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)
1904 #### If Content-Length is out of range, it most likely
1905 means that the file is larger than 2G and that we're
1906 compiled without LFS. In that case we should probably
1907 refuse to even attempt to download the file. */
1910 else if (parsed < 0)
1912 /* Negative Content-Length; nonsensical, so we can't
1913 assume any information about the content to receive. */
1920 /* Check for keep-alive related responses. */
1921 if (!inhibit_keep_alive && contlen != -1)
1923 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1925 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1927 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1932 /* The server has promised that it will not close the connection
1933 when we're done. This means that we can register it. */
1934 register_persistent (conn->host, conn->port, sock, using_ssl);
1936 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1938 /* Authorization is required. */
1939 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1940 CLOSE_FINISH (sock);
1942 CLOSE_INVALIDATE (sock);
1943 pconn.authorized = false;
1944 if (!auth_finished && (user && passwd))
1946 /* IIS sends multiple copies of WWW-Authenticate, one with
1947 the value "negotiate", and other(s) with data. Loop over
1948 all the occurrences and pick the one we recognize. */
1950 const char *wabeg, *waend;
1951 char *www_authenticate = NULL;
1953 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1954 &wabeg, &waend)) != -1;
1956 if (known_authentication_scheme_p (wabeg, waend))
1958 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1962 if (!www_authenticate)
1964 /* If the authentication header is missing or
1965 unrecognized, there's no sense in retrying. */
1966 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1968 else if (!basic_auth_finished
1969 || !BEGINS_WITH (www_authenticate, "Basic"))
1972 pth = url_full_path (u);
1973 request_set_header (req, "Authorization",
1974 create_authorization_line (www_authenticate,
1976 request_method (req),
1980 if (BEGINS_WITH (www_authenticate, "NTLM"))
1982 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
1984 /* Need to register this host as using basic auth,
1985 * so we automatically send creds next time. */
1986 register_basic_auth_host (u->host);
1989 goto retry_with_auth;
1993 /* We already did Basic auth, and it failed. Gotta
1997 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2001 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2003 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2005 pconn.authorized = true;
2009 hs->statcode = statcode;
2011 hs->error = xstrdup (_("Malformed status line"));
2013 hs->error = xstrdup (_("(no description)"));
2015 hs->error = xstrdup (message);
2016 xfree_null (message);
2018 type = resp_header_strdup (resp, "Content-Type");
2021 char *tmp = strchr (type, ';');
2024 while (tmp > type && ISSPACE (tmp[-1]))
2029 hs->newloc = resp_header_strdup (resp, "Location");
2030 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2032 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2036 const char *scbeg, *scend;
2037 /* The jar should have been created by now. */
2038 assert (wget_cookie_jar != NULL);
2040 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2041 &scbeg, &scend)) != -1;
2044 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2045 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2046 u->path, set_cookie);
2050 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2052 wgint first_byte_pos, last_byte_pos, entity_length;
2053 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2055 contrange = first_byte_pos;
2059 /* 20x responses are counted among successful by default. */
2060 if (H_20X (statcode))
2063 /* Return if redirected. */
2064 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2066 /* RFC2068 says that in case of the 300 (multiple choices)
2067 response, the server can output a preferred URL through
2068 `Location' header; otherwise, the request should be treated
2069 like GET. So, if the location is set, it will be a
2070 redirection; otherwise, just proceed normally. */
2071 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2075 logprintf (LOG_VERBOSE,
2076 _("Location: %s%s\n"),
2077 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2078 hs->newloc ? _(" [following]") : "");
2079 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2080 CLOSE_FINISH (sock);
2082 CLOSE_INVALIDATE (sock);
2088 /* If content-type is not given, assume text/html. This is because
2089 of the multitude of broken CGI's that "forget" to generate the
2092 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2093 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2098 if (opt.html_extension && (*dt & TEXTHTML))
2099 /* -E / --html-extension / html_extension = on was specified, and this is a
2100 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2101 already the file's suffix, tack on ".html". */
2103 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2105 if (last_period_in_local_filename == NULL
2106 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2107 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2109 int local_filename_len = strlen (hs->local_file);
2110 /* Resize the local file, allowing for ".html" preceded by
2111 optional ".NUMBER". */
2112 hs->local_file = xrealloc (hs->local_file,
2113 local_filename_len + 24 + sizeof (".html"));
2114 strcpy(hs->local_file + local_filename_len, ".html");
2115 /* If clobbering is not allowed and the file, as named,
2116 exists, tack on ".NUMBER.html" instead. */
2117 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2121 sprintf (hs->local_file + local_filename_len,
2122 ".%d.html", ext_num++);
2123 while (file_exists_p (hs->local_file));
2125 *dt |= ADDED_HTML_EXTENSION;
2129 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2131 /* If `-c' is in use and the file has been fully downloaded (or
2132 the remote file has shrunk), Wget effectively requests bytes
2133 after the end of file and the server response with 416. */
2134 logputs (LOG_VERBOSE, _("\
2135 \n The file is already fully retrieved; nothing to do.\n\n"));
2136 /* In case the caller inspects. */
2139 /* Mark as successfully retrieved. */
2142 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2143 might be more bytes in the body. */
2144 return RETRUNNEEDED;
2146 if ((contrange != 0 && contrange != hs->restval)
2147 || (H_PARTIAL (statcode) && !contrange))
2149 /* The Range request was somehow misunderstood by the server.
2152 CLOSE_INVALIDATE (sock);
2155 hs->contlen = contlen + contrange;
2161 /* No need to print this output if the body won't be
2162 downloaded at all, or if the original server response is
2164 logputs (LOG_VERBOSE, _("Length: "));
2167 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2168 if (contlen + contrange >= 1024)
2169 logprintf (LOG_VERBOSE, " (%s)",
2170 human_readable (contlen + contrange));
2173 if (contlen >= 1024)
2174 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2175 number_to_static_string (contlen),
2176 human_readable (contlen));
2178 logprintf (LOG_VERBOSE, _(", %s remaining"),
2179 number_to_static_string (contlen));
2183 logputs (LOG_VERBOSE,
2184 opt.ignore_length ? _("ignored") : _("unspecified"));
2186 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2188 logputs (LOG_VERBOSE, "\n");
2192 type = NULL; /* We don't need it any more. */
2194 /* Return if we have no intention of further downloading. */
2195 if (!(*dt & RETROKF) || head_only)
2197 /* In case the caller cares to look... */
2202 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2203 servers not to send body in response to a HEAD request, and
2204 those that do will likely be caught by test_socket_open.
2205 If not, they can be worked around using
2206 `--no-http-keep-alive'. */
2207 CLOSE_FINISH (sock);
2208 else if (keep_alive && skip_short_body (sock, contlen))
2209 /* Successfully skipped the body; also keep using the socket. */
2210 CLOSE_FINISH (sock);
2212 CLOSE_INVALIDATE (sock);
2213 return RETRFINISHED;
2216 /* Open the local file. */
2219 mkalldirs (hs->local_file);
2221 rotate_backups (hs->local_file);
2223 fp = fopen (hs->local_file, "ab");
2224 else if (ALLOW_CLOBBER)
2225 fp = fopen (hs->local_file, "wb");
2228 fp = fopen_excl (hs->local_file, true);
2229 if (!fp && errno == EEXIST)
2231 /* We cannot just invent a new name and use it (which is
2232 what functions like unique_create typically do)
2233 because we told the user we'd use this name.
2234 Instead, return and retry the download. */
2235 logprintf (LOG_NOTQUIET,
2236 _("%s has sprung into existence.\n"),
2238 CLOSE_INVALIDATE (sock);
2239 return FOPEN_EXCL_ERR;
2244 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2245 CLOSE_INVALIDATE (sock);
2252 /* Print fetch message, if opt.verbose. */
2255 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2256 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2259 /* This confuses the timestamping code that checks for file size.
2260 #### The timestamping code should be smarter about file size. */
2261 if (opt.save_headers && hs->restval == 0)
2262 fwrite (head, 1, strlen (head), fp);
2264 /* Now we no longer need to store the response header. */
2267 /* Download the request body. */
2270 /* If content-length is present, read that much; otherwise, read
2271 until EOF. The HTTP spec doesn't require the server to
2272 actually close the connection when it's done sending data. */
2273 flags |= rb_read_exactly;
2274 if (hs->restval > 0 && contrange == 0)
2275 /* If the server ignored our range request, instruct fd_read_body
2276 to skip the first RESTVAL bytes of body. */
2277 flags |= rb_skip_startpos;
2278 hs->len = hs->restval;
2280 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2281 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2285 CLOSE_FINISH (sock);
2289 hs->rderrmsg = xstrdup (fd_errstr (sock));
2290 CLOSE_INVALIDATE (sock);
2297 return RETRFINISHED;
2300 /* The genuine HTTP loop! This is the part where the retrieval is
2301 retried, and retried, and retried, and... */
2303 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2304 int *dt, struct url *proxy)
2307 bool got_head = false; /* used for time-stamping and filename detection */
2308 bool time_came_from_head = false;
2309 bool got_name = false;
2312 uerr_t err, ret = TRYLIMEXC;
2313 time_t tmr = -1; /* remote time-stamp */
2314 struct http_stat hstat; /* HTTP status */
2316 bool send_head_first = true;
2318 /* Assert that no value for *LOCAL_FILE was passed. */
2319 assert (local_file == NULL || *local_file == NULL);
2321 /* Set LOCAL_FILE parameter. */
2322 if (local_file && opt.output_document)
2323 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2325 /* Reset NEWLOC parameter. */
2328 /* This used to be done in main(), but it's a better idea to do it
2329 here so that we don't go through the hoops if we're just using
2334 /* Warn on (likely bogus) wildcard usage in HTTP. */
2335 if (opt.ftp_glob && has_wildcards_p (u->path))
2336 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2338 /* Setup hstat struct. */
2340 hstat.referer = referer;
2342 if (opt.output_document)
2344 hstat.local_file = xstrdup (opt.output_document);
2348 /* Reset the counter. */
2351 /* Reset the document type. */
2354 /* Skip preliminary HEAD request if we're not in spider mode AND
2355 * if -O was given or HTTP Content-Disposition support is disabled. */
2357 && (got_name || !opt.content_disposition))
2358 send_head_first = false;
2360 /* Send preliminary HEAD request if -N is given and we have an existing
2361 * destination file. */
2362 if (opt.timestamping
2363 && !opt.content_disposition
2364 && file_exists_p (url_file_name (u)))
2365 send_head_first = true;
2370 /* Increment the pass counter. */
2372 sleep_between_retrievals (count);
2374 /* Get the current time string. */
2375 tms = time_str (time (NULL));
2377 if (opt.spider && !got_head)
2378 logprintf (LOG_VERBOSE, _("\
2379 Spider mode enabled. Check if remote file exists.\n"));
2381 /* Print fetch message, if opt.verbose. */
2384 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2389 sprintf (tmp, _("(try:%2d)"), count);
2390 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2395 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2400 ws_changetitle (hurl);
2405 /* Default document type is empty. However, if spider mode is
2406 on or time-stamping is employed, HEAD_ONLY commands is
2407 encoded within *dt. */
2408 if (send_head_first && !got_head)
2413 /* Decide whether or not to restart. */
2416 && stat (hstat.local_file, &st) == 0
2417 && S_ISREG (st.st_mode))
2418 /* When -c is used, continue from on-disk size. (Can't use
2419 hstat.len even if count>1 because we don't want a failed
2420 first attempt to clobber existing data.) */
2421 hstat.restval = st.st_size;
2423 /* otherwise, continue where the previous try left off */
2424 hstat.restval = hstat.len;
2428 /* Decide whether to send the no-cache directive. We send it in
2430 a) we're using a proxy, and we're past our first retrieval.
2431 Some proxies are notorious for caching incomplete data, so
2432 we require a fresh get.
2433 b) caching is explicitly inhibited. */
2434 if ((proxy && count > 1) /* a */
2435 || !opt.allow_cache) /* b */
2436 *dt |= SEND_NOCACHE;
2438 *dt &= ~SEND_NOCACHE;
2440 /* Try fetching the document, or at least its head. */
2441 err = gethttp (u, &hstat, dt, proxy);
2444 tms = time_str (time (NULL));
2446 /* Get the new location (with or without the redirection). */
2448 *newloc = xstrdup (hstat.newloc);
2452 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2453 case CONERROR: case READERR: case WRITEFAILED:
2454 case RANGEERR: case FOPEN_EXCL_ERR:
2455 /* Non-fatal errors continue executing the loop, which will
2456 bring them to "while" statement at the end, to judge
2457 whether the number of tries was exceeded. */
2458 printwhat (count, opt.ntry);
2460 case FWRITEERR: case FOPENERR:
2461 /* Another fatal error. */
2462 logputs (LOG_VERBOSE, "\n");
2463 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2464 hstat.local_file, strerror (errno));
2465 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2466 case SSLINITFAILED: case CONTNOTSUPPORTED:
2467 /* Fatal errors just return from the function. */
2471 /* Another fatal error. */
2472 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2476 /* Return the new location to the caller. */
2479 logprintf (LOG_NOTQUIET,
2480 _("ERROR: Redirection (%d) without location.\n"),
2490 /* The file was already fully retrieved. */
2494 /* Deal with you later. */
2497 /* All possibilities should have been exhausted. */
2501 if (!(*dt & RETROKF))
2506 /* #### Ugly ugly ugly! */
2507 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2508 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2511 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2513 && (hstat.statcode == 500 || hstat.statcode == 501))
2518 /* Maybe we should always keep track of broken links, not just in
2520 else if (opt.spider)
2522 /* #### Again: ugly ugly ugly! */
2524 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2525 nonexisting_url (hurl);
2526 logprintf (LOG_NOTQUIET, _("\
2527 Remote file does not exist -- broken link!!!\n"));
2531 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2532 tms, hstat.statcode, escnonprint (hstat.error));
2534 logputs (LOG_VERBOSE, "\n");
2540 /* Did we get the time-stamp? */
2541 if (send_head_first && !got_head)
2543 bool restart_loop = false;
2545 if (opt.timestamping && !hstat.remote_time)
2547 logputs (LOG_NOTQUIET, _("\
2548 Last-modified header missing -- time-stamps turned off.\n"));
2550 else if (hstat.remote_time)
2552 /* Convert the date-string into struct tm. */
2553 tmr = http_atotm (hstat.remote_time);
2554 if (tmr == (time_t) (-1))
2555 logputs (LOG_VERBOSE, _("\
2556 Last-modified header invalid -- time-stamp ignored.\n"));
2557 if (*dt & HEAD_ONLY)
2558 time_came_from_head = true;
2561 /* The time-stamping section. */
2562 if (opt.timestamping)
2564 if (hstat.orig_file_name) /* Perform the following checks only
2565 if the file we're supposed to
2566 download already exists. */
2568 if (hstat.remote_time &&
2569 tmr != (time_t) (-1))
2571 /* Now time-stamping can be used validly. Time-stamping
2572 means that if the sizes of the local and remote file
2573 match, and local file is newer than the remote file,
2574 it will not be retrieved. Otherwise, the normal
2575 download procedure is resumed. */
2576 if (hstat.orig_file_tstamp >= tmr)
2578 if (hstat.contlen == -1
2579 || hstat.orig_file_size == hstat.contlen)
2581 logprintf (LOG_VERBOSE, _("\
2582 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2583 hstat.orig_file_name);
2589 logprintf (LOG_VERBOSE, _("\
2590 The sizes do not match (local %s) -- retrieving.\n"),
2591 number_to_static_string (hstat.orig_file_size));
2595 logputs (LOG_VERBOSE,
2596 _("Remote file is newer, retrieving.\n"));
2598 logputs (LOG_VERBOSE, "\n");
2602 /* free_hstat (&hstat); */
2603 hstat.timestamp_checked = true;
2604 restart_loop = true;
2613 logputs (LOG_VERBOSE, _("\
2614 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2615 restart_loop = true;
2619 logprintf (LOG_VERBOSE, _("\
2620 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2621 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2627 logprintf (LOG_VERBOSE, _("\
2628 Remote file exists but recursion is disabled -- not retrieving.\n\n"));
2629 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2634 if (send_head_first)
2637 restart_loop = true;
2640 got_head = true; /* no more time-stamping */
2642 count = 0; /* the retrieve count for HEAD is reset */
2648 if ((tmr != (time_t) (-1))
2649 && ((hstat.len == hstat.contlen) ||
2650 ((hstat.res == 0) && (hstat.contlen == -1))))
2652 /* #### This code repeats in http.c and ftp.c. Move it to a
2654 const char *fl = NULL;
2655 if (opt.output_document)
2657 if (output_stream_regular)
2658 fl = opt.output_document;
2661 fl = hstat.local_file;
2665 /* Reparse time header, in case it's changed. */
2666 if (time_came_from_head
2667 && hstat.remote_time && hstat.remote_time[0])
2669 newtmr = http_atotm (hstat.remote_time);
2676 /* End of time-stamping section. */
2678 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2679 total_download_time += hstat.dltime;
2681 if (hstat.len == hstat.contlen)
2685 logprintf (LOG_VERBOSE,
2686 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2687 tms, tmrate, hstat.local_file,
2688 number_to_static_string (hstat.len),
2689 number_to_static_string (hstat.contlen));
2690 logprintf (LOG_NONVERBOSE,
2691 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2693 number_to_static_string (hstat.len),
2694 number_to_static_string (hstat.contlen),
2695 hstat.local_file, count);
2698 total_downloaded_bytes += hstat.len;
2700 /* Remember that we downloaded the file for later ".orig" code. */
2701 if (*dt & ADDED_HTML_EXTENSION)
2702 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2704 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2709 else if (hstat.res == 0) /* No read error */
2711 if (hstat.contlen == -1) /* We don't know how much we were supposed
2712 to get, so assume we succeeded. */
2716 logprintf (LOG_VERBOSE,
2717 _("%s (%s) - `%s' saved [%s]\n\n"),
2718 tms, tmrate, hstat.local_file,
2719 number_to_static_string (hstat.len));
2720 logprintf (LOG_NONVERBOSE,
2721 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2722 tms, u->url, number_to_static_string (hstat.len),
2723 hstat.local_file, count);
2726 total_downloaded_bytes += hstat.len;
2728 /* Remember that we downloaded the file for later ".orig" code. */
2729 if (*dt & ADDED_HTML_EXTENSION)
2730 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2732 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2737 else if (hstat.len < hstat.contlen) /* meaning we lost the
2738 connection too soon */
2740 logprintf (LOG_VERBOSE,
2741 _("%s (%s) - Connection closed at byte %s. "),
2742 tms, tmrate, number_to_static_string (hstat.len));
2743 printwhat (count, opt.ntry);
2747 /* Getting here would mean reading more data than
2748 requested with content-length, which we never do. */
2751 else /* from now on hstat.res can only be -1 */
2753 if (hstat.contlen == -1)
2755 logprintf (LOG_VERBOSE,
2756 _("%s (%s) - Read error at byte %s (%s)."),
2757 tms, tmrate, number_to_static_string (hstat.len),
2759 printwhat (count, opt.ntry);
2762 else /* hstat.res == -1 and contlen is given */
2764 logprintf (LOG_VERBOSE,
2765 _("%s (%s) - Read error at byte %s/%s (%s). "),
2767 number_to_static_string (hstat.len),
2768 number_to_static_string (hstat.contlen),
2770 printwhat (count, opt.ntry);
2776 while (!opt.ntry || (count < opt.ntry));
2780 *local_file = xstrdup (hstat.local_file);
2781 free_hstat (&hstat);
2786 /* Check whether the result of strptime() indicates success.
2787 strptime() returns the pointer to how far it got to in the string.
2788 The processing has been successful if the string is at `GMT' or
2789 `+X', or at the end of the string.
2791 In extended regexp parlance, the function returns 1 if P matches
2792 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2793 can return) is considered a failure and 0 is returned. */
2795 check_end (const char *p)
2799 while (ISSPACE (*p))
2802 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2803 || ((p[0] == '+' || p[0] == '-') && ISDIGIT (p[1])))
2809 /* Convert the textual specification of time in TIME_STRING to the
2810 number of seconds since the Epoch.
2812 TIME_STRING can be in any of the three formats RFC2616 allows the
2813 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2814 as well as the time format used in the Set-Cookie header.
2815 Timezones are ignored, and should be GMT.
2817 Return the computed time_t representation, or -1 if the conversion
2820 This function uses strptime with various string formats for parsing
2821 TIME_STRING. This results in a parser that is not as lenient in
2822 interpreting TIME_STRING as I would like it to be. Being based on
2823 strptime, it always allows shortened months, one-digit days, etc.,
2824 but due to the multitude of formats in which time can be
2825 represented, an ideal HTTP time parser would be even more
2826 forgiving. It should completely ignore things like week days and
2827 concentrate only on the various forms of representing years,
2828 months, days, hours, minutes, and seconds. For example, it would
2829 be nice if it accepted ISO 8601 out of the box.
2831 I've investigated free and PD code for this purpose, but none was
2832 usable. getdate was big and unwieldy, and had potential copyright
2833 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2834 distributed with phttpd, is excellent, but we cannot use it because
2835 it is not assigned to the FSF. So I stuck it with strptime. */
2838 http_atotm (const char *time_string)
2840 /* NOTE: Solaris strptime man page claims that %n and %t match white
2841 space, but that's not universally available. Instead, we simply
2842 use ` ' to mean "skip all WS", which works under all strptime
2843 implementations I've tested. */
2845 static const char *time_formats[] = {
2846 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2847 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2848 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2849 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2850 (used in Set-Cookie, defined in the
2851 Netscape cookie specification.) */
2853 const char *oldlocale;
2855 time_t ret = (time_t) -1;
2857 /* Solaris strptime fails to recognize English month names in
2858 non-English locales, which we work around by temporarily setting
2859 locale to C before invoking strptime. */
2860 oldlocale = setlocale (LC_TIME, NULL);
2861 setlocale (LC_TIME, "C");
2863 for (i = 0; i < countof (time_formats); i++)
2867 /* Some versions of strptime use the existing contents of struct
2868 tm to recalculate the date according to format. Zero it out
2869 to prevent stack garbage from influencing strptime. */
2872 if (check_end (strptime (time_string, time_formats[i], &t)))
2879 /* Restore the previous locale. */
2880 setlocale (LC_TIME, oldlocale);
2885 /* Authorization support: We support three authorization schemes:
2887 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2889 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2890 consisting of answering to the server's challenge with the proper
2893 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2894 Stenberg for libcurl. Like digest, NTLM is based on a
2895 challenge-response mechanism, but unlike digest, it is non-standard
2896 (authenticates TCP connections rather than requests), undocumented
2897 and Microsoft-specific. */
2899 /* Create the authentication header contents for the `Basic' scheme.
2900 This is done by encoding the string "USER:PASS" to base64 and
2901 prepending the string "Basic " in front of it. */
2904 basic_authentication_encode (const char *user, const char *passwd)
2907 int len1 = strlen (user) + 1 + strlen (passwd);
2909 t1 = (char *)alloca (len1 + 1);
2910 sprintf (t1, "%s:%s", user, passwd);
2912 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2913 base64_encode (t1, len1, t2);
2915 return concat_strings ("Basic ", t2, (char *) 0);
2918 #define SKIP_WS(x) do { \
2919 while (ISSPACE (*(x))) \
2923 #ifdef ENABLE_DIGEST
2924 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2925 an array of 16 bytes containing the hash keys, and BUF should be a
2926 buffer of 33 writable characters (32 for hex digits plus one for
2927 zero termination). */
2929 dump_hash (char *buf, const unsigned char *hash)
2933 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2935 *buf++ = XNUM_TO_digit (*hash >> 4);
2936 *buf++ = XNUM_TO_digit (*hash & 0xf);
2941 /* Take the line apart to find the challenge, and compose a digest
2942 authorization header. See RFC2069 section 2.1.2. */
2944 digest_authentication_encode (const char *au, const char *user,
2945 const char *passwd, const char *method,
2948 static char *realm, *opaque, *nonce;
2953 { "realm", &realm },
2954 { "opaque", &opaque },
2958 param_token name, value;
2960 realm = opaque = nonce = NULL;
2962 au += 6; /* skip over `Digest' */
2963 while (extract_param (&au, &name, &value, ','))
2966 for (i = 0; i < countof (options); i++)
2967 if (name.e - name.b == strlen (options[i].name)
2968 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
2970 *options[i].variable = strdupdelim (value.b, value.e);
2974 if (!realm || !nonce || !user || !passwd || !path || !method)
2977 xfree_null (opaque);
2982 /* Calculate the digest value. */
2984 ALLOCA_MD5_CONTEXT (ctx);
2985 unsigned char hash[MD5_HASHLEN];
2986 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
2987 char response_digest[MD5_HASHLEN * 2 + 1];
2989 /* A1BUF = H(user ":" realm ":" password) */
2991 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
2992 gen_md5_update ((unsigned char *)":", 1, ctx);
2993 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
2994 gen_md5_update ((unsigned char *)":", 1, ctx);
2995 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
2996 gen_md5_finish (ctx, hash);
2997 dump_hash (a1buf, hash);
2999 /* A2BUF = H(method ":" path) */
3001 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3002 gen_md5_update ((unsigned char *)":", 1, ctx);
3003 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3004 gen_md5_finish (ctx, hash);
3005 dump_hash (a2buf, hash);
3007 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3009 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3010 gen_md5_update ((unsigned char *)":", 1, ctx);
3011 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3012 gen_md5_update ((unsigned char *)":", 1, ctx);
3013 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3014 gen_md5_finish (ctx, hash);
3015 dump_hash (response_digest, hash);
3017 res = xmalloc (strlen (user)
3022 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3023 + (opaque ? strlen (opaque) : 0)
3025 sprintf (res, "Digest \
3026 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3027 user, realm, nonce, path, response_digest);
3030 char *p = res + strlen (res);
3031 strcat (p, ", opaque=\"");
3038 #endif /* ENABLE_DIGEST */
3040 /* Computing the size of a string literal must take into account that
3041 value returned by sizeof includes the terminating \0. */
3042 #define STRSIZE(literal) (sizeof (literal) - 1)
3044 /* Whether chars in [b, e) begin with the literal string provided as
3045 first argument and are followed by whitespace or terminating \0.
3046 The comparison is case-insensitive. */
3047 #define STARTS(literal, b, e) \
3048 ((e) - (b) >= STRSIZE (literal) \
3049 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3050 && ((e) - (b) == STRSIZE (literal) \
3051 || ISSPACE (b[STRSIZE (literal)])))
3054 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3056 return STARTS ("Basic", hdrbeg, hdrend)
3057 #ifdef ENABLE_DIGEST
3058 || STARTS ("Digest", hdrbeg, hdrend)
3061 || STARTS ("NTLM", hdrbeg, hdrend)
3068 /* Create the HTTP authorization request header. When the
3069 `WWW-Authenticate' response header is seen, according to the
3070 authorization scheme specified in that header (`Basic' and `Digest'
3071 are supported by the current implementation), produce an
3072 appropriate HTTP authorization request header. */
3074 create_authorization_line (const char *au, const char *user,
3075 const char *passwd, const char *method,
3076 const char *path, bool *finished)
3078 /* We are called only with known schemes, so we can dispatch on the
3080 switch (TOUPPER (*au))
3082 case 'B': /* Basic */
3084 return basic_authentication_encode (user, passwd);
3085 #ifdef ENABLE_DIGEST
3086 case 'D': /* Digest */
3088 return digest_authentication_encode (au, user, passwd, method, path);
3091 case 'N': /* NTLM */
3092 if (!ntlm_input (&pconn.ntlm, au))
3097 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3100 /* We shouldn't get here -- this function should be only called
3101 with values approved by known_authentication_scheme_p. */
3109 if (!wget_cookie_jar)
3110 wget_cookie_jar = cookie_jar_new ();
3111 if (opt.cookies_input && !cookies_loaded_p)
3113 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3114 cookies_loaded_p = true;
3121 if (wget_cookie_jar)
3122 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3128 xfree_null (pconn.host);
3129 if (wget_cookie_jar)
3130 cookie_jar_delete (wget_cookie_jar);
3137 test_parse_content_disposition()
3142 char *opt_dir_prefix;
3146 { "filename=\"file.ext\"", NULL, "file.ext", true },
3147 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3148 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3149 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3150 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3151 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3152 { "attachment", NULL, NULL, false },
3153 { "attachment", "somedir", NULL, false },
3156 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3161 opt.dir_prefix = test_array[i].opt_dir_prefix;
3162 res = parse_content_disposition (test_array[i].hdrval, &filename);
3164 mu_assert ("test_parse_content_disposition: wrong result",
3165 res == test_array[i].result
3167 || 0 == strcmp (test_array[i].filename, filename)));
3173 #endif /* TESTING */
3176 * vim: et sts=2 sw=2 cino+={s