2 Copyright (C) 1996-2006 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 3 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, see <http://www.gnu.org/licenses/>.
19 In addition, as a special exception, the Free Software Foundation
20 gives permission to link the code of its release of Wget with the
21 OpenSSL project's "OpenSSL" library (or with modified versions of it
22 that use the same license as the "OpenSSL" library), and distribute
23 the linked executables. You must obey the GNU General Public License
24 in all respects for all of the code used other than "OpenSSL". If you
25 modify this file, you may extend this exception to your version of the
26 file, but you are not obligated to do so. If you do not wish to do
27 so, delete this exception statement from your version. */
55 # include "http-ntlm.h"
68 extern char *version_string;
71 static char *create_authorization_line (const char *, const char *,
72 const char *, const char *,
73 const char *, bool *);
74 static char *basic_authentication_encode (const char *, const char *);
75 static bool known_authentication_scheme_p (const char *, const char *);
76 static void load_cookies (void);
79 # define MIN(x, y) ((x) > (y) ? (y) : (x))
83 static bool cookies_loaded_p;
84 static struct cookie_jar *wget_cookie_jar;
86 #define TEXTHTML_S "text/html"
87 #define TEXTXHTML_S "application/xhtml+xml"
89 /* Some status code validation macros: */
90 #define H_20X(x) (((x) >= 200) && ((x) < 300))
91 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
92 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
93 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
94 || (x) == HTTP_STATUS_SEE_OTHER \
95 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
97 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
99 #define HTTP_STATUS_OK 200
100 #define HTTP_STATUS_CREATED 201
101 #define HTTP_STATUS_ACCEPTED 202
102 #define HTTP_STATUS_NO_CONTENT 204
103 #define HTTP_STATUS_PARTIAL_CONTENTS 206
105 /* Redirection 3xx. */
106 #define HTTP_STATUS_MULTIPLE_CHOICES 300
107 #define HTTP_STATUS_MOVED_PERMANENTLY 301
108 #define HTTP_STATUS_MOVED_TEMPORARILY 302
109 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
110 #define HTTP_STATUS_NOT_MODIFIED 304
111 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
113 /* Client error 4xx. */
114 #define HTTP_STATUS_BAD_REQUEST 400
115 #define HTTP_STATUS_UNAUTHORIZED 401
116 #define HTTP_STATUS_FORBIDDEN 403
117 #define HTTP_STATUS_NOT_FOUND 404
118 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
120 /* Server errors 5xx. */
121 #define HTTP_STATUS_INTERNAL 500
122 #define HTTP_STATUS_NOT_IMPLEMENTED 501
123 #define HTTP_STATUS_BAD_GATEWAY 502
124 #define HTTP_STATUS_UNAVAILABLE 503
127 rel_none, rel_name, rel_value, rel_both
134 struct request_header {
136 enum rp release_policy;
138 int hcount, hcapacity;
141 /* Create a new, empty request. At least request_set_method must be
142 called before the request can be used. */
144 static struct request *
147 struct request *req = xnew0 (struct request);
149 req->headers = xnew_array (struct request_header, req->hcapacity);
153 /* Set the request's method and its arguments. METH should be a
154 literal string (or it should outlive the request) because it will
155 not be freed. ARG will be freed by request_free. */
158 request_set_method (struct request *req, const char *meth, char *arg)
164 /* Return the method string passed with the last call to
165 request_set_method. */
168 request_method (const struct request *req)
173 /* Free one header according to the release policy specified with
174 request_set_header. */
177 release_header (struct request_header *hdr)
179 switch (hdr->release_policy)
196 /* Set the request named NAME to VALUE. Specifically, this means that
197 a "NAME: VALUE\r\n" header line will be used in the request. If a
198 header with the same name previously existed in the request, its
199 value will be replaced by this one. A NULL value means do nothing.
201 RELEASE_POLICY determines whether NAME and VALUE should be released
202 (freed) with request_free. Allowed values are:
204 - rel_none - don't free NAME or VALUE
205 - rel_name - free NAME when done
206 - rel_value - free VALUE when done
207 - rel_both - free both NAME and VALUE when done
209 Setting release policy is useful when arguments come from different
210 sources. For example:
212 // Don't free literal strings!
213 request_set_header (req, "Pragma", "no-cache", rel_none);
215 // Don't free a global variable, we'll need it later.
216 request_set_header (req, "Referer", opt.referer, rel_none);
218 // Value freshly allocated, free it when done.
219 request_set_header (req, "Range",
220 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
225 request_set_header (struct request *req, char *name, char *value,
226 enum rp release_policy)
228 struct request_header *hdr;
233 /* A NULL value is a no-op; if freeing the name is requested,
234 free it now to avoid leaks. */
235 if (release_policy == rel_name || release_policy == rel_both)
240 for (i = 0; i < req->hcount; i++)
242 hdr = &req->headers[i];
243 if (0 == strcasecmp (name, hdr->name))
245 /* Replace existing header. */
246 release_header (hdr);
249 hdr->release_policy = release_policy;
254 /* Install new header. */
256 if (req->hcount >= req->hcapacity)
258 req->hcapacity <<= 1;
259 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
261 hdr = &req->headers[req->hcount++];
264 hdr->release_policy = release_policy;
267 /* Like request_set_header, but sets the whole header line, as
268 provided by the user using the `--header' option. For example,
269 request_set_user_header (req, "Foo: bar") works just like
270 request_set_header (req, "Foo", "bar"). */
273 request_set_user_header (struct request *req, const char *header)
276 const char *p = strchr (header, ':');
279 BOUNDED_TO_ALLOCA (header, p, name);
283 request_set_header (req, xstrdup (name), (char *) p, rel_name);
286 /* Remove the header with specified name from REQ. Returns true if
287 the header was actually removed, false otherwise. */
290 request_remove_header (struct request *req, char *name)
293 for (i = 0; i < req->hcount; i++)
295 struct request_header *hdr = &req->headers[i];
296 if (0 == strcasecmp (name, hdr->name))
298 release_header (hdr);
299 /* Move the remaining headers by one. */
300 if (i < req->hcount - 1)
301 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
309 #define APPEND(p, str) do { \
310 int A_len = strlen (str); \
311 memcpy (p, str, A_len); \
315 /* Construct the request and write it to FD using fd_write. */
318 request_send (const struct request *req, int fd)
320 char *request_string, *p;
321 int i, size, write_error;
323 /* Count the request size. */
326 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
327 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
329 for (i = 0; i < req->hcount; i++)
331 struct request_header *hdr = &req->headers[i];
332 /* NAME ": " VALUE "\r\n" */
333 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
339 p = request_string = alloca_array (char, size);
341 /* Generate the request. */
343 APPEND (p, req->method); *p++ = ' ';
344 APPEND (p, req->arg); *p++ = ' ';
345 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
347 for (i = 0; i < req->hcount; i++)
349 struct request_header *hdr = &req->headers[i];
350 APPEND (p, hdr->name);
351 *p++ = ':', *p++ = ' ';
352 APPEND (p, hdr->value);
353 *p++ = '\r', *p++ = '\n';
356 *p++ = '\r', *p++ = '\n', *p++ = '\0';
357 assert (p - request_string == size);
361 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
363 /* Send the request to the server. */
365 write_error = fd_write (fd, request_string, size - 1, -1);
367 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
372 /* Release the resources used by REQ. */
375 request_free (struct request *req)
378 xfree_null (req->arg);
379 for (i = 0; i < req->hcount; i++)
380 release_header (&req->headers[i]);
381 xfree_null (req->headers);
385 static struct hash_table *basic_authed_hosts;
387 /* Find out if this host has issued a Basic challenge yet; if so, give
388 * it the username, password. A temporary measure until we can get
389 * proper authentication in place. */
392 maybe_send_basic_creds (const char *hostname, const char *user,
393 const char *passwd, struct request *req)
395 int did_challenge = 0;
397 if (basic_authed_hosts
398 && hash_table_contains(basic_authed_hosts, hostname))
400 DEBUGP(("Found `%s' in basic_authed_hosts.\n", hostname));
401 request_set_header (req, "Authorization",
402 basic_authentication_encode (user, passwd),
408 DEBUGP(("Host `%s' has not issued a general basic challenge.\n",
411 return did_challenge;
415 register_basic_auth_host (const char *hostname)
417 if (!basic_authed_hosts)
419 basic_authed_hosts = make_nocase_string_hash_table (1);
421 if (!hash_table_contains(basic_authed_hosts, hostname))
423 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
424 DEBUGP(("Inserted `%s' into basic_authed_hosts\n", hostname));
429 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
430 PROMISED_SIZE bytes are sent over the wire -- if the file is
431 longer, read only that much; if the file is shorter, report an error. */
434 post_file (int sock, const char *file_name, wgint promised_size)
436 static char chunk[8192];
441 DEBUGP (("[writing POST file %s ... ", file_name));
443 fp = fopen (file_name, "rb");
446 while (!feof (fp) && written < promised_size)
449 int length = fread (chunk, 1, sizeof (chunk), fp);
452 towrite = MIN (promised_size - written, length);
453 write_error = fd_write (sock, chunk, towrite, -1);
463 /* If we've written less than was promised, report a (probably
464 nonsensical) error rather than break the promise. */
465 if (written < promised_size)
471 assert (written == promised_size);
472 DEBUGP (("done]\n"));
476 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
477 If so, return the pointer to the position after the line, otherwise
478 return NULL. This is used as callback to fd_read_hunk. The data
479 between START and PEEKED has been read and cannot be "unread"; the
480 data after PEEKED has only been peeked. */
483 response_head_terminator (const char *start, const char *peeked, int peeklen)
487 /* If at first peek, verify whether HUNK starts with "HTTP". If
488 not, this is a HTTP/0.9 request and we must bail out without
490 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
493 /* Look for "\n[\r]\n", and return the following position if found.
494 Start two chars before the current to cover the possibility that
495 part of the terminator (e.g. "\n\r") arrived in the previous
497 p = peeked - start < 2 ? start : peeked - 2;
498 end = peeked + peeklen;
500 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
501 for (; p < end - 2; p++)
504 if (p[1] == '\r' && p[2] == '\n')
506 else if (p[1] == '\n')
509 /* p==end-2: check for \n\n directly preceding END. */
510 if (p[0] == '\n' && p[1] == '\n')
516 /* The maximum size of a single HTTP response we care to read. Rather
517 than being a limit of the reader implementation, this limit
518 prevents Wget from slurping all available memory upon encountering
519 malicious or buggy server output, thus protecting the user. Define
520 it to 0 to remove the limit. */
522 #define HTTP_RESPONSE_MAX_SIZE 65536
524 /* Read the HTTP request head from FD and return it. The error
525 conditions are the same as with fd_read_hunk.
527 To support HTTP/0.9 responses, this function tries to make sure
528 that the data begins with "HTTP". If this is not the case, no data
529 is read and an empty request is returned, so that the remaining
530 data can be treated as body. */
533 read_http_response_head (int fd)
535 return fd_read_hunk (fd, response_head_terminator, 512,
536 HTTP_RESPONSE_MAX_SIZE);
540 /* The response data. */
543 /* The array of pointers that indicate where each header starts.
544 For example, given this HTTP response:
551 The headers are located like this:
553 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
555 headers[0] headers[1] headers[2] headers[3]
557 I.e. headers[0] points to the beginning of the request,
558 headers[1] points to the end of the first header and the
559 beginning of the second one, etc. */
561 const char **headers;
564 /* Create a new response object from the text of the HTTP response,
565 available in HEAD. That text is automatically split into
566 constituent header lines for fast retrieval using
569 static struct response *
570 resp_new (const char *head)
575 struct response *resp = xnew0 (struct response);
580 /* Empty head means that we're dealing with a headerless
581 (HTTP/0.9) response. In that case, don't set HEADERS at
586 /* Split HEAD into header lines, so that resp_header_* functions
587 don't need to do this over and over again. */
593 DO_REALLOC (resp->headers, size, count + 1, const char *);
594 resp->headers[count++] = hdr;
596 /* Break upon encountering an empty line. */
597 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
600 /* Find the end of HDR, including continuations. */
603 const char *end = strchr (hdr, '\n');
609 while (*hdr == ' ' || *hdr == '\t');
611 DO_REALLOC (resp->headers, size, count + 1, const char *);
612 resp->headers[count] = NULL;
617 /* Locate the header named NAME in the request data, starting with
618 position START. This allows the code to loop through the request
619 data, filtering for all requests of a given name. Returns the
620 found position, or -1 for failure. The code that uses this
621 function typically looks like this:
623 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
624 ... do something with header ...
626 If you only care about one header, use resp_header_get instead of
630 resp_header_locate (const struct response *resp, const char *name, int start,
631 const char **begptr, const char **endptr)
634 const char **headers = resp->headers;
637 if (!headers || !headers[1])
640 name_len = strlen (name);
646 for (; headers[i + 1]; i++)
648 const char *b = headers[i];
649 const char *e = headers[i + 1];
651 && b[name_len] == ':'
652 && 0 == strncasecmp (b, name, name_len))
655 while (b < e && ISSPACE (*b))
657 while (b < e && ISSPACE (e[-1]))
667 /* Find and retrieve the header named NAME in the request data. If
668 found, set *BEGPTR to its starting, and *ENDPTR to its ending
669 position, and return true. Otherwise return false.
671 This function is used as a building block for resp_header_copy
672 and resp_header_strdup. */
675 resp_header_get (const struct response *resp, const char *name,
676 const char **begptr, const char **endptr)
678 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
682 /* Copy the response header named NAME to buffer BUF, no longer than
683 BUFSIZE (BUFSIZE includes the terminating 0). If the header
684 exists, true is returned, false otherwise. If there should be no
685 limit on the size of the header, use resp_header_strdup instead.
687 If BUFSIZE is 0, no data is copied, but the boolean indication of
688 whether the header is present is still returned. */
691 resp_header_copy (const struct response *resp, const char *name,
692 char *buf, int bufsize)
695 if (!resp_header_get (resp, name, &b, &e))
699 int len = MIN (e - b, bufsize - 1);
700 memcpy (buf, b, len);
706 /* Return the value of header named NAME in RESP, allocated with
707 malloc. If such a header does not exist in RESP, return NULL. */
710 resp_header_strdup (const struct response *resp, const char *name)
713 if (!resp_header_get (resp, name, &b, &e))
715 return strdupdelim (b, e);
718 /* Parse the HTTP status line, which is of format:
720 HTTP-Version SP Status-Code SP Reason-Phrase
722 The function returns the status-code, or -1 if the status line
723 appears malformed. The pointer to "reason-phrase" message is
724 returned in *MESSAGE. */
727 resp_status (const struct response *resp, char **message)
734 /* For a HTTP/0.9 response, assume status 200. */
736 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
740 p = resp->headers[0];
741 end = resp->headers[1];
747 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
751 /* Match the HTTP version. This is optional because Gnutella
752 servers have been reported to not specify HTTP version. */
753 if (p < end && *p == '/')
756 while (p < end && ISDIGIT (*p))
758 if (p < end && *p == '.')
760 while (p < end && ISDIGIT (*p))
764 while (p < end && ISSPACE (*p))
766 if (end - p < 3 || !ISDIGIT (p[0]) || !ISDIGIT (p[1]) || !ISDIGIT (p[2]))
769 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
774 while (p < end && ISSPACE (*p))
776 while (p < end && ISSPACE (end[-1]))
778 *message = strdupdelim (p, end);
784 /* Release the resources used by RESP. */
787 resp_free (struct response *resp)
789 xfree_null (resp->headers);
793 /* Print a single line of response, the characters [b, e). We tried
795 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
796 but that failed to escape the non-printable characters and, in fact,
797 caused crashes in UTF-8 locales. */
800 print_response_line(const char *prefix, const char *b, const char *e)
803 BOUNDED_TO_ALLOCA(b, e, copy);
804 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
807 /* Print the server response, line by line, omitting the trailing CRLF
808 from individual header lines, and prefixed with PREFIX. */
811 print_server_response (const struct response *resp, const char *prefix)
816 for (i = 0; resp->headers[i + 1]; i++)
818 const char *b = resp->headers[i];
819 const char *e = resp->headers[i + 1];
821 if (b < e && e[-1] == '\n')
823 if (b < e && e[-1] == '\r')
825 print_response_line(prefix, b, e);
829 /* Parse the `Content-Range' header and extract the information it
830 contains. Returns true if successful, false otherwise. */
832 parse_content_range (const char *hdr, wgint *first_byte_ptr,
833 wgint *last_byte_ptr, wgint *entity_length_ptr)
837 /* Ancient versions of Netscape proxy server, presumably predating
838 rfc2068, sent out `Content-Range' without the "bytes"
840 if (0 == strncasecmp (hdr, "bytes", 5))
843 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
847 while (ISSPACE (*hdr))
854 for (num = 0; ISDIGIT (*hdr); hdr++)
855 num = 10 * num + (*hdr - '0');
856 if (*hdr != '-' || !ISDIGIT (*(hdr + 1)))
858 *first_byte_ptr = num;
860 for (num = 0; ISDIGIT (*hdr); hdr++)
861 num = 10 * num + (*hdr - '0');
862 if (*hdr != '/' || !ISDIGIT (*(hdr + 1)))
864 *last_byte_ptr = num;
866 for (num = 0; ISDIGIT (*hdr); hdr++)
867 num = 10 * num + (*hdr - '0');
868 *entity_length_ptr = num;
872 /* Read the body of the request, but don't store it anywhere and don't
873 display a progress gauge. This is useful for reading the bodies of
874 administrative responses to which we will soon issue another
875 request. The response is not useful to the user, but reading it
876 allows us to continue using the same connection to the server.
878 If reading fails, false is returned, true otherwise. In debug
879 mode, the body is displayed for debugging purposes. */
882 skip_short_body (int fd, wgint contlen)
885 SKIP_SIZE = 512, /* size of the download buffer */
886 SKIP_THRESHOLD = 4096 /* the largest size we read */
888 char dlbuf[SKIP_SIZE + 1];
889 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
891 /* We shouldn't get here with unknown contlen. (This will change
892 with HTTP/1.1, which supports "chunked" transfer.) */
893 assert (contlen != -1);
895 /* If the body is too large, it makes more sense to simply close the
896 connection than to try to read the body. */
897 if (contlen > SKIP_THRESHOLD)
900 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
904 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
907 /* Don't normally report the error since this is an
908 optimization that should be invisible to the user. */
909 DEBUGP (("] aborting (%s).\n",
910 ret < 0 ? fd_errstr (fd) : "EOF received"));
914 /* Safe even if %.*s bogusly expects terminating \0 because
915 we've zero-terminated dlbuf above. */
916 DEBUGP (("%.*s", ret, dlbuf));
919 DEBUGP (("] done.\n"));
923 /* Extract a parameter from the string (typically an HTTP header) at
924 **SOURCE and advance SOURCE to the next parameter. Return false
925 when there are no more parameters to extract. The name of the
926 parameter is returned in NAME, and the value in VALUE. If the
927 parameter has no value, the token's value is zeroed out.
929 For example, if *SOURCE points to the string "attachment;
930 filename=\"foo bar\"", the first call to this function will return
931 the token named "attachment" and no value, and the second call will
932 return the token named "filename" and value "foo bar". The third
933 call will return false, indicating no more valid tokens. */
936 extract_param (const char **source, param_token *name, param_token *value,
939 const char *p = *source;
941 while (ISSPACE (*p)) ++p;
945 return false; /* no error; nothing more to extract */
950 while (*p && !ISSPACE (*p) && *p != '=' && *p != separator) ++p;
952 if (name->b == name->e)
953 return false; /* empty name: error */
954 while (ISSPACE (*p)) ++p;
955 if (*p == separator || !*p) /* no value */
958 if (*p == separator) ++p;
963 return false; /* error */
965 /* *p is '=', extract value */
967 while (ISSPACE (*p)) ++p;
968 if (*p == '"') /* quoted */
971 while (*p && *p != '"') ++p;
975 /* Currently at closing quote; find the end of param. */
976 while (ISSPACE (*p)) ++p;
977 while (*p && *p != separator) ++p;
981 /* garbage after closed quote, e.g. foo="bar"baz */
987 while (*p && *p != separator) ++p;
989 while (value->e != value->b && ISSPACE (value->e[-1]))
991 if (*p == separator) ++p;
998 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1000 /* Parse the contents of the `Content-Disposition' header, extracting
1001 the information useful to Wget. Content-Disposition is a header
1002 borrowed from MIME; when used in HTTP, it typically serves for
1003 specifying the desired file name of the resource. For example:
1005 Content-Disposition: attachment; filename="flora.jpg"
1007 Wget will skip the tokens it doesn't care about, such as
1008 "attachment" in the previous example; it will also skip other
1009 unrecognized params. If the header is syntactically correct and
1010 contains a file name, a copy of the file name is stored in
1011 *filename and true is returned. Otherwise, the function returns
1014 The file name is stripped of directory components and must not be
1018 parse_content_disposition (const char *hdr, char **filename)
1020 param_token name, value;
1021 while (extract_param (&hdr, &name, &value, ';'))
1022 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1024 /* Make the file name begin at the last slash or backslash. */
1025 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1026 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1027 if (last_slash && last_bs)
1028 value.b = 1 + MAX (last_slash, last_bs);
1029 else if (last_slash || last_bs)
1030 value.b = 1 + (last_slash ? last_slash : last_bs);
1031 if (value.b == value.e)
1033 /* Start with the directory prefix, if specified. */
1036 int prefix_length = strlen (opt.dir_prefix);
1037 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1042 total_length = prefix_length + (value.e - value.b);
1043 *filename = xmalloc (total_length + 1);
1044 strcpy (*filename, opt.dir_prefix);
1046 (*filename)[prefix_length - 1] = '/';
1047 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1048 (*filename)[total_length] = '\0';
1051 *filename = strdupdelim (value.b, value.e);
1057 /* Persistent connections. Currently, we cache the most recently used
1058 connection as persistent, provided that the HTTP server agrees to
1059 make it such. The persistence data is stored in the variables
1060 below. Ideally, it should be possible to cache an arbitrary fixed
1061 number of these connections. */
1063 /* Whether a persistent connection is active. */
1064 static bool pconn_active;
1067 /* The socket of the connection. */
1070 /* Host and port of the currently active persistent connection. */
1074 /* Whether a ssl handshake has occoured on this connection. */
1077 /* Whether the connection was authorized. This is only done by
1078 NTLM, which authorizes *connections* rather than individual
1079 requests. (That practice is peculiar for HTTP, but it is a
1080 useful optimization.) */
1084 /* NTLM data of the current connection. */
1085 struct ntlmdata ntlm;
1089 /* Mark the persistent connection as invalid and free the resources it
1090 uses. This is used by the CLOSE_* macros after they forcefully
1091 close a registered persistent connection. */
1094 invalidate_persistent (void)
1096 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1097 pconn_active = false;
1098 fd_close (pconn.socket);
1103 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1104 persistent. This will enable someone to use the same connection
1105 later. In the context of HTTP, this must be called only AFTER the
1106 response has been received and the server has promised that the
1107 connection will remain alive.
1109 If a previous connection was persistent, it is closed. */
1112 register_persistent (const char *host, int port, int fd, bool ssl)
1116 if (pconn.socket == fd)
1118 /* The connection FD is already registered. */
1123 /* The old persistent connection is still active; close it
1124 first. This situation arises whenever a persistent
1125 connection exists, but we then connect to a different
1126 host, and try to register a persistent connection to that
1128 invalidate_persistent ();
1132 pconn_active = true;
1134 pconn.host = xstrdup (host);
1137 pconn.authorized = false;
1139 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1142 /* Return true if a persistent connection is available for connecting
1146 persistent_available_p (const char *host, int port, bool ssl,
1147 bool *host_lookup_failed)
1149 /* First, check whether a persistent connection is active at all. */
1153 /* If we want SSL and the last connection wasn't or vice versa,
1154 don't use it. Checking for host and port is not enough because
1155 HTTP and HTTPS can apparently coexist on the same port. */
1156 if (ssl != pconn.ssl)
1159 /* If we're not connecting to the same port, we're not interested. */
1160 if (port != pconn.port)
1163 /* If the host is the same, we're in business. If not, there is
1164 still hope -- read below. */
1165 if (0 != strcasecmp (host, pconn.host))
1167 /* Check if pconn.socket is talking to HOST under another name.
1168 This happens often when both sites are virtual hosts
1169 distinguished only by name and served by the same network
1170 interface, and hence the same web server (possibly set up by
1171 the ISP and serving many different web sites). This
1172 admittedly unconventional optimization does not contradict
1173 HTTP and works well with popular server software. */
1177 struct address_list *al;
1180 /* Don't try to talk to two different SSL sites over the same
1181 secure connection! (Besides, it's not clear that
1182 name-based virtual hosting is even possible with SSL.) */
1185 /* If pconn.socket's peer is one of the IP addresses HOST
1186 resolves to, pconn.socket is for all intents and purposes
1187 already talking to HOST. */
1189 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1191 /* Can't get the peer's address -- something must be very
1192 wrong with the connection. */
1193 invalidate_persistent ();
1196 al = lookup_host (host, 0);
1199 *host_lookup_failed = true;
1203 found = address_list_contains (al, &ip);
1204 address_list_release (al);
1209 /* The persistent connection's peer address was found among the
1210 addresses HOST resolved to; therefore, pconn.sock is in fact
1211 already talking to HOST -- no need to reconnect. */
1214 /* Finally, check whether the connection is still open. This is
1215 important because most servers implement liberal (short) timeout
1216 on persistent connections. Wget can of course always reconnect
1217 if the connection doesn't work out, but it's nicer to know in
1218 advance. This test is a logical followup of the first test, but
1219 is "expensive" and therefore placed at the end of the list.
1221 (Current implementation of test_socket_open has a nice side
1222 effect that it treats sockets with pending data as "closed".
1223 This is exactly what we want: if a broken server sends message
1224 body in response to HEAD, or if it sends more than conent-length
1225 data, we won't reuse the corrupted connection.) */
1227 if (!test_socket_open (pconn.socket))
1229 /* Oops, the socket is no longer open. Now that we know that,
1230 let's invalidate the persistent connection before returning
1232 invalidate_persistent ();
1239 /* The idea behind these two CLOSE macros is to distinguish between
1240 two cases: one when the job we've been doing is finished, and we
1241 want to close the connection and leave, and two when something is
1242 seriously wrong and we're closing the connection as part of
1245 In case of keep_alive, CLOSE_FINISH should leave the connection
1246 open, while CLOSE_INVALIDATE should still close it.
1248 Note that the semantics of the flag `keep_alive' is "this
1249 connection *will* be reused (the server has promised not to close
1250 the connection once we're done)", while the semantics of
1251 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1252 active, registered connection". */
1254 #define CLOSE_FINISH(fd) do { \
1257 if (pconn_active && (fd) == pconn.socket) \
1258 invalidate_persistent (); \
1267 #define CLOSE_INVALIDATE(fd) do { \
1268 if (pconn_active && (fd) == pconn.socket) \
1269 invalidate_persistent (); \
1277 wgint len; /* received length */
1278 wgint contlen; /* expected length */
1279 wgint restval; /* the restart value */
1280 int res; /* the result of last read */
1281 char *rderrmsg; /* error message from read error */
1282 char *newloc; /* new location (redirection) */
1283 char *remote_time; /* remote time-stamp string */
1284 char *error; /* textual HTTP error */
1285 int statcode; /* status code */
1286 wgint rd_size; /* amount of data read from socket */
1287 double dltime; /* time it took to download the data */
1288 const char *referer; /* value of the referer header. */
1289 char *local_file; /* local file name. */
1290 bool timestamp_checked; /* true if pre-download time-stamping checks
1291 * have already been performed */
1292 char *orig_file_name; /* name of file to compare for time-stamping
1293 * (might be != local_file if -K is set) */
1294 wgint orig_file_size; /* size of file to compare for time-stamping */
1295 time_t orig_file_tstamp; /* time-stamp of file to compare for
1300 free_hstat (struct http_stat *hs)
1302 xfree_null (hs->newloc);
1303 xfree_null (hs->remote_time);
1304 xfree_null (hs->error);
1305 xfree_null (hs->rderrmsg);
1306 xfree_null (hs->local_file);
1307 xfree_null (hs->orig_file_name);
1309 /* Guard against being called twice. */
1311 hs->remote_time = NULL;
1315 #define BEGINS_WITH(line, string_constant) \
1316 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1317 && (ISSPACE (line[sizeof (string_constant) - 1]) \
1318 || !line[sizeof (string_constant) - 1]))
1320 #define SET_USER_AGENT(req) do { \
1321 if (!opt.useragent) \
1322 request_set_header (req, "User-Agent", \
1323 aprintf ("Wget/%s", version_string), rel_value); \
1324 else if (*opt.useragent) \
1325 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1328 /* The flags that allow clobbering the file (opening with "wb").
1329 Defined here to avoid repetition later. #### This will require
1331 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1332 || opt.dirstruct || opt.output_document)
1334 /* Retrieve a document through HTTP protocol. It recognizes status
1335 code, and correctly handles redirections. It closes the network
1336 socket. If it receives an error from the functions below it, it
1337 will print it if there is enough information to do so (almost
1338 always), returning the error to the caller (i.e. http_loop).
1340 Various HTTP parameters are stored to hs.
1342 If PROXY is non-NULL, the connection will be made to the proxy
1343 server, and u->url will be requested. */
1345 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1347 struct request *req;
1350 char *user, *passwd;
1354 wgint contlen, contrange;
1361 /* Set to 1 when the authorization has already been sent and should
1362 not be tried again. */
1363 bool auth_finished = false;
1365 /* Set to 1 when just globally-set Basic authorization has been sent;
1366 * should prevent further Basic negotiations, but not other
1368 bool basic_auth_finished = false;
1370 /* Whether NTLM authentication is used for this request. */
1371 bool ntlm_seen = false;
1373 /* Whether our connection to the remote host is through SSL. */
1374 bool using_ssl = false;
1376 /* Whether a HEAD request will be issued (as opposed to GET or
1378 bool head_only = !!(*dt & HEAD_ONLY);
1381 struct response *resp;
1385 /* Whether this connection will be kept alive after the HTTP request
1389 /* Whether keep-alive should be inhibited.
1391 RFC 2068 requests that 1.0 clients not send keep-alive requests
1392 to proxies. This is because many 1.0 proxies do not interpret
1393 the Connection header and transfer it to the remote server,
1394 causing it to not close the connection and leave both the proxy
1395 and the client hanging. */
1396 bool inhibit_keep_alive =
1397 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1399 /* Headers sent when using POST. */
1400 wgint post_data_size = 0;
1402 bool host_lookup_failed = false;
1405 if (u->scheme == SCHEME_HTTPS)
1407 /* Initialize the SSL context. After this has once been done,
1408 it becomes a no-op. */
1411 scheme_disable (SCHEME_HTTPS);
1412 logprintf (LOG_NOTQUIET,
1413 _("Disabling SSL due to encountered errors.\n"));
1414 return SSLINITFAILED;
1417 #endif /* HAVE_SSL */
1419 /* Initialize certain elements of struct http_stat. */
1423 hs->rderrmsg = NULL;
1425 hs->remote_time = NULL;
1430 /* Prepare the request to send. */
1432 req = request_new ();
1435 const char *meth = "GET";
1438 else if (opt.post_file_name || opt.post_data)
1440 /* Use the full path, i.e. one that includes the leading slash and
1441 the query string. E.g. if u->path is "foo/bar" and u->query is
1442 "param=value", full_path will be "/foo/bar?param=value". */
1445 /* When using SSL over proxy, CONNECT establishes a direct
1446 connection to the HTTPS server. Therefore use the same
1447 argument as when talking to the server directly. */
1448 && u->scheme != SCHEME_HTTPS
1451 meth_arg = xstrdup (u->url);
1453 meth_arg = url_full_path (u);
1454 request_set_method (req, meth, meth_arg);
1457 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1458 if (*dt & SEND_NOCACHE)
1459 request_set_header (req, "Pragma", "no-cache", rel_none);
1461 request_set_header (req, "Range",
1462 aprintf ("bytes=%s-",
1463 number_to_static_string (hs->restval)),
1465 SET_USER_AGENT (req);
1466 request_set_header (req, "Accept", "*/*", rel_none);
1468 /* Find the username and password for authentication. */
1471 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1472 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1473 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1476 && !u->user) /* We only do "site-wide" authentication with "global"
1477 user/password values; URL user/password info overrides. */
1479 /* If this is a host for which we've already received a Basic
1480 * challenge, we'll go ahead and send Basic authentication creds. */
1481 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1487 char *proxy_user, *proxy_passwd;
1488 /* For normal username and password, URL components override
1489 command-line/wgetrc parameters. With proxy
1490 authentication, it's the reverse, because proxy URLs are
1491 normally the "permanent" ones, so command-line args
1492 should take precedence. */
1493 if (opt.proxy_user && opt.proxy_passwd)
1495 proxy_user = opt.proxy_user;
1496 proxy_passwd = opt.proxy_passwd;
1500 proxy_user = proxy->user;
1501 proxy_passwd = proxy->passwd;
1503 /* #### This does not appear right. Can't the proxy request,
1504 say, `Digest' authentication? */
1505 if (proxy_user && proxy_passwd)
1506 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1508 /* If we're using a proxy, we will be connecting to the proxy
1512 /* Proxy authorization over SSL is handled below. */
1514 if (u->scheme != SCHEME_HTTPS)
1516 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1519 /* Generate the Host header, HOST:PORT. Take into account that:
1521 - Broken server-side software often doesn't recognize the PORT
1522 argument, so we must generate "Host: www.server.com" instead of
1523 "Host: www.server.com:80" (and likewise for https port).
1525 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1526 becomes ambiguous and needs to be rewritten as "Host:
1527 [3ffe:8100:200:2::2]:1234". */
1529 /* Formats arranged for hfmt[add_port][add_squares]. */
1530 static const char *hfmt[][2] = {
1531 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1533 int add_port = u->port != scheme_default_port (u->scheme);
1534 int add_squares = strchr (u->host, ':') != NULL;
1535 request_set_header (req, "Host",
1536 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1540 if (!inhibit_keep_alive)
1541 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1544 request_set_header (req, "Cookie",
1545 cookie_header (wget_cookie_jar,
1546 u->host, u->port, u->path,
1548 u->scheme == SCHEME_HTTPS
1555 if (opt.post_data || opt.post_file_name)
1557 request_set_header (req, "Content-Type",
1558 "application/x-www-form-urlencoded", rel_none);
1560 post_data_size = strlen (opt.post_data);
1563 post_data_size = file_size (opt.post_file_name);
1564 if (post_data_size == -1)
1566 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1567 opt.post_file_name, strerror (errno));
1571 request_set_header (req, "Content-Length",
1572 xstrdup (number_to_static_string (post_data_size)),
1576 /* Add the user headers. */
1577 if (opt.user_headers)
1580 for (i = 0; opt.user_headers[i]; i++)
1581 request_set_user_header (req, opt.user_headers[i]);
1585 /* We need to come back here when the initial attempt to retrieve
1586 without authorization header fails. (Expected to happen at least
1587 for the Digest authorization scheme.) */
1591 /* Establish the connection. */
1593 if (!inhibit_keep_alive)
1595 /* Look for a persistent connection to target host, unless a
1596 proxy is used. The exception is when SSL is in use, in which
1597 case the proxy is nothing but a passthrough to the target
1598 host, registered as a connection to the latter. */
1599 struct url *relevant = conn;
1601 if (u->scheme == SCHEME_HTTPS)
1605 if (persistent_available_p (relevant->host, relevant->port,
1607 relevant->scheme == SCHEME_HTTPS,
1611 &host_lookup_failed))
1613 sock = pconn.socket;
1614 using_ssl = pconn.ssl;
1615 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1616 escnonprint (pconn.host), pconn.port);
1617 DEBUGP (("Reusing fd %d.\n", sock));
1618 if (pconn.authorized)
1619 /* If the connection is already authorized, the "Basic"
1620 authorization added by code above is unnecessary and
1622 request_remove_header (req, "Authorization");
1628 /* In its current implementation, persistent_available_p will
1629 look up conn->host in some cases. If that lookup failed, we
1630 don't need to bother with connect_to_host. */
1631 if (host_lookup_failed)
1637 sock = connect_to_host (conn->host, conn->port);
1646 return (retryable_socket_connect_error (errno)
1647 ? CONERROR : CONIMPOSSIBLE);
1651 if (proxy && u->scheme == SCHEME_HTTPS)
1653 /* When requesting SSL URLs through proxies, use the
1654 CONNECT method to request passthrough. */
1655 struct request *connreq = request_new ();
1656 request_set_method (connreq, "CONNECT",
1657 aprintf ("%s:%d", u->host, u->port));
1658 SET_USER_AGENT (connreq);
1661 request_set_header (connreq, "Proxy-Authorization",
1662 proxyauth, rel_value);
1663 /* Now that PROXYAUTH is part of the CONNECT request,
1664 zero it out so we don't send proxy authorization with
1665 the regular request below. */
1668 /* Examples in rfc2817 use the Host header in CONNECT
1669 requests. I don't see how that gains anything, given
1670 that the contents of Host would be exactly the same as
1671 the contents of CONNECT. */
1673 write_error = request_send (connreq, sock);
1674 request_free (connreq);
1675 if (write_error < 0)
1677 CLOSE_INVALIDATE (sock);
1681 head = read_http_response_head (sock);
1684 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1686 CLOSE_INVALIDATE (sock);
1695 DEBUGP (("proxy responded with: [%s]\n", head));
1697 resp = resp_new (head);
1698 statcode = resp_status (resp, &message);
1701 if (statcode != 200)
1704 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1705 message ? escnonprint (message) : "?");
1706 xfree_null (message);
1709 xfree_null (message);
1711 /* SOCK is now *really* connected to u->host, so update CONN
1712 to reflect this. That way register_persistent will
1713 register SOCK as being connected to u->host:u->port. */
1717 if (conn->scheme == SCHEME_HTTPS)
1719 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1726 #endif /* HAVE_SSL */
1729 /* Send the request to server. */
1730 write_error = request_send (req, sock);
1732 if (write_error >= 0)
1736 DEBUGP (("[POST data: %s]\n", opt.post_data));
1737 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1739 else if (opt.post_file_name && post_data_size != 0)
1740 write_error = post_file (sock, opt.post_file_name, post_data_size);
1743 if (write_error < 0)
1745 CLOSE_INVALIDATE (sock);
1749 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1750 proxy ? "Proxy" : "HTTP");
1755 head = read_http_response_head (sock);
1760 logputs (LOG_NOTQUIET, _("No data received.\n"));
1761 CLOSE_INVALIDATE (sock);
1767 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1769 CLOSE_INVALIDATE (sock);
1774 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1776 resp = resp_new (head);
1778 /* Check for status line. */
1780 statcode = resp_status (resp, &message);
1781 if (!opt.server_response)
1782 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1783 message ? escnonprint (message) : "");
1786 logprintf (LOG_VERBOSE, "\n");
1787 print_server_response (resp, " ");
1790 /* Determine the local filename if needed. Notice that if -O is used
1791 * hstat.local_file is set by http_loop to the argument of -O. */
1792 if (!hs->local_file)
1794 /* Honor Content-Disposition whether possible. */
1795 if (!opt.content_disposition
1796 || !resp_header_copy (resp, "Content-Disposition",
1797 hdrval, sizeof (hdrval))
1798 || !parse_content_disposition (hdrval, &hs->local_file))
1800 /* The Content-Disposition header is missing or broken.
1801 * Choose unique file name according to given URL. */
1802 hs->local_file = url_file_name (u);
1806 /* TODO: perform this check only once. */
1807 if (file_exists_p (hs->local_file))
1811 /* If opt.noclobber is turned on and file already exists, do not
1812 retrieve the file */
1813 logprintf (LOG_VERBOSE, _("\
1814 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1815 /* If the file is there, we suppose it's retrieved OK. */
1818 /* #### Bogusness alert. */
1819 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1820 if (has_html_suffix_p (hs->local_file))
1825 else if (!ALLOW_CLOBBER)
1827 char *unique = unique_name (hs->local_file, true);
1828 if (unique != hs->local_file)
1829 xfree (hs->local_file);
1830 hs->local_file = unique;
1834 /* Support timestamping */
1835 /* TODO: move this code out of gethttp. */
1836 if (opt.timestamping && !hs->timestamp_checked)
1838 size_t filename_len = strlen (hs->local_file);
1839 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1840 bool local_dot_orig_file_exists = false;
1841 char *local_filename = NULL;
1844 if (opt.backup_converted)
1845 /* If -K is specified, we'll act on the assumption that it was specified
1846 last time these files were downloaded as well, and instead of just
1847 comparing local file X against server file X, we'll compare local
1848 file X.orig (if extant, else X) against server file X. If -K
1849 _wasn't_ specified last time, or the server contains files called
1850 *.orig, -N will be back to not operating correctly with -k. */
1852 /* Would a single s[n]printf() call be faster? --dan
1854 Definitely not. sprintf() is horribly slow. It's a
1855 different question whether the difference between the two
1856 affects a program. Usually I'd say "no", but at one
1857 point I profiled Wget, and found that a measurable and
1858 non-negligible amount of time was lost calling sprintf()
1859 in url.c. Replacing sprintf with inline calls to
1860 strcpy() and number_to_string() made a difference.
1862 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1863 memcpy (filename_plus_orig_suffix + filename_len,
1864 ".orig", sizeof (".orig"));
1866 /* Try to stat() the .orig file. */
1867 if (stat (filename_plus_orig_suffix, &st) == 0)
1869 local_dot_orig_file_exists = true;
1870 local_filename = filename_plus_orig_suffix;
1874 if (!local_dot_orig_file_exists)
1875 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1876 if (stat (hs->local_file, &st) == 0)
1877 local_filename = hs->local_file;
1879 if (local_filename != NULL)
1880 /* There was a local file, so we'll check later to see if the version
1881 the server has is the same version we already have, allowing us to
1884 hs->orig_file_name = xstrdup (local_filename);
1885 hs->orig_file_size = st.st_size;
1886 hs->orig_file_tstamp = st.st_mtime;
1888 /* Modification time granularity is 2 seconds for Windows, so
1889 increase local time by 1 second for later comparison. */
1890 ++hs->orig_file_tstamp;
1895 if (!opt.ignore_length
1896 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1900 parsed = str_to_wgint (hdrval, NULL, 10);
1901 if (parsed == WGINT_MAX && errno == ERANGE)
1903 #### If Content-Length is out of range, it most likely
1904 means that the file is larger than 2G and that we're
1905 compiled without LFS. In that case we should probably
1906 refuse to even attempt to download the file. */
1912 /* Check for keep-alive related responses. */
1913 if (!inhibit_keep_alive && contlen != -1)
1915 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1917 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1919 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1924 /* The server has promised that it will not close the connection
1925 when we're done. This means that we can register it. */
1926 register_persistent (conn->host, conn->port, sock, using_ssl);
1928 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1930 /* Authorization is required. */
1931 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1932 CLOSE_FINISH (sock);
1934 CLOSE_INVALIDATE (sock);
1935 pconn.authorized = false;
1936 if (!auth_finished && (user && passwd))
1938 /* IIS sends multiple copies of WWW-Authenticate, one with
1939 the value "negotiate", and other(s) with data. Loop over
1940 all the occurrences and pick the one we recognize. */
1942 const char *wabeg, *waend;
1943 char *www_authenticate = NULL;
1945 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1946 &wabeg, &waend)) != -1;
1948 if (known_authentication_scheme_p (wabeg, waend))
1950 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1954 if (!www_authenticate)
1956 /* If the authentication header is missing or
1957 unrecognized, there's no sense in retrying. */
1958 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1960 else if (!basic_auth_finished
1961 || !BEGINS_WITH (www_authenticate, "Basic"))
1964 pth = url_full_path (u);
1965 request_set_header (req, "Authorization",
1966 create_authorization_line (www_authenticate,
1968 request_method (req),
1972 if (BEGINS_WITH (www_authenticate, "NTLM"))
1974 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
1976 /* Need to register this host as using basic auth,
1977 * so we automatically send creds next time. */
1978 register_basic_auth_host (u->host);
1981 goto retry_with_auth;
1985 /* We already did Basic auth, and it failed. Gotta
1989 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
1993 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
1995 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
1997 pconn.authorized = true;
2001 hs->statcode = statcode;
2003 hs->error = xstrdup (_("Malformed status line"));
2005 hs->error = xstrdup (_("(no description)"));
2007 hs->error = xstrdup (message);
2008 xfree_null (message);
2010 type = resp_header_strdup (resp, "Content-Type");
2013 char *tmp = strchr (type, ';');
2016 while (tmp > type && ISSPACE (tmp[-1]))
2021 hs->newloc = resp_header_strdup (resp, "Location");
2022 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2024 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2028 const char *scbeg, *scend;
2029 /* The jar should have been created by now. */
2030 assert (wget_cookie_jar != NULL);
2032 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2033 &scbeg, &scend)) != -1;
2036 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2037 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2038 u->path, set_cookie);
2042 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2044 wgint first_byte_pos, last_byte_pos, entity_length;
2045 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2047 contrange = first_byte_pos;
2051 /* 20x responses are counted among successful by default. */
2052 if (H_20X (statcode))
2055 /* Return if redirected. */
2056 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2058 /* RFC2068 says that in case of the 300 (multiple choices)
2059 response, the server can output a preferred URL through
2060 `Location' header; otherwise, the request should be treated
2061 like GET. So, if the location is set, it will be a
2062 redirection; otherwise, just proceed normally. */
2063 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2067 logprintf (LOG_VERBOSE,
2068 _("Location: %s%s\n"),
2069 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2070 hs->newloc ? _(" [following]") : "");
2071 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2072 CLOSE_FINISH (sock);
2074 CLOSE_INVALIDATE (sock);
2080 /* If content-type is not given, assume text/html. This is because
2081 of the multitude of broken CGI's that "forget" to generate the
2084 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2085 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2090 if (opt.html_extension && (*dt & TEXTHTML))
2091 /* -E / --html-extension / html_extension = on was specified, and this is a
2092 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2093 already the file's suffix, tack on ".html". */
2095 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2097 if (last_period_in_local_filename == NULL
2098 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2099 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2101 int local_filename_len = strlen (hs->local_file);
2102 /* Resize the local file, allowing for ".html" preceded by
2103 optional ".NUMBER". */
2104 hs->local_file = xrealloc (hs->local_file,
2105 local_filename_len + 24 + sizeof (".html"));
2106 strcpy(hs->local_file + local_filename_len, ".html");
2107 /* If clobbering is not allowed and the file, as named,
2108 exists, tack on ".NUMBER.html" instead. */
2109 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2113 sprintf (hs->local_file + local_filename_len,
2114 ".%d.html", ext_num++);
2115 while (file_exists_p (hs->local_file));
2117 *dt |= ADDED_HTML_EXTENSION;
2121 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2123 /* If `-c' is in use and the file has been fully downloaded (or
2124 the remote file has shrunk), Wget effectively requests bytes
2125 after the end of file and the server response with 416. */
2126 logputs (LOG_VERBOSE, _("\
2127 \n The file is already fully retrieved; nothing to do.\n\n"));
2128 /* In case the caller inspects. */
2131 /* Mark as successfully retrieved. */
2134 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2135 might be more bytes in the body. */
2136 return RETRUNNEEDED;
2138 if ((contrange != 0 && contrange != hs->restval)
2139 || (H_PARTIAL (statcode) && !contrange))
2141 /* The Range request was somehow misunderstood by the server.
2144 CLOSE_INVALIDATE (sock);
2147 hs->contlen = contlen + contrange;
2153 /* No need to print this output if the body won't be
2154 downloaded at all, or if the original server response is
2156 logputs (LOG_VERBOSE, _("Length: "));
2159 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2160 if (contlen + contrange >= 1024)
2161 logprintf (LOG_VERBOSE, " (%s)",
2162 human_readable (contlen + contrange));
2165 if (contlen >= 1024)
2166 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2167 number_to_static_string (contlen),
2168 human_readable (contlen));
2170 logprintf (LOG_VERBOSE, _(", %s remaining"),
2171 number_to_static_string (contlen));
2175 logputs (LOG_VERBOSE,
2176 opt.ignore_length ? _("ignored") : _("unspecified"));
2178 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2180 logputs (LOG_VERBOSE, "\n");
2184 type = NULL; /* We don't need it any more. */
2186 /* Return if we have no intention of further downloading. */
2187 if (!(*dt & RETROKF) || head_only)
2189 /* In case the caller cares to look... */
2194 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2195 servers not to send body in response to a HEAD request, and
2196 those that do will likely be caught by test_socket_open.
2197 If not, they can be worked around using
2198 `--no-http-keep-alive'. */
2199 CLOSE_FINISH (sock);
2200 else if (keep_alive && skip_short_body (sock, contlen))
2201 /* Successfully skipped the body; also keep using the socket. */
2202 CLOSE_FINISH (sock);
2204 CLOSE_INVALIDATE (sock);
2205 return RETRFINISHED;
2208 /* Open the local file. */
2211 mkalldirs (hs->local_file);
2213 rotate_backups (hs->local_file);
2215 fp = fopen (hs->local_file, "ab");
2216 else if (ALLOW_CLOBBER)
2217 fp = fopen (hs->local_file, "wb");
2220 fp = fopen_excl (hs->local_file, true);
2221 if (!fp && errno == EEXIST)
2223 /* We cannot just invent a new name and use it (which is
2224 what functions like unique_create typically do)
2225 because we told the user we'd use this name.
2226 Instead, return and retry the download. */
2227 logprintf (LOG_NOTQUIET,
2228 _("%s has sprung into existence.\n"),
2230 CLOSE_INVALIDATE (sock);
2231 return FOPEN_EXCL_ERR;
2236 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2237 CLOSE_INVALIDATE (sock);
2244 /* Print fetch message, if opt.verbose. */
2247 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2248 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2251 /* This confuses the timestamping code that checks for file size.
2252 #### The timestamping code should be smarter about file size. */
2253 if (opt.save_headers && hs->restval == 0)
2254 fwrite (head, 1, strlen (head), fp);
2256 /* Now we no longer need to store the response header. */
2259 /* Download the request body. */
2262 /* If content-length is present, read that much; otherwise, read
2263 until EOF. The HTTP spec doesn't require the server to
2264 actually close the connection when it's done sending data. */
2265 flags |= rb_read_exactly;
2266 if (hs->restval > 0 && contrange == 0)
2267 /* If the server ignored our range request, instruct fd_read_body
2268 to skip the first RESTVAL bytes of body. */
2269 flags |= rb_skip_startpos;
2270 hs->len = hs->restval;
2272 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2273 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2277 CLOSE_FINISH (sock);
2281 hs->rderrmsg = xstrdup (fd_errstr (sock));
2282 CLOSE_INVALIDATE (sock);
2289 return RETRFINISHED;
2292 /* The genuine HTTP loop! This is the part where the retrieval is
2293 retried, and retried, and retried, and... */
2295 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2296 int *dt, struct url *proxy)
2299 bool got_head = false; /* used for time-stamping and filename detection */
2300 bool got_name = false;
2303 uerr_t err, ret = TRYLIMEXC;
2304 time_t tmr = -1; /* remote time-stamp */
2305 wgint local_size = 0; /* the size of the local file */
2306 struct http_stat hstat; /* HTTP status */
2309 /* Assert that no value for *LOCAL_FILE was passed. */
2310 assert (local_file == NULL || *local_file == NULL);
2312 /* Set LOCAL_FILE parameter. */
2313 if (local_file && opt.output_document)
2314 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2316 /* Reset NEWLOC parameter. */
2319 /* This used to be done in main(), but it's a better idea to do it
2320 here so that we don't go through the hoops if we're just using
2325 /* Warn on (likely bogus) wildcard usage in HTTP. */
2326 if (opt.ftp_glob && has_wildcards_p (u->path))
2327 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2329 /* Setup hstat struct. */
2331 hstat.referer = referer;
2333 if (opt.output_document)
2335 hstat.local_file = xstrdup (opt.output_document);
2339 /* Reset the counter. */
2342 /* Reset the document type. */
2348 /* Increment the pass counter. */
2350 sleep_between_retrievals (count);
2352 /* Get the current time string. */
2353 tms = time_str (time (NULL));
2355 if (opt.spider && !got_head)
2356 logprintf (LOG_VERBOSE, _("\
2357 Spider mode enabled. Check if remote file exists.\n"));
2359 /* Print fetch message, if opt.verbose. */
2362 char *hurl = url_string (u, true);
2367 sprintf (tmp, _("(try:%2d)"), count);
2368 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2373 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2378 ws_changetitle (hurl);
2383 /* Default document type is empty. However, if spider mode is
2384 on or time-stamping is employed, HEAD_ONLY commands is
2385 encoded within *dt. */
2386 if (((opt.spider || opt.timestamping) && !got_head)
2387 || (opt.always_rest && !got_name))
2392 /* Decide whether or not to restart. */
2395 && stat (hstat.local_file, &st) == 0
2396 && S_ISREG (st.st_mode))
2397 /* When -c is used, continue from on-disk size. (Can't use
2398 hstat.len even if count>1 because we don't want a failed
2399 first attempt to clobber existing data.) */
2400 hstat.restval = st.st_size;
2402 /* otherwise, continue where the previous try left off */
2403 hstat.restval = hstat.len;
2407 /* Decide whether to send the no-cache directive. We send it in
2409 a) we're using a proxy, and we're past our first retrieval.
2410 Some proxies are notorious for caching incomplete data, so
2411 we require a fresh get.
2412 b) caching is explicitly inhibited. */
2413 if ((proxy && count > 1) /* a */
2414 || !opt.allow_cache) /* b */
2415 *dt |= SEND_NOCACHE;
2417 *dt &= ~SEND_NOCACHE;
2419 /* Try fetching the document, or at least its head. */
2420 err = gethttp (u, &hstat, dt, proxy);
2423 tms = time_str (time (NULL));
2425 /* Get the new location (with or without the redirection). */
2427 *newloc = xstrdup (hstat.newloc);
2431 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2432 case CONERROR: case READERR: case WRITEFAILED:
2433 case RANGEERR: case FOPEN_EXCL_ERR:
2434 /* Non-fatal errors continue executing the loop, which will
2435 bring them to "while" statement at the end, to judge
2436 whether the number of tries was exceeded. */
2437 printwhat (count, opt.ntry);
2439 case FWRITEERR: case FOPENERR:
2440 /* Another fatal error. */
2441 logputs (LOG_VERBOSE, "\n");
2442 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2443 hstat.local_file, strerror (errno));
2444 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2445 case SSLINITFAILED: case CONTNOTSUPPORTED:
2446 /* Fatal errors just return from the function. */
2450 /* Another fatal error. */
2451 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2455 /* Return the new location to the caller. */
2458 logprintf (LOG_NOTQUIET,
2459 _("ERROR: Redirection (%d) without location.\n"),
2469 /* The file was already fully retrieved. */
2473 /* Deal with you later. */
2476 /* All possibilities should have been exhausted. */
2480 if (!(*dt & RETROKF))
2485 /* #### Ugly ugly ugly! */
2486 hurl = url_string (u, true);
2487 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2489 /* Maybe we should always keep track of broken links, not just in
2493 /* #### Again: ugly ugly ugly! */
2495 hurl = url_string (u, true);
2496 nonexisting_url (hurl);
2497 logprintf (LOG_NOTQUIET, _("\
2498 Remote file does not exist -- broken link!!!\n"));
2502 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2503 tms, hstat.statcode, escnonprint (hstat.error));
2505 logputs (LOG_VERBOSE, "\n");
2511 /* Did we get the time-stamp? */
2514 bool restart_loop = false;
2516 if (opt.timestamping && !hstat.remote_time)
2518 logputs (LOG_NOTQUIET, _("\
2519 Last-modified header missing -- time-stamps turned off.\n"));
2521 else if (hstat.remote_time)
2523 /* Convert the date-string into struct tm. */
2524 tmr = http_atotm (hstat.remote_time);
2525 if (tmr == (time_t) (-1))
2526 logputs (LOG_VERBOSE, _("\
2527 Last-modified header invalid -- time-stamp ignored.\n"));
2530 /* The time-stamping section. */
2531 if (opt.timestamping)
2533 if (hstat.orig_file_name) /* Perform the following checks only
2534 if the file we're supposed to
2535 download already exists. */
2537 if (hstat.remote_time &&
2538 tmr != (time_t) (-1))
2540 /* Now time-stamping can be used validly. Time-stamping
2541 means that if the sizes of the local and remote file
2542 match, and local file is newer than the remote file,
2543 it will not be retrieved. Otherwise, the normal
2544 download procedure is resumed. */
2545 if (hstat.orig_file_tstamp >= tmr)
2547 if (hstat.contlen == -1
2548 || hstat.orig_file_size == hstat.contlen)
2550 logprintf (LOG_VERBOSE, _("\
2551 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2552 hstat.orig_file_name);
2558 logprintf (LOG_VERBOSE, _("\
2559 The sizes do not match (local %s) -- retrieving.\n"),
2560 number_to_static_string (local_size));
2564 logputs (LOG_VERBOSE,
2565 _("Remote file is newer, retrieving.\n"));
2567 logputs (LOG_VERBOSE, "\n");
2571 /* free_hstat (&hstat); */
2572 hstat.timestamp_checked = true;
2573 restart_loop = true;
2576 if (opt.always_rest)
2579 restart_loop = true;
2588 logputs (LOG_VERBOSE, _("\
2589 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2590 restart_loop = true;
2594 logprintf (LOG_VERBOSE, _("\
2595 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2602 logprintf (LOG_VERBOSE, _("\
2603 Remote file exists but recursion is disabled -- not retrieving.\n\n"));
2609 got_head = true; /* no more time-stamping */
2611 count = 0; /* the retrieve count for HEAD is reset */
2617 if ((tmr != (time_t) (-1))
2618 && ((hstat.len == hstat.contlen) ||
2619 ((hstat.res == 0) && (hstat.contlen == -1))))
2621 /* #### This code repeats in http.c and ftp.c. Move it to a
2623 const char *fl = NULL;
2624 if (opt.output_document)
2626 if (output_stream_regular)
2627 fl = opt.output_document;
2630 fl = hstat.local_file;
2634 /* End of time-stamping section. */
2636 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2637 total_download_time += hstat.dltime;
2639 if (hstat.len == hstat.contlen)
2643 logprintf (LOG_VERBOSE,
2644 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2645 tms, tmrate, hstat.local_file,
2646 number_to_static_string (hstat.len),
2647 number_to_static_string (hstat.contlen));
2648 logprintf (LOG_NONVERBOSE,
2649 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2651 number_to_static_string (hstat.len),
2652 number_to_static_string (hstat.contlen),
2653 hstat.local_file, count);
2656 total_downloaded_bytes += hstat.len;
2658 /* Remember that we downloaded the file for later ".orig" code. */
2659 if (*dt & ADDED_HTML_EXTENSION)
2660 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2662 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2667 else if (hstat.res == 0) /* No read error */
2669 if (hstat.contlen == -1) /* We don't know how much we were supposed
2670 to get, so assume we succeeded. */
2674 logprintf (LOG_VERBOSE,
2675 _("%s (%s) - `%s' saved [%s]\n\n"),
2676 tms, tmrate, hstat.local_file,
2677 number_to_static_string (hstat.len));
2678 logprintf (LOG_NONVERBOSE,
2679 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2680 tms, u->url, number_to_static_string (hstat.len),
2681 hstat.local_file, count);
2684 total_downloaded_bytes += hstat.len;
2686 /* Remember that we downloaded the file for later ".orig" code. */
2687 if (*dt & ADDED_HTML_EXTENSION)
2688 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2690 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2695 else if (hstat.len < hstat.contlen) /* meaning we lost the
2696 connection too soon */
2698 logprintf (LOG_VERBOSE,
2699 _("%s (%s) - Connection closed at byte %s. "),
2700 tms, tmrate, number_to_static_string (hstat.len));
2701 printwhat (count, opt.ntry);
2705 /* Getting here would mean reading more data than
2706 requested with content-length, which we never do. */
2709 else /* from now on hstat.res can only be -1 */
2711 if (hstat.contlen == -1)
2713 logprintf (LOG_VERBOSE,
2714 _("%s (%s) - Read error at byte %s (%s)."),
2715 tms, tmrate, number_to_static_string (hstat.len),
2717 printwhat (count, opt.ntry);
2720 else /* hstat.res == -1 and contlen is given */
2722 logprintf (LOG_VERBOSE,
2723 _("%s (%s) - Read error at byte %s/%s (%s). "),
2725 number_to_static_string (hstat.len),
2726 number_to_static_string (hstat.contlen),
2728 printwhat (count, opt.ntry);
2734 while (!opt.ntry || (count < opt.ntry));
2738 *local_file = xstrdup (hstat.local_file);
2739 free_hstat (&hstat);
2744 /* Check whether the result of strptime() indicates success.
2745 strptime() returns the pointer to how far it got to in the string.
2746 The processing has been successful if the string is at `GMT' or
2747 `+X', or at the end of the string.
2749 In extended regexp parlance, the function returns 1 if P matches
2750 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2751 can return) is considered a failure and 0 is returned. */
2753 check_end (const char *p)
2757 while (ISSPACE (*p))
2760 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2761 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2767 /* Convert the textual specification of time in TIME_STRING to the
2768 number of seconds since the Epoch.
2770 TIME_STRING can be in any of the three formats RFC2616 allows the
2771 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2772 as well as the time format used in the Set-Cookie header.
2773 Timezones are ignored, and should be GMT.
2775 Return the computed time_t representation, or -1 if the conversion
2778 This function uses strptime with various string formats for parsing
2779 TIME_STRING. This results in a parser that is not as lenient in
2780 interpreting TIME_STRING as I would like it to be. Being based on
2781 strptime, it always allows shortened months, one-digit days, etc.,
2782 but due to the multitude of formats in which time can be
2783 represented, an ideal HTTP time parser would be even more
2784 forgiving. It should completely ignore things like week days and
2785 concentrate only on the various forms of representing years,
2786 months, days, hours, minutes, and seconds. For example, it would
2787 be nice if it accepted ISO 8601 out of the box.
2789 I've investigated free and PD code for this purpose, but none was
2790 usable. getdate was big and unwieldy, and had potential copyright
2791 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2792 distributed with phttpd, is excellent, but we cannot use it because
2793 it is not assigned to the FSF. So I stuck it with strptime. */
2796 http_atotm (const char *time_string)
2798 /* NOTE: Solaris strptime man page claims that %n and %t match white
2799 space, but that's not universally available. Instead, we simply
2800 use ` ' to mean "skip all WS", which works under all strptime
2801 implementations I've tested. */
2803 static const char *time_formats[] = {
2804 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2805 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2806 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2807 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2808 (used in Set-Cookie, defined in the
2809 Netscape cookie specification.) */
2811 const char *oldlocale;
2813 time_t ret = (time_t) -1;
2815 /* Solaris strptime fails to recognize English month names in
2816 non-English locales, which we work around by temporarily setting
2817 locale to C before invoking strptime. */
2818 oldlocale = setlocale (LC_TIME, NULL);
2819 setlocale (LC_TIME, "C");
2821 for (i = 0; i < countof (time_formats); i++)
2825 /* Some versions of strptime use the existing contents of struct
2826 tm to recalculate the date according to format. Zero it out
2827 to prevent stack garbage from influencing strptime. */
2830 if (check_end (strptime (time_string, time_formats[i], &t)))
2837 /* Restore the previous locale. */
2838 setlocale (LC_TIME, oldlocale);
2843 /* Authorization support: We support three authorization schemes:
2845 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2847 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2848 consisting of answering to the server's challenge with the proper
2851 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2852 Stenberg for libcurl. Like digest, NTLM is based on a
2853 challenge-response mechanism, but unlike digest, it is non-standard
2854 (authenticates TCP connections rather than requests), undocumented
2855 and Microsoft-specific. */
2857 /* Create the authentication header contents for the `Basic' scheme.
2858 This is done by encoding the string "USER:PASS" to base64 and
2859 prepending the string "Basic " in front of it. */
2862 basic_authentication_encode (const char *user, const char *passwd)
2865 int len1 = strlen (user) + 1 + strlen (passwd);
2867 t1 = (char *)alloca (len1 + 1);
2868 sprintf (t1, "%s:%s", user, passwd);
2870 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2871 base64_encode (t1, len1, t2);
2873 return concat_strings ("Basic ", t2, (char *) 0);
2876 #define SKIP_WS(x) do { \
2877 while (ISSPACE (*(x))) \
2881 #ifdef ENABLE_DIGEST
2882 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2883 an array of 16 bytes containing the hash keys, and BUF should be a
2884 buffer of 33 writable characters (32 for hex digits plus one for
2885 zero termination). */
2887 dump_hash (char *buf, const unsigned char *hash)
2891 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2893 *buf++ = XNUM_TO_digit (*hash >> 4);
2894 *buf++ = XNUM_TO_digit (*hash & 0xf);
2899 /* Take the line apart to find the challenge, and compose a digest
2900 authorization header. See RFC2069 section 2.1.2. */
2902 digest_authentication_encode (const char *au, const char *user,
2903 const char *passwd, const char *method,
2906 static char *realm, *opaque, *nonce;
2911 { "realm", &realm },
2912 { "opaque", &opaque },
2916 param_token name, value;
2918 realm = opaque = nonce = NULL;
2920 au += 6; /* skip over `Digest' */
2921 while (extract_param (&au, &name, &value, ','))
2924 for (i = 0; i < countof (options); i++)
2925 if (name.e - name.b == strlen (options[i].name)
2926 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
2928 *options[i].variable = strdupdelim (value.b, value.e);
2932 if (!realm || !nonce || !user || !passwd || !path || !method)
2935 xfree_null (opaque);
2940 /* Calculate the digest value. */
2942 ALLOCA_MD5_CONTEXT (ctx);
2943 unsigned char hash[MD5_HASHLEN];
2944 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2945 char response_digest[MD5_HASHLEN * 2 + 1];
2947 /* A1BUF = H(user ":" realm ":" password) */
2949 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2950 gen_md5_update ((unsigned char *)":", 1, ctx);
2951 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2952 gen_md5_update ((unsigned char *)":", 1, ctx);
2953 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2954 gen_md5_finish (ctx, hash);
2955 dump_hash (a1buf, hash);
2957 /* A2BUF = H(method ":" path) */
2959 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
2960 gen_md5_update ((unsigned char *)":", 1, ctx);
2961 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
2962 gen_md5_finish (ctx, hash);
2963 dump_hash (a2buf, hash);
2965 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
2967 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
2968 gen_md5_update ((unsigned char *)":", 1, ctx);
2969 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
2970 gen_md5_update ((unsigned char *)":", 1, ctx);
2971 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
2972 gen_md5_finish (ctx, hash);
2973 dump_hash (response_digest, hash);
2975 res = xmalloc (strlen (user)
2980 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
2981 + (opaque ? strlen (opaque) : 0)
2983 sprintf (res, "Digest \
2984 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
2985 user, realm, nonce, path, response_digest);
2988 char *p = res + strlen (res);
2989 strcat (p, ", opaque=\"");
2996 #endif /* ENABLE_DIGEST */
2998 /* Computing the size of a string literal must take into account that
2999 value returned by sizeof includes the terminating \0. */
3000 #define STRSIZE(literal) (sizeof (literal) - 1)
3002 /* Whether chars in [b, e) begin with the literal string provided as
3003 first argument and are followed by whitespace or terminating \0.
3004 The comparison is case-insensitive. */
3005 #define STARTS(literal, b, e) \
3006 ((e) - (b) >= STRSIZE (literal) \
3007 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3008 && ((e) - (b) == STRSIZE (literal) \
3009 || ISSPACE (b[STRSIZE (literal)])))
3012 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3014 return STARTS ("Basic", hdrbeg, hdrend)
3015 #ifdef ENABLE_DIGEST
3016 || STARTS ("Digest", hdrbeg, hdrend)
3019 || STARTS ("NTLM", hdrbeg, hdrend)
3026 /* Create the HTTP authorization request header. When the
3027 `WWW-Authenticate' response header is seen, according to the
3028 authorization scheme specified in that header (`Basic' and `Digest'
3029 are supported by the current implementation), produce an
3030 appropriate HTTP authorization request header. */
3032 create_authorization_line (const char *au, const char *user,
3033 const char *passwd, const char *method,
3034 const char *path, bool *finished)
3036 /* We are called only with known schemes, so we can dispatch on the
3038 switch (TOUPPER (*au))
3040 case 'B': /* Basic */
3042 return basic_authentication_encode (user, passwd);
3043 #ifdef ENABLE_DIGEST
3044 case 'D': /* Digest */
3046 return digest_authentication_encode (au, user, passwd, method, path);
3049 case 'N': /* NTLM */
3050 if (!ntlm_input (&pconn.ntlm, au))
3055 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3058 /* We shouldn't get here -- this function should be only called
3059 with values approved by known_authentication_scheme_p. */
3067 if (!wget_cookie_jar)
3068 wget_cookie_jar = cookie_jar_new ();
3069 if (opt.cookies_input && !cookies_loaded_p)
3071 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3072 cookies_loaded_p = true;
3079 if (wget_cookie_jar)
3080 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3086 xfree_null (pconn.host);
3087 if (wget_cookie_jar)
3088 cookie_jar_delete (wget_cookie_jar);
3095 test_parse_content_disposition()
3100 char *opt_dir_prefix;
3104 { "filename=\"file.ext\"", NULL, "file.ext", true },
3105 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3106 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3107 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3108 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3109 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3110 { "attachment", NULL, NULL, false },
3111 { "attachment", "somedir", NULL, false },
3114 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3119 opt.dir_prefix = test_array[i].opt_dir_prefix;
3120 res = parse_content_disposition (test_array[i].hdrval, &filename);
3122 mu_assert ("test_parse_content_disposition: wrong result",
3123 res == test_array[i].result
3125 || 0 == strcmp (test_array[i].filename, filename)));
3131 #endif /* TESTING */
3134 * vim: et sts=2 sw=2 cino+={s