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 char *message; /* status message */
1300 wgint rd_size; /* amount of data read from socket */
1301 double dltime; /* time it took to download the data */
1302 const char *referer; /* value of the referer header. */
1303 char *local_file; /* local file name. */
1304 bool existence_checked; /* true if we already checked for a file's
1305 existence after having begun to download
1306 (needed in gethttp for when connection is
1307 interrupted/restarted. */
1308 bool timestamp_checked; /* true if pre-download time-stamping checks
1309 * have already been performed */
1310 char *orig_file_name; /* name of file to compare for time-stamping
1311 * (might be != local_file if -K is set) */
1312 wgint orig_file_size; /* size of file to compare for time-stamping */
1313 time_t orig_file_tstamp; /* time-stamp of file to compare for
1318 free_hstat (struct http_stat *hs)
1320 xfree_null (hs->newloc);
1321 xfree_null (hs->remote_time);
1322 xfree_null (hs->error);
1323 xfree_null (hs->rderrmsg);
1324 xfree_null (hs->local_file);
1325 xfree_null (hs->orig_file_name);
1326 xfree_null (hs->message);
1328 /* Guard against being called twice. */
1330 hs->remote_time = NULL;
1334 #define BEGINS_WITH(line, string_constant) \
1335 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1336 && (c_isspace (line[sizeof (string_constant) - 1]) \
1337 || !line[sizeof (string_constant) - 1]))
1339 #define SET_USER_AGENT(req) do { \
1340 if (!opt.useragent) \
1341 request_set_header (req, "User-Agent", \
1342 aprintf ("Wget/%s", version_string), rel_value); \
1343 else if (*opt.useragent) \
1344 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1347 /* The flags that allow clobbering the file (opening with "wb").
1348 Defined here to avoid repetition later. #### This will require
1350 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1351 || opt.dirstruct || opt.output_document)
1353 /* Retrieve a document through HTTP protocol. It recognizes status
1354 code, and correctly handles redirections. It closes the network
1355 socket. If it receives an error from the functions below it, it
1356 will print it if there is enough information to do so (almost
1357 always), returning the error to the caller (i.e. http_loop).
1359 Various HTTP parameters are stored to hs.
1361 If PROXY is non-NULL, the connection will be made to the proxy
1362 server, and u->url will be requested. */
1364 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1366 struct request *req;
1369 char *user, *passwd;
1373 wgint contlen, contrange;
1380 /* Set to 1 when the authorization has already been sent and should
1381 not be tried again. */
1382 bool auth_finished = false;
1384 /* Set to 1 when just globally-set Basic authorization has been sent;
1385 * should prevent further Basic negotiations, but not other
1387 bool basic_auth_finished = false;
1389 /* Whether NTLM authentication is used for this request. */
1390 bool ntlm_seen = false;
1392 /* Whether our connection to the remote host is through SSL. */
1393 bool using_ssl = false;
1395 /* Whether a HEAD request will be issued (as opposed to GET or
1397 bool head_only = !!(*dt & HEAD_ONLY);
1400 struct response *resp;
1404 /* Whether this connection will be kept alive after the HTTP request
1408 /* Whether keep-alive should be inhibited.
1410 RFC 2068 requests that 1.0 clients not send keep-alive requests
1411 to proxies. This is because many 1.0 proxies do not interpret
1412 the Connection header and transfer it to the remote server,
1413 causing it to not close the connection and leave both the proxy
1414 and the client hanging. */
1415 bool inhibit_keep_alive =
1416 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1418 /* Headers sent when using POST. */
1419 wgint post_data_size = 0;
1421 bool host_lookup_failed = false;
1424 if (u->scheme == SCHEME_HTTPS)
1426 /* Initialize the SSL context. After this has once been done,
1427 it becomes a no-op. */
1430 scheme_disable (SCHEME_HTTPS);
1431 logprintf (LOG_NOTQUIET,
1432 _("Disabling SSL due to encountered errors.\n"));
1433 return SSLINITFAILED;
1436 #endif /* HAVE_SSL */
1438 /* Initialize certain elements of struct http_stat. */
1442 hs->rderrmsg = NULL;
1444 hs->remote_time = NULL;
1450 /* Prepare the request to send. */
1452 req = request_new ();
1455 const char *meth = "GET";
1458 else if (opt.post_file_name || opt.post_data)
1460 /* Use the full path, i.e. one that includes the leading slash and
1461 the query string. E.g. if u->path is "foo/bar" and u->query is
1462 "param=value", full_path will be "/foo/bar?param=value". */
1465 /* When using SSL over proxy, CONNECT establishes a direct
1466 connection to the HTTPS server. Therefore use the same
1467 argument as when talking to the server directly. */
1468 && u->scheme != SCHEME_HTTPS
1471 meth_arg = xstrdup (u->url);
1473 meth_arg = url_full_path (u);
1474 request_set_method (req, meth, meth_arg);
1477 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1478 if (*dt & SEND_NOCACHE)
1479 request_set_header (req, "Pragma", "no-cache", rel_none);
1481 request_set_header (req, "Range",
1482 aprintf ("bytes=%s-",
1483 number_to_static_string (hs->restval)),
1485 SET_USER_AGENT (req);
1486 request_set_header (req, "Accept", "*/*", rel_none);
1488 /* Find the username and password for authentication. */
1491 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1492 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1493 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1496 && !u->user) /* We only do "site-wide" authentication with "global"
1497 user/password values; URL user/password info overrides. */
1499 /* If this is a host for which we've already received a Basic
1500 * challenge, we'll go ahead and send Basic authentication creds. */
1501 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1504 /* Generate the Host header, HOST:PORT. Take into account that:
1506 - Broken server-side software often doesn't recognize the PORT
1507 argument, so we must generate "Host: www.server.com" instead of
1508 "Host: www.server.com:80" (and likewise for https port).
1510 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1511 becomes ambiguous and needs to be rewritten as "Host:
1512 [3ffe:8100:200:2::2]:1234". */
1514 /* Formats arranged for hfmt[add_port][add_squares]. */
1515 static const char *hfmt[][2] = {
1516 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1518 int add_port = u->port != scheme_default_port (u->scheme);
1519 int add_squares = strchr (u->host, ':') != NULL;
1520 request_set_header (req, "Host",
1521 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1525 if (!inhibit_keep_alive)
1526 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1529 request_set_header (req, "Cookie",
1530 cookie_header (wget_cookie_jar,
1531 u->host, u->port, u->path,
1533 u->scheme == SCHEME_HTTPS
1540 if (opt.post_data || opt.post_file_name)
1542 request_set_header (req, "Content-Type",
1543 "application/x-www-form-urlencoded", rel_none);
1545 post_data_size = strlen (opt.post_data);
1548 post_data_size = file_size (opt.post_file_name);
1549 if (post_data_size == -1)
1551 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1552 quote (opt.post_file_name), strerror (errno));
1556 request_set_header (req, "Content-Length",
1557 xstrdup (number_to_static_string (post_data_size)),
1561 /* Add the user headers. */
1562 if (opt.user_headers)
1565 for (i = 0; opt.user_headers[i]; i++)
1566 request_set_user_header (req, opt.user_headers[i]);
1570 /* We need to come back here when the initial attempt to retrieve
1571 without authorization header fails. (Expected to happen at least
1572 for the Digest authorization scheme.) */
1577 char *proxy_user, *proxy_passwd;
1578 /* For normal username and password, URL components override
1579 command-line/wgetrc parameters. With proxy
1580 authentication, it's the reverse, because proxy URLs are
1581 normally the "permanent" ones, so command-line args
1582 should take precedence. */
1583 if (opt.proxy_user && opt.proxy_passwd)
1585 proxy_user = opt.proxy_user;
1586 proxy_passwd = opt.proxy_passwd;
1590 proxy_user = proxy->user;
1591 proxy_passwd = proxy->passwd;
1593 /* #### This does not appear right. Can't the proxy request,
1594 say, `Digest' authentication? */
1595 if (proxy_user && proxy_passwd)
1596 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1598 /* If we're using a proxy, we will be connecting to the proxy
1602 /* Proxy authorization over SSL is handled below. */
1604 if (u->scheme != SCHEME_HTTPS)
1606 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1611 /* Establish the connection. */
1613 if (!inhibit_keep_alive)
1615 /* Look for a persistent connection to target host, unless a
1616 proxy is used. The exception is when SSL is in use, in which
1617 case the proxy is nothing but a passthrough to the target
1618 host, registered as a connection to the latter. */
1619 struct url *relevant = conn;
1621 if (u->scheme == SCHEME_HTTPS)
1625 if (persistent_available_p (relevant->host, relevant->port,
1627 relevant->scheme == SCHEME_HTTPS,
1631 &host_lookup_failed))
1633 sock = pconn.socket;
1634 using_ssl = pconn.ssl;
1635 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1636 quotearg_style (escape_quoting_style, pconn.host),
1638 DEBUGP (("Reusing fd %d.\n", sock));
1639 if (pconn.authorized)
1640 /* If the connection is already authorized, the "Basic"
1641 authorization added by code above is unnecessary and
1643 request_remove_header (req, "Authorization");
1645 else if (host_lookup_failed)
1648 logprintf(LOG_NOTQUIET,
1649 _("%s: unable to resolve host address %s\n"),
1650 exec_name, quote (relevant->host));
1657 sock = connect_to_host (conn->host, conn->port);
1666 return (retryable_socket_connect_error (errno)
1667 ? CONERROR : CONIMPOSSIBLE);
1671 if (proxy && u->scheme == SCHEME_HTTPS)
1673 /* When requesting SSL URLs through proxies, use the
1674 CONNECT method to request passthrough. */
1675 struct request *connreq = request_new ();
1676 request_set_method (connreq, "CONNECT",
1677 aprintf ("%s:%d", u->host, u->port));
1678 SET_USER_AGENT (connreq);
1681 request_set_header (connreq, "Proxy-Authorization",
1682 proxyauth, rel_value);
1683 /* Now that PROXYAUTH is part of the CONNECT request,
1684 zero it out so we don't send proxy authorization with
1685 the regular request below. */
1688 /* Examples in rfc2817 use the Host header in CONNECT
1689 requests. I don't see how that gains anything, given
1690 that the contents of Host would be exactly the same as
1691 the contents of CONNECT. */
1693 write_error = request_send (connreq, sock);
1694 request_free (connreq);
1695 if (write_error < 0)
1697 CLOSE_INVALIDATE (sock);
1701 head = read_http_response_head (sock);
1704 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1706 CLOSE_INVALIDATE (sock);
1715 DEBUGP (("proxy responded with: [%s]\n", head));
1717 resp = resp_new (head);
1718 statcode = resp_status (resp, &message);
1719 hs->message = xstrdup (message);
1722 if (statcode != 200)
1725 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1726 message ? quotearg_style (escape_quoting_style, message) : "?");
1727 xfree_null (message);
1730 xfree_null (message);
1732 /* SOCK is now *really* connected to u->host, so update CONN
1733 to reflect this. That way register_persistent will
1734 register SOCK as being connected to u->host:u->port. */
1738 if (conn->scheme == SCHEME_HTTPS)
1740 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1747 #endif /* HAVE_SSL */
1750 /* Send the request to server. */
1751 write_error = request_send (req, sock);
1753 if (write_error >= 0)
1757 DEBUGP (("[POST data: %s]\n", opt.post_data));
1758 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1760 else if (opt.post_file_name && post_data_size != 0)
1761 write_error = post_file (sock, opt.post_file_name, post_data_size);
1764 if (write_error < 0)
1766 CLOSE_INVALIDATE (sock);
1770 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1771 proxy ? "Proxy" : "HTTP");
1776 head = read_http_response_head (sock);
1781 logputs (LOG_NOTQUIET, _("No data received.\n"));
1782 CLOSE_INVALIDATE (sock);
1788 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1790 CLOSE_INVALIDATE (sock);
1795 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1797 resp = resp_new (head);
1799 /* Check for status line. */
1801 statcode = resp_status (resp, &message);
1802 hs->message = xstrdup (message);
1803 if (!opt.server_response)
1804 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1805 message ? quotearg_style (escape_quoting_style, message) : "");
1808 logprintf (LOG_VERBOSE, "\n");
1809 print_server_response (resp, " ");
1812 /* Determine the local filename if needed. Notice that if -O is used
1813 * hstat.local_file is set by http_loop to the argument of -O. */
1814 if (!hs->local_file)
1816 /* Honor Content-Disposition whether possible. */
1817 if (!opt.content_disposition
1818 || !resp_header_copy (resp, "Content-Disposition",
1819 hdrval, sizeof (hdrval))
1820 || !parse_content_disposition (hdrval, &hs->local_file))
1822 /* The Content-Disposition header is missing or broken.
1823 * Choose unique file name according to given URL. */
1824 hs->local_file = url_file_name (u);
1828 /* TODO: perform this check only once. */
1829 if (!hs->existence_checked && file_exists_p (hs->local_file))
1831 if (opt.noclobber && !opt.output_document)
1833 /* If opt.noclobber is turned on and file already exists, do not
1834 retrieve the file. But if the output_document was given, then this
1835 test was already done and the file didn't exist. Hence the !opt.output_document */
1836 logprintf (LOG_VERBOSE, _("\
1837 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1838 /* If the file is there, we suppose it's retrieved OK. */
1841 /* #### Bogusness alert. */
1842 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1843 if (has_html_suffix_p (hs->local_file))
1846 return RETRUNNEEDED;
1848 else if (!ALLOW_CLOBBER)
1850 char *unique = unique_name (hs->local_file, true);
1851 if (unique != hs->local_file)
1852 xfree (hs->local_file);
1853 hs->local_file = unique;
1856 hs->existence_checked = true;
1858 /* Support timestamping */
1859 /* TODO: move this code out of gethttp. */
1860 if (opt.timestamping && !hs->timestamp_checked)
1862 size_t filename_len = strlen (hs->local_file);
1863 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1864 bool local_dot_orig_file_exists = false;
1865 char *local_filename = NULL;
1868 if (opt.backup_converted)
1869 /* If -K is specified, we'll act on the assumption that it was specified
1870 last time these files were downloaded as well, and instead of just
1871 comparing local file X against server file X, we'll compare local
1872 file X.orig (if extant, else X) against server file X. If -K
1873 _wasn't_ specified last time, or the server contains files called
1874 *.orig, -N will be back to not operating correctly with -k. */
1876 /* Would a single s[n]printf() call be faster? --dan
1878 Definitely not. sprintf() is horribly slow. It's a
1879 different question whether the difference between the two
1880 affects a program. Usually I'd say "no", but at one
1881 point I profiled Wget, and found that a measurable and
1882 non-negligible amount of time was lost calling sprintf()
1883 in url.c. Replacing sprintf with inline calls to
1884 strcpy() and number_to_string() made a difference.
1886 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1887 memcpy (filename_plus_orig_suffix + filename_len,
1888 ".orig", sizeof (".orig"));
1890 /* Try to stat() the .orig file. */
1891 if (stat (filename_plus_orig_suffix, &st) == 0)
1893 local_dot_orig_file_exists = true;
1894 local_filename = filename_plus_orig_suffix;
1898 if (!local_dot_orig_file_exists)
1899 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1900 if (stat (hs->local_file, &st) == 0)
1901 local_filename = hs->local_file;
1903 if (local_filename != NULL)
1904 /* There was a local file, so we'll check later to see if the version
1905 the server has is the same version we already have, allowing us to
1908 hs->orig_file_name = xstrdup (local_filename);
1909 hs->orig_file_size = st.st_size;
1910 hs->orig_file_tstamp = st.st_mtime;
1912 /* Modification time granularity is 2 seconds for Windows, so
1913 increase local time by 1 second for later comparison. */
1914 ++hs->orig_file_tstamp;
1919 if (!opt.ignore_length
1920 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1924 parsed = str_to_wgint (hdrval, NULL, 10);
1925 if (parsed == WGINT_MAX && errno == ERANGE)
1928 #### If Content-Length is out of range, it most likely
1929 means that the file is larger than 2G and that we're
1930 compiled without LFS. In that case we should probably
1931 refuse to even attempt to download the file. */
1934 else if (parsed < 0)
1936 /* Negative Content-Length; nonsensical, so we can't
1937 assume any information about the content to receive. */
1944 /* Check for keep-alive related responses. */
1945 if (!inhibit_keep_alive && contlen != -1)
1947 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1949 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1951 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1956 /* The server has promised that it will not close the connection
1957 when we're done. This means that we can register it. */
1958 register_persistent (conn->host, conn->port, sock, using_ssl);
1960 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1962 /* Authorization is required. */
1963 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1964 CLOSE_FINISH (sock);
1966 CLOSE_INVALIDATE (sock);
1967 pconn.authorized = false;
1968 if (!auth_finished && (user && passwd))
1970 /* IIS sends multiple copies of WWW-Authenticate, one with
1971 the value "negotiate", and other(s) with data. Loop over
1972 all the occurrences and pick the one we recognize. */
1974 const char *wabeg, *waend;
1975 char *www_authenticate = NULL;
1977 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1978 &wabeg, &waend)) != -1;
1980 if (known_authentication_scheme_p (wabeg, waend))
1982 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1986 if (!www_authenticate)
1988 /* If the authentication header is missing or
1989 unrecognized, there's no sense in retrying. */
1990 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1992 else if (!basic_auth_finished
1993 || !BEGINS_WITH (www_authenticate, "Basic"))
1996 pth = url_full_path (u);
1997 request_set_header (req, "Authorization",
1998 create_authorization_line (www_authenticate,
2000 request_method (req),
2004 if (BEGINS_WITH (www_authenticate, "NTLM"))
2006 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2008 /* Need to register this host as using basic auth,
2009 * so we automatically send creds next time. */
2010 register_basic_auth_host (u->host);
2013 goto retry_with_auth;
2017 /* We already did Basic auth, and it failed. Gotta
2021 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2025 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2027 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2029 pconn.authorized = true;
2033 hs->statcode = statcode;
2035 hs->error = xstrdup (_("Malformed status line"));
2037 hs->error = xstrdup (_("(no description)"));
2039 hs->error = xstrdup (message);
2040 xfree_null (message);
2042 type = resp_header_strdup (resp, "Content-Type");
2045 char *tmp = strchr (type, ';');
2048 while (tmp > type && c_isspace (tmp[-1]))
2053 hs->newloc = resp_header_strdup (resp, "Location");
2054 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2056 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2060 const char *scbeg, *scend;
2061 /* The jar should have been created by now. */
2062 assert (wget_cookie_jar != NULL);
2064 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2065 &scbeg, &scend)) != -1;
2068 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2069 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2070 u->path, set_cookie);
2074 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2076 wgint first_byte_pos, last_byte_pos, entity_length;
2077 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2080 contrange = first_byte_pos;
2081 contlen = last_byte_pos - first_byte_pos + 1;
2086 /* 20x responses are counted among successful by default. */
2087 if (H_20X (statcode))
2090 /* Return if redirected. */
2091 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2093 /* RFC2068 says that in case of the 300 (multiple choices)
2094 response, the server can output a preferred URL through
2095 `Location' header; otherwise, the request should be treated
2096 like GET. So, if the location is set, it will be a
2097 redirection; otherwise, just proceed normally. */
2098 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2102 logprintf (LOG_VERBOSE,
2103 _("Location: %s%s\n"),
2104 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2105 hs->newloc ? _(" [following]") : "");
2106 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2107 CLOSE_FINISH (sock);
2109 CLOSE_INVALIDATE (sock);
2115 /* If content-type is not given, assume text/html. This is because
2116 of the multitude of broken CGI's that "forget" to generate the
2119 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2120 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2125 if (opt.html_extension && (*dt & TEXTHTML))
2126 /* -E / --html-extension / html_extension = on was specified, and this is a
2127 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2128 already the file's suffix, tack on ".html". */
2130 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2132 if (last_period_in_local_filename == NULL
2133 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2134 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2136 int local_filename_len = strlen (hs->local_file);
2137 /* Resize the local file, allowing for ".html" preceded by
2138 optional ".NUMBER". */
2139 hs->local_file = xrealloc (hs->local_file,
2140 local_filename_len + 24 + sizeof (".html"));
2141 strcpy(hs->local_file + local_filename_len, ".html");
2142 /* If clobbering is not allowed and the file, as named,
2143 exists, tack on ".NUMBER.html" instead. */
2144 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2148 sprintf (hs->local_file + local_filename_len,
2149 ".%d.html", ext_num++);
2150 while (file_exists_p (hs->local_file));
2152 *dt |= ADDED_HTML_EXTENSION;
2156 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2158 /* If `-c' is in use and the file has been fully downloaded (or
2159 the remote file has shrunk), Wget effectively requests bytes
2160 after the end of file and the server response with 416. */
2161 logputs (LOG_VERBOSE, _("\
2162 \n The file is already fully retrieved; nothing to do.\n\n"));
2163 /* In case the caller inspects. */
2166 /* Mark as successfully retrieved. */
2169 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2170 might be more bytes in the body. */
2171 return RETRUNNEEDED;
2173 if ((contrange != 0 && contrange != hs->restval)
2174 || (H_PARTIAL (statcode) && !contrange))
2176 /* The Range request was somehow misunderstood by the server.
2179 CLOSE_INVALIDATE (sock);
2185 hs->contlen = contlen + contrange;
2191 /* No need to print this output if the body won't be
2192 downloaded at all, or if the original server response is
2194 logputs (LOG_VERBOSE, _("Length: "));
2197 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2198 if (contlen + contrange >= 1024)
2199 logprintf (LOG_VERBOSE, " (%s)",
2200 human_readable (contlen + contrange));
2203 if (contlen >= 1024)
2204 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2205 number_to_static_string (contlen),
2206 human_readable (contlen));
2208 logprintf (LOG_VERBOSE, _(", %s remaining"),
2209 number_to_static_string (contlen));
2213 logputs (LOG_VERBOSE,
2214 opt.ignore_length ? _("ignored") : _("unspecified"));
2216 logprintf (LOG_VERBOSE, " [%s]\n", quotearg_style (escape_quoting_style, type));
2218 logputs (LOG_VERBOSE, "\n");
2222 type = NULL; /* We don't need it any more. */
2224 /* Return if we have no intention of further downloading. */
2225 if (!(*dt & RETROKF) || head_only)
2227 /* In case the caller cares to look... */
2232 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2233 servers not to send body in response to a HEAD request, and
2234 those that do will likely be caught by test_socket_open.
2235 If not, they can be worked around using
2236 `--no-http-keep-alive'. */
2237 CLOSE_FINISH (sock);
2238 else if (keep_alive && skip_short_body (sock, contlen))
2239 /* Successfully skipped the body; also keep using the socket. */
2240 CLOSE_FINISH (sock);
2242 CLOSE_INVALIDATE (sock);
2243 return RETRFINISHED;
2246 /* Open the local file. */
2249 mkalldirs (hs->local_file);
2251 rotate_backups (hs->local_file);
2253 fp = fopen (hs->local_file, "ab");
2254 else if (ALLOW_CLOBBER)
2255 fp = fopen (hs->local_file, "wb");
2258 fp = fopen_excl (hs->local_file, true);
2259 if (!fp && errno == EEXIST)
2261 /* We cannot just invent a new name and use it (which is
2262 what functions like unique_create typically do)
2263 because we told the user we'd use this name.
2264 Instead, return and retry the download. */
2265 logprintf (LOG_NOTQUIET,
2266 _("%s has sprung into existence.\n"),
2268 CLOSE_INVALIDATE (sock);
2269 return FOPEN_EXCL_ERR;
2274 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2275 CLOSE_INVALIDATE (sock);
2282 /* Print fetch message, if opt.verbose. */
2285 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2286 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2289 /* This confuses the timestamping code that checks for file size.
2290 #### The timestamping code should be smarter about file size. */
2291 if (opt.save_headers && hs->restval == 0)
2292 fwrite (head, 1, strlen (head), fp);
2294 /* Now we no longer need to store the response header. */
2297 /* Download the request body. */
2300 /* If content-length is present, read that much; otherwise, read
2301 until EOF. The HTTP spec doesn't require the server to
2302 actually close the connection when it's done sending data. */
2303 flags |= rb_read_exactly;
2304 if (hs->restval > 0 && contrange == 0)
2305 /* If the server ignored our range request, instruct fd_read_body
2306 to skip the first RESTVAL bytes of body. */
2307 flags |= rb_skip_startpos;
2308 hs->len = hs->restval;
2310 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2311 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2315 CLOSE_FINISH (sock);
2319 hs->rderrmsg = xstrdup (fd_errstr (sock));
2320 CLOSE_INVALIDATE (sock);
2327 return RETRFINISHED;
2330 /* The genuine HTTP loop! This is the part where the retrieval is
2331 retried, and retried, and retried, and... */
2333 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2334 int *dt, struct url *proxy)
2337 bool got_head = false; /* used for time-stamping and filename detection */
2338 bool time_came_from_head = false;
2339 bool got_name = false;
2342 uerr_t err, ret = TRYLIMEXC;
2343 time_t tmr = -1; /* remote time-stamp */
2344 struct http_stat hstat; /* HTTP status */
2346 bool send_head_first = true;
2348 /* Assert that no value for *LOCAL_FILE was passed. */
2349 assert (local_file == NULL || *local_file == NULL);
2351 /* Set LOCAL_FILE parameter. */
2352 if (local_file && opt.output_document)
2353 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2355 /* Reset NEWLOC parameter. */
2358 /* This used to be done in main(), but it's a better idea to do it
2359 here so that we don't go through the hoops if we're just using
2364 /* Warn on (likely bogus) wildcard usage in HTTP. */
2365 if (opt.ftp_glob && has_wildcards_p (u->path))
2366 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2368 /* Setup hstat struct. */
2370 hstat.referer = referer;
2372 if (opt.output_document)
2374 hstat.local_file = xstrdup (opt.output_document);
2377 else if (!opt.content_disposition)
2379 hstat.local_file = url_file_name (u);
2383 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2384 * screaming for some refactoring. */
2385 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber && !opt.output_document)
2387 /* If opt.noclobber is turned on and file already exists, do not
2388 retrieve the file. But if the output_document was given, then this
2389 test was already done and the file didn't exist. Hence the !opt.output_document */
2390 logprintf (LOG_VERBOSE, _("\
2391 File %s already there; not retrieving.\n\n"),
2392 quote (hstat.local_file));
2393 /* If the file is there, we suppose it's retrieved OK. */
2396 /* #### Bogusness alert. */
2397 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2398 if (has_html_suffix_p (hstat.local_file))
2405 /* Reset the counter. */
2408 /* Reset the document type. */
2411 /* Skip preliminary HEAD request if we're not in spider mode AND
2412 * if -O was given or HTTP Content-Disposition support is disabled. */
2414 && (got_name || !opt.content_disposition))
2415 send_head_first = false;
2417 /* Send preliminary HEAD request if -N is given and we have an existing
2418 * destination file. */
2419 if (opt.timestamping
2420 && !opt.content_disposition
2421 && file_exists_p (url_file_name (u)))
2422 send_head_first = true;
2427 /* Increment the pass counter. */
2429 sleep_between_retrievals (count);
2431 /* Get the current time string. */
2432 tms = datetime_str (time (NULL));
2434 if (opt.spider && !got_head)
2435 logprintf (LOG_VERBOSE, _("\
2436 Spider mode enabled. Check if remote file exists.\n"));
2438 /* Print fetch message, if opt.verbose. */
2441 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2446 sprintf (tmp, _("(try:%2d)"), count);
2447 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2452 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2457 ws_changetitle (hurl);
2462 /* Default document type is empty. However, if spider mode is
2463 on or time-stamping is employed, HEAD_ONLY commands is
2464 encoded within *dt. */
2465 if (send_head_first && !got_head)
2470 /* Decide whether or not to restart. */
2473 && stat (hstat.local_file, &st) == 0
2474 && S_ISREG (st.st_mode))
2475 /* When -c is used, continue from on-disk size. (Can't use
2476 hstat.len even if count>1 because we don't want a failed
2477 first attempt to clobber existing data.) */
2478 hstat.restval = st.st_size;
2480 /* otherwise, continue where the previous try left off */
2481 hstat.restval = hstat.len;
2485 /* Decide whether to send the no-cache directive. We send it in
2487 a) we're using a proxy, and we're past our first retrieval.
2488 Some proxies are notorious for caching incomplete data, so
2489 we require a fresh get.
2490 b) caching is explicitly inhibited. */
2491 if ((proxy && count > 1) /* a */
2492 || !opt.allow_cache) /* b */
2493 *dt |= SEND_NOCACHE;
2495 *dt &= ~SEND_NOCACHE;
2497 /* Try fetching the document, or at least its head. */
2498 err = gethttp (u, &hstat, dt, proxy);
2501 tms = datetime_str (time (NULL));
2503 /* Get the new location (with or without the redirection). */
2505 *newloc = xstrdup (hstat.newloc);
2509 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2510 case CONERROR: case READERR: case WRITEFAILED:
2511 case RANGEERR: case FOPEN_EXCL_ERR:
2512 /* Non-fatal errors continue executing the loop, which will
2513 bring them to "while" statement at the end, to judge
2514 whether the number of tries was exceeded. */
2515 printwhat (count, opt.ntry);
2517 case FWRITEERR: case FOPENERR:
2518 /* Another fatal error. */
2519 logputs (LOG_VERBOSE, "\n");
2520 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2521 quote (hstat.local_file), strerror (errno));
2522 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2523 case SSLINITFAILED: case CONTNOTSUPPORTED:
2524 /* Fatal errors just return from the function. */
2528 /* Another fatal error. */
2529 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2533 /* Return the new location to the caller. */
2536 logprintf (LOG_NOTQUIET,
2537 _("ERROR: Redirection (%d) without location.\n"),
2547 /* The file was already fully retrieved. */
2551 /* Deal with you later. */
2554 /* All possibilities should have been exhausted. */
2558 if (!(*dt & RETROKF))
2563 /* #### Ugly ugly ugly! */
2564 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2565 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2568 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2570 && (hstat.statcode == 500 || hstat.statcode == 501))
2575 /* Maybe we should always keep track of broken links, not just in
2577 else if (opt.spider)
2579 /* #### Again: ugly ugly ugly! */
2581 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2582 nonexisting_url (hurl);
2583 logprintf (LOG_NOTQUIET, _("\
2584 Remote file does not exist -- broken link!!!\n"));
2588 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2589 tms, hstat.statcode,
2590 quotearg_style (escape_quoting_style, hstat.error));
2592 logputs (LOG_VERBOSE, "\n");
2598 /* Did we get the time-stamp? */
2601 got_head = true; /* no more time-stamping */
2603 if (opt.timestamping && !hstat.remote_time)
2605 logputs (LOG_NOTQUIET, _("\
2606 Last-modified header missing -- time-stamps turned off.\n"));
2608 else if (hstat.remote_time)
2610 /* Convert the date-string into struct tm. */
2611 tmr = http_atotm (hstat.remote_time);
2612 if (tmr == (time_t) (-1))
2613 logputs (LOG_VERBOSE, _("\
2614 Last-modified header invalid -- time-stamp ignored.\n"));
2615 if (*dt & HEAD_ONLY)
2616 time_came_from_head = true;
2619 if (send_head_first)
2621 /* The time-stamping section. */
2622 if (opt.timestamping)
2624 if (hstat.orig_file_name) /* Perform the following
2625 checks only if the file
2627 download already exists. */
2629 if (hstat.remote_time &&
2630 tmr != (time_t) (-1))
2632 /* Now time-stamping can be used validly.
2633 Time-stamping means that if the sizes of
2634 the local and remote file match, and local
2635 file is newer than the remote file, it will
2636 not be retrieved. Otherwise, the normal
2637 download procedure is resumed. */
2638 if (hstat.orig_file_tstamp >= tmr)
2640 if (hstat.contlen == -1
2641 || hstat.orig_file_size == hstat.contlen)
2643 logprintf (LOG_VERBOSE, _("\
2644 Server file no newer than local file %s -- not retrieving.\n\n"),
2645 quote (hstat.orig_file_name));
2651 logprintf (LOG_VERBOSE, _("\
2652 The sizes do not match (local %s) -- retrieving.\n"),
2653 number_to_static_string (hstat.orig_file_size));
2657 logputs (LOG_VERBOSE,
2658 _("Remote file is newer, retrieving.\n"));
2660 logputs (LOG_VERBOSE, "\n");
2664 /* free_hstat (&hstat); */
2665 hstat.timestamp_checked = true;
2670 bool finished = true;
2675 logputs (LOG_VERBOSE, _("\
2676 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2681 logprintf (LOG_VERBOSE, _("\
2682 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2683 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2690 logprintf (LOG_VERBOSE, _("\
2691 Remote file exists and could contain further links,\n\
2692 but recursion is disabled -- not retrieving.\n\n"));
2696 logprintf (LOG_VERBOSE, _("\
2697 Remote file exists.\n\n"));
2699 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2704 logprintf (LOG_NONVERBOSE,
2705 _("%s URL:%s %2d %s\n"),
2706 tms, u->url, hstat.statcode,
2707 hstat.message ? escnonprint (hstat.message) : "");
2714 count = 0; /* the retrieve count for HEAD is reset */
2716 } /* send_head_first */
2719 if ((tmr != (time_t) (-1))
2720 && ((hstat.len == hstat.contlen) ||
2721 ((hstat.res == 0) && (hstat.contlen == -1))))
2723 /* #### This code repeats in http.c and ftp.c. Move it to a
2725 const char *fl = NULL;
2726 if (opt.output_document)
2728 if (output_stream_regular)
2729 fl = opt.output_document;
2732 fl = hstat.local_file;
2736 /* Reparse time header, in case it's changed. */
2737 if (time_came_from_head
2738 && hstat.remote_time && hstat.remote_time[0])
2740 newtmr = http_atotm (hstat.remote_time);
2747 /* End of time-stamping section. */
2749 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2750 total_download_time += hstat.dltime;
2752 if (hstat.len == hstat.contlen)
2756 logprintf (LOG_VERBOSE,
2757 _("%s (%s) - %s saved [%s/%s]\n\n"),
2758 tms, tmrate, quote (hstat.local_file),
2759 number_to_static_string (hstat.len),
2760 number_to_static_string (hstat.contlen));
2761 logprintf (LOG_NONVERBOSE,
2762 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2764 number_to_static_string (hstat.len),
2765 number_to_static_string (hstat.contlen),
2766 hstat.local_file, count);
2769 total_downloaded_bytes += hstat.len;
2771 /* Remember that we downloaded the file for later ".orig" code. */
2772 if (*dt & ADDED_HTML_EXTENSION)
2773 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2775 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2780 else if (hstat.res == 0) /* No read error */
2782 if (hstat.contlen == -1) /* We don't know how much we were supposed
2783 to get, so assume we succeeded. */
2787 logprintf (LOG_VERBOSE,
2788 _("%s (%s) - %s saved [%s]\n\n"),
2789 tms, tmrate, quote (hstat.local_file),
2790 number_to_static_string (hstat.len));
2791 logprintf (LOG_NONVERBOSE,
2792 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2793 tms, u->url, number_to_static_string (hstat.len),
2794 hstat.local_file, count);
2797 total_downloaded_bytes += hstat.len;
2799 /* Remember that we downloaded the file for later ".orig" code. */
2800 if (*dt & ADDED_HTML_EXTENSION)
2801 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2803 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2808 else if (hstat.len < hstat.contlen) /* meaning we lost the
2809 connection too soon */
2811 logprintf (LOG_VERBOSE,
2812 _("%s (%s) - Connection closed at byte %s. "),
2813 tms, tmrate, number_to_static_string (hstat.len));
2814 printwhat (count, opt.ntry);
2817 else if (hstat.len != hstat.restval)
2818 /* Getting here would mean reading more data than
2819 requested with content-length, which we never do. */
2823 /* Getting here probably means that the content-length was
2824 * _less_ than the original, local size. We should probably
2825 * truncate or re-read, or something. FIXME */
2830 else /* from now on hstat.res can only be -1 */
2832 if (hstat.contlen == -1)
2834 logprintf (LOG_VERBOSE,
2835 _("%s (%s) - Read error at byte %s (%s)."),
2836 tms, tmrate, number_to_static_string (hstat.len),
2838 printwhat (count, opt.ntry);
2841 else /* hstat.res == -1 and contlen is given */
2843 logprintf (LOG_VERBOSE,
2844 _("%s (%s) - Read error at byte %s/%s (%s). "),
2846 number_to_static_string (hstat.len),
2847 number_to_static_string (hstat.contlen),
2849 printwhat (count, opt.ntry);
2855 while (!opt.ntry || (count < opt.ntry));
2859 *local_file = xstrdup (hstat.local_file);
2860 free_hstat (&hstat);
2865 /* Check whether the result of strptime() indicates success.
2866 strptime() returns the pointer to how far it got to in the string.
2867 The processing has been successful if the string is at `GMT' or
2868 `+X', or at the end of the string.
2870 In extended regexp parlance, the function returns 1 if P matches
2871 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2872 can return) is considered a failure and 0 is returned. */
2874 check_end (const char *p)
2878 while (c_isspace (*p))
2881 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2882 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2888 /* Convert the textual specification of time in TIME_STRING to the
2889 number of seconds since the Epoch.
2891 TIME_STRING can be in any of the three formats RFC2616 allows the
2892 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2893 as well as the time format used in the Set-Cookie header.
2894 Timezones are ignored, and should be GMT.
2896 Return the computed time_t representation, or -1 if the conversion
2899 This function uses strptime with various string formats for parsing
2900 TIME_STRING. This results in a parser that is not as lenient in
2901 interpreting TIME_STRING as I would like it to be. Being based on
2902 strptime, it always allows shortened months, one-digit days, etc.,
2903 but due to the multitude of formats in which time can be
2904 represented, an ideal HTTP time parser would be even more
2905 forgiving. It should completely ignore things like week days and
2906 concentrate only on the various forms of representing years,
2907 months, days, hours, minutes, and seconds. For example, it would
2908 be nice if it accepted ISO 8601 out of the box.
2910 I've investigated free and PD code for this purpose, but none was
2911 usable. getdate was big and unwieldy, and had potential copyright
2912 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2913 distributed with phttpd, is excellent, but we cannot use it because
2914 it is not assigned to the FSF. So I stuck it with strptime. */
2917 http_atotm (const char *time_string)
2919 /* NOTE: Solaris strptime man page claims that %n and %t match white
2920 space, but that's not universally available. Instead, we simply
2921 use ` ' to mean "skip all WS", which works under all strptime
2922 implementations I've tested. */
2924 static const char *time_formats[] = {
2925 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2926 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2927 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2928 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2929 (used in Set-Cookie, defined in the
2930 Netscape cookie specification.) */
2932 const char *oldlocale;
2934 time_t ret = (time_t) -1;
2936 /* Solaris strptime fails to recognize English month names in
2937 non-English locales, which we work around by temporarily setting
2938 locale to C before invoking strptime. */
2939 oldlocale = setlocale (LC_TIME, NULL);
2940 setlocale (LC_TIME, "C");
2942 for (i = 0; i < countof (time_formats); i++)
2946 /* Some versions of strptime use the existing contents of struct
2947 tm to recalculate the date according to format. Zero it out
2948 to prevent stack garbage from influencing strptime. */
2951 if (check_end (strptime (time_string, time_formats[i], &t)))
2958 /* Restore the previous locale. */
2959 setlocale (LC_TIME, oldlocale);
2964 /* Authorization support: We support three authorization schemes:
2966 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2968 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2969 consisting of answering to the server's challenge with the proper
2972 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2973 Stenberg for libcurl. Like digest, NTLM is based on a
2974 challenge-response mechanism, but unlike digest, it is non-standard
2975 (authenticates TCP connections rather than requests), undocumented
2976 and Microsoft-specific. */
2978 /* Create the authentication header contents for the `Basic' scheme.
2979 This is done by encoding the string "USER:PASS" to base64 and
2980 prepending the string "Basic " in front of it. */
2983 basic_authentication_encode (const char *user, const char *passwd)
2986 int len1 = strlen (user) + 1 + strlen (passwd);
2988 t1 = (char *)alloca (len1 + 1);
2989 sprintf (t1, "%s:%s", user, passwd);
2991 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2992 base64_encode (t1, len1, t2);
2994 return concat_strings ("Basic ", t2, (char *) 0);
2997 #define SKIP_WS(x) do { \
2998 while (c_isspace (*(x))) \
3002 #ifdef ENABLE_DIGEST
3003 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
3004 an array of 16 bytes containing the hash keys, and BUF should be a
3005 buffer of 33 writable characters (32 for hex digits plus one for
3006 zero termination). */
3008 dump_hash (char *buf, const unsigned char *hash)
3012 for (i = 0; i < MD5_HASHLEN; i++, hash++)
3014 *buf++ = XNUM_TO_digit (*hash >> 4);
3015 *buf++ = XNUM_TO_digit (*hash & 0xf);
3020 /* Take the line apart to find the challenge, and compose a digest
3021 authorization header. See RFC2069 section 2.1.2. */
3023 digest_authentication_encode (const char *au, const char *user,
3024 const char *passwd, const char *method,
3027 static char *realm, *opaque, *nonce;
3032 { "realm", &realm },
3033 { "opaque", &opaque },
3037 param_token name, value;
3039 realm = opaque = nonce = NULL;
3041 au += 6; /* skip over `Digest' */
3042 while (extract_param (&au, &name, &value, ','))
3045 size_t namelen = name.e - name.b;
3046 for (i = 0; i < countof (options); i++)
3047 if (namelen == strlen (options[i].name)
3048 && 0 == strncmp (name.b, options[i].name,
3051 *options[i].variable = strdupdelim (value.b, value.e);
3055 if (!realm || !nonce || !user || !passwd || !path || !method)
3058 xfree_null (opaque);
3063 /* Calculate the digest value. */
3065 ALLOCA_MD5_CONTEXT (ctx);
3066 unsigned char hash[MD5_HASHLEN];
3067 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3068 char response_digest[MD5_HASHLEN * 2 + 1];
3070 /* A1BUF = H(user ":" realm ":" password) */
3072 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3073 gen_md5_update ((unsigned char *)":", 1, ctx);
3074 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3075 gen_md5_update ((unsigned char *)":", 1, ctx);
3076 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3077 gen_md5_finish (ctx, hash);
3078 dump_hash (a1buf, hash);
3080 /* A2BUF = H(method ":" path) */
3082 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3083 gen_md5_update ((unsigned char *)":", 1, ctx);
3084 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3085 gen_md5_finish (ctx, hash);
3086 dump_hash (a2buf, hash);
3088 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3090 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3091 gen_md5_update ((unsigned char *)":", 1, ctx);
3092 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3093 gen_md5_update ((unsigned char *)":", 1, ctx);
3094 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3095 gen_md5_finish (ctx, hash);
3096 dump_hash (response_digest, hash);
3098 res = xmalloc (strlen (user)
3103 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3104 + (opaque ? strlen (opaque) : 0)
3106 sprintf (res, "Digest \
3107 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3108 user, realm, nonce, path, response_digest);
3111 char *p = res + strlen (res);
3112 strcat (p, ", opaque=\"");
3119 #endif /* ENABLE_DIGEST */
3121 /* Computing the size of a string literal must take into account that
3122 value returned by sizeof includes the terminating \0. */
3123 #define STRSIZE(literal) (sizeof (literal) - 1)
3125 /* Whether chars in [b, e) begin with the literal string provided as
3126 first argument and are followed by whitespace or terminating \0.
3127 The comparison is case-insensitive. */
3128 #define STARTS(literal, b, e) \
3130 && ((size_t) ((e) - (b))) >= STRSIZE (literal) \
3131 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3132 && ((size_t) ((e) - (b)) == STRSIZE (literal) \
3133 || c_isspace (b[STRSIZE (literal)])))
3136 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3138 return STARTS ("Basic", hdrbeg, hdrend)
3139 #ifdef ENABLE_DIGEST
3140 || STARTS ("Digest", hdrbeg, hdrend)
3143 || STARTS ("NTLM", hdrbeg, hdrend)
3150 /* Create the HTTP authorization request header. When the
3151 `WWW-Authenticate' response header is seen, according to the
3152 authorization scheme specified in that header (`Basic' and `Digest'
3153 are supported by the current implementation), produce an
3154 appropriate HTTP authorization request header. */
3156 create_authorization_line (const char *au, const char *user,
3157 const char *passwd, const char *method,
3158 const char *path, bool *finished)
3160 /* We are called only with known schemes, so we can dispatch on the
3162 switch (c_toupper (*au))
3164 case 'B': /* Basic */
3166 return basic_authentication_encode (user, passwd);
3167 #ifdef ENABLE_DIGEST
3168 case 'D': /* Digest */
3170 return digest_authentication_encode (au, user, passwd, method, path);
3173 case 'N': /* NTLM */
3174 if (!ntlm_input (&pconn.ntlm, au))
3179 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3182 /* We shouldn't get here -- this function should be only called
3183 with values approved by known_authentication_scheme_p. */
3191 if (!wget_cookie_jar)
3192 wget_cookie_jar = cookie_jar_new ();
3193 if (opt.cookies_input && !cookies_loaded_p)
3195 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3196 cookies_loaded_p = true;
3203 if (wget_cookie_jar)
3204 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3210 xfree_null (pconn.host);
3211 if (wget_cookie_jar)
3212 cookie_jar_delete (wget_cookie_jar);
3219 test_parse_content_disposition()
3224 char *opt_dir_prefix;
3228 { "filename=\"file.ext\"", NULL, "file.ext", true },
3229 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3230 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3231 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3232 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3233 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3234 { "attachment", NULL, NULL, false },
3235 { "attachment", "somedir", NULL, false },
3238 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3243 opt.dir_prefix = test_array[i].opt_dir_prefix;
3244 res = parse_content_disposition (test_array[i].hdrval, &filename);
3246 mu_assert ("test_parse_content_disposition: wrong result",
3247 res == test_array[i].result
3249 || 0 == strcmp (test_array[i].filename, filename)));
3255 #endif /* TESTING */
3258 * vim: et sts=2 sw=2 cino+={s