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);
932 remaining_chunk_size = strtol (line, &endl, 16);
933 if (remaining_chunk_size == 0)
940 contlen = MIN (remaining_chunk_size, SKIP_SIZE);
943 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
945 ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
948 /* Don't normally report the error since this is an
949 optimization that should be invisible to the user. */
950 DEBUGP (("] aborting (%s).\n",
951 ret < 0 ? fd_errstr (fd) : "EOF received"));
958 remaining_chunk_size -= ret;
959 if (remaining_chunk_size == 0)
960 if (fd_read_line (fd) == NULL)
964 /* Safe even if %.*s bogusly expects terminating \0 because
965 we've zero-terminated dlbuf above. */
966 DEBUGP (("%.*s", ret, dlbuf));
969 DEBUGP (("] done.\n"));
973 #define NOT_RFC2231 0
974 #define RFC2231_NOENCODING 1
975 #define RFC2231_ENCODING 2
977 /* extract_param extracts the parameter name into NAME.
978 However, if the parameter name is in RFC2231 format then
979 this function adjusts NAME by stripping of the trailing
980 characters that are not part of the name but are present to
981 indicate the presence of encoding information in the value
982 or a fragment of a long parameter value
985 modify_param_name(param_token *name)
987 const char *delim1 = memchr (name->b, '*', name->e - name->b);
988 const char *delim2 = memrchr (name->b, '*', name->e - name->b);
994 result = NOT_RFC2231;
996 else if(delim1 == delim2)
998 if ((name->e - 1) == delim1)
1000 result = RFC2231_ENCODING;
1004 result = RFC2231_NOENCODING;
1011 result = RFC2231_ENCODING;
1016 /* extract_param extract the paramater value into VALUE.
1017 Like modify_param_name this function modifies VALUE by
1018 stripping off the encoding information from the actual value
1021 modify_param_value (param_token *value, int encoding_type )
1023 if (RFC2231_ENCODING == encoding_type)
1025 const char *delim = memrchr (value->b, '\'', value->e - value->b);
1026 if ( delim != NULL )
1028 value->b = (delim+1);
1033 /* Extract a parameter from the string (typically an HTTP header) at
1034 **SOURCE and advance SOURCE to the next parameter. Return false
1035 when there are no more parameters to extract. The name of the
1036 parameter is returned in NAME, and the value in VALUE. If the
1037 parameter has no value, the token's value is zeroed out.
1039 For example, if *SOURCE points to the string "attachment;
1040 filename=\"foo bar\"", the first call to this function will return
1041 the token named "attachment" and no value, and the second call will
1042 return the token named "filename" and value "foo bar". The third
1043 call will return false, indicating no more valid tokens. */
1046 extract_param (const char **source, param_token *name, param_token *value,
1049 const char *p = *source;
1051 while (c_isspace (*p)) ++p;
1055 return false; /* no error; nothing more to extract */
1060 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
1062 if (name->b == name->e)
1063 return false; /* empty name: error */
1064 while (c_isspace (*p)) ++p;
1065 if (*p == separator || !*p) /* no value */
1068 if (*p == separator) ++p;
1073 return false; /* error */
1075 /* *p is '=', extract value */
1077 while (c_isspace (*p)) ++p;
1078 if (*p == '"') /* quoted */
1081 while (*p && *p != '"') ++p;
1085 /* Currently at closing quote; find the end of param. */
1086 while (c_isspace (*p)) ++p;
1087 while (*p && *p != separator) ++p;
1088 if (*p == separator)
1091 /* garbage after closed quote, e.g. foo="bar"baz */
1097 while (*p && *p != separator) ++p;
1099 while (value->e != value->b && c_isspace (value->e[-1]))
1101 if (*p == separator) ++p;
1105 int param_type = modify_param_name(name);
1106 if (NOT_RFC2231 != param_type)
1108 modify_param_value(value, param_type);
1114 #undef RFC2231_NOENCODING
1115 #undef RFC2231_ENCODING
1117 /* Appends the string represented by VALUE to FILENAME */
1120 append_value_to_filename (char **filename, param_token const * const value)
1122 int original_length = strlen(*filename);
1123 int new_length = strlen(*filename) + (value->e - value->b);
1124 *filename = xrealloc (*filename, new_length+1);
1125 memcpy (*filename + original_length, value->b, (value->e - value->b));
1126 (*filename)[new_length] = '\0';
1130 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1132 /* Parse the contents of the `Content-Disposition' header, extracting
1133 the information useful to Wget. Content-Disposition is a header
1134 borrowed from MIME; when used in HTTP, it typically serves for
1135 specifying the desired file name of the resource. For example:
1137 Content-Disposition: attachment; filename="flora.jpg"
1139 Wget will skip the tokens it doesn't care about, such as
1140 "attachment" in the previous example; it will also skip other
1141 unrecognized params. If the header is syntactically correct and
1142 contains a file name, a copy of the file name is stored in
1143 *filename and true is returned. Otherwise, the function returns
1146 The file name is stripped of directory components and must not be
1149 Historically, this function returned filename prefixed with opt.dir_prefix,
1150 now that logic is handled by the caller, new code should pay attention,
1151 changed by crq, Sep 2010.
1155 parse_content_disposition (const char *hdr, char **filename)
1157 param_token name, value;
1159 while (extract_param (&hdr, &name, &value, ';'))
1161 int isFilename = BOUNDED_EQUAL_NO_CASE ( name.b, name.e, "filename" );
1162 if ( isFilename && value.b != NULL)
1164 /* Make the file name begin at the last slash or backslash. */
1165 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1166 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1167 if (last_slash && last_bs)
1168 value.b = 1 + MAX (last_slash, last_bs);
1169 else if (last_slash || last_bs)
1170 value.b = 1 + (last_slash ? last_slash : last_bs);
1171 if (value.b == value.e)
1175 append_value_to_filename (filename, &value);
1177 *filename = strdupdelim (value.b, value.e);
1188 /* Persistent connections. Currently, we cache the most recently used
1189 connection as persistent, provided that the HTTP server agrees to
1190 make it such. The persistence data is stored in the variables
1191 below. Ideally, it should be possible to cache an arbitrary fixed
1192 number of these connections. */
1194 /* Whether a persistent connection is active. */
1195 static bool pconn_active;
1198 /* The socket of the connection. */
1201 /* Host and port of the currently active persistent connection. */
1205 /* Whether a ssl handshake has occoured on this connection. */
1208 /* Whether the connection was authorized. This is only done by
1209 NTLM, which authorizes *connections* rather than individual
1210 requests. (That practice is peculiar for HTTP, but it is a
1211 useful optimization.) */
1215 /* NTLM data of the current connection. */
1216 struct ntlmdata ntlm;
1220 /* Mark the persistent connection as invalid and free the resources it
1221 uses. This is used by the CLOSE_* macros after they forcefully
1222 close a registered persistent connection. */
1225 invalidate_persistent (void)
1227 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1228 pconn_active = false;
1229 fd_close (pconn.socket);
1234 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1235 persistent. This will enable someone to use the same connection
1236 later. In the context of HTTP, this must be called only AFTER the
1237 response has been received and the server has promised that the
1238 connection will remain alive.
1240 If a previous connection was persistent, it is closed. */
1243 register_persistent (const char *host, int port, int fd, bool ssl)
1247 if (pconn.socket == fd)
1249 /* The connection FD is already registered. */
1254 /* The old persistent connection is still active; close it
1255 first. This situation arises whenever a persistent
1256 connection exists, but we then connect to a different
1257 host, and try to register a persistent connection to that
1259 invalidate_persistent ();
1263 pconn_active = true;
1265 pconn.host = xstrdup (host);
1268 pconn.authorized = false;
1270 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1273 /* Return true if a persistent connection is available for connecting
1277 persistent_available_p (const char *host, int port, bool ssl,
1278 bool *host_lookup_failed)
1280 /* First, check whether a persistent connection is active at all. */
1284 /* If we want SSL and the last connection wasn't or vice versa,
1285 don't use it. Checking for host and port is not enough because
1286 HTTP and HTTPS can apparently coexist on the same port. */
1287 if (ssl != pconn.ssl)
1290 /* If we're not connecting to the same port, we're not interested. */
1291 if (port != pconn.port)
1294 /* If the host is the same, we're in business. If not, there is
1295 still hope -- read below. */
1296 if (0 != strcasecmp (host, pconn.host))
1298 /* Check if pconn.socket is talking to HOST under another name.
1299 This happens often when both sites are virtual hosts
1300 distinguished only by name and served by the same network
1301 interface, and hence the same web server (possibly set up by
1302 the ISP and serving many different web sites). This
1303 admittedly unconventional optimization does not contradict
1304 HTTP and works well with popular server software. */
1308 struct address_list *al;
1311 /* Don't try to talk to two different SSL sites over the same
1312 secure connection! (Besides, it's not clear that
1313 name-based virtual hosting is even possible with SSL.) */
1316 /* If pconn.socket's peer is one of the IP addresses HOST
1317 resolves to, pconn.socket is for all intents and purposes
1318 already talking to HOST. */
1320 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1322 /* Can't get the peer's address -- something must be very
1323 wrong with the connection. */
1324 invalidate_persistent ();
1327 al = lookup_host (host, 0);
1330 *host_lookup_failed = true;
1334 found = address_list_contains (al, &ip);
1335 address_list_release (al);
1340 /* The persistent connection's peer address was found among the
1341 addresses HOST resolved to; therefore, pconn.sock is in fact
1342 already talking to HOST -- no need to reconnect. */
1345 /* Finally, check whether the connection is still open. This is
1346 important because most servers implement liberal (short) timeout
1347 on persistent connections. Wget can of course always reconnect
1348 if the connection doesn't work out, but it's nicer to know in
1349 advance. This test is a logical followup of the first test, but
1350 is "expensive" and therefore placed at the end of the list.
1352 (Current implementation of test_socket_open has a nice side
1353 effect that it treats sockets with pending data as "closed".
1354 This is exactly what we want: if a broken server sends message
1355 body in response to HEAD, or if it sends more than conent-length
1356 data, we won't reuse the corrupted connection.) */
1358 if (!test_socket_open (pconn.socket))
1360 /* Oops, the socket is no longer open. Now that we know that,
1361 let's invalidate the persistent connection before returning
1363 invalidate_persistent ();
1370 /* The idea behind these two CLOSE macros is to distinguish between
1371 two cases: one when the job we've been doing is finished, and we
1372 want to close the connection and leave, and two when something is
1373 seriously wrong and we're closing the connection as part of
1376 In case of keep_alive, CLOSE_FINISH should leave the connection
1377 open, while CLOSE_INVALIDATE should still close it.
1379 Note that the semantics of the flag `keep_alive' is "this
1380 connection *will* be reused (the server has promised not to close
1381 the connection once we're done)", while the semantics of
1382 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1383 active, registered connection". */
1385 #define CLOSE_FINISH(fd) do { \
1388 if (pconn_active && (fd) == pconn.socket) \
1389 invalidate_persistent (); \
1398 #define CLOSE_INVALIDATE(fd) do { \
1399 if (pconn_active && (fd) == pconn.socket) \
1400 invalidate_persistent (); \
1408 wgint len; /* received length */
1409 wgint contlen; /* expected length */
1410 wgint restval; /* the restart value */
1411 int res; /* the result of last read */
1412 char *rderrmsg; /* error message from read error */
1413 char *newloc; /* new location (redirection) */
1414 char *remote_time; /* remote time-stamp string */
1415 char *error; /* textual HTTP error */
1416 int statcode; /* status code */
1417 char *message; /* status message */
1418 wgint rd_size; /* amount of data read from socket */
1419 double dltime; /* time it took to download the data */
1420 const char *referer; /* value of the referer header. */
1421 char *local_file; /* local file name. */
1422 bool existence_checked; /* true if we already checked for a file's
1423 existence after having begun to download
1424 (needed in gethttp for when connection is
1425 interrupted/restarted. */
1426 bool timestamp_checked; /* true if pre-download time-stamping checks
1427 * have already been performed */
1428 char *orig_file_name; /* name of file to compare for time-stamping
1429 * (might be != local_file if -K is set) */
1430 wgint orig_file_size; /* size of file to compare for time-stamping */
1431 time_t orig_file_tstamp; /* time-stamp of file to compare for
1436 free_hstat (struct http_stat *hs)
1438 xfree_null (hs->newloc);
1439 xfree_null (hs->remote_time);
1440 xfree_null (hs->error);
1441 xfree_null (hs->rderrmsg);
1442 xfree_null (hs->local_file);
1443 xfree_null (hs->orig_file_name);
1444 xfree_null (hs->message);
1446 /* Guard against being called twice. */
1448 hs->remote_time = NULL;
1452 #define BEGINS_WITH(line, string_constant) \
1453 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1454 && (c_isspace (line[sizeof (string_constant) - 1]) \
1455 || !line[sizeof (string_constant) - 1]))
1458 #define SET_USER_AGENT(req) do { \
1459 if (!opt.useragent) \
1460 request_set_header (req, "User-Agent", \
1461 aprintf ("Wget/%s (VMS %s %s)", \
1462 version_string, vms_arch(), vms_vers()), \
1464 else if (*opt.useragent) \
1465 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1467 #else /* def __VMS */
1468 #define SET_USER_AGENT(req) do { \
1469 if (!opt.useragent) \
1470 request_set_header (req, "User-Agent", \
1471 aprintf ("Wget/%s (%s)", \
1472 version_string, OS_TYPE), \
1474 else if (*opt.useragent) \
1475 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1477 #endif /* def __VMS [else] */
1479 /* The flags that allow clobbering the file (opening with "wb").
1480 Defined here to avoid repetition later. #### This will require
1482 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1483 || opt.dirstruct || opt.output_document)
1485 /* Retrieve a document through HTTP protocol. It recognizes status
1486 code, and correctly handles redirections. It closes the network
1487 socket. If it receives an error from the functions below it, it
1488 will print it if there is enough information to do so (almost
1489 always), returning the error to the caller (i.e. http_loop).
1491 Various HTTP parameters are stored to hs.
1493 If PROXY is non-NULL, the connection will be made to the proxy
1494 server, and u->url will be requested. */
1496 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy,
1499 struct request *req;
1502 char *user, *passwd;
1506 wgint contlen, contrange;
1513 /* Set to 1 when the authorization has already been sent and should
1514 not be tried again. */
1515 bool auth_finished = false;
1517 /* Set to 1 when just globally-set Basic authorization has been sent;
1518 * should prevent further Basic negotiations, but not other
1520 bool basic_auth_finished = false;
1522 /* Whether NTLM authentication is used for this request. */
1523 bool ntlm_seen = false;
1525 /* Whether our connection to the remote host is through SSL. */
1526 bool using_ssl = false;
1528 /* Whether a HEAD request will be issued (as opposed to GET or
1530 bool head_only = !!(*dt & HEAD_ONLY);
1533 struct response *resp;
1537 /* Whether this connection will be kept alive after the HTTP request
1541 /* Is the server using the chunked transfer encoding? */
1542 bool chunked_transfer_encoding = false;
1544 /* Whether keep-alive should be inhibited. */
1545 bool inhibit_keep_alive =
1546 !opt.http_keep_alive || opt.ignore_length;
1548 /* Headers sent when using POST. */
1549 wgint post_data_size = 0;
1551 bool host_lookup_failed = false;
1554 if (u->scheme == SCHEME_HTTPS)
1556 /* Initialize the SSL context. After this has once been done,
1557 it becomes a no-op. */
1560 scheme_disable (SCHEME_HTTPS);
1561 logprintf (LOG_NOTQUIET,
1562 _("Disabling SSL due to encountered errors.\n"));
1563 return SSLINITFAILED;
1566 #endif /* HAVE_SSL */
1568 /* Initialize certain elements of struct http_stat. */
1572 hs->rderrmsg = NULL;
1574 hs->remote_time = NULL;
1580 /* Prepare the request to send. */
1582 req = request_new ();
1585 const char *meth = "GET";
1588 else if (opt.post_file_name || opt.post_data)
1590 /* Use the full path, i.e. one that includes the leading slash and
1591 the query string. E.g. if u->path is "foo/bar" and u->query is
1592 "param=value", full_path will be "/foo/bar?param=value". */
1595 /* When using SSL over proxy, CONNECT establishes a direct
1596 connection to the HTTPS server. Therefore use the same
1597 argument as when talking to the server directly. */
1598 && u->scheme != SCHEME_HTTPS
1601 meth_arg = xstrdup (u->url);
1603 meth_arg = url_full_path (u);
1604 request_set_method (req, meth, meth_arg);
1607 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1608 if (*dt & SEND_NOCACHE)
1609 request_set_header (req, "Pragma", "no-cache", rel_none);
1610 if (hs->restval && !opt.timestamping)
1611 request_set_header (req, "Range",
1612 aprintf ("bytes=%s-",
1613 number_to_static_string (hs->restval)),
1615 SET_USER_AGENT (req);
1616 request_set_header (req, "Accept", "*/*", rel_none);
1618 /* Find the username and password for authentication. */
1621 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1622 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1623 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1625 /* We only do "site-wide" authentication with "global" user/password
1626 * values unless --auth-no-challange has been requested; URL user/password
1627 * info overrides. */
1628 if (user && passwd && (!u->user || opt.auth_without_challenge))
1630 /* If this is a host for which we've already received a Basic
1631 * challenge, we'll go ahead and send Basic authentication creds. */
1632 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1635 /* Generate the Host header, HOST:PORT. Take into account that:
1637 - Broken server-side software often doesn't recognize the PORT
1638 argument, so we must generate "Host: www.server.com" instead of
1639 "Host: www.server.com:80" (and likewise for https port).
1641 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1642 becomes ambiguous and needs to be rewritten as "Host:
1643 [3ffe:8100:200:2::2]:1234". */
1645 /* Formats arranged for hfmt[add_port][add_squares]. */
1646 static const char *hfmt[][2] = {
1647 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1649 int add_port = u->port != scheme_default_port (u->scheme);
1650 int add_squares = strchr (u->host, ':') != NULL;
1651 request_set_header (req, "Host",
1652 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1656 if (inhibit_keep_alive)
1657 request_set_header (req, "Connection", "Close", rel_none);
1661 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1664 request_set_header (req, "Connection", "Close", rel_none);
1665 request_set_header (req, "Proxy-Connection", "Keep-Alive", rel_none);
1669 if (opt.post_data || opt.post_file_name)
1671 request_set_header (req, "Content-Type",
1672 "application/x-www-form-urlencoded", rel_none);
1674 post_data_size = strlen (opt.post_data);
1677 post_data_size = file_size (opt.post_file_name);
1678 if (post_data_size == -1)
1680 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1681 quote (opt.post_file_name), strerror (errno));
1685 request_set_header (req, "Content-Length",
1686 xstrdup (number_to_static_string (post_data_size)),
1691 /* We need to come back here when the initial attempt to retrieve
1692 without authorization header fails. (Expected to happen at least
1693 for the Digest authorization scheme.) */
1696 request_set_header (req, "Cookie",
1697 cookie_header (wget_cookie_jar,
1698 u->host, u->port, u->path,
1700 u->scheme == SCHEME_HTTPS
1707 /* Add the user headers. */
1708 if (opt.user_headers)
1711 for (i = 0; opt.user_headers[i]; i++)
1712 request_set_user_header (req, opt.user_headers[i]);
1718 char *proxy_user, *proxy_passwd;
1719 /* For normal username and password, URL components override
1720 command-line/wgetrc parameters. With proxy
1721 authentication, it's the reverse, because proxy URLs are
1722 normally the "permanent" ones, so command-line args
1723 should take precedence. */
1724 if (opt.proxy_user && opt.proxy_passwd)
1726 proxy_user = opt.proxy_user;
1727 proxy_passwd = opt.proxy_passwd;
1731 proxy_user = proxy->user;
1732 proxy_passwd = proxy->passwd;
1734 /* #### This does not appear right. Can't the proxy request,
1735 say, `Digest' authentication? */
1736 if (proxy_user && proxy_passwd)
1737 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1739 /* If we're using a proxy, we will be connecting to the proxy
1743 /* Proxy authorization over SSL is handled below. */
1745 if (u->scheme != SCHEME_HTTPS)
1747 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1752 /* Establish the connection. */
1754 if (inhibit_keep_alive)
1758 /* Look for a persistent connection to target host, unless a
1759 proxy is used. The exception is when SSL is in use, in which
1760 case the proxy is nothing but a passthrough to the target
1761 host, registered as a connection to the latter. */
1762 struct url *relevant = conn;
1764 if (u->scheme == SCHEME_HTTPS)
1768 if (persistent_available_p (relevant->host, relevant->port,
1770 relevant->scheme == SCHEME_HTTPS,
1774 &host_lookup_failed))
1776 sock = pconn.socket;
1777 using_ssl = pconn.ssl;
1778 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1779 quotearg_style (escape_quoting_style, pconn.host),
1781 DEBUGP (("Reusing fd %d.\n", sock));
1782 if (pconn.authorized)
1783 /* If the connection is already authorized, the "Basic"
1784 authorization added by code above is unnecessary and
1786 request_remove_header (req, "Authorization");
1788 else if (host_lookup_failed)
1791 logprintf(LOG_NOTQUIET,
1792 _("%s: unable to resolve host address %s\n"),
1793 exec_name, quote (relevant->host));
1800 sock = connect_to_host (conn->host, conn->port);
1809 return (retryable_socket_connect_error (errno)
1810 ? CONERROR : CONIMPOSSIBLE);
1814 if (proxy && u->scheme == SCHEME_HTTPS)
1816 /* When requesting SSL URLs through proxies, use the
1817 CONNECT method to request passthrough. */
1818 struct request *connreq = request_new ();
1819 request_set_method (connreq, "CONNECT",
1820 aprintf ("%s:%d", u->host, u->port));
1821 SET_USER_AGENT (connreq);
1824 request_set_header (connreq, "Proxy-Authorization",
1825 proxyauth, rel_value);
1826 /* Now that PROXYAUTH is part of the CONNECT request,
1827 zero it out so we don't send proxy authorization with
1828 the regular request below. */
1831 /* Examples in rfc2817 use the Host header in CONNECT
1832 requests. I don't see how that gains anything, given
1833 that the contents of Host would be exactly the same as
1834 the contents of CONNECT. */
1836 write_error = request_send (connreq, sock);
1837 request_free (connreq);
1838 if (write_error < 0)
1840 CLOSE_INVALIDATE (sock);
1844 head = read_http_response_head (sock);
1847 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1849 CLOSE_INVALIDATE (sock);
1858 DEBUGP (("proxy responded with: [%s]\n", head));
1860 resp = resp_new (head);
1861 statcode = resp_status (resp, &message);
1864 char *tms = datetime_str (time (NULL));
1865 logprintf (LOG_VERBOSE, "%d\n", statcode);
1866 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1867 quotearg_style (escape_quoting_style,
1868 _("Malformed status line")));
1872 hs->message = xstrdup (message);
1875 if (statcode != 200)
1878 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1879 message ? quotearg_style (escape_quoting_style, message) : "?");
1880 xfree_null (message);
1883 xfree_null (message);
1885 /* SOCK is now *really* connected to u->host, so update CONN
1886 to reflect this. That way register_persistent will
1887 register SOCK as being connected to u->host:u->port. */
1891 if (conn->scheme == SCHEME_HTTPS)
1893 if (!ssl_connect_wget (sock))
1898 else if (!ssl_check_certificate (sock, u->host))
1901 return VERIFCERTERR;
1905 #endif /* HAVE_SSL */
1908 /* Send the request to server. */
1909 write_error = request_send (req, sock);
1911 if (write_error >= 0)
1915 DEBUGP (("[POST data: %s]\n", opt.post_data));
1916 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1918 else if (opt.post_file_name && post_data_size != 0)
1919 write_error = post_file (sock, opt.post_file_name, post_data_size);
1922 if (write_error < 0)
1924 CLOSE_INVALIDATE (sock);
1928 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1929 proxy ? "Proxy" : "HTTP");
1935 head = read_http_response_head (sock);
1940 logputs (LOG_NOTQUIET, _("No data received.\n"));
1941 CLOSE_INVALIDATE (sock);
1947 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1949 CLOSE_INVALIDATE (sock);
1954 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1956 resp = resp_new (head);
1958 /* Check for status line. */
1960 statcode = resp_status (resp, &message);
1963 char *tms = datetime_str (time (NULL));
1964 logprintf (LOG_VERBOSE, "%d\n", statcode);
1965 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"), tms, statcode,
1966 quotearg_style (escape_quoting_style,
1967 _("Malformed status line")));
1968 CLOSE_INVALIDATE (sock);
1973 if (H_10X (statcode))
1975 DEBUGP (("Ignoring response\n"));
1979 hs->message = xstrdup (message);
1980 if (!opt.server_response)
1981 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1982 message ? quotearg_style (escape_quoting_style, message) : "");
1985 logprintf (LOG_VERBOSE, "\n");
1986 print_server_response (resp, " ");
1989 if (!opt.ignore_length
1990 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1994 parsed = str_to_wgint (hdrval, NULL, 10);
1995 if (parsed == WGINT_MAX && errno == ERANGE)
1998 #### If Content-Length is out of range, it most likely
1999 means that the file is larger than 2G and that we're
2000 compiled without LFS. In that case we should probably
2001 refuse to even attempt to download the file. */
2004 else if (parsed < 0)
2006 /* Negative Content-Length; nonsensical, so we can't
2007 assume any information about the content to receive. */
2014 /* Check for keep-alive related responses. */
2015 if (!inhibit_keep_alive && contlen != -1)
2017 if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
2019 if (0 == strcasecmp (hdrval, "Close"))
2024 resp_header_copy (resp, "Transfer-Encoding", hdrval, sizeof (hdrval));
2025 if (0 == strcasecmp (hdrval, "chunked"))
2026 chunked_transfer_encoding = true;
2028 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2032 const char *scbeg, *scend;
2033 /* The jar should have been created by now. */
2034 assert (wget_cookie_jar != NULL);
2036 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2037 &scbeg, &scend)) != -1;
2040 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2041 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2042 u->path, set_cookie);
2047 /* The server has promised that it will not close the connection
2048 when we're done. This means that we can register it. */
2049 register_persistent (conn->host, conn->port, sock, using_ssl);
2051 if (statcode == HTTP_STATUS_UNAUTHORIZED)
2053 /* Authorization is required. */
2054 if (keep_alive && !head_only
2055 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2056 CLOSE_FINISH (sock);
2058 CLOSE_INVALIDATE (sock);
2059 pconn.authorized = false;
2060 if (!auth_finished && (user && passwd))
2062 /* IIS sends multiple copies of WWW-Authenticate, one with
2063 the value "negotiate", and other(s) with data. Loop over
2064 all the occurrences and pick the one we recognize. */
2066 const char *wabeg, *waend;
2067 char *www_authenticate = NULL;
2069 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
2070 &wabeg, &waend)) != -1;
2072 if (known_authentication_scheme_p (wabeg, waend))
2074 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
2078 if (!www_authenticate)
2080 /* If the authentication header is missing or
2081 unrecognized, there's no sense in retrying. */
2082 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
2084 else if (!basic_auth_finished
2085 || !BEGINS_WITH (www_authenticate, "Basic"))
2088 pth = url_full_path (u);
2089 request_set_header (req, "Authorization",
2090 create_authorization_line (www_authenticate,
2092 request_method (req),
2096 if (BEGINS_WITH (www_authenticate, "NTLM"))
2098 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2100 /* Need to register this host as using basic auth,
2101 * so we automatically send creds next time. */
2102 register_basic_auth_host (u->host);
2105 xfree_null (message);
2108 goto retry_with_auth;
2112 /* We already did Basic auth, and it failed. Gotta
2116 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2118 xfree_null (message);
2123 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2125 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2127 pconn.authorized = true;
2130 /* Determine the local filename if needed. Notice that if -O is used
2131 * hstat.local_file is set by http_loop to the argument of -O. */
2132 if (!hs->local_file)
2134 char *local_file = NULL;
2136 /* Honor Content-Disposition whether possible. */
2137 if (!opt.content_disposition
2138 || !resp_header_copy (resp, "Content-Disposition",
2139 hdrval, sizeof (hdrval))
2140 || !parse_content_disposition (hdrval, &local_file))
2142 /* The Content-Disposition header is missing or broken.
2143 * Choose unique file name according to given URL. */
2144 hs->local_file = url_file_name (u, NULL);
2148 DEBUGP (("Parsed filename from Content-Disposition: %s\n",
2150 hs->local_file = url_file_name (u, local_file);
2154 /* TODO: perform this check only once. */
2155 if (!hs->existence_checked && file_exists_p (hs->local_file))
2157 if (opt.noclobber && !opt.output_document)
2159 /* If opt.noclobber is turned on and file already exists, do not
2160 retrieve the file. But if the output_document was given, then this
2161 test was already done and the file didn't exist. Hence the !opt.output_document */
2162 logprintf (LOG_VERBOSE, _("\
2163 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
2164 /* If the file is there, we suppose it's retrieved OK. */
2167 /* #### Bogusness alert. */
2168 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2169 if (has_html_suffix_p (hs->local_file))
2173 xfree_null (message);
2174 return RETRUNNEEDED;
2176 else if (!ALLOW_CLOBBER)
2178 char *unique = unique_name (hs->local_file, true);
2179 if (unique != hs->local_file)
2180 xfree (hs->local_file);
2181 hs->local_file = unique;
2184 hs->existence_checked = true;
2186 /* Support timestamping */
2187 /* TODO: move this code out of gethttp. */
2188 if (opt.timestamping && !hs->timestamp_checked)
2190 size_t filename_len = strlen (hs->local_file);
2191 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (ORIG_SFX));
2192 bool local_dot_orig_file_exists = false;
2193 char *local_filename = NULL;
2196 if (opt.backup_converted)
2197 /* If -K is specified, we'll act on the assumption that it was specified
2198 last time these files were downloaded as well, and instead of just
2199 comparing local file X against server file X, we'll compare local
2200 file X.orig (if extant, else X) against server file X. If -K
2201 _wasn't_ specified last time, or the server contains files called
2202 *.orig, -N will be back to not operating correctly with -k. */
2204 /* Would a single s[n]printf() call be faster? --dan
2206 Definitely not. sprintf() is horribly slow. It's a
2207 different question whether the difference between the two
2208 affects a program. Usually I'd say "no", but at one
2209 point I profiled Wget, and found that a measurable and
2210 non-negligible amount of time was lost calling sprintf()
2211 in url.c. Replacing sprintf with inline calls to
2212 strcpy() and number_to_string() made a difference.
2214 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
2215 memcpy (filename_plus_orig_suffix + filename_len,
2216 ORIG_SFX, sizeof (ORIG_SFX));
2218 /* Try to stat() the .orig file. */
2219 if (stat (filename_plus_orig_suffix, &st) == 0)
2221 local_dot_orig_file_exists = true;
2222 local_filename = filename_plus_orig_suffix;
2226 if (!local_dot_orig_file_exists)
2227 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
2228 if (stat (hs->local_file, &st) == 0)
2229 local_filename = hs->local_file;
2231 if (local_filename != NULL)
2232 /* There was a local file, so we'll check later to see if the version
2233 the server has is the same version we already have, allowing us to
2236 hs->orig_file_name = xstrdup (local_filename);
2237 hs->orig_file_size = st.st_size;
2238 hs->orig_file_tstamp = st.st_mtime;
2240 /* Modification time granularity is 2 seconds for Windows, so
2241 increase local time by 1 second for later comparison. */
2242 ++hs->orig_file_tstamp;
2249 hs->statcode = statcode;
2251 hs->error = xstrdup (_("Malformed status line"));
2253 hs->error = xstrdup (_("(no description)"));
2255 hs->error = xstrdup (message);
2256 xfree_null (message);
2258 type = resp_header_strdup (resp, "Content-Type");
2261 char *tmp = strchr (type, ';');
2264 /* sXXXav: only needed if IRI support is enabled */
2265 char *tmp2 = tmp + 1;
2267 while (tmp > type && c_isspace (tmp[-1]))
2271 /* Try to get remote encoding if needed */
2272 if (opt.enable_iri && !opt.encoding_remote)
2274 tmp = parse_charset (tmp2);
2276 set_content_encoding (iri, tmp);
2280 hs->newloc = resp_header_strdup (resp, "Location");
2281 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2283 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2285 wgint first_byte_pos, last_byte_pos, entity_length;
2286 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2289 contrange = first_byte_pos;
2290 contlen = last_byte_pos - first_byte_pos + 1;
2295 /* 20x responses are counted among successful by default. */
2296 if (H_20X (statcode))
2299 /* Return if redirected. */
2300 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2302 /* RFC2068 says that in case of the 300 (multiple choices)
2303 response, the server can output a preferred URL through
2304 `Location' header; otherwise, the request should be treated
2305 like GET. So, if the location is set, it will be a
2306 redirection; otherwise, just proceed normally. */
2307 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2311 logprintf (LOG_VERBOSE,
2312 _("Location: %s%s\n"),
2313 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2314 hs->newloc ? _(" [following]") : "");
2315 if (keep_alive && !head_only
2316 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2317 CLOSE_FINISH (sock);
2319 CLOSE_INVALIDATE (sock);
2326 /* If content-type is not given, assume text/html. This is because
2327 of the multitude of broken CGI's that "forget" to generate the
2330 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2331 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2337 0 == strncasecmp (type, TEXTCSS_S, strlen (TEXTCSS_S)))
2342 if (opt.adjust_extension)
2345 /* -E / --adjust-extension / adjust_extension = on was specified,
2346 and this is a text/html file. If some case-insensitive
2347 variation on ".htm[l]" isn't already the file's suffix,
2350 ensure_extension (hs, ".html", dt);
2352 else if (*dt & TEXTCSS)
2354 ensure_extension (hs, ".css", dt);
2358 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE
2359 || (!opt.timestamping && hs->restval > 0 && statcode == HTTP_STATUS_OK
2360 && contrange == 0 && contlen >= 0 && hs->restval >= contlen))
2362 /* If `-c' is in use and the file has been fully downloaded (or
2363 the remote file has shrunk), Wget effectively requests bytes
2364 after the end of file and the server response with 416
2365 (or 200 with a <= Content-Length. */
2366 logputs (LOG_VERBOSE, _("\
2367 \n The file is already fully retrieved; nothing to do.\n\n"));
2368 /* In case the caller inspects. */
2371 /* Mark as successfully retrieved. */
2374 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2375 might be more bytes in the body. */
2377 return RETRUNNEEDED;
2379 if ((contrange != 0 && contrange != hs->restval)
2380 || (H_PARTIAL (statcode) && !contrange))
2382 /* The Range request was somehow misunderstood by the server.
2385 CLOSE_INVALIDATE (sock);
2392 hs->contlen = contlen + contrange;
2398 /* No need to print this output if the body won't be
2399 downloaded at all, or if the original server response is
2401 logputs (LOG_VERBOSE, _("Length: "));
2404 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2405 if (contlen + contrange >= 1024)
2406 logprintf (LOG_VERBOSE, " (%s)",
2407 human_readable (contlen + contrange));
2410 if (contlen >= 1024)
2411 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2412 number_to_static_string (contlen),
2413 human_readable (contlen));
2415 logprintf (LOG_VERBOSE, _(", %s remaining"),
2416 number_to_static_string (contlen));
2420 logputs (LOG_VERBOSE,
2421 opt.ignore_length ? _("ignored") : _("unspecified"));
2423 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2425 logputs (LOG_VERBOSE, "\n");
2429 type = NULL; /* We don't need it any more. */
2431 /* Return if we have no intention of further downloading. */
2432 if (!(*dt & RETROKF) || head_only)
2434 /* In case the caller cares to look... */
2439 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2440 servers not to send body in response to a HEAD request, and
2441 those that do will likely be caught by test_socket_open.
2442 If not, they can be worked around using
2443 `--no-http-keep-alive'. */
2444 CLOSE_FINISH (sock);
2446 && skip_short_body (sock, contlen, chunked_transfer_encoding))
2447 /* Successfully skipped the body; also keep using the socket. */
2448 CLOSE_FINISH (sock);
2450 CLOSE_INVALIDATE (sock);
2452 return RETRFINISHED;
2456 For VMS, define common fopen() optional arguments.
2459 # define FOPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
2460 # define FOPEN_BIN_FLAG 3
2461 #else /* def __VMS */
2462 # define FOPEN_BIN_FLAG true
2463 #endif /* def __VMS [else] */
2465 /* Open the local file. */
2468 mkalldirs (hs->local_file);
2470 rotate_backups (hs->local_file);
2477 fp = fopen (hs->local_file, "ab", FOPEN_OPT_ARGS);
2478 #else /* def __VMS */
2479 fp = fopen (hs->local_file, "ab");
2480 #endif /* def __VMS [else] */
2482 else if (ALLOW_CLOBBER)
2484 if (opt.unlink && file_exists_p (hs->local_file))
2486 int res = unlink (hs->local_file);
2489 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file,
2491 CLOSE_INVALIDATE (sock);
2501 fp = fopen (hs->local_file, "wb", FOPEN_OPT_ARGS);
2502 #else /* def __VMS */
2503 fp = fopen (hs->local_file, "wb");
2504 #endif /* def __VMS [else] */
2508 fp = fopen_excl (hs->local_file, FOPEN_BIN_FLAG);
2509 if (!fp && errno == EEXIST)
2511 /* We cannot just invent a new name and use it (which is
2512 what functions like unique_create typically do)
2513 because we told the user we'd use this name.
2514 Instead, return and retry the download. */
2515 logprintf (LOG_NOTQUIET,
2516 _("%s has sprung into existence.\n"),
2518 CLOSE_INVALIDATE (sock);
2520 return FOPEN_EXCL_ERR;
2525 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2526 CLOSE_INVALIDATE (sock);
2534 /* Print fetch message, if opt.verbose. */
2537 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2538 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2541 /* This confuses the timestamping code that checks for file size.
2542 #### The timestamping code should be smarter about file size. */
2543 if (opt.save_headers && hs->restval == 0)
2544 fwrite (head, 1, strlen (head), fp);
2546 /* Now we no longer need to store the response header. */
2549 /* Download the request body. */
2552 /* If content-length is present, read that much; otherwise, read
2553 until EOF. The HTTP spec doesn't require the server to
2554 actually close the connection when it's done sending data. */
2555 flags |= rb_read_exactly;
2556 if (hs->restval > 0 && contrange == 0)
2557 /* If the server ignored our range request, instruct fd_read_body
2558 to skip the first RESTVAL bytes of body. */
2559 flags |= rb_skip_startpos;
2561 if (chunked_transfer_encoding)
2562 flags |= rb_chunked_transfer_encoding;
2564 hs->len = hs->restval;
2566 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2567 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2571 CLOSE_FINISH (sock);
2575 hs->rderrmsg = xstrdup (fd_errstr (sock));
2576 CLOSE_INVALIDATE (sock);
2583 return RETRFINISHED;
2586 /* The genuine HTTP loop! This is the part where the retrieval is
2587 retried, and retried, and retried, and... */
2589 http_loop (struct url *u, struct url *original_url, char **newloc,
2590 char **local_file, const char *referer, int *dt, struct url *proxy,
2594 bool got_head = false; /* used for time-stamping and filename detection */
2595 bool time_came_from_head = false;
2596 bool got_name = false;
2599 uerr_t err, ret = TRYLIMEXC;
2600 time_t tmr = -1; /* remote time-stamp */
2601 struct http_stat hstat; /* HTTP status */
2603 bool send_head_first = true;
2605 bool force_full_retrieve = false;
2607 /* Assert that no value for *LOCAL_FILE was passed. */
2608 assert (local_file == NULL || *local_file == NULL);
2610 /* Set LOCAL_FILE parameter. */
2611 if (local_file && opt.output_document)
2612 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2614 /* Reset NEWLOC parameter. */
2617 /* This used to be done in main(), but it's a better idea to do it
2618 here so that we don't go through the hoops if we're just using
2623 /* Warn on (likely bogus) wildcard usage in HTTP. */
2624 if (opt.ftp_glob && has_wildcards_p (u->path))
2625 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2627 /* Setup hstat struct. */
2629 hstat.referer = referer;
2631 if (opt.output_document)
2633 hstat.local_file = xstrdup (opt.output_document);
2636 else if (!opt.content_disposition)
2639 url_file_name (opt.trustservernames ? u : original_url, NULL);
2643 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2644 * screaming for some refactoring. */
2645 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2647 /* If opt.noclobber is turned on and file already exists, do not
2648 retrieve the file. But if the output_document was given, then this
2649 test was already done and the file didn't exist. Hence the !opt.output_document */
2650 logprintf (LOG_VERBOSE, _("\
2651 File %s already there; not retrieving.\n\n"),
2652 quote (hstat.local_file));
2653 /* If the file is there, we suppose it's retrieved OK. */
2656 /* #### Bogusness alert. */
2657 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2658 if (has_html_suffix_p (hstat.local_file))
2665 /* Reset the counter. */
2668 /* Reset the document type. */
2671 /* Skip preliminary HEAD request if we're not in spider mode. */
2673 send_head_first = false;
2675 /* Send preliminary HEAD request if -N is given and we have an existing
2676 * destination file. */
2677 file_name = url_file_name (opt.trustservernames ? u : original_url, NULL);
2678 if (opt.timestamping && (file_exists_p (file_name)
2679 || opt.content_disposition))
2680 send_head_first = true;
2686 /* Increment the pass counter. */
2688 sleep_between_retrievals (count);
2690 /* Get the current time string. */
2691 tms = datetime_str (time (NULL));
2693 if (opt.spider && !got_head)
2694 logprintf (LOG_VERBOSE, _("\
2695 Spider mode enabled. Check if remote file exists.\n"));
2697 /* Print fetch message, if opt.verbose. */
2700 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2705 sprintf (tmp, _("(try:%2d)"), count);
2706 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2711 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2716 ws_changetitle (hurl);
2721 /* Default document type is empty. However, if spider mode is
2722 on or time-stamping is employed, HEAD_ONLY commands is
2723 encoded within *dt. */
2724 if (send_head_first && !got_head)
2729 /* Decide whether or not to restart. */
2730 if (force_full_retrieve)
2731 hstat.restval = hstat.len;
2732 else if (opt.always_rest
2734 && stat (hstat.local_file, &st) == 0
2735 && S_ISREG (st.st_mode))
2736 /* When -c is used, continue from on-disk size. (Can't use
2737 hstat.len even if count>1 because we don't want a failed
2738 first attempt to clobber existing data.) */
2739 hstat.restval = st.st_size;
2741 /* otherwise, continue where the previous try left off */
2742 hstat.restval = hstat.len;
2746 /* Decide whether to send the no-cache directive. We send it in
2748 a) we're using a proxy, and we're past our first retrieval.
2749 Some proxies are notorious for caching incomplete data, so
2750 we require a fresh get.
2751 b) caching is explicitly inhibited. */
2752 if ((proxy && count > 1) /* a */
2753 || !opt.allow_cache) /* b */
2754 *dt |= SEND_NOCACHE;
2756 *dt &= ~SEND_NOCACHE;
2758 /* Try fetching the document, or at least its head. */
2759 err = gethttp (u, &hstat, dt, proxy, iri);
2762 tms = datetime_str (time (NULL));
2764 /* Get the new location (with or without the redirection). */
2766 *newloc = xstrdup (hstat.newloc);
2770 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2771 case CONERROR: case READERR: case WRITEFAILED:
2772 case RANGEERR: case FOPEN_EXCL_ERR:
2773 /* Non-fatal errors continue executing the loop, which will
2774 bring them to "while" statement at the end, to judge
2775 whether the number of tries was exceeded. */
2776 printwhat (count, opt.ntry);
2778 case FWRITEERR: case FOPENERR:
2779 /* Another fatal error. */
2780 logputs (LOG_VERBOSE, "\n");
2781 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2782 quote (hstat.local_file), strerror (errno));
2783 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2784 case SSLINITFAILED: case CONTNOTSUPPORTED: case VERIFCERTERR:
2785 /* Fatal errors just return from the function. */
2789 /* Another fatal error. */
2790 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2794 /* Another fatal error. */
2795 logputs (LOG_VERBOSE, "\n");
2796 logprintf (LOG_NOTQUIET, _("Cannot unlink %s (%s).\n"),
2797 quote (hstat.local_file), strerror (errno));
2801 /* Return the new location to the caller. */
2804 logprintf (LOG_NOTQUIET,
2805 _("ERROR: Redirection (%d) without location.\n"),
2815 /* The file was already fully retrieved. */
2819 /* Deal with you later. */
2822 /* All possibilities should have been exhausted. */
2826 if (!(*dt & RETROKF))
2831 /* #### Ugly ugly ugly! */
2832 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2833 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2836 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2838 && (hstat.statcode == 500 || hstat.statcode == 501))
2843 /* Maybe we should always keep track of broken links, not just in
2845 * Don't log error if it was UTF-8 encoded because we will try
2846 * once unencoded. */
2847 else if (opt.spider && !iri->utf8_encode)
2849 /* #### Again: ugly ugly ugly! */
2851 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2852 nonexisting_url (hurl);
2853 logprintf (LOG_NOTQUIET, _("\
2854 Remote file does not exist -- broken link!!!\n"));
2858 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2859 tms, hstat.statcode,
2860 quotearg_style (escape_quoting_style, hstat.error));
2862 logputs (LOG_VERBOSE, "\n");
2868 /* Did we get the time-stamp? */
2871 got_head = true; /* no more time-stamping */
2873 if (opt.timestamping && !hstat.remote_time)
2875 logputs (LOG_NOTQUIET, _("\
2876 Last-modified header missing -- time-stamps turned off.\n"));
2878 else if (hstat.remote_time)
2880 /* Convert the date-string into struct tm. */
2881 tmr = http_atotm (hstat.remote_time);
2882 if (tmr == (time_t) (-1))
2883 logputs (LOG_VERBOSE, _("\
2884 Last-modified header invalid -- time-stamp ignored.\n"));
2885 if (*dt & HEAD_ONLY)
2886 time_came_from_head = true;
2889 if (send_head_first)
2891 /* The time-stamping section. */
2892 if (opt.timestamping)
2894 if (hstat.orig_file_name) /* Perform the following
2895 checks only if the file
2897 download already exists. */
2899 if (hstat.remote_time &&
2900 tmr != (time_t) (-1))
2902 /* Now time-stamping can be used validly.
2903 Time-stamping means that if the sizes of
2904 the local and remote file match, and local
2905 file is newer than the remote file, it will
2906 not be retrieved. Otherwise, the normal
2907 download procedure is resumed. */
2908 if (hstat.orig_file_tstamp >= tmr)
2910 if (hstat.contlen == -1
2911 || hstat.orig_file_size == hstat.contlen)
2913 logprintf (LOG_VERBOSE, _("\
2914 Server file no newer than local file %s -- not retrieving.\n\n"),
2915 quote (hstat.orig_file_name));
2921 logprintf (LOG_VERBOSE, _("\
2922 The sizes do not match (local %s) -- retrieving.\n"),
2923 number_to_static_string (hstat.orig_file_size));
2928 force_full_retrieve = true;
2929 logputs (LOG_VERBOSE,
2930 _("Remote file is newer, retrieving.\n"));
2933 logputs (LOG_VERBOSE, "\n");
2937 /* free_hstat (&hstat); */
2938 hstat.timestamp_checked = true;
2943 bool finished = true;
2948 logputs (LOG_VERBOSE, _("\
2949 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2954 logprintf (LOG_VERBOSE, _("\
2955 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2956 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2963 logprintf (LOG_VERBOSE, _("\
2964 Remote file exists and could contain further links,\n\
2965 but recursion is disabled -- not retrieving.\n\n"));
2969 logprintf (LOG_VERBOSE, _("\
2970 Remote file exists.\n\n"));
2972 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2977 logprintf (LOG_NONVERBOSE,
2978 _("%s URL: %s %2d %s\n"),
2979 tms, u->url, hstat.statcode,
2980 hstat.message ? quotearg_style (escape_quoting_style, hstat.message) : "");
2987 count = 0; /* the retrieve count for HEAD is reset */
2989 } /* send_head_first */
2992 if (opt.useservertimestamps
2993 && (tmr != (time_t) (-1))
2994 && ((hstat.len == hstat.contlen) ||
2995 ((hstat.res == 0) && (hstat.contlen == -1))))
2997 const char *fl = NULL;
2998 set_local_file (&fl, hstat.local_file);
3002 /* Reparse time header, in case it's changed. */
3003 if (time_came_from_head
3004 && hstat.remote_time && hstat.remote_time[0])
3006 newtmr = http_atotm (hstat.remote_time);
3007 if (newtmr != (time_t)-1)
3013 /* End of time-stamping section. */
3015 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
3016 total_download_time += hstat.dltime;
3018 if (hstat.len == hstat.contlen)
3022 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3024 logprintf (LOG_VERBOSE,
3026 ? _("%s (%s) - written to stdout %s[%s/%s]\n\n")
3027 : _("%s (%s) - %s saved [%s/%s]\n\n"),
3029 write_to_stdout ? "" : quote (hstat.local_file),
3030 number_to_static_string (hstat.len),
3031 number_to_static_string (hstat.contlen));
3032 logprintf (LOG_NONVERBOSE,
3033 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
3035 number_to_static_string (hstat.len),
3036 number_to_static_string (hstat.contlen),
3037 hstat.local_file, count);
3040 total_downloaded_bytes += hstat.rd_size;
3042 /* Remember that we downloaded the file for later ".orig" code. */
3043 if (*dt & ADDED_HTML_EXTENSION)
3044 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3046 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3051 else if (hstat.res == 0) /* No read error */
3053 if (hstat.contlen == -1) /* We don't know how much we were supposed
3054 to get, so assume we succeeded. */
3058 bool write_to_stdout = (opt.output_document && HYPHENP (opt.output_document));
3060 logprintf (LOG_VERBOSE,
3062 ? _("%s (%s) - written to stdout %s[%s]\n\n")
3063 : _("%s (%s) - %s saved [%s]\n\n"),
3065 write_to_stdout ? "" : quote (hstat.local_file),
3066 number_to_static_string (hstat.len));
3067 logprintf (LOG_NONVERBOSE,
3068 "%s URL:%s [%s] -> \"%s\" [%d]\n",
3069 tms, u->url, number_to_static_string (hstat.len),
3070 hstat.local_file, count);
3073 total_downloaded_bytes += hstat.rd_size;
3075 /* Remember that we downloaded the file for later ".orig" code. */
3076 if (*dt & ADDED_HTML_EXTENSION)
3077 downloaded_file (FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
3079 downloaded_file (FILE_DOWNLOADED_NORMALLY, hstat.local_file);
3084 else if (hstat.len < hstat.contlen) /* meaning we lost the
3085 connection too soon */
3087 logprintf (LOG_VERBOSE,
3088 _("%s (%s) - Connection closed at byte %s. "),
3089 tms, tmrate, number_to_static_string (hstat.len));
3090 printwhat (count, opt.ntry);
3093 else if (hstat.len != hstat.restval)
3094 /* Getting here would mean reading more data than
3095 requested with content-length, which we never do. */
3099 /* Getting here probably means that the content-length was
3100 * _less_ than the original, local size. We should probably
3101 * truncate or re-read, or something. FIXME */
3106 else /* from now on hstat.res can only be -1 */
3108 if (hstat.contlen == -1)
3110 logprintf (LOG_VERBOSE,
3111 _("%s (%s) - Read error at byte %s (%s)."),
3112 tms, tmrate, number_to_static_string (hstat.len),
3114 printwhat (count, opt.ntry);
3117 else /* hstat.res == -1 and contlen is given */
3119 logprintf (LOG_VERBOSE,
3120 _("%s (%s) - Read error at byte %s/%s (%s). "),
3122 number_to_static_string (hstat.len),
3123 number_to_static_string (hstat.contlen),
3125 printwhat (count, opt.ntry);
3131 while (!opt.ntry || (count < opt.ntry));
3134 if (ret == RETROK && local_file)
3135 *local_file = xstrdup (hstat.local_file);
3136 free_hstat (&hstat);
3141 /* Check whether the result of strptime() indicates success.
3142 strptime() returns the pointer to how far it got to in the string.
3143 The processing has been successful if the string is at `GMT' or
3144 `+X', or at the end of the string.
3146 In extended regexp parlance, the function returns 1 if P matches
3147 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
3148 can return) is considered a failure and 0 is returned. */
3150 check_end (const char *p)
3154 while (c_isspace (*p))
3157 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
3158 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
3164 /* Convert the textual specification of time in TIME_STRING to the
3165 number of seconds since the Epoch.
3167 TIME_STRING can be in any of the three formats RFC2616 allows the
3168 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
3169 as well as the time format used in the Set-Cookie header.
3170 Timezones are ignored, and should be GMT.
3172 Return the computed time_t representation, or -1 if the conversion
3175 This function uses strptime with various string formats for parsing
3176 TIME_STRING. This results in a parser that is not as lenient in
3177 interpreting TIME_STRING as I would like it to be. Being based on
3178 strptime, it always allows shortened months, one-digit days, etc.,
3179 but due to the multitude of formats in which time can be
3180 represented, an ideal HTTP time parser would be even more
3181 forgiving. It should completely ignore things like week days and
3182 concentrate only on the various forms of representing years,
3183 months, days, hours, minutes, and seconds. For example, it would
3184 be nice if it accepted ISO 8601 out of the box.
3186 I've investigated free and PD code for this purpose, but none was
3187 usable. getdate was big and unwieldy, and had potential copyright
3188 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
3189 distributed with phttpd, is excellent, but we cannot use it because
3190 it is not assigned to the FSF. So I stuck it with strptime. */
3193 http_atotm (const char *time_string)
3195 /* NOTE: Solaris strptime man page claims that %n and %t match white
3196 space, but that's not universally available. Instead, we simply
3197 use ` ' to mean "skip all WS", which works under all strptime
3198 implementations I've tested. */
3200 static const char *time_formats[] = {
3201 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
3202 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
3203 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
3204 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
3205 (used in Set-Cookie, defined in the
3206 Netscape cookie specification.) */
3208 const char *oldlocale;
3209 char savedlocale[256];
3211 time_t ret = (time_t) -1;
3213 /* Solaris strptime fails to recognize English month names in
3214 non-English locales, which we work around by temporarily setting
3215 locale to C before invoking strptime. */
3216 oldlocale = setlocale (LC_TIME, NULL);
3219 size_t l = strlen (oldlocale) + 1;
3220 if (l >= sizeof savedlocale)
3221 savedlocale[0] = '\0';
3223 memcpy (savedlocale, oldlocale, l);
3225 else savedlocale[0] = '\0';
3227 setlocale (LC_TIME, "C");
3229 for (i = 0; i < countof (time_formats); i++)
3233 /* Some versions of strptime use the existing contents of struct
3234 tm to recalculate the date according to format. Zero it out
3235 to prevent stack garbage from influencing strptime. */
3238 if (check_end (strptime (time_string, time_formats[i], &t)))
3245 /* Restore the previous locale. */
3247 setlocale (LC_TIME, savedlocale);
3252 /* Authorization support: We support three authorization schemes:
3254 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
3256 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
3257 consisting of answering to the server's challenge with the proper
3260 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
3261 Stenberg for libcurl. Like digest, NTLM is based on a
3262 challenge-response mechanism, but unlike digest, it is non-standard
3263 (authenticates TCP connections rather than requests), undocumented
3264 and Microsoft-specific. */
3266 /* Create the authentication header contents for the `Basic' scheme.
3267 This is done by encoding the string "USER:PASS" to base64 and
3268 prepending the string "Basic " in front of it. */
3271 basic_authentication_encode (const char *user, const char *passwd)
3274 int len1 = strlen (user) + 1 + strlen (passwd);
3276 t1 = (char *)alloca (len1 + 1);
3277 sprintf (t1, "%s:%s", user, passwd);
3279 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
3280 base64_encode (t1, len1, t2);
3282 return concat_strings ("Basic ", t2, (char *) 0);
3285 #define SKIP_WS(x) do { \
3286 while (c_isspace (*(x))) \
3290 #ifdef ENABLE_DIGEST
3291 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3292 an array of 16 bytes containing the hash keys, and BUF should be a
3293 buffer of 33 writable characters (32 for hex digits plus one for
3294 zero termination). */
3296 dump_hash (char *buf, const unsigned char *hash)
3300 for (i = 0; i < MD5_DIGEST_SIZE; i++, hash++)
3302 *buf++ = XNUM_TO_digit (*hash >> 4);
3303 *buf++ = XNUM_TO_digit (*hash & 0xf);
3308 /* Take the line apart to find the challenge, and compose a digest
3309 authorization header. See RFC2069 section 2.1.2. */
3311 digest_authentication_encode (const char *au, const char *user,
3312 const char *passwd, const char *method,
3315 static char *realm, *opaque, *nonce;
3320 { "realm", &realm },
3321 { "opaque", &opaque },
3325 param_token name, value;
3327 realm = opaque = nonce = NULL;
3329 au += 6; /* skip over `Digest' */
3330 while (extract_param (&au, &name, &value, ','))
3333 size_t namelen = name.e - name.b;
3334 for (i = 0; i < countof (options); i++)
3335 if (namelen == strlen (options[i].name)
3336 && 0 == strncmp (name.b, options[i].name,
3339 *options[i].variable = strdupdelim (value.b, value.e);
3343 if (!realm || !nonce || !user || !passwd || !path || !method)
3346 xfree_null (opaque);
3351 /* Calculate the digest value. */
3354 unsigned char hash[MD5_DIGEST_SIZE];
3355 char a1buf[MD5_DIGEST_SIZE * 2 + 1], a2buf[MD5_DIGEST_SIZE * 2 + 1];
3356 char response_digest[MD5_DIGEST_SIZE * 2 + 1];
3358 /* A1BUF = H(user ":" realm ":" password) */
3359 md5_init_ctx (&ctx);
3360 md5_process_bytes ((unsigned char *)user, strlen (user), &ctx);
3361 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3362 md5_process_bytes ((unsigned char *)realm, strlen (realm), &ctx);
3363 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3364 md5_process_bytes ((unsigned char *)passwd, strlen (passwd), &ctx);
3365 md5_finish_ctx (&ctx, hash);
3366 dump_hash (a1buf, hash);
3368 /* A2BUF = H(method ":" path) */
3369 md5_init_ctx (&ctx);
3370 md5_process_bytes ((unsigned char *)method, strlen (method), &ctx);
3371 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3372 md5_process_bytes ((unsigned char *)path, strlen (path), &ctx);
3373 md5_finish_ctx (&ctx, hash);
3374 dump_hash (a2buf, hash);
3376 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3377 md5_init_ctx (&ctx);
3378 md5_process_bytes ((unsigned char *)a1buf, MD5_DIGEST_SIZE * 2, &ctx);
3379 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3380 md5_process_bytes ((unsigned char *)nonce, strlen (nonce), &ctx);
3381 md5_process_bytes ((unsigned char *)":", 1, &ctx);
3382 md5_process_bytes ((unsigned char *)a2buf, MD5_DIGEST_SIZE * 2, &ctx);
3383 md5_finish_ctx (&ctx, hash);
3384 dump_hash (response_digest, hash);
3386 res = xmalloc (strlen (user)
3391 + 2 * MD5_DIGEST_SIZE /*strlen (response_digest)*/
3392 + (opaque ? strlen (opaque) : 0)
3394 sprintf (res, "Digest \
3395 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3396 user, realm, nonce, path, response_digest);
3399 char *p = res + strlen (res);
3400 strcat (p, ", opaque=\"");
3407 #endif /* ENABLE_DIGEST */
3409 /* Computing the size of a string literal must take into account that
3410 value returned by sizeof includes the terminating \0. */
3411 #define STRSIZE(literal) (sizeof (literal) - 1)
3413 /* Whether chars in [b, e) begin with the literal string provided as
3414 first argument and are followed by whitespace or terminating \0.
3415 The comparison is case-insensitive. */
3416 #define STARTS(literal, b, e) \
3418 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3419 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3420 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3421 || c_isspace (b[STRSIZE (literal)])))
3424 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3426 return STARTS ("Basic", hdrbeg, hdrend)
3427 #ifdef ENABLE_DIGEST
3428 || STARTS ("Digest", hdrbeg, hdrend)
3431 || STARTS ("NTLM", hdrbeg, hdrend)
3438 /* Create the HTTP authorization request header. When the
3439 `WWW-Authenticate' response header is seen, according to the
3440 authorization scheme specified in that header (`Basic' and `Digest'
3441 are supported by the current implementation), produce an
3442 appropriate HTTP authorization request header. */
3444 create_authorization_line (const char *au, const char *user,
3445 const char *passwd, const char *method,
3446 const char *path, bool *finished)
3448 /* We are called only with known schemes, so we can dispatch on the
3450 switch (c_toupper (*au))
3452 case 'B': /* Basic */
3454 return basic_authentication_encode (user, passwd);
3455 #ifdef ENABLE_DIGEST
3456 case 'D': /* Digest */
3458 return digest_authentication_encode (au, user, passwd, method, path);
3461 case 'N': /* NTLM */
3462 if (!ntlm_input (&pconn.ntlm, au))
3467 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3470 /* We shouldn't get here -- this function should be only called
3471 with values approved by known_authentication_scheme_p. */
3479 if (!wget_cookie_jar)
3480 wget_cookie_jar = cookie_jar_new ();
3481 if (opt.cookies_input && !cookies_loaded_p)
3483 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3484 cookies_loaded_p = true;
3491 if (wget_cookie_jar)
3492 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3498 xfree_null (pconn.host);
3499 if (wget_cookie_jar)
3500 cookie_jar_delete (wget_cookie_jar);
3504 ensure_extension (struct http_stat *hs, const char *ext, int *dt)
3506 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
3508 int len = strlen (ext);
3511 strncpy (shortext, ext, len - 1);
3512 shortext[len - 2] = '\0';
3515 if (last_period_in_local_filename == NULL
3516 || !(0 == strcasecmp (last_period_in_local_filename, shortext)
3517 || 0 == strcasecmp (last_period_in_local_filename, ext)))
3519 int local_filename_len = strlen (hs->local_file);
3520 /* Resize the local file, allowing for ".html" preceded by
3521 optional ".NUMBER". */
3522 hs->local_file = xrealloc (hs->local_file,
3523 local_filename_len + 24 + len);
3524 strcpy (hs->local_file + local_filename_len, ext);
3525 /* If clobbering is not allowed and the file, as named,
3526 exists, tack on ".NUMBER.html" instead. */
3527 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
3531 sprintf (hs->local_file + local_filename_len,
3532 ".%d%s", ext_num++, ext);
3533 while (file_exists_p (hs->local_file));
3535 *dt |= ADDED_HTML_EXTENSION;
3543 test_parse_content_disposition()
3551 { "filename=\"file.ext\"", "file.ext", true },
3552 { "attachment; filename=\"file.ext\"", "file.ext", true },
3553 { "attachment; filename=\"file.ext\"; dummy", "file.ext", true },
3554 { "attachment", NULL, false },
3555 { "attachement; filename*=UTF-8'en-US'hello.txt", "hello.txt", true },
3556 { "attachement; filename*0=\"hello\"; filename*1=\"world.txt\"", "helloworld.txt", true },
3559 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3564 res = parse_content_disposition (test_array[i].hdrval, &filename);
3566 mu_assert ("test_parse_content_disposition: wrong result",
3567 res == test_array[i].result
3569 || 0 == strcmp (test_array[i].filename, filename)));
3575 #endif /* TESTING */
3578 * vim: et sts=2 sw=2 cino+={s