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_VERBOSE, "%s%s\n", prefix, escnonprint(copy));
816 /* Print the server response, line by line, omitting the trailing CRLF
817 from individual header lines, and prefixed with PREFIX. */
820 print_server_response (const struct response *resp, const char *prefix)
825 for (i = 0; resp->headers[i + 1]; i++)
827 const char *b = resp->headers[i];
828 const char *e = resp->headers[i + 1];
830 if (b < e && e[-1] == '\n')
832 if (b < e && e[-1] == '\r')
834 print_response_line(prefix, b, e);
838 /* Parse the `Content-Range' header and extract the information it
839 contains. Returns true if successful, false otherwise. */
841 parse_content_range (const char *hdr, wgint *first_byte_ptr,
842 wgint *last_byte_ptr, wgint *entity_length_ptr)
846 /* Ancient versions of Netscape proxy server, presumably predating
847 rfc2068, sent out `Content-Range' without the "bytes"
849 if (0 == strncasecmp (hdr, "bytes", 5))
852 /* "JavaWebServer/1.1.1" sends "bytes: x-y/z", contrary to the
856 while (c_isspace (*hdr))
861 if (!c_isdigit (*hdr))
863 for (num = 0; c_isdigit (*hdr); hdr++)
864 num = 10 * num + (*hdr - '0');
865 if (*hdr != '-' || !c_isdigit (*(hdr + 1)))
867 *first_byte_ptr = num;
869 for (num = 0; c_isdigit (*hdr); hdr++)
870 num = 10 * num + (*hdr - '0');
871 if (*hdr != '/' || !c_isdigit (*(hdr + 1)))
873 *last_byte_ptr = num;
878 for (num = 0; c_isdigit (*hdr); hdr++)
879 num = 10 * num + (*hdr - '0');
880 *entity_length_ptr = num;
884 /* Read the body of the request, but don't store it anywhere and don't
885 display a progress gauge. This is useful for reading the bodies of
886 administrative responses to which we will soon issue another
887 request. The response is not useful to the user, but reading it
888 allows us to continue using the same connection to the server.
890 If reading fails, false is returned, true otherwise. In debug
891 mode, the body is displayed for debugging purposes. */
894 skip_short_body (int fd, wgint contlen)
897 SKIP_SIZE = 512, /* size of the download buffer */
898 SKIP_THRESHOLD = 4096 /* the largest size we read */
900 char dlbuf[SKIP_SIZE + 1];
901 dlbuf[SKIP_SIZE] = '\0'; /* so DEBUGP can safely print it */
903 /* We shouldn't get here with unknown contlen. (This will change
904 with HTTP/1.1, which supports "chunked" transfer.) */
905 assert (contlen != -1);
907 /* If the body is too large, it makes more sense to simply close the
908 connection than to try to read the body. */
909 if (contlen > SKIP_THRESHOLD)
912 DEBUGP (("Skipping %s bytes of body: [", number_to_static_string (contlen)));
916 int ret = fd_read (fd, dlbuf, MIN (contlen, SKIP_SIZE), -1);
919 /* Don't normally report the error since this is an
920 optimization that should be invisible to the user. */
921 DEBUGP (("] aborting (%s).\n",
922 ret < 0 ? fd_errstr (fd) : "EOF received"));
926 /* Safe even if %.*s bogusly expects terminating \0 because
927 we've zero-terminated dlbuf above. */
928 DEBUGP (("%.*s", ret, dlbuf));
931 DEBUGP (("] done.\n"));
935 /* Extract a parameter from the string (typically an HTTP header) at
936 **SOURCE and advance SOURCE to the next parameter. Return false
937 when there are no more parameters to extract. The name of the
938 parameter is returned in NAME, and the value in VALUE. If the
939 parameter has no value, the token's value is zeroed out.
941 For example, if *SOURCE points to the string "attachment;
942 filename=\"foo bar\"", the first call to this function will return
943 the token named "attachment" and no value, and the second call will
944 return the token named "filename" and value "foo bar". The third
945 call will return false, indicating no more valid tokens. */
948 extract_param (const char **source, param_token *name, param_token *value,
951 const char *p = *source;
953 while (c_isspace (*p)) ++p;
957 return false; /* no error; nothing more to extract */
962 while (*p && !c_isspace (*p) && *p != '=' && *p != separator) ++p;
964 if (name->b == name->e)
965 return false; /* empty name: error */
966 while (c_isspace (*p)) ++p;
967 if (*p == separator || !*p) /* no value */
970 if (*p == separator) ++p;
975 return false; /* error */
977 /* *p is '=', extract value */
979 while (c_isspace (*p)) ++p;
980 if (*p == '"') /* quoted */
983 while (*p && *p != '"') ++p;
987 /* Currently at closing quote; find the end of param. */
988 while (c_isspace (*p)) ++p;
989 while (*p && *p != separator) ++p;
993 /* garbage after closed quote, e.g. foo="bar"baz */
999 while (*p && *p != separator) ++p;
1001 while (value->e != value->b && c_isspace (value->e[-1]))
1003 if (*p == separator) ++p;
1010 #define MAX(p, q) ((p) > (q) ? (p) : (q))
1012 /* Parse the contents of the `Content-Disposition' header, extracting
1013 the information useful to Wget. Content-Disposition is a header
1014 borrowed from MIME; when used in HTTP, it typically serves for
1015 specifying the desired file name of the resource. For example:
1017 Content-Disposition: attachment; filename="flora.jpg"
1019 Wget will skip the tokens it doesn't care about, such as
1020 "attachment" in the previous example; it will also skip other
1021 unrecognized params. If the header is syntactically correct and
1022 contains a file name, a copy of the file name is stored in
1023 *filename and true is returned. Otherwise, the function returns
1026 The file name is stripped of directory components and must not be
1030 parse_content_disposition (const char *hdr, char **filename)
1032 param_token name, value;
1033 while (extract_param (&hdr, &name, &value, ';'))
1034 if (BOUNDED_EQUAL_NO_CASE (name.b, name.e, "filename") && value.b != NULL)
1036 /* Make the file name begin at the last slash or backslash. */
1037 const char *last_slash = memrchr (value.b, '/', value.e - value.b);
1038 const char *last_bs = memrchr (value.b, '\\', value.e - value.b);
1039 if (last_slash && last_bs)
1040 value.b = 1 + MAX (last_slash, last_bs);
1041 else if (last_slash || last_bs)
1042 value.b = 1 + (last_slash ? last_slash : last_bs);
1043 if (value.b == value.e)
1045 /* Start with the directory prefix, if specified. */
1048 int prefix_length = strlen (opt.dir_prefix);
1049 bool add_slash = (opt.dir_prefix[prefix_length - 1] != '/');
1054 total_length = prefix_length + (value.e - value.b);
1055 *filename = xmalloc (total_length + 1);
1056 strcpy (*filename, opt.dir_prefix);
1058 (*filename)[prefix_length - 1] = '/';
1059 memcpy (*filename + prefix_length, value.b, (value.e - value.b));
1060 (*filename)[total_length] = '\0';
1063 *filename = strdupdelim (value.b, value.e);
1069 /* Persistent connections. Currently, we cache the most recently used
1070 connection as persistent, provided that the HTTP server agrees to
1071 make it such. The persistence data is stored in the variables
1072 below. Ideally, it should be possible to cache an arbitrary fixed
1073 number of these connections. */
1075 /* Whether a persistent connection is active. */
1076 static bool pconn_active;
1079 /* The socket of the connection. */
1082 /* Host and port of the currently active persistent connection. */
1086 /* Whether a ssl handshake has occoured on this connection. */
1089 /* Whether the connection was authorized. This is only done by
1090 NTLM, which authorizes *connections* rather than individual
1091 requests. (That practice is peculiar for HTTP, but it is a
1092 useful optimization.) */
1096 /* NTLM data of the current connection. */
1097 struct ntlmdata ntlm;
1101 /* Mark the persistent connection as invalid and free the resources it
1102 uses. This is used by the CLOSE_* macros after they forcefully
1103 close a registered persistent connection. */
1106 invalidate_persistent (void)
1108 DEBUGP (("Disabling further reuse of socket %d.\n", pconn.socket));
1109 pconn_active = false;
1110 fd_close (pconn.socket);
1115 /* Register FD, which should be a TCP/IP connection to HOST:PORT, as
1116 persistent. This will enable someone to use the same connection
1117 later. In the context of HTTP, this must be called only AFTER the
1118 response has been received and the server has promised that the
1119 connection will remain alive.
1121 If a previous connection was persistent, it is closed. */
1124 register_persistent (const char *host, int port, int fd, bool ssl)
1128 if (pconn.socket == fd)
1130 /* The connection FD is already registered. */
1135 /* The old persistent connection is still active; close it
1136 first. This situation arises whenever a persistent
1137 connection exists, but we then connect to a different
1138 host, and try to register a persistent connection to that
1140 invalidate_persistent ();
1144 pconn_active = true;
1146 pconn.host = xstrdup (host);
1149 pconn.authorized = false;
1151 DEBUGP (("Registered socket %d for persistent reuse.\n", fd));
1154 /* Return true if a persistent connection is available for connecting
1158 persistent_available_p (const char *host, int port, bool ssl,
1159 bool *host_lookup_failed)
1161 /* First, check whether a persistent connection is active at all. */
1165 /* If we want SSL and the last connection wasn't or vice versa,
1166 don't use it. Checking for host and port is not enough because
1167 HTTP and HTTPS can apparently coexist on the same port. */
1168 if (ssl != pconn.ssl)
1171 /* If we're not connecting to the same port, we're not interested. */
1172 if (port != pconn.port)
1175 /* If the host is the same, we're in business. If not, there is
1176 still hope -- read below. */
1177 if (0 != strcasecmp (host, pconn.host))
1179 /* Check if pconn.socket is talking to HOST under another name.
1180 This happens often when both sites are virtual hosts
1181 distinguished only by name and served by the same network
1182 interface, and hence the same web server (possibly set up by
1183 the ISP and serving many different web sites). This
1184 admittedly unconventional optimization does not contradict
1185 HTTP and works well with popular server software. */
1189 struct address_list *al;
1192 /* Don't try to talk to two different SSL sites over the same
1193 secure connection! (Besides, it's not clear that
1194 name-based virtual hosting is even possible with SSL.) */
1197 /* If pconn.socket's peer is one of the IP addresses HOST
1198 resolves to, pconn.socket is for all intents and purposes
1199 already talking to HOST. */
1201 if (!socket_ip_address (pconn.socket, &ip, ENDPOINT_PEER))
1203 /* Can't get the peer's address -- something must be very
1204 wrong with the connection. */
1205 invalidate_persistent ();
1208 al = lookup_host (host, 0);
1211 *host_lookup_failed = true;
1215 found = address_list_contains (al, &ip);
1216 address_list_release (al);
1221 /* The persistent connection's peer address was found among the
1222 addresses HOST resolved to; therefore, pconn.sock is in fact
1223 already talking to HOST -- no need to reconnect. */
1226 /* Finally, check whether the connection is still open. This is
1227 important because most servers implement liberal (short) timeout
1228 on persistent connections. Wget can of course always reconnect
1229 if the connection doesn't work out, but it's nicer to know in
1230 advance. This test is a logical followup of the first test, but
1231 is "expensive" and therefore placed at the end of the list.
1233 (Current implementation of test_socket_open has a nice side
1234 effect that it treats sockets with pending data as "closed".
1235 This is exactly what we want: if a broken server sends message
1236 body in response to HEAD, or if it sends more than conent-length
1237 data, we won't reuse the corrupted connection.) */
1239 if (!test_socket_open (pconn.socket))
1241 /* Oops, the socket is no longer open. Now that we know that,
1242 let's invalidate the persistent connection before returning
1244 invalidate_persistent ();
1251 /* The idea behind these two CLOSE macros is to distinguish between
1252 two cases: one when the job we've been doing is finished, and we
1253 want to close the connection and leave, and two when something is
1254 seriously wrong and we're closing the connection as part of
1257 In case of keep_alive, CLOSE_FINISH should leave the connection
1258 open, while CLOSE_INVALIDATE should still close it.
1260 Note that the semantics of the flag `keep_alive' is "this
1261 connection *will* be reused (the server has promised not to close
1262 the connection once we're done)", while the semantics of
1263 `pc_active_p && (fd) == pc_last_fd' is "we're *now* using an
1264 active, registered connection". */
1266 #define CLOSE_FINISH(fd) do { \
1269 if (pconn_active && (fd) == pconn.socket) \
1270 invalidate_persistent (); \
1279 #define CLOSE_INVALIDATE(fd) do { \
1280 if (pconn_active && (fd) == pconn.socket) \
1281 invalidate_persistent (); \
1289 wgint len; /* received length */
1290 wgint contlen; /* expected length */
1291 wgint restval; /* the restart value */
1292 int res; /* the result of last read */
1293 char *rderrmsg; /* error message from read error */
1294 char *newloc; /* new location (redirection) */
1295 char *remote_time; /* remote time-stamp string */
1296 char *error; /* textual HTTP error */
1297 int statcode; /* status code */
1298 wgint rd_size; /* amount of data read from socket */
1299 double dltime; /* time it took to download the data */
1300 const char *referer; /* value of the referer header. */
1301 char *local_file; /* local file name. */
1302 bool existence_checked; /* true if we already checked for a file's
1303 existence after having begun to download
1304 (needed in gethttp for when connection is
1305 interrupted/restarted. */
1306 bool timestamp_checked; /* true if pre-download time-stamping checks
1307 * have already been performed */
1308 char *orig_file_name; /* name of file to compare for time-stamping
1309 * (might be != local_file if -K is set) */
1310 wgint orig_file_size; /* size of file to compare for time-stamping */
1311 time_t orig_file_tstamp; /* time-stamp of file to compare for
1316 free_hstat (struct http_stat *hs)
1318 xfree_null (hs->newloc);
1319 xfree_null (hs->remote_time);
1320 xfree_null (hs->error);
1321 xfree_null (hs->rderrmsg);
1322 xfree_null (hs->local_file);
1323 xfree_null (hs->orig_file_name);
1325 /* Guard against being called twice. */
1327 hs->remote_time = NULL;
1331 #define BEGINS_WITH(line, string_constant) \
1332 (!strncasecmp (line, string_constant, sizeof (string_constant) - 1) \
1333 && (c_isspace (line[sizeof (string_constant) - 1]) \
1334 || !line[sizeof (string_constant) - 1]))
1336 #define SET_USER_AGENT(req) do { \
1337 if (!opt.useragent) \
1338 request_set_header (req, "User-Agent", \
1339 aprintf ("Wget/%s", version_string), rel_value); \
1340 else if (*opt.useragent) \
1341 request_set_header (req, "User-Agent", opt.useragent, rel_none); \
1344 /* The flags that allow clobbering the file (opening with "wb").
1345 Defined here to avoid repetition later. #### This will require
1347 #define ALLOW_CLOBBER (opt.noclobber || opt.always_rest || opt.timestamping \
1348 || opt.dirstruct || opt.output_document)
1350 /* Retrieve a document through HTTP protocol. It recognizes status
1351 code, and correctly handles redirections. It closes the network
1352 socket. If it receives an error from the functions below it, it
1353 will print it if there is enough information to do so (almost
1354 always), returning the error to the caller (i.e. http_loop).
1356 Various HTTP parameters are stored to hs.
1358 If PROXY is non-NULL, the connection will be made to the proxy
1359 server, and u->url will be requested. */
1361 gethttp (struct url *u, struct http_stat *hs, int *dt, struct url *proxy)
1363 struct request *req;
1366 char *user, *passwd;
1370 wgint contlen, contrange;
1377 /* Set to 1 when the authorization has already been sent and should
1378 not be tried again. */
1379 bool auth_finished = false;
1381 /* Set to 1 when just globally-set Basic authorization has been sent;
1382 * should prevent further Basic negotiations, but not other
1384 bool basic_auth_finished = false;
1386 /* Whether NTLM authentication is used for this request. */
1387 bool ntlm_seen = false;
1389 /* Whether our connection to the remote host is through SSL. */
1390 bool using_ssl = false;
1392 /* Whether a HEAD request will be issued (as opposed to GET or
1394 bool head_only = !!(*dt & HEAD_ONLY);
1397 struct response *resp;
1401 /* Whether this connection will be kept alive after the HTTP request
1405 /* Whether keep-alive should be inhibited.
1407 RFC 2068 requests that 1.0 clients not send keep-alive requests
1408 to proxies. This is because many 1.0 proxies do not interpret
1409 the Connection header and transfer it to the remote server,
1410 causing it to not close the connection and leave both the proxy
1411 and the client hanging. */
1412 bool inhibit_keep_alive =
1413 !opt.http_keep_alive || opt.ignore_length || proxy != NULL;
1415 /* Headers sent when using POST. */
1416 wgint post_data_size = 0;
1418 bool host_lookup_failed = false;
1421 if (u->scheme == SCHEME_HTTPS)
1423 /* Initialize the SSL context. After this has once been done,
1424 it becomes a no-op. */
1427 scheme_disable (SCHEME_HTTPS);
1428 logprintf (LOG_NOTQUIET,
1429 _("Disabling SSL due to encountered errors.\n"));
1430 return SSLINITFAILED;
1433 #endif /* HAVE_SSL */
1435 /* Initialize certain elements of struct http_stat. */
1439 hs->rderrmsg = NULL;
1441 hs->remote_time = NULL;
1446 /* Prepare the request to send. */
1448 req = request_new ();
1451 const char *meth = "GET";
1454 else if (opt.post_file_name || opt.post_data)
1456 /* Use the full path, i.e. one that includes the leading slash and
1457 the query string. E.g. if u->path is "foo/bar" and u->query is
1458 "param=value", full_path will be "/foo/bar?param=value". */
1461 /* When using SSL over proxy, CONNECT establishes a direct
1462 connection to the HTTPS server. Therefore use the same
1463 argument as when talking to the server directly. */
1464 && u->scheme != SCHEME_HTTPS
1467 meth_arg = xstrdup (u->url);
1469 meth_arg = url_full_path (u);
1470 request_set_method (req, meth, meth_arg);
1473 request_set_header (req, "Referer", (char *) hs->referer, rel_none);
1474 if (*dt & SEND_NOCACHE)
1475 request_set_header (req, "Pragma", "no-cache", rel_none);
1477 request_set_header (req, "Range",
1478 aprintf ("bytes=%s-",
1479 number_to_static_string (hs->restval)),
1481 SET_USER_AGENT (req);
1482 request_set_header (req, "Accept", "*/*", rel_none);
1484 /* Find the username and password for authentication. */
1487 search_netrc (u->host, (const char **)&user, (const char **)&passwd, 0);
1488 user = user ? user : (opt.http_user ? opt.http_user : opt.user);
1489 passwd = passwd ? passwd : (opt.http_passwd ? opt.http_passwd : opt.passwd);
1492 && !u->user) /* We only do "site-wide" authentication with "global"
1493 user/password values; URL user/password info overrides. */
1495 /* If this is a host for which we've already received a Basic
1496 * challenge, we'll go ahead and send Basic authentication creds. */
1497 basic_auth_finished = maybe_send_basic_creds(u->host, user, passwd, req);
1503 char *proxy_user, *proxy_passwd;
1504 /* For normal username and password, URL components override
1505 command-line/wgetrc parameters. With proxy
1506 authentication, it's the reverse, because proxy URLs are
1507 normally the "permanent" ones, so command-line args
1508 should take precedence. */
1509 if (opt.proxy_user && opt.proxy_passwd)
1511 proxy_user = opt.proxy_user;
1512 proxy_passwd = opt.proxy_passwd;
1516 proxy_user = proxy->user;
1517 proxy_passwd = proxy->passwd;
1519 /* #### This does not appear right. Can't the proxy request,
1520 say, `Digest' authentication? */
1521 if (proxy_user && proxy_passwd)
1522 proxyauth = basic_authentication_encode (proxy_user, proxy_passwd);
1524 /* If we're using a proxy, we will be connecting to the proxy
1528 /* Proxy authorization over SSL is handled below. */
1530 if (u->scheme != SCHEME_HTTPS)
1532 request_set_header (req, "Proxy-Authorization", proxyauth, rel_value);
1535 /* Generate the Host header, HOST:PORT. Take into account that:
1537 - Broken server-side software often doesn't recognize the PORT
1538 argument, so we must generate "Host: www.server.com" instead of
1539 "Host: www.server.com:80" (and likewise for https port).
1541 - IPv6 addresses contain ":", so "Host: 3ffe:8100:200:2::2:1234"
1542 becomes ambiguous and needs to be rewritten as "Host:
1543 [3ffe:8100:200:2::2]:1234". */
1545 /* Formats arranged for hfmt[add_port][add_squares]. */
1546 static const char *hfmt[][2] = {
1547 { "%s", "[%s]" }, { "%s:%d", "[%s]:%d" }
1549 int add_port = u->port != scheme_default_port (u->scheme);
1550 int add_squares = strchr (u->host, ':') != NULL;
1551 request_set_header (req, "Host",
1552 aprintf (hfmt[add_port][add_squares], u->host, u->port),
1556 if (!inhibit_keep_alive)
1557 request_set_header (req, "Connection", "Keep-Alive", rel_none);
1560 request_set_header (req, "Cookie",
1561 cookie_header (wget_cookie_jar,
1562 u->host, u->port, u->path,
1564 u->scheme == SCHEME_HTTPS
1571 if (opt.post_data || opt.post_file_name)
1573 request_set_header (req, "Content-Type",
1574 "application/x-www-form-urlencoded", rel_none);
1576 post_data_size = strlen (opt.post_data);
1579 post_data_size = file_size (opt.post_file_name);
1580 if (post_data_size == -1)
1582 logprintf (LOG_NOTQUIET, _("POST data file %s missing: %s\n"),
1583 quote (opt.post_file_name), strerror (errno));
1587 request_set_header (req, "Content-Length",
1588 xstrdup (number_to_static_string (post_data_size)),
1592 /* Add the user headers. */
1593 if (opt.user_headers)
1596 for (i = 0; opt.user_headers[i]; i++)
1597 request_set_user_header (req, opt.user_headers[i]);
1601 /* We need to come back here when the initial attempt to retrieve
1602 without authorization header fails. (Expected to happen at least
1603 for the Digest authorization scheme.) */
1607 /* Establish the connection. */
1609 if (!inhibit_keep_alive)
1611 /* Look for a persistent connection to target host, unless a
1612 proxy is used. The exception is when SSL is in use, in which
1613 case the proxy is nothing but a passthrough to the target
1614 host, registered as a connection to the latter. */
1615 struct url *relevant = conn;
1617 if (u->scheme == SCHEME_HTTPS)
1621 if (persistent_available_p (relevant->host, relevant->port,
1623 relevant->scheme == SCHEME_HTTPS,
1627 &host_lookup_failed))
1629 sock = pconn.socket;
1630 using_ssl = pconn.ssl;
1631 logprintf (LOG_VERBOSE, _("Reusing existing connection to %s:%d.\n"),
1632 escnonprint (pconn.host), pconn.port);
1633 DEBUGP (("Reusing fd %d.\n", sock));
1634 if (pconn.authorized)
1635 /* If the connection is already authorized, the "Basic"
1636 authorization added by code above is unnecessary and
1638 request_remove_header (req, "Authorization");
1640 else if (host_lookup_failed)
1643 logprintf(LOG_NOTQUIET,
1644 _("%s: unable to resolve host address %s\n"),
1645 exec_name, quote (relevant->host));
1652 sock = connect_to_host (conn->host, conn->port);
1661 return (retryable_socket_connect_error (errno)
1662 ? CONERROR : CONIMPOSSIBLE);
1666 if (proxy && u->scheme == SCHEME_HTTPS)
1668 /* When requesting SSL URLs through proxies, use the
1669 CONNECT method to request passthrough. */
1670 struct request *connreq = request_new ();
1671 request_set_method (connreq, "CONNECT",
1672 aprintf ("%s:%d", u->host, u->port));
1673 SET_USER_AGENT (connreq);
1676 request_set_header (connreq, "Proxy-Authorization",
1677 proxyauth, rel_value);
1678 /* Now that PROXYAUTH is part of the CONNECT request,
1679 zero it out so we don't send proxy authorization with
1680 the regular request below. */
1683 /* Examples in rfc2817 use the Host header in CONNECT
1684 requests. I don't see how that gains anything, given
1685 that the contents of Host would be exactly the same as
1686 the contents of CONNECT. */
1688 write_error = request_send (connreq, sock);
1689 request_free (connreq);
1690 if (write_error < 0)
1692 CLOSE_INVALIDATE (sock);
1696 head = read_http_response_head (sock);
1699 logprintf (LOG_VERBOSE, _("Failed reading proxy response: %s\n"),
1701 CLOSE_INVALIDATE (sock);
1710 DEBUGP (("proxy responded with: [%s]\n", head));
1712 resp = resp_new (head);
1713 statcode = resp_status (resp, &message);
1716 if (statcode != 200)
1719 logprintf (LOG_NOTQUIET, _("Proxy tunneling failed: %s"),
1720 message ? escnonprint (message) : "?");
1721 xfree_null (message);
1724 xfree_null (message);
1726 /* SOCK is now *really* connected to u->host, so update CONN
1727 to reflect this. That way register_persistent will
1728 register SOCK as being connected to u->host:u->port. */
1732 if (conn->scheme == SCHEME_HTTPS)
1734 if (!ssl_connect (sock) || !ssl_check_certificate (sock, u->host))
1741 #endif /* HAVE_SSL */
1744 /* Send the request to server. */
1745 write_error = request_send (req, sock);
1747 if (write_error >= 0)
1751 DEBUGP (("[POST data: %s]\n", opt.post_data));
1752 write_error = fd_write (sock, opt.post_data, post_data_size, -1);
1754 else if (opt.post_file_name && post_data_size != 0)
1755 write_error = post_file (sock, opt.post_file_name, post_data_size);
1758 if (write_error < 0)
1760 CLOSE_INVALIDATE (sock);
1764 logprintf (LOG_VERBOSE, _("%s request sent, awaiting response... "),
1765 proxy ? "Proxy" : "HTTP");
1770 head = read_http_response_head (sock);
1775 logputs (LOG_NOTQUIET, _("No data received.\n"));
1776 CLOSE_INVALIDATE (sock);
1782 logprintf (LOG_NOTQUIET, _("Read error (%s) in headers.\n"),
1784 CLOSE_INVALIDATE (sock);
1789 DEBUGP (("\n---response begin---\n%s---response end---\n", head));
1791 resp = resp_new (head);
1793 /* Check for status line. */
1795 statcode = resp_status (resp, &message);
1796 if (!opt.server_response)
1797 logprintf (LOG_VERBOSE, "%2d %s\n", statcode,
1798 message ? escnonprint (message) : "");
1801 logprintf (LOG_VERBOSE, "\n");
1802 print_server_response (resp, " ");
1805 /* Determine the local filename if needed. Notice that if -O is used
1806 * hstat.local_file is set by http_loop to the argument of -O. */
1807 if (!hs->local_file)
1809 /* Honor Content-Disposition whether possible. */
1810 if (!opt.content_disposition
1811 || !resp_header_copy (resp, "Content-Disposition",
1812 hdrval, sizeof (hdrval))
1813 || !parse_content_disposition (hdrval, &hs->local_file))
1815 /* The Content-Disposition header is missing or broken.
1816 * Choose unique file name according to given URL. */
1817 hs->local_file = url_file_name (u);
1821 /* TODO: perform this check only once. */
1822 if (!hs->existence_checked && file_exists_p (hs->local_file))
1826 /* If opt.noclobber is turned on and file already exists, do not
1827 retrieve the file */
1828 logprintf (LOG_VERBOSE, _("\
1829 File %s already there; not retrieving.\n\n"), quote (hs->local_file));
1830 /* If the file is there, we suppose it's retrieved OK. */
1833 /* #### Bogusness alert. */
1834 /* If its suffix is "html" or "htm" or similar, assume text/html. */
1835 if (has_html_suffix_p (hs->local_file))
1838 return RETRUNNEEDED;
1840 else if (!ALLOW_CLOBBER)
1842 char *unique = unique_name (hs->local_file, true);
1843 if (unique != hs->local_file)
1844 xfree (hs->local_file);
1845 hs->local_file = unique;
1848 hs->existence_checked = true;
1850 /* Support timestamping */
1851 /* TODO: move this code out of gethttp. */
1852 if (opt.timestamping && !hs->timestamp_checked)
1854 size_t filename_len = strlen (hs->local_file);
1855 char *filename_plus_orig_suffix = alloca (filename_len + sizeof (".orig"));
1856 bool local_dot_orig_file_exists = false;
1857 char *local_filename = NULL;
1860 if (opt.backup_converted)
1861 /* If -K is specified, we'll act on the assumption that it was specified
1862 last time these files were downloaded as well, and instead of just
1863 comparing local file X against server file X, we'll compare local
1864 file X.orig (if extant, else X) against server file X. If -K
1865 _wasn't_ specified last time, or the server contains files called
1866 *.orig, -N will be back to not operating correctly with -k. */
1868 /* Would a single s[n]printf() call be faster? --dan
1870 Definitely not. sprintf() is horribly slow. It's a
1871 different question whether the difference between the two
1872 affects a program. Usually I'd say "no", but at one
1873 point I profiled Wget, and found that a measurable and
1874 non-negligible amount of time was lost calling sprintf()
1875 in url.c. Replacing sprintf with inline calls to
1876 strcpy() and number_to_string() made a difference.
1878 memcpy (filename_plus_orig_suffix, hs->local_file, filename_len);
1879 memcpy (filename_plus_orig_suffix + filename_len,
1880 ".orig", sizeof (".orig"));
1882 /* Try to stat() the .orig file. */
1883 if (stat (filename_plus_orig_suffix, &st) == 0)
1885 local_dot_orig_file_exists = true;
1886 local_filename = filename_plus_orig_suffix;
1890 if (!local_dot_orig_file_exists)
1891 /* Couldn't stat() <file>.orig, so try to stat() <file>. */
1892 if (stat (hs->local_file, &st) == 0)
1893 local_filename = hs->local_file;
1895 if (local_filename != NULL)
1896 /* There was a local file, so we'll check later to see if the version
1897 the server has is the same version we already have, allowing us to
1900 hs->orig_file_name = xstrdup (local_filename);
1901 hs->orig_file_size = st.st_size;
1902 hs->orig_file_tstamp = st.st_mtime;
1904 /* Modification time granularity is 2 seconds for Windows, so
1905 increase local time by 1 second for later comparison. */
1906 ++hs->orig_file_tstamp;
1911 if (!opt.ignore_length
1912 && resp_header_copy (resp, "Content-Length", hdrval, sizeof (hdrval)))
1916 parsed = str_to_wgint (hdrval, NULL, 10);
1917 if (parsed == WGINT_MAX && errno == ERANGE)
1920 #### If Content-Length is out of range, it most likely
1921 means that the file is larger than 2G and that we're
1922 compiled without LFS. In that case we should probably
1923 refuse to even attempt to download the file. */
1926 else if (parsed < 0)
1928 /* Negative Content-Length; nonsensical, so we can't
1929 assume any information about the content to receive. */
1936 /* Check for keep-alive related responses. */
1937 if (!inhibit_keep_alive && contlen != -1)
1939 if (resp_header_copy (resp, "Keep-Alive", NULL, 0))
1941 else if (resp_header_copy (resp, "Connection", hdrval, sizeof (hdrval)))
1943 if (0 == strcasecmp (hdrval, "Keep-Alive"))
1948 /* The server has promised that it will not close the connection
1949 when we're done. This means that we can register it. */
1950 register_persistent (conn->host, conn->port, sock, using_ssl);
1952 if (statcode == HTTP_STATUS_UNAUTHORIZED)
1954 /* Authorization is required. */
1955 if (keep_alive && !head_only && skip_short_body (sock, contlen))
1956 CLOSE_FINISH (sock);
1958 CLOSE_INVALIDATE (sock);
1959 pconn.authorized = false;
1960 if (!auth_finished && (user && passwd))
1962 /* IIS sends multiple copies of WWW-Authenticate, one with
1963 the value "negotiate", and other(s) with data. Loop over
1964 all the occurrences and pick the one we recognize. */
1966 const char *wabeg, *waend;
1967 char *www_authenticate = NULL;
1969 (wapos = resp_header_locate (resp, "WWW-Authenticate", wapos,
1970 &wabeg, &waend)) != -1;
1972 if (known_authentication_scheme_p (wabeg, waend))
1974 BOUNDED_TO_ALLOCA (wabeg, waend, www_authenticate);
1978 if (!www_authenticate)
1980 /* If the authentication header is missing or
1981 unrecognized, there's no sense in retrying. */
1982 logputs (LOG_NOTQUIET, _("Unknown authentication scheme.\n"));
1984 else if (!basic_auth_finished
1985 || !BEGINS_WITH (www_authenticate, "Basic"))
1988 pth = url_full_path (u);
1989 request_set_header (req, "Authorization",
1990 create_authorization_line (www_authenticate,
1992 request_method (req),
1996 if (BEGINS_WITH (www_authenticate, "NTLM"))
1998 else if (!u->user && BEGINS_WITH (www_authenticate, "Basic"))
2000 /* Need to register this host as using basic auth,
2001 * so we automatically send creds next time. */
2002 register_basic_auth_host (u->host);
2005 goto retry_with_auth;
2009 /* We already did Basic auth, and it failed. Gotta
2013 logputs (LOG_NOTQUIET, _("Authorization failed.\n"));
2017 else /* statcode != HTTP_STATUS_UNAUTHORIZED */
2019 /* Kludge: if NTLM is used, mark the TCP connection as authorized. */
2021 pconn.authorized = true;
2025 hs->statcode = statcode;
2027 hs->error = xstrdup (_("Malformed status line"));
2029 hs->error = xstrdup (_("(no description)"));
2031 hs->error = xstrdup (message);
2032 xfree_null (message);
2034 type = resp_header_strdup (resp, "Content-Type");
2037 char *tmp = strchr (type, ';');
2040 while (tmp > type && c_isspace (tmp[-1]))
2045 hs->newloc = resp_header_strdup (resp, "Location");
2046 hs->remote_time = resp_header_strdup (resp, "Last-Modified");
2048 /* Handle (possibly multiple instances of) the Set-Cookie header. */
2052 const char *scbeg, *scend;
2053 /* The jar should have been created by now. */
2054 assert (wget_cookie_jar != NULL);
2056 (scpos = resp_header_locate (resp, "Set-Cookie", scpos,
2057 &scbeg, &scend)) != -1;
2060 char *set_cookie; BOUNDED_TO_ALLOCA (scbeg, scend, set_cookie);
2061 cookie_handle_set_cookie (wget_cookie_jar, u->host, u->port,
2062 u->path, set_cookie);
2066 if (resp_header_copy (resp, "Content-Range", hdrval, sizeof (hdrval)))
2068 wgint first_byte_pos, last_byte_pos, entity_length;
2069 if (parse_content_range (hdrval, &first_byte_pos, &last_byte_pos,
2072 contrange = first_byte_pos;
2073 contlen = last_byte_pos - first_byte_pos + 1;
2078 /* 20x responses are counted among successful by default. */
2079 if (H_20X (statcode))
2082 /* Return if redirected. */
2083 if (H_REDIRECTED (statcode) || statcode == HTTP_STATUS_MULTIPLE_CHOICES)
2085 /* RFC2068 says that in case of the 300 (multiple choices)
2086 response, the server can output a preferred URL through
2087 `Location' header; otherwise, the request should be treated
2088 like GET. So, if the location is set, it will be a
2089 redirection; otherwise, just proceed normally. */
2090 if (statcode == HTTP_STATUS_MULTIPLE_CHOICES && !hs->newloc)
2094 logprintf (LOG_VERBOSE,
2095 _("Location: %s%s\n"),
2096 hs->newloc ? escnonprint_uri (hs->newloc) : _("unspecified"),
2097 hs->newloc ? _(" [following]") : "");
2098 if (keep_alive && !head_only && skip_short_body (sock, contlen))
2099 CLOSE_FINISH (sock);
2101 CLOSE_INVALIDATE (sock);
2107 /* If content-type is not given, assume text/html. This is because
2108 of the multitude of broken CGI's that "forget" to generate the
2111 0 == strncasecmp (type, TEXTHTML_S, strlen (TEXTHTML_S)) ||
2112 0 == strncasecmp (type, TEXTXHTML_S, strlen (TEXTXHTML_S)))
2117 if (opt.html_extension && (*dt & TEXTHTML))
2118 /* -E / --html-extension / html_extension = on was specified, and this is a
2119 text/html file. If some case-insensitive variation on ".htm[l]" isn't
2120 already the file's suffix, tack on ".html". */
2122 char *last_period_in_local_filename = strrchr (hs->local_file, '.');
2124 if (last_period_in_local_filename == NULL
2125 || !(0 == strcasecmp (last_period_in_local_filename, ".htm")
2126 || 0 == strcasecmp (last_period_in_local_filename, ".html")))
2128 int local_filename_len = strlen (hs->local_file);
2129 /* Resize the local file, allowing for ".html" preceded by
2130 optional ".NUMBER". */
2131 hs->local_file = xrealloc (hs->local_file,
2132 local_filename_len + 24 + sizeof (".html"));
2133 strcpy(hs->local_file + local_filename_len, ".html");
2134 /* If clobbering is not allowed and the file, as named,
2135 exists, tack on ".NUMBER.html" instead. */
2136 if (!ALLOW_CLOBBER && file_exists_p (hs->local_file))
2140 sprintf (hs->local_file + local_filename_len,
2141 ".%d.html", ext_num++);
2142 while (file_exists_p (hs->local_file));
2144 *dt |= ADDED_HTML_EXTENSION;
2148 if (statcode == HTTP_STATUS_RANGE_NOT_SATISFIABLE)
2150 /* If `-c' is in use and the file has been fully downloaded (or
2151 the remote file has shrunk), Wget effectively requests bytes
2152 after the end of file and the server response with 416. */
2153 logputs (LOG_VERBOSE, _("\
2154 \n The file is already fully retrieved; nothing to do.\n\n"));
2155 /* In case the caller inspects. */
2158 /* Mark as successfully retrieved. */
2161 CLOSE_INVALIDATE (sock); /* would be CLOSE_FINISH, but there
2162 might be more bytes in the body. */
2163 return RETRUNNEEDED;
2165 if ((contrange != 0 && contrange != hs->restval)
2166 || (H_PARTIAL (statcode) && !contrange))
2168 /* The Range request was somehow misunderstood by the server.
2171 CLOSE_INVALIDATE (sock);
2177 hs->contlen = contlen + contrange;
2183 /* No need to print this output if the body won't be
2184 downloaded at all, or if the original server response is
2186 logputs (LOG_VERBOSE, _("Length: "));
2189 logputs (LOG_VERBOSE, number_to_static_string (contlen + contrange));
2190 if (contlen + contrange >= 1024)
2191 logprintf (LOG_VERBOSE, " (%s)",
2192 human_readable (contlen + contrange));
2195 if (contlen >= 1024)
2196 logprintf (LOG_VERBOSE, _(", %s (%s) remaining"),
2197 number_to_static_string (contlen),
2198 human_readable (contlen));
2200 logprintf (LOG_VERBOSE, _(", %s remaining"),
2201 number_to_static_string (contlen));
2205 logputs (LOG_VERBOSE,
2206 opt.ignore_length ? _("ignored") : _("unspecified"));
2208 logprintf (LOG_VERBOSE, " [%s]\n", escnonprint (type));
2210 logputs (LOG_VERBOSE, "\n");
2214 type = NULL; /* We don't need it any more. */
2216 /* Return if we have no intention of further downloading. */
2217 if (!(*dt & RETROKF) || head_only)
2219 /* In case the caller cares to look... */
2224 /* Pre-1.10 Wget used CLOSE_INVALIDATE here. Now we trust the
2225 servers not to send body in response to a HEAD request, and
2226 those that do will likely be caught by test_socket_open.
2227 If not, they can be worked around using
2228 `--no-http-keep-alive'. */
2229 CLOSE_FINISH (sock);
2230 else if (keep_alive && skip_short_body (sock, contlen))
2231 /* Successfully skipped the body; also keep using the socket. */
2232 CLOSE_FINISH (sock);
2234 CLOSE_INVALIDATE (sock);
2235 return RETRFINISHED;
2238 /* Open the local file. */
2241 mkalldirs (hs->local_file);
2243 rotate_backups (hs->local_file);
2245 fp = fopen (hs->local_file, "ab");
2246 else if (ALLOW_CLOBBER)
2247 fp = fopen (hs->local_file, "wb");
2250 fp = fopen_excl (hs->local_file, true);
2251 if (!fp && errno == EEXIST)
2253 /* We cannot just invent a new name and use it (which is
2254 what functions like unique_create typically do)
2255 because we told the user we'd use this name.
2256 Instead, return and retry the download. */
2257 logprintf (LOG_NOTQUIET,
2258 _("%s has sprung into existence.\n"),
2260 CLOSE_INVALIDATE (sock);
2261 return FOPEN_EXCL_ERR;
2266 logprintf (LOG_NOTQUIET, "%s: %s\n", hs->local_file, strerror (errno));
2267 CLOSE_INVALIDATE (sock);
2274 /* Print fetch message, if opt.verbose. */
2277 logprintf (LOG_NOTQUIET, _("Saving to: %s\n"),
2278 HYPHENP (hs->local_file) ? quote ("STDOUT") : quote (hs->local_file));
2281 /* This confuses the timestamping code that checks for file size.
2282 #### The timestamping code should be smarter about file size. */
2283 if (opt.save_headers && hs->restval == 0)
2284 fwrite (head, 1, strlen (head), fp);
2286 /* Now we no longer need to store the response header. */
2289 /* Download the request body. */
2292 /* If content-length is present, read that much; otherwise, read
2293 until EOF. The HTTP spec doesn't require the server to
2294 actually close the connection when it's done sending data. */
2295 flags |= rb_read_exactly;
2296 if (hs->restval > 0 && contrange == 0)
2297 /* If the server ignored our range request, instruct fd_read_body
2298 to skip the first RESTVAL bytes of body. */
2299 flags |= rb_skip_startpos;
2300 hs->len = hs->restval;
2302 hs->res = fd_read_body (sock, fp, contlen != -1 ? contlen : 0,
2303 hs->restval, &hs->rd_size, &hs->len, &hs->dltime,
2307 CLOSE_FINISH (sock);
2311 hs->rderrmsg = xstrdup (fd_errstr (sock));
2312 CLOSE_INVALIDATE (sock);
2319 return RETRFINISHED;
2322 /* The genuine HTTP loop! This is the part where the retrieval is
2323 retried, and retried, and retried, and... */
2325 http_loop (struct url *u, char **newloc, char **local_file, const char *referer,
2326 int *dt, struct url *proxy)
2329 bool got_head = false; /* used for time-stamping and filename detection */
2330 bool time_came_from_head = false;
2331 bool got_name = false;
2334 uerr_t err, ret = TRYLIMEXC;
2335 time_t tmr = -1; /* remote time-stamp */
2336 struct http_stat hstat; /* HTTP status */
2338 bool send_head_first = true;
2340 /* Assert that no value for *LOCAL_FILE was passed. */
2341 assert (local_file == NULL || *local_file == NULL);
2343 /* Set LOCAL_FILE parameter. */
2344 if (local_file && opt.output_document)
2345 *local_file = HYPHENP (opt.output_document) ? NULL : xstrdup (opt.output_document);
2347 /* Reset NEWLOC parameter. */
2350 /* This used to be done in main(), but it's a better idea to do it
2351 here so that we don't go through the hoops if we're just using
2356 /* Warn on (likely bogus) wildcard usage in HTTP. */
2357 if (opt.ftp_glob && has_wildcards_p (u->path))
2358 logputs (LOG_VERBOSE, _("Warning: wildcards not supported in HTTP.\n"));
2360 /* Setup hstat struct. */
2362 hstat.referer = referer;
2364 if (opt.output_document)
2366 hstat.local_file = xstrdup (opt.output_document);
2369 else if (!opt.content_disposition)
2371 hstat.local_file = url_file_name (u);
2375 /* TODO: Ick! This code is now in both gethttp and http_loop, and is
2376 * screaming for some refactoring. */
2377 if (got_name && file_exists_p (hstat.local_file) && opt.noclobber)
2379 /* If opt.noclobber is turned on and file already exists, do not
2380 retrieve the file */
2381 logprintf (LOG_VERBOSE, _("\
2382 File %s already there; not retrieving.\n\n"),
2383 quote (hstat.local_file));
2384 /* If the file is there, we suppose it's retrieved OK. */
2387 /* #### Bogusness alert. */
2388 /* If its suffix is "html" or "htm" or similar, assume text/html. */
2389 if (has_html_suffix_p (hstat.local_file))
2392 return RETRUNNEEDED;
2395 /* Reset the counter. */
2398 /* Reset the document type. */
2401 /* Skip preliminary HEAD request if we're not in spider mode AND
2402 * if -O was given or HTTP Content-Disposition support is disabled. */
2404 && (got_name || !opt.content_disposition))
2405 send_head_first = false;
2407 /* Send preliminary HEAD request if -N is given and we have an existing
2408 * destination file. */
2409 if (opt.timestamping
2410 && !opt.content_disposition
2411 && file_exists_p (url_file_name (u)))
2412 send_head_first = true;
2417 /* Increment the pass counter. */
2419 sleep_between_retrievals (count);
2421 /* Get the current time string. */
2422 tms = datetime_str (time (NULL));
2424 if (opt.spider && !got_head)
2425 logprintf (LOG_VERBOSE, _("\
2426 Spider mode enabled. Check if remote file exists.\n"));
2428 /* Print fetch message, if opt.verbose. */
2431 char *hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2436 sprintf (tmp, _("(try:%2d)"), count);
2437 logprintf (LOG_NOTQUIET, "--%s-- %s %s\n",
2442 logprintf (LOG_NOTQUIET, "--%s-- %s\n",
2447 ws_changetitle (hurl);
2452 /* Default document type is empty. However, if spider mode is
2453 on or time-stamping is employed, HEAD_ONLY commands is
2454 encoded within *dt. */
2455 if (send_head_first && !got_head)
2460 /* Decide whether or not to restart. */
2463 && stat (hstat.local_file, &st) == 0
2464 && S_ISREG (st.st_mode))
2465 /* When -c is used, continue from on-disk size. (Can't use
2466 hstat.len even if count>1 because we don't want a failed
2467 first attempt to clobber existing data.) */
2468 hstat.restval = st.st_size;
2470 /* otherwise, continue where the previous try left off */
2471 hstat.restval = hstat.len;
2475 /* Decide whether to send the no-cache directive. We send it in
2477 a) we're using a proxy, and we're past our first retrieval.
2478 Some proxies are notorious for caching incomplete data, so
2479 we require a fresh get.
2480 b) caching is explicitly inhibited. */
2481 if ((proxy && count > 1) /* a */
2482 || !opt.allow_cache) /* b */
2483 *dt |= SEND_NOCACHE;
2485 *dt &= ~SEND_NOCACHE;
2487 /* Try fetching the document, or at least its head. */
2488 err = gethttp (u, &hstat, dt, proxy);
2491 tms = datetime_str (time (NULL));
2493 /* Get the new location (with or without the redirection). */
2495 *newloc = xstrdup (hstat.newloc);
2499 case HERR: case HEOF: case CONSOCKERR: case CONCLOSED:
2500 case CONERROR: case READERR: case WRITEFAILED:
2501 case RANGEERR: case FOPEN_EXCL_ERR:
2502 /* Non-fatal errors continue executing the loop, which will
2503 bring them to "while" statement at the end, to judge
2504 whether the number of tries was exceeded. */
2505 printwhat (count, opt.ntry);
2507 case FWRITEERR: case FOPENERR:
2508 /* Another fatal error. */
2509 logputs (LOG_VERBOSE, "\n");
2510 logprintf (LOG_NOTQUIET, _("Cannot write to %s (%s).\n"),
2511 quote (hstat.local_file), strerror (errno));
2512 case HOSTERR: case CONIMPOSSIBLE: case PROXERR: case AUTHFAILED:
2513 case SSLINITFAILED: case CONTNOTSUPPORTED:
2514 /* Fatal errors just return from the function. */
2518 /* Another fatal error. */
2519 logprintf (LOG_NOTQUIET, _("Unable to establish SSL connection.\n"));
2523 /* Return the new location to the caller. */
2526 logprintf (LOG_NOTQUIET,
2527 _("ERROR: Redirection (%d) without location.\n"),
2537 /* The file was already fully retrieved. */
2541 /* Deal with you later. */
2544 /* All possibilities should have been exhausted. */
2548 if (!(*dt & RETROKF))
2553 /* #### Ugly ugly ugly! */
2554 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2555 logprintf (LOG_NONVERBOSE, "%s:\n", hurl);
2558 /* Fall back to GET if HEAD fails with a 500 or 501 error code. */
2560 && (hstat.statcode == 500 || hstat.statcode == 501))
2565 /* Maybe we should always keep track of broken links, not just in
2567 else if (opt.spider)
2569 /* #### Again: ugly ugly ugly! */
2571 hurl = url_string (u, URL_AUTH_HIDE_PASSWD);
2572 nonexisting_url (hurl);
2573 logprintf (LOG_NOTQUIET, _("\
2574 Remote file does not exist -- broken link!!!\n"));
2578 logprintf (LOG_NOTQUIET, _("%s ERROR %d: %s.\n"),
2579 tms, hstat.statcode, escnonprint (hstat.error));
2581 logputs (LOG_VERBOSE, "\n");
2587 /* Did we get the time-stamp? */
2590 got_head = true; /* no more time-stamping */
2592 if (opt.timestamping && !hstat.remote_time)
2594 logputs (LOG_NOTQUIET, _("\
2595 Last-modified header missing -- time-stamps turned off.\n"));
2597 else if (hstat.remote_time)
2599 /* Convert the date-string into struct tm. */
2600 tmr = http_atotm (hstat.remote_time);
2601 if (tmr == (time_t) (-1))
2602 logputs (LOG_VERBOSE, _("\
2603 Last-modified header invalid -- time-stamp ignored.\n"));
2604 if (*dt & HEAD_ONLY)
2605 time_came_from_head = true;
2608 if (send_head_first)
2610 /* The time-stamping section. */
2611 if (opt.timestamping)
2613 if (hstat.orig_file_name) /* Perform the following
2614 checks only if the file
2616 download already exists. */
2618 if (hstat.remote_time &&
2619 tmr != (time_t) (-1))
2621 /* Now time-stamping can be used validly.
2622 Time-stamping means that if the sizes of
2623 the local and remote file match, and local
2624 file is newer than the remote file, it will
2625 not be retrieved. Otherwise, the normal
2626 download procedure is resumed. */
2627 if (hstat.orig_file_tstamp >= tmr)
2629 if (hstat.contlen == -1
2630 || hstat.orig_file_size == hstat.contlen)
2632 logprintf (LOG_VERBOSE, _("\
2633 Server file no newer than local file %s -- not retrieving.\n\n"),
2634 quote (hstat.orig_file_name));
2640 logprintf (LOG_VERBOSE, _("\
2641 The sizes do not match (local %s) -- retrieving.\n"),
2642 number_to_static_string (hstat.orig_file_size));
2646 logputs (LOG_VERBOSE,
2647 _("Remote file is newer, retrieving.\n"));
2649 logputs (LOG_VERBOSE, "\n");
2653 /* free_hstat (&hstat); */
2654 hstat.timestamp_checked = true;
2663 logputs (LOG_VERBOSE, _("\
2664 Remote file exists and could contain links to other resources -- retrieving.\n\n"));
2668 logprintf (LOG_VERBOSE, _("\
2669 Remote file exists but does not contain any link -- not retrieving.\n\n"));
2670 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2678 logprintf (LOG_VERBOSE, _("\
2679 Remote file exists and could contain further links,\n\
2680 but recursion is disabled -- not retrieving.\n\n"));
2684 logprintf (LOG_VERBOSE, _("\
2685 Remote file exists.\n\n"));
2687 ret = RETROK; /* RETRUNNEEDED is not for caller. */
2694 count = 0; /* the retrieve count for HEAD is reset */
2696 } /* send_head_first */
2699 if ((tmr != (time_t) (-1))
2700 && ((hstat.len == hstat.contlen) ||
2701 ((hstat.res == 0) && (hstat.contlen == -1))))
2703 /* #### This code repeats in http.c and ftp.c. Move it to a
2705 const char *fl = NULL;
2706 if (opt.output_document)
2708 if (output_stream_regular)
2709 fl = opt.output_document;
2712 fl = hstat.local_file;
2716 /* Reparse time header, in case it's changed. */
2717 if (time_came_from_head
2718 && hstat.remote_time && hstat.remote_time[0])
2720 newtmr = http_atotm (hstat.remote_time);
2727 /* End of time-stamping section. */
2729 tmrate = retr_rate (hstat.rd_size, hstat.dltime);
2730 total_download_time += hstat.dltime;
2732 if (hstat.len == hstat.contlen)
2736 logprintf (LOG_VERBOSE,
2737 _("%s (%s) - %s saved [%s/%s]\n\n"),
2738 tms, tmrate, quote (hstat.local_file),
2739 number_to_static_string (hstat.len),
2740 number_to_static_string (hstat.contlen));
2741 logprintf (LOG_NONVERBOSE,
2742 "%s URL:%s [%s/%s] -> \"%s\" [%d]\n",
2744 number_to_static_string (hstat.len),
2745 number_to_static_string (hstat.contlen),
2746 hstat.local_file, count);
2749 total_downloaded_bytes += hstat.len;
2751 /* Remember that we downloaded the file for later ".orig" code. */
2752 if (*dt & ADDED_HTML_EXTENSION)
2753 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2755 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2760 else if (hstat.res == 0) /* No read error */
2762 if (hstat.contlen == -1) /* We don't know how much we were supposed
2763 to get, so assume we succeeded. */
2767 logprintf (LOG_VERBOSE,
2768 _("%s (%s) - %s saved [%s]\n\n"),
2769 tms, tmrate, quote (hstat.local_file),
2770 number_to_static_string (hstat.len));
2771 logprintf (LOG_NONVERBOSE,
2772 "%s URL:%s [%s] -> \"%s\" [%d]\n",
2773 tms, u->url, number_to_static_string (hstat.len),
2774 hstat.local_file, count);
2777 total_downloaded_bytes += hstat.len;
2779 /* Remember that we downloaded the file for later ".orig" code. */
2780 if (*dt & ADDED_HTML_EXTENSION)
2781 downloaded_file(FILE_DOWNLOADED_AND_HTML_EXTENSION_ADDED, hstat.local_file);
2783 downloaded_file(FILE_DOWNLOADED_NORMALLY, hstat.local_file);
2788 else if (hstat.len < hstat.contlen) /* meaning we lost the
2789 connection too soon */
2791 logprintf (LOG_VERBOSE,
2792 _("%s (%s) - Connection closed at byte %s. "),
2793 tms, tmrate, number_to_static_string (hstat.len));
2794 printwhat (count, opt.ntry);
2798 /* Getting here would mean reading more data than
2799 requested with content-length, which we never do. */
2802 else /* from now on hstat.res can only be -1 */
2804 if (hstat.contlen == -1)
2806 logprintf (LOG_VERBOSE,
2807 _("%s (%s) - Read error at byte %s (%s)."),
2808 tms, tmrate, number_to_static_string (hstat.len),
2810 printwhat (count, opt.ntry);
2813 else /* hstat.res == -1 and contlen is given */
2815 logprintf (LOG_VERBOSE,
2816 _("%s (%s) - Read error at byte %s/%s (%s). "),
2818 number_to_static_string (hstat.len),
2819 number_to_static_string (hstat.contlen),
2821 printwhat (count, opt.ntry);
2827 while (!opt.ntry || (count < opt.ntry));
2831 *local_file = xstrdup (hstat.local_file);
2832 free_hstat (&hstat);
2837 /* Check whether the result of strptime() indicates success.
2838 strptime() returns the pointer to how far it got to in the string.
2839 The processing has been successful if the string is at `GMT' or
2840 `+X', or at the end of the string.
2842 In extended regexp parlance, the function returns 1 if P matches
2843 "^ *(GMT|[+-][0-9]|$)", 0 otherwise. P being NULL (which strptime
2844 can return) is considered a failure and 0 is returned. */
2846 check_end (const char *p)
2850 while (c_isspace (*p))
2853 || (p[0] == 'G' && p[1] == 'M' && p[2] == 'T')
2854 || ((p[0] == '+' || p[0] == '-') && c_isdigit (p[1])))
2860 /* Convert the textual specification of time in TIME_STRING to the
2861 number of seconds since the Epoch.
2863 TIME_STRING can be in any of the three formats RFC2616 allows the
2864 HTTP servers to emit -- RFC1123-date, RFC850-date or asctime-date,
2865 as well as the time format used in the Set-Cookie header.
2866 Timezones are ignored, and should be GMT.
2868 Return the computed time_t representation, or -1 if the conversion
2871 This function uses strptime with various string formats for parsing
2872 TIME_STRING. This results in a parser that is not as lenient in
2873 interpreting TIME_STRING as I would like it to be. Being based on
2874 strptime, it always allows shortened months, one-digit days, etc.,
2875 but due to the multitude of formats in which time can be
2876 represented, an ideal HTTP time parser would be even more
2877 forgiving. It should completely ignore things like week days and
2878 concentrate only on the various forms of representing years,
2879 months, days, hours, minutes, and seconds. For example, it would
2880 be nice if it accepted ISO 8601 out of the box.
2882 I've investigated free and PD code for this purpose, but none was
2883 usable. getdate was big and unwieldy, and had potential copyright
2884 issues, or so I was informed. Dr. Marcus Hennecke's atotm(),
2885 distributed with phttpd, is excellent, but we cannot use it because
2886 it is not assigned to the FSF. So I stuck it with strptime. */
2889 http_atotm (const char *time_string)
2891 /* NOTE: Solaris strptime man page claims that %n and %t match white
2892 space, but that's not universally available. Instead, we simply
2893 use ` ' to mean "skip all WS", which works under all strptime
2894 implementations I've tested. */
2896 static const char *time_formats[] = {
2897 "%a, %d %b %Y %T", /* rfc1123: Thu, 29 Jan 1998 22:12:57 */
2898 "%A, %d-%b-%y %T", /* rfc850: Thursday, 29-Jan-98 22:12:57 */
2899 "%a %b %d %T %Y", /* asctime: Thu Jan 29 22:12:57 1998 */
2900 "%a, %d-%b-%Y %T" /* cookies: Thu, 29-Jan-1998 22:12:57
2901 (used in Set-Cookie, defined in the
2902 Netscape cookie specification.) */
2904 const char *oldlocale;
2906 time_t ret = (time_t) -1;
2908 /* Solaris strptime fails to recognize English month names in
2909 non-English locales, which we work around by temporarily setting
2910 locale to C before invoking strptime. */
2911 oldlocale = setlocale (LC_TIME, NULL);
2912 setlocale (LC_TIME, "C");
2914 for (i = 0; i < countof (time_formats); i++)
2918 /* Some versions of strptime use the existing contents of struct
2919 tm to recalculate the date according to format. Zero it out
2920 to prevent stack garbage from influencing strptime. */
2923 if (check_end (strptime (time_string, time_formats[i], &t)))
2930 /* Restore the previous locale. */
2931 setlocale (LC_TIME, oldlocale);
2936 /* Authorization support: We support three authorization schemes:
2938 * `Basic' scheme, consisting of base64-ing USER:PASSWORD string;
2940 * `Digest' scheme, added by Junio Hamano <junio@twinsun.com>,
2941 consisting of answering to the server's challenge with the proper
2944 * `NTLM' ("NT Lan Manager") scheme, based on code written by Daniel
2945 Stenberg for libcurl. Like digest, NTLM is based on a
2946 challenge-response mechanism, but unlike digest, it is non-standard
2947 (authenticates TCP connections rather than requests), undocumented
2948 and Microsoft-specific. */
2950 /* Create the authentication header contents for the `Basic' scheme.
2951 This is done by encoding the string "USER:PASS" to base64 and
2952 prepending the string "Basic " in front of it. */
2955 basic_authentication_encode (const char *user, const char *passwd)
2958 int len1 = strlen (user) + 1 + strlen (passwd);
2960 t1 = (char *)alloca (len1 + 1);
2961 sprintf (t1, "%s:%s", user, passwd);
2963 t2 = (char *)alloca (BASE64_LENGTH (len1) + 1);
2964 base64_encode (t1, len1, t2);
2966 return concat_strings ("Basic ", t2, (char *) 0);
2969 #define SKIP_WS(x) do { \
2970 while (c_isspace (*(x))) \
2974 #ifdef ENABLE_DIGEST
2975 /* Dump the hexadecimal representation of HASH to BUF. HASH should be
2976 an array of 16 bytes containing the hash keys, and BUF should be a
2977 buffer of 33 writable characters (32 for hex digits plus one for
2978 zero termination). */
2980 dump_hash (char *buf, const unsigned char *hash)
2984 for (i = 0; i < MD5_HASHLEN; i++, hash++)
2986 *buf++ = XNUM_TO_digit (*hash >> 4);
2987 *buf++ = XNUM_TO_digit (*hash & 0xf);
2992 /* Take the line apart to find the challenge, and compose a digest
2993 authorization header. See RFC2069 section 2.1.2. */
2995 digest_authentication_encode (const char *au, const char *user,
2996 const char *passwd, const char *method,
2999 static char *realm, *opaque, *nonce;
3004 { "realm", &realm },
3005 { "opaque", &opaque },
3009 param_token name, value;
3011 realm = opaque = nonce = NULL;
3013 au += 6; /* skip over `Digest' */
3014 while (extract_param (&au, &name, &value, ','))
3017 for (i = 0; i < countof (options); i++)
3018 if (name.e - name.b == strlen (options[i].name)
3019 && 0 == strncmp (name.b, options[i].name, name.e - name.b))
3021 *options[i].variable = strdupdelim (value.b, value.e);
3025 if (!realm || !nonce || !user || !passwd || !path || !method)
3028 xfree_null (opaque);
3033 /* Calculate the digest value. */
3035 ALLOCA_MD5_CONTEXT (ctx);
3036 unsigned char hash[MD5_HASHLEN];
3037 char a1buf[MD5_HASHLEN * 2 + 1], a2buf[MD5_HASHLEN * 2 + 1];
3038 char response_digest[MD5_HASHLEN * 2 + 1];
3040 /* A1BUF = H(user ":" realm ":" password) */
3042 gen_md5_update ((unsigned char *)user, strlen (user), ctx);
3043 gen_md5_update ((unsigned char *)":", 1, ctx);
3044 gen_md5_update ((unsigned char *)realm, strlen (realm), ctx);
3045 gen_md5_update ((unsigned char *)":", 1, ctx);
3046 gen_md5_update ((unsigned char *)passwd, strlen (passwd), ctx);
3047 gen_md5_finish (ctx, hash);
3048 dump_hash (a1buf, hash);
3050 /* A2BUF = H(method ":" path) */
3052 gen_md5_update ((unsigned char *)method, strlen (method), ctx);
3053 gen_md5_update ((unsigned char *)":", 1, ctx);
3054 gen_md5_update ((unsigned char *)path, strlen (path), ctx);
3055 gen_md5_finish (ctx, hash);
3056 dump_hash (a2buf, hash);
3058 /* RESPONSE_DIGEST = H(A1BUF ":" nonce ":" A2BUF) */
3060 gen_md5_update ((unsigned char *)a1buf, MD5_HASHLEN * 2, ctx);
3061 gen_md5_update ((unsigned char *)":", 1, ctx);
3062 gen_md5_update ((unsigned char *)nonce, strlen (nonce), ctx);
3063 gen_md5_update ((unsigned char *)":", 1, ctx);
3064 gen_md5_update ((unsigned char *)a2buf, MD5_HASHLEN * 2, ctx);
3065 gen_md5_finish (ctx, hash);
3066 dump_hash (response_digest, hash);
3068 res = xmalloc (strlen (user)
3073 + 2 * MD5_HASHLEN /*strlen (response_digest)*/
3074 + (opaque ? strlen (opaque) : 0)
3076 sprintf (res, "Digest \
3077 username=\"%s\", realm=\"%s\", nonce=\"%s\", uri=\"%s\", response=\"%s\"",
3078 user, realm, nonce, path, response_digest);
3081 char *p = res + strlen (res);
3082 strcat (p, ", opaque=\"");
3089 #endif /* ENABLE_DIGEST */
3091 /* Computing the size of a string literal must take into account that
3092 value returned by sizeof includes the terminating \0. */
3093 #define STRSIZE(literal) (sizeof (literal) - 1)
3095 /* Whether chars in [b, e) begin with the literal string provided as
3096 first argument and are followed by whitespace or terminating \0.
3097 The comparison is case-insensitive. */
3098 #define STARTS(literal, b, e) \
3099 ((e) - (b) >= STRSIZE (literal) \
3100 && 0 == strncasecmp (b, literal, STRSIZE (literal)) \
3101 && ((e) - (b) == STRSIZE (literal) \
3102 || c_isspace (b[STRSIZE (literal)])))
3105 known_authentication_scheme_p (const char *hdrbeg, const char *hdrend)
3107 return STARTS ("Basic", hdrbeg, hdrend)
3108 #ifdef ENABLE_DIGEST
3109 || STARTS ("Digest", hdrbeg, hdrend)
3112 || STARTS ("NTLM", hdrbeg, hdrend)
3119 /* Create the HTTP authorization request header. When the
3120 `WWW-Authenticate' response header is seen, according to the
3121 authorization scheme specified in that header (`Basic' and `Digest'
3122 are supported by the current implementation), produce an
3123 appropriate HTTP authorization request header. */
3125 create_authorization_line (const char *au, const char *user,
3126 const char *passwd, const char *method,
3127 const char *path, bool *finished)
3129 /* We are called only with known schemes, so we can dispatch on the
3131 switch (c_toupper (*au))
3133 case 'B': /* Basic */
3135 return basic_authentication_encode (user, passwd);
3136 #ifdef ENABLE_DIGEST
3137 case 'D': /* Digest */
3139 return digest_authentication_encode (au, user, passwd, method, path);
3142 case 'N': /* NTLM */
3143 if (!ntlm_input (&pconn.ntlm, au))
3148 return ntlm_output (&pconn.ntlm, user, passwd, finished);
3151 /* We shouldn't get here -- this function should be only called
3152 with values approved by known_authentication_scheme_p. */
3160 if (!wget_cookie_jar)
3161 wget_cookie_jar = cookie_jar_new ();
3162 if (opt.cookies_input && !cookies_loaded_p)
3164 cookie_jar_load (wget_cookie_jar, opt.cookies_input);
3165 cookies_loaded_p = true;
3172 if (wget_cookie_jar)
3173 cookie_jar_save (wget_cookie_jar, opt.cookies_output);
3179 xfree_null (pconn.host);
3180 if (wget_cookie_jar)
3181 cookie_jar_delete (wget_cookie_jar);
3188 test_parse_content_disposition()
3193 char *opt_dir_prefix;
3197 { "filename=\"file.ext\"", NULL, "file.ext", true },
3198 { "filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3199 { "attachment; filename=\"file.ext\"", NULL, "file.ext", true },
3200 { "attachment; filename=\"file.ext\"", "somedir", "somedir/file.ext", true },
3201 { "attachment; filename=\"file.ext\"; dummy", NULL, "file.ext", true },
3202 { "attachment; filename=\"file.ext\"; dummy", "somedir", "somedir/file.ext", true },
3203 { "attachment", NULL, NULL, false },
3204 { "attachment", "somedir", NULL, false },
3207 for (i = 0; i < sizeof(test_array)/sizeof(test_array[0]); ++i)
3212 opt.dir_prefix = test_array[i].opt_dir_prefix;
3213 res = parse_content_disposition (test_array[i].hdrval, &filename);
3215 mu_assert ("test_parse_content_disposition: wrong result",
3216 res == test_array[i].result
3218 || 0 == strcmp (test_array[i].filename, filename)));
3224 #endif /* TESTING */
3227 * vim: et sts=2 sw=2 cino+={s