2 Copyright (C) 2005 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. */
53 # include "http-ntlm.h"
61 extern char *version_string;
63 extern FILE *output_stream;
64 extern bool output_stream_regular;
67 # define MIN(x, y) ((x) > (y) ? (y) : (x))
71 static bool cookies_loaded_p;
72 static struct cookie_jar *wget_cookie_jar;
74 #define TEXTHTML_S "text/html"
75 #define TEXTXHTML_S "application/xhtml+xml"
77 /* Some status code validation macros: */
78 #define H_20X(x) (((x) >= 200) && ((x) < 300))
79 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
80 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
81 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
82 || (x) == HTTP_STATUS_SEE_OTHER \
83 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
85 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
87 #define HTTP_STATUS_OK 200
88 #define HTTP_STATUS_CREATED 201
89 #define HTTP_STATUS_ACCEPTED 202
90 #define HTTP_STATUS_NO_CONTENT 204
91 #define HTTP_STATUS_PARTIAL_CONTENTS 206
93 /* Redirection 3xx. */
94 #define HTTP_STATUS_MULTIPLE_CHOICES 300
95 #define HTTP_STATUS_MOVED_PERMANENTLY 301
96 #define HTTP_STATUS_MOVED_TEMPORARILY 302
97 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
98 #define HTTP_STATUS_NOT_MODIFIED 304
99 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
101 /* Client error 4xx. */
102 #define HTTP_STATUS_BAD_REQUEST 400
103 #define HTTP_STATUS_UNAUTHORIZED 401
104 #define HTTP_STATUS_FORBIDDEN 403
105 #define HTTP_STATUS_NOT_FOUND 404
106 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
108 /* Server errors 5xx. */
109 #define HTTP_STATUS_INTERNAL 500
110 #define HTTP_STATUS_NOT_IMPLEMENTED 501
111 #define HTTP_STATUS_BAD_GATEWAY 502
112 #define HTTP_STATUS_UNAVAILABLE 503
115 rel_none, rel_name, rel_value, rel_both
122 struct request_header {
124 enum rp release_policy;
126 int hcount, hcapacity;
129 /* Create a new, empty request. At least request_set_method must be
130 called before the request can be used. */
132 static struct request *
135 struct request *req = xnew0 (struct request);
137 req->headers = xnew_array (struct request_header, req->hcapacity);
141 /* Set the request's method and its arguments. METH should be a
142 literal string (or it should outlive the request) because it will
143 not be freed. ARG will be freed by request_free. */
146 request_set_method (struct request *req, const char *meth, char *arg)
152 /* Return the method string passed with the last call to
153 request_set_method. */
156 request_method (const struct request *req)
161 /* Free one header according to the release policy specified with
162 request_set_header. */
165 release_header (struct request_header *hdr)
167 switch (hdr->release_policy)
184 /* Set the request named NAME to VALUE. Specifically, this means that
185 a "NAME: VALUE\r\n" header line will be used in the request. If a
186 header with the same name previously existed in the request, its
187 value will be replaced by this one. A NULL value means do nothing.
189 RELEASE_POLICY determines whether NAME and VALUE should be released
190 (freed) with request_free. Allowed values are:
192 - rel_none - don't free NAME or VALUE
193 - rel_name - free NAME when done
194 - rel_value - free VALUE when done
195 - rel_both - free both NAME and VALUE when done
197 Setting release policy is useful when arguments come from different
198 sources. For example:
200 // Don't free literal strings!
201 request_set_header (req, "Pragma", "no-cache", rel_none);
203 // Don't free a global variable, we'll need it later.
204 request_set_header (req, "Referer", opt.referer, rel_none);
206 // Value freshly allocated, free it when done.
207 request_set_header (req, "Range",
208 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
213 request_set_header (struct request *req, char *name, char *value,
214 enum rp release_policy)
216 struct request_header *hdr;
221 /* A NULL value is a no-op; if freeing the name is requested,
222 free it now to avoid leaks. */
223 if (release_policy == rel_name || release_policy == rel_both)
228 for (i = 0; i < req->hcount; i++)
230 hdr = &req->headers[i];
231 if (0 == strcasecmp (name, hdr->name))
233 /* Replace existing header. */
234 release_header (hdr);
237 hdr->release_policy = release_policy;
242 /* Install new header. */
244 if (req->hcount >= req->hcapacity)
246 req->hcapacity <<= 1;
247 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
249 hdr = &req->headers[req->hcount++];
252 hdr->release_policy = release_policy;
255 /* Like request_set_header, but sets the whole header line, as
256 provided by the user using the `--header' option. For example,
257 request_set_user_header (req, "Foo: bar") works just like
258 request_set_header (req, "Foo", "bar"). */
261 request_set_user_header (struct request *req, const char *header)
264 const char *p = strchr (header, ':');
267 BOUNDED_TO_ALLOCA (header, p, name);
271 request_set_header (req, xstrdup (name), (char *) p, rel_name);
274 /* Remove the header with specified name from REQ. Returns true if
275 the header was actually removed, false otherwise. */
278 request_remove_header (struct request *req, char *name)
281 for (i = 0; i < req->hcount; i++)
283 struct request_header *hdr = &req->headers[i];
284 if (0 == strcasecmp (name, hdr->name))
286 release_header (hdr);
287 /* Move the remaining headers by one. */
288 if (i < req->hcount - 1)
289 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
297 #define APPEND(p, str) do { \
298 int A_len = strlen (str); \
299 memcpy (p, str, A_len); \
303 /* Construct the request and write it to FD using fd_write. */
306 request_send (const struct request *req, int fd)
308 char *request_string, *p;
309 int i, size, write_error;
311 /* Count the request size. */
314 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
315 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
317 for (i = 0; i < req->hcount; i++)
319 struct request_header *hdr = &req->headers[i];
320 /* NAME ": " VALUE "\r\n" */
321 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
327 p = request_string = alloca_array (char, size);
329 /* Generate the request. */
331 APPEND (p, req->method); *p++ = ' ';
332 APPEND (p, req->arg); *p++ = ' ';
333 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
335 for (i = 0; i < req->hcount; i++)
337 struct request_header *hdr = &req->headers[i];
338 APPEND (p, hdr->name);
339 *p++ = ':', *p++ = ' ';
340 APPEND (p, hdr->value);
341 *p++ = '\r', *p++ = '\n';
344 *p++ = '\r', *p++ = '\n', *p++ = '\0';
345 assert (p - request_string == size);
349 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
351 /* Send the request to the server. */
353 write_error = fd_write (fd, request_string, size - 1, -1);
355 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
360 /* Release the resources used by REQ. */
363 request_free (struct request *req)
366 xfree_null (req->arg);
367 for (i = 0; i < req->hcount; i++)
368 release_header (&req->headers[i]);
369 xfree_null (req->headers);
373 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
374 PROMISED_SIZE bytes are sent over the wire -- if the file is
375 longer, read only that much; if the file is shorter, report an error. */
378 post_file (int sock, const char *file_name, wgint promised_size)
380 static char chunk[8192];
385 DEBUGP (("[writing POST file %s ... ", file_name));
387 fp = fopen (file_name, "rb");
390 while (!feof (fp) && written < promised_size)
393 int length = fread (chunk, 1, sizeof (chunk), fp);
396 towrite = MIN (promised_size - written, length);
397 write_error = fd_write (sock, chunk, towrite, -1);
407 /* If we've written less than was promised, report a (probably
408 nonsensical) error rather than break the promise. */
409 if (written < promised_size)
415 assert (written == promised_size);
416 DEBUGP (("done]\n"));
421 response_head_terminator (const char *hunk, int oldlen, int peeklen)
423 const char *start, *end;
425 /* If at first peek, verify whether HUNK starts with "HTTP". If
426 not, this is a HTTP/0.9 request and we must bail out without
428 if (oldlen == 0 && 0 != memcmp (hunk, "HTTP", MIN (peeklen, 4)))
434 start = hunk + oldlen - 4;
435 end = hunk + oldlen + peeklen;
437 for (; start < end - 1; start++)
444 if (start[1] == '\n')
450 /* The maximum size of a single HTTP response we care to read. This
451 is not meant to impose an arbitrary limit, but to protect the user
452 from Wget slurping up available memory upon encountering malicious
453 or buggy server output. Define it to 0 to remove the limit. */
455 #define HTTP_RESPONSE_MAX_SIZE 65536
457 /* Read the HTTP request head from FD and return it. The error
458 conditions are the same as with fd_read_hunk.
460 To support HTTP/0.9 responses, this function tries to make sure
461 that the data begins with "HTTP". If this is not the case, no data
462 is read and an empty request is returned, so that the remaining
463 data can be treated as body. */
466 read_http_response_head (int fd)
468 return fd_read_hunk (fd, response_head_terminator, 512,
469 HTTP_RESPONSE_MAX_SIZE);
473 /* The response data. */
476 /* The array of pointers that indicate where each header starts.
477 For example, given this HTTP response:
484 The headers are located like this:
486 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
488 headers[0] headers[1] headers[2] headers[3]
490 I.e. headers[0] points to the beginning of the request,
491 headers[1] points to the end of the first header and the
492 beginning of the second one, etc. */
494 const char **headers;
497 /* Create a new response object from the text of the HTTP response,
498 available in HEAD. That text is automatically split into
499 constituent header lines for fast retrieval using
502 static struct response *
503 resp_new (const char *head)
508 struct response *resp = xnew0 (struct response);
513 /* Empty head means that we're dealing with a headerless
514 (HTTP/0.9) response. In that case, don't set HEADERS at
519 /* Split HEAD into header lines, so that resp_header_* functions
520 don't need to do this over and over again. */
526 DO_REALLOC (resp->headers, size, count + 1, const char *);
527 resp->headers[count++] = hdr;
529 /* Break upon encountering an empty line. */
530 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
533 /* Find the end of HDR, including continuations. */
536 const char *end = strchr (hdr, '\n');
542 while (*hdr == ' ' || *hdr == '\t');
544 DO_REALLOC (resp->headers, size, count + 1, const char *);
545 resp->headers[count] = NULL;
550 /* Locate the header named NAME in the request data, starting with
551 position START. This allows the code to loop through the request
552 data, filtering for all requests of a given name. Returns the
553 found position, or -1 for failure. The code that uses this
554 function typically looks like this:
556 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
557 ... do something with header ...
559 If you only care about one header, use resp_header_get instead of
563 resp_header_locate (const struct response *resp, const char *name, int start,
564 const char **begptr, const char **endptr)
567 const char **headers = resp->headers;
570 if (!headers || !headers[1])
573 name_len = strlen (name);
579 for (; headers[i + 1]; i++)
581 const char *b = headers[i];
582 const char *e = headers[i + 1];
584 && b[name_len] == ':'
585 && 0 == strncasecmp (b, name, name_len))
588 while (b < e && ISSPACE (*b))
590 while (b < e && ISSPACE (e[-1]))
600 /* Find and retrieve the header named NAME in the request data. If
601 found, set *BEGPTR to its starting, and *ENDPTR to its ending
602 position, and return true. Otherwise return false.
604 This function is used as a building block for resp_header_copy
605 and resp_header_strdup. */
608 resp_header_get (const struct response *resp, const char *name,
609 const char **begptr, const char **endptr)
611 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
615 /* Copy the response header named NAME to buffer BUF, no longer than
616 BUFSIZE (BUFSIZE includes the terminating 0). If the header
617 exists, true is returned, false otherwise. If there should be no
618 limit on the size of the header, use resp_header_strdup instead.
620 If BUFSIZE is 0, no data is copied, but the boolean indication of
621 whether the header is present is still returned. */
624 resp_header_copy (const struct response *resp, const char *name,
625 char *buf, int bufsize)
628 if (!resp_header_get (resp, name, &b, &e))
632 int len = MIN (e - b, bufsize - 1);
633 memcpy (buf, b, len);
639 /* Return the value of header named NAME in RESP, allocated with
640 malloc. If such a header does not exist in RESP, return NULL. */
643 resp_header_strdup (const struct response *resp, const char *name)
646 if (!resp_header_get (resp, name, &b, &e))
648 return strdupdelim (b, e);
651 /* Parse the HTTP status line, which is of format:
653 HTTP-Version SP Status-Code SP Reason-Phrase
655 The function returns the status-code, or -1 if the status line
656 appears malformed. The pointer to "reason-phrase" message is
657 returned in *MESSAGE. */
660 resp_status (const struct response *resp, char **message)
667 /* For a HTTP/0.9 response, assume status 200. */
669 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
673 p = resp->headers[0];
674 end = resp->headers[1];
680 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
684 /* Match the HTTP version. This is optional because Gnutella
685 servers have been reported to not specify HTTP version. */
686 if (p < end && *p == '/')
689 while (p < end && ISDIGIT (*p))
691 if (p < end && *p == '.')
693 while (p < end && ISDIGIT (*p))
697 while (p < end && ISSPACE (*p))
699 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
702 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
707 while (p < end && ISSPACE (*p))
709 while (p < end && ISSPACE (end[-1]))
711 *message = strdupdelim (p, end);
717 /* Release the resources used by RESP. */
720 resp_free (struct response *resp)
722 xfree_null (resp->headers);
726 /* Print the server response, line by line, omitting the trailing CRLF
727 from individual header lines, and prefixed with PREFIX. */
730 print_server_response (const struct response *resp, const char *prefix)
735 for (i = 0; resp->headers[i + 1]; i++)
737 const char *b = resp->headers[i];
738 const char *e = resp->headers[i + 1];
740 if (b < e && e[-1] == '\n')
742 if (b < e && e[-1] == '\r')
744 /* This is safe even on printfs with broken handling of "%.<n>s"
745 because resp->headers ends with \0. */
746 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, e - b, b);
750 /* Parse the `Content-Range' header and extract the information it
751 contains. Returns true if successful, false otherwise. */
753 parse_content_range (const char *hdr, wgint *first_byte_ptr,
754 wgint *last_byte_ptr, wgint *entity_length_ptr)
758 /* Ancient versions of Netscape proxy server, presumably predating
759 rfc2068, sent out `Content-Range' without the "bytes"
761 if (0 == strncasecmp (hdr, "bytes", 5))
764 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
768 while (ISSPACE (*hdr))
775 for (num = 0; ISDIGIT (*hdr); hdr++)
776 num = 10 * num + (*hdr - '0');
777 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
779 *first_byte_ptr = num;
781 for (num = 0; ISDIGIT (*hdr); hdr++)
782 num = 10 * num + (*hdr - '0');
783 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
785 *last_byte_ptr = num;
787 for (num = 0; ISDIGIT (*hdr); hdr++)
788 num = 10 * num + (*hdr - '0');
789 *entity_length_ptr = num;
793 /* Read the body of the request, but don't store it anywhere and don't
794 display a progress gauge. This is useful for reading the bodies of
795 administrative responses to which we will soon issue another
796 request. The response is not useful to the user, but reading it
797 allows us to continue using the same connection to the server.
799 If reading fails, false is returned, true otherwise. In debug
800 mode, the body is displayed for debugging purposes. */
803 skip_short_body (int fd, wgint contlen)
806 SKIP_SIZE = 512, /* size of the download buffer */
807 SKIP_THRESHOLD = 4096 /* the largest size we read */
809 char dlbuf[SKIP_SIZE + 1];
810 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
812 /* We shouldn't get here with unknown contlen. (This will change
813 with HTTP/1.1, which supports "chunked" transfer.) */
814 assert (contlen != -1);
816 /* If the body is too large, it makes more sense to simply close the
817 connection than to try to read the body. */
818 if (contlen > SKIP_THRESHOLD)
821 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
825 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
828 /* Don't normally report the error since this is an
829 optimization that should be invisible to the user. */
830 DEBUGP (("] aborting (%s).\n",
831 ret < 0 ? strerror (errno) : "EOF received"));
835 /* Safe even if %.*s bogusly expects terminating \0 because
836 we've zero-terminated dlbuf above. */
837 DEBUGP (("%.*s", ret, dlbuf));
840 DEBUGP (("] done.\n"));
844 /* Persistent connections. Currently, we cache the most recently used
845 connection as persistent, provided that the HTTP server agrees to
846 make it such. The persistence data is stored in the variables
847 below. Ideally, it should be possible to cache an arbitrary fixed
848 number of these connections. */
850 /* Whether a persistent connection is active. */
851 static bool pconn_active;
854 /* The socket of the connection. */
857 /* Host and port of the currently active persistent connection. */
861 /* Whether a ssl handshake has occoured on this connection. */
864 /* Whether the connection was authorized. This is only done by
865 NTLM, which authorizes *connections* rather than individual
866 requests. (That practice is peculiar for HTTP, but it is a
867 useful optimization.) */
871 /* NTLM data of the current connection. */
872 struct ntlmdata ntlm;
876 /* Mark the persistent connection as invalid and free the resources it
877 uses. This is used by the CLOSE_* macros after they forcefully
878 close a registered persistent connection. */
881 invalidate_persistent (void)
883 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
884 pconn_active = false;
885 fd_close (pconn.socket);
890 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
891 persistent. This will enable someone to use the same connection
892 later. In the context of HTTP, this must be called only AFTER the
893 response has been received and the server has promised that the
894 connection will remain alive.
896 If a previous connection was persistent, it is closed. */
899 register_persistent (const char *host, int port, int fd, bool ssl)
903 if (pconn.socket == fd)
905 /* The connection FD is already registered. */
910 /* The old persistent connection is still active; close it
911 first. This situation arises whenever a persistent
912 connection exists, but we then connect to a different
913 host, and try to register a persistent connection to that
915 invalidate_persistent ();
921 pconn.host = xstrdup (host);
924 pconn.authorized = false;
926 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
929 /* Return true if a persistent connection is available for connecting
933 persistent_available_p (const char *host, int port, bool ssl,
934 bool *host_lookup_failed)
936 /* First, check whether a persistent connection is active at all. */
940 /* If we want SSL and the last connection wasn't or vice versa,
941 don't use it. Checking for host and port is not enough because
942 HTTP and HTTPS can apparently coexist on the same port. */
943 if (ssl != pconn.ssl)
946 /* If we're not connecting to the same port, we're not interested. */
947 if (port != pconn.port)
950 /* If the host is the same, we're in business. If not, there is
951 still hope -- read below. */
952 if (0 != strcasecmp (host, pconn.host))
954 /* Check if pconn.socket is talking to HOST under another name.
955 This happens often when both sites are virtual hosts
956 distinguished only by name and served by the same network
957 interface, and hence the same web server (possibly set up by
958 the ISP and serving many different web sites). This
959 admittedly unconventional optimization does not contradict
960 HTTP and works well with popular server software. */
964 struct address_list *al;
967 /* Don't try to talk to two different SSL sites over the same
968 secure connection! (Besides, it's not clear that
969 name-based virtual hosting is even possible with SSL.) */
972 /* If pconn.socket's peer is one of the IP addresses HOST
973 resolves to, pconn.socket is for all intents and purposes
974 already talking to HOST. */
976 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
978 /* Can't get the peer's address -- something must be very
979 wrong with the connection. */
980 invalidate_persistent ();
983 al = lookup_host (host, 0);
986 *host_lookup_failed = true;
990 found = address_list_contains (al, &ip);
991 address_list_release (al);
996 /* The persistent connection's peer address was found among the
997 addresses HOST resolved to; therefore, pconn.sock is in fact
998 already talking to HOST -- no need to reconnect. */
1001 /* Finally, check whether the connection is still open. This is
1002 important because most server implement a liberal (short) timeout
1003 on persistent connections. Wget can of course always reconnect
1004 if the connection doesn't work out, but it's nicer to know in
1005 advance. This test is a logical followup of the first test, but
1006 is "expensive" and therefore placed at the end of the list. */
1008 if (!test_socket_open (pconn.socket))
1010 /* Oops, the socket is no longer open. Now that we know that,
1011 let's invalidate the persistent connection before returning
1013 invalidate_persistent ();
1020 /* The idea behind these two CLOSE macros is to distinguish between
1021 two cases: one when the job we've been doing is finished, and we
1022 want to close the connection and leave, and two when something is
1023 seriously wrong and we're closing the connection as part of
1026 In case of keep_alive, CLOSE_FINISH should leave the connection
1027 open, while CLOSE_INVALIDATE should still close it.
1029 Note that the semantics of the flag `keep_alive' is "this
1030 connection *will* be reused (the server has promised not to close
1031 the connection once we're done)", while the semantics of
1032 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1033 active, registered connection". */
1035 #define CLOSE_FINISH(fd) do { \
1038 if (pconn_active && (fd) == pconn.socket) \
1039 invalidate_persistent (); \
1048 #define CLOSE_INVALIDATE(fd) do { \
1049 if (pconn_active && (fd) == pconn.socket) \
1050 invalidate_persistent (); \
1058 wgint len; /* received length */
1059 wgint contlen; /* expected length */
1060 wgint restval; /* the restart value */
1061 int res; /* the result of last read */
1062 char *newloc; /* new location (redirection) */
1063 char *remote_time; /* remote time-stamp string */
1064 char *error; /* textual HTTP error */
1065 int statcode; /* status code */
1066 wgint rd_size; /* amount of data read from socket */
1067 double dltime; /* time it took to download the data */
1068 const char *referer; /* value of the referer header. */
1069 char **local_file; /* local file. */
1073 free_hstat (struct http_stat *hs)
1075 xfree_null (hs->newloc);
1076 xfree_null (hs->remote_time);
1077 xfree_null (hs->error);
1079 /* Guard against being called twice. */
1081 hs->remote_time = NULL;
1085 static char *create_authorization_line (const char *, const char *,
1086 const char *, const char *,
1087 const char *, bool *);
1088 static char *basic_authentication_encode (const char *, const char *);
1089 static bool known_authentication_scheme_p (const char *, const char *);
1091 time_t http_atotm (const char *);
1093 #define BEGINS_WITH(line, string_constant) \
1094 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1095 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1096 || !line[sizeof (string_constant) - 1]))
1098 #define SET_USER_AGENT(req) do { \
1099 if (!opt.useragent) \
1100 request_set_header (req, "User-Agent", \
1101 aprintf ("Wget/%s", version_string), rel_value); \
1102 else if (*opt.useragent) \
1103 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1106 /* The flags that allow clobbering the file (opening with "wb").
1107 Defined here to avoid repetition later. #### This will require
1109 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1110 || opt.dirstruct || opt.output_document)
1112 /* Retrieve a document through HTTP protocol. It recognizes status
1113 code, and correctly handles redirections. It closes the network
1114 socket. If it receives an error from the functions below it, it
1115 will print it if there is enough information to do so (almost
1116 always), returning the error to the caller (i.e. http_loop).
1118 Various HTTP parameters are stored to hs.
1120 If PROXY is non-NULL, the connection will be made to the proxy
1121 server, and u->url will be requested. */
1123 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1125 struct request *req;
1128 char *user, *passwd;
1132 wgint contlen, contrange;
1139 /* Set to 1 when the authorization has failed permanently and should
1140 not be tried again. */
1141 bool auth_finished = false;
1143 /* Whether NTLM authentication is used for this request. */
1144 bool ntlm_seen = false;
1146 /* Whether our connection to the remote host is through SSL. */
1147 bool using_ssl = false;
1149 /* Whether a HEAD request will be issued (as opposed to GET or
1151 bool head_only = !!(*dt & HEAD_ONLY);
1154 struct response *resp;
1158 /* Whether this connection will be kept alive after the HTTP request
1162 /* Whether keep-alive should be inhibited.
1164 RFC 2068 requests that 1.0 clients not send keep-alive requests
1165 to proxies. This is because many 1.0 proxies do not interpret
1166 the Connection header and transfer it to the remote server,
1167 causing it to not close the connection and leave both the proxy
1168 and the client hanging. */
1169 bool inhibit_keep_alive =
1170 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1172 /* Headers sent when using POST. */
1173 wgint post_data_size = 0;
1175 bool host_lookup_failed = false;
1178 if (u->scheme == SCHEME_HTTPS)
1180 /* Initialize the SSL context. After this has once been done,
1181 it becomes a no-op. */
1184 scheme_disable (SCHEME_HTTPS);
1185 logprintf (LOG_NOTQUIET,
1186 _("Disabling SSL due to encountered errors.\n"));
1187 return SSLINITFAILED;
1190 #endif /* HAVE_SSL */
1193 /* If we're doing a GET on the URL, as opposed to just a HEAD, we need to
1194 know the local filename so we can save to it. */
1195 assert (*hs->local_file != NULL);
1197 /* Initialize certain elements of struct http_stat. */
1202 hs->remote_time = NULL;
1207 /* Prepare the request to send. */
1209 req = request_new ();
1212 const char *meth = "GET";
1215 else if (opt.post_file_name || opt.post_data)
1217 /* Use the full path, i.e. one that includes the leading slash and
1218 the query string. E.g. if u->path is "foo/bar" and u->query is
1219 "param=value", full_path will be "/foo/bar?param=value". */
1222 /* When using SSL over proxy, CONNECT establishes a direct
1223 connection to the HTTPS server. Therefore use the same
1224 argument as when talking to the server directly. */
1225 && u->scheme != SCHEME_HTTPS
1228 meth_arg = xstrdup (u->url);
1230 meth_arg = url_full_path (u);
1231 request_set_method (req, meth, meth_arg);
1234 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1235 if (*dt & SEND_NOCACHE)
1236 request_set_header (req, "Pragma", "no-cache", rel_none);
1238 request_set_header (req, "Range",
1239 aprintf ("bytes=%s-",
1240 number_to_static_string (hs->restval)),
1242 SET_USER_AGENT (req);
1243 request_set_header (req, "Accept", "*/*", rel_none);
1245 /* Find the username and password for authentication. */
1248 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1249 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1250 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1254 /* We have the username and the password, but haven't tried
1255 any authorization yet. Let's see if the "Basic" method
1256 works. If not, we'll come back here and construct a
1257 proper authorization method with the right challenges.
1259 If we didn't employ this kind of logic, every URL that
1260 requires authorization would have to be processed twice,
1261 which is very suboptimal and generates a bunch of false
1262 "unauthorized" errors in the server log.
1264 #### But this logic also has a serious problem when used
1265 with stronger authentications: we *first* transmit the
1266 username and the password in clear text, and *then* attempt a
1267 stronger authentication scheme. That cannot be right! We
1268 are only fortunate that almost everyone still uses the
1269 `Basic' scheme anyway.
1271 There should be an option to prevent this from happening, for
1272 those who use strong authentication schemes and value their
1274 request_set_header (req, "Authorization",
1275 basic_authentication_encode (user, passwd),
1282 char *proxy_user, *proxy_passwd;
1283 /* For normal username and password, URL components override
1284 command-line/wgetrc parameters. With proxy
1285 authentication, it's the reverse, because proxy URLs are
1286 normally the "permanent" ones, so command-line args
1287 should take precedence. */
1288 if (opt.proxy_user && opt.proxy_passwd)
1290 proxy_user = opt.proxy_user;
1291 proxy_passwd = opt.proxy_passwd;
1295 proxy_user = proxy->user;
1296 proxy_passwd = proxy->passwd;
1298 /* #### This does not appear right. Can't the proxy request,
1299 say, `Digest' authentication? */
1300 if (proxy_user && proxy_passwd)
1301 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1303 /* If we're using a proxy, we will be connecting to the proxy
1307 /* Proxy authorization over SSL is handled below. */
1309 if (u->scheme != SCHEME_HTTPS)
1311 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1315 /* Whether we need to print the host header with braces around
1316 host, e.g. "Host: [3ffe:8100:200:2::2]:1234" instead of the
1317 usual "Host: symbolic-name:1234". */
1318 bool squares = strchr (u->host, ':') != NULL;
1319 if (u->port == scheme_default_port (u->scheme))
1320 request_set_header (req, "Host",
1321 aprintf (squares ? "[%s]" : "%s", u->host),
1324 request_set_header (req, "Host",
1325 aprintf (squares ? "[%s]:%d" : "%s:%d",
1330 if (!inhibit_keep_alive)
1331 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1334 request_set_header (req, "Cookie",
1335 cookie_header (wget_cookie_jar,
1336 u->host, u->port, u->path,
1338 u->scheme == SCHEME_HTTPS
1345 if (opt.post_data || opt.post_file_name)
1347 request_set_header (req, "Content-Type",
1348 "application/x-www-form-urlencoded", rel_none);
1350 post_data_size = strlen (opt.post_data);
1353 post_data_size = file_size (opt.post_file_name);
1354 if (post_data_size == -1)
1356 logprintf (LOG_NOTQUIET, _("POST data file missing: %s (%s)\n"),
1357 opt.post_file_name, strerror (errno));
1361 request_set_header (req, "Content-Length",
1362 xstrdup (number_to_static_string (post_data_size)),
1366 /* Add the user headers. */
1367 if (opt.user_headers)
1370 for (i = 0; opt.user_headers[i]; i++)
1371 request_set_user_header (req, opt.user_headers[i]);
1375 /* We need to come back here when the initial attempt to retrieve
1376 without authorization header fails. (Expected to happen at least
1377 for the Digest authorization scheme.) */
1381 /* Establish the connection. */
1383 if (!inhibit_keep_alive)
1385 /* Look for a persistent connection to target host, unless a
1386 proxy is used. The exception is when SSL is in use, in which
1387 case the proxy is nothing but a passthrough to the target
1388 host, registered as a connection to the latter. */
1389 struct url *relevant = conn;
1391 if (u->scheme == SCHEME_HTTPS)
1395 if (persistent_available_p (relevant->host, relevant->port,
1397 relevant->scheme == SCHEME_HTTPS,
1401 &host_lookup_failed))
1403 sock = pconn.socket;
1404 using_ssl = pconn.ssl;
1405 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1406 escnonprint (pconn.host), pconn.port);
1407 DEBUGP (("Reusing fd %d.\n", sock));
1408 if (pconn.authorized)
1409 /* If the connection is already authorized, the "Basic"
1410 authorization added by code above is unnecessary and
1412 request_remove_header (req, "Authorization");
1418 /* In its current implementation, persistent_available_p will
1419 look up conn->host in some cases. If that lookup failed, we
1420 don't need to bother with connect_to_host. */
1421 if (host_lookup_failed)
1427 sock = connect_to_host (conn->host, conn->port);
1436 return (retryable_socket_connect_error (errno)
1437 ? CONERROR : CONIMPOSSIBLE);
1441 if (proxy && u->scheme == SCHEME_HTTPS)
1443 /* When requesting SSL URLs through proxies, use the
1444 CONNECT method to request passthrough. */
1445 struct request *connreq = request_new ();
1446 request_set_method (connreq, "CONNECT",
1447 aprintf ("%s:%d", u->host, u->port));
1448 SET_USER_AGENT (connreq);
1451 request_set_header (connreq, "Proxy-Authorization",
1452 proxyauth, rel_value);
1453 /* Now that PROXYAUTH is part of the CONNECT request,
1454 zero it out so we don't send proxy authorization with
1455 the regular request below. */
1458 /* Examples in rfc2817 use the Host header in CONNECT
1459 requests. I don't see how that gains anything, given
1460 that the contents of Host would be exactly the same as
1461 the contents of CONNECT. */
1463 write_error = request_send (connreq, sock);
1464 request_free (connreq);
1465 if (write_error < 0)
1467 logprintf (LOG_VERBOSE, _("Failed writing to proxy: %s.\n"),
1469 CLOSE_INVALIDATE (sock);
1473 head = read_http_response_head (sock);
1476 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1478 CLOSE_INVALIDATE (sock);
1487 DEBUGP (("proxy responded with: [%s]\n", head));
1489 resp = resp_new (head);
1490 statcode = resp_status (resp, &message);
1493 if (statcode != 200)
1496 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1497 message ? escnonprint (message) : "?");
1498 xfree_null (message);
1501 xfree_null (message);
1503 /* SOCK is now *really* connected to u->host, so update CONN
1504 to reflect this. That way register_persistent will
1505 register SOCK as being connected to u->host:u->port. */
1509 if (conn->scheme == SCHEME_HTTPS)
1511 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1518 #endif /* HAVE_SSL */
1521 /* Send the request to server. */
1522 write_error = request_send (req, sock);
1524 if (write_error >= 0)
1528 DEBUGP (("[POST data: %s]\n", opt.post_data));
1529 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1531 else if (opt.post_file_name && post_data_size != 0)
1532 write_error = post_file (sock, opt.post_file_name, post_data_size);
1535 if (write_error < 0)
1537 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
1539 CLOSE_INVALIDATE (sock);
1543 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1544 proxy ? "Proxy" : "HTTP");
1549 head = read_http_response_head (sock);
1554 logputs (LOG_NOTQUIET, _("No data received.\n"));
1555 CLOSE_INVALIDATE (sock);
1561 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1563 CLOSE_INVALIDATE (sock);
1568 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1570 resp = resp_new (head);
1572 /* Check for status line. */
1574 statcode = resp_status (resp, &message);
1575 if (!opt.server_response)
1576 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1577 message ? escnonprint (message) : "");
1580 logprintf (LOG_VERBOSE, "\n");
1581 print_server_response (resp, " ");
1584 if (!opt.ignore_length
1585 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1589 parsed = str_to_wgint (hdrval, NULL, 10);
1590 if (parsed == WGINT_MAX && errno == ERANGE)
1592 #### If Content-Length is out of range, it most likely
1593 means that the file is larger than 2G and that we're
1594 compiled without LFS. In that case we should probably
1595 refuse to even attempt to download the file. */
1601 /* Check for keep-alive related responses. */
1602 if (!inhibit_keep_alive && contlen != -1)
1604 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1606 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1608 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1613 /* The server has promised that it will not close the connection
1614 when we're done. This means that we can register it. */
1615 register_persistent (conn->host, conn->port, sock, using_ssl);
1617 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1619 /* Authorization is required. */
1620 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1621 CLOSE_FINISH (sock);
1623 CLOSE_INVALIDATE (sock);
1624 pconn.authorized = false;
1625 if (!auth_finished && (user && passwd))
1627 /* IIS sends multiple copies of WWW-Authenticate, one with
1628 the value "negotiate", and other(s) with data. Loop over
1629 all the occurrences and pick the one we recognize. */
1631 const char *wabeg, *waend;
1632 char *www_authenticate = NULL;
1634 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1635 &wabeg, &waend)) != -1;
1637 if (known_authentication_scheme_p (wabeg, waend))
1639 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1643 if (!www_authenticate)
1644 /* If the authentication header is missing or
1645 unrecognized, there's no sense in retrying. */
1646 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1647 else if (BEGINS_WITH (www_authenticate, "Basic"))
1648 /* If the authentication scheme is "Basic", which we send
1649 by default, there's no sense in retrying either. (This
1650 should be changed when we stop sending "Basic" data by
1656 pth = url_full_path (u);
1657 request_set_header (req, "Authorization",
1658 create_authorization_line (www_authenticate,
1660 request_method (req),
1664 if (BEGINS_WITH (www_authenticate, "NTLM"))
1667 goto retry_with_auth;
1670 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1674 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1676 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1678 pconn.authorized = true;
1682 hs->statcode = statcode;
1684 hs->error = xstrdup (_("Malformed status line"));
1686 hs->error = xstrdup (_("(no description)"));
1688 hs->error = xstrdup (message);
1691 type = resp_header_strdup (resp, "Content-Type");
1694 char *tmp = strchr (type, ';');
1697 while (tmp > type && ISSPACE (tmp[-1]))
1702 hs->newloc = resp_header_strdup (resp, "Location");
1703 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
1705 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1709 const char *scbeg, *scend;
1710 /* The jar should have been created by now. */
1711 assert (wget_cookie_jar != NULL);
1713 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1714 &scbeg, &scend)) != -1;
1717 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
1718 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
1719 u->path, set_cookie);
1723 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
1725 wgint first_byte_pos, last_byte_pos, entity_length;
1726 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
1728 contrange = first_byte_pos;
1732 /* 20x responses are counted among successful by default. */
1733 if (H_20X (statcode))
1736 /* Return if redirected. */
1737 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
1739 /* RFC2068 says that in case of the 300 (multiple choices)
1740 response, the server can output a preferred URL through
1741 `Location' header; otherwise, the request should be treated
1742 like GET. So, if the location is set, it will be a
1743 redirection; otherwise, just proceed normally. */
1744 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
1748 logprintf (LOG_VERBOSE,
1749 _("Location: %s%s\n"),
1750 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
1751 hs->newloc ? _(" [following]") : "");
1752 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1753 CLOSE_FINISH (sock);
1755 CLOSE_INVALIDATE (sock);
1761 /* If content-type is not given, assume text/html. This is because
1762 of the multitude of broken CGI's that "forget" to generate the
1765 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
1766 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
1771 if (opt.html_extension && (*dt & TEXTHTML))
1772 /* -E / --html-extension / html_extension = on was specified, and this is a
1773 text/html file. If some case-insensitive variation on ".htm[l]" isn't
1774 already the file's suffix, tack on ".html". */
1776 char *last_period_in_local_filename = strrchr (*hs->local_file, '.');
1778 if (last_period_in_local_filename == NULL
1779 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
1780 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
1782 int local_filename_len = strlen (*hs->local_file);
1783 /* Resize the local file, allowing for ".html" preceded by
1784 optional ".NUMBER". */
1785 *hs->local_file = xrealloc (*hs->local_file,
1786 local_filename_len + 24 + sizeof (".html"));
1787 strcpy(*hs->local_file + local_filename_len, ".html");
1788 /* If clobbering is not allowed and the file, as named,
1789 exists, tack on ".NUMBER.html" instead. */
1794 sprintf (*hs->local_file + local_filename_len,
1795 ".%d.html", ext_num++);
1796 while (file_exists_p (*hs->local_file));
1798 *dt |= ADDED_HTML_EXTENSION;
1802 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
1804 /* If `-c' is in use and the file has been fully downloaded (or
1805 the remote file has shrunk), Wget effectively requests bytes
1806 after the end of file and the server response with 416. */
1807 logputs (LOG_VERBOSE, _("\
1808 \n The file is already fully retrieved; nothing to do.\n\n"));
1809 /* In case the caller inspects. */
1812 /* Mark as successfully retrieved. */
1815 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
1816 might be more bytes in the body. */
1817 return RETRUNNEEDED;
1819 if ((contrange != 0 && contrange != hs->restval)
1820 || (H_PARTIAL (statcode) && !contrange))
1822 /* The Range request was somehow misunderstood by the server.
1825 CLOSE_INVALIDATE (sock);
1828 hs->contlen = contlen + contrange;
1834 /* No need to print this output if the body won't be
1835 downloaded at all, or if the original server response is
1837 logputs (LOG_VERBOSE, _("Length: "));
1840 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
1841 if (contlen + contrange >= 1024)
1842 logprintf (LOG_VERBOSE, " (%s)",
1843 human_readable (contlen + contrange));
1846 if (contlen >= 1024)
1847 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
1848 number_to_static_string (contlen),
1849 human_readable (contlen));
1851 logprintf (LOG_VERBOSE, _(", %s remaining"),
1852 number_to_static_string (contlen));
1856 logputs (LOG_VERBOSE,
1857 opt.ignore_length ? _("ignored") : _("unspecified"));
1859 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
1861 logputs (LOG_VERBOSE, "\n");
1865 type = NULL; /* We don't need it any more. */
1867 /* Return if we have no intention of further downloading. */
1868 if (!(*dt & RETROKF) || head_only)
1870 /* In case the caller cares to look... */
1874 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
1875 servers not to send body in response to a HEAD request. If
1876 you encounter such a server (more likely a broken CGI), use
1877 `--no-http-keep-alive'. */
1878 CLOSE_FINISH (sock);
1879 return RETRFINISHED;
1882 /* Open the local file. */
1885 mkalldirs (*hs->local_file);
1887 rotate_backups (*hs->local_file);
1889 fp = fopen (*hs->local_file, "ab");
1890 else if (ALLOW_CLOBBER)
1891 fp = fopen (*hs->local_file, "wb");
1894 fp = fopen_excl (*hs->local_file, true);
1895 if (!fp && errno == EEXIST)
1897 /* We cannot just invent a new name and use it (which is
1898 what functions like unique_create typically do)
1899 because we told the user we'd use this name.
1900 Instead, return and retry the download. */
1901 logprintf (LOG_NOTQUIET,
1902 _("%s has sprung into existence.\n"),
1904 CLOSE_INVALIDATE (sock);
1905 return FOPEN_EXCL_ERR;
1910 logprintf (LOG_NOTQUIET, "%s: %s\n", *hs->local_file, strerror (errno));
1911 CLOSE_INVALIDATE (sock);
1918 /* #### This confuses the timestamping code that checks for file
1919 size. Maybe we should save some additional information? */
1920 if (opt.save_headers)
1921 fwrite (head, 1, strlen (head), fp);
1923 /* Now we no longer need to store the response header. */
1926 /* Download the request body. */
1929 flags |= rb_read_exactly;
1930 if (hs->restval > 0 && contrange == 0)
1931 /* If the server ignored our range request, instruct fd_read_body
1932 to skip the first RESTVAL bytes of body. */
1933 flags |= rb_skip_startpos;
1934 hs->len = hs->restval;
1936 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
1937 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
1941 CLOSE_FINISH (sock);
1943 CLOSE_INVALIDATE (sock);
1946 /* Close or flush the file. We have to be careful to check for
1947 error here. Checking the result of fwrite() is not enough --
1948 errors could go unnoticed! */
1951 flush_res = fclose (fp);
1953 flush_res = fflush (fp);
1954 if (flush_res == EOF)
1959 return RETRFINISHED;
1962 /* The genuine HTTP loop! This is the part where the retrieval is
1963 retried, and retried, and retried, and... */
1965 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
1966 int *dt, struct url *proxy)
1969 bool use_ts, got_head = false;/* time-stamping info */
1970 char *filename_plus_orig_suffix;
1971 char *local_filename = NULL;
1975 time_t tml = -1, tmr = -1; /* local and remote time-stamps */
1976 wgint local_size = 0; /* the size of the local file */
1977 size_t filename_len;
1978 struct http_stat hstat; /* HTTP status */
1982 /* This used to be done in main(), but it's a better idea to do it
1983 here so that we don't go through the hoops if we're just using
1987 if (!wget_cookie_jar)
1988 wget_cookie_jar = cookie_jar_new ();
1989 if (opt.cookies_input && !cookies_loaded_p)
1991 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
1992 cookies_loaded_p = true;
1998 /* Warn on (likely bogus) wildcard usage in HTTP. */
1999 if (opt.ftp_glob && has_wildcards_p (u->path))
2000 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2004 /* Determine the local filename. */
2005 if (local_file && *local_file)
2006 hstat.local_file = local_file;
2007 else if (local_file && !opt.output_document)
2009 *local_file = url_file_name (u);
2010 hstat.local_file = local_file;
2014 dummy = url_file_name (u);
2015 hstat.local_file = &dummy;
2016 /* be honest about where we will save the file */
2017 if (local_file && opt.output_document)
2018 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2021 if (!opt.output_document)
2022 locf = *hstat.local_file;
2024 locf = opt.output_document;
2026 hstat.referer = referer;
2028 filename_len = strlen (*hstat.local_file);
2029 filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
2031 if (opt.noclobber && file_exists_p (*hstat.local_file))
2033 /* If opt.noclobber is turned on and file already exists, do not
2034 retrieve the file */
2035 logprintf (LOG_VERBOSE, _("\
2036 File `%s' already there; not retrieving.\n\n"), *hstat.local_file);
2037 /* If the file is there, we suppose it's retrieved OK. */
2040 /* #### Bogusness alert. */
2041 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2042 if (has_html_suffix_p (*hstat.local_file))
2050 if (opt.timestamping)
2052 bool local_dot_orig_file_exists = false;
2054 if (opt.backup_converted)
2055 /* If -K is specified, we'll act on the assumption that it was specified
2056 last time these files were downloaded as well, and instead of just
2057 comparing local file X against server file X, we'll compare local
2058 file X.orig (if extant, else X) against server file X. If -K
2059 _wasn't_ specified last time, or the server contains files called
2060 *.orig, -N will be back to not operating correctly with -k. */
2062 /* Would a single s[n]printf() call be faster? --dan
2064 Definitely not. sprintf() is horribly slow. It's a
2065 different question whether the difference between the two
2066 affects a program. Usually I'd say "no", but at one
2067 point I profiled Wget, and found that a measurable and
2068 non-negligible amount of time was lost calling sprintf()
2069 in url.c. Replacing sprintf with inline calls to
2070 strcpy() and number_to_string() made a difference.
2072 memcpy (filename_plus_orig_suffix, *hstat.local_file, filename_len);
2073 memcpy (filename_plus_orig_suffix + filename_len,
2074 ".orig", sizeof (".orig"));
2076 /* Try to stat() the .orig file. */
2077 if (stat (filename_plus_orig_suffix, &st) == 0)
2079 local_dot_orig_file_exists = 1;
2080 local_filename = filename_plus_orig_suffix;
2084 if (!local_dot_orig_file_exists)
2085 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2086 if (stat (*hstat.local_file, &st) == 0)
2087 local_filename = *hstat.local_file;
2089 if (local_filename != NULL)
2090 /* There was a local file, so we'll check later to see if the version
2091 the server has is the same version we already have, allowing us to
2097 /* Modification time granularity is 2 seconds for Windows, so
2098 increase local time by 1 second for later comparison. */
2101 local_size = st.st_size;
2105 /* Reset the counter. */
2111 /* Increment the pass counter. */
2113 sleep_between_retrievals (count);
2114 /* Get the current time string. */
2115 tms = time_str (NULL);
2116 /* Print fetch message, if opt.verbose. */
2119 char *hurl = url_string (u, true);
2123 sprintf (tmp, _("(try:%2d)"), count);
2124 logprintf (LOG_VERBOSE, "--%s-- %s\n %s => `%s'\n",
2125 tms, hurl, tmp, locf);
2127 ws_changetitle (hurl);
2132 /* Default document type is empty. However, if spider mode is
2133 on or time-stamping is employed, HEAD_ONLY commands is
2134 encoded within *dt. */
2135 if (opt.spider || (use_ts && !got_head))
2140 /* Decide whether or not to restart. */
2142 && stat (locf, &st) == 0
2143 && S_ISREG (st.st_mode))
2144 /* When -c is used, continue from on-disk size. (Can't use
2145 hstat.len even if count>1 because we don't want a failed
2146 first attempt to clobber existing data.) */
2147 hstat.restval = st.st_size;
2149 /* otherwise, continue where the previous try left off */
2150 hstat.restval = hstat.len;
2154 /* Decide whether to send the no-cache directive. We send it in
2156 a) we're using a proxy, and we're past our first retrieval.
2157 Some proxies are notorious for caching incomplete data, so
2158 we require a fresh get.
2159 b) caching is explicitly inhibited. */
2160 if ((proxy && count > 1) /* a */
2161 || !opt.allow_cache /* b */
2163 *dt |= SEND_NOCACHE;
2165 *dt &= ~SEND_NOCACHE;
2167 /* Try fetching the document, or at least its head. */
2168 err = gethttp (u, &hstat, dt, proxy);
2170 /* It's unfortunate that wget determines the local filename before finding
2171 out the Content-Type of the file. Barring a major restructuring of the
2172 code, we need to re-set locf here, since gethttp() may have xrealloc()d
2173 *hstat.local_file to tack on ".html". */
2174 if (!opt.output_document)
2175 locf = *hstat.local_file;
2178 tms = time_str (NULL);
2179 /* Get the new location (with or without the redirection). */
2181 *newloc = xstrdup (hstat.newloc);
2184 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2185 case CONERROR: case READERR: case WRITEFAILED:
2186 case RANGEERR: case FOPEN_EXCL_ERR:
2187 /* Non-fatal errors continue executing the loop, which will
2188 bring them to "while" statement at the end, to judge
2189 whether the number of tries was exceeded. */
2190 free_hstat (&hstat);
2191 printwhat (count, opt.ntry);
2192 if (err == FOPEN_EXCL_ERR)
2194 /* Re-determine the file name. */
2195 if (local_file && *local_file)
2197 xfree (*local_file);
2198 *local_file = url_file_name (u);
2199 hstat.local_file = local_file;
2204 dummy = url_file_name (u);
2205 hstat.local_file = &dummy;
2207 /* be honest about where we will save the file */
2208 if (local_file && opt.output_document)
2209 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2210 if (!opt.output_document)
2211 locf = *hstat.local_file;
2213 locf = opt.output_document;
2216 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2217 case SSLINITFAILED: case CONTNOTSUPPORTED:
2218 /* Fatal errors just return from the function. */
2219 free_hstat (&hstat);
2222 case FWRITEERR: case FOPENERR:
2223 /* Another fatal error. */
2224 logputs (LOG_VERBOSE, "\n");
2225 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2226 *hstat.local_file, strerror (errno));
2227 free_hstat (&hstat);
2231 /* Another fatal error. */
2232 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2233 free_hstat (&hstat);
2237 /* Return the new location to the caller. */
2240 logprintf (LOG_NOTQUIET,
2241 _("ERROR: Redirection (%d) without location.\n"),
2243 free_hstat (&hstat);
2247 free_hstat (&hstat);
2251 /* The file was already fully retrieved. */
2252 free_hstat (&hstat);
2256 /* Deal with you later. */
2259 /* All possibilities should have been exhausted. */
2262 if (!(*dt & RETROKF))
2266 /* #### Ugly ugly ugly! */
2267 char *hurl = url_string (u, true);
2268 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2271 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2272 tms, hstat.statcode, escnonprint (hstat.error));
2273 logputs (LOG_VERBOSE, "\n");
2274 free_hstat (&hstat);
2279 /* Did we get the time-stamp? */
2282 if (opt.timestamping && !hstat.remote_time)
2284 logputs (LOG_NOTQUIET, _("\
2285 Last-modified header missing -- time-stamps turned off.\n"));
2287 else if (hstat.remote_time)
2289 /* Convert the date-string into struct tm. */
2290 tmr = http_atotm (hstat.remote_time);
2291 if (tmr == (time_t) (-1))
2292 logputs (LOG_VERBOSE, _("\
2293 Last-modified header invalid -- time-stamp ignored.\n"));
2297 /* The time-stamping section. */
2302 use_ts = false; /* no more time-stamping */
2303 count = 0; /* the retrieve count for HEAD is
2305 if (hstat.remote_time && tmr != (time_t) (-1))
2307 /* Now time-stamping can be used validly. Time-stamping
2308 means that if the sizes of the local and remote file
2309 match, and local file is newer than the remote file,
2310 it will not be retrieved. Otherwise, the normal
2311 download procedure is resumed. */
2313 (hstat.contlen == -1 || local_size == hstat.contlen))
2315 logprintf (LOG_VERBOSE, _("\
2316 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2318 free_hstat (&hstat);
2322 else if (tml >= tmr)
2323 logprintf (LOG_VERBOSE, _("\
2324 The sizes do not match (local %s) -- retrieving.\n"),
2325 number_to_static_string (local_size));
2327 logputs (LOG_VERBOSE,
2328 _("Remote file is newer, retrieving.\n"));
2330 free_hstat (&hstat);
2333 if ((tmr != (time_t) (-1))
2335 && ((hstat.len == hstat.contlen) ||
2336 ((hstat.res == 0) &&
2337 ((hstat.contlen == -1) ||
2338 (hstat.len >= hstat.contlen && !opt.kill_longer)))))
2340 /* #### This code repeats in http.c and ftp.c. Move it to a
2342 const char *fl = NULL;
2343 if (opt.output_document)
2345 if (output_stream_regular)
2346 fl = opt.output_document;
2349 fl = *hstat.local_file;
2353 /* End of time-stamping section. */
2357 logprintf (LOG_NOTQUIET, "%d %s\n\n", hstat.statcode,
2358 escnonprint (hstat.error));
2363 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2364 total_download_time += hstat.dltime;
2366 if (hstat.len == hstat.contlen)
2370 logprintf (LOG_VERBOSE,
2371 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2373 number_to_static_string (hstat.len),
2374 number_to_static_string (hstat.contlen));
2375 logprintf (LOG_NONVERBOSE,
2376 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2378 number_to_static_string (hstat.len),
2379 number_to_static_string (hstat.contlen),
2383 total_downloaded_bytes += hstat.len;
2385 /* Remember that we downloaded the file for later ".orig" code. */
2386 if (*dt & ADDED_HTML_EXTENSION)
2387 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2389 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2391 free_hstat (&hstat);
2395 else if (hstat.res == 0) /* No read error */
2397 if (hstat.contlen == -1) /* We don't know how much we were supposed
2398 to get, so assume we succeeded. */
2402 logprintf (LOG_VERBOSE,
2403 _("%s (%s) - `%s' saved [%s]\n\n"),
2405 number_to_static_string (hstat.len));
2406 logprintf (LOG_NONVERBOSE,
2407 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2408 tms, u->url, number_to_static_string (hstat.len),
2412 total_downloaded_bytes += hstat.len;
2414 /* Remember that we downloaded the file for later ".orig" code. */
2415 if (*dt & ADDED_HTML_EXTENSION)
2416 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2418 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2420 free_hstat (&hstat);
2424 else if (hstat.len < hstat.contlen) /* meaning we lost the
2425 connection too soon */
2427 logprintf (LOG_VERBOSE,
2428 _("%s (%s) - Connection closed at byte %s. "),
2429 tms, tmrate, number_to_static_string (hstat.len));
2430 printwhat (count, opt.ntry);
2431 free_hstat (&hstat);
2434 else if (!opt.kill_longer) /* meaning we got more than expected */
2436 logprintf (LOG_VERBOSE,
2437 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2439 number_to_static_string (hstat.len),
2440 number_to_static_string (hstat.contlen));
2441 logprintf (LOG_NONVERBOSE,
2442 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2444 number_to_static_string (hstat.len),
2445 number_to_static_string (hstat.contlen),
2448 total_downloaded_bytes += hstat.len;
2450 /* Remember that we downloaded the file for later ".orig" code. */
2451 if (*dt & ADDED_HTML_EXTENSION)
2452 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, locf);
2454 downloaded_file(FILE_DOWNLOADED_NORMALLY, locf);
2456 free_hstat (&hstat);
2460 else /* the same, but not accepted */
2462 logprintf (LOG_VERBOSE,
2463 _("%s (%s) - Connection closed at byte %s/%s. "),
2465 number_to_static_string (hstat.len),
2466 number_to_static_string (hstat.contlen));
2467 printwhat (count, opt.ntry);
2468 free_hstat (&hstat);
2472 else /* now hstat.res can only be -1 */
2474 if (hstat.contlen == -1)
2476 logprintf (LOG_VERBOSE,
2477 _("%s (%s) - Read error at byte %s (%s)."),
2478 tms, tmrate, number_to_static_string (hstat.len),
2480 printwhat (count, opt.ntry);
2481 free_hstat (&hstat);
2484 else /* hstat.res == -1 and contlen is given */
2486 logprintf (LOG_VERBOSE,
2487 _("%s (%s) - Read error at byte %s/%s (%s). "),
2489 number_to_static_string (hstat.len),
2490 number_to_static_string (hstat.contlen),
2492 printwhat (count, opt.ntry);
2493 free_hstat (&hstat);
2499 while (!opt.ntry || (count < opt.ntry));
2503 /* Converts struct tm to time_t, assuming the data in tm is UTC rather
2504 than local timezone.
2506 mktime is similar but assumes struct tm, also known as the
2507 "broken-down" form of time, is in local time zone. mktime_from_utc
2508 uses mktime to make the conversion understanding that an offset
2509 will be introduced by the local time assumption.
2511 mktime_from_utc then measures the introduced offset by applying
2512 gmtime to the initial result and applying mktime to the resulting
2513 "broken-down" form. The difference between the two mktime results
2514 is the measured offset which is then subtracted from the initial
2515 mktime result to yield a calendar time which is the value returned.
2517 tm_isdst in struct tm is set to 0 to force mktime to introduce a
2518 consistent offset (the non DST offset) since tm and tm+o might be
2519 on opposite sides of a DST change.
2521 Some implementations of mktime return -1 for the nonexistent
2522 localtime hour at the beginning of DST. In this event, use
2523 mktime(tm - 1hr) + 3600.
2527 gmtime(t+o) --> tm+o
2528 mktime(tm+o) --> t+2o
2529 t+o - (t+2o - t+o) = t
2531 Note that glibc contains a function of the same purpose named
2532 `timegm' (reverse of gmtime). But obviously, it is not universally
2533 available, and unfortunately it is not straightforwardly
2534 extractable for use here. Perhaps configure should detect timegm
2535 and use it where available.
2537 Contributed by Roger Beeman <beeman@cisco.com>, with the help of
2538 Mark Baushke <mdb@cisco.com> and the rest of the Gurus at CISCO.
2539 Further improved by Roger with assistance from Edward J. Sabol
2540 based on input by Jamie Zawinski. */
2543 mktime_from_utc (struct tm *t)
2554 return -1; /* can't deal with output from strptime */
2565 return -1; /* can't deal with output from gmtime */
2568 return (tl - (tb - tl));
2571 /* Check whether the result of strptime() indicates success.
2572 strptime() returns the pointer to how far it got to in the string.
2573 The processing has been successful if the string is at `GMT' or
2574 `+X', or at the end of the string.
2576 In extended regexp parlance, the function returns 1 if P matches
2577 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2578 can return) is considered a failure and 0 is returned. */
2580 check_end (const char *p)
2584 while (ISSPACE (*p))
2587 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2588 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2594 /* Convert the textual specification of time in TIME_STRING to the
2595 number of seconds since the Epoch.
2597 TIME_STRING can be in any of the three formats RFC2616 allows the
2598 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2599 as well as the time format used in the Set-Cookie header.
2600 Timezones are ignored, and should be GMT.
2602 Return the computed time_t representation, or -1 if the conversion
2605 This function uses strptime with various string formats for parsing
2606 TIME_STRING. This results in a parser that is not as lenient in
2607 interpreting TIME_STRING as I would like it to be. Being based on
2608 strptime, it always allows shortened months, one-digit days, etc.,
2609 but due to the multitude of formats in which time can be
2610 represented, an ideal HTTP time parser would be even more
2611 forgiving. It should completely ignore things like week days and
2612 concentrate only on the various forms of representing years,
2613 months, days, hours, minutes, and seconds. For example, it would
2614 be nice if it accepted ISO 8601 out of the box.
2616 I've investigated free and PD code for this purpose, but none was
2617 usable. getdate was big and unwieldy, and had potential copyright
2618 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2619 distributed with phttpd, is excellent, but we cannot use it because
2620 it is not assigned to the FSF. So I stuck it with strptime. */
2623 http_atotm (const char *time_string)
2625 /* NOTE: Solaris strptime man page claims that %n and %t match white
2626 space, but that's not universally available. Instead, we simply
2627 use ` ' to mean "skip all WS", which works under all strptime
2628 implementations I've tested. */
2630 static const char *time_formats[] = {
2631 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2632 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2633 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2634 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2635 (used in Set-Cookie, defined in the
2636 Netscape cookie specification.) */
2640 for (i = 0; i < countof (time_formats); i++)
2644 /* Some versions of strptime use the existing contents of struct
2645 tm to recalculate the date according to format. Zero it out
2646 to prevent garbage from the stack influencing strptime. */
2649 /* Solaris strptime fails to recognize English month names in
2650 non-English locales, which we work around by not setting the
2651 LC_TIME category. Another way would be to temporarily set
2652 locale to C before invoking strptime, but that's slow and
2653 messy. GNU strptime does not have this problem because it
2654 recognizes English month names along with the local ones. */
2656 if (check_end (strptime (time_string, time_formats[i], &t)))
2657 return mktime_from_utc (&t);
2660 /* All formats have failed. */
2664 /* Authorization support: We support three authorization schemes:
2666 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2668 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2669 consisting of answering to the server's challenge with the proper
2672 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2673 Stenberg for libcurl. Like digest, NTLM is based on a
2674 challenge-response mechanism, but unlike digest, it is non-standard
2675 (authenticates TCP connections rather than requests), undocumented
2676 and Microsoft-specific. */
2678 /* Create the authentication header contents for the `Basic' scheme.
2679 This is done by encoding the string "USER:PASS" to base64 and
2680 prepending the string "Basic " in front of it. */
2683 basic_authentication_encode (const char *user, const char *passwd)
2686 int len1 = strlen (user) + 1 + strlen (passwd);
2688 t1 = (char *)alloca (len1 + 1);
2689 sprintf (t1, "%s:%s", user, passwd);
2691 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2692 base64_encode (t1, len1, t2);
2694 return concat_strings ("Basic ", t2, (char *) 0);
2697 #define SKIP_WS(x) do { \
2698 while (ISSPACE (*(x))) \
2702 #ifdef ENABLE_DIGEST
2703 /* Parse HTTP `WWW-Authenticate:' header. AU points to the beginning
2704 of a field in such a header. If the field is the one specified by
2705 ATTR_NAME ("realm", "opaque", and "nonce" are used by the current
2706 digest authorization code), extract its value in the (char*)
2707 variable pointed by RET. Returns negative on a malformed header,
2708 or number of bytes that have been parsed by this call. */
2710 extract_header_attr (const char *au, const char *attr_name, char **ret)
2713 const char *cp = au;
2715 if (strncmp (cp, attr_name, strlen (attr_name)) == 0)
2717 cp += strlen (attr_name);
2730 for (ep = cp; *ep && *ep != '\"'; ep++)
2735 *ret = strdupdelim (cp, ep);
2742 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2743 an array of 16 bytes containing the hash keys, and BUF should be a
2744 buffer of 33 writable characters (32 for hex digits plus one for
2745 zero termination). */
2747 dump_hash (unsigned char *buf, const unsigned char *hash)
2751 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2753 *buf++ = XNUM_TO_digit (*hash >> 4);
2754 *buf++ = XNUM_TO_digit (*hash & 0xf);
2759 /* Take the line apart to find the challenge, and compose a digest
2760 authorization header. See RFC2069 section 2.1.2. */
2762 digest_authentication_encode (const char *au, const char *user,
2763 const char *passwd, const char *method,
2766 static char *realm, *opaque, *nonce;
2771 { "realm", &realm },
2772 { "opaque", &opaque },
2777 realm = opaque = nonce = NULL;
2779 au += 6; /* skip over `Digest' */
2785 for (i = 0; i < countof (options); i++)
2787 int skip = extract_header_attr (au, options[i].name,
2788 options[i].variable);
2792 xfree_null (opaque);
2802 if (i == countof (options))
2804 while (*au && *au != '=')
2812 while (*au && *au != '\"')
2819 while (*au && *au != ',')
2824 if (!realm || !nonce || !user || !passwd || !path || !method)
2827 xfree_null (opaque);
2832 /* Calculate the digest value. */
2834 ALLOCA_MD5_CONTEXT (ctx);
2835 unsigned char hash[MD5_HASHLEN];
2836 unsigned char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2837 unsigned char response_digest[MD5_HASHLEN * 2 + 1];
2839 /* A1BUF = H(user ":" realm ":" password) */
2841 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2842 gen_md5_update ((unsigned char *)":", 1, ctx);
2843 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2844 gen_md5_update ((unsigned char *)":", 1, ctx);
2845 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2846 gen_md5_finish (ctx, hash);
2847 dump_hash (a1buf, hash);
2849 /* A2BUF = H(method ":" path) */
2851 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2852 gen_md5_update ((unsigned char *)":", 1, ctx);
2853 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2854 gen_md5_finish (ctx, hash);
2855 dump_hash (a2buf, hash);
2857 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2859 gen_md5_update (a1buf, MD5_HASHLEN * 2, ctx);
2860 gen_md5_update ((unsigned char *)":", 1, ctx);
2861 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2862 gen_md5_update ((unsigned char *)":", 1, ctx);
2863 gen_md5_update (a2buf, MD5_HASHLEN * 2, ctx);
2864 gen_md5_finish (ctx, hash);
2865 dump_hash (response_digest, hash);
2867 res = xmalloc (strlen (user)
2872 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2873 + (opaque ? strlen (opaque) : 0)
2875 sprintf (res, "Digest \
2876 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2877 user, realm, nonce, path, response_digest);
2880 char *p = res + strlen (res);
2881 strcat (p, ", opaque=\"");
2888 #endif /* ENABLE_DIGEST */
2890 /* Computing the size of a string literal must take into account that
2891 value returned by sizeof includes the terminating \0. */
2892 #define STRSIZE(literal) (sizeof (literal) - 1)
2894 /* Whether chars in [b, e) begin with the literal string provided as
2895 first argument and are followed by whitespace or terminating \0.
2896 The comparison is case-insensitive. */
2897 #define STARTS(literal, b, e) \
2898 ((e) - (b) >= STRSIZE (literal) \
2899 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
2900 && ((e) - (b) == STRSIZE (literal) \
2901 || ISSPACE (b[STRSIZE (literal)])))
2904 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
2906 return STARTS ("Basic", hdrbeg, hdrend)
2907 #ifdef ENABLE_DIGEST
2908 || STARTS ("Digest", hdrbeg, hdrend)
2911 || STARTS ("NTLM", hdrbeg, hdrend)
2918 /* Create the HTTP authorization request header. When the
2919 `WWW-Authenticate' response header is seen, according to the
2920 authorization scheme specified in that header (`Basic' and `Digest'
2921 are supported by the current implementation), produce an
2922 appropriate HTTP authorization request header. */
2924 create_authorization_line (const char *au, const char *user,
2925 const char *passwd, const char *method,
2926 const char *path, bool *finished)
2928 /* We are called only with known schemes, so we can dispatch on the
2930 switch (TOUPPER (*au))
2932 case 'B': /* Basic */
2934 return basic_authentication_encode (user, passwd);
2935 #ifdef ENABLE_DIGEST
2936 case 'D': /* Digest */
2938 return digest_authentication_encode (au, user, passwd, method, path);
2941 case 'N': /* NTLM */
2942 if (!ntlm_input (&pconn.ntlm, au))
2947 return ntlm_output (&pconn.ntlm, user, passwd, finished);
2950 /* We shouldn't get here -- this function should be only called
2951 with values approved by known_authentication_scheme_p. */
2959 if (wget_cookie_jar)
2960 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
2966 xfree_null (pconn.host);
2967 if (wget_cookie_jar)
2968 cookie_jar_delete (wget_cookie_jar);