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, escnonprint (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 escnonprint (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 ? escnonprint (message) : "?");
1381 xfree_null (message);
1384 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,
1462 message ? escnonprint (message) : "");
1465 logprintf (LOG_VERBOSE, "\n");
1466 print_server_response (resp, " ");
1469 if (!opt.ignore_length
1470 && response_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1474 parsed = str_to_wgint (hdrval, NULL, 10);
1475 if (parsed == WGINT_MAX && errno == ERANGE)
1477 #### If Content-Length is out of range, it most likely
1478 means that the file is larger than 2G and that we're
1479 compiled without LFS. In that case we should probably
1480 refuse to even attempt to download the file. */
1486 /* Check for keep-alive related responses. */
1487 if (!inhibit_keep_alive && contlen != -1)
1489 if (response_header_copy (resp, "Keep-Alive", NULL, 0))
1491 else if (response_header_copy (resp, "Connection", hdrval,
1494 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1499 /* The server has promised that it will not close the connection
1500 when we're done. This means that we can register it. */
1501 register_persistent (conn->host, conn->port, sock, using_ssl);
1503 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1505 /* Authorization is required. */
1506 skip_short_body (sock, contlen);
1507 CLOSE_FINISH (sock);
1508 if (auth_tried_already || !(user && passwd))
1510 /* If we have tried it already, then there is not point
1512 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1516 char *www_authenticate = response_header_strdup (resp,
1517 "WWW-Authenticate");
1518 /* If the authentication scheme is unknown or if it's the
1519 "Basic" authentication (which we try by default), there's
1520 no sense in retrying. */
1521 if (!www_authenticate
1522 || !known_authentication_scheme_p (www_authenticate)
1523 || BEGINS_WITH (www_authenticate, "Basic"))
1525 xfree_null (www_authenticate);
1526 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1531 auth_tried_already = 1;
1532 pth = url_full_path (u);
1533 request_set_header (req, "Authorization",
1534 create_authorization_line (www_authenticate,
1536 request_method (req),
1540 xfree (www_authenticate);
1541 goto retry_with_auth;
1549 hs->statcode = statcode;
1551 hs->error = xstrdup (_("Malformed status line"));
1553 hs->error = xstrdup (_("(no description)"));
1555 hs->error = xstrdup (message);
1557 type = response_header_strdup (resp, "Content-Type");
1560 char *tmp = strchr (type, ';');
1563 while (tmp > type && ISSPACE (tmp[-1]))
1568 hs->newloc = response_header_strdup (resp, "Location");
1569 hs->remote_time = response_header_strdup (resp, "Last-Modified");
1571 char *set_cookie = response_header_strdup (resp, "Set-Cookie");
1574 /* The jar should have been created by now. */
1575 assert (wget_cookie_jar != NULL);
1576 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port, u->path,
1581 if (response_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1583 wgint first_byte_pos, last_byte_pos, entity_length;
1584 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1586 contrange = first_byte_pos;
1588 response_free (resp);
1590 /* 20x responses are counted among successful by default. */
1591 if (H_20X (statcode))
1594 /* Return if redirected. */
1595 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1597 /* RFC2068 says that in case of the 300 (multiple choices)
1598 response, the server can output a preferred URL through
1599 `Location' header; otherwise, the request should be treated
1600 like GET. So, if the location is set, it will be a
1601 redirection; otherwise, just proceed normally. */
1602 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1606 logprintf (LOG_VERBOSE,
1607 _("Location: %s%s\n"),
1608 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1609 hs->newloc ? _(" [following]") : "");
1611 skip_short_body (sock, contlen);
1612 CLOSE_FINISH (sock);
1618 /* If content-type is not given, assume text/html. This is because
1619 of the multitude of broken CGI's that "forget" to generate the
1622 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1623 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1628 if (opt.html_extension && (*dt & TEXTHTML))
1629 /* -E / --html-extension / html_extension = on was specified, and this is a
1630 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1631 already the file's suffix, tack on ".html". */
1633 char* last_period_in_local_filename = strrchr(*hs->local_file, '.');
1635 if (last_period_in_local_filename == NULL
1636 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1637 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1639 size_t local_filename_len = strlen(*hs->local_file);
1641 *hs->local_file = xrealloc(*hs->local_file,
1642 local_filename_len + sizeof(".html"));
1643 strcpy(*hs->local_file + local_filename_len, ".html");
1645 *dt |= ADDED_HTML_EXTENSION;
1649 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1651 /* If `-c' is in use and the file has been fully downloaded (or
1652 the remote file has shrunk), Wget effectively requests bytes
1653 after the end of file and the server response with 416. */
1654 logputs (LOG_VERBOSE, _("\
1655 \n The file is already fully retrieved; nothing to do.\n\n"));
1656 /* In case the caller inspects. */
1659 /* Mark as successfully retrieved. */
1662 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1663 might be more bytes in the body. */
1664 return RETRUNNEEDED;
1666 if ((contrange != 0 && contrange != hs->restval)
1667 || (H_PARTIAL (statcode) && !contrange))
1669 /* The Range request was somehow misunderstood by the server.
1672 CLOSE_INVALIDATE (sock);
1675 hs->contlen = contlen + contrange;
1681 /* No need to print this output if the body won't be
1682 downloaded at all, or if the original server response is
1684 logputs (LOG_VERBOSE, _("Length: "));
1687 logputs (LOG_VERBOSE, legible (contlen + contrange));
1689 logprintf (LOG_VERBOSE, _(" (%s to go)"), legible (contlen));
1692 logputs (LOG_VERBOSE,
1693 opt.ignore_length ? _("ignored") : _("unspecified"));
1695 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1697 logputs (LOG_VERBOSE, "\n");
1701 type = NULL; /* We don't need it any more. */
1703 /* Return if we have no intention of further downloading. */
1704 if (!(*dt & RETROKF) || (*dt & HEAD_ONLY))
1706 /* In case the caller cares to look... */
1710 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1711 servers not to send body in response to a HEAD request. If
1712 you encounter such a server (more likely a broken CGI), use
1713 `--no-http-keep-alive'. */
1714 CLOSE_FINISH (sock);
1715 return RETRFINISHED;
1718 /* Open the local file. */
1721 mkalldirs (*hs->local_file);
1723 rotate_backups (*hs->local_file);
1725 fp = fopen (*hs->local_file, "ab");
1726 else if (opt.noclobber || opt.always_rest || opt.timestamping || opt.dirstruct
1727 || opt.output_document)
1728 fp = fopen (*hs->local_file, "wb");
1731 fp = fopen_excl (*hs->local_file, 0);
1732 if (!fp && errno == EEXIST)
1734 /* We cannot just invent a new name and use it (which is
1735 what functions like unique_create typically do)
1736 because we told the user we'd use this name.
1737 Instead, return and retry the download. */
1738 logprintf (LOG_NOTQUIET,
1739 _("%s has sprung into existence.\n"),
1741 CLOSE_INVALIDATE (sock);
1742 return FOPEN_EXCL_ERR;
1747 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1748 CLOSE_INVALIDATE (sock);
1755 /* #### This confuses the timestamping code that checks for file
1756 size. Maybe we should save some additional information? */
1757 if (opt.save_headers)
1758 fwrite (head, 1, strlen (head), fp);
1760 /* Download the request body. */
1763 flags |= rb_read_exactly;
1764 if (hs->restval > 0 && contrange == 0)
1765 /* If the server ignored our range request, instruct fd_read_body
1766 to skip the first RESTVAL bytes of body. */
1767 flags |= rb_skip_startpos;
1768 hs->len = hs->restval;
1770 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1771 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1775 CLOSE_FINISH (sock);
1777 CLOSE_INVALIDATE (sock);
1780 /* Close or flush the file. We have to be careful to check for
1781 error here. Checking the result of fwrite() is not enough --
1782 errors could go unnoticed! */
1785 flush_res = fclose (fp);
1787 flush_res = fflush (fp);
1788 if (flush_res == EOF)
1793 return RETRFINISHED;
1796 /* The genuine HTTP loop! This is the part where the retrieval is
1797 retried, and retried, and retried, and... */
1799 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1800 int *dt, struct url *proxy)
1803 int use_ts, got_head = 0; /* time-stamping info */
1804 char *filename_plus_orig_suffix;
1805 char *local_filename = NULL;
1806 char *tms, *locf, *tmrate;
1808 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1809 wgint local_size = 0; /* the size of the local file */
1810 size_t filename_len;
1811 struct http_stat hstat; /* HTTP status */
1815 /* This used to be done in main(), but it's a better idea to do it
1816 here so that we don't go through the hoops if we're just using
1820 if (!wget_cookie_jar)
1821 wget_cookie_jar = cookie_jar_new ();
1822 if (opt.cookies_input && !cookies_loaded_p)
1824 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1825 cookies_loaded_p = 1;
1831 /* Warn on (likely bogus) wildcard usage in HTTP. Don't use
1832 has_wildcards_p because it would also warn on `?', and we know that
1833 shows up in CGI paths a *lot*. */
1834 if (strchr (u->url, '*'))
1835 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
1839 /* Determine the local filename. */
1840 if (local_file && *local_file)
1841 hstat.local_file = local_file;
1842 else if (local_file && !opt.output_document)
1844 *local_file = url_file_name (u);
1845 hstat.local_file = local_file;
1849 dummy = url_file_name (u);
1850 hstat.local_file = &dummy;
1851 /* be honest about where we will save the file */
1852 if (local_file && opt.output_document)
1853 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
1856 if (!opt.output_document)
1857 locf = *hstat.local_file;
1859 locf = opt.output_document;
1861 hstat.referer = referer;
1863 filename_len = strlen (*hstat.local_file);
1864 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1866 if (opt.noclobber && file_exists_p (*hstat.local_file))
1868 /* If opt.noclobber is turned on and file already exists, do not
1869 retrieve the file */
1870 logprintf (LOG_VERBOSE, _("\
1871 File `%s' already there, will not retrieve.\n"), *hstat.local_file);
1872 /* If the file is there, we suppose it's retrieved OK. */
1875 /* #### Bogusness alert. */
1876 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1877 if (has_html_suffix_p (*hstat.local_file))
1885 if (opt.timestamping)
1887 int local_dot_orig_file_exists = 0;
1889 if (opt.backup_converted)
1890 /* If -K is specified, we'll act on the assumption that it was specified
1891 last time these files were downloaded as well, and instead of just
1892 comparing local file X against server file X, we'll compare local
1893 file X.orig (if extant, else X) against server file X. If -K
1894 _wasn't_ specified last time, or the server contains files called
1895 *.orig, -N will be back to not operating correctly with -k. */
1897 /* Would a single s[n]printf() call be faster? --dan
1899 Definitely not. sprintf() is horribly slow. It's a
1900 different question whether the difference between the two
1901 affects a program. Usually I'd say "no", but at one
1902 point I profiled Wget, and found that a measurable and
1903 non-negligible amount of time was lost calling sprintf()
1904 in url.c. Replacing sprintf with inline calls to
1905 strcpy() and number_to_string() made a difference.
1907 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
1908 memcpy (filename_plus_orig_suffix + filename_len,
1909 ".orig", sizeof (".orig"));
1911 /* Try to stat() the .orig file. */
1912 if (stat (filename_plus_orig_suffix, &st) == 0)
1914 local_dot_orig_file_exists = 1;
1915 local_filename = filename_plus_orig_suffix;
1919 if (!local_dot_orig_file_exists)
1920 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1921 if (stat (*hstat.local_file, &st) == 0)
1922 local_filename = *hstat.local_file;
1924 if (local_filename != NULL)
1925 /* There was a local file, so we'll check later to see if the version
1926 the server has is the same version we already have, allowing us to
1932 /* Modification time granularity is 2 seconds for Windows, so
1933 increase local time by 1 second for later comparison. */
1936 local_size = st.st_size;
1940 /* Reset the counter. */
1946 /* Increment the pass counter. */
1948 sleep_between_retrievals (count);
1949 /* Get the current time string. */
1950 tms = time_str (NULL);
1951 /* Print fetch message, if opt.verbose. */
1954 char *hurl = url_string (u, 1);
1958 sprintf (tmp, _("(try:%2d)"), count);
1959 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
1960 tms, hurl, tmp, locf);
1962 ws_changetitle (hurl);
1967 /* Default document type is empty. However, if spider mode is
1968 on or time-stamping is employed, HEAD_ONLY commands is
1969 encoded within *dt. */
1970 if (opt.spider || (use_ts && !got_head))
1975 /* Decide whether or not to restart. */
1978 hstat.restval = hstat.len; /* continue where we left off */
1979 else if (opt.always_rest
1980 && stat (locf, &st) == 0
1981 && S_ISREG (st.st_mode))
1982 hstat.restval = st.st_size;
1984 /* Decide whether to send the no-cache directive. We send it in
1986 a) we're using a proxy, and we're past our first retrieval.
1987 Some proxies are notorious for caching incomplete data, so
1988 we require a fresh get.
1989 b) caching is explicitly inhibited. */
1990 if ((proxy && count > 1) /* a */
1991 || !opt.allow_cache /* b */
1993 *dt |= SEND_NOCACHE;
1995 *dt &= ~SEND_NOCACHE;
1997 /* Try fetching the document, or at least its head. */
1998 err = gethttp (u, &hstat, dt, proxy);
2000 /* It's unfortunate that wget determines the local filename before finding
2001 out the Content-Type of the file. Barring a major restructuring of the
2002 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2003 *hstat.local_file to tack on ".html". */
2004 if (!opt.output_document)
2005 locf = *hstat.local_file;
2008 tms = time_str (NULL);
2009 /* Get the new location (with or without the redirection). */
2011 *newloc = xstrdup (hstat.newloc);
2014 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2015 case CONERROR: case READERR: case WRITEFAILED:
2016 case RANGEERR: case FOPEN_EXCL_ERR:
2017 /* Non-fatal errors continue executing the loop, which will
2018 bring them to "while" statement at the end, to judge
2019 whether the number of tries was exceeded. */
2020 free_hstat (&hstat);
2021 printwhat (count, opt.ntry);
2022 if (err == FOPEN_EXCL_ERR)
2024 /* Re-determine the file name. */
2025 if (local_file && *local_file)
2027 xfree (*local_file);
2028 *local_file = url_file_name (u);
2029 hstat.local_file = local_file;
2034 dummy = url_file_name (u);
2035 hstat.local_file = &dummy;
2037 /* be honest about where we will save the file */
2038 if (local_file && opt.output_document)
2039 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2040 if (!opt.output_document)
2041 locf = *hstat.local_file;
2043 locf = opt.output_document;
2047 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2048 case SSLERRCTXCREATE: case CONTNOTSUPPORTED:
2049 /* Fatal errors just return from the function. */
2050 free_hstat (&hstat);
2054 case FWRITEERR: case FOPENERR:
2055 /* Another fatal error. */
2056 logputs (LOG_VERBOSE, "\n");
2057 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2058 *hstat.local_file, strerror (errno));
2059 free_hstat (&hstat);
2064 /* Another fatal error. */
2065 logputs (LOG_VERBOSE, "\n");
2066 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2067 free_hstat (&hstat);
2072 /* Return the new location to the caller. */
2075 logprintf (LOG_NOTQUIET,
2076 _("ERROR: Redirection (%d) without location.\n"),
2078 free_hstat (&hstat);
2082 free_hstat (&hstat);
2087 /* The file was already fully retrieved. */
2088 free_hstat (&hstat);
2093 /* Deal with you later. */
2096 /* All possibilities should have been exhausted. */
2099 if (!(*dt & RETROKF))
2103 /* #### Ugly ugly ugly! */
2104 char *hurl = url_string (u, 1);
2105 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2108 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2109 tms, hstat.statcode, escnonprint (hstat.error));
2110 logputs (LOG_VERBOSE, "\n");
2111 free_hstat (&hstat);
2116 /* Did we get the time-stamp? */
2119 if (opt.timestamping && !hstat.remote_time)
2121 logputs (LOG_NOTQUIET, _("\
2122 Last-modified header missing -- time-stamps turned off.\n"));
2124 else if (hstat.remote_time)
2126 /* Convert the date-string into struct tm. */
2127 tmr = http_atotm (hstat.remote_time);
2128 if (tmr == (time_t) (-1))
2129 logputs (LOG_VERBOSE, _("\
2130 Last-modified header invalid -- time-stamp ignored.\n"));
2134 /* The time-stamping section. */
2139 use_ts = 0; /* no more time-stamping */
2140 count = 0; /* the retrieve count for HEAD is
2142 if (hstat.remote_time && tmr != (time_t) (-1))
2144 /* Now time-stamping can be used validly. Time-stamping
2145 means that if the sizes of the local and remote file
2146 match, and local file is newer than the remote file,
2147 it will not be retrieved. Otherwise, the normal
2148 download procedure is resumed. */
2150 (hstat.contlen == -1 || local_size == hstat.contlen))
2152 logprintf (LOG_VERBOSE, _("\
2153 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2155 free_hstat (&hstat);
2159 else if (tml >= tmr)
2160 logprintf (LOG_VERBOSE, _("\
2161 The sizes do not match (local %s) -- retrieving.\n"),
2162 number_to_static_string (local_size));
2164 logputs (LOG_VERBOSE,
2165 _("Remote file is newer, retrieving.\n"));
2167 free_hstat (&hstat);
2170 if ((tmr != (time_t) (-1))
2172 && ((hstat.len == hstat.contlen) ||
2173 ((hstat.res == 0) &&
2174 ((hstat.contlen == -1) ||
2175 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2177 /* #### This code repeats in http.c and ftp.c. Move it to a
2179 const char *fl = NULL;
2180 if (opt.output_document)
2182 if (output_stream_regular)
2183 fl = opt.output_document;
2186 fl = *hstat.local_file;
2190 /* End of time-stamping section. */
2194 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2195 escnonprint (hstat.error));
2200 tmrate = retr_rate (hstat.rd_size, hstat.dltime, 0);
2202 if (hstat.len == hstat.contlen)
2206 logprintf (LOG_VERBOSE,
2207 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2209 number_to_static_string (hstat.len),
2210 number_to_static_string (hstat.contlen));
2211 logprintf (LOG_NONVERBOSE,
2212 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2214 number_to_static_string (hstat.len),
2215 number_to_static_string (hstat.contlen),
2219 total_downloaded_bytes += hstat.len;
2221 /* Remember that we downloaded the file for later ".orig" code. */
2222 if (*dt & ADDED_HTML_EXTENSION)
2223 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2225 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2227 free_hstat (&hstat);
2231 else if (hstat.res == 0) /* No read error */
2233 if (hstat.contlen == -1) /* We don't know how much we were supposed
2234 to get, so assume we succeeded. */
2238 logprintf (LOG_VERBOSE,
2239 _("%s (%s) - `%s' saved [%s]\n\n"),
2241 number_to_static_string (hstat.len));
2242 logprintf (LOG_NONVERBOSE,
2243 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2244 tms, u->url, number_to_static_string (hstat.len),
2248 total_downloaded_bytes += hstat.len;
2250 /* Remember that we downloaded the file for later ".orig" code. */
2251 if (*dt & ADDED_HTML_EXTENSION)
2252 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2254 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2256 free_hstat (&hstat);
2260 else if (hstat.len < hstat.contlen) /* meaning we lost the
2261 connection too soon */
2263 logprintf (LOG_VERBOSE,
2264 _("%s (%s) - Connection closed at byte %s. "),
2265 tms, tmrate, number_to_static_string (hstat.len));
2266 printwhat (count, opt.ntry);
2267 free_hstat (&hstat);
2270 else if (!opt.kill_longer) /* meaning we got more than expected */
2272 logprintf (LOG_VERBOSE,
2273 _("%s (%s) - `%s' saved [%s/%s])\n\n"),
2275 number_to_static_string (hstat.len),
2276 number_to_static_string (hstat.contlen));
2277 logprintf (LOG_NONVERBOSE,
2278 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2280 number_to_static_string (hstat.len),
2281 number_to_static_string (hstat.contlen),
2284 total_downloaded_bytes += hstat.len;
2286 /* Remember that we downloaded the file for later ".orig" code. */
2287 if (*dt & ADDED_HTML_EXTENSION)
2288 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2290 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2292 free_hstat (&hstat);
2296 else /* the same, but not accepted */
2298 logprintf (LOG_VERBOSE,
2299 _("%s (%s) - Connection closed at byte %s/%s. "),
2301 number_to_static_string (hstat.len),
2302 number_to_static_string (hstat.contlen));
2303 printwhat (count, opt.ntry);
2304 free_hstat (&hstat);
2308 else /* now hstat.res can only be -1 */
2310 if (hstat.contlen == -1)
2312 logprintf (LOG_VERBOSE,
2313 _("%s (%s) - Read error at byte %s (%s)."),
2314 tms, tmrate, number_to_static_string (hstat.len),
2316 printwhat (count, opt.ntry);
2317 free_hstat (&hstat);
2320 else /* hstat.res == -1 and contlen is given */
2322 logprintf (LOG_VERBOSE,
2323 _("%s (%s) - Read error at byte %s/%s (%s). "),
2325 number_to_static_string (hstat.len),
2326 number_to_static_string (hstat.contlen),
2328 printwhat (count, opt.ntry);
2329 free_hstat (&hstat);
2336 while (!opt.ntry || (count < opt.ntry));
2340 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2341 than local timezone.
2343 mktime is similar but assumes struct tm, also known as the
2344 "broken-down" form of time, is in local time zone. mktime_from_utc
2345 uses mktime to make the conversion understanding that an offset
2346 will be introduced by the local time assumption.
2348 mktime_from_utc then measures the introduced offset by applying
2349 gmtime to the initial result and applying mktime to the resulting
2350 "broken-down" form. The difference between the two mktime results
2351 is the measured offset which is then subtracted from the initial
2352 mktime result to yield a calendar time which is the value returned.
2354 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2355 consistent offset (the non DST offset) since tm and tm+o might be
2356 on opposite sides of a DST change.
2358 Some implementations of mktime return -1 for the nonexistent
2359 localtime hour at the beginning of DST. In this event, use
2360 mktime(tm - 1hr) + 3600.
2364 gmtime(t+o) --> tm+o
2365 mktime(tm+o) --> t+2o
2366 t+o - (t+2o - t+o) = t
2368 Note that glibc contains a function of the same purpose named
2369 `timegm' (reverse of gmtime). But obviously, it is not universally
2370 available, and unfortunately it is not straightforwardly
2371 extractable for use here. Perhaps configure should detect timegm
2372 and use it where available.
2374 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2375 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2376 Further improved by Roger with assistance from Edward J. Sabol
2377 based on input by Jamie Zawinski. */
2380 mktime_from_utc (struct tm *t)
2391 return -1; /* can't deal with output from strptime */
2402 return -1; /* can't deal with output from gmtime */
2405 return (tl - (tb - tl));
2408 /* Check whether the result of strptime() indicates success.
2409 strptime() returns the pointer to how far it got to in the string.
2410 The processing has been successful if the string is at `GMT' or
2411 `+X', or at the end of the string.
2413 In extended regexp parlance, the function returns 1 if P matches
2414 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2415 can return) is considered a failure and 0 is returned. */
2417 check_end (const char *p)
2421 while (ISSPACE (*p))
2424 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2425 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2431 /* Convert the textual specification of time in TIME_STRING to the
2432 number of seconds since the Epoch.
2434 TIME_STRING can be in any of the three formats RFC2068 allows the
2435 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date.
2436 Timezones are ignored, and should be GMT.
2438 Return the computed time_t representation, or -1 if the conversion
2441 This function uses strptime with various string formats for parsing
2442 TIME_STRING. This results in a parser that is not as lenient in
2443 interpreting TIME_STRING as I would like it to be. Being based on
2444 strptime, it always allows shortened months, one-digit days, etc.,
2445 but due to the multitude of formats in which time can be
2446 represented, an ideal HTTP time parser would be even more
2447 forgiving. It should completely ignore things like week days and
2448 concentrate only on the various forms of representing years,
2449 months, days, hours, minutes, and seconds. For example, it would
2450 be nice if it accepted ISO 8601 out of the box.
2452 I've investigated free and PD code for this purpose, but none was
2453 usable. getdate was big and unwieldy, and had potential copyright
2454 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2455 distributed with phttpd, is excellent, but we cannot use it because
2456 it is not assigned to the FSF. So I stuck it with strptime. */
2459 http_atotm (const char *time_string)
2461 /* NOTE: Solaris strptime man page claims that %n and %t match white
2462 space, but that's not universally available. Instead, we simply
2463 use ` ' to mean "skip all WS", which works under all strptime
2464 implementations I've tested. */
2466 static const char *time_formats[] = {
2467 "%a, %d %b %Y %T", /* RFC1123: Thu, 29 Jan 1998 22:12:57 */
2468 "%A, %d-%b-%y %T", /* RFC850: Thursday, 29-Jan-98 22:12:57 */
2469 "%a, %d-%b-%Y %T", /* pseudo-RFC850: Thu, 29-Jan-1998 22:12:57
2470 (google.com uses this for their cookies.) */
2471 "%a %b %d %T %Y" /* asctime: Thu Jan 29 22:12:57 1998 */
2477 /* According to Roger Beeman, we need to initialize tm_isdst, since
2478 strptime won't do it. */
2481 /* Note that under foreign locales Solaris strptime() fails to
2482 recognize English dates, which renders this function useless. We
2483 solve this by being careful not to affect LC_TIME when
2484 initializing locale.
2486 Another solution would be to temporarily set locale to C, invoke
2487 strptime(), and restore it back. This is slow and dirty,
2488 however, and locale support other than LC_MESSAGES can mess other
2489 things, so I rather chose to stick with just setting LC_MESSAGES.
2491 GNU strptime does not have this problem because it recognizes
2492 both international and local dates. */
2494 for (i = 0; i < countof (time_formats); i++)
2495 if (check_end (strptime (time_string, time_formats[i], &t)))
2496 return mktime_from_utc (&t);
2498 /* All formats have failed. */
2502 /* Authorization support: We support two authorization schemes:
2504 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2506 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2507 consisting of answering to the server's challenge with the proper
2510 /* How many bytes it will take to store LEN bytes in base64. */
2511 #define BASE64_LENGTH(len) (4 * (((len) + 2) / 3))
2513 /* Encode the string S of length LENGTH to base64 format and place it
2514 to STORE. STORE will be 0-terminated, and must point to a writable
2515 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2517 base64_encode (const char *s, char *store, int length)
2519 /* Conversion table. */
2520 static char tbl[64] = {
2521 'A','B','C','D','E','F','G','H',
2522 'I','J','K','L','M','N','O','P',
2523 'Q','R','S','T','U','V','W','X',
2524 'Y','Z','a','b','c','d','e','f',
2525 'g','h','i','j','k','l','m','n',
2526 'o','p','q','r','s','t','u','v',
2527 'w','x','y','z','0','1','2','3',
2528 '4','5','6','7','8','9','+','/'
2531 unsigned char *p = (unsigned char *)store;
2533 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2534 for (i = 0; i < length; i += 3)
2536 *p++ = tbl[s[0] >> 2];
2537 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2538 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2539 *p++ = tbl[s[2] & 0x3f];
2542 /* Pad the result if necessary... */
2543 if (i == length + 1)
2545 else if (i == length + 2)
2546 *(p - 1) = *(p - 2) = '=';
2547 /* ...and zero-terminate it. */
2551 /* Create the authentication header contents for the `Basic' scheme.
2552 This is done by encoding the string `USER:PASS' in base64 and
2553 prepending `HEADER: Basic ' to it. */
2555 basic_authentication_encode (const char *user, const char *passwd)
2557 char *t1, *t2, *res;
2558 int len1 = strlen (user) + 1 + strlen (passwd);
2559 int len2 = BASE64_LENGTH (len1);
2561 t1 = (char *)alloca (len1 + 1);
2562 sprintf (t1, "%s:%s", user, passwd);
2564 t2 = (char *)alloca (len2 + 1);
2565 base64_encode (t1, t2, len1);
2567 res = (char *)xmalloc (6 + len2 + 1);
2568 sprintf (res, "Basic %s", t2);
2573 #define SKIP_WS(x) do { \
2574 while (ISSPACE (*(x))) \
2579 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2580 of a field in such a header. If the field is the one specified by
2581 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2582 digest authorization code), extract its value in the (char*)
2583 variable pointed by RET. Returns negative on a malformed header,
2584 or number of bytes that have been parsed by this call. */
2586 extract_header_attr (const char *au, const char *attr_name, char **ret)
2588 const char *cp, *ep;
2592 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2594 cp += strlen (attr_name);
2607 for (ep = cp; *ep && *ep != '\"'; ep++)
2612 *ret = strdupdelim (cp, ep);
2619 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2620 an array of 16 bytes containing the hash keys, and BUF should be a
2621 buffer of 33 writable characters (32 for hex digits plus one for
2622 zero termination). */
2624 dump_hash (unsigned char *buf, const unsigned char *hash)
2628 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2630 *buf++ = XNUM_TO_digit (*hash >> 4);
2631 *buf++ = XNUM_TO_digit (*hash & 0xf);
2636 /* Take the line apart to find the challenge, and compose a digest
2637 authorization header. See RFC2069 section 2.1.2. */
2639 digest_authentication_encode (const char *au, const char *user,
2640 const char *passwd, const char *method,
2643 static char *realm, *opaque, *nonce;
2648 { "realm", &realm },
2649 { "opaque", &opaque },
2654 realm = opaque = nonce = NULL;
2656 au += 6; /* skip over `Digest' */
2662 for (i = 0; i < countof (options); i++)
2664 int skip = extract_header_attr (au, options[i].name,
2665 options[i].variable);
2669 xfree_null (opaque);
2679 if (i == countof (options))
2681 while (*au && *au != '=')
2689 while (*au && *au != '\"')
2696 while (*au && *au != ',')
2701 if (!realm || !nonce || !user || !passwd || !path || !method)
2704 xfree_null (opaque);
2709 /* Calculate the digest value. */
2711 ALLOCA_MD5_CONTEXT (ctx);
2712 unsigned char hash[MD5_HASHLEN];
2713 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2714 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2716 /* A1BUF = H(user ":" realm ":" password) */
2718 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2719 gen_md5_update ((unsigned char *)":", 1, ctx);
2720 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2721 gen_md5_update ((unsigned char *)":", 1, ctx);
2722 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2723 gen_md5_finish (ctx, hash);
2724 dump_hash (a1buf, hash);
2726 /* A2BUF = H(method ":" path) */
2728 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2729 gen_md5_update ((unsigned char *)":", 1, ctx);
2730 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2731 gen_md5_finish (ctx, hash);
2732 dump_hash (a2buf, hash);
2734 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2736 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2737 gen_md5_update ((unsigned char *)":", 1, ctx);
2738 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2739 gen_md5_update ((unsigned char *)":", 1, ctx);
2740 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2741 gen_md5_finish (ctx, hash);
2742 dump_hash (response_digest, hash);
2744 res = (char*) xmalloc (strlen (user)
2749 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2750 + (opaque ? strlen (opaque) : 0)
2752 sprintf (res, "Digest \
2753 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2754 user, realm, nonce, path, response_digest);
2757 char *p = res + strlen (res);
2758 strcat (p, ", opaque=\"");
2765 #endif /* USE_DIGEST */
2768 #define BEGINS_WITH(line, string_constant) \
2769 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
2770 && (ISSPACE (line[sizeof (string_constant) - 1]) \
2771 || !line[sizeof (string_constant) - 1]))
2774 known_authentication_scheme_p (const char *au)
2776 return BEGINS_WITH (au, "Basic")
2777 || BEGINS_WITH (au, "Digest")
2778 || BEGINS_WITH (au, "NTLM");
2783 /* Create the HTTP authorization request header. When the
2784 `WWW-Authenticate' response header is seen, according to the
2785 authorization scheme specified in that header (`Basic' and `Digest'
2786 are supported by the current implementation), produce an
2787 appropriate HTTP authorization request header. */
2789 create_authorization_line (const char *au, const char *user,
2790 const char *passwd, const char *method,
2793 if (0 == strncasecmp (au, "Basic", 5))
2794 return basic_authentication_encode (user, passwd);
2796 if (0 == strncasecmp (au, "Digest", 6))
2797 return digest_authentication_encode (au, user, passwd, method, path);
2798 #endif /* USE_DIGEST */