
218 // Don't free a global variable, we'll need it later.
219 request_set_header (req, "Referer", opt.referer, rel_none);
221 // Value freshly allocated, free it when done.
222 request_set_header (req, "Range",
223 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
228 request_set_header (struct request *req, char *name, char *value,
229 enum rp release_policy)
231 struct request_header *hdr;
235 for (i = 0; i < req->hcount; i++)
237 hdr = &req->headers[i];
238 if (0 == strcasecmp (name, hdr->name))
240 /* Replace existing header. */
241 release_header (hdr);
244 hdr->release_policy = release_policy;
249 /* Install new header. */
251 if (req->hcount >= req->hcount)
253 req->hcapacity <<= 1;
254 req->headers = xrealloc (req->headers,
255 req->hcapacity * sizeof (struct request_header));
257 hdr = &req->headers[req->hcount++];
260 hdr->release_policy = release_policy;
263 /* Like request_set_header, but sets the whole header line, as
264 provided by the user using the `--header' option. For example,
265 request_set_user_header (req, "Foo: bar") works just like
266 request_set_header (req, "Foo", "bar"). */
269 request_set_user_header (struct request *req, const char *header)
272 const char *p = strchr (header, ':');
275 BOUNDED_TO_ALLOCA (header, p, name);
279 request_set_header (req, xstrdup (name), (char *) p, rel_name);
282 #define APPEND(p, str) do { \
283 int A_len = strlen (str); \
284 memcpy (p, str, A_len); \
288 /* Construct the request and write it to FD using fd_write. */
291 request_send (const struct request *req, int fd)
293 char *request_string, *p;
294 int i, size, write_error;
296 /* Count the request size. */
299 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
300 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
302 for (i = 0; i < req->hcount; i++)
304 struct request_header *hdr = &req->headers[i];
305 /* NAME ": " VALUE "\r\n" */
306 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
312 p = request_string = alloca_array (char, size);
314 /* Generate the request. */
316 APPEND (p, req->method); *p++ = ' ';
317 APPEND (p, req->arg); *p++ = ' ';
318 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
320 for (i = 0; i < req->hcount; i++)
322 struct request_header *hdr = &req->headers[i];
323 APPEND (p, hdr->name);
324 *p++ = ':', *p++ = ' ';
325 APPEND (p, hdr->value);
326 *p++ = '\r', *p++ = '\n';
329 *p++ = '\r', *p++ = '\n', *p++ = '\0';
330 assert (p - request_string == size);
334 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
336 /* Send the request to the server. */
338 write_error = fd_write (fd, request_string, size - 1, -1);
340 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
345 /* Release the resources used by REQ. */
348 request_free (struct request *req)
351 xfree_null (req->arg);
352 for (i = 0; i < req->hcount; i++)
353 release_header (&req->headers[i]);
354 xfree_null (req->headers);
358 /* Send the contents of FILE_NAME to SOCK/SSL. Make sure that exactly
359 PROMISED_SIZE bytes are sent over the wire -- if the file is
360 longer, read only that much; if the file is shorter, report an error. */
363 post_file (int sock, const char *file_name, wgint promised_size)
365 static char chunk[8192];
370 DEBUGP (("[writing POST file %s ... ", file_name));
372 fp = fopen (file_name, "rb");
375 while (!feof (fp) && written < promised_size)
378 int length = fread (chunk, 1, sizeof (chunk), fp);
381 towrite = MIN (promised_size - written, length);
382 write_error = fd_write (sock, chunk, towrite, -1);
392 /* If we've written less than was promised, report a (probably
393 nonsensical) error rather than break the promise. */
394 if (written < promised_size)
400 assert (written == promised_size);
401 DEBUGP (("done]\n"));
406 head_terminator (const char *hunk, int oldlen, int peeklen)
408 const char *start, *end;
410 /* If at first peek, verify whether HUNK starts with "HTTP". If
411 not, this is a HTTP/0.9 request and we must bail out without
413 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
419 start = hunk + oldlen - 4;
420 end = hunk + oldlen + peeklen;
422 for (; start < end - 1; start++)
429 if (start[1] == '\n')
435 /* Read the HTTP request head from FD and return it. The error
436 conditions are the same as with fd_read_hunk.
438 To support HTTP/0.9 responses, this function tries to make sure
439 that the data begins with "HTTP". If this is not the case, no data
440 is read and an empty request is returned, so that the remaining
441 data can be treated as body. */
444 fd_read_http_head (int fd)
446 return fd_read_hunk (fd, head_terminator, 512);
450 /* The response data. */
453 /* The array of pointers that indicate where each header starts.
454 For example, given this HTTP response:
461 The headers are located like this:
463 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
465 headers[0] headers[1] headers[2] headers[3]
467 I.e. headers[0] points to the beginning of the request,
468 headers[1] points to the end of the first header and the
469 beginning of the second one, etc. */
471 const char **headers;
474 /* Create a new response object from the text of the HTTP response,
475 available in HEAD. That text is automatically split into
476 constituent header lines for fast retrieval using
477 response_header_*. */
479 static struct response *
480 response_new (const char *head)
485 struct response *resp = xnew0 (struct response);
490 /* Empty head means that we're dealing with a headerless
491 (HTTP/0.9) response. In that case, don't set HEADERS at
496 /* Split HEAD into header lines, so that response_header_* functions
497 don't need to do this over and over again. */
503 DO_REALLOC (resp->headers, size, count + 1, const char *);
504 resp->headers[count++] = hdr;
506 /* Break upon encountering an empty line. */
507 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
510 /* Find the end of HDR, including continuations. */
513 const char *end = strchr (hdr, '\n');
519 while (*hdr == ' ' || *hdr == '\t');
521 DO_REALLOC (resp->headers, size, count + 1, const char *);
522 resp->headers[count++] = NULL;
527 /* Locate the header named NAME in the request data. If found, set
528 *BEGPTR to its starting, and *ENDPTR to its ending position, and
529 return 1. Otherwise return 0.
531 This function is used as a building block for response_header_copy
532 and response_header_strdup. */
535 response_header_bounds (const struct response *resp, const char *name,
536 const char **begptr, const char **endptr)
539 const char **headers = resp->headers;
542 if (!headers || !headers[1])
545 name_len = strlen (name);
547 for (i = 1; headers[i + 1]; i++)
549 const char *b = headers[i];
550 const char *e = headers[i + 1];
552 && b[name_len] == ':'
553 && 0 == strncasecmp (b, name, name_len))
556 while (b < e && ISSPACE (*b))
558 while (b < e && ISSPACE (e[-1]))
568 /* Copy the response header named NAME to buffer BUF, no longer than
569 BUFSIZE (BUFSIZE includes the terminating 0). If the header
570 exists, 1 is returned, otherwise 0. If there should be no limit on
571 the size of the header, use response_header_strdup instead.
573 If BUFSIZE is 0, no data is copied, but the boolean indication of
574 whether the header is present is still returned. */
577 response_header_copy (const struct response *resp, const char *name,
578 char *buf, int bufsize)
581 if (!response_header_bounds (resp, name, &b, &e))
585 int len = MIN (e - b, bufsize - 1);
586 memcpy (buf, b, len);
592 /* Return the value of header named NAME in RESP, allocated with
593 malloc. If such a header does not exist in RESP, return NULL. */
596 response_header_strdup (const struct response *resp, const char *name)
599 if (!response_header_bounds (resp, name, &b, &e))
601 return strdupdelim (b, e);
604 /* Parse the HTTP status line, which is of format:
606 HTTP-Version SP Status-Code SP Reason-Phrase
608 The function returns the status-code, or -1 if the status line
609 appears malformed. The pointer to "reason-phrase" message is
610 returned in *MESSAGE. */
613 response_status (const struct response *resp, char **message)
620 /* For a HTTP/0.9 response, assume status 200. */
622 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
626 p = resp->headers[0];
627 end = resp->headers[1];
633 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
637 /* Match the HTTP version. This is optional because Gnutella
638 servers have been reported to not specify HTTP version. */
639 if (p < end && *p == '/')
642 while (p < end && ISDIGIT (*p))
644 if (p < end && *p == '.')
646 while (p < end && ISDIGIT (*p))
650 while (p < end && ISSPACE (*p))
652 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
655 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
660 while (p < end && ISSPACE (*p))
662 while (p < end && ISSPACE (end[-1]))
664 *message = strdupdelim (p, end);
670 /* Release the resources used by RESP. */
673 response_free (struct response *resp)
675 xfree_null (resp->headers);
679 /* Print [b, e) to the log, omitting the trailing CRLF. */
682 print_server_response_1 (const char *prefix, const char *b, const char *e)
685 if (b < e && e[-1] == '\n')
687 if (b < e && e[-1] == '\r')
689 BOUNDED_TO_ALLOCA (b, e, ln);
690 logprintf (LOG_VERBOSE, "%s%s\n", prefix, ln);
693 /* Print the server response, line by line, omitting the trailing CR
694 characters, prefixed with PREFIX. */
697 print_server_response (const struct response *resp, const char *prefix)
702 for (i = 0; resp->headers[i + 1]; i++)
703 print_server_response_1 (prefix, resp->headers[i], resp->headers[i + 1]);
706 /* Parse the `Content-Range' header and extract the information it
707 contains. Returns 1 if successful, -1 otherwise. */
709 parse_content_range (const char *hdr, wgint *first_byte_ptr,
710 wgint *last_byte_ptr, wgint *entity_length_ptr)
714 /* Ancient versions of Netscape proxy server, presumably predating
715 rfc2068, sent out `Content-Range' without the "bytes"
717 if (!strncasecmp (hdr, "bytes", 5))
720 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
724 while (ISSPACE (*hdr))
731 for (num = 0; ISDIGIT (*hdr); hdr++)
732 num = 10 * num + (*hdr - '0');
733 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
735 *first_byte_ptr = num;
737 for (num = 0; ISDIGIT (*hdr); hdr++)
738 num = 10 * num + (*hdr - '0');
739 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
741 *last_byte_ptr = num;
743 for (num = 0; ISDIGIT (*hdr); hdr++)
744 num = 10 * num + (*hdr - '0');
745 *entity_length_ptr = num;
749 /* Read the body of the request, but don't store it anywhere and don't
750 display a progress gauge. This is useful for reading the error
751 responses whose bodies don't need to be displayed or logged, but
752 which need to be read anyway. */
755 skip_short_body (int fd, wgint contlen)
757 /* Skipping the body doesn't make sense if the content length is
758 unknown because, in that case, persistent connections cannot be
759 used. (#### This is not the case with HTTP/1.1 where they can
760 still be used with the magic of the "chunked" transfer!) */
763 DEBUGP (("Skipping %s bytes of body data... ", number_to_static_string (contlen)));
768 int ret = fd_read (fd, dlbuf, MIN (contlen, sizeof (dlbuf)), -1);
773 DEBUGP (("done.\n"));
776 /* Persistent connections. Currently, we cache the most recently used
777 connection as persistent, provided that the HTTP server agrees to
778 make it such. The persistence data is stored in the variables
779 below. Ideally, it should be possible to cache an arbitrary fixed
780 number of these connections. */
782 /* Whether a persistent connection is active. */
783 static int pconn_active;
786 /* The socket of the connection. */
789 /* Host and port of the currently active persistent connection. */
793 /* Whether a ssl handshake has occoured on this connection. */
797 /* Mark the persistent connection as invalid and free the resources it
798 uses. This is used by the CLOSE_* macros after they forcefully
799 close a registered persistent connection. */
802 invalidate_persistent (void)
804 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
806 fd_close (pconn.socket);
811 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
812 persistent. This will enable someone to use the same connection
813 later. In the context of HTTP, this must be called only AFTER the
814 response has been received and the server has promised that the
815 connection will remain alive.
817 If a previous connection was persistent, it is closed. */
820 register_persistent (const char *host, int port, int fd, int ssl)
824 if (pconn.socket == fd)
826 /* The connection FD is already registered. */
831 /* The old persistent connection is still active; close it
832 first. This situation arises whenever a persistent
833 connection exists, but we then connect to a different
834 host, and try to register a persistent connection to that
836 invalidate_persistent ();
842 pconn.host = xstrdup (host);
846 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
849 /* Return non-zero if a persistent connection is available for
850 connecting to HOST:PORT. */
853 persistent_available_p (const char *host, int port, int ssl,
854 int *host_lookup_failed)
856 /* First, check whether a persistent connection is active at all. */
860 /* If we want SSL and the last connection wasn't or vice versa,
861 don't use it. Checking for host and port is not enough because
862 HTTP and HTTPS can apparently coexist on the same port. */
863 if (ssl != pconn.ssl)
866 /* If we're not connecting to the same port, we're not interested. */
867 if (port != pconn.port)
870 /* If the host is the same, we're in business. If not, there is
871 still hope -- read below. */
872 if (0 != strcasecmp (host, pconn.host))
874 /* If pconn.socket is already talking to HOST, we needn't
875 reconnect. This happens often when both sites are virtual
876 hosts distinguished only by name and served by the same
877 network interface, and hence the same web server (possibly
878 set up by the ISP and serving many different web sites).
879 This admittedly non-standard optimization does not contradict
880 HTTP and works well with popular server software. */
884 struct address_list *al;
887 /* Don't try to talk to two different SSL sites over the same
888 secure connection! (Besides, it's not clear if name-based
889 virtual hosting is even possible with SSL.) */
892 /* If pconn.socket's peer is one of the IP addresses HOST
893 resolves to, pconn.socket is for all intents and purposes
894 already talking to HOST. */
896 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
898 /* Can't get the peer's address -- something must be very
899 wrong with the connection. */
900 invalidate_persistent ();
903 al = lookup_host (host, 0);
906 *host_lookup_failed = 1;
910 found = address_list_contains (al, &ip);
911 address_list_release (al);
916 /* The persistent connection's peer address was found among the
917 addresses HOST resolved to; therefore, pconn.sock is in fact
918 already talking to HOST -- no need to reconnect. */
921 /* Finally, check whether the connection is still open. This is
922 important because most server implement a liberal (short) timeout
923 on persistent connections. Wget can of course always reconnect
924 if the connection doesn't work out, but it's nicer to know in
925 advance. This test is a logical followup of the first test, but
926 is "expensive" and therefore placed at the end of the list. */
928 if (!test_socket_open (pconn.socket))
930 /* Oops, the socket is no longer open. Now that we know that,
931 let's invalidate the persistent connection before returning
933 invalidate_persistent ();
940 /* The idea behind these two CLOSE macros is to distinguish between
941 two cases: one when the job we've been doing is finished, and we
942 want to close the connection and leave, and two when something is
943 seriously wrong and we're closing the connection as part of
946 In case of keep_alive, CLOSE_FINISH should leave the connection
947 open, while CLOSE_INVALIDATE should still close it.
949 Note that the semantics of the flag `keep_alive' is "this
950 connection *will* be reused (the server has promised not to close
951 the connection once we're done)", while the semantics of
952 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
953 active, registered connection". */
955 #define CLOSE_FINISH(fd) do { \
958 if (pconn_active && (fd) == pconn.socket) \
959 invalidate_persistent (); \
968 #define CLOSE_INVALIDATE(fd) do { \
969 if (pconn_active && (fd) == pconn.socket) \
970 invalidate_persistent (); \
978 wgint len; /* received length */
979 wgint contlen; /* expected length */
980 wgint restval; /* the restart value */
981 int res; /* the result of last read */
982 char *newloc; /* new location (redirection) */
983 char *remote_time; /* remote time-stamp string */
984 char *error; /* textual HTTP error */
985 int statcode; /* status code */
986 wgint rd_size; /* amount of data read from socket */
987 double dltime; /* time it took to download the data */
988 const char *referer; /* value of the referer header. */
989 char **local_file; /* local file. */
993 free_hstat (struct http_stat *hs)
995 xfree_null (hs->newloc);
996 xfree_null (hs->remote_time);
997 xfree_null (hs->error);
999 /* Guard against being called twice. */
1001 hs->remote_time = NULL;
1005 static char *create_authorization_line PARAMS ((const char *, const char *,
1006 const char *, const char *,
1008 static char *basic_authentication_encode PARAMS ((const char *, const char *));
1009 static int known_authentication_scheme_p PARAMS ((const char *));
1011 time_t http_atotm PARAMS ((const char *));
1013 #define BEGINS_WITH(line, string_constant) \
1014 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1015 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1016 || !line[sizeof (string_constant) - 1]))
1018 /* Retrieve a document through HTTP protocol. It recognizes status
1019 code, and correctly handles redirections. It closes the network
1020 socket. If it receives an error from the functions below it, it
1021 will print it if there is enough information to do so (almost
1022 always), returning the error to the caller (i.e. http_loop).
1024 Various HTTP parameters are stored to hs.
1026 If PROXY is non-NULL, the connection will be made to the proxy
1027 server, and u->url will be requested. */
1029 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1031 struct request *req;
1034 char *user, *passwd;
1038 wgint contlen, contrange;
1045 /* Whether authorization has been already tried. */
1046 int auth_tried_already = 0;
1048 /* Whether our connection to the remote host is through SSL. */
1052 struct response *resp;
1056 /* Whether this connection will be kept alive after the HTTP request
1060 /* Whether keep-alive should be inhibited. */
1061 int inhibit_keep_alive = !opt.http_keep_alive || opt.ignore_length;
1063 /* Headers sent when using POST. */
1064 wgint post_data_size = 0;
1066 int host_lookup_failed = 0;
1069 if (u->scheme == SCHEME_HTTPS)
1071 /* Initialize the SSL context. After this has once been done,
1072 it becomes a no-op. */
1073 switch (ssl_init ())
1075 case SSLERRCTXCREATE:
1077 logprintf (LOG_NOTQUIET, _("Failed to set up an SSL context\n"));
1078 return SSLERRCTXCREATE;
1079 case SSLERRCERTFILE:
1080 /* try without certfile */
1081 logprintf (LOG_NOTQUIET,
1082 _("Failed to load certificates from %s\n"),
1084 logprintf (LOG_NOTQUIET,
1085 _("Trying without the specified certificate\n"));
1088 logprintf (LOG_NOTQUIET,
1089 _("Failed to get certificate key from %s\n"),
1091 logprintf (LOG_NOTQUIET,
1092 _("Trying without the specified certificate\n"));
1098 #endif /* HAVE_SSL */
1100 if (!(*dt & HEAD_ONLY))
1101 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1102 know the local filename so we can save to it. */
1103 assert (*hs->local_file != NULL);
1105 auth_tried_already = 0;
1107 /* Initialize certain elements of struct http_stat. */
1112 hs->remote_time = NULL;
1117 /* Prepare the request to send. */
1119 req = request_new ();
1121 const char *meth = "GET";
1122 if (*dt & HEAD_ONLY)
1124 else if (opt.post_file_name || opt.post_data)
1126 /* Use the full path, i.e. one that includes the leading slash and
1127 the query string. E.g. if u->path is "foo/bar" and u->query is
1128 "param=value", full_path will be "/foo/bar?param=value". */
1129 request_set_method (req, meth,
1130 proxy ? xstrdup (u->url) : url_full_path (u));
1133 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1134 if (*dt & SEND_NOCACHE)
1135 request_set_header (req, "Pragma", "no-cache", rel_none);
1137 request_set_header (req, "Range",
1138 aprintf ("bytes=%s-",
1139 number_to_static_string (hs->restval)),
1142 request_set_header (req, "User-Agent", opt.useragent, rel_none);
1144 request_set_header (req, "User-Agent",
1145 aprintf ("Wget/%s", version_string), rel_value);
1146 request_set_header (req, "Accept", "*/*", rel_none);
1148 /* Find the username and password for authentication. */
1151 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1152 user = user ? user : opt.http_user;
1153 passwd = passwd ? passwd : opt.http_passwd;
1157 /* We have the username and the password, but haven't tried
1158 any authorization yet. Let's see if the "Basic" method
1159 works. If not, we'll come back here and construct a
1160 proper authorization method with the right challenges.
1162 If we didn't employ this kind of logic, every URL that
1163 requires authorization would have to be processed twice,
1164 which is very suboptimal and generates a bunch of false
1165 "unauthorized" errors in the server log.
1167 #### But this logic also has a serious problem when used
1168 with stronger authentications: we *first* transmit the
1169 username and the password in clear text, and *then* attempt a
1170 stronger authentication scheme. That cannot be right! We
1171 are only fortunate that almost everyone still uses the
1172 `Basic' scheme anyway.
1174 There should be an option to prevent this from happening, for
1175 those who use strong authentication schemes and value their
1177 request_set_header (req, "Authorization",
1178 basic_authentication_encode (user, passwd),
1185 char *proxy_user, *proxy_passwd;
1186 /* For normal username and password, URL components override
1187 command-line/wgetrc parameters. With proxy
1188 authentication, it's the reverse, because proxy URLs are
1189 normally the "permanent" ones, so command-line args
1190 should take precedence. */
1191 if (opt.proxy_user && opt.proxy_passwd)
1193 proxy_user = opt.proxy_user;
1194 proxy_passwd = opt.proxy_passwd;
1198 proxy_user = proxy->user;
1199 proxy_passwd = proxy->passwd;
1201 /* #### This does not appear right. Can't the proxy request,
1202 say, `Digest' authentication? */
1203 if (proxy_user && proxy_passwd)
1204 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1206 /* If we're using a proxy, we will be connecting to the proxy
1210 /* Proxy authorization over SSL is handled below. */
1212 if (u->scheme != SCHEME_HTTPS)
1214 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1218 /* Whether we need to print the host header with braces around
1219 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1220 usual "Host: symbolic-name:1234". */
1221 int squares = strchr (u->host, ':') != NULL;
1222 if (u->port == scheme_default_port (u->scheme))
1223 request_set_header (req, "Host",
1224 aprintf (squares ? "[%s]" : "%s", u->host),
1227 request_set_header (req, "Host",
1228 aprintf (squares ? "[%s]:%d" : "%s:%d",
1233 if (!inhibit_keep_alive)
1234 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1237 request_set_header (req, "Cookie",
1238 cookie_header (wget_cookie_jar,
1239 u->host, u->port, u->path,
1241 u->scheme == SCHEME_HTTPS
1248 if (opt.post_data || opt.post_file_name)
1250 request_set_header (req, "Content-Type",
1251 "application/x-www-form-urlencoded", rel_none);
1253 post_data_size = strlen (opt.post_data);
1256 post_data_size = file_size (opt.post_file_name);
1257 if (post_data_size == -1)
1259 logprintf (LOG_NOTQUIET, "POST data file missing: %s\n",
1260 opt.post_file_name);
1264 request_set_header (req, "Content-Length",
1265 xstrdup (number_to_static_string (post_data_size)),
1269 /* Add the user headers. */
1270 if (opt.user_headers)
1273 for (i = 0; opt.user_headers[i]; i++)
1274 request_set_user_header (req, opt.user_headers[i]);
1278 /* We need to come back here when the initial attempt to retrieve
1279 without authorization header fails. (Expected to happen at least
1280 for the Digest authorization scheme.) */
1284 /* Establish the connection. */
1286 if (!inhibit_keep_alive)
1288 /* Look for a persistent connection to target host, unless a
1289 proxy is used. The exception is when SSL is in use, in which
1290 case the proxy is nothing but a passthrough to the target
1291 host, registered as a connection to the latter. */
1292 struct url *relevant = conn;
1294 if (u->scheme == SCHEME_HTTPS)
1298 if (persistent_available_p (relevant->host, relevant->port,
1300 relevant->scheme == SCHEME_HTTPS,
1304 &host_lookup_failed))
1306 sock = pconn.socket;
1307 using_ssl = pconn.ssl;
1308 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1309 pconn.host, pconn.port);
1310 DEBUGP (("Reusing fd %d.\n", sock));
1316 /* In its current implementation, persistent_available_p will
1317 look up conn->host in some cases. If that lookup failed, we
1318 don't need to bother with connect_to_host. */
1319 if (host_lookup_failed)
1322 sock = connect_to_host (conn->host, conn->port);
1326 return (retryable_socket_connect_error (errno)
1327 ? CONERROR : CONIMPOSSIBLE);
1330 if (proxy && u->scheme == SCHEME_HTTPS)
1332 /* When requesting SSL URLs through proxies, use the
1333 CONNECT method to request passthrough. */
1334 struct request *connreq = request_new ();
1335 request_set_method (connreq, "CONNECT",
1336 aprintf ("%s:%d", u->host, u->port));
1339 request_set_header (connreq, "Proxy-Authorization",
1340 proxyauth, rel_value);
1341 /* Now that PROXYAUTH is part of the CONNECT request,
1342 zero it out so we don't send proxy authorization with
1343 the regular request below. */
1347 write_error = request_send (connreq, sock);
1348 request_free (connreq);
1349 if (write_error < 0)
1351 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1353 CLOSE_INVALIDATE (sock);
1357 head = fd_read_http_head (sock);
1360 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1362 CLOSE_INVALIDATE (sock);
1371 DEBUGP (("proxy responded with: [%s]\n", head));
1373 resp = response_new (head);
1374 statcode = response_status (resp, &message);
1375 response_free (resp);
1376 if (statcode != 200)
1379 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1380 message ? message : "?");
1381 xfree_null (message);
1386 /* SOCK is now *really* connected to u->host, so update CONN
1387 to reflect this. That way register_persistent will
1388 register SOCK as being connected to u->host:u->port. */
1392 if (conn->scheme == SCHEME_HTTPS)
1394 if (!ssl_connect (sock))
1401 #endif /* HAVE_SSL */
1404 /* Send the request to server. */
1405 write_error = request_send (req, sock);
1407 if (write_error >= 0)
1411 DEBUGP (("[POST data: %s]\n", opt.post_data));
1412 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1414 else if (opt.post_file_name && post_data_size != 0)
1415 write_error = post_file (sock, opt.post_file_name, post_data_size);
1418 if (write_error < 0)
1420 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1422 CLOSE_INVALIDATE (sock);
1426 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1427 proxy ? "Proxy" : "HTTP");
1434 head = fd_read_http_head (sock);
1439 logputs (LOG_NOTQUIET, _("No data received.\n"));
1440 CLOSE_INVALIDATE (sock);
1446 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1448 CLOSE_INVALIDATE (sock);
1453 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1455 resp = response_new (head);
1457 /* Check for status line. */
1459 statcode = response_status (resp, &message);
1460 if (!opt.server_response)
1461 logprintf (LOG_VERBOSE, "%2d %s\n", statcode, message ? message : "");
1464 logprintf (LOG_VERBOSE, "\n");
1465 print_server_response (resp, " ");
1468 if (!opt.ignore_length
1469 && response_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1473 parsed = str_to_wgint (hdrval, NULL, 10);
1474 if (parsed == WGINT_MAX && errno == ERANGE)
1476 #### If Content-Length is out of range, it most likely
1477 means that the file is larger than 2G and that we're
1478 compiled without LFS. In that case we should probably
1479 refuse to even attempt to download the file. */
1485 /* Check for keep-alive related responses. */
1486 if (!inhibit_keep_alive && contlen != -1)
1488 if (response_header_copy (resp, "Keep-Alive", NULL, 0))
1490 else if (response_header_copy (resp, "Connection", hdrval,
1493 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1498 /* The server has promised that it will not close the connection
1499 when we're done. This means that we can register it. */
1500 register_persistent (conn->host, conn->port, sock, using_ssl);
1502 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1504 /* Authorization is required. */
1505 skip_short_body (sock, contlen);
1506 CLOSE_FINISH (sock);
1507 if (auth_tried_already || !(user && passwd))
1509 /* If we have tried it already, then there is not point
1511 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1515 char *www_authenticate = response_header_strdup (resp,
1516 "WWW-Authenticate");
1517 /* If the authentication scheme is unknown or if it's the
1518 "Basic" authentication (which we try by default), there's
1519 no sense in retrying. */
1520 if (!www_authenticate
1521 || !known_authentication_scheme_p (www_authenticate)
1522 || BEGINS_WITH (www_authenticate, "Basic"))
1524 xfree_null (www_authenticate);
1525 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1530 auth_tried_already = 1;
1531 pth = url_full_path (u);
1532 request_set_header (req, "Authorization",
1533 create_authorization_line (www_authenticate,
1535 request_method (req),
1539 xfree (www_authenticate);
1540 goto retry_with_auth;
1548 hs->statcode = statcode;
1550 hs->error = xstrdup (_("Malformed status line"));
1552 hs->error = xstrdup (_("(no description)"));
1554 hs->error = xstrdup (message);
1556 type = response_header_strdup (resp, "Content-Type");
1559 char *tmp = strchr (type, ';');
1562 while (tmp > type && ISSPACE (tmp[-1]))
1567 hs->newloc = response_header_strdup (resp, "Location");
1568 hs->remote_time = response_header_strdup (resp, "Last-Modified");
1570 char *set_cookie = response_header_strdup (resp, "Set-Cookie");
1573 /* The jar should have been created by now. */
1574 assert (wget_cookie_jar != NULL);
1575 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
1580 if (response_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1582 wgint first_byte_pos, last_byte_pos, entity_length;
1583 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1585 contrange = first_byte_pos;
1587 response_free (resp);
1589 /* 20x responses are counted among successful by default. */
1590 if (H_20X (statcode))
1593 /* Return if redirected. */
1594 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1596 /* RFC2068 says that in case of the 300 (multiple choices)
1597 response, the server can output a preferred URL through
1598 `Location' header; otherwise, the request should be treated
1599 like GET. So, if the location is set, it will be a
1600 redirection; otherwise, just proceed normally. */
1601 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1605 logprintf (LOG_VERBOSE,
1606 _("Location: %s%s\n"),
1607 hs->newloc ? hs->newloc : _("unspecified"),
1608 hs->newloc ? _(" [following]") : "");
1610 skip_short_body (sock, contlen);
1611 CLOSE_FINISH (sock);
1617 /* If content-type is not given, assume text/html. This is because
1618 of the multitude of broken CGI's that "forget" to generate the
1621 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1622 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1627 if (opt.html_extension && (*dt & TEXTHTML))
1628 /* -E / --html-extension / html_extension = on was specified, and this is a
1629 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1630 already the file's suffix, tack on ".html". */
1632 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1634 if (last_period_in_local_filename == NULL
1635 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1636 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1638 size_t local_filename_len = strlen(*hs->local_file);
1640 *hs->local_file = xrealloc(*hs->local_file,
1641 local_filename_len + sizeof(".html"));
1642 strcpy(*hs->local_file + local_filename_len, ".html");
1644 *dt |= ADDED_HTML_EXTENSION;
1648 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1650 /* If `-c' is in use and the file has been fully downloaded (or
1651 the remote file has shrunk), Wget effectively requests bytes
1652 after the end of file and the server response with 416. */
1653 logputs (LOG_VERBOSE, _("\
1654 \n The file is already fully retrieved; nothing to do.\n\n"));
1655 /* In case the caller inspects. */
1658 /* Mark as successfully retrieved. */
1661 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1662 might be more bytes in the body. */
1663 return RETRUNNEEDED;
1665 if ((contrange != 0 && contrange != hs->restval)
1666 || (H_PARTIAL (statcode) && !contrange))
1668 /* The Range request was somehow misunderstood by the server.
1671 CLOSE_INVALIDATE (sock);
1674 hs->contlen = contlen + contrange;
1680 /* No need to print this output if the body won't be
1681 downloaded at all, or if the original server response is
1683 logputs (LOG_VERBOSE, _("Length: "));
1686 logputs (LOG_VERBOSE, legible (contlen + contrange));
1688 logprintf (LOG_VERBOSE, _(" (%s to go)"), legible (contlen));
1691 logputs (LOG_VERBOSE,
1692 opt.ignore_length ? _("ignored") : _("unspecified"));
1694 logprintf (LOG_VERBOSE, " [%s]\n", type);
1696 logputs (LOG_VERBOSE, "\n");
1700 type = NULL; /* We don't need it any more. */
1702 /* Return if we have no intention of further downloading. */
1703 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1705 /* In case the caller cares to look... */
1709 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1710 servers not to send body in response to a HEAD request. If
1711 you encounter such a server (more likely a broken CGI), use
1712 `--no-http-keep-alive'. */
1713 CLOSE_FINISH (sock);
1714 return RETRFINISHED;
1717 /* Open the local file. */
1720 mkalldirs (*hs->local_file);
1722 rotate_backups (*hs->local_file);
1723 fp = fopen (*hs->local_file, hs->restval ? "ab" : "wb");
1726 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1727 CLOSE_INVALIDATE (sock);
1734 /* #### This confuses the timestamping code that checks for file
1735 size. Maybe we should save some additional information? */
1736 if (opt.save_headers)
1737 fwrite (head, 1, strlen (head), fp);
1739 /* Download the request body. */
1742 flags |= rb_read_exactly;
1743 if (hs->restval > 0 && contrange == 0)
1744 /* If the server ignored our range request, instruct fd_read_body
1745 to skip the first RESTVAL bytes of body. */
1746 flags |= rb_skip_startpos;
1747 hs->len = hs->restval;
1749 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1750 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1754 CLOSE_FINISH (sock);
1756 CLOSE_INVALIDATE (sock);
1759 /* Close or flush the file. We have to be careful to check for
1760 error here. Checking the result of fwrite() is not enough --
1761 errors could go unnoticed! */
1764 flush_res = fclose (fp);
1766 flush_res = fflush (fp);
1767 if (flush_res == EOF)
1772 return RETRFINISHED;
1775 /* The genuine HTTP loop! This is the part where the retrieval is
1776 retried, and retried, and retried, and... */
1778 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1779 int *dt, struct url *proxy)
1782 int use_ts, got_head = 0; /* time-stamping info */
1783 char *filename_plus_orig_suffix;
1784 char *local_filename = NULL;
1785 char *tms, *locf, *tmrate;
1787 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1788 wgint local_size = 0; /* the size of the local file */
1789 size_t filename_len;
1790 struct http_stat hstat; /* HTTP status */
1794 /* This used to be done in main(), but it's a better idea to do it
1795 here so that we don't go through the hoops if we're just using
1799 if (!wget_cookie_jar)
1800 wget_cookie_jar = cookie_jar_new ();
1801 if (opt.cookies_input && !cookies_loaded_p)
1803 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1804 cookies_loaded_p = 1;
1810 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1811 has_wildcards_p because it would also warn on `?', and we know that
1812 shows up in CGI paths a *lot*. */
1813 if (strchr (u->url, '*'))
1814 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1818 /* Determine the local filename. */
1819 if (local_file && *local_file)
1820 hstat.local_file = local_file;
1821 else if (local_file && !opt.output_document)
1823 *local_file = url_file_name (u);
1824 hstat.local_file = local_file;
1828 dummy = url_file_name (u);
1829 hstat.local_file = &dummy;
1830 /* be honest about where we will save the file */
1831 if (local_file && opt.output_document)
1832 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
1835 if (!opt.output_document)
1836 locf = *hstat.local_file;
1838 locf = opt.output_document;
1840 hstat.referer = referer;
1842 filename_len = strlen (*hstat.local_file);
1843 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1845 if (opt.noclobber && file_exists_p (*hstat.local_file))
1847 /* If opt.noclobber is turned on and file already exists, do not
1848 retrieve the file */
1849 logprintf (LOG_VERBOSE, _("\
1850 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1851 /* If the file is there, we suppose it's retrieved OK. */
1854 /* #### Bogusness alert. */
1855 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1856 if (has_html_suffix_p (*hstat.local_file))
1864 if (opt.timestamping)
1866 int local_dot_orig_file_exists = 0;
1868 if (opt.backup_converted)
1869 /* If -K is specified, we'll act on the assumption that it was specified
1870 last time these files were downloaded as well, and instead of just
1871 comparing local file X against server file X, we'll compare local
1872 file X.orig (if extant, else X) against server file X. If -K
1873 _wasn't_ specified last time, or the server contains files called
1874 *.orig, -N will be back to not operating correctly with -k. */
1876 /* Would a single s[n]printf() call be faster? --dan
1878 Definitely not. sprintf() is horribly slow. It's a
1879 different question whether the difference between the two
1880 affects a program. Usually I'd say "no", but at one
1881 point I profiled Wget, and found that a measurable and
1882 non-negligible amount of time was lost calling sprintf()
1883 in url.c. Replacing sprintf with inline calls to
1884 strcpy() and number_to_string() made a difference.
1886 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1887 memcpy (filename_plus_orig_suffix + filename_len,
1888 ".orig", sizeof (".orig"));
1890 /* Try to stat() the .orig file. */
1891 if (stat (filename_plus_orig_suffix, &st) == 0)
1893 local_dot_orig_file_exists = 1;
1894 local_filename = filename_plus_orig_suffix;
1898 if (!local_dot_orig_file_exists)
1899 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1900 if (stat (*hstat.local_file, &st) == 0)
1901 local_filename = *hstat.local_file;
1903 if (local_filename != NULL)
1904 /* There was a local file, so we'll check later to see if the version
1905 the server has is the same version we already have, allowing us to
1911 /* Modification time granularity is 2 seconds for Windows, so
1912 increase local time by 1 second for later comparison. */
1915 local_size = st.st_size;
1919 /* Reset the counter. */
1925 /* Increment the pass counter. */
1927 sleep_between_retrievals (count);
1928 /* Get the current time string. */
1929 tms = time_str (NULL);
1930 /* Print fetch message, if opt.verbose. */
1933 char *hurl = url_string (u, 1);
1937 sprintf (tmp, _("(try:%2d)"), count);
1938 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1939 tms, hurl, tmp, locf);
1941 ws_changetitle (hurl);
1946 /* Default document type is empty. However, if spider mode is
1947 on or time-stamping is employed, HEAD_ONLY commands is
1948 encoded within *dt. */
1949 if (opt.spider || (use_ts && !got_head))
1954 /* Decide whether or not to restart. */
1957 hstat.restval = hstat.len; /* continue where we left off */
1958 else if (opt.always_rest
1959 && stat (locf, &st) == 0
1960 && S_ISREG (st.st_mode))
1961 hstat.restval = st.st_size;
1963 /* Decide whether to send the no-cache directive. We send it in
1965 a) we're using a proxy, and we're past our first retrieval.
1966 Some proxies are notorious for caching incomplete data, so
1967 we require a fresh get.
1968 b) caching is explicitly inhibited. */
1969 if ((proxy && count > 1) /* a */
1970 || !opt.allow_cache /* b */
1972 *dt |= SEND_NOCACHE;
1974 *dt &= ~SEND_NOCACHE;
1976 /* Try fetching the document, or at least its head. */
1977 err = gethttp (u, &hstat, dt, proxy);
1979 /* It's unfortunate that wget determines the local filename before finding
1980 out the Content-Type of the file. Barring a major restructuring of the
1981 code, we need to re-set locf here, since gethttp() may have xrealloc()d
1982 *hstat.local_file to tack on ".html". */
1983 if (!opt.output_document)
1984 locf = *hstat.local_file;
1987 tms = time_str (NULL);
1988 /* Get the new location (with or without the redirection). */
1990 *newloc = xstrdup (hstat.newloc);
1993 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
1994 case CONERROR: case READERR: case WRITEFAILED:
1996 /* Non-fatal errors continue executing the loop, which will
1997 bring them to "while" statement at the end, to judge
1998 whether the number of tries was exceeded. */
1999 free_hstat (&hstat);
2000 printwhat (count, opt.ntry);
2003 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2004 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
2005 /* Fatal errors just return from the function. */
2006 free_hstat (&hstat);
2010 case FWRITEERR: case FOPENERR:
2011 /* Another fatal error. */
2012 logputs (LOG_VERBOSE, "\n");
2013 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2014 *hstat.local_file, strerror (errno));
2015 free_hstat (&hstat);
2020 /* Another fatal error. */
2021 logputs (LOG_VERBOSE, "\n");
2022 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2023 free_hstat (&hstat);
2028 /* Return the new location to the caller. */
2031 logprintf (LOG_NOTQUIET,
2032 _("ERROR: Redirection (%d) without location.\n"),
2034 free_hstat (&hstat);
2038 free_hstat (&hstat);
2043 /* The file was already fully retrieved. */
2044 free_hstat (&hstat);
2049 /* Deal with you later. */
2052 /* All possibilities should have been exhausted. */
2055 if (!(*dt & RETROKF))
2059 /* #### Ugly ugly ugly! */
2060 char *hurl = url_string (u, 1);
2061 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2064 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2065 tms, hstat.statcode, hstat.error);
2066 logputs (LOG_VERBOSE, "\n");
2067 free_hstat (&hstat);
2072 /* Did we get the time-stamp? */
2075 if (opt.timestamping && !hstat.remote_time)
2077 logputs (LOG_NOTQUIET, _("\
2078 Last-modified header missing -- time-stamps turned off.\n"));
2080 else if (hstat.remote_time)
2082 /* Convert the date-string into struct tm. */
2083 tmr = http_atotm (hstat.remote_time);
2084 if (tmr == (time_t) (-1))
2085 logputs (LOG_VERBOSE, _("\
2086 Last-modified header invalid -- time-stamp ignored.\n"));
2090 /* The time-stamping section. */
2095 use_ts = 0; /* no more time-stamping */
2096 count = 0; /* the retrieve count for HEAD is
2098 if (hstat.remote_time && tmr != (time_t) (-1))
2100 /* Now time-stamping can be used validly. Time-stamping
2101 means that if the sizes of the local and remote file
2102 match, and local file is newer than the remote file,
2103 it will not be retrieved. Otherwise, the normal
2104 download procedure is resumed. */
2106 (hstat.contlen == -1 || local_size == hstat.contlen))
2108 logprintf (LOG_VERBOSE, _("\
2109 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2111 free_hstat (&hstat);
2115 else if (tml >= tmr)
2116 logprintf (LOG_VERBOSE, _("\
2117 The sizes do not match (local %s) -- retrieving.\n"),
2118 number_to_static_string (local_size));
2120 logputs (LOG_VERBOSE,
2121 _("Remote file is newer, retrieving.\n"));
2123 free_hstat (&hstat);
2126 if ((tmr != (time_t) (-1))
2128 && ((hstat.len == hstat.contlen) ||
2129 ((hstat.res == 0) &&
2130 ((hstat.contlen == -1) ||
2131 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2133 /* #### This code repeats in http.c and ftp.c. Move it to a
2135 const char *fl = NULL;
2136 if (opt.output_document)
2138 if (output_stream_regular)
2139 fl = opt.output_document;
2142 fl = *hstat.local_file;
2146 /* End of time-stamping section. */
2150 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode, hstat.error);
2155 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2157 if (hstat.len == hstat.contlen)
2161 logprintf (LOG_VERBOSE,
2162 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2164 number_to_static_string (hstat.len),
2165 number_to_static_string (hstat.contlen));
2166 logprintf (LOG_NONVERBOSE,
2167 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2169 number_to_static_string (hstat.len),
2170 number_to_static_string (hstat.contlen),
2174 total_downloaded_bytes += hstat.len;
2176 /* Remember that we downloaded the file for later ".orig" code. */
2177 if (*dt & ADDED_HTML_EXTENSION)
2178 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2180 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2182 free_hstat (&hstat);
2186 else if (hstat.res == 0) /* No read error */
2188 if (hstat.contlen == -1) /* We don't know how much we were supposed
2189 to get, so assume we succeeded. */
2193 logprintf (LOG_VERBOSE,
2194 _("%s (%s) - `%s' saved [%s]\n\n"),
2196 number_to_static_string (hstat.len));
2197 logprintf (LOG_NONVERBOSE,
2198 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2199 tms, u->url, number_to_static_string (hstat.len),
2203 total_downloaded_bytes += hstat.len;
2205 /* Remember that we downloaded the file for later ".orig" code. */
2206 if (*dt & ADDED_HTML_EXTENSION)
2207 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2209 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2211 free_hstat (&hstat);
2215 else if (hstat.len < hstat.contlen) /* meaning we lost the
2216 connection too soon */
2218 logprintf (LOG_VERBOSE,
2219 _("%s (%s) - Connection closed at byte %s. "),
2220 tms, tmrate, number_to_static_string (hstat.len));
2221 printwhat (count, opt.ntry);
2222 free_hstat (&hstat);
2225 else if (!opt.kill_longer) /* meaning we got more than expected */
2227 logprintf (LOG_VERBOSE,
2228 _("%s (%s) - `%s' saved [%s/%s])\n\n"),
2230 number_to_static_string (hstat.len),
2231 number_to_static_string (hstat.contlen));
2232 logprintf (LOG_NONVERBOSE,
2233 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2235 number_to_static_string (hstat.len),
2236 number_to_static_string (hstat.contlen),
2239 total_downloaded_bytes += hstat.len;
2241 /* Remember that we downloaded the file for later ".orig" code. */
2242 if (*dt & ADDED_HTML_EXTENSION)
2243 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2245 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2247 free_hstat (&hstat);
2251 else /* the same, but not accepted */
2253 logprintf (LOG_VERBOSE,
2254 _("%s (%s) - Connection closed at byte %s/%s. "),
2256 number_to_static_string (hstat.len),
2257 number_to_static_string (hstat.contlen));
2258 printwhat (count, opt.ntry);
2259 free_hstat (&hstat);
2263 else /* now hstat.res can only be -1 */
2265 if (hstat.contlen == -1)
2267 logprintf (LOG_VERBOSE,
2268 _("%s (%s) - Read error at byte %s (%s)."),
2269 tms, tmrate, number_to_static_string (hstat.len),
2271 printwhat (count, opt.ntry);
2272 free_hstat (&hstat);
2275 else /* hstat.res == -1 and contlen is given */
2277 logprintf (LOG_VERBOSE,
2278 _("%s (%s) - Read error at byte %s/%s (%s). "),
2280 number_to_static_string (hstat.len),
2281 number_to_static_string (hstat.contlen),
2283 printwhat (count, opt.ntry);
2284 free_hstat (&hstat);
2291 while (!opt.ntry || (count < opt.ntry));
2295 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2296 than local timezone.
2298 mktime is similar but assumes struct tm, also known as the
2299 "broken-down" form of time, is in local time zone. mktime_from_utc
2300 uses mktime to make the conversion understanding that an offset
2301 will be introduced by the local time assumption.
2303 mktime_from_utc then measures the introduced offset by applying
2304 gmtime to the initial result and applying mktime to the resulting
2305 "broken-down" form. The difference between the two mktime results
2306 is the measured offset which is then subtracted from the initial
2307 mktime result to yield a calendar time which is the value returned.
2309 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2310 consistent offset (the non DST offset) since tm and tm+o might be
2311 on opposite sides of a DST change.
2313 Some implementations of mktime return -1 for the nonexistent
2314 localtime hour at the beginning of DST. In this event, use
2315 mktime(tm - 1hr) + 3600.
2319 gmtime(t+o) --> tm+o
2320 mktime(tm+o) --> t+2o
2321 t+o - (t+2o - t+o) = t
2323 Note that glibc contains a function of the same purpose named
2324 `timegm' (reverse of gmtime). But obviously, it is not universally
2325 available, and unfortunately it is not straightforwardly
2326 extractable for use here. Perhaps configure should detect timegm
2327 and use it where available.
2329 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2330 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2331 Further improved by Roger with assistance from Edward J. Sabol
2332 based on input by Jamie Zawinski. */
2335 mktime_from_utc (struct tm *t)
2346 return -1; /* can't deal with output from strptime */
2357 return -1; /* can't deal with output from gmtime */
2360 return (tl - (tb - tl));
2363 /* Check whether the result of strptime() indicates success.
2364 strptime() returns the pointer to how far it got to in the string.
2365 The processing has been successful if the string is at `GMT' or
2366 `+X', or at the end of the string.
2368 In extended regexp parlance, the function returns 1 if P matches
2369 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2370 can return) is considered a failure and 0 is returned. */
2372 check_end (const char *p)
2376 while (ISSPACE (*p))
2379 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2380 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2386 /* Convert the textual specification of time in TIME_STRING to the
2387 number of seconds since the Epoch.
2389 TIME_STRING can be in any of the three formats RFC2068 allows the
2390 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2391 Timezones are ignored, and should be GMT.
2393 Return the computed time_t representation, or -1 if the conversion
2396 This function uses strptime with various string formats for parsing
2397 TIME_STRING. This results in a parser that is not as lenient in
2398 interpreting TIME_STRING as I would like it to be. Being based on
2399 strptime, it always allows shortened months, one-digit days, etc.,
2400 but due to the multitude of formats in which time can be
2401 represented, an ideal HTTP time parser would be even more
2402 forgiving. It should completely ignore things like week days and
2403 concentrate only on the various forms of representing years,
2404 months, days, hours, minutes, and seconds. For example, it would
2405 be nice if it accepted ISO 8601 out of the box.
2407 I've investigated free and PD code for this purpose, but none was
2408 usable. getdate was big and unwieldy, and had potential copyright
2409 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2410 distributed with phttpd, is excellent, but we cannot use it because
2411 it is not assigned to the FSF. So I stuck it with strptime. */
2414 http_atotm (const char *time_string)
2416 /* NOTE: Solaris strptime man page claims that %n and %t match white
2417 space, but that's not universally available. Instead, we simply
2418 use ` ' to mean "skip all WS", which works under all strptime
2419 implementations I've tested. */
2421 static const char *time_formats[] = {
2422 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2423 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2424 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2425 (google.com uses this for their cookies.) */
2426 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2432 /* According to Roger Beeman, we need to initialize tm_isdst, since
2433 strptime won't do it. */
2436 /* Note that under foreign locales Solaris strptime() fails to
2437 recognize English dates, which renders this function useless. We
2438 solve this by being careful not to affect LC_TIME when
2439 initializing locale.
2441 Another solution would be to temporarily set locale to C, invoke
2442 strptime(), and restore it back. This is slow and dirty,
2443 however, and locale support other than LC_MESSAGES can mess other
2444 things, so I rather chose to stick with just setting LC_MESSAGES.
2446 GNU strptime does not have this problem because it recognizes
2447 both international and local dates. */
2449 for (i = 0; i < countof (time_formats); i++)
2450 if (check_end (strptime (time_string, time_formats[i], &t)))
2451 return mktime_from_utc (&t);
2453 /* All formats have failed. */
2457 /* Authorization support: We support two authorization schemes:
2459 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2461 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2462 consisting of answering to the server's challenge with the proper
2465 /* How many bytes it will take to store LEN bytes in base64. */
2466 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2468 /* Encode the string S of length LENGTH to base64 format and place it
2469 to STORE. STORE will be 0-terminated, and must point to a writable
2470 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2472 base64_encode (const char *s, char *store, int length)
2474 /* Conversion table. */
2475 static char tbl[64] = {
2476 'A','B','C','D','E','F','G','H',
2477 'I','J','K','L','M','N','O','P',
2478 'Q','R','S','T','U','V','W','X',
2479 'Y','Z','a','b','c','d','e','f',
2480 'g','h','i','j','k','l','m','n',
2481 'o','p','q','r','s','t','u','v',
2482 'w','x','y','z','0','1','2','3',
2483 '4','5','6','7','8','9','+','/'
2486 unsigned char *p = (unsigned char *)store;
2488 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2489 for (i = 0; i < length; i += 3)
2491 *p++ = tbl[s[0] >> 2];
2492 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2493 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2494 *p++ = tbl[s[2] & 0x3f];
2497 /* Pad the result if necessary... */
2498 if (i == length + 1)
2500 else if (i == length + 2)
2501 *(p - 1) = *(p - 2) = '=';
2502 /* ...and zero-terminate it. */
2506 /* Create the authentication header contents for the `Basic' scheme.
2507 This is done by encoding the string `USER:PASS' in base64 and
2508 prepending `HEADER: Basic ' to it. */
2510 basic_authentication_encode (const char *user, const char *passwd)
2512 char *t1, *t2, *res;
2513 int len1 = strlen (user) + 1 + strlen (passwd);
2514 int len2 = BASE64_LENGTH (len1);
2516 t1 = (char *)alloca (len1 + 1);
2517 sprintf (t1, "%s:%s", user, passwd);
2519 t2 = (char *)alloca (len2 + 1);
2520 base64_encode (t1, t2, len1);
2522 res = (char *)xmalloc (6 + len2 + 1);
2523 sprintf (res, "Basic %s", t2);
2528 #define SKIP_WS(x) do { \
2529 while (ISSPACE (*(x))) \
2534 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2535 of a field in such a header. If the field is the one specified by
2536 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2537 digest authorization code), extract its value in the (char*)
2538 variable pointed by RET. Returns negative on a malformed header,
2539 or number of bytes that have been parsed by this call. */
2541 extract_header_attr (const char *au, const char *attr_name, char **ret)
2543 const char *cp, *ep;
2547 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2549 cp += strlen (attr_name);
2562 for (ep = cp; *ep && *ep != '\"'; ep++)
2567 *ret = strdupdelim (cp, ep);
2574 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2575 an array of 16 bytes containing the hash keys, and BUF should be a
2576 buffer of 33 writable characters (32 for hex digits plus one for
2577 zero termination). */
2579 dump_hash (unsigned char *buf, const unsigned char *hash)
2583 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2585 *buf++ = XNUM_TO_digit (*hash >> 4);
2586 *buf++ = XNUM_TO_digit (*hash & 0xf);
2591 /* Take the line apart to find the challenge, and compose a digest
2592 authorization header. See RFC2069 section 2.1.2. */
2594 digest_authentication_encode (const char *au, const char *user,
2595 const char *passwd, const char *method,
2598 static char *realm, *opaque, *nonce;
2603 { "realm", &realm },
2604 { "opaque", &opaque },
2609 realm = opaque = nonce = NULL;
2611 au += 6; /* skip over `Digest' */
2617 for (i = 0; i < countof (options); i++)
2619 int skip = extract_header_attr (au, options[i].name,
2620 options[i].variable);
2624 xfree_null (opaque);
2634 if (i == countof (options))
2636 while (*au && *au != '=')
2644 while (*au && *au != '\"')
2651 while (*au && *au != ',')
2656 if (!realm || !nonce || !user || !passwd || !path || !method)
2659 xfree_null (opaque);
2664 /* Calculate the digest value. */
2666 ALLOCA_MD5_CONTEXT (ctx);
2667 unsigned char hash[MD5_HASHLEN];
2668 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2669 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2671 /* A1BUF = H(user ":" realm ":" password) */
2673 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2674 gen_md5_update ((unsigned char *)":", 1, ctx);
2675 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2676 gen_md5_update ((unsigned char *)":", 1, ctx);
2677 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2678 gen_md5_finish (ctx, hash);
2679 dump_hash (a1buf, hash);
2681 /* A2BUF = H(method ":" path) */
2683 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2684 gen_md5_update ((unsigned char *)":", 1, ctx);
2685 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2686 gen_md5_finish (ctx, hash);
2687 dump_hash (a2buf, hash);
2689 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2691 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2692 gen_md5_update ((unsigned char *)":", 1, ctx);
2693 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2694 gen_md5_update ((unsigned char *)":", 1, ctx);
2695 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2696 gen_md5_finish (ctx, hash);
2697 dump_hash (response_digest, hash);
2699 res = (char*) xmalloc (strlen (user)
2704 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2705 + (opaque ? strlen (opaque) : 0)
2707 sprintf (res, "Digest \
2708 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2709 user, realm, nonce, path, response_digest);
2712 char *p = res + strlen (res);
2713 strcat (p, ", opaque=\"");
2720 #endif /* USE_DIGEST */
2723 #define BEGINS_WITH(line, string_constant) \
2724 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2725 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2726 || !line[sizeof (string_constant) - 1]))
2729 known_authentication_scheme_p (const char *au)
2731 return BEGINS_WITH (au, "Basic")
2732 || BEGINS_WITH (au, "Digest")
2733 || BEGINS_WITH (au, "NTLM");
2738 /* Create the HTTP authorization request header. When the
2739 `WWW-Authenticate' response header is seen, according to the
2740 authorization scheme specified in that header (`Basic' and `Digest'
2741 are supported by the current implementation), produce an
2742 appropriate HTTP authorization request header. */
2744 create_authorization_line (const char *au, const char *user,
2745 const char *passwd, const char *method,
2748 if (0 == strncasecmp (au, "Basic", 5))
2749 return basic_authentication_encode (user, passwd);
2751 if (0 == strncasecmp (au, "Digest", 6))
2752 return digest_authentication_encode (au, user, passwd, method, path);
2753 #endif /* USE_DIGEST */