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 bool do_challenge = false;
398 if (opt.auth_without_challenge)
400 DEBUGP(("Auth-without-challenge set, sending Basic credentials.\n"));
403 else if (basic_authed_hosts
404 && hash_table_contains(basic_authed_hosts, hostname))
406 DEBUGP(("Found %s in basic_authed_hosts.\n", quote (hostname)));
411 DEBUGP(("Host %s has not issued a general basic challenge.\n",
416 request_set_header (req, "Authorization",
417 basic_authentication_encode (user, passwd),
424 register_basic_auth_host (const char *hostname)
426 if (!basic_authed_hosts)
428 basic_authed_hosts = make_nocase_string_hash_table (1);
430 if (!hash_table_contains(basic_authed_hosts, hostname))
432 hash_table_put (basic_authed_hosts, xstrdup(hostname), NULL);
433 DEBUGP(("Inserted %s into basic_authed_hosts\n", quote (hostname)));
438 /* Send the contents of FILE_NAME to SOCK. Make sure that exactly
439 PROMISED_SIZE bytes are sent over the wire -- if the file is
440 longer, read only that much; if the file is shorter, report an error. */
443 post_file (int sock, const char *file_name, wgint promised_size)
445 static char chunk[8192];
450 DEBUGP (("[writing POST file %s ... ", file_name));
452 fp = fopen (file_name, "rb");
455 while (!feof (fp) && written < promised_size)
458 int length = fread (chunk, 1, sizeof (chunk), fp);
461 towrite = MIN (promised_size - written, length);
462 write_error = fd_write (sock, chunk, towrite, -1);
472 /* If we've written less than was promised, report a (probably
473 nonsensical) error rather than break the promise. */
474 if (written < promised_size)
480 assert (written == promised_size);
481 DEBUGP (("done]\n"));
485 /* Determine whether [START, PEEKED + PEEKLEN) contains an empty line.
486 If so, return the pointer to the position after the line, otherwise
487 return NULL. This is used as callback to fd_read_hunk. The data
488 between START and PEEKED has been read and cannot be "unread"; the
489 data after PEEKED has only been peeked. */
492 response_head_terminator (const char *start, const char *peeked, int peeklen)
496 /* If at first peek, verify whether HUNK starts with "HTTP". If
497 not, this is a HTTP/0.9 request and we must bail out without
499 if (start == peeked && 0 != memcmp (start, "HTTP", MIN (peeklen, 4)))
502 /* Look for "\n[\r]\n", and return the following position if found.
503 Start two chars before the current to cover the possibility that
504 part of the terminator (e.g. "\n\r") arrived in the previous
506 p = peeked - start < 2 ? start : peeked - 2;
507 end = peeked + peeklen;
509 /* Check for \n\r\n or \n\n anywhere in [p, end-2). */
510 for (; p < end - 2; p++)
513 if (p[1] == '\r' && p[2] == '\n')
515 else if (p[1] == '\n')
518 /* p==end-2: check for \n\n directly preceding END. */
519 if (p[0] == '\n' && p[1] == '\n')
525 /* The maximum size of a single HTTP response we care to read. Rather
526 than being a limit of the reader implementation, this limit
527 prevents Wget from slurping all available memory upon encountering
528 malicious or buggy server output, thus protecting the user. Define
529 it to 0 to remove the limit. */
531 #define HTTP_RESPONSE_MAX_SIZE 65536
533 /* Read the HTTP request head from FD and return it. The error
534 conditions are the same as with fd_read_hunk.
536 To support HTTP/0.9 responses, this function tries to make sure
537 that the data begins with "HTTP". If this is not the case, no data
538 is read and an empty request is returned, so that the remaining
539 data can be treated as body. */
542 read_http_response_head (int fd)
544 return fd_read_hunk (fd, response_head_terminator, 512,
545 HTTP_RESPONSE_MAX_SIZE);
549 /* The response data. */
552 /* The array of pointers that indicate where each header starts.
553 For example, given this HTTP response:
560 The headers are located like this:
562 "HTTP/1.0 200 Ok\r\nDescription: some\r\n text\r\nEtag: x\r\n\r\n"
564 headers[0] headers[1] headers[2] headers[3]
566 I.e. headers[0] points to the beginning of the request,
567 headers[1] points to the end of the first header and the
568 beginning of the second one, etc. */
570 const char **headers;
573 /* Create a new response object from the text of the HTTP response,
574 available in HEAD. That text is automatically split into
575 constituent header lines for fast retrieval using
578 static struct response *
579 resp_new (const char *head)
584 struct response *resp = xnew0 (struct response);
589 /* Empty head means that we're dealing with a headerless
590 (HTTP/0.9) response. In that case, don't set HEADERS at
595 /* Split HEAD into header lines, so that resp_header_* functions
596 don't need to do this over and over again. */
602 DO_REALLOC (resp->headers, size, count + 1, const char *);
603 resp->headers[count++] = hdr;
605 /* Break upon encountering an empty line. */
606 if (!hdr[0] || (hdr[0] == '\r' && hdr[1] == '\n') || hdr[0] == '\n')
609 /* Find the end of HDR, including continuations. */
612 const char *end = strchr (hdr, '\n');
618 while (*hdr == ' ' || *hdr == '\t');
620 DO_REALLOC (resp->headers, size, count + 1, const char *);
621 resp->headers[count] = NULL;
626 /* Locate the header named NAME in the request data, starting with
627 position START. This allows the code to loop through the request
628 data, filtering for all requests of a given name. Returns the
629 found position, or -1 for failure. The code that uses this
630 function typically looks like this:
632 for (pos = 0; (pos = resp_header_locate (...)) != -1; pos++)
633 ... do something with header ...
635 If you only care about one header, use resp_header_get instead of
639 resp_header_locate (const struct response *resp, const char *name, int start,
640 const char **begptr, const char **endptr)
643 const char **headers = resp->headers;
646 if (!headers || !headers[1])
649 name_len = strlen (name);
655 for (; headers[i + 1]; i++)
657 const char *b = headers[i];
658 const char *e = headers[i + 1];
660 && b[name_len] == ':'
661 && 0 == strncasecmp (b, name, name_len))
664 while (b < e && c_isspace (*b))
666 while (b < e && c_isspace (e[-1]))
676 /* Find and retrieve the header named NAME in the request data. If
677 found, set *BEGPTR to its starting, and *ENDPTR to its ending
678 position, and return true. Otherwise return false.
680 This function is used as a building block for resp_header_copy
681 and resp_header_strdup. */
684 resp_header_get (const struct response *resp, const char *name,
685 const char **begptr, const char **endptr)
687 int pos = resp_header_locate (resp, name, 0, begptr, endptr);
691 /* Copy the response header named NAME to buffer BUF, no longer than
692 BUFSIZE (BUFSIZE includes the terminating 0). If the header
693 exists, true is returned, false otherwise. If there should be no
694 limit on the size of the header, use resp_header_strdup instead.
696 If BUFSIZE is 0, no data is copied, but the boolean indication of
697 whether the header is present is still returned. */
700 resp_header_copy (const struct response *resp, const char *name,
701 char *buf, int bufsize)
704 if (!resp_header_get (resp, name, &b, &e))
708 int len = MIN (e - b, bufsize - 1);
709 memcpy (buf, b, len);
715 /* Return the value of header named NAME in RESP, allocated with
716 malloc. If such a header does not exist in RESP, return NULL. */
719 resp_header_strdup (const struct response *resp, const char *name)
722 if (!resp_header_get (resp, name, &b, &e))
724 return strdupdelim (b, e);
727 /* Parse the HTTP status line, which is of format:
729 HTTP-Version SP Status-Code SP Reason-Phrase
731 The function returns the status-code, or -1 if the status line
732 appears malformed. The pointer to "reason-phrase" message is
733 returned in *MESSAGE. */
736 resp_status (const struct response *resp, char **message)
743 /* For a HTTP/0.9 response, assume status 200. */
745 *message = xstrdup (_("No headers, assuming HTTP/0.9"));
749 p = resp->headers[0];
750 end = resp->headers[1];
756 if (end - p < 4 || 0 != strncmp (p, "HTTP", 4))
760 /* Match the HTTP version. This is optional because Gnutella
761 servers have been reported to not specify HTTP version. */
762 if (p < end && *p == '/')
765 while (p < end && c_isdigit (*p))
767 if (p < end && *p == '.')
769 while (p < end && c_isdigit (*p))
773 while (p < end && c_isspace (*p))
775 if (end - p < 3 || !c_isdigit (p[0]) || !c_isdigit (p[1]) || !c_isdigit (p[2]))
778 status = 100 * (p[0] - '0') + 10 * (p[1] - '0') + (p[2] - '0');
783 while (p < end && c_isspace (*p))
785 while (p < end && c_isspace (end[-1]))
787 *message = strdupdelim (p, end);
793 /* Release the resources used by RESP. */
796 resp_free (struct response *resp)
798 xfree_null (resp->headers);
802 /* Print a single line of response, the characters [b, e). We tried
804 logprintf (LOG_VERBOSE, "%s%.*s\n", prefix, (int) (e - b), b);
805 but that failed to escape the non-printable characters and, in fact,
806 caused crashes in UTF-8 locales. */
809 print_response_line(const char *prefix, const char *b, const char *e)
812 BOUNDED_TO_ALLOCA(b, e, copy);
813 logprintf (LOG_ALWAYS, "%s%s\n", prefix,
814 quotearg_style (escape_quoting_style, copy));
817 /* Print the server response, line by line, omitting the trailing CRLF
818 from individual header lines, and prefixed with PREFIX. */
821 print_server_response (const struct response *resp, const char *prefix)
826 for (i = 0; resp->headers[i + 1]; i++)
828 const char *b = resp->headers[i];
829 const char *e = resp->headers[i + 1];
831 if (b < e && e[-1] == '\n')
833 if (b < e && e[-1] == '\r')
835 print_response_line(prefix, b, e);
839 /* Parse the `Content-Range' header and extract the information it
840 contains. Returns true if successful, false otherwise. */
842 parse_content_range (const char *hdr, wgint *first_byte_ptr,
843 wgint *last_byte_ptr, wgint *entity_length_ptr)
847 /* Ancient versions of Netscape proxy server, presumably predating
848 rfc2068, sent out `Content-Range' without the "bytes"
850 if (0 == strncasecmp (hdr, "bytes", 5))
853 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
857 while (c_isspace (*hdr))
862 if (!c_isdigit (*hdr))
864 for (num = 0; c_isdigit (*hdr); hdr++)
865 num = 10 * num + (*hdr - '0');
866 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
868 *first_byte_ptr = num;
870 for (num = 0; c_isdigit (*hdr); hdr++)
871 num = 10 * num + (*hdr - '0');
872 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
874 *last_byte_ptr = num;
879 for (num = 0; c_isdigit (*hdr); hdr++)
880 num = 10 * num + (*hdr - '0');
881 *entity_length_ptr = num;
885 /* Read the body of the request, but don't store it anywhere and don't
886 display a progress gauge. This is useful for reading the bodies of
887 administrative responses to which we will soon issue another
888 request. The response is not useful to the user, but reading it
889 allows us to continue using the same connection to the server.
891 If reading fails, false is returned, true otherwise. In debug
892 mode, the body is displayed for debugging purposes. */
895 skip_short_body (int fd, wgint contlen)
898 SKIP_SIZE = 512, /* size of the download buffer */
899 SKIP_THRESHOLD = 4096 /* the largest size we read */
901 char dlbuf[SKIP_SIZE + 1];
902 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
904 /* We shouldn't get here with unknown contlen. (This will change
905 with HTTP/1.1, which supports "chunked" transfer.) */
906 assert (contlen != -1);
908 /* If the body is too large, it makes more sense to simply close the
909 connection than to try to read the body. */
910 if (contlen > SKIP_THRESHOLD)
913 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
917 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
920 /* Don't normally report the error since this is an
921 optimization that should be invisible to the user. */
922 DEBUGP (("] aborting (%s).\n",
923 ret < 0 ? fd_errstr (fd) : "EOF received"));
927 /* Safe even if %.*s bogusly expects terminating \0 because
928 we've zero-terminated dlbuf above. */
929 DEBUGP (("%.*s", ret, dlbuf));
932 DEBUGP (("] done.\n"));
936 /* Extract a parameter from the string (typically an HTTP header) at
937 **SOURCE and advance SOURCE to the next parameter. Return false
938 when there are no more parameters to extract. The name of the
939 parameter is returned in NAME, and the value in VALUE. If the
940 parameter has no value, the token's value is zeroed out.
942 For example, if *SOURCE points to the string "attachment;
943 filename=\"foo bar\"", the first call to this function will return
944 the token named "attachment" and no value, and the second call will
945 return the token named "filename" and value "foo bar". The third
946 call will return false, indicating no more valid tokens. */
949 extract_param (const char **source, param_token *name, param_token *value,
952 const char *p = *source;
954 while (c_isspace (*p)) ++p;
958 return false; /* no error; nothing more to extract */
963 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
965 if (name->b == name->e)
966 return false; /* empty name: error */
967 while (c_isspace (*p)) ++p;
968 if (*p == separator || !*p) /* no value */
971 if (*p == separator) ++p;
976 return false; /* error */
978 /* *p is '=', extract value */
980 while (c_isspace (*p)) ++p;
981 if (*p == '"') /* quoted */
984 while (*p && *p != '"') ++p;
988 /* Currently at closing quote; find the end of param. */
989 while (c_isspace (*p)) ++p;
990 while (*p && *p != separator) ++p;
994 /* garbage after closed quote, e.g. foo="bar"baz */
1000 while (*p && *p != separator) ++p;
1002 while (value->e != value->b && c_isspace (value->e[-1]))
1004 if (*p == separator) ++p;
1011 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1013 /* Parse the contents of the `Content-Disposition' header, extracting
1014 the information useful to Wget. Content-Disposition is a header
1015 borrowed from MIME; when used in HTTP, it typically serves for
1016 specifying the desired file name of the resource. For example:
1018 Content-Disposition: attachment; filename="flora.jpg"
1020 Wget will skip the tokens it doesn't care about, such as
1021 "attachment" in the previous example; it will also skip other
1022 unrecognized params. If the header is syntactically correct and
1023 contains a file name, a copy of the file name is stored in
1024 *filename and true is returned. Otherwise, the function returns
1027 The file name is stripped of directory components and must not be
1031 parse_content_disposition (const char *hdr, char **filename)
1033 param_token name, value;
1034 while (extract_param (&hdr, &name, &value, ';'))
1035 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1037 /* Make the file name begin at the last slash or backslash. */
1038 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1039 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1040 if (last_slash && last_bs)
1041 value.b = 1 + MAX (last_slash, last_bs);
1042 else if (last_slash || last_bs)
1043 value.b = 1 + (last_slash ? last_slash : last_bs);
1044 if (value.b == value.e)
1046 /* Start with the directory prefix, if specified. */
1049 int prefix_length = strlen (opt.dir_prefix);
1050 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1055 total_length = prefix_length + (value.e - value.b);
1056 *filename = xmalloc (total_length + 1);
1057 strcpy (*filename, opt.dir_prefix);
1059 (*filename)[prefix_length - 1] = '/';
1060 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1061 (*filename)[total_length] = '\0';
1064 *filename = strdupdelim (value.b, value.e);
1070 /* Persistent connections. Currently, we cache the most recently used
1071 connection as persistent, provided that the HTTP server agrees to
1072 make it such. The persistence data is stored in the variables
1073 below. Ideally, it should be possible to cache an arbitrary fixed
1074 number of these connections. */
1076 /* Whether a persistent connection is active. */
1077 static bool pconn_active;
1080 /* The socket of the connection. */
1083 /* Host and port of the currently active persistent connection. */
1087 /* Whether a ssl handshake has occoured on this connection. */
1090 /* Whether the connection was authorized. This is only done by
1091 NTLM, which authorizes *connections* rather than individual
1092 requests. (That practice is peculiar for HTTP, but it is a
1093 useful optimization.) */
1097 /* NTLM data of the current connection. */
1098 struct ntlmdata ntlm;
1102 /* Mark the persistent connection as invalid and free the resources it
1103 uses. This is used by the CLOSE_* macros after they forcefully
1104 close a registered persistent connection. */
1107 invalidate_persistent (void)
1109 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1110 pconn_active = false;
1111 fd_close (pconn.socket);
1116 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1117 persistent. This will enable someone to use the same connection
1118 later. In the context of HTTP, this must be called only AFTER the
1119 response has been received and the server has promised that the
1120 connection will remain alive.
1122 If a previous connection was persistent, it is closed. */
1125 register_persistent (const char *host, int port, int fd, bool ssl)
1129 if (pconn.socket == fd)
1131 /* The connection FD is already registered. */
1136 /* The old persistent connection is still active; close it
1137 first. This situation arises whenever a persistent
1138 connection exists, but we then connect to a different
1139 host, and try to register a persistent connection to that
1141 invalidate_persistent ();
1145 pconn_active = true;
1147 pconn.host = xstrdup (host);
1150 pconn.authorized = false;
1152 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1155 /* Return true if a persistent connection is available for connecting
1159 persistent_available_p (const char *host, int port, bool ssl,
1160 bool *host_lookup_failed)
1162 /* First, check whether a persistent connection is active at all. */
1166 /* If we want SSL and the last connection wasn't or vice versa,
1167 don't use it. Checking for host and port is not enough because
1168 HTTP and HTTPS can apparently coexist on the same port. */
1169 if (ssl != pconn.ssl)
1172 /* If we're not connecting to the same port, we're not interested. */
1173 if (port != pconn.port)
1176 /* If the host is the same, we're in business. If not, there is
1177 still hope -- read below. */
1178 if (0 != strcasecmp (host, pconn.host))
1180 /* Check if pconn.socket is talking to HOST under another name.
1181 This happens often when both sites are virtual hosts
1182 distinguished only by name and served by the same network
1183 interface, and hence the same web server (possibly set up by
1184 the ISP and serving many different web sites). This
1185 admittedly unconventional optimization does not contradict
1186 HTTP and works well with popular server software. */
1190 struct address_list *al;
1193 /* Don't try to talk to two different SSL sites over the same
1194 secure connection! (Besides, it's not clear that
1195 name-based virtual hosting is even possible with SSL.) */
1198 /* If pconn.socket's peer is one of the IP addresses HOST
1199 resolves to, pconn.socket is for all intents and purposes
1200 already talking to HOST. */
1202 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1204 /* Can't get the peer's address -- something must be very
1205 wrong with the connection. */
1206 invalidate_persistent ();
1209 al = lookup_host (host, 0);
1212 *host_lookup_failed = true;
1216 found = address_list_contains (al, &ip);
1217 address_list_release (al);
1222 /* The persistent connection's peer address was found among the
1223 addresses HOST resolved to; therefore, pconn.sock is in fact
1224 already talking to HOST -- no need to reconnect. */
1227 /* Finally, check whether the connection is still open. This is
1228 important because most servers implement liberal (short) timeout
1229 on persistent connections. Wget can of course always reconnect
1230 if the connection doesn't work out, but it's nicer to know in
1231 advance. This test is a logical followup of the first test, but
1232 is "expensive" and therefore placed at the end of the list.
1234 (Current implementation of test_socket_open has a nice side
1235 effect that it treats sockets with pending data as "closed".
1236 This is exactly what we want: if a broken server sends message
1237 body in response to HEAD, or if it sends more than conent-length
1238 data, we won't reuse the corrupted connection.) */
1240 if (!test_socket_open (pconn.socket))
1242 /* Oops, the socket is no longer open. Now that we know that,
1243 let's invalidate the persistent connection before returning
1245 invalidate_persistent ();
1252 /* The idea behind these two CLOSE macros is to distinguish between
1253 two cases: one when the job we've been doing is finished, and we
1254 want to close the connection and leave, and two when something is
1255 seriously wrong and we're closing the connection as part of
1258 In case of keep_alive, CLOSE_FINISH should leave the connection
1259 open, while CLOSE_INVALIDATE should still close it.
1261 Note that the semantics of the flag `keep_alive' is "this
1262 connection *will* be reused (the server has promised not to close
1263 the connection once we're done)", while the semantics of
1264 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1265 active, registered connection". */
1267 #define CLOSE_FINISH(fd) do { \
1270 if (pconn_active && (fd) == pconn.socket) \
1271 invalidate_persistent (); \
1280 #define CLOSE_INVALIDATE(fd) do { \
1281 if (pconn_active && (fd) == pconn.socket) \
1282 invalidate_persistent (); \
1290 wgint len; /* received length */
1291 wgint contlen; /* expected length */
1292 wgint restval; /* the restart value */
1293 int res; /* the result of last read */
1294 char *rderrmsg; /* error message from read error */
1295 char *newloc; /* new location (redirection) */
1296 char *remote_time; /* remote time-stamp string */
1297 char *error; /* textual HTTP error */
1298 int statcode; /* status code */
1299 wgint rd_size; /* amount of data read from socket */
1300 double dltime; /* time it took to download the data */
1301 const char *referer; /* value of the referer header. */
1302 char *local_file; /* local file name. */
1303 bool existence_checked; /* true if we already checked for a file's
1304 existence after having begun to download
1305 (needed in gethttp for when connection is
1306 interrupted/restarted. */
1307 bool timestamp_checked; /* true if pre-download time-stamping checks
1308 * have already been performed */
1309 char *orig_file_name; /* name of file to compare for time-stamping
1310 * (might be != local_file if -K is set) */
1311 wgint orig_file_size; /* size of file to compare for time-stamping */
1312 time_t orig_file_tstamp; /* time-stamp of file to compare for
1317 free_hstat (struct http_stat *hs)
1319 xfree_null (hs->newloc);
1320 xfree_null (hs->remote_time);
1321 xfree_null (hs->error);
1322 xfree_null (hs->rderrmsg);
1323 xfree_null (hs->local_file);
1324 xfree_null (hs->orig_file_name);
1326 /* Guard against being called twice. */
1328 hs->remote_time = NULL;
1332 #define BEGINS_WITH(line, string_constant) \
1333 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1334 && (c_isspace (line[sizeof (string_constant) - 1]) \
1335 || !line[sizeof (string_constant) - 1]))
1337 #define SET_USER_AGENT(req) do { \
1338 if (!opt.useragent) \
1339 request_set_header (req, "User-Agent", \
1340 aprintf ("Wget/%s", version_string), rel_value); \
1341 else if (*opt.useragent) \
1342 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1345 /* The flags that allow clobbering the file (opening with "wb").
1346 Defined here to avoid repetition later. #### This will require
1348 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1349 || opt.dirstruct || opt.output_document)
1351 /* Retrieve a document through HTTP protocol. It recognizes status
1352 code, and correctly handles redirections. It closes the network
1353 socket. If it receives an error from the functions below it, it
1354 will print it if there is enough information to do so (almost
1355 always), returning the error to the caller (i.e. http_loop).
1357 Various HTTP parameters are stored to hs.
1359 If PROXY is non-NULL, the connection will be made to the proxy
1360 server, and u->url will be requested. */
1362 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1364 struct request *req;
1367 char *user, *passwd;
1371 wgint contlen, contrange;
1378 /* Set to 1 when the authorization has already been sent and should
1379 not be tried again. */
1380 bool auth_finished = false;
1382 /* Set to 1 when just globally-set Basic authorization has been sent;
1383 * should prevent further Basic negotiations, but not other
1385 bool basic_auth_finished = false;
1387 /* Whether NTLM authentication is used for this request. */
1388 bool ntlm_seen = false;
1390 /* Whether our connection to the remote host is through SSL. */
1391 bool using_ssl = false;
1393 /* Whether a HEAD request will be issued (as opposed to GET or
1395 bool head_only = !!(*dt & HEAD_ONLY);
1398 struct response *resp;
1402 /* Whether this connection will be kept alive after the HTTP request
1406 /* Whether keep-alive should be inhibited.
1408 RFC 2068 requests that 1.0 clients not send keep-alive requests
1409 to proxies. This is because many 1.0 proxies do not interpret
1410 the Connection header and transfer it to the remote server,
1411 causing it to not close the connection and leave both the proxy
1412 and the client hanging. */
1413 bool inhibit_keep_alive =
1414 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1416 /* Headers sent when using POST. */
1417 wgint post_data_size = 0;
1419 bool host_lookup_failed = false;
1422 if (u->scheme == SCHEME_HTTPS)
1424 /* Initialize the SSL context. After this has once been done,
1425 it becomes a no-op. */
1428 scheme_disable (SCHEME_HTTPS);
1429 logprintf (LOG_NOTQUIET,
1430 _("Disabling SSL due to encountered errors.\n"));
1431 return SSLINITFAILED;
1434 #endif /* HAVE_SSL */
1436 /* Initialize certain elements of struct http_stat. */
1440 hs->rderrmsg = NULL;
1442 hs->remote_time = NULL;
1447 /* Prepare the request to send. */
1449 req = request_new ();
1452 const char *meth = "GET";
1455 else if (opt.post_file_name || opt.post_data)
1457 /* Use the full path, i.e. one that includes the leading slash and
1458 the query string. E.g. if u->path is "foo/bar" and u->query is
1459 "param=value", full_path will be "/foo/bar?param=value". */
1462 /* When using SSL over proxy, CONNECT establishes a direct
1463 connection to the HTTPS server. Therefore use the same
1464 argument as when talking to the server directly. */
1465 && u->scheme != SCHEME_HTTPS
1468 meth_arg = xstrdup (u->url);
1470 meth_arg = url_full_path (u);
1471 request_set_method (req, meth, meth_arg);
1474 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1475 if (*dt & SEND_NOCACHE)
1476 request_set_header (req, "Pragma", "no-cache", rel_none);
1478 request_set_header (req, "Range",
1479 aprintf ("bytes=%s-",
1480 number_to_static_string (hs->restval)),
1482 SET_USER_AGENT (req);
1483 request_set_header (req, "Accept", "*/*", rel_none);
1485 /* Find the username and password for authentication. */
1488 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1489 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1490 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1493 && !u->user) /* We only do "site-wide" authentication with "global"
1494 user/password values; URL user/password info overrides. */
1496 /* If this is a host for which we've already received a Basic
1497 * challenge, we'll go ahead and send Basic authentication creds. */
1498 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1501 /* Generate the Host header, HOST:PORT. Take into account that:
1503 - Broken server-side software often doesn't recognize the PORT
1504 argument, so we must generate "Host: www.server.com" instead of
1505 "Host: www.server.com:80" (and likewise for https port).
1507 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1508 becomes ambiguous and needs to be rewritten as "Host:
1509 [3ffe:8100:200:2::2]:1234". */
1511 /* Formats arranged for hfmt[add_port][add_squares]. */
1512 static const char *hfmt[][2] = {
1513 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1515 int add_port = u->port != scheme_default_port (u->scheme);
1516 int add_squares = strchr (u->host, ':') != NULL;
1517 request_set_header (req, "Host",
1518 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1522 if (!inhibit_keep_alive)
1523 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1526 request_set_header (req, "Cookie",
1527 cookie_header (wget_cookie_jar,
1528 u->host, u->port, u->path,
1530 u->scheme == SCHEME_HTTPS
1537 if (opt.post_data || opt.post_file_name)
1539 request_set_header (req, "Content-Type",
1540 "application/x-www-form-urlencoded", rel_none);
1542 post_data_size = strlen (opt.post_data);
1545 post_data_size = file_size (opt.post_file_name);
1546 if (post_data_size == -1)
1548 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1549 quote (opt.post_file_name), strerror (errno));
1553 request_set_header (req, "Content-Length",
1554 xstrdup (number_to_static_string (post_data_size)),
1558 /* Add the user headers. */
1559 if (opt.user_headers)
1562 for (i = 0; opt.user_headers[i]; i++)
1563 request_set_user_header (req, opt.user_headers[i]);
1567 /* We need to come back here when the initial attempt to retrieve
1568 without authorization header fails. (Expected to happen at least
1569 for the Digest authorization scheme.) */
1574 char *proxy_user, *proxy_passwd;
1575 /* For normal username and password, URL components override
1576 command-line/wgetrc parameters. With proxy
1577 authentication, it's the reverse, because proxy URLs are
1578 normally the "permanent" ones, so command-line args
1579 should take precedence. */
1580 if (opt.proxy_user && opt.proxy_passwd)
1582 proxy_user = opt.proxy_user;
1583 proxy_passwd = opt.proxy_passwd;
1587 proxy_user = proxy->user;
1588 proxy_passwd = proxy->passwd;
1590 /* #### This does not appear right. Can't the proxy request,
1591 say, `Digest' authentication? */
1592 if (proxy_user && proxy_passwd)
1593 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1595 /* If we're using a proxy, we will be connecting to the proxy
1599 /* Proxy authorization over SSL is handled below. */
1601 if (u->scheme != SCHEME_HTTPS)
1603 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1608 /* Establish the connection. */
1610 if (!inhibit_keep_alive)
1612 /* Look for a persistent connection to target host, unless a
1613 proxy is used. The exception is when SSL is in use, in which
1614 case the proxy is nothing but a passthrough to the target
1615 host, registered as a connection to the latter. */
1616 struct url *relevant = conn;
1618 if (u->scheme == SCHEME_HTTPS)
1622 if (persistent_available_p (relevant->host, relevant->port,
1624 relevant->scheme == SCHEME_HTTPS,
1628 &host_lookup_failed))
1630 sock = pconn.socket;
1631 using_ssl = pconn.ssl;
1632 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1633 quotearg_style (escape_quoting_style, pconn.host),
1635 DEBUGP (("Reusing fd %d.\n", sock));
1636 if (pconn.authorized)
1637 /* If the connection is already authorized, the "Basic"
1638 authorization added by code above is unnecessary and
1640 request_remove_header (req, "Authorization");
1642 else if (host_lookup_failed)
1645 logprintf(LOG_NOTQUIET,
1646 _("%s: unable to resolve host address %s\n"),
1647 exec_name, quote (relevant->host));
1654 sock = connect_to_host (conn->host, conn->port);
1663 return (retryable_socket_connect_error (errno)
1664 ? CONERROR : CONIMPOSSIBLE);
1668 if (proxy && u->scheme == SCHEME_HTTPS)
1670 /* When requesting SSL URLs through proxies, use the
1671 CONNECT method to request passthrough. */
1672 struct request *connreq = request_new ();
1673 request_set_method (connreq, "CONNECT",
1674 aprintf ("%s:%d", u->host, u->port));
1675 SET_USER_AGENT (connreq);
1678 request_set_header (connreq, "Proxy-Authorization",
1679 proxyauth, rel_value);
1680 /* Now that PROXYAUTH is part of the CONNECT request,
1681 zero it out so we don't send proxy authorization with
1682 the regular request below. */
1685 /* Examples in rfc2817 use the Host header in CONNECT
1686 requests. I don't see how that gains anything, given
1687 that the contents of Host would be exactly the same as
1688 the contents of CONNECT. */
1690 write_error = request_send (connreq, sock);
1691 request_free (connreq);
1692 if (write_error < 0)
1694 CLOSE_INVALIDATE (sock);
1698 head = read_http_response_head (sock);
1701 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1703 CLOSE_INVALIDATE (sock);
1712 DEBUGP (("proxy responded with: [%s]\n", head));
1714 resp = resp_new (head);
1715 statcode = resp_status (resp, &message);
1718 if (statcode != 200)
1721 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1722 message ? quotearg_style (escape_quoting_style, message) : "?");
1723 xfree_null (message);
1726 xfree_null (message);
1728 /* SOCK is now *really* connected to u->host, so update CONN
1729 to reflect this. That way register_persistent will
1730 register SOCK as being connected to u->host:u->port. */
1734 if (conn->scheme == SCHEME_HTTPS)
1736 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1743 #endif /* HAVE_SSL */
1746 /* Send the request to server. */
1747 write_error = request_send (req, sock);
1749 if (write_error >= 0)
1753 DEBUGP (("[POST data: %s]\n", opt.post_data));
1754 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1756 else if (opt.post_file_name && post_data_size != 0)
1757 write_error = post_file (sock, opt.post_file_name, post_data_size);
1760 if (write_error < 0)
1762 CLOSE_INVALIDATE (sock);
1766 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1767 proxy ? "Proxy" : "HTTP");
1772 head = read_http_response_head (sock);
1777 logputs (LOG_NOTQUIET, _("No data received.\n"));
1778 CLOSE_INVALIDATE (sock);
1784 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1786 CLOSE_INVALIDATE (sock);
1791 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1793 resp = resp_new (head);
1795 /* Check for status line. */
1797 statcode = resp_status (resp, &message);
1798 if (!opt.server_response)
1799 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1800 message ? quotearg_style (escape_quoting_style, message) : "");
1803 logprintf (LOG_VERBOSE, "\n");
1804 print_server_response (resp, " ");
1807 /* Determine the local filename if needed. Notice that if -O is used
1808 * hstat.local_file is set by http_loop to the argument of -O. */
1809 if (!hs->local_file)
1811 /* Honor Content-Disposition whether possible. */
1812 if (!opt.content_disposition
1813 || !resp_header_copy (resp, "Content-Disposition",
1814 hdrval, sizeof (hdrval))
1815 || !parse_content_disposition (hdrval, &hs->local_file))
1817 /* The Content-Disposition header is missing or broken.
1818 * Choose unique file name according to given URL. */
1819 hs->local_file = url_file_name (u);
1823 /* TODO: perform this check only once. */
1824 if (!hs->existence_checked && file_exists_p (hs->local_file))
1826 if (opt.noclobber && !opt.output_document)
1828 /* If opt.noclobber is turned on and file already exists, do not
1829 retrieve the file. But if the output_document was given, then this
1830 test was already done and the file didn't exist. Hence the !opt.output_document */
1831 logprintf (LOG_VERBOSE, _("\
1832 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1833 /* If the file is there, we suppose it's retrieved OK. */
1836 /* #### Bogusness alert. */
1837 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1838 if (has_html_suffix_p (hs->local_file))
1841 return RETRUNNEEDED;
1843 else if (!ALLOW_CLOBBER)
1845 char *unique = unique_name (hs->local_file, true);
1846 if (unique != hs->local_file)
1847 xfree (hs->local_file);
1848 hs->local_file = unique;
1851 hs->existence_checked = true;
1853 /* Support timestamping */
1854 /* TODO: move this code out of gethttp. */
1855 if (opt.timestamping && !hs->timestamp_checked)
1857 size_t filename_len = strlen (hs->local_file);
1858 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1859 bool local_dot_orig_file_exists = false;
1860 char *local_filename = NULL;
1863 if (opt.backup_converted)
1864 /* If -K is specified, we'll act on the assumption that it was specified
1865 last time these files were downloaded as well, and instead of just
1866 comparing local file X against server file X, we'll compare local
1867 file X.orig (if extant, else X) against server file X. If -K
1868 _wasn't_ specified last time, or the server contains files called
1869 *.orig, -N will be back to not operating correctly with -k. */
1871 /* Would a single s[n]printf() call be faster? --dan
1873 Definitely not. sprintf() is horribly slow. It's a
1874 different question whether the difference between the two
1875 affects a program. Usually I'd say "no", but at one
1876 point I profiled Wget, and found that a measurable and
1877 non-negligible amount of time was lost calling sprintf()
1878 in url.c. Replacing sprintf with inline calls to
1879 strcpy() and number_to_string() made a difference.
1881 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1882 memcpy (filename_plus_orig_suffix + filename_len,
1883 ".orig", sizeof (".orig"));
1885 /* Try to stat() the .orig file. */
1886 if (stat (filename_plus_orig_suffix, &st) == 0)
1888 local_dot_orig_file_exists = true;
1889 local_filename = filename_plus_orig_suffix;
1893 if (!local_dot_orig_file_exists)
1894 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1895 if (stat (hs->local_file, &st) == 0)
1896 local_filename = hs->local_file;
1898 if (local_filename != NULL)
1899 /* There was a local file, so we'll check later to see if the version
1900 the server has is the same version we already have, allowing us to
1903 hs->orig_file_name = xstrdup (local_filename);
1904 hs->orig_file_size = st.st_size;
1905 hs->orig_file_tstamp = st.st_mtime;
1907 /* Modification time granularity is 2 seconds for Windows, so
1908 increase local time by 1 second for later comparison. */
1909 ++hs->orig_file_tstamp;
1914 if (!opt.ignore_length
1915 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1919 parsed = str_to_wgint (hdrval, NULL, 10);
1920 if (parsed == WGINT_MAX && errno == ERANGE)
1923 #### If Content-Length is out of range, it most likely
1924 means that the file is larger than 2G and that we're
1925 compiled without LFS. In that case we should probably
1926 refuse to even attempt to download the file. */
1929 else if (parsed < 0)
1931 /* Negative Content-Length; nonsensical, so we can't
1932 assume any information about the content to receive. */
1939 /* Check for keep-alive related responses. */
1940 if (!inhibit_keep_alive && contlen != -1)
1942 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1944 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1946 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1951 /* The server has promised that it will not close the connection
1952 when we're done. This means that we can register it. */
1953 register_persistent (conn->host, conn->port, sock, using_ssl);
1955 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1957 /* Authorization is required. */
1958 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1959 CLOSE_FINISH (sock);
1961 CLOSE_INVALIDATE (sock);
1962 pconn.authorized = false;
1963 if (!auth_finished && (user && passwd))
1965 /* IIS sends multiple copies of WWW-Authenticate, one with
1966 the value "negotiate", and other(s) with data. Loop over
1967 all the occurrences and pick the one we recognize. */
1969 const char *wabeg, *waend;
1970 char *www_authenticate = NULL;
1972 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1973 &wabeg, &waend)) != -1;
1975 if (known_authentication_scheme_p (wabeg, waend))
1977 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1981 if (!www_authenticate)
1983 /* If the authentication header is missing or
1984 unrecognized, there's no sense in retrying. */
1985 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1987 else if (!basic_auth_finished
1988 || !BEGINS_WITH (www_authenticate, "Basic"))
1991 pth = url_full_path (u);
1992 request_set_header (req, "Authorization",
1993 create_authorization_line (www_authenticate,
1995 request_method (req),
1999 if (BEGINS_WITH (www_authenticate, "NTLM"))
2001 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2003 /* Need to register this host as using basic auth,
2004 * so we automatically send creds next time. */
2005 register_basic_auth_host (u->host);
2008 goto retry_with_auth;
2012 /* We already did Basic auth, and it failed. Gotta
2016 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2020 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2022 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2024 pconn.authorized = true;
2028 hs->statcode = statcode;
2030 hs->error = xstrdup (_("Malformed status line"));
2032 hs->error = xstrdup (_("(no description)"));
2034 hs->error = xstrdup (message);
2035 xfree_null (message);
2037 type = resp_header_strdup (resp, "Content-Type");
2040 char *tmp = strchr (type, ';');
2044 char *tmp2 = tmp + 1;
2047 while (tmp > type && c_isspace (tmp[-1]))
2052 if (opt.enable_iri && *tmp2 != '\0' &&
2053 (tmp = strstr (tmp2, "charset=")) != NULL)
2058 while (*tmp2 && !c_isspace (*tmp2))
2064 /* sXXXav : check given charset */
2065 logprintf (LOG_VERBOSE, "HTTP charset: `%s'\n", tmp);
2071 hs->newloc = resp_header_strdup (resp, "Location");
2072 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2074 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2078 const char *scbeg, *scend;
2079 /* The jar should have been created by now. */
2080 assert (wget_cookie_jar != NULL);
2082 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2083 &scbeg, &scend)) != -1;
2086 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2087 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2088 u->path, set_cookie);
2092 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2094 wgint first_byte_pos, last_byte_pos, entity_length;
2095 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2098 contrange = first_byte_pos;
2099 contlen = last_byte_pos - first_byte_pos + 1;
2104 /* 20x responses are counted among successful by default. */
2105 if (H_20X (statcode))
2108 /* Return if redirected. */
2109 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2111 /* RFC2068 says that in case of the 300 (multiple choices)
2112 response, the server can output a preferred URL through
2113 `Location' header; otherwise, the request should be treated
2114 like GET. So, if the location is set, it will be a
2115 redirection; otherwise, just proceed normally. */
2116 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2120 logprintf (LOG_VERBOSE,
2121 _("Location: %s%s\n"),
2122 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2123 hs->newloc ? _(" [following]") : "");
2124 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2125 CLOSE_FINISH (sock);
2127 CLOSE_INVALIDATE (sock);
2133 /* If content-type is not given, assume text/html. This is because
2134 of the multitude of broken CGI's that "forget" to generate the
2137 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2138 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2143 if (opt.html_extension && (*dt & TEXTHTML))
2144 /* -E / --html-extension / html_extension = on was specified, and this is a
2145 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2146 already the file's suffix, tack on ".html". */
2148 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2150 if (last_period_in_local_filename == NULL
2151 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2152 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2154 int local_filename_len = strlen (hs->local_file);
2155 /* Resize the local file, allowing for ".html" preceded by
2156 optional ".NUMBER". */
2157 hs->local_file = xrealloc (hs->local_file,
2158 local_filename_len + 24 + sizeof (".html"));
2159 strcpy(hs->local_file + local_filename_len, ".html");
2160 /* If clobbering is not allowed and the file, as named,
2161 exists, tack on ".NUMBER.html" instead. */
2162 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2166 sprintf (hs->local_file + local_filename_len,
2167 ".%d.html", ext_num++);
2168 while (file_exists_p (hs->local_file));
2170 *dt |= ADDED_HTML_EXTENSION;
2174 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2176 /* If `-c' is in use and the file has been fully downloaded (or
2177 the remote file has shrunk), Wget effectively requests bytes
2178 after the end of file and the server response with 416. */
2179 logputs (LOG_VERBOSE, _("\
2180 \n The file is already fully retrieved; nothing to do.\n\n"));
2181 /* In case the caller inspects. */
2184 /* Mark as successfully retrieved. */
2187 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2188 might be more bytes in the body. */
2189 return RETRUNNEEDED;
2191 if ((contrange != 0 && contrange != hs->restval)
2192 || (H_PARTIAL (statcode) && !contrange))
2194 /* The Range request was somehow misunderstood by the server.
2197 CLOSE_INVALIDATE (sock);
2203 hs->contlen = contlen + contrange;
2209 /* No need to print this output if the body won't be
2210 downloaded at all, or if the original server response is
2212 logputs (LOG_VERBOSE, _("Length: "));
2215 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2216 if (contlen + contrange >= 1024)
2217 logprintf (LOG_VERBOSE, " (%s)",
2218 human_readable (contlen + contrange));
2221 if (contlen >= 1024)
2222 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2223 number_to_static_string (contlen),
2224 human_readable (contlen));
2226 logprintf (LOG_VERBOSE, _(", %s remaining"),
2227 number_to_static_string (contlen));
2231 logputs (LOG_VERBOSE,
2232 opt.ignore_length ? _("ignored") : _("unspecified"));
2234 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2236 logputs (LOG_VERBOSE, "\n");
2240 type = NULL; /* We don't need it any more. */
2242 /* Return if we have no intention of further downloading. */
2243 if (!(*dt & RETROKF) || head_only)
2245 /* In case the caller cares to look... */
2250 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2251 servers not to send body in response to a HEAD request, and
2252 those that do will likely be caught by test_socket_open.
2253 If not, they can be worked around using
2254 `--no-http-keep-alive'. */
2255 CLOSE_FINISH (sock);
2256 else if (keep_alive && skip_short_body (sock, contlen))
2257 /* Successfully skipped the body; also keep using the socket. */
2258 CLOSE_FINISH (sock);
2260 CLOSE_INVALIDATE (sock);
2261 return RETRFINISHED;
2264 /* Open the local file. */
2267 mkalldirs (hs->local_file);
2269 rotate_backups (hs->local_file);
2271 fp = fopen (hs->local_file, "ab");
2272 else if (ALLOW_CLOBBER)
2273 fp = fopen (hs->local_file, "wb");
2276 fp = fopen_excl (hs->local_file, true);
2277 if (!fp && errno == EEXIST)
2279 /* We cannot just invent a new name and use it (which is
2280 what functions like unique_create typically do)
2281 because we told the user we'd use this name.
2282 Instead, return and retry the download. */
2283 logprintf (LOG_NOTQUIET,
2284 _("%s has sprung into existence.\n"),
2286 CLOSE_INVALIDATE (sock);
2287 return FOPEN_EXCL_ERR;
2292 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2293 CLOSE_INVALIDATE (sock);
2300 /* Print fetch message, if opt.verbose. */
2303 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2304 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2307 /* This confuses the timestamping code that checks for file size.
2308 #### The timestamping code should be smarter about file size. */
2309 if (opt.save_headers && hs->restval == 0)
2310 fwrite (head, 1, strlen (head), fp);
2312 /* Now we no longer need to store the response header. */
2315 /* Download the request body. */
2318 /* If content-length is present, read that much; otherwise, read
2319 until EOF. The HTTP spec doesn't require the server to
2320 actually close the connection when it's done sending data. */
2321 flags |= rb_read_exactly;
2322 if (hs->restval > 0 && contrange == 0)
2323 /* If the server ignored our range request, instruct fd_read_body
2324 to skip the first RESTVAL bytes of body. */
2325 flags |= rb_skip_startpos;
2326 hs->len = hs->restval;
2328 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2329 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2333 CLOSE_FINISH (sock);
2337 hs->rderrmsg = xstrdup (fd_errstr (sock));
2338 CLOSE_INVALIDATE (sock);
2345 return RETRFINISHED;
2348 /* The genuine HTTP loop! This is the part where the retrieval is
2349 retried, and retried, and retried, and... */
2351 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2352 int *dt, struct url *proxy)
2355 bool got_head = false; /* used for time-stamping and filename detection */
2356 bool time_came_from_head = false;
2357 bool got_name = false;
2360 uerr_t err, ret = TRYLIMEXC;
2361 time_t tmr = -1; /* remote time-stamp */
2362 struct http_stat hstat; /* HTTP status */
2364 bool send_head_first = true;
2366 /* Assert that no value for *LOCAL_FILE was passed. */
2367 assert (local_file == NULL || *local_file == NULL);
2369 /* Set LOCAL_FILE parameter. */
2370 if (local_file && opt.output_document)
2371 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2373 /* Reset NEWLOC parameter. */
2376 /* This used to be done in main(), but it's a better idea to do it
2377 here so that we don't go through the hoops if we're just using
2382 /* Warn on (likely bogus) wildcard usage in HTTP. */
2383 if (opt.ftp_glob && has_wildcards_p (u->path))
2384 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2386 /* Setup hstat struct. */
2388 hstat.referer = referer;
2390 if (opt.output_document)
2392 hstat.local_file = xstrdup (opt.output_document);
2395 else if (!opt.content_disposition)
2397 hstat.local_file = url_file_name (u);
2401 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2402 * screaming for some refactoring. */
2403 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2405 /* If opt.noclobber is turned on and file already exists, do not
2406 retrieve the file. But if the output_document was given, then this
2407 test was already done and the file didn't exist. Hence the !opt.output_document */
2408 logprintf (LOG_VERBOSE, _("\
2409 File %s already there; not retrieving.\n\n"),
2410 quote (hstat.local_file));
2411 /* If the file is there, we suppose it's retrieved OK. */
2414 /* #### Bogusness alert. */
2415 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2416 if (has_html_suffix_p (hstat.local_file))
2423 /* Reset the counter. */
2426 /* Reset the document type. */
2429 /* Skip preliminary HEAD request if we're not in spider mode AND
2430 * if -O was given or HTTP Content-Disposition support is disabled. */
2432 && (got_name || !opt.content_disposition))
2433 send_head_first = false;
2435 /* Send preliminary HEAD request if -N is given and we have an existing
2436 * destination file. */
2437 if (opt.timestamping
2438 && !opt.content_disposition
2439 && file_exists_p (url_file_name (u)))
2440 send_head_first = true;
2445 /* Increment the pass counter. */
2447 sleep_between_retrievals (count);
2449 /* Get the current time string. */
2450 tms = datetime_str (time (NULL));
2452 if (opt.spider && !got_head)
2453 logprintf (LOG_VERBOSE, _("\
2454 Spider mode enabled. Check if remote file exists.\n"));
2456 /* Print fetch message, if opt.verbose. */
2459 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2464 sprintf (tmp, _("(try:%2d)"), count);
2465 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2470 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2475 ws_changetitle (hurl);
2480 /* Default document type is empty. However, if spider mode is
2481 on or time-stamping is employed, HEAD_ONLY commands is
2482 encoded within *dt. */
2483 if (send_head_first && !got_head)
2488 /* Decide whether or not to restart. */
2491 && stat (hstat.local_file, &st) == 0
2492 && S_ISREG (st.st_mode))
2493 /* When -c is used, continue from on-disk size. (Can't use
2494 hstat.len even if count>1 because we don't want a failed
2495 first attempt to clobber existing data.) */
2496 hstat.restval = st.st_size;
2498 /* otherwise, continue where the previous try left off */
2499 hstat.restval = hstat.len;
2503 /* Decide whether to send the no-cache directive. We send it in
2505 a) we're using a proxy, and we're past our first retrieval.
2506 Some proxies are notorious for caching incomplete data, so
2507 we require a fresh get.
2508 b) caching is explicitly inhibited. */
2509 if ((proxy && count > 1) /* a */
2510 || !opt.allow_cache) /* b */
2511 *dt |= SEND_NOCACHE;
2513 *dt &= ~SEND_NOCACHE;
2515 /* Try fetching the document, or at least its head. */
2516 err = gethttp (u, &hstat, dt, proxy);
2519 tms = datetime_str (time (NULL));
2521 /* Get the new location (with or without the redirection). */
2523 *newloc = xstrdup (hstat.newloc);
2527 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2528 case CONERROR: case READERR: case WRITEFAILED:
2529 case RANGEERR: case FOPEN_EXCL_ERR:
2530 /* Non-fatal errors continue executing the loop, which will
2531 bring them to "while" statement at the end, to judge
2532 whether the number of tries was exceeded. */
2533 printwhat (count, opt.ntry);
2535 case FWRITEERR: case FOPENERR:
2536 /* Another fatal error. */
2537 logputs (LOG_VERBOSE, "\n");
2538 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2539 quote (hstat.local_file), strerror (errno));
2540 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2541 case SSLINITFAILED: case CONTNOTSUPPORTED:
2542 /* Fatal errors just return from the function. */
2546 /* Another fatal error. */
2547 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2551 /* Return the new location to the caller. */
2554 logprintf (LOG_NOTQUIET,
2555 _("ERROR: Redirection (%d) without location.\n"),
2565 /* The file was already fully retrieved. */
2569 /* Deal with you later. */
2572 /* All possibilities should have been exhausted. */
2576 if (!(*dt & RETROKF))
2581 /* #### Ugly ugly ugly! */
2582 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2583 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2586 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2588 && (hstat.statcode == 500 || hstat.statcode == 501))
2593 /* Maybe we should always keep track of broken links, not just in
2595 else if (opt.spider)
2597 /* #### Again: ugly ugly ugly! */
2599 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2600 nonexisting_url (hurl);
2601 logprintf (LOG_NOTQUIET, _("\
2602 Remote file does not exist -- broken link!!!\n"));
2606 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2607 tms, hstat.statcode,
2608 quotearg_style (escape_quoting_style, hstat.error));
2610 logputs (LOG_VERBOSE, "\n");
2616 /* Did we get the time-stamp? */
2619 got_head = true; /* no more time-stamping */
2621 if (opt.timestamping && !hstat.remote_time)
2623 logputs (LOG_NOTQUIET, _("\
2624 Last-modified header missing -- time-stamps turned off.\n"));
2626 else if (hstat.remote_time)
2628 /* Convert the date-string into struct tm. */
2629 tmr = http_atotm (hstat.remote_time);
2630 if (tmr == (time_t) (-1))
2631 logputs (LOG_VERBOSE, _("\
2632 Last-modified header invalid -- time-stamp ignored.\n"));
2633 if (*dt & HEAD_ONLY)
2634 time_came_from_head = true;
2637 if (send_head_first)
2639 /* The time-stamping section. */
2640 if (opt.timestamping)
2642 if (hstat.orig_file_name) /* Perform the following
2643 checks only if the file
2645 download already exists. */
2647 if (hstat.remote_time &&
2648 tmr != (time_t) (-1))
2650 /* Now time-stamping can be used validly.
2651 Time-stamping means that if the sizes of
2652 the local and remote file match, and local
2653 file is newer than the remote file, it will
2654 not be retrieved. Otherwise, the normal
2655 download procedure is resumed. */
2656 if (hstat.orig_file_tstamp >= tmr)
2658 if (hstat.contlen == -1
2659 || hstat.orig_file_size == hstat.contlen)
2661 logprintf (LOG_VERBOSE, _("\
2662 Server file no newer than local file %s -- not retrieving.\n\n"),
2663 quote (hstat.orig_file_name));
2669 logprintf (LOG_VERBOSE, _("\
2670 The sizes do not match (local %s) -- retrieving.\n"),
2671 number_to_static_string (hstat.orig_file_size));
2675 logputs (LOG_VERBOSE,
2676 _("Remote file is newer, retrieving.\n"));
2678 logputs (LOG_VERBOSE, "\n");
2682 /* free_hstat (&hstat); */
2683 hstat.timestamp_checked = true;
2692 logputs (LOG_VERBOSE, _("\
2693 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2697 logprintf (LOG_VERBOSE, _("\
2698 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2699 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2707 logprintf (LOG_VERBOSE, _("\
2708 Remote file exists and could contain further links,\n\
2709 but recursion is disabled -- not retrieving.\n\n"));
2713 logprintf (LOG_VERBOSE, _("\
2714 Remote file exists.\n\n"));
2716 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2723 count = 0; /* the retrieve count for HEAD is reset */
2725 } /* send_head_first */
2728 if ((tmr != (time_t) (-1))
2729 && ((hstat.len == hstat.contlen) ||
2730 ((hstat.res == 0) && (hstat.contlen == -1))))
2732 /* #### This code repeats in http.c and ftp.c. Move it to a
2734 const char *fl = NULL;
2735 if (opt.output_document)
2737 if (output_stream_regular)
2738 fl = opt.output_document;
2741 fl = hstat.local_file;
2745 /* Reparse time header, in case it's changed. */
2746 if (time_came_from_head
2747 && hstat.remote_time && hstat.remote_time[0])
2749 newtmr = http_atotm (hstat.remote_time);
2756 /* End of time-stamping section. */
2758 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2759 total_download_time += hstat.dltime;
2761 if (hstat.len == hstat.contlen)
2765 logprintf (LOG_VERBOSE,
2766 _("%s (%s) - %s saved [%s/%s]\n\n"),
2767 tms, tmrate, quote (hstat.local_file),
2768 number_to_static_string (hstat.len),
2769 number_to_static_string (hstat.contlen));
2770 logprintf (LOG_NONVERBOSE,
2771 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2773 number_to_static_string (hstat.len),
2774 number_to_static_string (hstat.contlen),
2775 hstat.local_file, count);
2778 total_downloaded_bytes += hstat.len;
2780 /* Remember that we downloaded the file for later ".orig" code. */
2781 if (*dt & ADDED_HTML_EXTENSION)
2782 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2784 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2789 else if (hstat.res == 0) /* No read error */
2791 if (hstat.contlen == -1) /* We don't know how much we were supposed
2792 to get, so assume we succeeded. */
2796 logprintf (LOG_VERBOSE,
2797 _("%s (%s) - %s saved [%s]\n\n"),
2798 tms, tmrate, quote (hstat.local_file),
2799 number_to_static_string (hstat.len));
2800 logprintf (LOG_NONVERBOSE,
2801 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2802 tms, u->url, number_to_static_string (hstat.len),
2803 hstat.local_file, count);
2806 total_downloaded_bytes += hstat.len;
2808 /* Remember that we downloaded the file for later ".orig" code. */
2809 if (*dt & ADDED_HTML_EXTENSION)
2810 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2812 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2817 else if (hstat.len < hstat.contlen) /* meaning we lost the
2818 connection too soon */
2820 logprintf (LOG_VERBOSE,
2821 _("%s (%s) - Connection closed at byte %s. "),
2822 tms, tmrate, number_to_static_string (hstat.len));
2823 printwhat (count, opt.ntry);
2827 /* Getting here would mean reading more data than
2828 requested with content-length, which we never do. */
2831 else /* from now on hstat.res can only be -1 */
2833 if (hstat.contlen == -1)
2835 logprintf (LOG_VERBOSE,
2836 _("%s (%s) - Read error at byte %s (%s)."),
2837 tms, tmrate, number_to_static_string (hstat.len),
2839 printwhat (count, opt.ntry);
2842 else /* hstat.res == -1 and contlen is given */
2844 logprintf (LOG_VERBOSE,
2845 _("%s (%s) - Read error at byte %s/%s (%s). "),
2847 number_to_static_string (hstat.len),
2848 number_to_static_string (hstat.contlen),
2850 printwhat (count, opt.ntry);
2856 while (!opt.ntry || (count < opt.ntry));
2860 *local_file = xstrdup (hstat.local_file);
2861 free_hstat (&hstat);
2866 /* Check whether the result of strptime() indicates success.
2867 strptime() returns the pointer to how far it got to in the string.
2868 The processing has been successful if the string is at `GMT' or
2869 `+X', or at the end of the string.
2871 In extended regexp parlance, the function returns 1 if P matches
2872 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2873 can return) is considered a failure and 0 is returned. */
2875 check_end (const char *p)
2879 while (c_isspace (*p))
2882 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2883 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2889 /* Convert the textual specification of time in TIME_STRING to the
2890 number of seconds since the Epoch.
2892 TIME_STRING can be in any of the three formats RFC2616 allows the
2893 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2894 as well as the time format used in the Set-Cookie header.
2895 Timezones are ignored, and should be GMT.
2897 Return the computed time_t representation, or -1 if the conversion
2900 This function uses strptime with various string formats for parsing
2901 TIME_STRING. This results in a parser that is not as lenient in
2902 interpreting TIME_STRING as I would like it to be. Being based on
2903 strptime, it always allows shortened months, one-digit days, etc.,
2904 but due to the multitude of formats in which time can be
2905 represented, an ideal HTTP time parser would be even more
2906 forgiving. It should completely ignore things like week days and
2907 concentrate only on the various forms of representing years,
2908 months, days, hours, minutes, and seconds. For example, it would
2909 be nice if it accepted ISO 8601 out of the box.
2911 I've investigated free and PD code for this purpose, but none was
2912 usable. getdate was big and unwieldy, and had potential copyright
2913 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2914 distributed with phttpd, is excellent, but we cannot use it because
2915 it is not assigned to the FSF. So I stuck it with strptime. */
2918 http_atotm (const char *time_string)
2920 /* NOTE: Solaris strptime man page claims that %n and %t match white
2921 space, but that's not universally available. Instead, we simply
2922 use ` ' to mean "skip all WS", which works under all strptime
2923 implementations I've tested. */
2925 static const char *time_formats[] = {
2926 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2927 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2928 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2929 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2930 (used in Set-Cookie, defined in the
2931 Netscape cookie specification.) */
2933 const char *oldlocale;
2935 time_t ret = (time_t) -1;
2937 /* Solaris strptime fails to recognize English month names in
2938 non-English locales, which we work around by temporarily setting
2939 locale to C before invoking strptime. */
2940 oldlocale = setlocale (LC_TIME, NULL);
2941 setlocale (LC_TIME, "C");
2943 for (i = 0; i < countof (time_formats); i++)
2947 /* Some versions of strptime use the existing contents of struct
2948 tm to recalculate the date according to format. Zero it out
2949 to prevent stack garbage from influencing strptime. */
2952 if (check_end (strptime (time_string, time_formats[i], &t)))
2959 /* Restore the previous locale. */
2960 setlocale (LC_TIME, oldlocale);
2965 /* Authorization support: We support three authorization schemes:
2967 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2969 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2970 consisting of answering to the server's challenge with the proper
2973 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2974 Stenberg for libcurl. Like digest, NTLM is based on a
2975 challenge-response mechanism, but unlike digest, it is non-standard
2976 (authenticates TCP connections rather than requests), undocumented
2977 and Microsoft-specific. */
2979 /* Create the authentication header contents for the `Basic' scheme.
2980 This is done by encoding the string "USER:PASS" to base64 and
2981 prepending the string "Basic " in front of it. */
2984 basic_authentication_encode (const char *user, const char *passwd)
2987 int len1 = strlen (user) + 1 + strlen (passwd);
2989 t1 = (char *)alloca (len1 + 1);
2990 sprintf (t1, "%s:%s", user, passwd);
2992 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2993 base64_encode (t1, len1, t2);
2995 return concat_strings ("Basic ", t2, (char *) 0);
2998 #define SKIP_WS(x) do { \
2999 while (c_isspace (*(x))) \
3003 #ifdef ENABLE_DIGEST
3004 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3005 an array of 16 bytes containing the hash keys, and BUF should be a
3006 buffer of 33 writable characters (32 for hex digits plus one for
3007 zero termination). */
3009 dump_hash (char *buf, const unsigned char *hash)
3013 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3015 *buf++ = XNUM_TO_digit (*hash >> 4);
3016 *buf++ = XNUM_TO_digit (*hash & 0xf);
3021 /* Take the line apart to find the challenge, and compose a digest
3022 authorization header. See RFC2069 section 2.1.2. */
3024 digest_authentication_encode (const char *au, const char *user,
3025 const char *passwd, const char *method,
3028 static char *realm, *opaque, *nonce;
3033 { "realm", &realm },
3034 { "opaque", &opaque },
3038 param_token name, value;
3040 realm = opaque = nonce = NULL;
3042 au += 6; /* skip over `Digest' */
3043 while (extract_param (&au, &name, &value, ','))
3046 size_t namelen = name.e - name.b;
3047 for (i = 0; i < countof (options); i++)
3048 if (namelen == strlen (options[i].name)
3049 && 0 == strncmp (name.b, options[i].name,
3052 *options[i].variable = strdupdelim (value.b, value.e);
3056 if (!realm || !nonce || !user || !passwd || !path || !method)
3059 xfree_null (opaque);
3064 /* Calculate the digest value. */
3066 ALLOCA_MD5_CONTEXT (ctx);
3067 unsigned char hash[MD5_HASHLEN];
3068 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3069 char response_digest[MD5_HASHLEN * 2 + 1];
3071 /* A1BUF = H(user ":" realm ":" password) */
3073 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3074 gen_md5_update ((unsigned char *)":", 1, ctx);
3075 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3076 gen_md5_update ((unsigned char *)":", 1, ctx);
3077 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3078 gen_md5_finish (ctx, hash);
3079 dump_hash (a1buf, hash);
3081 /* A2BUF = H(method ":" path) */
3083 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3084 gen_md5_update ((unsigned char *)":", 1, ctx);
3085 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3086 gen_md5_finish (ctx, hash);
3087 dump_hash (a2buf, hash);
3089 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3091 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3092 gen_md5_update ((unsigned char *)":", 1, ctx);
3093 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3094 gen_md5_update ((unsigned char *)":", 1, ctx);
3095 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3096 gen_md5_finish (ctx, hash);
3097 dump_hash (response_digest, hash);
3099 res = xmalloc (strlen (user)
3104 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3105 + (opaque ? strlen (opaque) : 0)
3107 sprintf (res, "Digest \
3108 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3109 user, realm, nonce, path, response_digest);
3112 char *p = res + strlen (res);
3113 strcat (p, ", opaque=\"");
3120 #endif /* ENABLE_DIGEST */
3122 /* Computing the size of a string literal must take into account that
3123 value returned by sizeof includes the terminating \0. */
3124 #define STRSIZE(literal) (sizeof (literal) - 1)
3126 /* Whether chars in [b, e) begin with the literal string provided as
3127 first argument and are followed by whitespace or terminating \0.
3128 The comparison is case-insensitive. */
3129 #define STARTS(literal, b, e) \
3131 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3132 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3133 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3134 || c_isspace (b[STRSIZE (literal)])))
3137 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3139 return STARTS ("Basic", hdrbeg, hdrend)
3140 #ifdef ENABLE_DIGEST
3141 || STARTS ("Digest", hdrbeg, hdrend)
3144 || STARTS ("NTLM", hdrbeg, hdrend)
3151 /* Create the HTTP authorization request header. When the
3152 `WWW-Authenticate' response header is seen, according to the
3153 authorization scheme specified in that header (`Basic' and `Digest'
3154 are supported by the current implementation), produce an
3155 appropriate HTTP authorization request header. */
3157 create_authorization_line (const char *au, const char *user,
3158 const char *passwd, const char *method,
3159 const char *path, bool *finished)
3161 /* We are called only with known schemes, so we can dispatch on the
3163 switch (c_toupper (*au))
3165 case 'B': /* Basic */
3167 return basic_authentication_encode (user, passwd);
3168 #ifdef ENABLE_DIGEST
3169 case 'D': /* Digest */
3171 return digest_authentication_encode (au, user, passwd, method, path);
3174 case 'N': /* NTLM */
3175 if (!ntlm_input (&pconn.ntlm, au))
3180 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3183 /* We shouldn't get here -- this function should be only called
3184 with values approved by known_authentication_scheme_p. */
3192 if (!wget_cookie_jar)
3193 wget_cookie_jar = cookie_jar_new ();
3194 if (opt.cookies_input && !cookies_loaded_p)
3196 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3197 cookies_loaded_p = true;
3204 if (wget_cookie_jar)
3205 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3211 xfree_null (pconn.host);
3212 if (wget_cookie_jar)
3213 cookie_jar_delete (wget_cookie_jar);
3220 test_parse_content_disposition()
3225 char *opt_dir_prefix;
3229 { "filename=\"file.ext\"", NULL, "file.ext", true },
3230 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3231 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3232 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3233 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3234 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3235 { "attachment", NULL, NULL, false },
3236 { "attachment", "somedir", NULL, false },
3239 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3244 opt.dir_prefix = test_array[i].opt_dir_prefix;
3245 res = parse_content_disposition (test_array[i].hdrval, &filename);
3247 mu_assert ("test_parse_content_disposition: wrong result",
3248 res == test_array[i].result
3250 || 0 == strcmp (test_array[i].filename, filename)));
3256 #endif /* TESTING */
3259 * vim: et sts=2 sw=2 cino+={s