2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009, 2010 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"
69 #endif /* def __VMS */
71 extern char *version_string;
75 static char *create_authorization_line (const char *, const char *,
76 const char *, const char *,
77 const char *, bool *);
78 static char *basic_authentication_encode (const char *, const char *);
79 static bool known_authentication_scheme_p (const char *, const char *);
80 static void ensure_extension (struct http_stat *, const char *, int *);
81 static void load_cookies (void);
84 # define MIN(x, y) ((x) > (y) ? (y) : (x))
88 static bool cookies_loaded_p;
89 static struct cookie_jar *wget_cookie_jar;
91 #define TEXTHTML_S "text/html"
92 #define TEXTXHTML_S "application/xhtml+xml"
93 #define TEXTCSS_S "text/css"
95 /* Some status code validation macros: */
96 #define H_10X(x) (((x) >= 100) && ((x) < 200))
97 #define H_20X(x) (((x) >= 200) && ((x) < 300))
98 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
99 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
100 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
101 || (x) == HTTP_STATUS_SEE_OTHER \
102 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
104 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
105 /* Successful 2xx. */
106 #define HTTP_STATUS_OK 200
107 #define HTTP_STATUS_CREATED 201
108 #define HTTP_STATUS_ACCEPTED 202
109 #define HTTP_STATUS_NO_CONTENT 204
110 #define HTTP_STATUS_PARTIAL_CONTENTS 206
112 /* Redirection 3xx. */
113 #define HTTP_STATUS_MULTIPLE_CHOICES 300
114 #define HTTP_STATUS_MOVED_PERMANENTLY 301
115 #define HTTP_STATUS_MOVED_TEMPORARILY 302
116 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
117 #define HTTP_STATUS_NOT_MODIFIED 304
118 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
120 /* Client error 4xx. */
121 #define HTTP_STATUS_BAD_REQUEST 400
122 #define HTTP_STATUS_UNAUTHORIZED 401
123 #define HTTP_STATUS_FORBIDDEN 403
124 #define HTTP_STATUS_NOT_FOUND 404
125 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
127 /* Server errors 5xx. */
128 #define HTTP_STATUS_INTERNAL 500
129 #define HTTP_STATUS_NOT_IMPLEMENTED 501
130 #define HTTP_STATUS_BAD_GATEWAY 502
131 #define HTTP_STATUS_UNAVAILABLE 503
134 rel_none, rel_name, rel_value, rel_both
141 struct request_header {
143 enum rp release_policy;
145 int hcount, hcapacity;
150 /* Create a new, empty request. At least request_set_method must be
151 called before the request can be used. */
153 static struct request *
156 struct request *req = xnew0 (struct request);
158 req->headers = xnew_array (struct request_header, req->hcapacity);
162 /* Set the request's method and its arguments. METH should be a
163 literal string (or it should outlive the request) because it will
164 not be freed. ARG will be freed by request_free. */
167 request_set_method (struct request *req, const char *meth, char *arg)
173 /* Return the method string passed with the last call to
174 request_set_method. */
177 request_method (const struct request *req)
182 /* Free one header according to the release policy specified with
183 request_set_header. */
186 release_header (struct request_header *hdr)
188 switch (hdr->release_policy)
205 /* Set the request named NAME to VALUE. Specifically, this means that
206 a "NAME: VALUE\r\n" header line will be used in the request. If a
207 header with the same name previously existed in the request, its
208 value will be replaced by this one. A NULL value means do nothing.
210 RELEASE_POLICY determines whether NAME and VALUE should be released
211 (freed) with request_free. Allowed values are:
213 - rel_none - don't free NAME or VALUE
214 - rel_name - free NAME when done
215 - rel_value - free VALUE when done
216 - rel_both - free both NAME and VALUE when done
218 Setting release policy is useful when arguments come from different
219 sources. For example:
221 // Don't free literal strings!
222 request_set_header (req, "Pragma", "no-cache", rel_none);
224 // Don't free a global variable, we'll need it later.
225 request_set_header (req, "Referer", opt.referer, rel_none);
227 // Value freshly allocated, free it when done.
228 request_set_header (req, "Range",
229 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
234 request_set_header (struct request *req, char *name, char *value,
235 enum rp release_policy)
237 struct request_header *hdr;
242 /* A NULL value is a no-op; if freeing the name is requested,
243 free it now to avoid leaks. */
244 if (release_policy == rel_name || release_policy == rel_both)
249 for (i = 0; i < req->hcount; i++)
251 hdr = &req->headers[i];
252 if (0 == strcasecmp (name, hdr->name))
254 /* Replace existing header. */
255 release_header (hdr);
258 hdr->release_policy = release_policy;
263 /* Install new header. */
265 if (req->hcount >= req->hcapacity)
267 req->hcapacity <<= 1;
268 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
270 hdr = &req->headers[req->hcount++];
273 hdr->release_policy = release_policy;
276 /* Like request_set_header, but sets the whole header line, as
277 provided by the user using the `--header' option. For example,
278 request_set_user_header (req, "Foo: bar") works just like
279 request_set_header (req, "Foo", "bar"). */
282 request_set_user_header (struct request *req, const char *header)
285 const char *p = strchr (header, ':');
288 BOUNDED_TO_ALLOCA (header, p, name);
290 while (c_isspace (*p))
292 request_set_header (req, xstrdup (name), (char *) p, rel_name);
295 /* Remove the header with specified name from REQ. Returns true if
296 the header was actually removed, false otherwise. */
299 request_remove_header (struct request *req, char *name)
302 for (i = 0; i < req->hcount; i++)
304 struct request_header *hdr = &req->headers[i];
305 if (0 == strcasecmp (name, hdr->name))
307 release_header (hdr);
308 /* Move the remaining headers by one. */
309 if (i < req->hcount - 1)
310 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
318 #define APPEND(p, str) do { \
319 int A_len = strlen (str); \
320 memcpy (p, str, A_len); \
324 /* Construct the request and write it to FD using fd_write. */
327 request_send (const struct request *req, int fd)
329 char *request_string, *p;
330 int i, size, write_error;
332 /* Count the request size. */
335 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
336 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
338 for (i = 0; i < req->hcount; i++)
340 struct request_header *hdr = &req->headers[i];
341 /* NAME ": " VALUE "\r\n" */
342 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
348 p = request_string = alloca_array (char, size);
350 /* Generate the request. */
352 APPEND (p, req->method); *p++ = ' ';
353 APPEND (p, req->arg); *p++ = ' ';
354 memcpy (p, "HTTP/1.1\r\n", 10); p += 10;
356 for (i = 0; i < req->hcount; i++)
358 struct request_header *hdr = &req->headers[i];
359 APPEND (p, hdr->name);
360 *p++ = ':', *p++ = ' ';
361 APPEND (p, hdr->value);
362 *p++ = '\r', *p++ = '\n';
365 *p++ = '\r', *p++ = '\n', *p++ = '\0';
366 assert (p - request_string == size);
370 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
372 /* Send the request to the server. */
374 write_error = fd_write (fd, request_string, size - 1, -1);
376 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
381 /* Release the resources used by REQ. */
384 request_free (struct request *req)
387 xfree_null (req->arg);
388 for (i = 0; i < req->hcount; i++)
389 release_header (&req->headers[i]);
390 xfree_null (req->headers);
394 static struct hash_table *basic_authed_hosts;
396 /* Find out if this host has issued a Basic challenge yet; if so, give
397 * it the username, password. A temporary measure until we can get
398 * proper authentication in place. */
401 maybe_send_basic_creds (const char *hostname, const char *user,
402 const char *passwd, struct request *req)
404 bool do_challenge = false;
406 if (opt.auth_without_challenge)
408 DEBUGP (("Auth-without-challenge set, sending Basic credentials.\n"));
411 else if (basic_authed_hosts
412 && hash_table_contains(basic_authed_hosts, hostname))
414 DEBUGP (("Found %s in basic_authed_hosts.\n", quote (hostname)));
419 DEBUGP (("Host %s has not issued a general basic challenge.\n",
424 request_set_header (req, "Authorization",
425 basic_authentication_encode (user, passwd),
432 register_basic_auth_host (const char *hostname)
434 if (!basic_authed_hosts)
436 basic_authed_hosts = make_nocase_string_hash_table (1);
438 if (!hash_table_contains(basic_authed_hosts, hostname))
440 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
441 DEBUGP (("Inserted %s into basic_authed_hosts\n", quote (hostname)));
446 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
447 PROMISED_SIZE bytes are sent over the wire -- if the file is
448 longer, read only that much; if the file is shorter, report an error. */
451 post_file (int sock, const char *file_name, wgint promised_size)
453 static char chunk[8192];
458 DEBUGP (("[writing POST file %s ... ", file_name));
460 fp = fopen (file_name, "rb");
463 while (!feof (fp) && written < promised_size)
466 int length = fread (chunk, 1, sizeof (chunk), fp);
469 towrite = MIN (promised_size - written, length);
470 write_error = fd_write (sock, chunk, towrite, -1);
480 /* If we've written less than was promised, report a (probably
481 nonsensical) error rather than break the promise. */
482 if (written < promised_size)
488 assert (written == promised_size);
489 DEBUGP (("done]\n"));
493 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
494 If so, return the pointer to the position after the line, otherwise
495 return NULL. This is used as callback to fd_read_hunk. The data
496 between START and PEEKED has been read and cannot be "unread"; the
497 data after PEEKED has only been peeked. */
500 response_head_terminator (const char *start, const char *peeked, int peeklen)
504 /* If at first peek, verify whether HUNK starts with "HTTP". If
505 not, this is a HTTP/0.9 request and we must bail out without
507 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
510 /* Look for "\n[\r]\n", and return the following position if found.
511 Start two chars before the current to cover the possibility that
512 part of the terminator (e.g. "\n\r") arrived in the previous
514 p = peeked - start < 2 ? start : peeked - 2;
515 end = peeked + peeklen;
517 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
518 for (; p < end - 2; p++)
521 if (p[1] == '\r' && p[2] == '\n')
523 else if (p[1] == '\n')
526 /* p==end-2: check for \n\n directly preceding END. */
527 if (p[0] == '\n' && p[1] == '\n')
533 /* The maximum size of a single HTTP response we care to read. Rather
534 than being a limit of the reader implementation, this limit
535 prevents Wget from slurping all available memory upon encountering
536 malicious or buggy server output, thus protecting the user. Define
537 it to 0 to remove the limit. */
539 #define HTTP_RESPONSE_MAX_SIZE 65536
541 /* Read the HTTP request head from FD and return it. The error
542 conditions are the same as with fd_read_hunk.
544 To support HTTP/0.9 responses, this function tries to make sure
545 that the data begins with "HTTP". If this is not the case, no data
546 is read and an empty request is returned, so that the remaining
547 data can be treated as body. */
550 read_http_response_head (int fd)
552 return fd_read_hunk (fd, response_head_terminator, 512,
553 HTTP_RESPONSE_MAX_SIZE);
557 /* The response data. */
560 /* The array of pointers that indicate where each header starts.
561 For example, given this HTTP response:
568 The headers are located like this:
570 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
572 headers[0] headers[1] headers[2] headers[3]
574 I.e. headers[0] points to the beginning of the request,
575 headers[1] points to the end of the first header and the
576 beginning of the second one, etc. */
578 const char **headers;
581 /* Create a new response object from the text of the HTTP response,
582 available in HEAD. That text is automatically split into
583 constituent header lines for fast retrieval using
586 static struct response *
587 resp_new (const char *head)
592 struct response *resp = xnew0 (struct response);
597 /* Empty head means that we're dealing with a headerless
598 (HTTP/0.9) response. In that case, don't set HEADERS at
603 /* Split HEAD into header lines, so that resp_header_* functions
604 don't need to do this over and over again. */
610 DO_REALLOC (resp->headers, size, count + 1, const char *);
611 resp->headers[count++] = hdr;
613 /* Break upon encountering an empty line. */
614 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
617 /* Find the end of HDR, including continuations. */
620 const char *end = strchr (hdr, '\n');
626 while (*hdr == ' ' || *hdr == '\t');
628 DO_REALLOC (resp->headers, size, count + 1, const char *);
629 resp->headers[count] = NULL;
634 /* Locate the header named NAME in the request data, starting with
635 position START. This allows the code to loop through the request
636 data, filtering for all requests of a given name. Returns the
637 found position, or -1 for failure. The code that uses this
638 function typically looks like this:
640 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
641 ... do something with header ...
643 If you only care about one header, use resp_header_get instead of
647 resp_header_locate (const struct response *resp, const char *name, int start,
648 const char **begptr, const char **endptr)
651 const char **headers = resp->headers;
654 if (!headers || !headers[1])
657 name_len = strlen (name);
663 for (; headers[i + 1]; i++)
665 const char *b = headers[i];
666 const char *e = headers[i + 1];
668 && b[name_len] == ':'
669 && 0 == strncasecmp (b, name, name_len))
672 while (b < e && c_isspace (*b))
674 while (b < e && c_isspace (e[-1]))
684 /* Find and retrieve the header named NAME in the request data. If
685 found, set *BEGPTR to its starting, and *ENDPTR to its ending
686 position, and return true. Otherwise return false.
688 This function is used as a building block for resp_header_copy
689 and resp_header_strdup. */
692 resp_header_get (const struct response *resp, const char *name,
693 const char **begptr, const char **endptr)
695 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
699 /* Copy the response header named NAME to buffer BUF, no longer than
700 BUFSIZE (BUFSIZE includes the terminating 0). If the header
701 exists, true is returned, false otherwise. If there should be no
702 limit on the size of the header, use resp_header_strdup instead.
704 If BUFSIZE is 0, no data is copied, but the boolean indication of
705 whether the header is present is still returned. */
708 resp_header_copy (const struct response *resp, const char *name,
709 char *buf, int bufsize)
712 if (!resp_header_get (resp, name, &b, &e))
716 int len = MIN (e - b, bufsize - 1);
717 memcpy (buf, b, len);
723 /* Return the value of header named NAME in RESP, allocated with
724 malloc. If such a header does not exist in RESP, return NULL. */
727 resp_header_strdup (const struct response *resp, const char *name)
730 if (!resp_header_get (resp, name, &b, &e))
732 return strdupdelim (b, e);
735 /* Parse the HTTP status line, which is of format:
737 HTTP-Version SP Status-Code SP Reason-Phrase
739 The function returns the status-code, or -1 if the status line
740 appears malformed. The pointer to "reason-phrase" message is
741 returned in *MESSAGE. */
744 resp_status (const struct response *resp, char **message)
751 /* For a HTTP/0.9 response, assume status 200. */
753 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
757 p = resp->headers[0];
758 end = resp->headers[1];
764 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
768 /* Match the HTTP version. This is optional because Gnutella
769 servers have been reported to not specify HTTP version. */
770 if (p < end && *p == '/')
773 while (p < end && c_isdigit (*p))
775 if (p < end && *p == '.')
777 while (p < end && c_isdigit (*p))
781 while (p < end && c_isspace (*p))
783 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
786 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
791 while (p < end && c_isspace (*p))
793 while (p < end && c_isspace (end[-1]))
795 *message = strdupdelim (p, end);
801 /* Release the resources used by RESP. */
804 resp_free (struct response *resp)
806 xfree_null (resp->headers);
810 /* Print a single line of response, the characters [b, e). We tried
812 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
813 but that failed to escape the non-printable characters and, in fact,
814 caused crashes in UTF-8 locales. */
817 print_response_line(const char *prefix, const char *b, const char *e)
820 BOUNDED_TO_ALLOCA(b, e, copy);
821 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
822 quotearg_style (escape_quoting_style, copy));
825 /* Print the server response, line by line, omitting the trailing CRLF
826 from individual header lines, and prefixed with PREFIX. */
829 print_server_response (const struct response *resp, const char *prefix)
834 for (i = 0; resp->headers[i + 1]; i++)
836 const char *b = resp->headers[i];
837 const char *e = resp->headers[i + 1];
839 if (b < e && e[-1] == '\n')
841 if (b < e && e[-1] == '\r')
843 print_response_line(prefix, b, e);
847 /* Parse the `Content-Range' header and extract the information it
848 contains. Returns true if successful, false otherwise. */
850 parse_content_range (const char *hdr, wgint *first_byte_ptr,
851 wgint *last_byte_ptr, wgint *entity_length_ptr)
855 /* Ancient versions of Netscape proxy server, presumably predating
856 rfc2068, sent out `Content-Range' without the "bytes"
858 if (0 == strncasecmp (hdr, "bytes", 5))
861 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
865 while (c_isspace (*hdr))
870 if (!c_isdigit (*hdr))
872 for (num = 0; c_isdigit (*hdr); hdr++)
873 num = 10 * num + (*hdr - '0');
874 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
876 *first_byte_ptr = num;
878 for (num = 0; c_isdigit (*hdr); hdr++)
879 num = 10 * num + (*hdr - '0');
880 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
882 *last_byte_ptr = num;
887 for (num = 0; c_isdigit (*hdr); hdr++)
888 num = 10 * num + (*hdr - '0');
889 *entity_length_ptr = num;
893 /* Read the body of the request, but don't store it anywhere and don't
894 display a progress gauge. This is useful for reading the bodies of
895 administrative responses to which we will soon issue another
896 request. The response is not useful to the user, but reading it
897 allows us to continue using the same connection to the server.
899 If reading fails, false is returned, true otherwise. In debug
900 mode, the body is displayed for debugging purposes. */
903 skip_short_body (int fd, wgint contlen, bool chunked)
906 SKIP_SIZE = 512, /* size of the download buffer */
907 SKIP_THRESHOLD = 4096 /* the largest size we read */
909 wgint remaining_chunk_size = 0;
910 char dlbuf[SKIP_SIZE + 1];
911 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
913 assert (contlen != -1 || contlen);
915 /* If the body is too large, it makes more sense to simply close the
916 connection than to try to read the body. */
917 if (contlen > SKIP_THRESHOLD)
920 while (contlen > 0 || chunked)
925 if (remaining_chunk_size == 0)
927 char *line = fd_read_line (fd);
935 remaining_chunk_size = strtol (line, &endl, 16);
936 if (remaining_chunk_size == 0)
939 if (fd_read_line (fd) == NULL)
945 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
948 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
950 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
953 /* Don't normally report the error since this is an
954 optimization that should be invisible to the user. */
955 DEBUGP (("] aborting (%s).\n",
956 ret < 0 ? fd_errstr (fd) : "EOF received"));
963 remaining_chunk_size -= ret;
964 if (remaining_chunk_size == 0)
965 if (fd_read_line (fd) == NULL)
969 /* Safe even if %.*s bogusly expects terminating \0 because
970 we've zero-terminated dlbuf above. */
971 DEBUGP (("%.*s", ret, dlbuf));
974 DEBUGP (("] done.\n"));
978 #define NOT_RFC2231 0
979 #define RFC2231_NOENCODING 1
980 #define RFC2231_ENCODING 2
982 /* extract_param extracts the parameter name into NAME.
983 However, if the parameter name is in RFC2231 format then
984 this function adjusts NAME by stripping of the trailing
985 characters that are not part of the name but are present to
986 indicate the presence of encoding information in the value
987 or a fragment of a long parameter value
990 modify_param_name(param_token *name)
992 const char *delim1 = memchr (name->b, '*', name->e - name->b);
993 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
999 result = NOT_RFC2231;
1001 else if(delim1 == delim2)
1003 if ((name->e - 1) == delim1)
1005 result = RFC2231_ENCODING;
1009 result = RFC2231_NOENCODING;
1016 result = RFC2231_ENCODING;
1021 /* extract_param extract the paramater value into VALUE.
1022 Like modify_param_name this function modifies VALUE by
1023 stripping off the encoding information from the actual value
1026 modify_param_value (param_token *value, int encoding_type )
1028 if (RFC2231_ENCODING == encoding_type)
1030 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1031 if ( delim != NULL )
1033 value->b = (delim+1);
1038 /* Extract a parameter from the string (typically an HTTP header) at
1039 **SOURCE and advance SOURCE to the next parameter. Return false
1040 when there are no more parameters to extract. The name of the
1041 parameter is returned in NAME, and the value in VALUE. If the
1042 parameter has no value, the token's value is zeroed out.
1044 For example, if *SOURCE points to the string "attachment;
1045 filename=\"foo bar\"", the first call to this function will return
1046 the token named "attachment" and no value, and the second call will
1047 return the token named "filename" and value "foo bar". The third
1048 call will return false, indicating no more valid tokens. */
1051 extract_param (const char **source, param_token *name, param_token *value,
1054 const char *p = *source;
1056 while (c_isspace (*p)) ++p;
1060 return false; /* no error; nothing more to extract */
1065 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1067 if (name->b == name->e)
1068 return false; /* empty name: error */
1069 while (c_isspace (*p)) ++p;
1070 if (*p == separator || !*p) /* no value */
1073 if (*p == separator) ++p;
1078 return false; /* error */
1080 /* *p is '=', extract value */
1082 while (c_isspace (*p)) ++p;
1083 if (*p == '"') /* quoted */
1086 while (*p && *p != '"') ++p;
1090 /* Currently at closing quote; find the end of param. */
1091 while (c_isspace (*p)) ++p;
1092 while (*p && *p != separator) ++p;
1093 if (*p == separator)
1096 /* garbage after closed quote, e.g. foo="bar"baz */
1102 while (*p && *p != separator) ++p;
1104 while (value->e != value->b && c_isspace (value->e[-1]))
1106 if (*p == separator) ++p;
1110 int param_type = modify_param_name(name);
1111 if (NOT_RFC2231 != param_type)
1113 modify_param_value(value, param_type);
1119 #undef RFC2231_NOENCODING
1120 #undef RFC2231_ENCODING
1122 /* Appends the string represented by VALUE to FILENAME */
1125 append_value_to_filename (char **filename, param_token const * const value)
1127 int original_length = strlen(*filename);
1128 int new_length = strlen(*filename) + (value->e - value->b);
1129 *filename = xrealloc (*filename, new_length+1);
1130 memcpy (*filename + original_length, value->b, (value->e - value->b));
1131 (*filename)[new_length] = '\0';
1135 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1137 /* Parse the contents of the `Content-Disposition' header, extracting
1138 the information useful to Wget. Content-Disposition is a header
1139 borrowed from MIME; when used in HTTP, it typically serves for
1140 specifying the desired file name of the resource. For example:
1142 Content-Disposition: attachment; filename="flora.jpg"
1144 Wget will skip the tokens it doesn't care about, such as
1145 "attachment" in the previous example; it will also skip other
1146 unrecognized params. If the header is syntactically correct and
1147 contains a file name, a copy of the file name is stored in
1148 *filename and true is returned. Otherwise, the function returns
1151 The file name is stripped of directory components and must not be
1154 parse_content_disposition (const char *hdr, char **filename)
1157 param_token name, value;
1158 while (extract_param (&hdr, &name, &value, ';'))
1160 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1161 if ( isFilename && value.b != NULL)
1163 /* Make the file name begin at the last slash or backslash. */
1164 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1165 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1166 if (last_slash && last_bs)
1167 value.b = 1 + MAX (last_slash, last_bs);
1168 else if (last_slash || last_bs)
1169 value.b = 1 + (last_slash ? last_slash : last_bs);
1170 if (value.b == value.e)
1172 /* Start with the directory prefix, if specified. */
1177 int prefix_length = strlen (opt.dir_prefix);
1178 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1183 total_length = prefix_length + (value.e - value.b);
1184 *filename = xmalloc (total_length + 1);
1185 strcpy (*filename, opt.dir_prefix);
1187 (*filename)[prefix_length - 1] = '/';
1188 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1189 (*filename)[total_length] = '\0';
1193 append_value_to_filename (filename, &value);
1200 append_value_to_filename (filename, &value);
1204 *filename = strdupdelim (value.b, value.e);
1220 /* Persistent connections. Currently, we cache the most recently used
1221 connection as persistent, provided that the HTTP server agrees to
1222 make it such. The persistence data is stored in the variables
1223 below. Ideally, it should be possible to cache an arbitrary fixed
1224 number of these connections. */
1226 /* Whether a persistent connection is active. */
1227 static bool pconn_active;
1230 /* The socket of the connection. */
1233 /* Host and port of the currently active persistent connection. */
1237 /* Whether a ssl handshake has occoured on this connection. */
1240 /* Whether the connection was authorized. This is only done by
1241 NTLM, which authorizes *connections* rather than individual
1242 requests. (That practice is peculiar for HTTP, but it is a
1243 useful optimization.) */
1247 /* NTLM data of the current connection. */
1248 struct ntlmdata ntlm;
1252 /* Mark the persistent connection as invalid and free the resources it
1253 uses. This is used by the CLOSE_* macros after they forcefully
1254 close a registered persistent connection. */
1257 invalidate_persistent (void)
1259 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1260 pconn_active = false;
1261 fd_close (pconn.socket);
1266 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1267 persistent. This will enable someone to use the same connection
1268 later. In the context of HTTP, this must be called only AFTER the
1269 response has been received and the server has promised that the
1270 connection will remain alive.
1272 If a previous connection was persistent, it is closed. */
1275 register_persistent (const char *host, int port, int fd, bool ssl)
1279 if (pconn.socket == fd)
1281 /* The connection FD is already registered. */
1286 /* The old persistent connection is still active; close it
1287 first. This situation arises whenever a persistent
1288 connection exists, but we then connect to a different
1289 host, and try to register a persistent connection to that
1291 invalidate_persistent ();
1295 pconn_active = true;
1297 pconn.host = xstrdup (host);
1300 pconn.authorized = false;
1302 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1305 /* Return true if a persistent connection is available for connecting
1309 persistent_available_p (const char *host, int port, bool ssl,
1310 bool *host_lookup_failed)
1312 /* First, check whether a persistent connection is active at all. */
1316 /* If we want SSL and the last connection wasn't or vice versa,
1317 don't use it. Checking for host and port is not enough because
1318 HTTP and HTTPS can apparently coexist on the same port. */
1319 if (ssl != pconn.ssl)
1322 /* If we're not connecting to the same port, we're not interested. */
1323 if (port != pconn.port)
1326 /* If the host is the same, we're in business. If not, there is
1327 still hope -- read below. */
1328 if (0 != strcasecmp (host, pconn.host))
1330 /* Check if pconn.socket is talking to HOST under another name.
1331 This happens often when both sites are virtual hosts
1332 distinguished only by name and served by the same network
1333 interface, and hence the same web server (possibly set up by
1334 the ISP and serving many different web sites). This
1335 admittedly unconventional optimization does not contradict
1336 HTTP and works well with popular server software. */
1340 struct address_list *al;
1343 /* Don't try to talk to two different SSL sites over the same
1344 secure connection! (Besides, it's not clear that
1345 name-based virtual hosting is even possible with SSL.) */
1348 /* If pconn.socket's peer is one of the IP addresses HOST
1349 resolves to, pconn.socket is for all intents and purposes
1350 already talking to HOST. */
1352 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1354 /* Can't get the peer's address -- something must be very
1355 wrong with the connection. */
1356 invalidate_persistent ();
1359 al = lookup_host (host, 0);
1362 *host_lookup_failed = true;
1366 found = address_list_contains (al, &ip);
1367 address_list_release (al);
1372 /* The persistent connection's peer address was found among the
1373 addresses HOST resolved to; therefore, pconn.sock is in fact
1374 already talking to HOST -- no need to reconnect. */
1377 /* Finally, check whether the connection is still open. This is
1378 important because most servers implement liberal (short) timeout
1379 on persistent connections. Wget can of course always reconnect
1380 if the connection doesn't work out, but it's nicer to know in
1381 advance. This test is a logical followup of the first test, but
1382 is "expensive" and therefore placed at the end of the list.
1384 (Current implementation of test_socket_open has a nice side
1385 effect that it treats sockets with pending data as "closed".
1386 This is exactly what we want: if a broken server sends message
1387 body in response to HEAD, or if it sends more than conent-length
1388 data, we won't reuse the corrupted connection.) */
1390 if (!test_socket_open (pconn.socket))
1392 /* Oops, the socket is no longer open. Now that we know that,
1393 let's invalidate the persistent connection before returning
1395 invalidate_persistent ();
1402 /* The idea behind these two CLOSE macros is to distinguish between
1403 two cases: one when the job we've been doing is finished, and we
1404 want to close the connection and leave, and two when something is
1405 seriously wrong and we're closing the connection as part of
1408 In case of keep_alive, CLOSE_FINISH should leave the connection
1409 open, while CLOSE_INVALIDATE should still close it.
1411 Note that the semantics of the flag `keep_alive' is "this
1412 connection *will* be reused (the server has promised not to close
1413 the connection once we're done)", while the semantics of
1414 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1415 active, registered connection". */
1417 #define CLOSE_FINISH(fd) do { \
1420 if (pconn_active && (fd) == pconn.socket) \
1421 invalidate_persistent (); \
1430 #define CLOSE_INVALIDATE(fd) do { \
1431 if (pconn_active && (fd) == pconn.socket) \
1432 invalidate_persistent (); \
1440 wgint len; /* received length */
1441 wgint contlen; /* expected length */
1442 wgint restval; /* the restart value */
1443 int res; /* the result of last read */
1444 char *rderrmsg; /* error message from read error */
1445 char *newloc; /* new location (redirection) */
1446 char *remote_time; /* remote time-stamp string */
1447 char *error; /* textual HTTP error */
1448 int statcode; /* status code */
1449 char *message; /* status message */
1450 wgint rd_size; /* amount of data read from socket */
1451 double dltime; /* time it took to download the data */
1452 const char *referer; /* value of the referer header. */
1453 char *local_file; /* local file name. */
1454 bool existence_checked; /* true if we already checked for a file's
1455 existence after having begun to download
1456 (needed in gethttp for when connection is
1457 interrupted/restarted. */
1458 bool timestamp_checked; /* true if pre-download time-stamping checks
1459 * have already been performed */
1460 char *orig_file_name; /* name of file to compare for time-stamping
1461 * (might be != local_file if -K is set) */
1462 wgint orig_file_size; /* size of file to compare for time-stamping */
1463 time_t orig_file_tstamp; /* time-stamp of file to compare for
1468 free_hstat (struct http_stat *hs)
1470 xfree_null (hs->newloc);
1471 xfree_null (hs->remote_time);
1472 xfree_null (hs->error);
1473 xfree_null (hs->rderrmsg);
1474 xfree_null (hs->local_file);
1475 xfree_null (hs->orig_file_name);
1476 xfree_null (hs->message);
1478 /* Guard against being called twice. */
1480 hs->remote_time = NULL;
1484 #define BEGINS_WITH(line, string_constant) \
1485 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1486 && (c_isspace (line[sizeof (string_constant) - 1]) \
1487 || !line[sizeof (string_constant) - 1]))
1490 #define SET_USER_AGENT(req) do { \
1491 if (!opt.useragent) \
1492 request_set_header (req, "User-Agent", \
1493 aprintf ("Wget/%s (VMS %s %s)", \
1494 version_string, vms_arch(), vms_vers()), \
1496 else if (*opt.useragent) \
1497 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1499 #else /* def __VMS */
1500 #define SET_USER_AGENT(req) do { \
1501 if (!opt.useragent) \
1502 request_set_header (req, "User-Agent", \
1503 aprintf ("Wget/%s (%s)", \
1504 version_string, OS_TYPE), \
1506 else if (*opt.useragent) \
1507 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1509 #endif /* def __VMS [else] */
1511 /* The flags that allow clobbering the file (opening with "wb").
1512 Defined here to avoid repetition later. #### This will require
1514 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1515 || opt.dirstruct || opt.output_document)
1517 /* Retrieve a document through HTTP protocol. It recognizes status
1518 code, and correctly handles redirections. It closes the network
1519 socket. If it receives an error from the functions below it, it
1520 will print it if there is enough information to do so (almost
1521 always), returning the error to the caller (i.e. http_loop).
1523 Various HTTP parameters are stored to hs.
1525 If PROXY is non-NULL, the connection will be made to the proxy
1526 server, and u->url will be requested. */
1528 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1531 struct request *req;
1534 char *user, *passwd;
1538 wgint contlen, contrange;
1545 /* Set to 1 when the authorization has already been sent and should
1546 not be tried again. */
1547 bool auth_finished = false;
1549 /* Set to 1 when just globally-set Basic authorization has been sent;
1550 * should prevent further Basic negotiations, but not other
1552 bool basic_auth_finished = false;
1554 /* Whether NTLM authentication is used for this request. */
1555 bool ntlm_seen = false;
1557 /* Whether our connection to the remote host is through SSL. */
1558 bool using_ssl = false;
1560 /* Whether a HEAD request will be issued (as opposed to GET or
1562 bool head_only = !!(*dt & HEAD_ONLY);
1565 struct response *resp;
1569 /* Whether this connection will be kept alive after the HTTP request
1573 /* Is the server using the chunked transfer encoding? */
1574 bool chunked_transfer_encoding = false;
1576 /* Whether keep-alive should be inhibited. */
1577 bool inhibit_keep_alive =
1578 !opt.http_keep_alive || opt.ignore_length;
1580 /* Headers sent when using POST. */
1581 wgint post_data_size = 0;
1583 bool host_lookup_failed = false;
1586 if (u->scheme == SCHEME_HTTPS)
1588 /* Initialize the SSL context. After this has once been done,
1589 it becomes a no-op. */
1592 scheme_disable (SCHEME_HTTPS);
1593 logprintf (LOG_NOTQUIET,
1594 _("Disabling SSL due to encountered errors.\n"));
1595 return SSLINITFAILED;
1598 #endif /* HAVE_SSL */
1600 /* Initialize certain elements of struct http_stat. */
1604 hs->rderrmsg = NULL;
1606 hs->remote_time = NULL;
1612 /* Prepare the request to send. */
1614 req = request_new ();
1617 const char *meth = "GET";
1620 else if (opt.post_file_name || opt.post_data)
1622 /* Use the full path, i.e. one that includes the leading slash and
1623 the query string. E.g. if u->path is "foo/bar" and u->query is
1624 "param=value", full_path will be "/foo/bar?param=value". */
1627 /* When using SSL over proxy, CONNECT establishes a direct
1628 connection to the HTTPS server. Therefore use the same
1629 argument as when talking to the server directly. */
1630 && u->scheme != SCHEME_HTTPS
1633 meth_arg = xstrdup (u->url);
1635 meth_arg = url_full_path (u);
1636 request_set_method (req, meth, meth_arg);
1639 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1640 if (*dt & SEND_NOCACHE)
1641 request_set_header (req, "Pragma", "no-cache", rel_none);
1642 if (hs->restval && !opt.timestamping)
1643 request_set_header (req, "Range",
1644 aprintf ("bytes=%s-",
1645 number_to_static_string (hs->restval)),
1647 SET_USER_AGENT (req);
1648 request_set_header (req, "Accept", "*/*", rel_none);
1650 /* Find the username and password for authentication. */
1653 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1654 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1655 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1657 /* We only do "site-wide" authentication with "global" user/password
1658 * values unless --auth-no-challange has been requested; URL user/password
1659 * info overrides. */
1660 if (user && passwd && (!u->user || opt.auth_without_challenge))
1662 /* If this is a host for which we've already received a Basic
1663 * challenge, we'll go ahead and send Basic authentication creds. */
1664 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1667 /* Generate the Host header, HOST:PORT. Take into account that:
1669 - Broken server-side software often doesn't recognize the PORT
1670 argument, so we must generate "Host: www.server.com" instead of
1671 "Host: www.server.com:80" (and likewise for https port).
1673 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1674 becomes ambiguous and needs to be rewritten as "Host:
1675 [3ffe:8100:200:2::2]:1234". */
1677 /* Formats arranged for hfmt[add_port][add_squares]. */
1678 static const char *hfmt[][2] = {
1679 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1681 int add_port = u->port != scheme_default_port (u->scheme);
1682 int add_squares = strchr (u->host, ':') != NULL;
1683 request_set_header (req, "Host",
1684 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1688 if (inhibit_keep_alive)
1689 request_set_header (req, "Connection", "Close", rel_none);
1693 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1696 request_set_header (req, "Connection", "Close", rel_none);
1697 request_set_header (req, "Proxy-Connection", "Keep-Alive", rel_none);
1701 if (opt.post_data || opt.post_file_name)
1703 request_set_header (req, "Content-Type",
1704 "application/x-www-form-urlencoded", rel_none);
1706 post_data_size = strlen (opt.post_data);
1709 post_data_size = file_size (opt.post_file_name);
1710 if (post_data_size == -1)
1712 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1713 quote (opt.post_file_name), strerror (errno));
1717 request_set_header (req, "Content-Length",
1718 xstrdup (number_to_static_string (post_data_size)),
1723 /* We need to come back here when the initial attempt to retrieve
1724 without authorization header fails. (Expected to happen at least
1725 for the Digest authorization scheme.) */
1728 request_set_header (req, "Cookie",
1729 cookie_header (wget_cookie_jar,
1730 u->host, u->port, u->path,
1732 u->scheme == SCHEME_HTTPS
1739 /* Add the user headers. */
1740 if (opt.user_headers)
1743 for (i = 0; opt.user_headers[i]; i++)
1744 request_set_user_header (req, opt.user_headers[i]);
1750 char *proxy_user, *proxy_passwd;
1751 /* For normal username and password, URL components override
1752 command-line/wgetrc parameters. With proxy
1753 authentication, it's the reverse, because proxy URLs are
1754 normally the "permanent" ones, so command-line args
1755 should take precedence. */
1756 if (opt.proxy_user && opt.proxy_passwd)
1758 proxy_user = opt.proxy_user;
1759 proxy_passwd = opt.proxy_passwd;
1763 proxy_user = proxy->user;
1764 proxy_passwd = proxy->passwd;
1766 /* #### This does not appear right. Can't the proxy request,
1767 say, `Digest' authentication? */
1768 if (proxy_user && proxy_passwd)
1769 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1771 /* If we're using a proxy, we will be connecting to the proxy
1775 /* Proxy authorization over SSL is handled below. */
1777 if (u->scheme != SCHEME_HTTPS)
1779 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1784 /* Establish the connection. */
1786 if (inhibit_keep_alive)
1790 /* Look for a persistent connection to target host, unless a
1791 proxy is used. The exception is when SSL is in use, in which
1792 case the proxy is nothing but a passthrough to the target
1793 host, registered as a connection to the latter. */
1794 struct url *relevant = conn;
1796 if (u->scheme == SCHEME_HTTPS)
1800 if (persistent_available_p (relevant->host, relevant->port,
1802 relevant->scheme == SCHEME_HTTPS,
1806 &host_lookup_failed))
1808 sock = pconn.socket;
1809 using_ssl = pconn.ssl;
1810 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1811 quotearg_style (escape_quoting_style, pconn.host),
1813 DEBUGP (("Reusing fd %d.\n", sock));
1814 if (pconn.authorized)
1815 /* If the connection is already authorized, the "Basic"
1816 authorization added by code above is unnecessary and
1818 request_remove_header (req, "Authorization");
1820 else if (host_lookup_failed)
1823 logprintf(LOG_NOTQUIET,
1824 _("%s: unable to resolve host address %s\n"),
1825 exec_name, quote (relevant->host));
1832 sock = connect_to_host (conn->host, conn->port);
1841 return (retryable_socket_connect_error (errno)
1842 ? CONERROR : CONIMPOSSIBLE);
1846 if (proxy && u->scheme == SCHEME_HTTPS)
1848 /* When requesting SSL URLs through proxies, use the
1849 CONNECT method to request passthrough. */
1850 struct request *connreq = request_new ();
1851 request_set_method (connreq, "CONNECT",
1852 aprintf ("%s:%d", u->host, u->port));
1853 SET_USER_AGENT (connreq);
1856 request_set_header (connreq, "Proxy-Authorization",
1857 proxyauth, rel_value);
1858 /* Now that PROXYAUTH is part of the CONNECT request,
1859 zero it out so we don't send proxy authorization with
1860 the regular request below. */
1863 /* Examples in rfc2817 use the Host header in CONNECT
1864 requests. I don't see how that gains anything, given
1865 that the contents of Host would be exactly the same as
1866 the contents of CONNECT. */
1868 write_error = request_send (connreq, sock);
1869 request_free (connreq);
1870 if (write_error < 0)
1872 CLOSE_INVALIDATE (sock);
1876 head = read_http_response_head (sock);
1879 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1881 CLOSE_INVALIDATE (sock);
1890 DEBUGP (("proxy responded with: [%s]\n", head));
1892 resp = resp_new (head);
1893 statcode = resp_status (resp, &message);
1896 char *tms = datetime_str (time (NULL));
1897 logprintf (LOG_VERBOSE, "%d\n", statcode);
1898 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1899 quotearg_style (escape_quoting_style,
1900 _("Malformed status line")));
1904 hs->message = xstrdup (message);
1907 if (statcode != 200)
1910 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1911 message ? quotearg_style (escape_quoting_style, message) : "?");
1912 xfree_null (message);
1915 xfree_null (message);
1917 /* SOCK is now *really* connected to u->host, so update CONN
1918 to reflect this. That way register_persistent will
1919 register SOCK as being connected to u->host:u->port. */
1923 if (conn->scheme == SCHEME_HTTPS)
1925 if (!ssl_connect_wget (sock))
1930 else if (!ssl_check_certificate (sock, u->host))
1933 return VERIFCERTERR;
1937 #endif /* HAVE_SSL */
1940 /* Send the request to server. */
1941 write_error = request_send (req, sock);
1943 if (write_error >= 0)
1947 DEBUGP (("[POST data: %s]\n", opt.post_data));
1948 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1950 else if (opt.post_file_name && post_data_size != 0)
1951 write_error = post_file (sock, opt.post_file_name, post_data_size);
1954 if (write_error < 0)
1956 CLOSE_INVALIDATE (sock);
1960 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1961 proxy ? "Proxy" : "HTTP");
1967 head = read_http_response_head (sock);
1972 logputs (LOG_NOTQUIET, _("No data received.\n"));
1973 CLOSE_INVALIDATE (sock);
1979 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1981 CLOSE_INVALIDATE (sock);
1986 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1988 resp = resp_new (head);
1990 /* Check for status line. */
1992 statcode = resp_status (resp, &message);
1995 char *tms = datetime_str (time (NULL));
1996 logprintf (LOG_VERBOSE, "%d\n", statcode);
1997 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1998 quotearg_style (escape_quoting_style,
1999 _("Malformed status line")));
2000 CLOSE_INVALIDATE (sock);
2005 if (H_10X (statcode))
2007 DEBUGP (("Ignoring response\n"));
2011 hs->message = xstrdup (message);
2012 if (!opt.server_response)
2013 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
2014 message ? quotearg_style (escape_quoting_style, message) : "");
2017 logprintf (LOG_VERBOSE, "\n");
2018 print_server_response (resp, " ");
2021 if (!opt.ignore_length
2022 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
2026 parsed = str_to_wgint (hdrval, NULL, 10);
2027 if (parsed == WGINT_MAX && errno == ERANGE)
2030 #### If Content-Length is out of range, it most likely
2031 means that the file is larger than 2G and that we're
2032 compiled without LFS. In that case we should probably
2033 refuse to even attempt to download the file. */
2036 else if (parsed < 0)
2038 /* Negative Content-Length; nonsensical, so we can't
2039 assume any information about the content to receive. */
2046 /* Check for keep-alive related responses. */
2047 if (!inhibit_keep_alive && contlen != -1)
2049 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2051 if (0 == strcasecmp (hdrval, "Close"))
2056 resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval));
2057 if (0 == strcasecmp (hdrval, "chunked"))
2058 chunked_transfer_encoding = true;
2060 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2064 const char *scbeg, *scend;
2065 /* The jar should have been created by now. */
2066 assert (wget_cookie_jar != NULL);
2068 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2069 &scbeg, &scend)) != -1;
2072 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2073 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2074 u->path, set_cookie);
2079 /* The server has promised that it will not close the connection
2080 when we're done. This means that we can register it. */
2081 register_persistent (conn->host, conn->port, sock, using_ssl);
2083 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2085 /* Authorization is required. */
2086 if (keep_alive && !head_only
2087 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2088 CLOSE_FINISH (sock);
2090 CLOSE_INVALIDATE (sock);
2091 pconn.authorized = false;
2092 if (!auth_finished && (user && passwd))
2094 /* IIS sends multiple copies of WWW-Authenticate, one with
2095 the value "negotiate", and other(s) with data. Loop over
2096 all the occurrences and pick the one we recognize. */
2098 const char *wabeg, *waend;
2099 char *www_authenticate = NULL;
2101 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2102 &wabeg, &waend)) != -1;
2104 if (known_authentication_scheme_p (wabeg, waend))
2106 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2110 if (!www_authenticate)
2112 /* If the authentication header is missing or
2113 unrecognized, there's no sense in retrying. */
2114 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2116 else if (!basic_auth_finished
2117 || !BEGINS_WITH (www_authenticate, "Basic"))
2120 pth = url_full_path (u);
2121 request_set_header (req, "Authorization",
2122 create_authorization_line (www_authenticate,
2124 request_method (req),
2128 if (BEGINS_WITH (www_authenticate, "NTLM"))
2130 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2132 /* Need to register this host as using basic auth,
2133 * so we automatically send creds next time. */
2134 register_basic_auth_host (u->host);
2137 xfree_null (message);
2140 goto retry_with_auth;
2144 /* We already did Basic auth, and it failed. Gotta
2148 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2150 xfree_null (message);
2155 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2157 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2159 pconn.authorized = true;
2162 /* Determine the local filename if needed. Notice that if -O is used
2163 * hstat.local_file is set by http_loop to the argument of -O. */
2164 if (!hs->local_file)
2166 /* Honor Content-Disposition whether possible. */
2167 if (!opt.content_disposition
2168 || !resp_header_copy (resp, "Content-Disposition",
2169 hdrval, sizeof (hdrval))
2170 || !parse_content_disposition (hdrval, &hs->local_file))
2172 /* The Content-Disposition header is missing or broken.
2173 * Choose unique file name according to given URL. */
2174 hs->local_file = url_file_name (u);
2178 /* TODO: perform this check only once. */
2179 if (!hs->existence_checked && file_exists_p (hs->local_file))
2181 if (opt.noclobber && !opt.output_document)
2183 /* If opt.noclobber is turned on and file already exists, do not
2184 retrieve the file. But if the output_document was given, then this
2185 test was already done and the file didn't exist. Hence the !opt.output_document */
2186 logprintf (LOG_VERBOSE, _("\
2187 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2188 /* If the file is there, we suppose it's retrieved OK. */
2191 /* #### Bogusness alert. */
2192 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2193 if (has_html_suffix_p (hs->local_file))
2197 xfree_null (message);
2198 return RETRUNNEEDED;
2200 else if (!ALLOW_CLOBBER)
2202 char *unique = unique_name (hs->local_file, true);
2203 if (unique != hs->local_file)
2204 xfree (hs->local_file);
2205 hs->local_file = unique;
2208 hs->existence_checked = true;
2210 /* Support timestamping */
2211 /* TODO: move this code out of gethttp. */
2212 if (opt.timestamping && !hs->timestamp_checked)
2214 size_t filename_len = strlen (hs->local_file);
2215 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2216 bool local_dot_orig_file_exists = false;
2217 char *local_filename = NULL;
2220 if (opt.backup_converted)
2221 /* If -K is specified, we'll act on the assumption that it was specified
2222 last time these files were downloaded as well, and instead of just
2223 comparing local file X against server file X, we'll compare local
2224 file X.orig (if extant, else X) against server file X. If -K
2225 _wasn't_ specified last time, or the server contains files called
2226 *.orig, -N will be back to not operating correctly with -k. */
2228 /* Would a single s[n]printf() call be faster? --dan
2230 Definitely not. sprintf() is horribly slow. It's a
2231 different question whether the difference between the two
2232 affects a program. Usually I'd say "no", but at one
2233 point I profiled Wget, and found that a measurable and
2234 non-negligible amount of time was lost calling sprintf()
2235 in url.c. Replacing sprintf with inline calls to
2236 strcpy() and number_to_string() made a difference.
2238 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2239 memcpy (filename_plus_orig_suffix + filename_len,
2240 ORIG_SFX, sizeof (ORIG_SFX));
2242 /* Try to stat() the .orig file. */
2243 if (stat (filename_plus_orig_suffix, &st) == 0)
2245 local_dot_orig_file_exists = true;
2246 local_filename = filename_plus_orig_suffix;
2250 if (!local_dot_orig_file_exists)
2251 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2252 if (stat (hs->local_file, &st) == 0)
2253 local_filename = hs->local_file;
2255 if (local_filename != NULL)
2256 /* There was a local file, so we'll check later to see if the version
2257 the server has is the same version we already have, allowing us to
2260 hs->orig_file_name = xstrdup (local_filename);
2261 hs->orig_file_size = st.st_size;
2262 hs->orig_file_tstamp = st.st_mtime;
2264 /* Modification time granularity is 2 seconds for Windows, so
2265 increase local time by 1 second for later comparison. */
2266 ++hs->orig_file_tstamp;
2273 hs->statcode = statcode;
2275 hs->error = xstrdup (_("Malformed status line"));
2277 hs->error = xstrdup (_("(no description)"));
2279 hs->error = xstrdup (message);
2280 xfree_null (message);
2282 type = resp_header_strdup (resp, "Content-Type");
2285 char *tmp = strchr (type, ';');
2288 /* sXXXav: only needed if IRI support is enabled */
2289 char *tmp2 = tmp + 1;
2291 while (tmp > type && c_isspace (tmp[-1]))
2295 /* Try to get remote encoding if needed */
2296 if (opt.enable_iri && !opt.encoding_remote)
2298 tmp = parse_charset (tmp2);
2300 set_content_encoding (iri, tmp);
2304 hs->newloc = resp_header_strdup (resp, "Location");
2305 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2307 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2309 wgint first_byte_pos, last_byte_pos, entity_length;
2310 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2313 contrange = first_byte_pos;
2314 contlen = last_byte_pos - first_byte_pos + 1;
2319 /* 20x responses are counted among successful by default. */
2320 if (H_20X (statcode))
2323 /* Return if redirected. */
2324 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2326 /* RFC2068 says that in case of the 300 (multiple choices)
2327 response, the server can output a preferred URL through
2328 `Location' header; otherwise, the request should be treated
2329 like GET. So, if the location is set, it will be a
2330 redirection; otherwise, just proceed normally. */
2331 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2335 logprintf (LOG_VERBOSE,
2336 _("Location: %s%s\n"),
2337 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2338 hs->newloc ? _(" [following]") : "");
2339 if (keep_alive && !head_only
2340 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2341 CLOSE_FINISH (sock);
2343 CLOSE_INVALIDATE (sock);
2350 /* If content-type is not given, assume text/html. This is because
2351 of the multitude of broken CGI's that "forget" to generate the
2354 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2355 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2361 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2366 if (opt.adjust_extension)
2369 /* -E / --adjust-extension / adjust_extension = on was specified,
2370 and this is a text/html file. If some case-insensitive
2371 variation on ".htm[l]" isn't already the file's suffix,
2374 ensure_extension (hs, ".html", dt);
2376 else if (*dt & TEXTCSS)
2378 ensure_extension (hs, ".css", dt);
2382 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2383 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK
2384 && contrange == 0 && contlen >= 0 && hs->restval >= contlen))
2386 /* If `-c' is in use and the file has been fully downloaded (or
2387 the remote file has shrunk), Wget effectively requests bytes
2388 after the end of file and the server response with 416
2389 (or 200 with a <= Content-Length. */
2390 logputs (LOG_VERBOSE, _("\
2391 \n The file is already fully retrieved; nothing to do.\n\n"));
2392 /* In case the caller inspects. */
2395 /* Mark as successfully retrieved. */
2398 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2399 might be more bytes in the body. */
2401 return RETRUNNEEDED;
2403 if ((contrange != 0 && contrange != hs->restval)
2404 || (H_PARTIAL (statcode) && !contrange))
2406 /* The Range request was somehow misunderstood by the server.
2409 CLOSE_INVALIDATE (sock);
2416 hs->contlen = contlen + contrange;
2422 /* No need to print this output if the body won't be
2423 downloaded at all, or if the original server response is
2425 logputs (LOG_VERBOSE, _("Length: "));
2428 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2429 if (contlen + contrange >= 1024)
2430 logprintf (LOG_VERBOSE, " (%s)",
2431 human_readable (contlen + contrange));
2434 if (contlen >= 1024)
2435 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2436 number_to_static_string (contlen),
2437 human_readable (contlen));
2439 logprintf (LOG_VERBOSE, _(", %s remaining"),
2440 number_to_static_string (contlen));
2444 logputs (LOG_VERBOSE,
2445 opt.ignore_length ? _("ignored") : _("unspecified"));
2447 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2449 logputs (LOG_VERBOSE, "\n");
2453 type = NULL; /* We don't need it any more. */
2455 /* Return if we have no intention of further downloading. */
2456 if (!(*dt & RETROKF) || head_only)
2458 /* In case the caller cares to look... */
2463 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2464 servers not to send body in response to a HEAD request, and
2465 those that do will likely be caught by test_socket_open.
2466 If not, they can be worked around using
2467 `--no-http-keep-alive'. */
2468 CLOSE_FINISH (sock);
2470 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2471 /* Successfully skipped the body; also keep using the socket. */
2472 CLOSE_FINISH (sock);
2474 CLOSE_INVALIDATE (sock);
2476 return RETRFINISHED;
2480 For VMS, define common fopen() optional arguments.
2483 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2484 # define FOPEN_BIN_FLAG 3
2485 #else /* def __VMS */
2486 # define FOPEN_BIN_FLAG true
2487 #endif /* def __VMS [else] */
2489 /* Open the local file. */
2492 mkalldirs (hs->local_file);
2494 rotate_backups (hs->local_file);
2501 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2502 #else /* def __VMS */
2503 fp = fopen (hs->local_file, "ab");
2504 #endif /* def __VMS [else] */
2506 else if (ALLOW_CLOBBER)
2512 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2513 #else /* def __VMS */
2514 fp = fopen (hs->local_file, "wb");
2515 #endif /* def __VMS [else] */
2519 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2520 if (!fp && errno == EEXIST)
2522 /* We cannot just invent a new name and use it (which is
2523 what functions like unique_create typically do)
2524 because we told the user we'd use this name.
2525 Instead, return and retry the download. */
2526 logprintf (LOG_NOTQUIET,
2527 _("%s has sprung into existence.\n"),
2529 CLOSE_INVALIDATE (sock);
2531 return FOPEN_EXCL_ERR;
2536 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2537 CLOSE_INVALIDATE (sock);
2545 /* Print fetch message, if opt.verbose. */
2548 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2549 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2552 /* This confuses the timestamping code that checks for file size.
2553 #### The timestamping code should be smarter about file size. */
2554 if (opt.save_headers && hs->restval == 0)
2555 fwrite (head, 1, strlen (head), fp);
2557 /* Now we no longer need to store the response header. */
2560 /* Download the request body. */
2563 /* If content-length is present, read that much; otherwise, read
2564 until EOF. The HTTP spec doesn't require the server to
2565 actually close the connection when it's done sending data. */
2566 flags |= rb_read_exactly;
2567 if (hs->restval > 0 && contrange == 0)
2568 /* If the server ignored our range request, instruct fd_read_body
2569 to skip the first RESTVAL bytes of body. */
2570 flags |= rb_skip_startpos;
2572 if (chunked_transfer_encoding)
2573 flags |= rb_chunked_transfer_encoding;
2575 hs->len = hs->restval;
2577 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2578 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2582 CLOSE_FINISH (sock);
2586 hs->rderrmsg = xstrdup (fd_errstr (sock));
2587 CLOSE_INVALIDATE (sock);
2594 return RETRFINISHED;
2597 /* The genuine HTTP loop! This is the part where the retrieval is
2598 retried, and retried, and retried, and... */
2600 http_loop (struct url *u, struct url *original_url, char **newloc,
2601 char **local_file, const char *referer, int *dt, struct url *proxy,
2605 bool got_head = false; /* used for time-stamping and filename detection */
2606 bool time_came_from_head = false;
2607 bool got_name = false;
2610 uerr_t err, ret = TRYLIMEXC;
2611 time_t tmr = -1; /* remote time-stamp */
2612 struct http_stat hstat; /* HTTP status */
2614 bool send_head_first = true;
2616 bool force_full_retrieve = false;
2618 /* Assert that no value for *LOCAL_FILE was passed. */
2619 assert (local_file == NULL || *local_file == NULL);
2621 /* Set LOCAL_FILE parameter. */
2622 if (local_file && opt.output_document)
2623 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2625 /* Reset NEWLOC parameter. */
2628 /* This used to be done in main(), but it's a better idea to do it
2629 here so that we don't go through the hoops if we're just using
2634 /* Warn on (likely bogus) wildcard usage in HTTP. */
2635 if (opt.ftp_glob && has_wildcards_p (u->path))
2636 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2638 /* Setup hstat struct. */
2640 hstat.referer = referer;
2642 if (opt.output_document)
2644 hstat.local_file = xstrdup (opt.output_document);
2647 else if (!opt.content_disposition)
2650 url_file_name (opt.trustservernames ? u : original_url);
2654 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2655 * screaming for some refactoring. */
2656 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2658 /* If opt.noclobber is turned on and file already exists, do not
2659 retrieve the file. But if the output_document was given, then this
2660 test was already done and the file didn't exist. Hence the !opt.output_document */
2661 logprintf (LOG_VERBOSE, _("\
2662 File %s already there; not retrieving.\n\n"),
2663 quote (hstat.local_file));
2664 /* If the file is there, we suppose it's retrieved OK. */
2667 /* #### Bogusness alert. */
2668 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2669 if (has_html_suffix_p (hstat.local_file))
2676 /* Reset the counter. */
2679 /* Reset the document type. */
2682 /* Skip preliminary HEAD request if we're not in spider mode. */
2684 send_head_first = false;
2686 /* Send preliminary HEAD request if -N is given and we have an existing
2687 * destination file. */
2688 file_name = url_file_name (opt.trustservernames ? u : original_url);
2689 if (opt.timestamping && (file_exists_p (file_name)
2690 || opt.content_disposition))
2691 send_head_first = true;
2697 /* Increment the pass counter. */
2699 sleep_between_retrievals (count);
2701 /* Get the current time string. */
2702 tms = datetime_str (time (NULL));
2704 if (opt.spider && !got_head)
2705 logprintf (LOG_VERBOSE, _("\
2706 Spider mode enabled. Check if remote file exists.\n"));
2708 /* Print fetch message, if opt.verbose. */
2711 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2716 sprintf (tmp, _("(try:%2d)"), count);
2717 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2722 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2727 ws_changetitle (hurl);
2732 /* Default document type is empty. However, if spider mode is
2733 on or time-stamping is employed, HEAD_ONLY commands is
2734 encoded within *dt. */
2735 if (send_head_first && !got_head)
2740 /* Decide whether or not to restart. */
2741 if (force_full_retrieve)
2742 hstat.restval = hstat.len;
2743 else if (opt.always_rest
2745 && stat (hstat.local_file, &st) == 0
2746 && S_ISREG (st.st_mode))
2747 /* When -c is used, continue from on-disk size. (Can't use
2748 hstat.len even if count>1 because we don't want a failed
2749 first attempt to clobber existing data.) */
2750 hstat.restval = st.st_size;
2752 /* otherwise, continue where the previous try left off */
2753 hstat.restval = hstat.len;
2757 /* Decide whether to send the no-cache directive. We send it in
2759 a) we're using a proxy, and we're past our first retrieval.
2760 Some proxies are notorious for caching incomplete data, so
2761 we require a fresh get.
2762 b) caching is explicitly inhibited. */
2763 if ((proxy && count > 1) /* a */
2764 || !opt.allow_cache) /* b */
2765 *dt |= SEND_NOCACHE;
2767 *dt &= ~SEND_NOCACHE;
2769 /* Try fetching the document, or at least its head. */
2770 err = gethttp (u, &hstat, dt, proxy, iri);
2773 tms = datetime_str (time (NULL));
2775 /* Get the new location (with or without the redirection). */
2777 *newloc = xstrdup (hstat.newloc);
2781 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2782 case CONERROR: case READERR: case WRITEFAILED:
2783 case RANGEERR: case FOPEN_EXCL_ERR:
2784 /* Non-fatal errors continue executing the loop, which will
2785 bring them to "while" statement at the end, to judge
2786 whether the number of tries was exceeded. */
2787 printwhat (count, opt.ntry);
2789 case FWRITEERR: case FOPENERR:
2790 /* Another fatal error. */
2791 logputs (LOG_VERBOSE, "\n");
2792 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2793 quote (hstat.local_file), strerror (errno));
2794 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2795 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2796 /* Fatal errors just return from the function. */
2800 /* Another fatal error. */
2801 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2805 /* Return the new location to the caller. */
2808 logprintf (LOG_NOTQUIET,
2809 _("ERROR: Redirection (%d) without location.\n"),
2819 /* The file was already fully retrieved. */
2823 /* Deal with you later. */
2826 /* All possibilities should have been exhausted. */
2830 if (!(*dt & RETROKF))
2835 /* #### Ugly ugly ugly! */
2836 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2837 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2840 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2842 && (hstat.statcode == 500 || hstat.statcode == 501))
2847 /* Maybe we should always keep track of broken links, not just in
2849 * Don't log error if it was UTF-8 encoded because we will try
2850 * once unencoded. */
2851 else if (opt.spider && !iri->utf8_encode)
2853 /* #### Again: ugly ugly ugly! */
2855 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2856 nonexisting_url (hurl);
2857 logprintf (LOG_NOTQUIET, _("\
2858 Remote file does not exist -- broken link!!!\n"));
2862 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2863 tms, hstat.statcode,
2864 quotearg_style (escape_quoting_style, hstat.error));
2866 logputs (LOG_VERBOSE, "\n");
2872 /* Did we get the time-stamp? */
2875 got_head = true; /* no more time-stamping */
2877 if (opt.timestamping && !hstat.remote_time)
2879 logputs (LOG_NOTQUIET, _("\
2880 Last-modified header missing -- time-stamps turned off.\n"));
2882 else if (hstat.remote_time)
2884 /* Convert the date-string into struct tm. */
2885 tmr = http_atotm (hstat.remote_time);
2886 if (tmr == (time_t) (-1))
2887 logputs (LOG_VERBOSE, _("\
2888 Last-modified header invalid -- time-stamp ignored.\n"));
2889 if (*dt & HEAD_ONLY)
2890 time_came_from_head = true;
2893 if (send_head_first)
2895 /* The time-stamping section. */
2896 if (opt.timestamping)
2898 if (hstat.orig_file_name) /* Perform the following
2899 checks only if the file
2901 download already exists. */
2903 if (hstat.remote_time &&
2904 tmr != (time_t) (-1))
2906 /* Now time-stamping can be used validly.
2907 Time-stamping means that if the sizes of
2908 the local and remote file match, and local
2909 file is newer than the remote file, it will
2910 not be retrieved. Otherwise, the normal
2911 download procedure is resumed. */
2912 if (hstat.orig_file_tstamp >= tmr)
2914 if (hstat.contlen == -1
2915 || hstat.orig_file_size == hstat.contlen)
2917 logprintf (LOG_VERBOSE, _("\
2918 Server file no newer than local file %s -- not retrieving.\n\n"),
2919 quote (hstat.orig_file_name));
2925 logprintf (LOG_VERBOSE, _("\
2926 The sizes do not match (local %s) -- retrieving.\n"),
2927 number_to_static_string (hstat.orig_file_size));
2932 force_full_retrieve = true;
2933 logputs (LOG_VERBOSE,
2934 _("Remote file is newer, retrieving.\n"));
2937 logputs (LOG_VERBOSE, "\n");
2941 /* free_hstat (&hstat); */
2942 hstat.timestamp_checked = true;
2947 bool finished = true;
2952 logputs (LOG_VERBOSE, _("\
2953 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2958 logprintf (LOG_VERBOSE, _("\
2959 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2960 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2967 logprintf (LOG_VERBOSE, _("\
2968 Remote file exists and could contain further links,\n\
2969 but recursion is disabled -- not retrieving.\n\n"));
2973 logprintf (LOG_VERBOSE, _("\
2974 Remote file exists.\n\n"));
2976 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2981 logprintf (LOG_NONVERBOSE,
2982 _("%s URL: %s %2d %s\n"),
2983 tms, u->url, hstat.statcode,
2984 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2991 count = 0; /* the retrieve count for HEAD is reset */
2993 } /* send_head_first */
2996 if (opt.useservertimestamps
2997 && (tmr != (time_t) (-1))
2998 && ((hstat.len == hstat.contlen) ||
2999 ((hstat.res == 0) && (hstat.contlen == -1))))
3001 const char *fl = NULL;
3002 set_local_file (&fl, hstat.local_file);
3006 /* Reparse time header, in case it's changed. */
3007 if (time_came_from_head
3008 && hstat.remote_time && hstat.remote_time[0])
3010 newtmr = http_atotm (hstat.remote_time);
3011 if (newtmr != (time_t)-1)
3017 /* End of time-stamping section. */
3019 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3020 total_download_time += hstat.dltime;
3022 if (hstat.len == hstat.contlen)
3026 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3028 logprintf (LOG_VERBOSE,
3030 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3031 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3033 write_to_stdout ? "" : quote (hstat.local_file),
3034 number_to_static_string (hstat.len),
3035 number_to_static_string (hstat.contlen));
3036 logprintf (LOG_NONVERBOSE,
3037 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3039 number_to_static_string (hstat.len),
3040 number_to_static_string (hstat.contlen),
3041 hstat.local_file, count);
3044 total_downloaded_bytes += hstat.rd_size;
3046 /* Remember that we downloaded the file for later ".orig" code. */
3047 if (*dt & ADDED_HTML_EXTENSION)
3048 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3050 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3055 else if (hstat.res == 0) /* No read error */
3057 if (hstat.contlen == -1) /* We don't know how much we were supposed
3058 to get, so assume we succeeded. */
3062 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3064 logprintf (LOG_VERBOSE,
3066 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3067 : _("%s (%s) - %s saved [%s]\n\n"),
3069 write_to_stdout ? "" : quote (hstat.local_file),
3070 number_to_static_string (hstat.len));
3071 logprintf (LOG_NONVERBOSE,
3072 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3073 tms, u->url, number_to_static_string (hstat.len),
3074 hstat.local_file, count);
3077 total_downloaded_bytes += hstat.rd_size;
3079 /* Remember that we downloaded the file for later ".orig" code. */
3080 if (*dt & ADDED_HTML_EXTENSION)
3081 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3083 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3088 else if (hstat.len < hstat.contlen) /* meaning we lost the
3089 connection too soon */
3091 logprintf (LOG_VERBOSE,
3092 _("%s (%s) - Connection closed at byte %s. "),
3093 tms, tmrate, number_to_static_string (hstat.len));
3094 printwhat (count, opt.ntry);
3097 else if (hstat.len != hstat.restval)
3098 /* Getting here would mean reading more data than
3099 requested with content-length, which we never do. */
3103 /* Getting here probably means that the content-length was
3104 * _less_ than the original, local size. We should probably
3105 * truncate or re-read, or something. FIXME */
3110 else /* from now on hstat.res can only be -1 */
3112 if (hstat.contlen == -1)
3114 logprintf (LOG_VERBOSE,
3115 _("%s (%s) - Read error at byte %s (%s)."),
3116 tms, tmrate, number_to_static_string (hstat.len),
3118 printwhat (count, opt.ntry);
3121 else /* hstat.res == -1 and contlen is given */
3123 logprintf (LOG_VERBOSE,
3124 _("%s (%s) - Read error at byte %s/%s (%s). "),
3126 number_to_static_string (hstat.len),
3127 number_to_static_string (hstat.contlen),
3129 printwhat (count, opt.ntry);
3135 while (!opt.ntry || (count < opt.ntry));
3139 *local_file = xstrdup (hstat.local_file);
3140 free_hstat (&hstat);
3145 /* Check whether the result of strptime() indicates success.
3146 strptime() returns the pointer to how far it got to in the string.
3147 The processing has been successful if the string is at `GMT' or
3148 `+X', or at the end of the string.
3150 In extended regexp parlance, the function returns 1 if P matches
3151 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3152 can return) is considered a failure and 0 is returned. */
3154 check_end (const char *p)
3158 while (c_isspace (*p))
3161 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3162 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3168 /* Convert the textual specification of time in TIME_STRING to the
3169 number of seconds since the Epoch.
3171 TIME_STRING can be in any of the three formats RFC2616 allows the
3172 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3173 as well as the time format used in the Set-Cookie header.
3174 Timezones are ignored, and should be GMT.
3176 Return the computed time_t representation, or -1 if the conversion
3179 This function uses strptime with various string formats for parsing
3180 TIME_STRING. This results in a parser that is not as lenient in
3181 interpreting TIME_STRING as I would like it to be. Being based on
3182 strptime, it always allows shortened months, one-digit days, etc.,
3183 but due to the multitude of formats in which time can be
3184 represented, an ideal HTTP time parser would be even more
3185 forgiving. It should completely ignore things like week days and
3186 concentrate only on the various forms of representing years,
3187 months, days, hours, minutes, and seconds. For example, it would
3188 be nice if it accepted ISO 8601 out of the box.
3190 I've investigated free and PD code for this purpose, but none was
3191 usable. getdate was big and unwieldy, and had potential copyright
3192 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3193 distributed with phttpd, is excellent, but we cannot use it because
3194 it is not assigned to the FSF. So I stuck it with strptime. */
3197 http_atotm (const char *time_string)
3199 /* NOTE: Solaris strptime man page claims that %n and %t match white
3200 space, but that's not universally available. Instead, we simply
3201 use ` ' to mean "skip all WS", which works under all strptime
3202 implementations I've tested. */
3204 static const char *time_formats[] = {
3205 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3206 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3207 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3208 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3209 (used in Set-Cookie, defined in the
3210 Netscape cookie specification.) */
3212 const char *oldlocale;
3213 char savedlocale[256];
3215 time_t ret = (time_t) -1;
3217 /* Solaris strptime fails to recognize English month names in
3218 non-English locales, which we work around by temporarily setting
3219 locale to C before invoking strptime. */
3220 oldlocale = setlocale (LC_TIME, NULL);
3223 size_t l = strlen (oldlocale) + 1;
3224 if (l >= sizeof savedlocale)
3225 savedlocale[0] = '\0';
3227 memcpy (savedlocale, oldlocale, l);
3229 else savedlocale[0] = '\0';
3231 setlocale (LC_TIME, "C");
3233 for (i = 0; i < countof (time_formats); i++)
3237 /* Some versions of strptime use the existing contents of struct
3238 tm to recalculate the date according to format. Zero it out
3239 to prevent stack garbage from influencing strptime. */
3242 if (check_end (strptime (time_string, time_formats[i], &t)))
3249 /* Restore the previous locale. */
3251 setlocale (LC_TIME, savedlocale);
3256 /* Authorization support: We support three authorization schemes:
3258 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3260 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3261 consisting of answering to the server's challenge with the proper
3264 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3265 Stenberg for libcurl. Like digest, NTLM is based on a
3266 challenge-response mechanism, but unlike digest, it is non-standard
3267 (authenticates TCP connections rather than requests), undocumented
3268 and Microsoft-specific. */
3270 /* Create the authentication header contents for the `Basic' scheme.
3271 This is done by encoding the string "USER:PASS" to base64 and
3272 prepending the string "Basic " in front of it. */
3275 basic_authentication_encode (const char *user, const char *passwd)
3278 int len1 = strlen (user) + 1 + strlen (passwd);
3280 t1 = (char *)alloca (len1 + 1);
3281 sprintf (t1, "%s:%s", user, passwd);
3283 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3284 base64_encode (t1, len1, t2);
3286 return concat_strings ("Basic ", t2, (char *) 0);
3289 #define SKIP_WS(x) do { \
3290 while (c_isspace (*(x))) \
3294 #ifdef ENABLE_DIGEST
3295 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3296 an array of 16 bytes containing the hash keys, and BUF should be a
3297 buffer of 33 writable characters (32 for hex digits plus one for
3298 zero termination). */
3300 dump_hash (char *buf, const unsigned char *hash)
3304 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3306 *buf++ = XNUM_TO_digit (*hash >> 4);
3307 *buf++ = XNUM_TO_digit (*hash & 0xf);
3312 /* Take the line apart to find the challenge, and compose a digest
3313 authorization header. See RFC2069 section 2.1.2. */
3315 digest_authentication_encode (const char *au, const char *user,
3316 const char *passwd, const char *method,
3319 static char *realm, *opaque, *nonce;
3324 { "realm", &realm },
3325 { "opaque", &opaque },
3329 param_token name, value;
3331 realm = opaque = nonce = NULL;
3333 au += 6; /* skip over `Digest' */
3334 while (extract_param (&au, &name, &value, ','))
3337 size_t namelen = name.e - name.b;
3338 for (i = 0; i < countof (options); i++)
3339 if (namelen == strlen (options[i].name)
3340 && 0 == strncmp (name.b, options[i].name,
3343 *options[i].variable = strdupdelim (value.b, value.e);
3347 if (!realm || !nonce || !user || !passwd || !path || !method)
3350 xfree_null (opaque);
3355 /* Calculate the digest value. */
3358 unsigned char hash[MD5_DIGEST_SIZE];
3359 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3360 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3362 /* A1BUF = H(user ":" realm ":" password) */
3363 md5_init_ctx (&ctx);
3364 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3365 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3366 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3367 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3368 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3369 md5_finish_ctx (&ctx, hash);
3370 dump_hash (a1buf, hash);
3372 /* A2BUF = H(method ":" path) */
3373 md5_init_ctx (&ctx);
3374 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3375 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3376 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3377 md5_finish_ctx (&ctx, hash);
3378 dump_hash (a2buf, hash);
3380 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3381 md5_init_ctx (&ctx);
3382 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3383 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3384 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3385 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3386 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3387 md5_finish_ctx (&ctx, hash);
3388 dump_hash (response_digest, hash);
3390 res = xmalloc (strlen (user)
3395 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3396 + (opaque ? strlen (opaque) : 0)
3398 sprintf (res, "Digest \
3399 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3400 user, realm, nonce, path, response_digest);
3403 char *p = res + strlen (res);
3404 strcat (p, ", opaque=\"");
3411 #endif /* ENABLE_DIGEST */
3413 /* Computing the size of a string literal must take into account that
3414 value returned by sizeof includes the terminating \0. */
3415 #define STRSIZE(literal) (sizeof (literal) - 1)
3417 /* Whether chars in [b, e) begin with the literal string provided as
3418 first argument and are followed by whitespace or terminating \0.
3419 The comparison is case-insensitive. */
3420 #define STARTS(literal, b, e) \
3422 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3423 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3424 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3425 || c_isspace (b[STRSIZE (literal)])))
3428 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3430 return STARTS ("Basic", hdrbeg, hdrend)
3431 #ifdef ENABLE_DIGEST
3432 || STARTS ("Digest", hdrbeg, hdrend)
3435 || STARTS ("NTLM", hdrbeg, hdrend)
3442 /* Create the HTTP authorization request header. When the
3443 `WWW-Authenticate' response header is seen, according to the
3444 authorization scheme specified in that header (`Basic' and `Digest'
3445 are supported by the current implementation), produce an
3446 appropriate HTTP authorization request header. */
3448 create_authorization_line (const char *au, const char *user,
3449 const char *passwd, const char *method,
3450 const char *path, bool *finished)
3452 /* We are called only with known schemes, so we can dispatch on the
3454 switch (c_toupper (*au))
3456 case 'B': /* Basic */
3458 return basic_authentication_encode (user, passwd);
3459 #ifdef ENABLE_DIGEST
3460 case 'D': /* Digest */
3462 return digest_authentication_encode (au, user, passwd, method, path);
3465 case 'N': /* NTLM */
3466 if (!ntlm_input (&pconn.ntlm, au))
3471 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3474 /* We shouldn't get here -- this function should be only called
3475 with values approved by known_authentication_scheme_p. */
3483 if (!wget_cookie_jar)
3484 wget_cookie_jar = cookie_jar_new ();
3485 if (opt.cookies_input && !cookies_loaded_p)
3487 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3488 cookies_loaded_p = true;
3495 if (wget_cookie_jar)
3496 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3502 xfree_null (pconn.host);
3503 if (wget_cookie_jar)
3504 cookie_jar_delete (wget_cookie_jar);
3508 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3510 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3512 int len = strlen (ext);
3515 strncpy (shortext, ext, len - 1);
3516 shortext[len - 2] = '\0';
3519 if (last_period_in_local_filename == NULL
3520 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3521 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3523 int local_filename_len = strlen (hs->local_file);
3524 /* Resize the local file, allowing for ".html" preceded by
3525 optional ".NUMBER". */
3526 hs->local_file = xrealloc (hs->local_file,
3527 local_filename_len + 24 + len);
3528 strcpy (hs->local_file + local_filename_len, ext);
3529 /* If clobbering is not allowed and the file, as named,
3530 exists, tack on ".NUMBER.html" instead. */
3531 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3535 sprintf (hs->local_file + local_filename_len,
3536 ".%d%s", ext_num++, ext);
3537 while (file_exists_p (hs->local_file));
3539 *dt |= ADDED_HTML_EXTENSION;
3547 test_parse_content_disposition()
3552 char *opt_dir_prefix;
3556 { "filename=\"file.ext\"", NULL, "file.ext", true },
3557 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3558 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3559 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3560 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3561 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3562 { "attachment", NULL, NULL, false },
3563 { "attachment", "somedir", NULL, false },
3564 { "attachement; filename*=UTF-8'en-US'hello.txt", NULL, "hello.txt", true },
3565 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", NULL, "helloworld.txt", true },
3568 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3573 opt.dir_prefix = test_array[i].opt_dir_prefix;
3574 res = parse_content_disposition (test_array[i].hdrval, &filename);
3576 mu_assert ("test_parse_content_disposition: wrong result",
3577 res == test_array[i].result
3579 || 0 == strcmp (test_array[i].filename, filename)));
3585 #endif /* TESTING */
3588 * vim: et sts=2 sw=2 cino+={s