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.
1578 RFC 2068 requests that 1.0 clients not send keep-alive requests
1579 to proxies. This is because many 1.0 proxies do not interpret
1580 the Connection header and transfer it to the remote server,
1581 causing it to not close the connection and leave both the proxy
1582 and the client hanging. */
1583 bool inhibit_keep_alive =
1584 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1586 /* Headers sent when using POST. */
1587 wgint post_data_size = 0;
1589 bool host_lookup_failed = false;
1592 if (u->scheme == SCHEME_HTTPS)
1594 /* Initialize the SSL context. After this has once been done,
1595 it becomes a no-op. */
1598 scheme_disable (SCHEME_HTTPS);
1599 logprintf (LOG_NOTQUIET,
1600 _("Disabling SSL due to encountered errors.\n"));
1601 return SSLINITFAILED;
1604 #endif /* HAVE_SSL */
1606 /* Initialize certain elements of struct http_stat. */
1610 hs->rderrmsg = NULL;
1612 hs->remote_time = NULL;
1618 /* Prepare the request to send. */
1620 req = request_new ();
1623 const char *meth = "GET";
1626 else if (opt.post_file_name || opt.post_data)
1628 /* Use the full path, i.e. one that includes the leading slash and
1629 the query string. E.g. if u->path is "foo/bar" and u->query is
1630 "param=value", full_path will be "/foo/bar?param=value". */
1633 /* When using SSL over proxy, CONNECT establishes a direct
1634 connection to the HTTPS server. Therefore use the same
1635 argument as when talking to the server directly. */
1636 && u->scheme != SCHEME_HTTPS
1639 meth_arg = xstrdup (u->url);
1641 meth_arg = url_full_path (u);
1642 request_set_method (req, meth, meth_arg);
1645 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1646 if (*dt & SEND_NOCACHE)
1647 request_set_header (req, "Pragma", "no-cache", rel_none);
1648 if (hs->restval && !opt.timestamping)
1649 request_set_header (req, "Range",
1650 aprintf ("bytes=%s-",
1651 number_to_static_string (hs->restval)),
1653 SET_USER_AGENT (req);
1654 request_set_header (req, "Accept", "*/*", rel_none);
1656 /* Find the username and password for authentication. */
1659 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1660 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1661 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1663 /* We only do "site-wide" authentication with "global" user/password
1664 * values unless --auth-no-challange has been requested; URL user/password
1665 * info overrides. */
1666 if (user && passwd && (!u->user || opt.auth_without_challenge))
1668 /* If this is a host for which we've already received a Basic
1669 * challenge, we'll go ahead and send Basic authentication creds. */
1670 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1673 /* Generate the Host header, HOST:PORT. Take into account that:
1675 - Broken server-side software often doesn't recognize the PORT
1676 argument, so we must generate "Host: www.server.com" instead of
1677 "Host: www.server.com:80" (and likewise for https port).
1679 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1680 becomes ambiguous and needs to be rewritten as "Host:
1681 [3ffe:8100:200:2::2]:1234". */
1683 /* Formats arranged for hfmt[add_port][add_squares]. */
1684 static const char *hfmt[][2] = {
1685 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1687 int add_port = u->port != scheme_default_port (u->scheme);
1688 int add_squares = strchr (u->host, ':') != NULL;
1689 request_set_header (req, "Host",
1690 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1694 if (!inhibit_keep_alive)
1695 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1697 if (opt.post_data || opt.post_file_name)
1699 request_set_header (req, "Content-Type",
1700 "application/x-www-form-urlencoded", rel_none);
1702 post_data_size = strlen (opt.post_data);
1705 post_data_size = file_size (opt.post_file_name);
1706 if (post_data_size == -1)
1708 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1709 quote (opt.post_file_name), strerror (errno));
1713 request_set_header (req, "Content-Length",
1714 xstrdup (number_to_static_string (post_data_size)),
1719 /* We need to come back here when the initial attempt to retrieve
1720 without authorization header fails. (Expected to happen at least
1721 for the Digest authorization scheme.) */
1724 request_set_header (req, "Cookie",
1725 cookie_header (wget_cookie_jar,
1726 u->host, u->port, u->path,
1728 u->scheme == SCHEME_HTTPS
1735 /* Add the user headers. */
1736 if (opt.user_headers)
1739 for (i = 0; opt.user_headers[i]; i++)
1740 request_set_user_header (req, opt.user_headers[i]);
1746 char *proxy_user, *proxy_passwd;
1747 /* For normal username and password, URL components override
1748 command-line/wgetrc parameters. With proxy
1749 authentication, it's the reverse, because proxy URLs are
1750 normally the "permanent" ones, so command-line args
1751 should take precedence. */
1752 if (opt.proxy_user && opt.proxy_passwd)
1754 proxy_user = opt.proxy_user;
1755 proxy_passwd = opt.proxy_passwd;
1759 proxy_user = proxy->user;
1760 proxy_passwd = proxy->passwd;
1762 /* #### This does not appear right. Can't the proxy request,
1763 say, `Digest' authentication? */
1764 if (proxy_user && proxy_passwd)
1765 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1767 /* If we're using a proxy, we will be connecting to the proxy
1771 /* Proxy authorization over SSL is handled below. */
1773 if (u->scheme != SCHEME_HTTPS)
1775 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1780 /* Establish the connection. */
1782 if (inhibit_keep_alive)
1786 /* Look for a persistent connection to target host, unless a
1787 proxy is used. The exception is when SSL is in use, in which
1788 case the proxy is nothing but a passthrough to the target
1789 host, registered as a connection to the latter. */
1790 struct url *relevant = conn;
1792 if (u->scheme == SCHEME_HTTPS)
1796 if (persistent_available_p (relevant->host, relevant->port,
1798 relevant->scheme == SCHEME_HTTPS,
1802 &host_lookup_failed))
1804 sock = pconn.socket;
1805 using_ssl = pconn.ssl;
1806 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1807 quotearg_style (escape_quoting_style, pconn.host),
1809 DEBUGP (("Reusing fd %d.\n", sock));
1810 if (pconn.authorized)
1811 /* If the connection is already authorized, the "Basic"
1812 authorization added by code above is unnecessary and
1814 request_remove_header (req, "Authorization");
1816 else if (host_lookup_failed)
1819 logprintf(LOG_NOTQUIET,
1820 _("%s: unable to resolve host address %s\n"),
1821 exec_name, quote (relevant->host));
1828 sock = connect_to_host (conn->host, conn->port);
1837 return (retryable_socket_connect_error (errno)
1838 ? CONERROR : CONIMPOSSIBLE);
1842 if (proxy && u->scheme == SCHEME_HTTPS)
1844 /* When requesting SSL URLs through proxies, use the
1845 CONNECT method to request passthrough. */
1846 struct request *connreq = request_new ();
1847 request_set_method (connreq, "CONNECT",
1848 aprintf ("%s:%d", u->host, u->port));
1849 SET_USER_AGENT (connreq);
1852 request_set_header (connreq, "Proxy-Authorization",
1853 proxyauth, rel_value);
1854 /* Now that PROXYAUTH is part of the CONNECT request,
1855 zero it out so we don't send proxy authorization with
1856 the regular request below. */
1859 /* Examples in rfc2817 use the Host header in CONNECT
1860 requests. I don't see how that gains anything, given
1861 that the contents of Host would be exactly the same as
1862 the contents of CONNECT. */
1864 write_error = request_send (connreq, sock);
1865 request_free (connreq);
1866 if (write_error < 0)
1868 CLOSE_INVALIDATE (sock);
1872 head = read_http_response_head (sock);
1875 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1877 CLOSE_INVALIDATE (sock);
1886 DEBUGP (("proxy responded with: [%s]\n", head));
1888 resp = resp_new (head);
1889 statcode = resp_status (resp, &message);
1892 char *tms = datetime_str (time (NULL));
1893 logprintf (LOG_VERBOSE, "%d\n", statcode);
1894 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1895 quotearg_style (escape_quoting_style,
1896 _("Malformed status line")));
1900 hs->message = xstrdup (message);
1903 if (statcode != 200)
1906 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1907 message ? quotearg_style (escape_quoting_style, message) : "?");
1908 xfree_null (message);
1911 xfree_null (message);
1913 /* SOCK is now *really* connected to u->host, so update CONN
1914 to reflect this. That way register_persistent will
1915 register SOCK as being connected to u->host:u->port. */
1919 if (conn->scheme == SCHEME_HTTPS)
1921 if (!ssl_connect_wget (sock))
1926 else if (!ssl_check_certificate (sock, u->host))
1929 return VERIFCERTERR;
1933 #endif /* HAVE_SSL */
1936 /* Send the request to server. */
1937 write_error = request_send (req, sock);
1939 if (write_error >= 0)
1943 DEBUGP (("[POST data: %s]\n", opt.post_data));
1944 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1946 else if (opt.post_file_name && post_data_size != 0)
1947 write_error = post_file (sock, opt.post_file_name, post_data_size);
1950 if (write_error < 0)
1952 CLOSE_INVALIDATE (sock);
1956 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1957 proxy ? "Proxy" : "HTTP");
1963 head = read_http_response_head (sock);
1968 logputs (LOG_NOTQUIET, _("No data received.\n"));
1969 CLOSE_INVALIDATE (sock);
1975 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1977 CLOSE_INVALIDATE (sock);
1982 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1984 resp = resp_new (head);
1986 /* Check for status line. */
1988 statcode = resp_status (resp, &message);
1991 char *tms = datetime_str (time (NULL));
1992 logprintf (LOG_VERBOSE, "%d\n", statcode);
1993 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1994 quotearg_style (escape_quoting_style,
1995 _("Malformed status line")));
1996 CLOSE_INVALIDATE (sock);
2001 if (H_10X (statcode))
2003 DEBUGP (("Ignoring response\n"));
2007 hs->message = xstrdup (message);
2008 if (!opt.server_response)
2009 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
2010 message ? quotearg_style (escape_quoting_style, message) : "");
2013 logprintf (LOG_VERBOSE, "\n");
2014 print_server_response (resp, " ");
2017 if (!opt.ignore_length
2018 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
2022 parsed = str_to_wgint (hdrval, NULL, 10);
2023 if (parsed == WGINT_MAX && errno == ERANGE)
2026 #### If Content-Length is out of range, it most likely
2027 means that the file is larger than 2G and that we're
2028 compiled without LFS. In that case we should probably
2029 refuse to even attempt to download the file. */
2032 else if (parsed < 0)
2034 /* Negative Content-Length; nonsensical, so we can't
2035 assume any information about the content to receive. */
2042 /* Check for keep-alive related responses. */
2043 if (!inhibit_keep_alive && contlen != -1)
2045 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2047 if (0 == strcasecmp (hdrval, "Close"))
2052 resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval));
2053 if (0 == strcasecmp (hdrval, "chunked"))
2054 chunked_transfer_encoding = true;
2056 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2060 const char *scbeg, *scend;
2061 /* The jar should have been created by now. */
2062 assert (wget_cookie_jar != NULL);
2064 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2065 &scbeg, &scend)) != -1;
2068 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2069 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2070 u->path, set_cookie);
2075 /* The server has promised that it will not close the connection
2076 when we're done. This means that we can register it. */
2077 register_persistent (conn->host, conn->port, sock, using_ssl);
2079 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2081 /* Authorization is required. */
2082 if (keep_alive && !head_only
2083 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2084 CLOSE_FINISH (sock);
2086 CLOSE_INVALIDATE (sock);
2087 pconn.authorized = false;
2088 if (!auth_finished && (user && passwd))
2090 /* IIS sends multiple copies of WWW-Authenticate, one with
2091 the value "negotiate", and other(s) with data. Loop over
2092 all the occurrences and pick the one we recognize. */
2094 const char *wabeg, *waend;
2095 char *www_authenticate = NULL;
2097 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2098 &wabeg, &waend)) != -1;
2100 if (known_authentication_scheme_p (wabeg, waend))
2102 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2106 if (!www_authenticate)
2108 /* If the authentication header is missing or
2109 unrecognized, there's no sense in retrying. */
2110 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2112 else if (!basic_auth_finished
2113 || !BEGINS_WITH (www_authenticate, "Basic"))
2116 pth = url_full_path (u);
2117 request_set_header (req, "Authorization",
2118 create_authorization_line (www_authenticate,
2120 request_method (req),
2124 if (BEGINS_WITH (www_authenticate, "NTLM"))
2126 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2128 /* Need to register this host as using basic auth,
2129 * so we automatically send creds next time. */
2130 register_basic_auth_host (u->host);
2133 xfree_null (message);
2136 goto retry_with_auth;
2140 /* We already did Basic auth, and it failed. Gotta
2144 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2146 xfree_null (message);
2151 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2153 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2155 pconn.authorized = true;
2158 /* Determine the local filename if needed. Notice that if -O is used
2159 * hstat.local_file is set by http_loop to the argument of -O. */
2160 if (!hs->local_file)
2162 /* Honor Content-Disposition whether possible. */
2163 if (!opt.content_disposition
2164 || !resp_header_copy (resp, "Content-Disposition",
2165 hdrval, sizeof (hdrval))
2166 || !parse_content_disposition (hdrval, &hs->local_file))
2168 /* The Content-Disposition header is missing or broken.
2169 * Choose unique file name according to given URL. */
2170 hs->local_file = url_file_name (u);
2174 /* TODO: perform this check only once. */
2175 if (!hs->existence_checked && file_exists_p (hs->local_file))
2177 if (opt.noclobber && !opt.output_document)
2179 /* If opt.noclobber is turned on and file already exists, do not
2180 retrieve the file. But if the output_document was given, then this
2181 test was already done and the file didn't exist. Hence the !opt.output_document */
2182 logprintf (LOG_VERBOSE, _("\
2183 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2184 /* If the file is there, we suppose it's retrieved OK. */
2187 /* #### Bogusness alert. */
2188 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2189 if (has_html_suffix_p (hs->local_file))
2193 xfree_null (message);
2194 return RETRUNNEEDED;
2196 else if (!ALLOW_CLOBBER)
2198 char *unique = unique_name (hs->local_file, true);
2199 if (unique != hs->local_file)
2200 xfree (hs->local_file);
2201 hs->local_file = unique;
2204 hs->existence_checked = true;
2206 /* Support timestamping */
2207 /* TODO: move this code out of gethttp. */
2208 if (opt.timestamping && !hs->timestamp_checked)
2210 size_t filename_len = strlen (hs->local_file);
2211 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2212 bool local_dot_orig_file_exists = false;
2213 char *local_filename = NULL;
2216 if (opt.backup_converted)
2217 /* If -K is specified, we'll act on the assumption that it was specified
2218 last time these files were downloaded as well, and instead of just
2219 comparing local file X against server file X, we'll compare local
2220 file X.orig (if extant, else X) against server file X. If -K
2221 _wasn't_ specified last time, or the server contains files called
2222 *.orig, -N will be back to not operating correctly with -k. */
2224 /* Would a single s[n]printf() call be faster? --dan
2226 Definitely not. sprintf() is horribly slow. It's a
2227 different question whether the difference between the two
2228 affects a program. Usually I'd say "no", but at one
2229 point I profiled Wget, and found that a measurable and
2230 non-negligible amount of time was lost calling sprintf()
2231 in url.c. Replacing sprintf with inline calls to
2232 strcpy() and number_to_string() made a difference.
2234 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2235 memcpy (filename_plus_orig_suffix + filename_len,
2236 ORIG_SFX, sizeof (ORIG_SFX));
2238 /* Try to stat() the .orig file. */
2239 if (stat (filename_plus_orig_suffix, &st) == 0)
2241 local_dot_orig_file_exists = true;
2242 local_filename = filename_plus_orig_suffix;
2246 if (!local_dot_orig_file_exists)
2247 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2248 if (stat (hs->local_file, &st) == 0)
2249 local_filename = hs->local_file;
2251 if (local_filename != NULL)
2252 /* There was a local file, so we'll check later to see if the version
2253 the server has is the same version we already have, allowing us to
2256 hs->orig_file_name = xstrdup (local_filename);
2257 hs->orig_file_size = st.st_size;
2258 hs->orig_file_tstamp = st.st_mtime;
2260 /* Modification time granularity is 2 seconds for Windows, so
2261 increase local time by 1 second for later comparison. */
2262 ++hs->orig_file_tstamp;
2269 hs->statcode = statcode;
2271 hs->error = xstrdup (_("Malformed status line"));
2273 hs->error = xstrdup (_("(no description)"));
2275 hs->error = xstrdup (message);
2276 xfree_null (message);
2278 type = resp_header_strdup (resp, "Content-Type");
2281 char *tmp = strchr (type, ';');
2284 /* sXXXav: only needed if IRI support is enabled */
2285 char *tmp2 = tmp + 1;
2287 while (tmp > type && c_isspace (tmp[-1]))
2291 /* Try to get remote encoding if needed */
2292 if (opt.enable_iri && !opt.encoding_remote)
2294 tmp = parse_charset (tmp2);
2296 set_content_encoding (iri, tmp);
2300 hs->newloc = resp_header_strdup (resp, "Location");
2301 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2303 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2305 wgint first_byte_pos, last_byte_pos, entity_length;
2306 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2309 contrange = first_byte_pos;
2310 contlen = last_byte_pos - first_byte_pos + 1;
2315 /* 20x responses are counted among successful by default. */
2316 if (H_20X (statcode))
2319 /* Return if redirected. */
2320 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2322 /* RFC2068 says that in case of the 300 (multiple choices)
2323 response, the server can output a preferred URL through
2324 `Location' header; otherwise, the request should be treated
2325 like GET. So, if the location is set, it will be a
2326 redirection; otherwise, just proceed normally. */
2327 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2331 logprintf (LOG_VERBOSE,
2332 _("Location: %s%s\n"),
2333 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2334 hs->newloc ? _(" [following]") : "");
2335 if (keep_alive && !head_only
2336 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2337 CLOSE_FINISH (sock);
2339 CLOSE_INVALIDATE (sock);
2346 /* If content-type is not given, assume text/html. This is because
2347 of the multitude of broken CGI's that "forget" to generate the
2350 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2351 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2357 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2362 if (opt.adjust_extension)
2365 /* -E / --adjust-extension / adjust_extension = on was specified,
2366 and this is a text/html file. If some case-insensitive
2367 variation on ".htm[l]" isn't already the file's suffix,
2370 ensure_extension (hs, ".html", dt);
2372 else if (*dt & TEXTCSS)
2374 ensure_extension (hs, ".css", dt);
2378 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2379 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK
2380 && contrange == 0 && contlen >= 0 && hs->restval >= contlen))
2382 /* If `-c' is in use and the file has been fully downloaded (or
2383 the remote file has shrunk), Wget effectively requests bytes
2384 after the end of file and the server response with 416
2385 (or 200 with a <= Content-Length. */
2386 logputs (LOG_VERBOSE, _("\
2387 \n The file is already fully retrieved; nothing to do.\n\n"));
2388 /* In case the caller inspects. */
2391 /* Mark as successfully retrieved. */
2394 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2395 might be more bytes in the body. */
2397 return RETRUNNEEDED;
2399 if ((contrange != 0 && contrange != hs->restval)
2400 || (H_PARTIAL (statcode) && !contrange))
2402 /* The Range request was somehow misunderstood by the server.
2405 CLOSE_INVALIDATE (sock);
2412 hs->contlen = contlen + contrange;
2418 /* No need to print this output if the body won't be
2419 downloaded at all, or if the original server response is
2421 logputs (LOG_VERBOSE, _("Length: "));
2424 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2425 if (contlen + contrange >= 1024)
2426 logprintf (LOG_VERBOSE, " (%s)",
2427 human_readable (contlen + contrange));
2430 if (contlen >= 1024)
2431 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2432 number_to_static_string (contlen),
2433 human_readable (contlen));
2435 logprintf (LOG_VERBOSE, _(", %s remaining"),
2436 number_to_static_string (contlen));
2440 logputs (LOG_VERBOSE,
2441 opt.ignore_length ? _("ignored") : _("unspecified"));
2443 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2445 logputs (LOG_VERBOSE, "\n");
2449 type = NULL; /* We don't need it any more. */
2451 /* Return if we have no intention of further downloading. */
2452 if (!(*dt & RETROKF) || head_only)
2454 /* In case the caller cares to look... */
2459 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2460 servers not to send body in response to a HEAD request, and
2461 those that do will likely be caught by test_socket_open.
2462 If not, they can be worked around using
2463 `--no-http-keep-alive'. */
2464 CLOSE_FINISH (sock);
2466 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2467 /* Successfully skipped the body; also keep using the socket. */
2468 CLOSE_FINISH (sock);
2470 CLOSE_INVALIDATE (sock);
2472 return RETRFINISHED;
2476 For VMS, define common fopen() optional arguments.
2479 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2480 # define FOPEN_BIN_FLAG 3
2481 #else /* def __VMS */
2482 # define FOPEN_BIN_FLAG true
2483 #endif /* def __VMS [else] */
2485 /* Open the local file. */
2488 mkalldirs (hs->local_file);
2490 rotate_backups (hs->local_file);
2497 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2498 #else /* def __VMS */
2499 fp = fopen (hs->local_file, "ab");
2500 #endif /* def __VMS [else] */
2502 else if (ALLOW_CLOBBER)
2508 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2509 #else /* def __VMS */
2510 fp = fopen (hs->local_file, "wb");
2511 #endif /* def __VMS [else] */
2515 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2516 if (!fp && errno == EEXIST)
2518 /* We cannot just invent a new name and use it (which is
2519 what functions like unique_create typically do)
2520 because we told the user we'd use this name.
2521 Instead, return and retry the download. */
2522 logprintf (LOG_NOTQUIET,
2523 _("%s has sprung into existence.\n"),
2525 CLOSE_INVALIDATE (sock);
2527 return FOPEN_EXCL_ERR;
2532 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2533 CLOSE_INVALIDATE (sock);
2541 /* Print fetch message, if opt.verbose. */
2544 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2545 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2548 /* This confuses the timestamping code that checks for file size.
2549 #### The timestamping code should be smarter about file size. */
2550 if (opt.save_headers && hs->restval == 0)
2551 fwrite (head, 1, strlen (head), fp);
2553 /* Now we no longer need to store the response header. */
2556 /* Download the request body. */
2559 /* If content-length is present, read that much; otherwise, read
2560 until EOF. The HTTP spec doesn't require the server to
2561 actually close the connection when it's done sending data. */
2562 flags |= rb_read_exactly;
2563 if (hs->restval > 0 && contrange == 0)
2564 /* If the server ignored our range request, instruct fd_read_body
2565 to skip the first RESTVAL bytes of body. */
2566 flags |= rb_skip_startpos;
2568 if (chunked_transfer_encoding)
2569 flags |= rb_chunked_transfer_encoding;
2571 hs->len = hs->restval;
2573 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2574 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2578 CLOSE_FINISH (sock);
2582 hs->rderrmsg = xstrdup (fd_errstr (sock));
2583 CLOSE_INVALIDATE (sock);
2590 return RETRFINISHED;
2593 /* The genuine HTTP loop! This is the part where the retrieval is
2594 retried, and retried, and retried, and... */
2596 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2597 int *dt, struct url *proxy, struct iri *iri)
2600 bool got_head = false; /* used for time-stamping and filename detection */
2601 bool time_came_from_head = false;
2602 bool got_name = false;
2605 uerr_t err, ret = TRYLIMEXC;
2606 time_t tmr = -1; /* remote time-stamp */
2607 struct http_stat hstat; /* HTTP status */
2609 bool send_head_first = true;
2611 bool force_full_retrieve = false;
2613 /* Assert that no value for *LOCAL_FILE was passed. */
2614 assert (local_file == NULL || *local_file == NULL);
2616 /* Set LOCAL_FILE parameter. */
2617 if (local_file && opt.output_document)
2618 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2620 /* Reset NEWLOC parameter. */
2623 /* This used to be done in main(), but it's a better idea to do it
2624 here so that we don't go through the hoops if we're just using
2629 /* Warn on (likely bogus) wildcard usage in HTTP. */
2630 if (opt.ftp_glob && has_wildcards_p (u->path))
2631 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2633 /* Setup hstat struct. */
2635 hstat.referer = referer;
2637 if (opt.output_document)
2639 hstat.local_file = xstrdup (opt.output_document);
2642 else if (!opt.content_disposition)
2644 hstat.local_file = url_file_name (u);
2648 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2649 * screaming for some refactoring. */
2650 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2652 /* If opt.noclobber is turned on and file already exists, do not
2653 retrieve the file. But if the output_document was given, then this
2654 test was already done and the file didn't exist. Hence the !opt.output_document */
2655 logprintf (LOG_VERBOSE, _("\
2656 File %s already there; not retrieving.\n\n"),
2657 quote (hstat.local_file));
2658 /* If the file is there, we suppose it's retrieved OK. */
2661 /* #### Bogusness alert. */
2662 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2663 if (has_html_suffix_p (hstat.local_file))
2670 /* Reset the counter. */
2673 /* Reset the document type. */
2676 /* Skip preliminary HEAD request if we're not in spider mode. */
2678 send_head_first = false;
2680 /* Send preliminary HEAD request if -N is given and we have an existing
2681 * destination file. */
2682 file_name = url_file_name (u);
2683 if (opt.timestamping && (file_exists_p (file_name)
2684 || opt.content_disposition))
2685 send_head_first = true;
2691 /* Increment the pass counter. */
2693 sleep_between_retrievals (count);
2695 /* Get the current time string. */
2696 tms = datetime_str (time (NULL));
2698 if (opt.spider && !got_head)
2699 logprintf (LOG_VERBOSE, _("\
2700 Spider mode enabled. Check if remote file exists.\n"));
2702 /* Print fetch message, if opt.verbose. */
2705 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2710 sprintf (tmp, _("(try:%2d)"), count);
2711 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2716 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2721 ws_changetitle (hurl);
2726 /* Default document type is empty. However, if spider mode is
2727 on or time-stamping is employed, HEAD_ONLY commands is
2728 encoded within *dt. */
2729 if (send_head_first && !got_head)
2734 /* Decide whether or not to restart. */
2735 if (force_full_retrieve)
2736 hstat.restval = hstat.len;
2737 else if (opt.always_rest
2739 && stat (hstat.local_file, &st) == 0
2740 && S_ISREG (st.st_mode))
2741 /* When -c is used, continue from on-disk size. (Can't use
2742 hstat.len even if count>1 because we don't want a failed
2743 first attempt to clobber existing data.) */
2744 hstat.restval = st.st_size;
2746 /* otherwise, continue where the previous try left off */
2747 hstat.restval = hstat.len;
2751 /* Decide whether to send the no-cache directive. We send it in
2753 a) we're using a proxy, and we're past our first retrieval.
2754 Some proxies are notorious for caching incomplete data, so
2755 we require a fresh get.
2756 b) caching is explicitly inhibited. */
2757 if ((proxy && count > 1) /* a */
2758 || !opt.allow_cache) /* b */
2759 *dt |= SEND_NOCACHE;
2761 *dt &= ~SEND_NOCACHE;
2763 /* Try fetching the document, or at least its head. */
2764 err = gethttp (u, &hstat, dt, proxy, iri);
2767 tms = datetime_str (time (NULL));
2769 /* Get the new location (with or without the redirection). */
2771 *newloc = xstrdup (hstat.newloc);
2775 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2776 case CONERROR: case READERR: case WRITEFAILED:
2777 case RANGEERR: case FOPEN_EXCL_ERR:
2778 /* Non-fatal errors continue executing the loop, which will
2779 bring them to "while" statement at the end, to judge
2780 whether the number of tries was exceeded. */
2781 printwhat (count, opt.ntry);
2783 case FWRITEERR: case FOPENERR:
2784 /* Another fatal error. */
2785 logputs (LOG_VERBOSE, "\n");
2786 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2787 quote (hstat.local_file), strerror (errno));
2788 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2789 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2790 /* Fatal errors just return from the function. */
2794 /* Another fatal error. */
2795 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2799 /* Return the new location to the caller. */
2802 logprintf (LOG_NOTQUIET,
2803 _("ERROR: Redirection (%d) without location.\n"),
2813 /* The file was already fully retrieved. */
2817 /* Deal with you later. */
2820 /* All possibilities should have been exhausted. */
2824 if (!(*dt & RETROKF))
2829 /* #### Ugly ugly ugly! */
2830 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2831 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2834 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2836 && (hstat.statcode == 500 || hstat.statcode == 501))
2841 /* Maybe we should always keep track of broken links, not just in
2843 * Don't log error if it was UTF-8 encoded because we will try
2844 * once unencoded. */
2845 else if (opt.spider && !iri->utf8_encode)
2847 /* #### Again: ugly ugly ugly! */
2849 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2850 nonexisting_url (hurl);
2851 logprintf (LOG_NOTQUIET, _("\
2852 Remote file does not exist -- broken link!!!\n"));
2856 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2857 tms, hstat.statcode,
2858 quotearg_style (escape_quoting_style, hstat.error));
2860 logputs (LOG_VERBOSE, "\n");
2866 /* Did we get the time-stamp? */
2869 got_head = true; /* no more time-stamping */
2871 if (opt.timestamping && !hstat.remote_time)
2873 logputs (LOG_NOTQUIET, _("\
2874 Last-modified header missing -- time-stamps turned off.\n"));
2876 else if (hstat.remote_time)
2878 /* Convert the date-string into struct tm. */
2879 tmr = http_atotm (hstat.remote_time);
2880 if (tmr == (time_t) (-1))
2881 logputs (LOG_VERBOSE, _("\
2882 Last-modified header invalid -- time-stamp ignored.\n"));
2883 if (*dt & HEAD_ONLY)
2884 time_came_from_head = true;
2887 if (send_head_first)
2889 /* The time-stamping section. */
2890 if (opt.timestamping)
2892 if (hstat.orig_file_name) /* Perform the following
2893 checks only if the file
2895 download already exists. */
2897 if (hstat.remote_time &&
2898 tmr != (time_t) (-1))
2900 /* Now time-stamping can be used validly.
2901 Time-stamping means that if the sizes of
2902 the local and remote file match, and local
2903 file is newer than the remote file, it will
2904 not be retrieved. Otherwise, the normal
2905 download procedure is resumed. */
2906 if (hstat.orig_file_tstamp >= tmr)
2908 if (hstat.contlen == -1
2909 || hstat.orig_file_size == hstat.contlen)
2911 logprintf (LOG_VERBOSE, _("\
2912 Server file no newer than local file %s -- not retrieving.\n\n"),
2913 quote (hstat.orig_file_name));
2919 logprintf (LOG_VERBOSE, _("\
2920 The sizes do not match (local %s) -- retrieving.\n"),
2921 number_to_static_string (hstat.orig_file_size));
2926 force_full_retrieve = true;
2927 logputs (LOG_VERBOSE,
2928 _("Remote file is newer, retrieving.\n"));
2931 logputs (LOG_VERBOSE, "\n");
2935 /* free_hstat (&hstat); */
2936 hstat.timestamp_checked = true;
2941 bool finished = true;
2946 logputs (LOG_VERBOSE, _("\
2947 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2952 logprintf (LOG_VERBOSE, _("\
2953 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2954 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2961 logprintf (LOG_VERBOSE, _("\
2962 Remote file exists and could contain further links,\n\
2963 but recursion is disabled -- not retrieving.\n\n"));
2967 logprintf (LOG_VERBOSE, _("\
2968 Remote file exists.\n\n"));
2970 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2975 logprintf (LOG_NONVERBOSE,
2976 _("%s URL: %s %2d %s\n"),
2977 tms, u->url, hstat.statcode,
2978 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2985 count = 0; /* the retrieve count for HEAD is reset */
2987 } /* send_head_first */
2990 if (opt.useservertimestamps
2991 && (tmr != (time_t) (-1))
2992 && ((hstat.len == hstat.contlen) ||
2993 ((hstat.res == 0) && (hstat.contlen == -1))))
2995 const char *fl = NULL;
2996 set_local_file (&fl, hstat.local_file);
3000 /* Reparse time header, in case it's changed. */
3001 if (time_came_from_head
3002 && hstat.remote_time && hstat.remote_time[0])
3004 newtmr = http_atotm (hstat.remote_time);
3005 if (newtmr != (time_t)-1)
3011 /* End of time-stamping section. */
3013 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3014 total_download_time += hstat.dltime;
3016 if (hstat.len == hstat.contlen)
3020 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3022 logprintf (LOG_VERBOSE,
3024 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3025 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3027 write_to_stdout ? "" : quote (hstat.local_file),
3028 number_to_static_string (hstat.len),
3029 number_to_static_string (hstat.contlen));
3030 logprintf (LOG_NONVERBOSE,
3031 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3033 number_to_static_string (hstat.len),
3034 number_to_static_string (hstat.contlen),
3035 hstat.local_file, count);
3038 total_downloaded_bytes += hstat.rd_size;
3040 /* Remember that we downloaded the file for later ".orig" code. */
3041 if (*dt & ADDED_HTML_EXTENSION)
3042 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3044 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3049 else if (hstat.res == 0) /* No read error */
3051 if (hstat.contlen == -1) /* We don't know how much we were supposed
3052 to get, so assume we succeeded. */
3056 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3058 logprintf (LOG_VERBOSE,
3060 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3061 : _("%s (%s) - %s saved [%s]\n\n"),
3063 write_to_stdout ? "" : quote (hstat.local_file),
3064 number_to_static_string (hstat.len));
3065 logprintf (LOG_NONVERBOSE,
3066 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3067 tms, u->url, number_to_static_string (hstat.len),
3068 hstat.local_file, count);
3071 total_downloaded_bytes += hstat.rd_size;
3073 /* Remember that we downloaded the file for later ".orig" code. */
3074 if (*dt & ADDED_HTML_EXTENSION)
3075 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3077 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3082 else if (hstat.len < hstat.contlen) /* meaning we lost the
3083 connection too soon */
3085 logprintf (LOG_VERBOSE,
3086 _("%s (%s) - Connection closed at byte %s. "),
3087 tms, tmrate, number_to_static_string (hstat.len));
3088 printwhat (count, opt.ntry);
3091 else if (hstat.len != hstat.restval)
3092 /* Getting here would mean reading more data than
3093 requested with content-length, which we never do. */
3097 /* Getting here probably means that the content-length was
3098 * _less_ than the original, local size. We should probably
3099 * truncate or re-read, or something. FIXME */
3104 else /* from now on hstat.res can only be -1 */
3106 if (hstat.contlen == -1)
3108 logprintf (LOG_VERBOSE,
3109 _("%s (%s) - Read error at byte %s (%s)."),
3110 tms, tmrate, number_to_static_string (hstat.len),
3112 printwhat (count, opt.ntry);
3115 else /* hstat.res == -1 and contlen is given */
3117 logprintf (LOG_VERBOSE,
3118 _("%s (%s) - Read error at byte %s/%s (%s). "),
3120 number_to_static_string (hstat.len),
3121 number_to_static_string (hstat.contlen),
3123 printwhat (count, opt.ntry);
3129 while (!opt.ntry || (count < opt.ntry));
3133 *local_file = xstrdup (hstat.local_file);
3134 free_hstat (&hstat);
3139 /* Check whether the result of strptime() indicates success.
3140 strptime() returns the pointer to how far it got to in the string.
3141 The processing has been successful if the string is at `GMT' or
3142 `+X', or at the end of the string.
3144 In extended regexp parlance, the function returns 1 if P matches
3145 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3146 can return) is considered a failure and 0 is returned. */
3148 check_end (const char *p)
3152 while (c_isspace (*p))
3155 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3156 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3162 /* Convert the textual specification of time in TIME_STRING to the
3163 number of seconds since the Epoch.
3165 TIME_STRING can be in any of the three formats RFC2616 allows the
3166 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3167 as well as the time format used in the Set-Cookie header.
3168 Timezones are ignored, and should be GMT.
3170 Return the computed time_t representation, or -1 if the conversion
3173 This function uses strptime with various string formats for parsing
3174 TIME_STRING. This results in a parser that is not as lenient in
3175 interpreting TIME_STRING as I would like it to be. Being based on
3176 strptime, it always allows shortened months, one-digit days, etc.,
3177 but due to the multitude of formats in which time can be
3178 represented, an ideal HTTP time parser would be even more
3179 forgiving. It should completely ignore things like week days and
3180 concentrate only on the various forms of representing years,
3181 months, days, hours, minutes, and seconds. For example, it would
3182 be nice if it accepted ISO 8601 out of the box.
3184 I've investigated free and PD code for this purpose, but none was
3185 usable. getdate was big and unwieldy, and had potential copyright
3186 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3187 distributed with phttpd, is excellent, but we cannot use it because
3188 it is not assigned to the FSF. So I stuck it with strptime. */
3191 http_atotm (const char *time_string)
3193 /* NOTE: Solaris strptime man page claims that %n and %t match white
3194 space, but that's not universally available. Instead, we simply
3195 use ` ' to mean "skip all WS", which works under all strptime
3196 implementations I've tested. */
3198 static const char *time_formats[] = {
3199 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3200 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3201 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3202 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3203 (used in Set-Cookie, defined in the
3204 Netscape cookie specification.) */
3206 const char *oldlocale;
3207 char savedlocale[256];
3209 time_t ret = (time_t) -1;
3211 /* Solaris strptime fails to recognize English month names in
3212 non-English locales, which we work around by temporarily setting
3213 locale to C before invoking strptime. */
3214 oldlocale = setlocale (LC_TIME, NULL);
3217 size_t l = strlen (oldlocale) + 1;
3218 if (l >= sizeof savedlocale)
3219 savedlocale[0] = '\0';
3221 memcpy (savedlocale, oldlocale, l);
3223 else savedlocale[0] = '\0';
3225 setlocale (LC_TIME, "C");
3227 for (i = 0; i < countof (time_formats); i++)
3231 /* Some versions of strptime use the existing contents of struct
3232 tm to recalculate the date according to format. Zero it out
3233 to prevent stack garbage from influencing strptime. */
3236 if (check_end (strptime (time_string, time_formats[i], &t)))
3243 /* Restore the previous locale. */
3245 setlocale (LC_TIME, savedlocale);
3250 /* Authorization support: We support three authorization schemes:
3252 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3254 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3255 consisting of answering to the server's challenge with the proper
3258 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3259 Stenberg for libcurl. Like digest, NTLM is based on a
3260 challenge-response mechanism, but unlike digest, it is non-standard
3261 (authenticates TCP connections rather than requests), undocumented
3262 and Microsoft-specific. */
3264 /* Create the authentication header contents for the `Basic' scheme.
3265 This is done by encoding the string "USER:PASS" to base64 and
3266 prepending the string "Basic " in front of it. */
3269 basic_authentication_encode (const char *user, const char *passwd)
3272 int len1 = strlen (user) + 1 + strlen (passwd);
3274 t1 = (char *)alloca (len1 + 1);
3275 sprintf (t1, "%s:%s", user, passwd);
3277 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3278 base64_encode (t1, len1, t2);
3280 return concat_strings ("Basic ", t2, (char *) 0);
3283 #define SKIP_WS(x) do { \
3284 while (c_isspace (*(x))) \
3288 #ifdef ENABLE_DIGEST
3289 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3290 an array of 16 bytes containing the hash keys, and BUF should be a
3291 buffer of 33 writable characters (32 for hex digits plus one for
3292 zero termination). */
3294 dump_hash (char *buf, const unsigned char *hash)
3298 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3300 *buf++ = XNUM_TO_digit (*hash >> 4);
3301 *buf++ = XNUM_TO_digit (*hash & 0xf);
3306 /* Take the line apart to find the challenge, and compose a digest
3307 authorization header. See RFC2069 section 2.1.2. */
3309 digest_authentication_encode (const char *au, const char *user,
3310 const char *passwd, const char *method,
3313 static char *realm, *opaque, *nonce;
3318 { "realm", &realm },
3319 { "opaque", &opaque },
3323 param_token name, value;
3325 realm = opaque = nonce = NULL;
3327 au += 6; /* skip over `Digest' */
3328 while (extract_param (&au, &name, &value, ','))
3331 size_t namelen = name.e - name.b;
3332 for (i = 0; i < countof (options); i++)
3333 if (namelen == strlen (options[i].name)
3334 && 0 == strncmp (name.b, options[i].name,
3337 *options[i].variable = strdupdelim (value.b, value.e);
3341 if (!realm || !nonce || !user || !passwd || !path || !method)
3344 xfree_null (opaque);
3349 /* Calculate the digest value. */
3352 unsigned char hash[MD5_DIGEST_SIZE];
3353 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3354 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3356 /* A1BUF = H(user ":" realm ":" password) */
3357 md5_init_ctx (&ctx);
3358 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3359 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3360 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3361 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3362 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3363 md5_finish_ctx (&ctx, hash);
3364 dump_hash (a1buf, hash);
3366 /* A2BUF = H(method ":" path) */
3367 md5_init_ctx (&ctx);
3368 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3369 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3370 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3371 md5_finish_ctx (&ctx, hash);
3372 dump_hash (a2buf, hash);
3374 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3375 md5_init_ctx (&ctx);
3376 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3377 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3378 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3379 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3380 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3381 md5_finish_ctx (&ctx, hash);
3382 dump_hash (response_digest, hash);
3384 res = xmalloc (strlen (user)
3389 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3390 + (opaque ? strlen (opaque) : 0)
3392 sprintf (res, "Digest \
3393 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3394 user, realm, nonce, path, response_digest);
3397 char *p = res + strlen (res);
3398 strcat (p, ", opaque=\"");
3405 #endif /* ENABLE_DIGEST */
3407 /* Computing the size of a string literal must take into account that
3408 value returned by sizeof includes the terminating \0. */
3409 #define STRSIZE(literal) (sizeof (literal) - 1)
3411 /* Whether chars in [b, e) begin with the literal string provided as
3412 first argument and are followed by whitespace or terminating \0.
3413 The comparison is case-insensitive. */
3414 #define STARTS(literal, b, e) \
3416 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3417 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3418 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3419 || c_isspace (b[STRSIZE (literal)])))
3422 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3424 return STARTS ("Basic", hdrbeg, hdrend)
3425 #ifdef ENABLE_DIGEST
3426 || STARTS ("Digest", hdrbeg, hdrend)
3429 || STARTS ("NTLM", hdrbeg, hdrend)
3436 /* Create the HTTP authorization request header. When the
3437 `WWW-Authenticate' response header is seen, according to the
3438 authorization scheme specified in that header (`Basic' and `Digest'
3439 are supported by the current implementation), produce an
3440 appropriate HTTP authorization request header. */
3442 create_authorization_line (const char *au, const char *user,
3443 const char *passwd, const char *method,
3444 const char *path, bool *finished)
3446 /* We are called only with known schemes, so we can dispatch on the
3448 switch (c_toupper (*au))
3450 case 'B': /* Basic */
3452 return basic_authentication_encode (user, passwd);
3453 #ifdef ENABLE_DIGEST
3454 case 'D': /* Digest */
3456 return digest_authentication_encode (au, user, passwd, method, path);
3459 case 'N': /* NTLM */
3460 if (!ntlm_input (&pconn.ntlm, au))
3465 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3468 /* We shouldn't get here -- this function should be only called
3469 with values approved by known_authentication_scheme_p. */
3477 if (!wget_cookie_jar)
3478 wget_cookie_jar = cookie_jar_new ();
3479 if (opt.cookies_input && !cookies_loaded_p)
3481 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3482 cookies_loaded_p = true;
3489 if (wget_cookie_jar)
3490 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3496 xfree_null (pconn.host);
3497 if (wget_cookie_jar)
3498 cookie_jar_delete (wget_cookie_jar);
3502 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3504 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3506 int len = strlen (ext);
3509 strncpy (shortext, ext, len - 1);
3510 shortext[len - 2] = '\0';
3513 if (last_period_in_local_filename == NULL
3514 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3515 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3517 int local_filename_len = strlen (hs->local_file);
3518 /* Resize the local file, allowing for ".html" preceded by
3519 optional ".NUMBER". */
3520 hs->local_file = xrealloc (hs->local_file,
3521 local_filename_len + 24 + len);
3522 strcpy (hs->local_file + local_filename_len, ext);
3523 /* If clobbering is not allowed and the file, as named,
3524 exists, tack on ".NUMBER.html" instead. */
3525 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3529 sprintf (hs->local_file + local_filename_len,
3530 ".%d%s", ext_num++, ext);
3531 while (file_exists_p (hs->local_file));
3533 *dt |= ADDED_HTML_EXTENSION;
3541 test_parse_content_disposition()
3546 char *opt_dir_prefix;
3550 { "filename=\"file.ext\"", NULL, "file.ext", true },
3551 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3552 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3553 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3554 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3555 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3556 { "attachment", NULL, NULL, false },
3557 { "attachment", "somedir", NULL, false },
3558 { "attachement; filename*=UTF-8'en-US'hello.txt", NULL, "hello.txt", true },
3559 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", NULL, "helloworld.txt", true },
3562 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3567 opt.dir_prefix = test_array[i].opt_dir_prefix;
3568 res = parse_content_disposition (test_array[i].hdrval, &filename);
3570 mu_assert ("test_parse_content_disposition: wrong result",
3571 res == test_array[i].result
3573 || 0 == strcmp (test_array[i].filename, filename)));
3579 #endif /* TESTING */
3582 * vim: et sts=2 sw=2 cino+={s