2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003,
3 2004, 2005, 2006, 2007, 2008 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 extern char *version_string;
72 static char *create_authorization_line (const char *, const char *,
73 const char *, const char *,
74 const char *, bool *);
75 static char *basic_authentication_encode (const char *, const char *);
76 static bool known_authentication_scheme_p (const char *, const char *);
77 static void load_cookies (void);
80 # define MIN(x, y) ((x) > (y) ? (y) : (x))
84 static bool cookies_loaded_p;
85 static struct cookie_jar *wget_cookie_jar;
87 #define TEXTHTML_S "text/html"
88 #define TEXTXHTML_S "application/xhtml+xml"
90 /* Some status code validation macros: */
91 #define H_20X(x) (((x) >= 200) && ((x) < 300))
92 #define H_PARTIAL(x) ((x) == HTTP_STATUS_PARTIAL_CONTENTS)
93 #define H_REDIRECTED(x) ((x) == HTTP_STATUS_MOVED_PERMANENTLY \
94 || (x) == HTTP_STATUS_MOVED_TEMPORARILY \
95 || (x) == HTTP_STATUS_SEE_OTHER \
96 || (x) == HTTP_STATUS_TEMPORARY_REDIRECT)
98 /* HTTP/1.0 status codes from RFC1945, provided for reference. */
100 #define HTTP_STATUS_OK 200
101 #define HTTP_STATUS_CREATED 201
102 #define HTTP_STATUS_ACCEPTED 202
103 #define HTTP_STATUS_NO_CONTENT 204
104 #define HTTP_STATUS_PARTIAL_CONTENTS 206
106 /* Redirection 3xx. */
107 #define HTTP_STATUS_MULTIPLE_CHOICES 300
108 #define HTTP_STATUS_MOVED_PERMANENTLY 301
109 #define HTTP_STATUS_MOVED_TEMPORARILY 302
110 #define HTTP_STATUS_SEE_OTHER 303 /* from HTTP/1.1 */
111 #define HTTP_STATUS_NOT_MODIFIED 304
112 #define HTTP_STATUS_TEMPORARY_REDIRECT 307 /* from HTTP/1.1 */
114 /* Client error 4xx. */
115 #define HTTP_STATUS_BAD_REQUEST 400
116 #define HTTP_STATUS_UNAUTHORIZED 401
117 #define HTTP_STATUS_FORBIDDEN 403
118 #define HTTP_STATUS_NOT_FOUND 404
119 #define HTTP_STATUS_RANGE_NOT_SATISFIABLE 416
121 /* Server errors 5xx. */
122 #define HTTP_STATUS_INTERNAL 500
123 #define HTTP_STATUS_NOT_IMPLEMENTED 501
124 #define HTTP_STATUS_BAD_GATEWAY 502
125 #define HTTP_STATUS_UNAVAILABLE 503
128 rel_none, rel_name, rel_value, rel_both
135 struct request_header {
137 enum rp release_policy;
139 int hcount, hcapacity;
142 /* Create a new, empty request. At least request_set_method must be
143 called before the request can be used. */
145 static struct request *
148 struct request *req = xnew0 (struct request);
150 req->headers = xnew_array (struct request_header, req->hcapacity);
154 /* Set the request's method and its arguments. METH should be a
155 literal string (or it should outlive the request) because it will
156 not be freed. ARG will be freed by request_free. */
159 request_set_method (struct request *req, const char *meth, char *arg)
165 /* Return the method string passed with the last call to
166 request_set_method. */
169 request_method (const struct request *req)
174 /* Free one header according to the release policy specified with
175 request_set_header. */
178 release_header (struct request_header *hdr)
180 switch (hdr->release_policy)
197 /* Set the request named NAME to VALUE. Specifically, this means that
198 a "NAME: VALUE\r\n" header line will be used in the request. If a
199 header with the same name previously existed in the request, its
200 value will be replaced by this one. A NULL value means do nothing.
202 RELEASE_POLICY determines whether NAME and VALUE should be released
203 (freed) with request_free. Allowed values are:
205 - rel_none - don't free NAME or VALUE
206 - rel_name - free NAME when done
207 - rel_value - free VALUE when done
208 - rel_both - free both NAME and VALUE when done
210 Setting release policy is useful when arguments come from different
211 sources. For example:
213 // Don't free literal strings!
214 request_set_header (req, "Pragma", "no-cache", rel_none);
216 // Don't free a global variable, we'll need it later.
217 request_set_header (req, "Referer", opt.referer, rel_none);
219 // Value freshly allocated, free it when done.
220 request_set_header (req, "Range",
221 aprintf ("bytes=%s-", number_to_static_string (hs->restval)),
226 request_set_header (struct request *req, char *name, char *value,
227 enum rp release_policy)
229 struct request_header *hdr;
234 /* A NULL value is a no-op; if freeing the name is requested,
235 free it now to avoid leaks. */
236 if (release_policy == rel_name || release_policy == rel_both)
241 for (i = 0; i < req->hcount; i++)
243 hdr = &req->headers[i];
244 if (0 == strcasecmp (name, hdr->name))
246 /* Replace existing header. */
247 release_header (hdr);
250 hdr->release_policy = release_policy;
255 /* Install new header. */
257 if (req->hcount >= req->hcapacity)
259 req->hcapacity <<= 1;
260 req->headers = xrealloc (req->headers, req->hcapacity * sizeof (*hdr));
262 hdr = &req->headers[req->hcount++];
265 hdr->release_policy = release_policy;
268 /* Like request_set_header, but sets the whole header line, as
269 provided by the user using the `--header' option. For example,
270 request_set_user_header (req, "Foo: bar") works just like
271 request_set_header (req, "Foo", "bar"). */
274 request_set_user_header (struct request *req, const char *header)
277 const char *p = strchr (header, ':');
280 BOUNDED_TO_ALLOCA (header, p, name);
282 while (c_isspace (*p))
284 request_set_header (req, xstrdup (name), (char *) p, rel_name);
287 /* Remove the header with specified name from REQ. Returns true if
288 the header was actually removed, false otherwise. */
291 request_remove_header (struct request *req, char *name)
294 for (i = 0; i < req->hcount; i++)
296 struct request_header *hdr = &req->headers[i];
297 if (0 == strcasecmp (name, hdr->name))
299 release_header (hdr);
300 /* Move the remaining headers by one. */
301 if (i < req->hcount - 1)
302 memmove (hdr, hdr + 1, (req->hcount - i - 1) * sizeof (*hdr));
310 #define APPEND(p, str) do { \
311 int A_len = strlen (str); \
312 memcpy (p, str, A_len); \
316 /* Construct the request and write it to FD using fd_write. */
319 request_send (const struct request *req, int fd)
321 char *request_string, *p;
322 int i, size, write_error;
324 /* Count the request size. */
327 /* METHOD " " ARG " " "HTTP/1.0" "\r\n" */
328 size += strlen (req->method) + 1 + strlen (req->arg) + 1 + 8 + 2;
330 for (i = 0; i < req->hcount; i++)
332 struct request_header *hdr = &req->headers[i];
333 /* NAME ": " VALUE "\r\n" */
334 size += strlen (hdr->name) + 2 + strlen (hdr->value) + 2;
340 p = request_string = alloca_array (char, size);
342 /* Generate the request. */
344 APPEND (p, req->method); *p++ = ' ';
345 APPEND (p, req->arg); *p++ = ' ';
346 memcpy (p, "HTTP/1.0\r\n", 10); p += 10;
348 for (i = 0; i < req->hcount; i++)
350 struct request_header *hdr = &req->headers[i];
351 APPEND (p, hdr->name);
352 *p++ = ':', *p++ = ' ';
353 APPEND (p, hdr->value);
354 *p++ = '\r', *p++ = '\n';
357 *p++ = '\r', *p++ = '\n', *p++ = '\0';
358 assert (p - request_string == size);
362 DEBUGP (("\n---request begin---\n%s---request end---\n", request_string));
364 /* Send the request to the server. */
366 write_error = fd_write (fd, request_string, size - 1, -1);
368 logprintf (LOG_VERBOSE, _("Failed writing HTTP request: %s.\n"),
373 /* Release the resources used by REQ. */
376 request_free (struct request *req)
379 xfree_null (req->arg);
380 for (i = 0; i < req->hcount; i++)
381 release_header (&req->headers[i]);
382 xfree_null (req->headers);
386 static struct hash_table *basic_authed_hosts;
388 /* Find out if this host has issued a Basic challenge yet; if so, give
389 * it the username, password. A temporary measure until we can get
390 * proper authentication in place. */
393 maybe_send_basic_creds (const char *hostname, const char *user,
394 const char *passwd, struct request *req)
396 int did_challenge = 0;
398 if (basic_authed_hosts
399 && hash_table_contains(basic_authed_hosts, hostname))
401 DEBUGP(("Found `%s' in basic_authed_hosts.\n", hostname));
402 request_set_header (req, "Authorization",
403 basic_authentication_encode (user, passwd),
409 DEBUGP(("Host `%s' has not issued a general basic challenge.\n",
412 return did_challenge;
416 register_basic_auth_host (const char *hostname)
418 if (!basic_authed_hosts)
420 basic_authed_hosts = make_nocase_string_hash_table (1);
422 if (!hash_table_contains(basic_authed_hosts, hostname))
424 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
425 DEBUGP(("Inserted `%s' into basic_authed_hosts\n", hostname));
430 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
431 PROMISED_SIZE bytes are sent over the wire -- if the file is
432 longer, read only that much; if the file is shorter, report an error. */
435 post_file (int sock, const char *file_name, wgint promised_size)
437 static char chunk[8192];
442 DEBUGP (("[writing POST file %s ... ", file_name));
444 fp = fopen (file_name, "rb");
447 while (!feof (fp) && written < promised_size)
450 int length = fread (chunk, 1, sizeof (chunk), fp);
453 towrite = MIN (promised_size - written, length);
454 write_error = fd_write (sock, chunk, towrite, -1);
464 /* If we've written less than was promised, report a (probably
465 nonsensical) error rather than break the promise. */
466 if (written < promised_size)
472 assert (written == promised_size);
473 DEBUGP (("done]\n"));
477 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
478 If so, return the pointer to the position after the line, otherwise
479 return NULL. This is used as callback to fd_read_hunk. The data
480 between START and PEEKED has been read and cannot be "unread"; the
481 data after PEEKED has only been peeked. */
484 response_head_terminator (const char *start, const char *peeked, int peeklen)
488 /* If at first peek, verify whether HUNK starts with "HTTP". If
489 not, this is a HTTP/0.9 request and we must bail out without
491 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
494 /* Look for "\n[\r]\n", and return the following position if found.
495 Start two chars before the current to cover the possibility that
496 part of the terminator (e.g. "\n\r") arrived in the previous
498 p = peeked - start < 2 ? start : peeked - 2;
499 end = peeked + peeklen;
501 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
502 for (; p < end - 2; p++)
505 if (p[1] == '\r' && p[2] == '\n')
507 else if (p[1] == '\n')
510 /* p==end-2: check for \n\n directly preceding END. */
511 if (p[0] == '\n' && p[1] == '\n')
517 /* The maximum size of a single HTTP response we care to read. Rather
518 than being a limit of the reader implementation, this limit
519 prevents Wget from slurping all available memory upon encountering
520 malicious or buggy server output, thus protecting the user. Define
521 it to 0 to remove the limit. */
523 #define HTTP_RESPONSE_MAX_SIZE 65536
525 /* Read the HTTP request head from FD and return it. The error
526 conditions are the same as with fd_read_hunk.
528 To support HTTP/0.9 responses, this function tries to make sure
529 that the data begins with "HTTP". If this is not the case, no data
530 is read and an empty request is returned, so that the remaining
531 data can be treated as body. */
534 read_http_response_head (int fd)
536 return fd_read_hunk (fd, response_head_terminator, 512,
537 HTTP_RESPONSE_MAX_SIZE);
541 /* The response data. */
544 /* The array of pointers that indicate where each header starts.
545 For example, given this HTTP response:
552 The headers are located like this:
554 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
556 headers[0] headers[1] headers[2] headers[3]
558 I.e. headers[0] points to the beginning of the request,
559 headers[1] points to the end of the first header and the
560 beginning of the second one, etc. */
562 const char **headers;
565 /* Create a new response object from the text of the HTTP response,
566 available in HEAD. That text is automatically split into
567 constituent header lines for fast retrieval using
570 static struct response *
571 resp_new (const char *head)
576 struct response *resp = xnew0 (struct response);
581 /* Empty head means that we're dealing with a headerless
582 (HTTP/0.9) response. In that case, don't set HEADERS at
587 /* Split HEAD into header lines, so that resp_header_* functions
588 don't need to do this over and over again. */
594 DO_REALLOC (resp->headers, size, count + 1, const char *);
595 resp->headers[count++] = hdr;
597 /* Break upon encountering an empty line. */
598 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
601 /* Find the end of HDR, including continuations. */
604 const char *end = strchr (hdr, '\n');
610 while (*hdr == ' ' || *hdr == '\t');
612 DO_REALLOC (resp->headers, size, count + 1, const char *);
613 resp->headers[count] = NULL;
618 /* Locate the header named NAME in the request data, starting with
619 position START. This allows the code to loop through the request
620 data, filtering for all requests of a given name. Returns the
621 found position, or -1 for failure. The code that uses this
622 function typically looks like this:
624 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
625 ... do something with header ...
627 If you only care about one header, use resp_header_get instead of
631 resp_header_locate (const struct response *resp, const char *name, int start,
632 const char **begptr, const char **endptr)
635 const char **headers = resp->headers;
638 if (!headers || !headers[1])
641 name_len = strlen (name);
647 for (; headers[i + 1]; i++)
649 const char *b = headers[i];
650 const char *e = headers[i + 1];
652 && b[name_len] == ':'
653 && 0 == strncasecmp (b, name, name_len))
656 while (b < e && c_isspace (*b))
658 while (b < e && c_isspace (e[-1]))
668 /* Find and retrieve the header named NAME in the request data. If
669 found, set *BEGPTR to its starting, and *ENDPTR to its ending
670 position, and return true. Otherwise return false.
672 This function is used as a building block for resp_header_copy
673 and resp_header_strdup. */
676 resp_header_get (const struct response *resp, const char *name,
677 const char **begptr, const char **endptr)
679 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
683 /* Copy the response header named NAME to buffer BUF, no longer than
684 BUFSIZE (BUFSIZE includes the terminating 0). If the header
685 exists, true is returned, false otherwise. If there should be no
686 limit on the size of the header, use resp_header_strdup instead.
688 If BUFSIZE is 0, no data is copied, but the boolean indication of
689 whether the header is present is still returned. */
692 resp_header_copy (const struct response *resp, const char *name,
693 char *buf, int bufsize)
696 if (!resp_header_get (resp, name, &b, &e))
700 int len = MIN (e - b, bufsize - 1);
701 memcpy (buf, b, len);
707 /* Return the value of header named NAME in RESP, allocated with
708 malloc. If such a header does not exist in RESP, return NULL. */
711 resp_header_strdup (const struct response *resp, const char *name)
714 if (!resp_header_get (resp, name, &b, &e))
716 return strdupdelim (b, e);
719 /* Parse the HTTP status line, which is of format:
721 HTTP-Version SP Status-Code SP Reason-Phrase
723 The function returns the status-code, or -1 if the status line
724 appears malformed. The pointer to "reason-phrase" message is
725 returned in *MESSAGE. */
728 resp_status (const struct response *resp, char **message)
735 /* For a HTTP/0.9 response, assume status 200. */
737 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
741 p = resp->headers[0];
742 end = resp->headers[1];
748 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
752 /* Match the HTTP version. This is optional because Gnutella
753 servers have been reported to not specify HTTP version. */
754 if (p < end && *p == '/')
757 while (p < end && c_isdigit (*p))
759 if (p < end && *p == '.')
761 while (p < end && c_isdigit (*p))
765 while (p < end && c_isspace (*p))
767 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
770 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
775 while (p < end && c_isspace (*p))
777 while (p < end && c_isspace (end[-1]))
779 *message = strdupdelim (p, end);
785 /* Release the resources used by RESP. */
788 resp_free (struct response *resp)
790 xfree_null (resp->headers);
794 /* Print a single line of response, the characters [b, e). We tried
796 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
797 but that failed to escape the non-printable characters and, in fact,
798 caused crashes in UTF-8 locales. */
801 print_response_line(const char *prefix, const char *b, const char *e)
804 BOUNDED_TO_ALLOCA(b, e, copy);
805 logprintf (LOG_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
808 /* Print the server response, line by line, omitting the trailing CRLF
809 from individual header lines, and prefixed with PREFIX. */
812 print_server_response (const struct response *resp, const char *prefix)
817 for (i = 0; resp->headers[i + 1]; i++)
819 const char *b = resp->headers[i];
820 const char *e = resp->headers[i + 1];
822 if (b < e && e[-1] == '\n')
824 if (b < e && e[-1] == '\r')
826 print_response_line(prefix, b, e);
830 /* Parse the `Content-Range' header and extract the information it
831 contains. Returns true if successful, false otherwise. */
833 parse_content_range (const char *hdr, wgint *first_byte_ptr,
834 wgint *last_byte_ptr, wgint *entity_length_ptr)
838 /* Ancient versions of Netscape proxy server, presumably predating
839 rfc2068, sent out `Content-Range' without the "bytes"
841 if (0 == strncasecmp (hdr, "bytes", 5))
844 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
848 while (c_isspace (*hdr))
853 if (!c_isdigit (*hdr))
855 for (num = 0; c_isdigit (*hdr); hdr++)
856 num = 10 * num + (*hdr - '0');
857 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
859 *first_byte_ptr = num;
861 for (num = 0; c_isdigit (*hdr); hdr++)
862 num = 10 * num + (*hdr - '0');
863 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
865 *last_byte_ptr = num;
870 for (num = 0; c_isdigit (*hdr); hdr++)
871 num = 10 * num + (*hdr - '0');
872 *entity_length_ptr = num;
876 /* Read the body of the request, but don't store it anywhere and don't
877 display a progress gauge. This is useful for reading the bodies of
878 administrative responses to which we will soon issue another
879 request. The response is not useful to the user, but reading it
880 allows us to continue using the same connection to the server.
882 If reading fails, false is returned, true otherwise. In debug
883 mode, the body is displayed for debugging purposes. */
886 skip_short_body (int fd, wgint contlen)
889 SKIP_SIZE = 512, /* size of the download buffer */
890 SKIP_THRESHOLD = 4096 /* the largest size we read */
892 char dlbuf[SKIP_SIZE + 1];
893 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
895 /* We shouldn't get here with unknown contlen. (This will change
896 with HTTP/1.1, which supports "chunked" transfer.) */
897 assert (contlen != -1);
899 /* If the body is too large, it makes more sense to simply close the
900 connection than to try to read the body. */
901 if (contlen > SKIP_THRESHOLD)
904 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
908 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
911 /* Don't normally report the error since this is an
912 optimization that should be invisible to the user. */
913 DEBUGP (("] aborting (%s).\n",
914 ret < 0 ? fd_errstr (fd) : "EOF received"));
918 /* Safe even if %.*s bogusly expects terminating \0 because
919 we've zero-terminated dlbuf above. */
920 DEBUGP (("%.*s", ret, dlbuf));
923 DEBUGP (("] done.\n"));
927 /* Extract a parameter from the string (typically an HTTP header) at
928 **SOURCE and advance SOURCE to the next parameter. Return false
929 when there are no more parameters to extract. The name of the
930 parameter is returned in NAME, and the value in VALUE. If the
931 parameter has no value, the token's value is zeroed out.
933 For example, if *SOURCE points to the string "attachment;
934 filename=\"foo bar\"", the first call to this function will return
935 the token named "attachment" and no value, and the second call will
936 return the token named "filename" and value "foo bar". The third
937 call will return false, indicating no more valid tokens. */
940 extract_param (const char **source, param_token *name, param_token *value,
943 const char *p = *source;
945 while (c_isspace (*p)) ++p;
949 return false; /* no error; nothing more to extract */
954 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
956 if (name->b == name->e)
957 return false; /* empty name: error */
958 while (c_isspace (*p)) ++p;
959 if (*p == separator || !*p) /* no value */
962 if (*p == separator) ++p;
967 return false; /* error */
969 /* *p is '=', extract value */
971 while (c_isspace (*p)) ++p;
972 if (*p == '"') /* quoted */
975 while (*p && *p != '"') ++p;
979 /* Currently at closing quote; find the end of param. */
980 while (c_isspace (*p)) ++p;
981 while (*p && *p != separator) ++p;
985 /* garbage after closed quote, e.g. foo="bar"baz */
991 while (*p && *p != separator) ++p;
993 while (value->e != value->b && c_isspace (value->e[-1]))
995 if (*p == separator) ++p;
1002 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1004 /* Parse the contents of the `Content-Disposition' header, extracting
1005 the information useful to Wget. Content-Disposition is a header
1006 borrowed from MIME; when used in HTTP, it typically serves for
1007 specifying the desired file name of the resource. For example:
1009 Content-Disposition: attachment; filename="flora.jpg"
1011 Wget will skip the tokens it doesn't care about, such as
1012 "attachment" in the previous example; it will also skip other
1013 unrecognized params. If the header is syntactically correct and
1014 contains a file name, a copy of the file name is stored in
1015 *filename and true is returned. Otherwise, the function returns
1018 The file name is stripped of directory components and must not be
1022 parse_content_disposition (const char *hdr, char **filename)
1024 param_token name, value;
1025 while (extract_param (&hdr, &name, &value, ';'))
1026 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1028 /* Make the file name begin at the last slash or backslash. */
1029 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1030 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1031 if (last_slash && last_bs)
1032 value.b = 1 + MAX (last_slash, last_bs);
1033 else if (last_slash || last_bs)
1034 value.b = 1 + (last_slash ? last_slash : last_bs);
1035 if (value.b == value.e)
1037 /* Start with the directory prefix, if specified. */
1040 int prefix_length = strlen (opt.dir_prefix);
1041 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1046 total_length = prefix_length + (value.e - value.b);
1047 *filename = xmalloc (total_length + 1);
1048 strcpy (*filename, opt.dir_prefix);
1050 (*filename)[prefix_length - 1] = '/';
1051 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1052 (*filename)[total_length] = '\0';
1055 *filename = strdupdelim (value.b, value.e);
1061 /* Persistent connections. Currently, we cache the most recently used
1062 connection as persistent, provided that the HTTP server agrees to
1063 make it such. The persistence data is stored in the variables
1064 below. Ideally, it should be possible to cache an arbitrary fixed
1065 number of these connections. */
1067 /* Whether a persistent connection is active. */
1068 static bool pconn_active;
1071 /* The socket of the connection. */
1074 /* Host and port of the currently active persistent connection. */
1078 /* Whether a ssl handshake has occoured on this connection. */
1081 /* Whether the connection was authorized. This is only done by
1082 NTLM, which authorizes *connections* rather than individual
1083 requests. (That practice is peculiar for HTTP, but it is a
1084 useful optimization.) */
1088 /* NTLM data of the current connection. */
1089 struct ntlmdata ntlm;
1093 /* Mark the persistent connection as invalid and free the resources it
1094 uses. This is used by the CLOSE_* macros after they forcefully
1095 close a registered persistent connection. */
1098 invalidate_persistent (void)
1100 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1101 pconn_active = false;
1102 fd_close (pconn.socket);
1107 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1108 persistent. This will enable someone to use the same connection
1109 later. In the context of HTTP, this must be called only AFTER the
1110 response has been received and the server has promised that the
1111 connection will remain alive.
1113 If a previous connection was persistent, it is closed. */
1116 register_persistent (const char *host, int port, int fd, bool ssl)
1120 if (pconn.socket == fd)
1122 /* The connection FD is already registered. */
1127 /* The old persistent connection is still active; close it
1128 first. This situation arises whenever a persistent
1129 connection exists, but we then connect to a different
1130 host, and try to register a persistent connection to that
1132 invalidate_persistent ();
1136 pconn_active = true;
1138 pconn.host = xstrdup (host);
1141 pconn.authorized = false;
1143 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1146 /* Return true if a persistent connection is available for connecting
1150 persistent_available_p (const char *host, int port, bool ssl,
1151 bool *host_lookup_failed)
1153 /* First, check whether a persistent connection is active at all. */
1157 /* If we want SSL and the last connection wasn't or vice versa,
1158 don't use it. Checking for host and port is not enough because
1159 HTTP and HTTPS can apparently coexist on the same port. */
1160 if (ssl != pconn.ssl)
1163 /* If we're not connecting to the same port, we're not interested. */
1164 if (port != pconn.port)
1167 /* If the host is the same, we're in business. If not, there is
1168 still hope -- read below. */
1169 if (0 != strcasecmp (host, pconn.host))
1171 /* Check if pconn.socket is talking to HOST under another name.
1172 This happens often when both sites are virtual hosts
1173 distinguished only by name and served by the same network
1174 interface, and hence the same web server (possibly set up by
1175 the ISP and serving many different web sites). This
1176 admittedly unconventional optimization does not contradict
1177 HTTP and works well with popular server software. */
1181 struct address_list *al;
1184 /* Don't try to talk to two different SSL sites over the same
1185 secure connection! (Besides, it's not clear that
1186 name-based virtual hosting is even possible with SSL.) */
1189 /* If pconn.socket's peer is one of the IP addresses HOST
1190 resolves to, pconn.socket is for all intents and purposes
1191 already talking to HOST. */
1193 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1195 /* Can't get the peer's address -- something must be very
1196 wrong with the connection. */
1197 invalidate_persistent ();
1200 al = lookup_host (host, 0);
1203 *host_lookup_failed = true;
1207 found = address_list_contains (al, &ip);
1208 address_list_release (al);
1213 /* The persistent connection's peer address was found among the
1214 addresses HOST resolved to; therefore, pconn.sock is in fact
1215 already talking to HOST -- no need to reconnect. */
1218 /* Finally, check whether the connection is still open. This is
1219 important because most servers implement liberal (short) timeout
1220 on persistent connections. Wget can of course always reconnect
1221 if the connection doesn't work out, but it's nicer to know in
1222 advance. This test is a logical followup of the first test, but
1223 is "expensive" and therefore placed at the end of the list.
1225 (Current implementation of test_socket_open has a nice side
1226 effect that it treats sockets with pending data as "closed".
1227 This is exactly what we want: if a broken server sends message
1228 body in response to HEAD, or if it sends more than conent-length
1229 data, we won't reuse the corrupted connection.) */
1231 if (!test_socket_open (pconn.socket))
1233 /* Oops, the socket is no longer open. Now that we know that,
1234 let's invalidate the persistent connection before returning
1236 invalidate_persistent ();
1243 /* The idea behind these two CLOSE macros is to distinguish between
1244 two cases: one when the job we've been doing is finished, and we
1245 want to close the connection and leave, and two when something is
1246 seriously wrong and we're closing the connection as part of
1249 In case of keep_alive, CLOSE_FINISH should leave the connection
1250 open, while CLOSE_INVALIDATE should still close it.
1252 Note that the semantics of the flag `keep_alive' is "this
1253 connection *will* be reused (the server has promised not to close
1254 the connection once we're done)", while the semantics of
1255 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1256 active, registered connection". */
1258 #define CLOSE_FINISH(fd) do { \
1261 if (pconn_active && (fd) == pconn.socket) \
1262 invalidate_persistent (); \
1271 #define CLOSE_INVALIDATE(fd) do { \
1272 if (pconn_active && (fd) == pconn.socket) \
1273 invalidate_persistent (); \
1281 wgint len; /* received length */
1282 wgint contlen; /* expected length */
1283 wgint restval; /* the restart value */
1284 int res; /* the result of last read */
1285 char *rderrmsg; /* error message from read error */
1286 char *newloc; /* new location (redirection) */
1287 char *remote_time; /* remote time-stamp string */
1288 char *error; /* textual HTTP error */
1289 int statcode; /* status code */
1290 wgint rd_size; /* amount of data read from socket */
1291 double dltime; /* time it took to download the data */
1292 const char *referer; /* value of the referer header. */
1293 char *local_file; /* local file name. */
1294 bool existence_checked; /* true if we already checked for a file's
1295 existence after having begun to download
1296 (needed in gethttp for when connection is
1297 interrupted/restarted. */
1298 bool timestamp_checked; /* true if pre-download time-stamping checks
1299 * have already been performed */
1300 char *orig_file_name; /* name of file to compare for time-stamping
1301 * (might be != local_file if -K is set) */
1302 wgint orig_file_size; /* size of file to compare for time-stamping */
1303 time_t orig_file_tstamp; /* time-stamp of file to compare for
1308 free_hstat (struct http_stat *hs)
1310 xfree_null (hs->newloc);
1311 xfree_null (hs->remote_time);
1312 xfree_null (hs->error);
1313 xfree_null (hs->rderrmsg);
1314 xfree_null (hs->local_file);
1315 xfree_null (hs->orig_file_name);
1317 /* Guard against being called twice. */
1319 hs->remote_time = NULL;
1323 #define BEGINS_WITH(line, string_constant) \
1324 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1325 && (c_isspace (line[sizeof (string_constant) - 1]) \
1326 || !line[sizeof (string_constant) - 1]))
1328 #define SET_USER_AGENT(req) do { \
1329 if (!opt.useragent) \
1330 request_set_header (req, "User-Agent", \
1331 aprintf ("Wget/%s", version_string), rel_value); \
1332 else if (*opt.useragent) \
1333 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1336 /* The flags that allow clobbering the file (opening with "wb").
1337 Defined here to avoid repetition later. #### This will require
1339 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1340 || opt.dirstruct || opt.output_document)
1342 /* Retrieve a document through HTTP protocol. It recognizes status
1343 code, and correctly handles redirections. It closes the network
1344 socket. If it receives an error from the functions below it, it
1345 will print it if there is enough information to do so (almost
1346 always), returning the error to the caller (i.e. http_loop).
1348 Various HTTP parameters are stored to hs.
1350 If PROXY is non-NULL, the connection will be made to the proxy
1351 server, and u->url will be requested. */
1353 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1355 struct request *req;
1358 char *user, *passwd;
1362 wgint contlen, contrange;
1369 /* Set to 1 when the authorization has already been sent and should
1370 not be tried again. */
1371 bool auth_finished = false;
1373 /* Set to 1 when just globally-set Basic authorization has been sent;
1374 * should prevent further Basic negotiations, but not other
1376 bool basic_auth_finished = false;
1378 /* Whether NTLM authentication is used for this request. */
1379 bool ntlm_seen = false;
1381 /* Whether our connection to the remote host is through SSL. */
1382 bool using_ssl = false;
1384 /* Whether a HEAD request will be issued (as opposed to GET or
1386 bool head_only = !!(*dt & HEAD_ONLY);
1389 struct response *resp;
1393 /* Whether this connection will be kept alive after the HTTP request
1397 /* Whether keep-alive should be inhibited.
1399 RFC 2068 requests that 1.0 clients not send keep-alive requests
1400 to proxies. This is because many 1.0 proxies do not interpret
1401 the Connection header and transfer it to the remote server,
1402 causing it to not close the connection and leave both the proxy
1403 and the client hanging. */
1404 bool inhibit_keep_alive =
1405 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1407 /* Headers sent when using POST. */
1408 wgint post_data_size = 0;
1410 bool host_lookup_failed = false;
1413 if (u->scheme == SCHEME_HTTPS)
1415 /* Initialize the SSL context. After this has once been done,
1416 it becomes a no-op. */
1419 scheme_disable (SCHEME_HTTPS);
1420 logprintf (LOG_NOTQUIET,
1421 _("Disabling SSL due to encountered errors.\n"));
1422 return SSLINITFAILED;
1425 #endif /* HAVE_SSL */
1427 /* Initialize certain elements of struct http_stat. */
1431 hs->rderrmsg = NULL;
1433 hs->remote_time = NULL;
1438 /* Prepare the request to send. */
1440 req = request_new ();
1443 const char *meth = "GET";
1446 else if (opt.post_file_name || opt.post_data)
1448 /* Use the full path, i.e. one that includes the leading slash and
1449 the query string. E.g. if u->path is "foo/bar" and u->query is
1450 "param=value", full_path will be "/foo/bar?param=value". */
1453 /* When using SSL over proxy, CONNECT establishes a direct
1454 connection to the HTTPS server. Therefore use the same
1455 argument as when talking to the server directly. */
1456 && u->scheme != SCHEME_HTTPS
1459 meth_arg = xstrdup (u->url);
1461 meth_arg = url_full_path (u);
1462 request_set_method (req, meth, meth_arg);
1465 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1466 if (*dt & SEND_NOCACHE)
1467 request_set_header (req, "Pragma", "no-cache", rel_none);
1469 request_set_header (req, "Range",
1470 aprintf ("bytes=%s-",
1471 number_to_static_string (hs->restval)),
1473 SET_USER_AGENT (req);
1474 request_set_header (req, "Accept", "*/*", rel_none);
1476 /* Find the username and password for authentication. */
1479 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1480 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1481 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1484 && !u->user) /* We only do "site-wide" authentication with "global"
1485 user/password values; URL user/password info overrides. */
1487 /* If this is a host for which we've already received a Basic
1488 * challenge, we'll go ahead and send Basic authentication creds. */
1489 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1495 char *proxy_user, *proxy_passwd;
1496 /* For normal username and password, URL components override
1497 command-line/wgetrc parameters. With proxy
1498 authentication, it's the reverse, because proxy URLs are
1499 normally the "permanent" ones, so command-line args
1500 should take precedence. */
1501 if (opt.proxy_user && opt.proxy_passwd)
1503 proxy_user = opt.proxy_user;
1504 proxy_passwd = opt.proxy_passwd;
1508 proxy_user = proxy->user;
1509 proxy_passwd = proxy->passwd;
1511 /* #### This does not appear right. Can't the proxy request,
1512 say, `Digest' authentication? */
1513 if (proxy_user && proxy_passwd)
1514 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1516 /* If we're using a proxy, we will be connecting to the proxy
1520 /* Proxy authorization over SSL is handled below. */
1522 if (u->scheme != SCHEME_HTTPS)
1524 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1527 /* Generate the Host header, HOST:PORT. Take into account that:
1529 - Broken server-side software often doesn't recognize the PORT
1530 argument, so we must generate "Host: www.server.com" instead of
1531 "Host: www.server.com:80" (and likewise for https port).
1533 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1534 becomes ambiguous and needs to be rewritten as "Host:
1535 [3ffe:8100:200:2::2]:1234". */
1537 /* Formats arranged for hfmt[add_port][add_squares]. */
1538 static const char *hfmt[][2] = {
1539 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1541 int add_port = u->port != scheme_default_port (u->scheme);
1542 int add_squares = strchr (u->host, ':') != NULL;
1543 request_set_header (req, "Host",
1544 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1548 if (!inhibit_keep_alive)
1549 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1552 request_set_header (req, "Cookie",
1553 cookie_header (wget_cookie_jar,
1554 u->host, u->port, u->path,
1556 u->scheme == SCHEME_HTTPS
1563 if (opt.post_data || opt.post_file_name)
1565 request_set_header (req, "Content-Type",
1566 "application/x-www-form-urlencoded", rel_none);
1568 post_data_size = strlen (opt.post_data);
1571 post_data_size = file_size (opt.post_file_name);
1572 if (post_data_size == -1)
1574 logprintf (LOG_NOTQUIET, _("POST data file `%s' missing: %s\n"),
1575 opt.post_file_name, strerror (errno));
1579 request_set_header (req, "Content-Length",
1580 xstrdup (number_to_static_string (post_data_size)),
1584 /* Add the user headers. */
1585 if (opt.user_headers)
1588 for (i = 0; opt.user_headers[i]; i++)
1589 request_set_user_header (req, opt.user_headers[i]);
1593 /* We need to come back here when the initial attempt to retrieve
1594 without authorization header fails. (Expected to happen at least
1595 for the Digest authorization scheme.) */
1599 /* Establish the connection. */
1601 if (!inhibit_keep_alive)
1603 /* Look for a persistent connection to target host, unless a
1604 proxy is used. The exception is when SSL is in use, in which
1605 case the proxy is nothing but a passthrough to the target
1606 host, registered as a connection to the latter. */
1607 struct url *relevant = conn;
1609 if (u->scheme == SCHEME_HTTPS)
1613 if (persistent_available_p (relevant->host, relevant->port,
1615 relevant->scheme == SCHEME_HTTPS,
1619 &host_lookup_failed))
1621 sock = pconn.socket;
1622 using_ssl = pconn.ssl;
1623 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1624 escnonprint (pconn.host), pconn.port);
1625 DEBUGP (("Reusing fd %d.\n", sock));
1626 if (pconn.authorized)
1627 /* If the connection is already authorized, the "Basic"
1628 authorization added by code above is unnecessary and
1630 request_remove_header (req, "Authorization");
1632 else if (host_lookup_failed)
1635 logprintf(LOG_NOTQUIET,
1636 _("%s: unable to resolve host address `%s'\n"),
1637 exec_name, relevant->host);
1644 sock = connect_to_host (conn->host, conn->port);
1653 return (retryable_socket_connect_error (errno)
1654 ? CONERROR : CONIMPOSSIBLE);
1658 if (proxy && u->scheme == SCHEME_HTTPS)
1660 /* When requesting SSL URLs through proxies, use the
1661 CONNECT method to request passthrough. */
1662 struct request *connreq = request_new ();
1663 request_set_method (connreq, "CONNECT",
1664 aprintf ("%s:%d", u->host, u->port));
1665 SET_USER_AGENT (connreq);
1668 request_set_header (connreq, "Proxy-Authorization",
1669 proxyauth, rel_value);
1670 /* Now that PROXYAUTH is part of the CONNECT request,
1671 zero it out so we don't send proxy authorization with
1672 the regular request below. */
1675 /* Examples in rfc2817 use the Host header in CONNECT
1676 requests. I don't see how that gains anything, given
1677 that the contents of Host would be exactly the same as
1678 the contents of CONNECT. */
1680 write_error = request_send (connreq, sock);
1681 request_free (connreq);
1682 if (write_error < 0)
1684 CLOSE_INVALIDATE (sock);
1688 head = read_http_response_head (sock);
1691 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1693 CLOSE_INVALIDATE (sock);
1702 DEBUGP (("proxy responded with: [%s]\n", head));
1704 resp = resp_new (head);
1705 statcode = resp_status (resp, &message);
1708 if (statcode != 200)
1711 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1712 message ? escnonprint (message) : "?");
1713 xfree_null (message);
1716 xfree_null (message);
1718 /* SOCK is now *really* connected to u->host, so update CONN
1719 to reflect this. That way register_persistent will
1720 register SOCK as being connected to u->host:u->port. */
1724 if (conn->scheme == SCHEME_HTTPS)
1726 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1733 #endif /* HAVE_SSL */
1736 /* Send the request to server. */
1737 write_error = request_send (req, sock);
1739 if (write_error >= 0)
1743 DEBUGP (("[POST data: %s]\n", opt.post_data));
1744 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1746 else if (opt.post_file_name && post_data_size != 0)
1747 write_error = post_file (sock, opt.post_file_name, post_data_size);
1750 if (write_error < 0)
1752 CLOSE_INVALIDATE (sock);
1756 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1757 proxy ? "Proxy" : "HTTP");
1762 head = read_http_response_head (sock);
1767 logputs (LOG_NOTQUIET, _("No data received.\n"));
1768 CLOSE_INVALIDATE (sock);
1774 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1776 CLOSE_INVALIDATE (sock);
1781 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1783 resp = resp_new (head);
1785 /* Check for status line. */
1787 statcode = resp_status (resp, &message);
1788 if (!opt.server_response)
1789 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1790 message ? escnonprint (message) : "");
1793 logprintf (LOG_VERBOSE, "\n");
1794 print_server_response (resp, " ");
1797 /* Determine the local filename if needed. Notice that if -O is used
1798 * hstat.local_file is set by http_loop to the argument of -O. */
1799 if (!hs->local_file)
1801 /* Honor Content-Disposition whether possible. */
1802 if (!opt.content_disposition
1803 || !resp_header_copy (resp, "Content-Disposition",
1804 hdrval, sizeof (hdrval))
1805 || !parse_content_disposition (hdrval, &hs->local_file))
1807 /* The Content-Disposition header is missing or broken.
1808 * Choose unique file name according to given URL. */
1809 hs->local_file = url_file_name (u);
1813 /* TODO: perform this check only once. */
1814 if (!hs->existence_checked && file_exists_p (hs->local_file))
1818 /* If opt.noclobber is turned on and file already exists, do not
1819 retrieve the file */
1820 logprintf (LOG_VERBOSE, _("\
1821 File `%s' already there; not retrieving.\n\n"), hs->local_file);
1822 /* If the file is there, we suppose it's retrieved OK. */
1825 /* #### Bogusness alert. */
1826 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1827 if (has_html_suffix_p (hs->local_file))
1830 return RETRUNNEEDED;
1832 else if (!ALLOW_CLOBBER)
1834 char *unique = unique_name (hs->local_file, true);
1835 if (unique != hs->local_file)
1836 xfree (hs->local_file);
1837 hs->local_file = unique;
1840 hs->existence_checked = true;
1842 /* Support timestamping */
1843 /* TODO: move this code out of gethttp. */
1844 if (opt.timestamping && !hs->timestamp_checked)
1846 size_t filename_len = strlen (hs->local_file);
1847 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1848 bool local_dot_orig_file_exists = false;
1849 char *local_filename = NULL;
1852 if (opt.backup_converted)
1853 /* If -K is specified, we'll act on the assumption that it was specified
1854 last time these files were downloaded as well, and instead of just
1855 comparing local file X against server file X, we'll compare local
1856 file X.orig (if extant, else X) against server file X. If -K
1857 _wasn't_ specified last time, or the server contains files called
1858 *.orig, -N will be back to not operating correctly with -k. */
1860 /* Would a single s[n]printf() call be faster? --dan
1862 Definitely not. sprintf() is horribly slow. It's a
1863 different question whether the difference between the two
1864 affects a program. Usually I'd say "no", but at one
1865 point I profiled Wget, and found that a measurable and
1866 non-negligible amount of time was lost calling sprintf()
1867 in url.c. Replacing sprintf with inline calls to
1868 strcpy() and number_to_string() made a difference.
1870 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1871 memcpy (filename_plus_orig_suffix + filename_len,
1872 ".orig", sizeof (".orig"));
1874 /* Try to stat() the .orig file. */
1875 if (stat (filename_plus_orig_suffix, &st) == 0)
1877 local_dot_orig_file_exists = true;
1878 local_filename = filename_plus_orig_suffix;
1882 if (!local_dot_orig_file_exists)
1883 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1884 if (stat (hs->local_file, &st) == 0)
1885 local_filename = hs->local_file;
1887 if (local_filename != NULL)
1888 /* There was a local file, so we'll check later to see if the version
1889 the server has is the same version we already have, allowing us to
1892 hs->orig_file_name = xstrdup (local_filename);
1893 hs->orig_file_size = st.st_size;
1894 hs->orig_file_tstamp = st.st_mtime;
1896 /* Modification time granularity is 2 seconds for Windows, so
1897 increase local time by 1 second for later comparison. */
1898 ++hs->orig_file_tstamp;
1903 if (!opt.ignore_length
1904 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1908 parsed = str_to_wgint (hdrval, NULL, 10);
1909 if (parsed == WGINT_MAX && errno == ERANGE)
1912 #### If Content-Length is out of range, it most likely
1913 means that the file is larger than 2G and that we're
1914 compiled without LFS. In that case we should probably
1915 refuse to even attempt to download the file. */
1918 else if (parsed < 0)
1920 /* Negative Content-Length; nonsensical, so we can't
1921 assume any information about the content to receive. */
1928 /* Check for keep-alive related responses. */
1929 if (!inhibit_keep_alive && contlen != -1)
1931 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1933 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1935 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1940 /* The server has promised that it will not close the connection
1941 when we're done. This means that we can register it. */
1942 register_persistent (conn->host, conn->port, sock, using_ssl);
1944 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1946 /* Authorization is required. */
1947 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1948 CLOSE_FINISH (sock);
1950 CLOSE_INVALIDATE (sock);
1951 pconn.authorized = false;
1952 if (!auth_finished && (user && passwd))
1954 /* IIS sends multiple copies of WWW-Authenticate, one with
1955 the value "negotiate", and other(s) with data. Loop over
1956 all the occurrences and pick the one we recognize. */
1958 const char *wabeg, *waend;
1959 char *www_authenticate = NULL;
1961 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1962 &wabeg, &waend)) != -1;
1964 if (known_authentication_scheme_p (wabeg, waend))
1966 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1970 if (!www_authenticate)
1972 /* If the authentication header is missing or
1973 unrecognized, there's no sense in retrying. */
1974 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1976 else if (!basic_auth_finished
1977 || !BEGINS_WITH (www_authenticate, "Basic"))
1980 pth = url_full_path (u);
1981 request_set_header (req, "Authorization",
1982 create_authorization_line (www_authenticate,
1984 request_method (req),
1988 if (BEGINS_WITH (www_authenticate, "NTLM"))
1990 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
1992 /* Need to register this host as using basic auth,
1993 * so we automatically send creds next time. */
1994 register_basic_auth_host (u->host);
1997 goto retry_with_auth;
2001 /* We already did Basic auth, and it failed. Gotta
2005 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2009 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2011 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2013 pconn.authorized = true;
2017 hs->statcode = statcode;
2019 hs->error = xstrdup (_("Malformed status line"));
2021 hs->error = xstrdup (_("(no description)"));
2023 hs->error = xstrdup (message);
2024 xfree_null (message);
2026 type = resp_header_strdup (resp, "Content-Type");
2029 char *tmp = strchr (type, ';');
2032 while (tmp > type && c_isspace (tmp[-1]))
2037 hs->newloc = resp_header_strdup (resp, "Location");
2038 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2040 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2044 const char *scbeg, *scend;
2045 /* The jar should have been created by now. */
2046 assert (wget_cookie_jar != NULL);
2048 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2049 &scbeg, &scend)) != -1;
2052 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2053 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2054 u->path, set_cookie);
2058 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2060 wgint first_byte_pos, last_byte_pos, entity_length;
2061 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2064 contrange = first_byte_pos;
2065 contlen = last_byte_pos - first_byte_pos + 1;
2070 /* 20x responses are counted among successful by default. */
2071 if (H_20X (statcode))
2074 /* Return if redirected. */
2075 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2077 /* RFC2068 says that in case of the 300 (multiple choices)
2078 response, the server can output a preferred URL through
2079 `Location' header; otherwise, the request should be treated
2080 like GET. So, if the location is set, it will be a
2081 redirection; otherwise, just proceed normally. */
2082 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2086 logprintf (LOG_VERBOSE,
2087 _("Location: %s%s\n"),
2088 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2089 hs->newloc ? _(" [following]") : "");
2090 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2091 CLOSE_FINISH (sock);
2093 CLOSE_INVALIDATE (sock);
2099 /* If content-type is not given, assume text/html. This is because
2100 of the multitude of broken CGI's that "forget" to generate the
2103 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2104 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2109 if (opt.html_extension && (*dt & TEXTHTML))
2110 /* -E / --html-extension / html_extension = on was specified, and this is a
2111 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2112 already the file's suffix, tack on ".html". */
2114 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2116 if (last_period_in_local_filename == NULL
2117 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2118 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2120 int local_filename_len = strlen (hs->local_file);
2121 /* Resize the local file, allowing for ".html" preceded by
2122 optional ".NUMBER". */
2123 hs->local_file = xrealloc (hs->local_file,
2124 local_filename_len + 24 + sizeof (".html"));
2125 strcpy(hs->local_file + local_filename_len, ".html");
2126 /* If clobbering is not allowed and the file, as named,
2127 exists, tack on ".NUMBER.html" instead. */
2128 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2132 sprintf (hs->local_file + local_filename_len,
2133 ".%d.html", ext_num++);
2134 while (file_exists_p (hs->local_file));
2136 *dt |= ADDED_HTML_EXTENSION;
2140 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2142 /* If `-c' is in use and the file has been fully downloaded (or
2143 the remote file has shrunk), Wget effectively requests bytes
2144 after the end of file and the server response with 416. */
2145 logputs (LOG_VERBOSE, _("\
2146 \n The file is already fully retrieved; nothing to do.\n\n"));
2147 /* In case the caller inspects. */
2150 /* Mark as successfully retrieved. */
2153 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2154 might be more bytes in the body. */
2155 return RETRUNNEEDED;
2157 if ((contrange != 0 && contrange != hs->restval)
2158 || (H_PARTIAL (statcode) && !contrange))
2160 /* The Range request was somehow misunderstood by the server.
2163 CLOSE_INVALIDATE (sock);
2169 hs->contlen = contlen + contrange;
2175 /* No need to print this output if the body won't be
2176 downloaded at all, or if the original server response is
2178 logputs (LOG_VERBOSE, _("Length: "));
2181 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2182 if (contlen + contrange >= 1024)
2183 logprintf (LOG_VERBOSE, " (%s)",
2184 human_readable (contlen + contrange));
2187 if (contlen >= 1024)
2188 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2189 number_to_static_string (contlen),
2190 human_readable (contlen));
2192 logprintf (LOG_VERBOSE, _(", %s remaining"),
2193 number_to_static_string (contlen));
2197 logputs (LOG_VERBOSE,
2198 opt.ignore_length ? _("ignored") : _("unspecified"));
2200 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2202 logputs (LOG_VERBOSE, "\n");
2206 type = NULL; /* We don't need it any more. */
2208 /* Return if we have no intention of further downloading. */
2209 if (!(*dt & RETROKF) || head_only)
2211 /* In case the caller cares to look... */
2216 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2217 servers not to send body in response to a HEAD request, and
2218 those that do will likely be caught by test_socket_open.
2219 If not, they can be worked around using
2220 `--no-http-keep-alive'. */
2221 CLOSE_FINISH (sock);
2222 else if (keep_alive && skip_short_body (sock, contlen))
2223 /* Successfully skipped the body; also keep using the socket. */
2224 CLOSE_FINISH (sock);
2226 CLOSE_INVALIDATE (sock);
2227 return RETRFINISHED;
2230 /* Open the local file. */
2233 mkalldirs (hs->local_file);
2235 rotate_backups (hs->local_file);
2237 fp = fopen (hs->local_file, "ab");
2238 else if (ALLOW_CLOBBER)
2239 fp = fopen (hs->local_file, "wb");
2242 fp = fopen_excl (hs->local_file, true);
2243 if (!fp && errno == EEXIST)
2245 /* We cannot just invent a new name and use it (which is
2246 what functions like unique_create typically do)
2247 because we told the user we'd use this name.
2248 Instead, return and retry the download. */
2249 logprintf (LOG_NOTQUIET,
2250 _("%s has sprung into existence.\n"),
2252 CLOSE_INVALIDATE (sock);
2253 return FOPEN_EXCL_ERR;
2258 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2259 CLOSE_INVALIDATE (sock);
2266 /* Print fetch message, if opt.verbose. */
2269 logprintf (LOG_NOTQUIET, _("Saving to: `%s'\n"),
2270 HYPHENP (hs->local_file) ? "STDOUT" : hs->local_file);
2273 /* This confuses the timestamping code that checks for file size.
2274 #### The timestamping code should be smarter about file size. */
2275 if (opt.save_headers && hs->restval == 0)
2276 fwrite (head, 1, strlen (head), fp);
2278 /* Now we no longer need to store the response header. */
2281 /* Download the request body. */
2284 /* If content-length is present, read that much; otherwise, read
2285 until EOF. The HTTP spec doesn't require the server to
2286 actually close the connection when it's done sending data. */
2287 flags |= rb_read_exactly;
2288 if (hs->restval > 0 && contrange == 0)
2289 /* If the server ignored our range request, instruct fd_read_body
2290 to skip the first RESTVAL bytes of body. */
2291 flags |= rb_skip_startpos;
2292 hs->len = hs->restval;
2294 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2295 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2299 CLOSE_FINISH (sock);
2303 hs->rderrmsg = xstrdup (fd_errstr (sock));
2304 CLOSE_INVALIDATE (sock);
2311 return RETRFINISHED;
2314 /* The genuine HTTP loop! This is the part where the retrieval is
2315 retried, and retried, and retried, and... */
2317 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2318 int *dt, struct url *proxy)
2321 bool got_head = false; /* used for time-stamping and filename detection */
2322 bool time_came_from_head = false;
2323 bool got_name = false;
2326 uerr_t err, ret = TRYLIMEXC;
2327 time_t tmr = -1; /* remote time-stamp */
2328 struct http_stat hstat; /* HTTP status */
2330 bool send_head_first = true;
2332 /* Assert that no value for *LOCAL_FILE was passed. */
2333 assert (local_file == NULL || *local_file == NULL);
2335 /* Set LOCAL_FILE parameter. */
2336 if (local_file && opt.output_document)
2337 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2339 /* Reset NEWLOC parameter. */
2342 /* This used to be done in main(), but it's a better idea to do it
2343 here so that we don't go through the hoops if we're just using
2348 /* Warn on (likely bogus) wildcard usage in HTTP. */
2349 if (opt.ftp_glob && has_wildcards_p (u->path))
2350 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2352 /* Setup hstat struct. */
2354 hstat.referer = referer;
2356 if (opt.output_document)
2358 hstat.local_file = xstrdup (opt.output_document);
2361 else if (!opt.content_disposition)
2363 hstat.local_file = url_file_name (u);
2367 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2368 * screaming for some refactoring. */
2369 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber)
2371 /* If opt.noclobber is turned on and file already exists, do not
2372 retrieve the file */
2373 logprintf (LOG_VERBOSE, _("\
2374 File `%s' already there; not retrieving.\n\n"),
2376 /* If the file is there, we suppose it's retrieved OK. */
2379 /* #### Bogusness alert. */
2380 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2381 if (has_html_suffix_p (hstat.local_file))
2384 return RETRUNNEEDED;
2387 /* Reset the counter. */
2390 /* Reset the document type. */
2393 /* Skip preliminary HEAD request if we're not in spider mode AND
2394 * if -O was given or HTTP Content-Disposition support is disabled. */
2396 && (got_name || !opt.content_disposition))
2397 send_head_first = false;
2399 /* Send preliminary HEAD request if -N is given and we have an existing
2400 * destination file. */
2401 if (opt.timestamping
2402 && !opt.content_disposition
2403 && file_exists_p (url_file_name (u)))
2404 send_head_first = true;
2409 /* Increment the pass counter. */
2411 sleep_between_retrievals (count);
2413 /* Get the current time string. */
2414 tms = datetime_str (time (NULL));
2416 if (opt.spider && !got_head)
2417 logprintf (LOG_VERBOSE, _("\
2418 Spider mode enabled. Check if remote file exists.\n"));
2420 /* Print fetch message, if opt.verbose. */
2423 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2428 sprintf (tmp, _("(try:%2d)"), count);
2429 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2434 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2439 ws_changetitle (hurl);
2444 /* Default document type is empty. However, if spider mode is
2445 on or time-stamping is employed, HEAD_ONLY commands is
2446 encoded within *dt. */
2447 if (send_head_first && !got_head)
2452 /* Decide whether or not to restart. */
2455 && stat (hstat.local_file, &st) == 0
2456 && S_ISREG (st.st_mode))
2457 /* When -c is used, continue from on-disk size. (Can't use
2458 hstat.len even if count>1 because we don't want a failed
2459 first attempt to clobber existing data.) */
2460 hstat.restval = st.st_size;
2462 /* otherwise, continue where the previous try left off */
2463 hstat.restval = hstat.len;
2467 /* Decide whether to send the no-cache directive. We send it in
2469 a) we're using a proxy, and we're past our first retrieval.
2470 Some proxies are notorious for caching incomplete data, so
2471 we require a fresh get.
2472 b) caching is explicitly inhibited. */
2473 if ((proxy && count > 1) /* a */
2474 || !opt.allow_cache) /* b */
2475 *dt |= SEND_NOCACHE;
2477 *dt &= ~SEND_NOCACHE;
2479 /* Try fetching the document, or at least its head. */
2480 err = gethttp (u, &hstat, dt, proxy);
2483 tms = datetime_str (time (NULL));
2485 /* Get the new location (with or without the redirection). */
2487 *newloc = xstrdup (hstat.newloc);
2491 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2492 case CONERROR: case READERR: case WRITEFAILED:
2493 case RANGEERR: case FOPEN_EXCL_ERR:
2494 /* Non-fatal errors continue executing the loop, which will
2495 bring them to "while" statement at the end, to judge
2496 whether the number of tries was exceeded. */
2497 printwhat (count, opt.ntry);
2499 case FWRITEERR: case FOPENERR:
2500 /* Another fatal error. */
2501 logputs (LOG_VERBOSE, "\n");
2502 logprintf (LOG_NOTQUIET, _("Cannot write to `%s' (%s).\n"),
2503 hstat.local_file, strerror (errno));
2504 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2505 case SSLINITFAILED: case CONTNOTSUPPORTED:
2506 /* Fatal errors just return from the function. */
2510 /* Another fatal error. */
2511 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2515 /* Return the new location to the caller. */
2518 logprintf (LOG_NOTQUIET,
2519 _("ERROR: Redirection (%d) without location.\n"),
2529 /* The file was already fully retrieved. */
2533 /* Deal with you later. */
2536 /* All possibilities should have been exhausted. */
2540 if (!(*dt & RETROKF))
2545 /* #### Ugly ugly ugly! */
2546 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2547 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2550 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2552 && (hstat.statcode == 500 || hstat.statcode == 501))
2557 /* Maybe we should always keep track of broken links, not just in
2559 else if (opt.spider)
2561 /* #### Again: ugly ugly ugly! */
2563 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2564 nonexisting_url (hurl);
2565 logprintf (LOG_NOTQUIET, _("\
2566 Remote file does not exist -- broken link!!!\n"));
2570 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2571 tms, hstat.statcode, escnonprint (hstat.error));
2573 logputs (LOG_VERBOSE, "\n");
2579 /* Did we get the time-stamp? */
2582 got_head = true; /* no more time-stamping */
2584 if (opt.timestamping && !hstat.remote_time)
2586 logputs (LOG_NOTQUIET, _("\
2587 Last-modified header missing -- time-stamps turned off.\n"));
2589 else if (hstat.remote_time)
2591 /* Convert the date-string into struct tm. */
2592 tmr = http_atotm (hstat.remote_time);
2593 if (tmr == (time_t) (-1))
2594 logputs (LOG_VERBOSE, _("\
2595 Last-modified header invalid -- time-stamp ignored.\n"));
2596 if (*dt & HEAD_ONLY)
2597 time_came_from_head = true;
2600 if (send_head_first)
2602 /* The time-stamping section. */
2603 if (opt.timestamping)
2605 if (hstat.orig_file_name) /* Perform the following
2606 checks only if the file
2608 download already exists. */
2610 if (hstat.remote_time &&
2611 tmr != (time_t) (-1))
2613 /* Now time-stamping can be used validly.
2614 Time-stamping means that if the sizes of
2615 the local and remote file match, and local
2616 file is newer than the remote file, it will
2617 not be retrieved. Otherwise, the normal
2618 download procedure is resumed. */
2619 if (hstat.orig_file_tstamp >= tmr)
2621 if (hstat.contlen == -1
2622 || hstat.orig_file_size == hstat.contlen)
2624 logprintf (LOG_VERBOSE, _("\
2625 Server file no newer than local file `%s' -- not retrieving.\n\n"),
2626 hstat.orig_file_name);
2632 logprintf (LOG_VERBOSE, _("\
2633 The sizes do not match (local %s) -- retrieving.\n"),
2634 number_to_static_string (hstat.orig_file_size));
2638 logputs (LOG_VERBOSE,
2639 _("Remote file is newer, retrieving.\n"));
2641 logputs (LOG_VERBOSE, "\n");
2645 /* free_hstat (&hstat); */
2646 hstat.timestamp_checked = true;
2655 logputs (LOG_VERBOSE, _("\
2656 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2660 logprintf (LOG_VERBOSE, _("\
2661 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2662 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2668 logprintf (LOG_VERBOSE, _("\
2669 Remote file exists but recursion is disabled -- not retrieving.\n\n"));
2670 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2677 count = 0; /* the retrieve count for HEAD is reset */
2679 } /* send_head_first */
2682 if ((tmr != (time_t) (-1))
2683 && ((hstat.len == hstat.contlen) ||
2684 ((hstat.res == 0) && (hstat.contlen == -1))))
2686 /* #### This code repeats in http.c and ftp.c. Move it to a
2688 const char *fl = NULL;
2689 if (opt.output_document)
2691 if (output_stream_regular)
2692 fl = opt.output_document;
2695 fl = hstat.local_file;
2699 /* Reparse time header, in case it's changed. */
2700 if (time_came_from_head
2701 && hstat.remote_time && hstat.remote_time[0])
2703 newtmr = http_atotm (hstat.remote_time);
2710 /* End of time-stamping section. */
2712 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2713 total_download_time += hstat.dltime;
2715 if (hstat.len == hstat.contlen)
2719 logprintf (LOG_VERBOSE,
2720 _("%s (%s) - `%s' saved [%s/%s]\n\n"),
2721 tms, tmrate, hstat.local_file,
2722 number_to_static_string (hstat.len),
2723 number_to_static_string (hstat.contlen));
2724 logprintf (LOG_NONVERBOSE,
2725 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2727 number_to_static_string (hstat.len),
2728 number_to_static_string (hstat.contlen),
2729 hstat.local_file, count);
2732 total_downloaded_bytes += hstat.len;
2734 /* Remember that we downloaded the file for later ".orig" code. */
2735 if (*dt & ADDED_HTML_EXTENSION)
2736 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2738 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2743 else if (hstat.res == 0) /* No read error */
2745 if (hstat.contlen == -1) /* We don't know how much we were supposed
2746 to get, so assume we succeeded. */
2750 logprintf (LOG_VERBOSE,
2751 _("%s (%s) - `%s' saved [%s]\n\n"),
2752 tms, tmrate, hstat.local_file,
2753 number_to_static_string (hstat.len));
2754 logprintf (LOG_NONVERBOSE,
2755 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2756 tms, u->url, number_to_static_string (hstat.len),
2757 hstat.local_file, count);
2760 total_downloaded_bytes += hstat.len;
2762 /* Remember that we downloaded the file for later ".orig" code. */
2763 if (*dt & ADDED_HTML_EXTENSION)
2764 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2766 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2771 else if (hstat.len < hstat.contlen) /* meaning we lost the
2772 connection too soon */
2774 logprintf (LOG_VERBOSE,
2775 _("%s (%s) - Connection closed at byte %s. "),
2776 tms, tmrate, number_to_static_string (hstat.len));
2777 printwhat (count, opt.ntry);
2781 /* Getting here would mean reading more data than
2782 requested with content-length, which we never do. */
2785 else /* from now on hstat.res can only be -1 */
2787 if (hstat.contlen == -1)
2789 logprintf (LOG_VERBOSE,
2790 _("%s (%s) - Read error at byte %s (%s)."),
2791 tms, tmrate, number_to_static_string (hstat.len),
2793 printwhat (count, opt.ntry);
2796 else /* hstat.res == -1 and contlen is given */
2798 logprintf (LOG_VERBOSE,
2799 _("%s (%s) - Read error at byte %s/%s (%s). "),
2801 number_to_static_string (hstat.len),
2802 number_to_static_string (hstat.contlen),
2804 printwhat (count, opt.ntry);
2810 while (!opt.ntry || (count < opt.ntry));
2814 *local_file = xstrdup (hstat.local_file);
2815 free_hstat (&hstat);
2820 /* Check whether the result of strptime() indicates success.
2821 strptime() returns the pointer to how far it got to in the string.
2822 The processing has been successful if the string is at `GMT' or
2823 `+X', or at the end of the string.
2825 In extended regexp parlance, the function returns 1 if P matches
2826 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2827 can return) is considered a failure and 0 is returned. */
2829 check_end (const char *p)
2833 while (c_isspace (*p))
2836 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2837 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2843 /* Convert the textual specification of time in TIME_STRING to the
2844 number of seconds since the Epoch.
2846 TIME_STRING can be in any of the three formats RFC2616 allows the
2847 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2848 as well as the time format used in the Set-Cookie header.
2849 Timezones are ignored, and should be GMT.
2851 Return the computed time_t representation, or -1 if the conversion
2854 This function uses strptime with various string formats for parsing
2855 TIME_STRING. This results in a parser that is not as lenient in
2856 interpreting TIME_STRING as I would like it to be. Being based on
2857 strptime, it always allows shortened months, one-digit days, etc.,
2858 but due to the multitude of formats in which time can be
2859 represented, an ideal HTTP time parser would be even more
2860 forgiving. It should completely ignore things like week days and
2861 concentrate only on the various forms of representing years,
2862 months, days, hours, minutes, and seconds. For example, it would
2863 be nice if it accepted ISO 8601 out of the box.
2865 I've investigated free and PD code for this purpose, but none was
2866 usable. getdate was big and unwieldy, and had potential copyright
2867 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2868 distributed with phttpd, is excellent, but we cannot use it because
2869 it is not assigned to the FSF. So I stuck it with strptime. */
2872 http_atotm (const char *time_string)
2874 /* NOTE: Solaris strptime man page claims that %n and %t match white
2875 space, but that's not universally available. Instead, we simply
2876 use ` ' to mean "skip all WS", which works under all strptime
2877 implementations I've tested. */
2879 static const char *time_formats[] = {
2880 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2881 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2882 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2883 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2884 (used in Set-Cookie, defined in the
2885 Netscape cookie specification.) */
2887 const char *oldlocale;
2889 time_t ret = (time_t) -1;
2891 /* Solaris strptime fails to recognize English month names in
2892 non-English locales, which we work around by temporarily setting
2893 locale to C before invoking strptime. */
2894 oldlocale = setlocale (LC_TIME, NULL);
2895 setlocale (LC_TIME, "C");
2897 for (i = 0; i < countof (time_formats); i++)
2901 /* Some versions of strptime use the existing contents of struct
2902 tm to recalculate the date according to format. Zero it out
2903 to prevent stack garbage from influencing strptime. */
2906 if (check_end (strptime (time_string, time_formats[i], &t)))
2913 /* Restore the previous locale. */
2914 setlocale (LC_TIME, oldlocale);
2919 /* Authorization support: We support three authorization schemes:
2921 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2923 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2924 consisting of answering to the server's challenge with the proper
2927 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2928 Stenberg for libcurl. Like digest, NTLM is based on a
2929 challenge-response mechanism, but unlike digest, it is non-standard
2930 (authenticates TCP connections rather than requests), undocumented
2931 and Microsoft-specific. */
2933 /* Create the authentication header contents for the `Basic' scheme.
2934 This is done by encoding the string "USER:PASS" to base64 and
2935 prepending the string "Basic " in front of it. */
2938 basic_authentication_encode (const char *user, const char *passwd)
2941 int len1 = strlen (user) + 1 + strlen (passwd);
2943 t1 = (char *)alloca (len1 + 1);
2944 sprintf (t1, "%s:%s", user, passwd);
2946 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2947 base64_encode (t1, len1, t2);
2949 return concat_strings ("Basic ", t2, (char *) 0);
2952 #define SKIP_WS(x) do { \
2953 while (c_isspace (*(x))) \
2957 #ifdef ENABLE_DIGEST
2958 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2959 an array of 16 bytes containing the hash keys, and BUF should be a
2960 buffer of 33 writable characters (32 for hex digits plus one for
2961 zero termination). */
2963 dump_hash (char *buf, const unsigned char *hash)
2967 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2969 *buf++ = XNUM_TO_digit (*hash >> 4);
2970 *buf++ = XNUM_TO_digit (*hash & 0xf);
2975 /* Take the line apart to find the challenge, and compose a digest
2976 authorization header. See RFC2069 section 2.1.2. */
2978 digest_authentication_encode (const char *au, const char *user,
2979 const char *passwd, const char *method,
2982 static char *realm, *opaque, *nonce;
2987 { "realm", &realm },
2988 { "opaque", &opaque },
2992 param_token name, value;
2994 realm = opaque = nonce = NULL;
2996 au += 6; /* skip over `Digest' */
2997 while (extract_param (&au, &name, &value, ','))
3000 for (i = 0; i < countof (options); i++)
3001 if (name.e - name.b == strlen (options[i].name)
3002 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
3004 *options[i].variable = strdupdelim (value.b, value.e);
3008 if (!realm || !nonce || !user || !passwd || !path || !method)
3011 xfree_null (opaque);
3016 /* Calculate the digest value. */
3018 ALLOCA_MD5_CONTEXT (ctx);
3019 unsigned char hash[MD5_HASHLEN];
3020 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3021 char response_digest[MD5_HASHLEN * 2 + 1];
3023 /* A1BUF = H(user ":" realm ":" password) */
3025 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3026 gen_md5_update ((unsigned char *)":", 1, ctx);
3027 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3028 gen_md5_update ((unsigned char *)":", 1, ctx);
3029 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3030 gen_md5_finish (ctx, hash);
3031 dump_hash (a1buf, hash);
3033 /* A2BUF = H(method ":" path) */
3035 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3036 gen_md5_update ((unsigned char *)":", 1, ctx);
3037 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3038 gen_md5_finish (ctx, hash);
3039 dump_hash (a2buf, hash);
3041 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3043 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3044 gen_md5_update ((unsigned char *)":", 1, ctx);
3045 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3046 gen_md5_update ((unsigned char *)":", 1, ctx);
3047 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3048 gen_md5_finish (ctx, hash);
3049 dump_hash (response_digest, hash);
3051 res = xmalloc (strlen (user)
3056 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3057 + (opaque ? strlen (opaque) : 0)
3059 sprintf (res, "Digest \
3060 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3061 user, realm, nonce, path, response_digest);
3064 char *p = res + strlen (res);
3065 strcat (p, ", opaque=\"");
3072 #endif /* ENABLE_DIGEST */
3074 /* Computing the size of a string literal must take into account that
3075 value returned by sizeof includes the terminating \0. */
3076 #define STRSIZE(literal) (sizeof (literal) - 1)
3078 /* Whether chars in [b, e) begin with the literal string provided as
3079 first argument and are followed by whitespace or terminating \0.
3080 The comparison is case-insensitive. */
3081 #define STARTS(literal, b, e) \
3082 ((e) - (b) >= STRSIZE (literal) \
3083 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3084 && ((e) - (b) == STRSIZE (literal) \
3085 || c_isspace (b[STRSIZE (literal)])))
3088 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3090 return STARTS ("Basic", hdrbeg, hdrend)
3091 #ifdef ENABLE_DIGEST
3092 || STARTS ("Digest", hdrbeg, hdrend)
3095 || STARTS ("NTLM", hdrbeg, hdrend)
3102 /* Create the HTTP authorization request header. When the
3103 `WWW-Authenticate' response header is seen, according to the
3104 authorization scheme specified in that header (`Basic' and `Digest'
3105 are supported by the current implementation), produce an
3106 appropriate HTTP authorization request header. */
3108 create_authorization_line (const char *au, const char *user,
3109 const char *passwd, const char *method,
3110 const char *path, bool *finished)
3112 /* We are called only with known schemes, so we can dispatch on the
3114 switch (c_toupper (*au))
3116 case 'B': /* Basic */
3118 return basic_authentication_encode (user, passwd);
3119 #ifdef ENABLE_DIGEST
3120 case 'D': /* Digest */
3122 return digest_authentication_encode (au, user, passwd, method, path);
3125 case 'N': /* NTLM */
3126 if (!ntlm_input (&pconn.ntlm, au))
3131 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3134 /* We shouldn't get here -- this function should be only called
3135 with values approved by known_authentication_scheme_p. */
3143 if (!wget_cookie_jar)
3144 wget_cookie_jar = cookie_jar_new ();
3145 if (opt.cookies_input && !cookies_loaded_p)
3147 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3148 cookies_loaded_p = true;
3155 if (wget_cookie_jar)
3156 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3162 xfree_null (pconn.host);
3163 if (wget_cookie_jar)
3164 cookie_jar_delete (wget_cookie_jar);
3171 test_parse_content_disposition()
3176 char *opt_dir_prefix;
3180 { "filename=\"file.ext\"", NULL, "file.ext", true },
3181 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3182 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3183 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3184 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3185 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3186 { "attachment", NULL, NULL, false },
3187 { "attachment", "somedir", NULL, false },
3190 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3195 opt.dir_prefix = test_array[i].opt_dir_prefix;
3196 res = parse_content_disposition (test_array[i].hdrval, &filename);
3198 mu_assert ("test_parse_content_disposition: wrong result",
3199 res == test_array[i].result
3201 || 0 == strcmp (test_array[i].filename, filename)));
3207 #endif /* TESTING */
3210 * vim: et sts=2 sw=2 cino+={s