2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008, 2009 Free Software Foundation, Inc.
5 This file is part of GNU Wget.
7 GNU Wget is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 GNU Wget is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with Wget. If not, see <http://www.gnu.org/licenses/>.
20 Additional permission under GNU GPL version 3 section 7
22 If you modify this program, or any covered work, by linking or
23 combining it with the OpenSSL project's OpenSSL library (or a
24 modified version of that library), containing parts covered by the
25 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
26 grants you additional permission to convey the resulting work.
27 Corresponding Source for a non-source form of such a combination
28 shall include the source code for the parts of OpenSSL used as well
29 as that of the covered work. */
56 # include "http-ntlm.h"
71 #endif /* def __VMS */
73 extern char *version_string;
77 static char *create_authorization_line (const char *, const char *,
78 const char *, const char *,
79 const char *, bool *);
80 static char *basic_authentication_encode (const char *, const char *);
81 static bool known_authentication_scheme_p (const char *, const char *);
82 static void ensure_extension (struct http_stat *, const char *, int *);
83 static void load_cookies (void);
86 # define MIN(x, y) ((x) > (y) ? (y) : (x))
90 static bool cookies_loaded_p;
91 static struct cookie_jar *wget_cookie_jar;
93 #define TEXTHTML_S "text/html"
94 #define TEXTXHTML_S "application/xhtml+xml"
95 #define TEXTCSS_S "text/css"
97 /* Some status code validation macros: */
98 #define H_20X(x) (((x) >= 200) && ((x) < 300))
99 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
100 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
101 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
102 || (x) == HTTP_STATUS_SEE_OTHER \
103 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
105 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
106 /* Successful 2xx. */
107 #define HTTP_STATUS_OK 200
108 #define HTTP_STATUS_CREATED 201
109 #define HTTP_STATUS_ACCEPTED 202
110 #define HTTP_STATUS_NO_CONTENT 204
111 #define HTTP_STATUS_PARTIAL_CONTENTS 206
113 /* Redirection 3xx. */
114 #define HTTP_STATUS_MULTIPLE_CHOICES 300
115 #define HTTP_STATUS_MOVED_PERMANENTLY 301
116 #define HTTP_STATUS_MOVED_TEMPORARILY 302
117 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
118 #define HTTP_STATUS_NOT_MODIFIED 304
119 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
121 /* Client error 4xx. */
122 #define HTTP_STATUS_BAD_REQUEST 400
123 #define HTTP_STATUS_UNAUTHORIZED 401
124 #define HTTP_STATUS_FORBIDDEN 403
125 #define HTTP_STATUS_NOT_FOUND 404
126 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
128 /* Server errors 5xx. */
129 #define HTTP_STATUS_INTERNAL 500
130 #define HTTP_STATUS_NOT_IMPLEMENTED 501
131 #define HTTP_STATUS_BAD_GATEWAY 502
132 #define HTTP_STATUS_UNAVAILABLE 503
135 rel_none, rel_name, rel_value, rel_both
142 struct request_header {
144 enum rp release_policy;
146 int hcount, hcapacity;
151 /* Create a new, empty request. At least request_set_method must be
152 called before the request can be used. */
154 static struct request *
157 struct request *req = xnew0 (struct request);
159 req->headers = xnew_array (struct request_header, req->hcapacity);
163 /* Set the request's method and its arguments. METH should be a
164 literal string (or it should outlive the request) because it will
165 not be freed. ARG will be freed by request_free. */
168 request_set_method (struct request *req, const char *meth, char *arg)
174 /* Return the method string passed with the last call to
175 request_set_method. */
178 request_method (const struct request *req)
183 /* Free one header according to the release policy specified with
184 request_set_header. */
187 release_header (struct request_header *hdr)
189 switch (hdr->release_policy)
206 /* Set the request named NAME to VALUE. Specifically, this means that
207 a "NAME: VALUE\r\n" header line will be used in the request. If a
208 header with the same name previously existed in the request, its
209 value will be replaced by this one. A NULL value means do nothing.
211 RELEASE_POLICY determines whether NAME and VALUE should be released
212 (freed) with request_free. Allowed values are:
214 - rel_none - don't free NAME or VALUE
215 - rel_name - free NAME when done
216 - rel_value - free VALUE when done
217 - rel_both - free both NAME and VALUE when done
219 Setting release policy is useful when arguments come from different
220 sources. For example:
222 // Don't free literal strings!
223 request_set_header (req, "Pragma", "no-cache", rel_none);
225 // Don't free a global variable, we'll need it later.
226 request_set_header (req, "Referer", opt.referer, rel_none);
228 // Value freshly allocated, free it when done.
229 request_set_header (req, "Range",
230 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
235 request_set_header (struct request *req, char *name, char *value,
236 enum rp release_policy)
238 struct request_header *hdr;
243 /* A NULL value is a no-op; if freeing the name is requested,
244 free it now to avoid leaks. */
245 if (release_policy == rel_name || release_policy == rel_both)
250 for (i = 0; i < req->hcount; i++)
252 hdr = &req->headers[i];
253 if (0 == strcasecmp (name, hdr->name))
255 /* Replace existing header. */
256 release_header (hdr);
259 hdr->release_policy = release_policy;
264 /* Install new header. */
266 if (req->hcount >= req->hcapacity)
268 req->hcapacity <<= 1;
269 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
271 hdr = &req->headers[req->hcount++];
274 hdr->release_policy = release_policy;
277 /* Like request_set_header, but sets the whole header line, as
278 provided by the user using the `--header' option. For example,
279 request_set_user_header (req, "Foo: bar") works just like
280 request_set_header (req, "Foo", "bar"). */
283 request_set_user_header (struct request *req, const char *header)
286 const char *p = strchr (header, ':');
289 BOUNDED_TO_ALLOCA (header, p, name);
291 while (c_isspace (*p))
293 request_set_header (req, xstrdup (name), (char *) p, rel_name);
296 /* Remove the header with specified name from REQ. Returns true if
297 the header was actually removed, false otherwise. */
300 request_remove_header (struct request *req, char *name)
303 for (i = 0; i < req->hcount; i++)
305 struct request_header *hdr = &req->headers[i];
306 if (0 == strcasecmp (name, hdr->name))
308 release_header (hdr);
309 /* Move the remaining headers by one. */
310 if (i < req->hcount - 1)
311 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
319 #define APPEND(p, str) do { \
320 int A_len = strlen (str); \
321 memcpy (p, str, A_len); \
325 /* Construct the request and write it to FD using fd_write. */
328 request_send (const struct request *req, int fd)
330 char *request_string, *p;
331 int i, size, write_error;
333 /* Count the request size. */
336 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
337 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
339 for (i = 0; i < req->hcount; i++)
341 struct request_header *hdr = &req->headers[i];
342 /* NAME ": " VALUE "\r\n" */
343 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
349 p = request_string = alloca_array (char, size);
351 /* Generate the request. */
353 APPEND (p, req->method); *p++ = ' ';
354 APPEND (p, req->arg); *p++ = ' ';
355 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
357 for (i = 0; i < req->hcount; i++)
359 struct request_header *hdr = &req->headers[i];
360 APPEND (p, hdr->name);
361 *p++ = ':', *p++ = ' ';
362 APPEND (p, hdr->value);
363 *p++ = '\r', *p++ = '\n';
366 *p++ = '\r', *p++ = '\n', *p++ = '\0';
367 assert (p - request_string == size);
371 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
373 /* Send the request to the server. */
375 write_error = fd_write (fd, request_string, size - 1, -1);
377 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
382 /* Release the resources used by REQ. */
385 request_free (struct request *req)
388 xfree_null (req->arg);
389 for (i = 0; i < req->hcount; i++)
390 release_header (&req->headers[i]);
391 xfree_null (req->headers);
395 static struct hash_table *basic_authed_hosts;
397 /* Find out if this host has issued a Basic challenge yet; if so, give
398 * it the username, password. A temporary measure until we can get
399 * proper authentication in place. */
402 maybe_send_basic_creds (const char *hostname, const char *user,
403 const char *passwd, struct request *req)
405 bool do_challenge = false;
407 if (opt.auth_without_challenge)
409 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
412 else if (basic_authed_hosts
413 && hash_table_contains(basic_authed_hosts, hostname))
415 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
420 DEBUGP(("Host %s has not issued a general basic challenge.\n",
425 request_set_header (req, "Authorization",
426 basic_authentication_encode (user, passwd),
433 register_basic_auth_host (const char *hostname)
435 if (!basic_authed_hosts)
437 basic_authed_hosts = make_nocase_string_hash_table (1);
439 if (!hash_table_contains(basic_authed_hosts, hostname))
441 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
442 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
447 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
448 PROMISED_SIZE bytes are sent over the wire -- if the file is
449 longer, read only that much; if the file is shorter, report an error. */
452 post_file (int sock, const char *file_name, wgint promised_size)
454 static char chunk[8192];
459 DEBUGP (("[writing POST file %s ... ", file_name));
461 fp = fopen (file_name, "rb");
464 while (!feof (fp) && written < promised_size)
467 int length = fread (chunk, 1, sizeof (chunk), fp);
470 towrite = MIN (promised_size - written, length);
471 write_error = fd_write (sock, chunk, towrite, -1);
481 /* If we've written less than was promised, report a (probably
482 nonsensical) error rather than break the promise. */
483 if (written < promised_size)
489 assert (written == promised_size);
490 DEBUGP (("done]\n"));
494 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
495 If so, return the pointer to the position after the line, otherwise
496 return NULL. This is used as callback to fd_read_hunk. The data
497 between START and PEEKED has been read and cannot be "unread"; the
498 data after PEEKED has only been peeked. */
501 response_head_terminator (const char *start, const char *peeked, int peeklen)
505 /* If at first peek, verify whether HUNK starts with "HTTP". If
506 not, this is a HTTP/0.9 request and we must bail out without
508 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
511 /* Look for "\n[\r]\n", and return the following position if found.
512 Start two chars before the current to cover the possibility that
513 part of the terminator (e.g. "\n\r") arrived in the previous
515 p = peeked - start < 2 ? start : peeked - 2;
516 end = peeked + peeklen;
518 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
519 for (; p < end - 2; p++)
522 if (p[1] == '\r' && p[2] == '\n')
524 else if (p[1] == '\n')
527 /* p==end-2: check for \n\n directly preceding END. */
528 if (p[0] == '\n' && p[1] == '\n')
534 /* The maximum size of a single HTTP response we care to read. Rather
535 than being a limit of the reader implementation, this limit
536 prevents Wget from slurping all available memory upon encountering
537 malicious or buggy server output, thus protecting the user. Define
538 it to 0 to remove the limit. */
540 #define HTTP_RESPONSE_MAX_SIZE 65536
542 /* Read the HTTP request head from FD and return it. The error
543 conditions are the same as with fd_read_hunk.
545 To support HTTP/0.9 responses, this function tries to make sure
546 that the data begins with "HTTP". If this is not the case, no data
547 is read and an empty request is returned, so that the remaining
548 data can be treated as body. */
551 read_http_response_head (int fd)
553 return fd_read_hunk (fd, response_head_terminator, 512,
554 HTTP_RESPONSE_MAX_SIZE);
558 /* The response data. */
561 /* The array of pointers that indicate where each header starts.
562 For example, given this HTTP response:
569 The headers are located like this:
571 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
573 headers[0] headers[1] headers[2] headers[3]
575 I.e. headers[0] points to the beginning of the request,
576 headers[1] points to the end of the first header and the
577 beginning of the second one, etc. */
579 const char **headers;
582 /* Create a new response object from the text of the HTTP response,
583 available in HEAD. That text is automatically split into
584 constituent header lines for fast retrieval using
587 static struct response *
588 resp_new (const char *head)
593 struct response *resp = xnew0 (struct response);
598 /* Empty head means that we're dealing with a headerless
599 (HTTP/0.9) response. In that case, don't set HEADERS at
604 /* Split HEAD into header lines, so that resp_header_* functions
605 don't need to do this over and over again. */
611 DO_REALLOC (resp->headers, size, count + 1, const char *);
612 resp->headers[count++] = hdr;
614 /* Break upon encountering an empty line. */
615 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
618 /* Find the end of HDR, including continuations. */
621 const char *end = strchr (hdr, '\n');
627 while (*hdr == ' ' || *hdr == '\t');
629 DO_REALLOC (resp->headers, size, count + 1, const char *);
630 resp->headers[count] = NULL;
635 /* Locate the header named NAME in the request data, starting with
636 position START. This allows the code to loop through the request
637 data, filtering for all requests of a given name. Returns the
638 found position, or -1 for failure. The code that uses this
639 function typically looks like this:
641 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
642 ... do something with header ...
644 If you only care about one header, use resp_header_get instead of
648 resp_header_locate (const struct response *resp, const char *name, int start,
649 const char **begptr, const char **endptr)
652 const char **headers = resp->headers;
655 if (!headers || !headers[1])
658 name_len = strlen (name);
664 for (; headers[i + 1]; i++)
666 const char *b = headers[i];
667 const char *e = headers[i + 1];
669 && b[name_len] == ':'
670 && 0 == strncasecmp (b, name, name_len))
673 while (b < e && c_isspace (*b))
675 while (b < e && c_isspace (e[-1]))
685 /* Find and retrieve the header named NAME in the request data. If
686 found, set *BEGPTR to its starting, and *ENDPTR to its ending
687 position, and return true. Otherwise return false.
689 This function is used as a building block for resp_header_copy
690 and resp_header_strdup. */
693 resp_header_get (const struct response *resp, const char *name,
694 const char **begptr, const char **endptr)
696 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
700 /* Copy the response header named NAME to buffer BUF, no longer than
701 BUFSIZE (BUFSIZE includes the terminating 0). If the header
702 exists, true is returned, false otherwise. If there should be no
703 limit on the size of the header, use resp_header_strdup instead.
705 If BUFSIZE is 0, no data is copied, but the boolean indication of
706 whether the header is present is still returned. */
709 resp_header_copy (const struct response *resp, const char *name,
710 char *buf, int bufsize)
713 if (!resp_header_get (resp, name, &b, &e))
717 int len = MIN (e - b, bufsize - 1);
718 memcpy (buf, b, len);
724 /* Return the value of header named NAME in RESP, allocated with
725 malloc. If such a header does not exist in RESP, return NULL. */
728 resp_header_strdup (const struct response *resp, const char *name)
731 if (!resp_header_get (resp, name, &b, &e))
733 return strdupdelim (b, e);
736 /* Parse the HTTP status line, which is of format:
738 HTTP-Version SP Status-Code SP Reason-Phrase
740 The function returns the status-code, or -1 if the status line
741 appears malformed. The pointer to "reason-phrase" message is
742 returned in *MESSAGE. */
745 resp_status (const struct response *resp, char **message)
752 /* For a HTTP/0.9 response, assume status 200. */
754 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
758 p = resp->headers[0];
759 end = resp->headers[1];
765 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
769 /* Match the HTTP version. This is optional because Gnutella
770 servers have been reported to not specify HTTP version. */
771 if (p < end && *p == '/')
774 while (p < end && c_isdigit (*p))
776 if (p < end && *p == '.')
778 while (p < end && c_isdigit (*p))
782 while (p < end && c_isspace (*p))
784 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
787 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
792 while (p < end && c_isspace (*p))
794 while (p < end && c_isspace (end[-1]))
796 *message = strdupdelim (p, end);
802 /* Release the resources used by RESP. */
805 resp_free (struct response *resp)
807 xfree_null (resp->headers);
811 /* Print a single line of response, the characters [b, e). We tried
813 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
814 but that failed to escape the non-printable characters and, in fact,
815 caused crashes in UTF-8 locales. */
818 print_response_line(const char *prefix, const char *b, const char *e)
821 BOUNDED_TO_ALLOCA(b, e, copy);
822 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
823 quotearg_style (escape_quoting_style, copy));
826 /* Print the server response, line by line, omitting the trailing CRLF
827 from individual header lines, and prefixed with PREFIX. */
830 print_server_response (const struct response *resp, const char *prefix)
835 for (i = 0; resp->headers[i + 1]; i++)
837 const char *b = resp->headers[i];
838 const char *e = resp->headers[i + 1];
840 if (b < e && e[-1] == '\n')
842 if (b < e && e[-1] == '\r')
844 print_response_line(prefix, b, e);
848 /* Parse the `Content-Range' header and extract the information it
849 contains. Returns true if successful, false otherwise. */
851 parse_content_range (const char *hdr, wgint *first_byte_ptr,
852 wgint *last_byte_ptr, wgint *entity_length_ptr)
856 /* Ancient versions of Netscape proxy server, presumably predating
857 rfc2068, sent out `Content-Range' without the "bytes"
859 if (0 == strncasecmp (hdr, "bytes", 5))
862 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
866 while (c_isspace (*hdr))
871 if (!c_isdigit (*hdr))
873 for (num = 0; c_isdigit (*hdr); hdr++)
874 num = 10 * num + (*hdr - '0');
875 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
877 *first_byte_ptr = num;
879 for (num = 0; c_isdigit (*hdr); hdr++)
880 num = 10 * num + (*hdr - '0');
881 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
883 *last_byte_ptr = num;
888 for (num = 0; c_isdigit (*hdr); hdr++)
889 num = 10 * num + (*hdr - '0');
890 *entity_length_ptr = num;
894 /* Read the body of the request, but don't store it anywhere and don't
895 display a progress gauge. This is useful for reading the bodies of
896 administrative responses to which we will soon issue another
897 request. The response is not useful to the user, but reading it
898 allows us to continue using the same connection to the server.
900 If reading fails, false is returned, true otherwise. In debug
901 mode, the body is displayed for debugging purposes. */
904 skip_short_body (int fd, wgint contlen)
907 SKIP_SIZE = 512, /* size of the download buffer */
908 SKIP_THRESHOLD = 4096 /* the largest size we read */
910 char dlbuf[SKIP_SIZE + 1];
911 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
913 /* We shouldn't get here with unknown contlen. (This will change
914 with HTTP/1.1, which supports "chunked" transfer.) */
915 assert (contlen != -1);
917 /* If the body is too large, it makes more sense to simply close the
918 connection than to try to read the body. */
919 if (contlen > SKIP_THRESHOLD)
922 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
926 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
929 /* Don't normally report the error since this is an
930 optimization that should be invisible to the user. */
931 DEBUGP (("] aborting (%s).\n",
932 ret < 0 ? fd_errstr (fd) : "EOF received"));
936 /* Safe even if %.*s bogusly expects terminating \0 because
937 we've zero-terminated dlbuf above. */
938 DEBUGP (("%.*s", ret, dlbuf));
941 DEBUGP (("] done.\n"));
945 #define NOT_RFC2231 0
946 #define RFC2231_NOENCODING 1
947 #define RFC2231_ENCODING 2
949 /* extract_param extracts the parameter name into NAME.
950 However, if the parameter name is in RFC2231 format then
951 this function adjusts NAME by stripping of the trailing
952 characters that are not part of the name but are present to
953 indicate the presence of encoding information in the value
954 or a fragment of a long parameter value
957 modify_param_name(param_token *name)
959 const char *delim1 = memchr (name->b, '*', name->e - name->b);
960 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
966 result = NOT_RFC2231;
968 else if(delim1 == delim2)
970 if ((name->e - 1) == delim1)
972 result = RFC2231_ENCODING;
976 result = RFC2231_NOENCODING;
983 result = RFC2231_ENCODING;
988 /* extract_param extract the paramater value into VALUE.
989 Like modify_param_name this function modifies VALUE by
990 stripping off the encoding information from the actual value
993 modify_param_value (param_token *value, int encoding_type )
995 if (RFC2231_ENCODING == encoding_type)
997 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1000 value->b = (delim+1);
1005 /* Extract a parameter from the string (typically an HTTP header) at
1006 **SOURCE and advance SOURCE to the next parameter. Return false
1007 when there are no more parameters to extract. The name of the
1008 parameter is returned in NAME, and the value in VALUE. If the
1009 parameter has no value, the token's value is zeroed out.
1011 For example, if *SOURCE points to the string "attachment;
1012 filename=\"foo bar\"", the first call to this function will return
1013 the token named "attachment" and no value, and the second call will
1014 return the token named "filename" and value "foo bar". The third
1015 call will return false, indicating no more valid tokens. */
1018 extract_param (const char **source, param_token *name, param_token *value,
1021 const char *p = *source;
1023 while (c_isspace (*p)) ++p;
1027 return false; /* no error; nothing more to extract */
1032 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1034 if (name->b == name->e)
1035 return false; /* empty name: error */
1036 while (c_isspace (*p)) ++p;
1037 if (*p == separator || !*p) /* no value */
1040 if (*p == separator) ++p;
1045 return false; /* error */
1047 /* *p is '=', extract value */
1049 while (c_isspace (*p)) ++p;
1050 if (*p == '"') /* quoted */
1053 while (*p && *p != '"') ++p;
1057 /* Currently at closing quote; find the end of param. */
1058 while (c_isspace (*p)) ++p;
1059 while (*p && *p != separator) ++p;
1060 if (*p == separator)
1063 /* garbage after closed quote, e.g. foo="bar"baz */
1069 while (*p && *p != separator) ++p;
1071 while (value->e != value->b && c_isspace (value->e[-1]))
1073 if (*p == separator) ++p;
1077 int param_type = modify_param_name(name);
1078 if (NOT_RFC2231 != param_type)
1080 modify_param_value(value, param_type);
1086 #undef RFC2231_NOENCODING
1087 #undef RFC2231_ENCODING
1089 /* Appends the string represented by VALUE to FILENAME */
1092 append_value_to_filename (char **filename, param_token const * const value)
1094 int original_length = strlen(*filename);
1095 int new_length = strlen(*filename) + (value->e - value->b);
1096 *filename = xrealloc (*filename, new_length+1);
1097 memcpy (*filename + original_length, value->b, (value->e - value->b));
1098 (*filename)[new_length] = '\0';
1102 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1104 /* Parse the contents of the `Content-Disposition' header, extracting
1105 the information useful to Wget. Content-Disposition is a header
1106 borrowed from MIME; when used in HTTP, it typically serves for
1107 specifying the desired file name of the resource. For example:
1109 Content-Disposition: attachment; filename="flora.jpg"
1111 Wget will skip the tokens it doesn't care about, such as
1112 "attachment" in the previous example; it will also skip other
1113 unrecognized params. If the header is syntactically correct and
1114 contains a file name, a copy of the file name is stored in
1115 *filename and true is returned. Otherwise, the function returns
1118 The file name is stripped of directory components and must not be
1121 parse_content_disposition (const char *hdr, char **filename)
1124 param_token name, value;
1125 while (extract_param (&hdr, &name, &value, ';'))
1127 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1128 if ( isFilename && value.b != NULL)
1130 /* Make the file name begin at the last slash or backslash. */
1131 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1132 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1133 if (last_slash && last_bs)
1134 value.b = 1 + MAX (last_slash, last_bs);
1135 else if (last_slash || last_bs)
1136 value.b = 1 + (last_slash ? last_slash : last_bs);
1137 if (value.b == value.e)
1139 /* Start with the directory prefix, if specified. */
1144 int prefix_length = strlen (opt.dir_prefix);
1145 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1150 total_length = prefix_length + (value.e - value.b);
1151 *filename = xmalloc (total_length + 1);
1152 strcpy (*filename, opt.dir_prefix);
1154 (*filename)[prefix_length - 1] = '/';
1155 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1156 (*filename)[total_length] = '\0';
1160 append_value_to_filename (filename, &value);
1167 append_value_to_filename (filename, &value);
1171 *filename = strdupdelim (value.b, value.e);
1187 /* Persistent connections. Currently, we cache the most recently used
1188 connection as persistent, provided that the HTTP server agrees to
1189 make it such. The persistence data is stored in the variables
1190 below. Ideally, it should be possible to cache an arbitrary fixed
1191 number of these connections. */
1193 /* Whether a persistent connection is active. */
1194 static bool pconn_active;
1197 /* The socket of the connection. */
1200 /* Host and port of the currently active persistent connection. */
1204 /* Whether a ssl handshake has occoured on this connection. */
1207 /* Whether the connection was authorized. This is only done by
1208 NTLM, which authorizes *connections* rather than individual
1209 requests. (That practice is peculiar for HTTP, but it is a
1210 useful optimization.) */
1214 /* NTLM data of the current connection. */
1215 struct ntlmdata ntlm;
1219 /* Mark the persistent connection as invalid and free the resources it
1220 uses. This is used by the CLOSE_* macros after they forcefully
1221 close a registered persistent connection. */
1224 invalidate_persistent (void)
1226 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1227 pconn_active = false;
1228 fd_close (pconn.socket);
1233 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1234 persistent. This will enable someone to use the same connection
1235 later. In the context of HTTP, this must be called only AFTER the
1236 response has been received and the server has promised that the
1237 connection will remain alive.
1239 If a previous connection was persistent, it is closed. */
1242 register_persistent (const char *host, int port, int fd, bool ssl)
1246 if (pconn.socket == fd)
1248 /* The connection FD is already registered. */
1253 /* The old persistent connection is still active; close it
1254 first. This situation arises whenever a persistent
1255 connection exists, but we then connect to a different
1256 host, and try to register a persistent connection to that
1258 invalidate_persistent ();
1262 pconn_active = true;
1264 pconn.host = xstrdup (host);
1267 pconn.authorized = false;
1269 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1272 /* Return true if a persistent connection is available for connecting
1276 persistent_available_p (const char *host, int port, bool ssl,
1277 bool *host_lookup_failed)
1279 /* First, check whether a persistent connection is active at all. */
1283 /* If we want SSL and the last connection wasn't or vice versa,
1284 don't use it. Checking for host and port is not enough because
1285 HTTP and HTTPS can apparently coexist on the same port. */
1286 if (ssl != pconn.ssl)
1289 /* If we're not connecting to the same port, we're not interested. */
1290 if (port != pconn.port)
1293 /* If the host is the same, we're in business. If not, there is
1294 still hope -- read below. */
1295 if (0 != strcasecmp (host, pconn.host))
1297 /* Check if pconn.socket is talking to HOST under another name.
1298 This happens often when both sites are virtual hosts
1299 distinguished only by name and served by the same network
1300 interface, and hence the same web server (possibly set up by
1301 the ISP and serving many different web sites). This
1302 admittedly unconventional optimization does not contradict
1303 HTTP and works well with popular server software. */
1307 struct address_list *al;
1310 /* Don't try to talk to two different SSL sites over the same
1311 secure connection! (Besides, it's not clear that
1312 name-based virtual hosting is even possible with SSL.) */
1315 /* If pconn.socket's peer is one of the IP addresses HOST
1316 resolves to, pconn.socket is for all intents and purposes
1317 already talking to HOST. */
1319 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1321 /* Can't get the peer's address -- something must be very
1322 wrong with the connection. */
1323 invalidate_persistent ();
1326 al = lookup_host (host, 0);
1329 *host_lookup_failed = true;
1333 found = address_list_contains (al, &ip);
1334 address_list_release (al);
1339 /* The persistent connection's peer address was found among the
1340 addresses HOST resolved to; therefore, pconn.sock is in fact
1341 already talking to HOST -- no need to reconnect. */
1344 /* Finally, check whether the connection is still open. This is
1345 important because most servers implement liberal (short) timeout
1346 on persistent connections. Wget can of course always reconnect
1347 if the connection doesn't work out, but it's nicer to know in
1348 advance. This test is a logical followup of the first test, but
1349 is "expensive" and therefore placed at the end of the list.
1351 (Current implementation of test_socket_open has a nice side
1352 effect that it treats sockets with pending data as "closed".
1353 This is exactly what we want: if a broken server sends message
1354 body in response to HEAD, or if it sends more than conent-length
1355 data, we won't reuse the corrupted connection.) */
1357 if (!test_socket_open (pconn.socket))
1359 /* Oops, the socket is no longer open. Now that we know that,
1360 let's invalidate the persistent connection before returning
1362 invalidate_persistent ();
1369 /* The idea behind these two CLOSE macros is to distinguish between
1370 two cases: one when the job we've been doing is finished, and we
1371 want to close the connection and leave, and two when something is
1372 seriously wrong and we're closing the connection as part of
1375 In case of keep_alive, CLOSE_FINISH should leave the connection
1376 open, while CLOSE_INVALIDATE should still close it.
1378 Note that the semantics of the flag `keep_alive' is "this
1379 connection *will* be reused (the server has promised not to close
1380 the connection once we're done)", while the semantics of
1381 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1382 active, registered connection". */
1384 #define CLOSE_FINISH(fd) do { \
1387 if (pconn_active && (fd) == pconn.socket) \
1388 invalidate_persistent (); \
1397 #define CLOSE_INVALIDATE(fd) do { \
1398 if (pconn_active && (fd) == pconn.socket) \
1399 invalidate_persistent (); \
1407 wgint len; /* received length */
1408 wgint contlen; /* expected length */
1409 wgint restval; /* the restart value */
1410 int res; /* the result of last read */
1411 char *rderrmsg; /* error message from read error */
1412 char *newloc; /* new location (redirection) */
1413 char *remote_time; /* remote time-stamp string */
1414 char *error; /* textual HTTP error */
1415 int statcode; /* status code */
1416 char *message; /* status message */
1417 wgint rd_size; /* amount of data read from socket */
1418 double dltime; /* time it took to download the data */
1419 const char *referer; /* value of the referer header. */
1420 char *local_file; /* local file name. */
1421 bool existence_checked; /* true if we already checked for a file's
1422 existence after having begun to download
1423 (needed in gethttp for when connection is
1424 interrupted/restarted. */
1425 bool timestamp_checked; /* true if pre-download time-stamping checks
1426 * have already been performed */
1427 char *orig_file_name; /* name of file to compare for time-stamping
1428 * (might be != local_file if -K is set) */
1429 wgint orig_file_size; /* size of file to compare for time-stamping */
1430 time_t orig_file_tstamp; /* time-stamp of file to compare for
1435 free_hstat (struct http_stat *hs)
1437 xfree_null (hs->newloc);
1438 xfree_null (hs->remote_time);
1439 xfree_null (hs->error);
1440 xfree_null (hs->rderrmsg);
1441 xfree_null (hs->local_file);
1442 xfree_null (hs->orig_file_name);
1443 xfree_null (hs->message);
1445 /* Guard against being called twice. */
1447 hs->remote_time = NULL;
1451 #define BEGINS_WITH(line, string_constant) \
1452 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1453 && (c_isspace (line[sizeof (string_constant) - 1]) \
1454 || !line[sizeof (string_constant) - 1]))
1457 #define SET_USER_AGENT(req) do { \
1458 if (!opt.useragent) \
1459 request_set_header (req, "User-Agent", \
1460 aprintf ("Wget/%s (VMS %s %s)", \
1461 version_string, vms_arch(), vms_vers()), \
1463 else if (*opt.useragent) \
1464 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1466 #else /* def __VMS */
1467 #define SET_USER_AGENT(req) do { \
1468 if (!opt.useragent) \
1469 request_set_header (req, "User-Agent", \
1470 aprintf ("Wget/%s (%s)", \
1471 version_string, OS_TYPE), \
1473 else if (*opt.useragent) \
1474 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1476 #endif /* def __VMS [else] */
1478 /* The flags that allow clobbering the file (opening with "wb").
1479 Defined here to avoid repetition later. #### This will require
1481 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1482 || opt.dirstruct || opt.output_document)
1484 /* Retrieve a document through HTTP protocol. It recognizes status
1485 code, and correctly handles redirections. It closes the network
1486 socket. If it receives an error from the functions below it, it
1487 will print it if there is enough information to do so (almost
1488 always), returning the error to the caller (i.e. http_loop).
1490 Various HTTP parameters are stored to hs.
1492 If PROXY is non-NULL, the connection will be made to the proxy
1493 server, and u->url will be requested. */
1495 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1498 struct request *req;
1501 char *user, *passwd;
1505 wgint contlen, contrange;
1512 /* Set to 1 when the authorization has already been sent and should
1513 not be tried again. */
1514 bool auth_finished = false;
1516 /* Set to 1 when just globally-set Basic authorization has been sent;
1517 * should prevent further Basic negotiations, but not other
1519 bool basic_auth_finished = false;
1521 /* Whether NTLM authentication is used for this request. */
1522 bool ntlm_seen = false;
1524 /* Whether our connection to the remote host is through SSL. */
1525 bool using_ssl = false;
1527 /* Whether a HEAD request will be issued (as opposed to GET or
1529 bool head_only = !!(*dt & HEAD_ONLY);
1532 struct response *resp;
1536 /* Whether this connection will be kept alive after the HTTP request
1540 /* Whether keep-alive should be inhibited.
1542 RFC 2068 requests that 1.0 clients not send keep-alive requests
1543 to proxies. This is because many 1.0 proxies do not interpret
1544 the Connection header and transfer it to the remote server,
1545 causing it to not close the connection and leave both the proxy
1546 and the client hanging. */
1547 bool inhibit_keep_alive =
1548 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1550 /* Headers sent when using POST. */
1551 wgint post_data_size = 0;
1553 bool host_lookup_failed = false;
1556 if (u->scheme == SCHEME_HTTPS)
1558 /* Initialize the SSL context. After this has once been done,
1559 it becomes a no-op. */
1562 scheme_disable (SCHEME_HTTPS);
1563 logprintf (LOG_NOTQUIET,
1564 _("Disabling SSL due to encountered errors.\n"));
1565 return SSLINITFAILED;
1568 #endif /* HAVE_SSL */
1570 /* Initialize certain elements of struct http_stat. */
1574 hs->rderrmsg = NULL;
1576 hs->remote_time = NULL;
1582 /* Prepare the request to send. */
1584 req = request_new ();
1587 const char *meth = "GET";
1590 else if (opt.post_file_name || opt.post_data)
1592 /* Use the full path, i.e. one that includes the leading slash and
1593 the query string. E.g. if u->path is "foo/bar" and u->query is
1594 "param=value", full_path will be "/foo/bar?param=value". */
1597 /* When using SSL over proxy, CONNECT establishes a direct
1598 connection to the HTTPS server. Therefore use the same
1599 argument as when talking to the server directly. */
1600 && u->scheme != SCHEME_HTTPS
1603 meth_arg = xstrdup (u->url);
1605 meth_arg = url_full_path (u);
1606 request_set_method (req, meth, meth_arg);
1609 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1610 if (*dt & SEND_NOCACHE)
1611 request_set_header (req, "Pragma", "no-cache", rel_none);
1613 request_set_header (req, "Range",
1614 aprintf ("bytes=%s-",
1615 number_to_static_string (hs->restval)),
1617 SET_USER_AGENT (req);
1618 request_set_header (req, "Accept", "*/*", rel_none);
1620 /* Find the username and password for authentication. */
1623 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1624 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1625 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1627 /* We only do "site-wide" authentication with "global" user/password
1628 * values unless --auth-no-challange has been requested; URL user/password
1629 * info overrides. */
1630 if (user && passwd && (!u->user || opt.auth_without_challenge))
1632 /* If this is a host for which we've already received a Basic
1633 * challenge, we'll go ahead and send Basic authentication creds. */
1634 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1637 /* Generate the Host header, HOST:PORT. Take into account that:
1639 - Broken server-side software often doesn't recognize the PORT
1640 argument, so we must generate "Host: www.server.com" instead of
1641 "Host: www.server.com:80" (and likewise for https port).
1643 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1644 becomes ambiguous and needs to be rewritten as "Host:
1645 [3ffe:8100:200:2::2]:1234". */
1647 /* Formats arranged for hfmt[add_port][add_squares]. */
1648 static const char *hfmt[][2] = {
1649 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1651 int add_port = u->port != scheme_default_port (u->scheme);
1652 int add_squares = strchr (u->host, ':') != NULL;
1653 request_set_header (req, "Host",
1654 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1658 if (!inhibit_keep_alive)
1659 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1662 request_set_header (req, "Cookie",
1663 cookie_header (wget_cookie_jar,
1664 u->host, u->port, u->path,
1666 u->scheme == SCHEME_HTTPS
1673 if (opt.post_data || opt.post_file_name)
1675 request_set_header (req, "Content-Type",
1676 "application/x-www-form-urlencoded", rel_none);
1678 post_data_size = strlen (opt.post_data);
1681 post_data_size = file_size (opt.post_file_name);
1682 if (post_data_size == -1)
1684 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1685 quote (opt.post_file_name), strerror (errno));
1689 request_set_header (req, "Content-Length",
1690 xstrdup (number_to_static_string (post_data_size)),
1694 /* Add the user headers. */
1695 if (opt.user_headers)
1698 for (i = 0; opt.user_headers[i]; i++)
1699 request_set_user_header (req, opt.user_headers[i]);
1703 /* We need to come back here when the initial attempt to retrieve
1704 without authorization header fails. (Expected to happen at least
1705 for the Digest authorization scheme.) */
1710 char *proxy_user, *proxy_passwd;
1711 /* For normal username and password, URL components override
1712 command-line/wgetrc parameters. With proxy
1713 authentication, it's the reverse, because proxy URLs are
1714 normally the "permanent" ones, so command-line args
1715 should take precedence. */
1716 if (opt.proxy_user && opt.proxy_passwd)
1718 proxy_user = opt.proxy_user;
1719 proxy_passwd = opt.proxy_passwd;
1723 proxy_user = proxy->user;
1724 proxy_passwd = proxy->passwd;
1726 /* #### This does not appear right. Can't the proxy request,
1727 say, `Digest' authentication? */
1728 if (proxy_user && proxy_passwd)
1729 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1731 /* If we're using a proxy, we will be connecting to the proxy
1735 /* Proxy authorization over SSL is handled below. */
1737 if (u->scheme != SCHEME_HTTPS)
1739 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1744 /* Establish the connection. */
1746 if (!inhibit_keep_alive)
1748 /* Look for a persistent connection to target host, unless a
1749 proxy is used. The exception is when SSL is in use, in which
1750 case the proxy is nothing but a passthrough to the target
1751 host, registered as a connection to the latter. */
1752 struct url *relevant = conn;
1754 if (u->scheme == SCHEME_HTTPS)
1758 if (persistent_available_p (relevant->host, relevant->port,
1760 relevant->scheme == SCHEME_HTTPS,
1764 &host_lookup_failed))
1766 sock = pconn.socket;
1767 using_ssl = pconn.ssl;
1768 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1769 quotearg_style (escape_quoting_style, pconn.host),
1771 DEBUGP (("Reusing fd %d.\n", sock));
1772 if (pconn.authorized)
1773 /* If the connection is already authorized, the "Basic"
1774 authorization added by code above is unnecessary and
1776 request_remove_header (req, "Authorization");
1778 else if (host_lookup_failed)
1781 logprintf(LOG_NOTQUIET,
1782 _("%s: unable to resolve host address %s\n"),
1783 exec_name, quote (relevant->host));
1790 sock = connect_to_host (conn->host, conn->port);
1799 return (retryable_socket_connect_error (errno)
1800 ? CONERROR : CONIMPOSSIBLE);
1804 if (proxy && u->scheme == SCHEME_HTTPS)
1806 /* When requesting SSL URLs through proxies, use the
1807 CONNECT method to request passthrough. */
1808 struct request *connreq = request_new ();
1809 request_set_method (connreq, "CONNECT",
1810 aprintf ("%s:%d", u->host, u->port));
1811 SET_USER_AGENT (connreq);
1814 request_set_header (connreq, "Proxy-Authorization",
1815 proxyauth, rel_value);
1816 /* Now that PROXYAUTH is part of the CONNECT request,
1817 zero it out so we don't send proxy authorization with
1818 the regular request below. */
1821 /* Examples in rfc2817 use the Host header in CONNECT
1822 requests. I don't see how that gains anything, given
1823 that the contents of Host would be exactly the same as
1824 the contents of CONNECT. */
1826 write_error = request_send (connreq, sock);
1827 request_free (connreq);
1828 if (write_error < 0)
1830 CLOSE_INVALIDATE (sock);
1834 head = read_http_response_head (sock);
1837 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1839 CLOSE_INVALIDATE (sock);
1848 DEBUGP (("proxy responded with: [%s]\n", head));
1850 resp = resp_new (head);
1851 statcode = resp_status (resp, &message);
1852 hs->message = xstrdup (message);
1855 if (statcode != 200)
1858 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1859 message ? quotearg_style (escape_quoting_style, message) : "?");
1860 xfree_null (message);
1863 xfree_null (message);
1865 /* SOCK is now *really* connected to u->host, so update CONN
1866 to reflect this. That way register_persistent will
1867 register SOCK as being connected to u->host:u->port. */
1871 if (conn->scheme == SCHEME_HTTPS)
1873 if (!ssl_connect_wget (sock))
1878 else if (!ssl_check_certificate (sock, u->host))
1881 return VERIFCERTERR;
1885 #endif /* HAVE_SSL */
1888 /* Send the request to server. */
1889 write_error = request_send (req, sock);
1891 if (write_error >= 0)
1895 DEBUGP (("[POST data: %s]\n", opt.post_data));
1896 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1898 else if (opt.post_file_name && post_data_size != 0)
1899 write_error = post_file (sock, opt.post_file_name, post_data_size);
1902 if (write_error < 0)
1904 CLOSE_INVALIDATE (sock);
1908 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1909 proxy ? "Proxy" : "HTTP");
1914 head = read_http_response_head (sock);
1919 logputs (LOG_NOTQUIET, _("No data received.\n"));
1920 CLOSE_INVALIDATE (sock);
1926 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1928 CLOSE_INVALIDATE (sock);
1933 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1935 resp = resp_new (head);
1937 /* Check for status line. */
1939 statcode = resp_status (resp, &message);
1940 hs->message = xstrdup (message);
1941 if (!opt.server_response)
1942 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1943 message ? quotearg_style (escape_quoting_style, message) : "");
1946 logprintf (LOG_VERBOSE, "\n");
1947 print_server_response (resp, " ");
1950 if (!opt.ignore_length
1951 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1955 parsed = str_to_wgint (hdrval, NULL, 10);
1956 if (parsed == WGINT_MAX && errno == ERANGE)
1959 #### If Content-Length is out of range, it most likely
1960 means that the file is larger than 2G and that we're
1961 compiled without LFS. In that case we should probably
1962 refuse to even attempt to download the file. */
1965 else if (parsed < 0)
1967 /* Negative Content-Length; nonsensical, so we can't
1968 assume any information about the content to receive. */
1975 /* Check for keep-alive related responses. */
1976 if (!inhibit_keep_alive && contlen != -1)
1978 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1980 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1982 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1987 /* Handle (possibly multiple instances of) the Set-Cookie header. */
1991 const char *scbeg, *scend;
1992 /* The jar should have been created by now. */
1993 assert (wget_cookie_jar != NULL);
1995 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
1996 &scbeg, &scend)) != -1;
1999 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2000 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2001 u->path, set_cookie);
2006 /* The server has promised that it will not close the connection
2007 when we're done. This means that we can register it. */
2008 register_persistent (conn->host, conn->port, sock, using_ssl);
2010 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2012 /* Authorization is required. */
2013 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2014 CLOSE_FINISH (sock);
2016 CLOSE_INVALIDATE (sock);
2017 pconn.authorized = false;
2018 if (!auth_finished && (user && passwd))
2020 /* IIS sends multiple copies of WWW-Authenticate, one with
2021 the value "negotiate", and other(s) with data. Loop over
2022 all the occurrences and pick the one we recognize. */
2024 const char *wabeg, *waend;
2025 char *www_authenticate = NULL;
2027 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2028 &wabeg, &waend)) != -1;
2030 if (known_authentication_scheme_p (wabeg, waend))
2032 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2036 if (!www_authenticate)
2038 /* If the authentication header is missing or
2039 unrecognized, there's no sense in retrying. */
2040 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2042 else if (!basic_auth_finished
2043 || !BEGINS_WITH (www_authenticate, "Basic"))
2046 pth = url_full_path (u);
2047 request_set_header (req, "Authorization",
2048 create_authorization_line (www_authenticate,
2050 request_method (req),
2054 if (BEGINS_WITH (www_authenticate, "NTLM"))
2056 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2058 /* Need to register this host as using basic auth,
2059 * so we automatically send creds next time. */
2060 register_basic_auth_host (u->host);
2063 xfree_null (message);
2066 goto retry_with_auth;
2070 /* We already did Basic auth, and it failed. Gotta
2074 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2076 xfree_null (message);
2081 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2083 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2085 pconn.authorized = true;
2088 /* Determine the local filename if needed. Notice that if -O is used
2089 * hstat.local_file is set by http_loop to the argument of -O. */
2090 if (!hs->local_file)
2092 /* Honor Content-Disposition whether possible. */
2093 if (!opt.content_disposition
2094 || !resp_header_copy (resp, "Content-Disposition",
2095 hdrval, sizeof (hdrval))
2096 || !parse_content_disposition (hdrval, &hs->local_file))
2098 /* The Content-Disposition header is missing or broken.
2099 * Choose unique file name according to given URL. */
2100 hs->local_file = url_file_name (u);
2104 /* TODO: perform this check only once. */
2105 if (!hs->existence_checked && file_exists_p (hs->local_file))
2107 if (opt.noclobber && !opt.output_document)
2109 /* If opt.noclobber is turned on and file already exists, do not
2110 retrieve the file. But if the output_document was given, then this
2111 test was already done and the file didn't exist. Hence the !opt.output_document */
2112 logprintf (LOG_VERBOSE, _("\
2113 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2114 /* If the file is there, we suppose it's retrieved OK. */
2117 /* #### Bogusness alert. */
2118 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2119 if (has_html_suffix_p (hs->local_file))
2123 xfree_null (message);
2124 return RETRUNNEEDED;
2126 else if (!ALLOW_CLOBBER)
2128 char *unique = unique_name (hs->local_file, true);
2129 if (unique != hs->local_file)
2130 xfree (hs->local_file);
2131 hs->local_file = unique;
2134 hs->existence_checked = true;
2136 /* Support timestamping */
2137 /* TODO: move this code out of gethttp. */
2138 if (opt.timestamping && !hs->timestamp_checked)
2140 size_t filename_len = strlen (hs->local_file);
2141 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2142 bool local_dot_orig_file_exists = false;
2143 char *local_filename = NULL;
2146 if (opt.backup_converted)
2147 /* If -K is specified, we'll act on the assumption that it was specified
2148 last time these files were downloaded as well, and instead of just
2149 comparing local file X against server file X, we'll compare local
2150 file X.orig (if extant, else X) against server file X. If -K
2151 _wasn't_ specified last time, or the server contains files called
2152 *.orig, -N will be back to not operating correctly with -k. */
2154 /* Would a single s[n]printf() call be faster? --dan
2156 Definitely not. sprintf() is horribly slow. It's a
2157 different question whether the difference between the two
2158 affects a program. Usually I'd say "no", but at one
2159 point I profiled Wget, and found that a measurable and
2160 non-negligible amount of time was lost calling sprintf()
2161 in url.c. Replacing sprintf with inline calls to
2162 strcpy() and number_to_string() made a difference.
2164 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2165 memcpy (filename_plus_orig_suffix + filename_len,
2166 ORIG_SFX, sizeof (ORIG_SFX));
2168 /* Try to stat() the .orig file. */
2169 if (stat (filename_plus_orig_suffix, &st) == 0)
2171 local_dot_orig_file_exists = true;
2172 local_filename = filename_plus_orig_suffix;
2176 if (!local_dot_orig_file_exists)
2177 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2178 if (stat (hs->local_file, &st) == 0)
2179 local_filename = hs->local_file;
2181 if (local_filename != NULL)
2182 /* There was a local file, so we'll check later to see if the version
2183 the server has is the same version we already have, allowing us to
2186 hs->orig_file_name = xstrdup (local_filename);
2187 hs->orig_file_size = st.st_size;
2188 hs->orig_file_tstamp = st.st_mtime;
2190 /* Modification time granularity is 2 seconds for Windows, so
2191 increase local time by 1 second for later comparison. */
2192 ++hs->orig_file_tstamp;
2199 hs->statcode = statcode;
2201 hs->error = xstrdup (_("Malformed status line"));
2203 hs->error = xstrdup (_("(no description)"));
2205 hs->error = xstrdup (message);
2206 xfree_null (message);
2208 type = resp_header_strdup (resp, "Content-Type");
2211 char *tmp = strchr (type, ';');
2214 /* sXXXav: only needed if IRI support is enabled */
2215 char *tmp2 = tmp + 1;
2217 while (tmp > type && c_isspace (tmp[-1]))
2221 /* Try to get remote encoding if needed */
2222 if (opt.enable_iri && !opt.encoding_remote)
2224 tmp = parse_charset (tmp2);
2226 set_content_encoding (iri, tmp);
2230 hs->newloc = resp_header_strdup (resp, "Location");
2231 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2233 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2235 wgint first_byte_pos, last_byte_pos, entity_length;
2236 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2239 contrange = first_byte_pos;
2240 contlen = last_byte_pos - first_byte_pos + 1;
2245 /* 20x responses are counted among successful by default. */
2246 if (H_20X (statcode))
2249 /* Return if redirected. */
2250 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2252 /* RFC2068 says that in case of the 300 (multiple choices)
2253 response, the server can output a preferred URL through
2254 `Location' header; otherwise, the request should be treated
2255 like GET. So, if the location is set, it will be a
2256 redirection; otherwise, just proceed normally. */
2257 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2261 logprintf (LOG_VERBOSE,
2262 _("Location: %s%s\n"),
2263 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2264 hs->newloc ? _(" [following]") : "");
2265 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2266 CLOSE_FINISH (sock);
2268 CLOSE_INVALIDATE (sock);
2275 /* If content-type is not given, assume text/html. This is because
2276 of the multitude of broken CGI's that "forget" to generate the
2279 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2280 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2286 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2291 if (opt.adjust_extension)
2294 /* -E / --adjust-extension / adjust_extension = on was specified,
2295 and this is a text/html file. If some case-insensitive
2296 variation on ".htm[l]" isn't already the file's suffix,
2299 ensure_extension (hs, ".html", dt);
2301 else if (*dt & TEXTCSS)
2303 ensure_extension (hs, ".css", dt);
2307 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2308 || (hs->restval > 0 && statcode == HTTP_STATUS_OK
2309 && contrange == 0 && hs->restval >= contlen)
2312 /* If `-c' is in use and the file has been fully downloaded (or
2313 the remote file has shrunk), Wget effectively requests bytes
2314 after the end of file and the server response with 416
2315 (or 200 with a <= Content-Length. */
2316 logputs (LOG_VERBOSE, _("\
2317 \n The file is already fully retrieved; nothing to do.\n\n"));
2318 /* In case the caller inspects. */
2321 /* Mark as successfully retrieved. */
2324 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2325 might be more bytes in the body. */
2327 return RETRUNNEEDED;
2329 if ((contrange != 0 && contrange != hs->restval)
2330 || (H_PARTIAL (statcode) && !contrange))
2332 /* The Range request was somehow misunderstood by the server.
2335 CLOSE_INVALIDATE (sock);
2342 hs->contlen = contlen + contrange;
2348 /* No need to print this output if the body won't be
2349 downloaded at all, or if the original server response is
2351 logputs (LOG_VERBOSE, _("Length: "));
2354 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2355 if (contlen + contrange >= 1024)
2356 logprintf (LOG_VERBOSE, " (%s)",
2357 human_readable (contlen + contrange));
2360 if (contlen >= 1024)
2361 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2362 number_to_static_string (contlen),
2363 human_readable (contlen));
2365 logprintf (LOG_VERBOSE, _(", %s remaining"),
2366 number_to_static_string (contlen));
2370 logputs (LOG_VERBOSE,
2371 opt.ignore_length ? _("ignored") : _("unspecified"));
2373 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2375 logputs (LOG_VERBOSE, "\n");
2379 type = NULL; /* We don't need it any more. */
2381 /* Return if we have no intention of further downloading. */
2382 if (!(*dt & RETROKF) || head_only)
2384 /* In case the caller cares to look... */
2389 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2390 servers not to send body in response to a HEAD request, and
2391 those that do will likely be caught by test_socket_open.
2392 If not, they can be worked around using
2393 `--no-http-keep-alive'. */
2394 CLOSE_FINISH (sock);
2395 else if (keep_alive && skip_short_body (sock, contlen))
2396 /* Successfully skipped the body; also keep using the socket. */
2397 CLOSE_FINISH (sock);
2399 CLOSE_INVALIDATE (sock);
2401 return RETRFINISHED;
2405 For VMS, define common fopen() optional arguments.
2408 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2409 # define FOPEN_BIN_FLAG 3
2410 #else /* def __VMS */
2411 # define FOPEN_BIN_FLAG true
2412 #endif /* def __VMS [else] */
2414 /* Open the local file. */
2417 mkalldirs (hs->local_file);
2419 rotate_backups (hs->local_file);
2426 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2427 #else /* def __VMS */
2428 fp = fopen (hs->local_file, "ab");
2429 #endif /* def __VMS [else] */
2431 else if (ALLOW_CLOBBER)
2437 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2438 #else /* def __VMS */
2439 fp = fopen (hs->local_file, "wb");
2440 #endif /* def __VMS [else] */
2444 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2445 if (!fp && errno == EEXIST)
2447 /* We cannot just invent a new name and use it (which is
2448 what functions like unique_create typically do)
2449 because we told the user we'd use this name.
2450 Instead, return and retry the download. */
2451 logprintf (LOG_NOTQUIET,
2452 _("%s has sprung into existence.\n"),
2454 CLOSE_INVALIDATE (sock);
2456 return FOPEN_EXCL_ERR;
2461 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2462 CLOSE_INVALIDATE (sock);
2470 /* Print fetch message, if opt.verbose. */
2473 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2474 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2477 /* This confuses the timestamping code that checks for file size.
2478 #### The timestamping code should be smarter about file size. */
2479 if (opt.save_headers && hs->restval == 0)
2480 fwrite (head, 1, strlen (head), fp);
2482 /* Now we no longer need to store the response header. */
2485 /* Download the request body. */
2488 /* If content-length is present, read that much; otherwise, read
2489 until EOF. The HTTP spec doesn't require the server to
2490 actually close the connection when it's done sending data. */
2491 flags |= rb_read_exactly;
2492 if (hs->restval > 0 && contrange == 0)
2493 /* If the server ignored our range request, instruct fd_read_body
2494 to skip the first RESTVAL bytes of body. */
2495 flags |= rb_skip_startpos;
2496 hs->len = hs->restval;
2498 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2499 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2503 CLOSE_FINISH (sock);
2507 hs->rderrmsg = xstrdup (fd_errstr (sock));
2508 CLOSE_INVALIDATE (sock);
2515 return RETRFINISHED;
2518 /* The genuine HTTP loop! This is the part where the retrieval is
2519 retried, and retried, and retried, and... */
2521 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2522 int *dt, struct url *proxy, struct iri *iri)
2525 bool got_head = false; /* used for time-stamping and filename detection */
2526 bool time_came_from_head = false;
2527 bool got_name = false;
2530 uerr_t err, ret = TRYLIMEXC;
2531 time_t tmr = -1; /* remote time-stamp */
2532 struct http_stat hstat; /* HTTP status */
2534 bool send_head_first = true;
2537 /* Assert that no value for *LOCAL_FILE was passed. */
2538 assert (local_file == NULL || *local_file == NULL);
2540 /* Set LOCAL_FILE parameter. */
2541 if (local_file && opt.output_document)
2542 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2544 /* Reset NEWLOC parameter. */
2547 /* This used to be done in main(), but it's a better idea to do it
2548 here so that we don't go through the hoops if we're just using
2553 /* Warn on (likely bogus) wildcard usage in HTTP. */
2554 if (opt.ftp_glob && has_wildcards_p (u->path))
2555 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2557 /* Setup hstat struct. */
2559 hstat.referer = referer;
2561 if (opt.output_document)
2563 hstat.local_file = xstrdup (opt.output_document);
2566 else if (!opt.content_disposition)
2568 hstat.local_file = url_file_name (u);
2572 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2573 * screaming for some refactoring. */
2574 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2576 /* If opt.noclobber is turned on and file already exists, do not
2577 retrieve the file. But if the output_document was given, then this
2578 test was already done and the file didn't exist. Hence the !opt.output_document */
2579 logprintf (LOG_VERBOSE, _("\
2580 File %s already there; not retrieving.\n\n"),
2581 quote (hstat.local_file));
2582 /* If the file is there, we suppose it's retrieved OK. */
2585 /* #### Bogusness alert. */
2586 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2587 if (has_html_suffix_p (hstat.local_file))
2594 /* Reset the counter. */
2597 /* Reset the document type. */
2600 /* Skip preliminary HEAD request if we're not in spider mode AND
2601 * if -O was given or HTTP Content-Disposition support is disabled. */
2603 && (got_name || !opt.content_disposition))
2604 send_head_first = false;
2606 /* Send preliminary HEAD request if -N is given and we have an existing
2607 * destination file. */
2608 file_name = url_file_name (u);
2609 if (opt.timestamping
2610 && !opt.content_disposition
2611 && file_exists_p (file_name))
2612 send_head_first = true;
2618 /* Increment the pass counter. */
2620 sleep_between_retrievals (count);
2622 /* Get the current time string. */
2623 tms = datetime_str (time (NULL));
2625 if (opt.spider && !got_head)
2626 logprintf (LOG_VERBOSE, _("\
2627 Spider mode enabled. Check if remote file exists.\n"));
2629 /* Print fetch message, if opt.verbose. */
2632 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2637 sprintf (tmp, _("(try:%2d)"), count);
2638 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2643 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2648 ws_changetitle (hurl);
2653 /* Default document type is empty. However, if spider mode is
2654 on or time-stamping is employed, HEAD_ONLY commands is
2655 encoded within *dt. */
2656 if (send_head_first && !got_head)
2661 /* Decide whether or not to restart. */
2664 && stat (hstat.local_file, &st) == 0
2665 && S_ISREG (st.st_mode))
2666 /* When -c is used, continue from on-disk size. (Can't use
2667 hstat.len even if count>1 because we don't want a failed
2668 first attempt to clobber existing data.) */
2669 hstat.restval = st.st_size;
2671 /* otherwise, continue where the previous try left off */
2672 hstat.restval = hstat.len;
2676 /* Decide whether to send the no-cache directive. We send it in
2678 a) we're using a proxy, and we're past our first retrieval.
2679 Some proxies are notorious for caching incomplete data, so
2680 we require a fresh get.
2681 b) caching is explicitly inhibited. */
2682 if ((proxy && count > 1) /* a */
2683 || !opt.allow_cache) /* b */
2684 *dt |= SEND_NOCACHE;
2686 *dt &= ~SEND_NOCACHE;
2688 /* Try fetching the document, or at least its head. */
2689 err = gethttp (u, &hstat, dt, proxy, iri);
2692 tms = datetime_str (time (NULL));
2694 /* Get the new location (with or without the redirection). */
2696 *newloc = xstrdup (hstat.newloc);
2700 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2701 case CONERROR: case READERR: case WRITEFAILED:
2702 case RANGEERR: case FOPEN_EXCL_ERR:
2703 /* Non-fatal errors continue executing the loop, which will
2704 bring them to "while" statement at the end, to judge
2705 whether the number of tries was exceeded. */
2706 printwhat (count, opt.ntry);
2708 case FWRITEERR: case FOPENERR:
2709 /* Another fatal error. */
2710 logputs (LOG_VERBOSE, "\n");
2711 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2712 quote (hstat.local_file), strerror (errno));
2713 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2714 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2715 /* Fatal errors just return from the function. */
2719 /* Another fatal error. */
2720 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2724 /* Return the new location to the caller. */
2727 logprintf (LOG_NOTQUIET,
2728 _("ERROR: Redirection (%d) without location.\n"),
2738 /* The file was already fully retrieved. */
2742 /* Deal with you later. */
2745 /* All possibilities should have been exhausted. */
2749 if (!(*dt & RETROKF))
2754 /* #### Ugly ugly ugly! */
2755 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2756 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2759 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2761 && (hstat.statcode == 500 || hstat.statcode == 501))
2766 /* Maybe we should always keep track of broken links, not just in
2768 * Don't log error if it was UTF-8 encoded because we will try
2769 * once unencoded. */
2770 else if (opt.spider && !iri->utf8_encode)
2772 /* #### Again: ugly ugly ugly! */
2774 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2775 nonexisting_url (hurl);
2776 logprintf (LOG_NOTQUIET, _("\
2777 Remote file does not exist -- broken link!!!\n"));
2781 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2782 tms, hstat.statcode,
2783 quotearg_style (escape_quoting_style, hstat.error));
2785 logputs (LOG_VERBOSE, "\n");
2791 /* Did we get the time-stamp? */
2794 got_head = true; /* no more time-stamping */
2796 if (opt.timestamping && !hstat.remote_time)
2798 logputs (LOG_NOTQUIET, _("\
2799 Last-modified header missing -- time-stamps turned off.\n"));
2801 else if (hstat.remote_time)
2803 /* Convert the date-string into struct tm. */
2804 tmr = http_atotm (hstat.remote_time);
2805 if (tmr == (time_t) (-1))
2806 logputs (LOG_VERBOSE, _("\
2807 Last-modified header invalid -- time-stamp ignored.\n"));
2808 if (*dt & HEAD_ONLY)
2809 time_came_from_head = true;
2812 if (send_head_first)
2814 /* The time-stamping section. */
2815 if (opt.timestamping)
2817 if (hstat.orig_file_name) /* Perform the following
2818 checks only if the file
2820 download already exists. */
2822 if (hstat.remote_time &&
2823 tmr != (time_t) (-1))
2825 /* Now time-stamping can be used validly.
2826 Time-stamping means that if the sizes of
2827 the local and remote file match, and local
2828 file is newer than the remote file, it will
2829 not be retrieved. Otherwise, the normal
2830 download procedure is resumed. */
2831 if (hstat.orig_file_tstamp >= tmr)
2833 if (hstat.contlen == -1
2834 || hstat.orig_file_size == hstat.contlen)
2836 logprintf (LOG_VERBOSE, _("\
2837 Server file no newer than local file %s -- not retrieving.\n\n"),
2838 quote (hstat.orig_file_name));
2844 logprintf (LOG_VERBOSE, _("\
2845 The sizes do not match (local %s) -- retrieving.\n"),
2846 number_to_static_string (hstat.orig_file_size));
2850 logputs (LOG_VERBOSE,
2851 _("Remote file is newer, retrieving.\n"));
2853 logputs (LOG_VERBOSE, "\n");
2857 /* free_hstat (&hstat); */
2858 hstat.timestamp_checked = true;
2863 bool finished = true;
2868 logputs (LOG_VERBOSE, _("\
2869 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2874 logprintf (LOG_VERBOSE, _("\
2875 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2876 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2883 logprintf (LOG_VERBOSE, _("\
2884 Remote file exists and could contain further links,\n\
2885 but recursion is disabled -- not retrieving.\n\n"));
2889 logprintf (LOG_VERBOSE, _("\
2890 Remote file exists.\n\n"));
2892 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2897 logprintf (LOG_NONVERBOSE,
2898 _("%s URL: %s %2d %s\n"),
2899 tms, u->url, hstat.statcode,
2900 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2907 count = 0; /* the retrieve count for HEAD is reset */
2909 } /* send_head_first */
2912 if (opt.useservertimestamps
2913 && (tmr != (time_t) (-1))
2914 && ((hstat.len == hstat.contlen) ||
2915 ((hstat.res == 0) && (hstat.contlen == -1))))
2917 const char *fl = NULL;
2918 set_local_file (&fl, hstat.local_file);
2922 /* Reparse time header, in case it's changed. */
2923 if (time_came_from_head
2924 && hstat.remote_time && hstat.remote_time[0])
2926 newtmr = http_atotm (hstat.remote_time);
2927 if (newtmr != (time_t)-1)
2933 /* End of time-stamping section. */
2935 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2936 total_download_time += hstat.dltime;
2938 if (hstat.len == hstat.contlen)
2942 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
2944 logprintf (LOG_VERBOSE,
2946 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
2947 : _("%s (%s) - %s saved [%s/%s]\n\n"),
2949 write_to_stdout ? "" : quote (hstat.local_file),
2950 number_to_static_string (hstat.len),
2951 number_to_static_string (hstat.contlen));
2952 logprintf (LOG_NONVERBOSE,
2953 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2955 number_to_static_string (hstat.len),
2956 number_to_static_string (hstat.contlen),
2957 hstat.local_file, count);
2960 total_downloaded_bytes += hstat.len;
2962 /* Remember that we downloaded the file for later ".orig" code. */
2963 if (*dt & ADDED_HTML_EXTENSION)
2964 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2966 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2971 else if (hstat.res == 0) /* No read error */
2973 if (hstat.contlen == -1) /* We don't know how much we were supposed
2974 to get, so assume we succeeded. */
2978 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
2980 logprintf (LOG_VERBOSE,
2982 ? _("%s (%s) - written to stdout %s[%s]\n\n")
2983 : _("%s (%s) - %s saved [%s]\n\n"),
2985 write_to_stdout ? "" : quote (hstat.local_file),
2986 number_to_static_string (hstat.len));
2987 logprintf (LOG_NONVERBOSE,
2988 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2989 tms, u->url, number_to_static_string (hstat.len),
2990 hstat.local_file, count);
2993 total_downloaded_bytes += hstat.len;
2995 /* Remember that we downloaded the file for later ".orig" code. */
2996 if (*dt & ADDED_HTML_EXTENSION)
2997 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2999 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3004 else if (hstat.len < hstat.contlen) /* meaning we lost the
3005 connection too soon */
3007 logprintf (LOG_VERBOSE,
3008 _("%s (%s) - Connection closed at byte %s. "),
3009 tms, tmrate, number_to_static_string (hstat.len));
3010 printwhat (count, opt.ntry);
3013 else if (hstat.len != hstat.restval)
3014 /* Getting here would mean reading more data than
3015 requested with content-length, which we never do. */
3019 /* Getting here probably means that the content-length was
3020 * _less_ than the original, local size. We should probably
3021 * truncate or re-read, or something. FIXME */
3026 else /* from now on hstat.res can only be -1 */
3028 if (hstat.contlen == -1)
3030 logprintf (LOG_VERBOSE,
3031 _("%s (%s) - Read error at byte %s (%s)."),
3032 tms, tmrate, number_to_static_string (hstat.len),
3034 printwhat (count, opt.ntry);
3037 else /* hstat.res == -1 and contlen is given */
3039 logprintf (LOG_VERBOSE,
3040 _("%s (%s) - Read error at byte %s/%s (%s). "),
3042 number_to_static_string (hstat.len),
3043 number_to_static_string (hstat.contlen),
3045 printwhat (count, opt.ntry);
3051 while (!opt.ntry || (count < opt.ntry));
3055 *local_file = xstrdup (hstat.local_file);
3056 free_hstat (&hstat);
3061 /* Check whether the result of strptime() indicates success.
3062 strptime() returns the pointer to how far it got to in the string.
3063 The processing has been successful if the string is at `GMT' or
3064 `+X', or at the end of the string.
3066 In extended regexp parlance, the function returns 1 if P matches
3067 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3068 can return) is considered a failure and 0 is returned. */
3070 check_end (const char *p)
3074 while (c_isspace (*p))
3077 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3078 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3084 /* Convert the textual specification of time in TIME_STRING to the
3085 number of seconds since the Epoch.
3087 TIME_STRING can be in any of the three formats RFC2616 allows the
3088 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3089 as well as the time format used in the Set-Cookie header.
3090 Timezones are ignored, and should be GMT.
3092 Return the computed time_t representation, or -1 if the conversion
3095 This function uses strptime with various string formats for parsing
3096 TIME_STRING. This results in a parser that is not as lenient in
3097 interpreting TIME_STRING as I would like it to be. Being based on
3098 strptime, it always allows shortened months, one-digit days, etc.,
3099 but due to the multitude of formats in which time can be
3100 represented, an ideal HTTP time parser would be even more
3101 forgiving. It should completely ignore things like week days and
3102 concentrate only on the various forms of representing years,
3103 months, days, hours, minutes, and seconds. For example, it would
3104 be nice if it accepted ISO 8601 out of the box.
3106 I've investigated free and PD code for this purpose, but none was
3107 usable. getdate was big and unwieldy, and had potential copyright
3108 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3109 distributed with phttpd, is excellent, but we cannot use it because
3110 it is not assigned to the FSF. So I stuck it with strptime. */
3113 http_atotm (const char *time_string)
3115 /* NOTE: Solaris strptime man page claims that %n and %t match white
3116 space, but that's not universally available. Instead, we simply
3117 use ` ' to mean "skip all WS", which works under all strptime
3118 implementations I've tested. */
3120 static const char *time_formats[] = {
3121 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3122 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3123 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3124 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3125 (used in Set-Cookie, defined in the
3126 Netscape cookie specification.) */
3128 const char *oldlocale;
3129 char savedlocale[256];
3131 time_t ret = (time_t) -1;
3133 /* Solaris strptime fails to recognize English month names in
3134 non-English locales, which we work around by temporarily setting
3135 locale to C before invoking strptime. */
3136 oldlocale = setlocale (LC_TIME, NULL);
3139 size_t l = strlen (oldlocale);
3140 if (l >= sizeof savedlocale)
3141 savedlocale[0] = '\0';
3143 memcpy (savedlocale, oldlocale, l);
3145 else savedlocale[0] = '\0';
3147 setlocale (LC_TIME, "C");
3149 for (i = 0; i < countof (time_formats); i++)
3153 /* Some versions of strptime use the existing contents of struct
3154 tm to recalculate the date according to format. Zero it out
3155 to prevent stack garbage from influencing strptime. */
3158 if (check_end (strptime (time_string, time_formats[i], &t)))
3165 /* Restore the previous locale. */
3167 setlocale (LC_TIME, savedlocale);
3172 /* Authorization support: We support three authorization schemes:
3174 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3176 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3177 consisting of answering to the server's challenge with the proper
3180 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3181 Stenberg for libcurl. Like digest, NTLM is based on a
3182 challenge-response mechanism, but unlike digest, it is non-standard
3183 (authenticates TCP connections rather than requests), undocumented
3184 and Microsoft-specific. */
3186 /* Create the authentication header contents for the `Basic' scheme.
3187 This is done by encoding the string "USER:PASS" to base64 and
3188 prepending the string "Basic " in front of it. */
3191 basic_authentication_encode (const char *user, const char *passwd)
3194 int len1 = strlen (user) + 1 + strlen (passwd);
3196 t1 = (char *)alloca (len1 + 1);
3197 sprintf (t1, "%s:%s", user, passwd);
3199 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3200 base64_encode (t1, len1, t2);
3202 return concat_strings ("Basic ", t2, (char *) 0);
3205 #define SKIP_WS(x) do { \
3206 while (c_isspace (*(x))) \
3210 #ifdef ENABLE_DIGEST
3211 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3212 an array of 16 bytes containing the hash keys, and BUF should be a
3213 buffer of 33 writable characters (32 for hex digits plus one for
3214 zero termination). */
3216 dump_hash (char *buf, const unsigned char *hash)
3220 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3222 *buf++ = XNUM_TO_digit (*hash >> 4);
3223 *buf++ = XNUM_TO_digit (*hash & 0xf);
3228 /* Take the line apart to find the challenge, and compose a digest
3229 authorization header. See RFC2069 section 2.1.2. */
3231 digest_authentication_encode (const char *au, const char *user,
3232 const char *passwd, const char *method,
3235 static char *realm, *opaque, *nonce;
3240 { "realm", &realm },
3241 { "opaque", &opaque },
3245 param_token name, value;
3247 realm = opaque = nonce = NULL;
3249 au += 6; /* skip over `Digest' */
3250 while (extract_param (&au, &name, &value, ','))
3253 size_t namelen = name.e - name.b;
3254 for (i = 0; i < countof (options); i++)
3255 if (namelen == strlen (options[i].name)
3256 && 0 == strncmp (name.b, options[i].name,
3259 *options[i].variable = strdupdelim (value.b, value.e);
3263 if (!realm || !nonce || !user || !passwd || !path || !method)
3266 xfree_null (opaque);
3271 /* Calculate the digest value. */
3273 ALLOCA_MD5_CONTEXT (ctx);
3274 unsigned char hash[MD5_HASHLEN];
3275 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3276 char response_digest[MD5_HASHLEN * 2 + 1];
3278 /* A1BUF = H(user ":" realm ":" password) */
3280 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3281 gen_md5_update ((unsigned char *)":", 1, ctx);
3282 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3283 gen_md5_update ((unsigned char *)":", 1, ctx);
3284 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3285 gen_md5_finish (ctx, hash);
3286 dump_hash (a1buf, hash);
3288 /* A2BUF = H(method ":" path) */
3290 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3291 gen_md5_update ((unsigned char *)":", 1, ctx);
3292 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3293 gen_md5_finish (ctx, hash);
3294 dump_hash (a2buf, hash);
3296 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3298 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3299 gen_md5_update ((unsigned char *)":", 1, ctx);
3300 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3301 gen_md5_update ((unsigned char *)":", 1, ctx);
3302 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3303 gen_md5_finish (ctx, hash);
3304 dump_hash (response_digest, hash);
3306 res = xmalloc (strlen (user)
3311 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3312 + (opaque ? strlen (opaque) : 0)
3314 sprintf (res, "Digest \
3315 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3316 user, realm, nonce, path, response_digest);
3319 char *p = res + strlen (res);
3320 strcat (p, ", opaque=\"");
3327 #endif /* ENABLE_DIGEST */
3329 /* Computing the size of a string literal must take into account that
3330 value returned by sizeof includes the terminating \0. */
3331 #define STRSIZE(literal) (sizeof (literal) - 1)
3333 /* Whether chars in [b, e) begin with the literal string provided as
3334 first argument and are followed by whitespace or terminating \0.
3335 The comparison is case-insensitive. */
3336 #define STARTS(literal, b, e) \
3338 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3339 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3340 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3341 || c_isspace (b[STRSIZE (literal)])))
3344 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3346 return STARTS ("Basic", hdrbeg, hdrend)
3347 #ifdef ENABLE_DIGEST
3348 || STARTS ("Digest", hdrbeg, hdrend)
3351 || STARTS ("NTLM", hdrbeg, hdrend)
3358 /* Create the HTTP authorization request header. When the
3359 `WWW-Authenticate' response header is seen, according to the
3360 authorization scheme specified in that header (`Basic' and `Digest'
3361 are supported by the current implementation), produce an
3362 appropriate HTTP authorization request header. */
3364 create_authorization_line (const char *au, const char *user,
3365 const char *passwd, const char *method,
3366 const char *path, bool *finished)
3368 /* We are called only with known schemes, so we can dispatch on the
3370 switch (c_toupper (*au))
3372 case 'B': /* Basic */
3374 return basic_authentication_encode (user, passwd);
3375 #ifdef ENABLE_DIGEST
3376 case 'D': /* Digest */
3378 return digest_authentication_encode (au, user, passwd, method, path);
3381 case 'N': /* NTLM */
3382 if (!ntlm_input (&pconn.ntlm, au))
3387 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3390 /* We shouldn't get here -- this function should be only called
3391 with values approved by known_authentication_scheme_p. */
3399 if (!wget_cookie_jar)
3400 wget_cookie_jar = cookie_jar_new ();
3401 if (opt.cookies_input && !cookies_loaded_p)
3403 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3404 cookies_loaded_p = true;
3411 if (wget_cookie_jar)
3412 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3418 xfree_null (pconn.host);
3419 if (wget_cookie_jar)
3420 cookie_jar_delete (wget_cookie_jar);
3424 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3426 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3428 int len = strlen (ext);
3431 strncpy (shortext, ext, len - 1);
3432 shortext[len - 2] = '\0';
3435 if (last_period_in_local_filename == NULL
3436 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3437 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3439 int local_filename_len = strlen (hs->local_file);
3440 /* Resize the local file, allowing for ".html" preceded by
3441 optional ".NUMBER". */
3442 hs->local_file = xrealloc (hs->local_file,
3443 local_filename_len + 24 + len);
3444 strcpy (hs->local_file + local_filename_len, ext);
3445 /* If clobbering is not allowed and the file, as named,
3446 exists, tack on ".NUMBER.html" instead. */
3447 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3451 sprintf (hs->local_file + local_filename_len,
3452 ".%d%s", ext_num++, ext);
3453 while (file_exists_p (hs->local_file));
3455 *dt |= ADDED_HTML_EXTENSION;
3463 test_parse_content_disposition()
3468 char *opt_dir_prefix;
3472 { "filename=\"file.ext\"", NULL, "file.ext", true },
3473 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3474 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3475 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3476 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3477 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3478 { "attachment", NULL, NULL, false },
3479 { "attachment", "somedir", NULL, false },
3480 { "attachement; filename*=UTF-8'en-US'hello.txt", NULL, "hello.txt", true },
3481 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", NULL, "helloworld.txt", true },
3484 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3489 opt.dir_prefix = test_array[i].opt_dir_prefix;
3490 res = parse_content_disposition (test_array[i].hdrval, &filename);
3492 mu_assert ("test_parse_content_disposition: wrong result",
3493 res == test_array[i].result
3495 || 0 == strcmp (test_array[i].filename, filename)));
3501 #endif /* TESTING */
3504 * vim: et sts=2 sw=2 cino+={s