1 /* Handling of recursive HTTP retrieving.
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. */
38 #endif /* HAVE_UNISTD_H */
53 /* Functions for maintaining the URL queue. */
55 struct queue_element {
56 const char *url; /* the URL to download */
57 const char *referer; /* the referring document */
58 int depth; /* the depth */
59 bool html_allowed; /* whether the document is allowed to
60 be treated as HTML. */
62 struct queue_element *next; /* next element in queue */
66 struct queue_element *head;
67 struct queue_element *tail;
71 /* Create a URL queue. */
73 static struct url_queue *
76 struct url_queue *queue = xnew0 (struct url_queue);
80 /* Delete a URL queue. */
83 url_queue_delete (struct url_queue *queue)
88 /* Enqueue a URL in the queue. The queue is FIFO: the items will be
89 retrieved ("dequeued") from the queue in the order they were placed
93 url_enqueue (struct url_queue *queue,
94 const char *url, const char *referer, int depth, bool html_allowed)
96 struct queue_element *qel = xnew (struct queue_element);
98 qel->referer = referer;
100 qel->html_allowed = html_allowed;
104 if (queue->count > queue->maxcount)
105 queue->maxcount = queue->count;
107 DEBUGP (("Enqueuing %s at depth %d\n", url, depth));
108 DEBUGP (("Queue count %d, maxcount %d.\n", queue->count, queue->maxcount));
111 queue->tail->next = qel;
115 queue->head = queue->tail;
118 /* Take a URL out of the queue. Return true if this operation
119 succeeded, or false if the queue is empty. */
122 url_dequeue (struct url_queue *queue,
123 const char **url, const char **referer, int *depth,
126 struct queue_element *qel = queue->head;
131 queue->head = queue->head->next;
136 *referer = qel->referer;
138 *html_allowed = qel->html_allowed;
142 DEBUGP (("Dequeuing %s at depth %d\n", qel->url, qel->depth));
143 DEBUGP (("Queue count %d, maxcount %d.\n", queue->count, queue->maxcount));
149 static bool download_child_p (const struct urlpos *, struct url *, int,
150 struct url *, struct hash_table *);
151 static bool descend_redirect_p (const char *, const char *, int,
152 struct url *, struct hash_table *);
155 /* Retrieve a part of the web beginning with START_URL. This used to
156 be called "recursive retrieval", because the old function was
157 recursive and implemented depth-first search. retrieve_tree on the
158 other hand implements breadth-search traversal of the tree, which
159 results in much nicer ordering of downloads.
161 The algorithm this function uses is simple:
163 1. put START_URL in the queue.
164 2. while there are URLs in the queue:
166 3. get next URL from the queue.
168 5. if the URL is HTML and its depth does not exceed maximum depth,
169 get the list of URLs embedded therein.
170 6. for each of those URLs do the following:
172 7. if the URL is not one of those downloaded before, and if it
173 satisfies the criteria specified by the various command-line
174 options, add it to the queue. */
177 retrieve_tree (const char *start_url)
179 uerr_t status = RETROK;
181 /* The queue of URLs we need to load. */
182 struct url_queue *queue;
184 /* The URLs we do not wish to enqueue, because they are already in
185 the queue, but haven't been downloaded yet. */
186 struct hash_table *blacklist;
189 struct url *start_url_parsed = url_parse (start_url, &up_error_code);
191 if (!start_url_parsed)
193 logprintf (LOG_NOTQUIET, "%s: %s.\n", start_url,
194 url_error (up_error_code));
198 queue = url_queue_new ();
199 blacklist = make_string_hash_table (0);
201 /* Enqueue the starting URL. Use start_url_parsed->url rather than
202 just URL so we enqueue the canonical form of the URL. */
203 url_enqueue (queue, xstrdup (start_url_parsed->url), NULL, 0, true);
204 string_set_add (blacklist, start_url_parsed->url);
208 bool descend = false;
209 char *url, *referer, *file = NULL;
212 bool dash_p_leaf_HTML = false;
214 if (opt.quota && total_downloaded_bytes > opt.quota)
216 if (status == FWRITEERR)
219 /* Get the next URL from the queue... */
221 if (!url_dequeue (queue,
222 (const char **)&url, (const char **)&referer,
223 &depth, &html_allowed))
226 /* ...and download it. Note that this download is in most cases
227 unconditional, as download_child_p already makes sure a file
228 doesn't get enqueued twice -- and yet this check is here, and
229 not in download_child_p. This is so that if you run `wget -r
230 URL1 URL2', and a random URL is encountered once under URL1
231 and again under URL2, but at a different (possibly smaller)
232 depth, we want the URL's children to be taken into account
234 if (dl_url_file_map && hash_table_contains (dl_url_file_map, url))
236 file = xstrdup (hash_table_get (dl_url_file_map, url));
238 DEBUGP (("Already downloaded \"%s\", reusing it from \"%s\".\n",
242 && downloaded_html_set
243 && string_set_contains (downloaded_html_set, file))
249 char *redirected = NULL;
251 status = retrieve_url (url, &file, &redirected, referer, &dt, false);
253 if (html_allowed && file && status == RETROK
254 && (dt & RETROKF) && (dt & TEXTHTML))
259 /* We have been redirected, possibly to another host, or
260 different path, or wherever. Check whether we really
261 want to follow it. */
264 if (!descend_redirect_p (redirected, url, depth,
265 start_url_parsed, blacklist))
268 /* Make sure that the old pre-redirect form gets
270 string_set_add (blacklist, url);
280 visited_url (url, referer);
284 && depth >= opt.reclevel && opt.reclevel != INFINITE_RECURSION)
286 if (opt.page_requisites
287 && (depth == opt.reclevel || depth == opt.reclevel + 1))
289 /* When -p is specified, we are allowed to exceed the
290 maximum depth, but only for the "inline" links,
291 i.e. those that are needed to display the page.
292 Originally this could exceed the depth at most by
293 one, but we allow one more level so that the leaf
294 pages that contain frames can be loaded
296 dash_p_leaf_HTML = true;
300 /* Either -p wasn't specified or it was and we've
301 already spent the two extra (pseudo-)levels that it
302 affords us, so we need to bail out. */
303 DEBUGP (("Not descending further; at depth %d, max. %d.\n",
304 depth, opt.reclevel));
309 /* If the downloaded document was HTML, parse it and enqueue the
310 links it contains. */
314 bool meta_disallow_follow = false;
315 struct urlpos *children
316 = get_urls_html (file, url, &meta_disallow_follow);
318 if (opt.use_robots && meta_disallow_follow)
320 free_urlpos (children);
326 struct urlpos *child = children;
327 struct url *url_parsed = url_parsed = url_parse (url, NULL);
328 char *referer_url = url;
329 bool strip_auth = (url_parsed != NULL
330 && url_parsed->user != NULL);
331 assert (url_parsed != NULL);
333 /* Strip auth info if present */
335 referer_url = url_string (url_parsed, URL_AUTH_HIDE);
337 for (; child; child = child->next)
339 if (child->ignore_when_downloading)
341 if (dash_p_leaf_HTML && !child->link_inline_p)
343 if (download_child_p (child, url_parsed, depth, start_url_parsed,
346 url_enqueue (queue, xstrdup (child->url->url),
347 xstrdup (referer_url), depth + 1,
348 child->link_expect_html);
349 /* We blacklist the URL we have enqueued, because we
350 don't want to enqueue (and hence download) the
352 string_set_add (blacklist, child->url->url);
358 url_free (url_parsed);
359 free_urlpos (children);
365 || opt.spider /* opt.recursive is implicitely true */
366 || !acceptable (file)))
368 /* Either --delete-after was specified, or we loaded this
369 (otherwise unneeded because of --spider or rejected by -R)
370 HTML file just to harvest its hyperlinks -- in either case,
371 delete the local file. */
372 DEBUGP (("Removing file due to %s in recursive_retrieve():\n",
373 opt.delete_after ? "--delete-after" :
374 (opt.spider ? "--spider" :
375 "recursive rejection criteria")));
376 logprintf (LOG_VERBOSE,
377 (opt.delete_after || opt.spider
378 ? _("Removing %s.\n")
379 : _("Removing %s since it should be rejected.\n")),
382 logprintf (LOG_NOTQUIET, "unlink: %s\n", strerror (errno));
383 logputs (LOG_VERBOSE, "\n");
384 register_delete_file (file);
388 xfree_null (referer);
392 /* If anything is left of the queue due to a premature exit, free it
398 while (url_dequeue (queue,
399 (const char **)&d1, (const char **)&d2, &d3, &d4))
405 url_queue_delete (queue);
407 if (start_url_parsed)
408 url_free (start_url_parsed);
409 string_set_free (blacklist);
411 if (opt.quota && total_downloaded_bytes > opt.quota)
413 else if (status == FWRITEERR)
419 /* Based on the context provided by retrieve_tree, decide whether a
420 URL is to be descended to. This is only ever called from
421 retrieve_tree, but is in a separate function for clarity.
423 The most expensive checks (such as those for robots) are memoized
424 by storing these URLs to BLACKLIST. This may or may not help. It
425 will help if those URLs are encountered many times. */
428 download_child_p (const struct urlpos *upos, struct url *parent, int depth,
429 struct url *start_url_parsed, struct hash_table *blacklist)
431 struct url *u = upos->url;
432 const char *url = u->url;
433 bool u_scheme_like_http;
435 DEBUGP (("Deciding whether to enqueue \"%s\".\n", url));
437 if (string_set_contains (blacklist, url))
441 char *referrer = url_string (parent, URL_AUTH_HIDE_PASSWD);
442 DEBUGP (("download_child_p: parent->url is: %s\n", quote (parent->url)));
443 visited_url (url, referrer);
446 DEBUGP (("Already on the black list.\n"));
450 /* Several things to check for:
451 1. if scheme is not http, and we don't load it
452 2. check for relative links (if relative_only is set)
454 4. check for no-parent
455 5. check for excludes && includes
457 7. check for same host (if spanhost is unset), with possible
458 gethostbyname baggage
459 8. check for robots.txt
461 Addendum: If the URL is FTP, and it is to be loaded, only the
462 domain and suffix settings are "stronger".
464 Note that .html files will get loaded regardless of suffix rules
465 (but that is remedied later with unlink) unless the depth equals
468 More time- and memory- consuming tests should be put later on
471 /* Determine whether URL under consideration has a HTTP-like scheme. */
472 u_scheme_like_http = schemes_are_similar_p (u->scheme, SCHEME_HTTP);
474 /* 1. Schemes other than HTTP are normally not recursed into. */
475 if (!u_scheme_like_http && !(u->scheme == SCHEME_FTP && opt.follow_ftp))
477 DEBUGP (("Not following non-HTTP schemes.\n"));
481 /* 2. If it is an absolute link and they are not followed, throw it
483 if (u_scheme_like_http)
484 if (opt.relative_only && !upos->link_relative_p)
486 DEBUGP (("It doesn't really look like a relative link.\n"));
490 /* 3. If its domain is not to be accepted/looked-up, chuck it
492 if (!accept_domain (u))
494 DEBUGP (("The domain was not accepted.\n"));
498 /* 4. Check for parent directory.
500 If we descended to a different host or changed the scheme, ignore
501 opt.no_parent. Also ignore it for documents needed to display
502 the parent page when in -p mode. */
504 && schemes_are_similar_p (u->scheme, start_url_parsed->scheme)
505 && 0 == strcasecmp (u->host, start_url_parsed->host)
506 && u->port == start_url_parsed->port
507 && !(opt.page_requisites && upos->link_inline_p))
509 if (!subdir_p (start_url_parsed->dir, u->dir))
511 DEBUGP (("Going to \"%s\" would escape \"%s\" with no_parent on.\n",
512 u->dir, start_url_parsed->dir));
517 /* 5. If the file does not match the acceptance list, or is on the
518 rejection list, chuck it out. The same goes for the directory
519 exclusion and inclusion lists. */
520 if (opt.includes || opt.excludes)
522 if (!accdir (u->dir))
524 DEBUGP (("%s (%s) is excluded/not-included.\n", url, u->dir));
529 /* 6. Check for acceptance/rejection rules. We ignore these rules
530 for directories (no file name to match) and for non-leaf HTMLs,
531 which can lead to other files that do need to be downloaded. (-p
532 automatically implies non-leaf because with -p we can, if
533 necesary, overstep the maximum depth to get the page requisites.) */
534 if (u->file[0] != '\0'
535 && !(has_html_suffix_p (u->file)
536 /* The exception only applies to non-leaf HTMLs (but -p
537 always implies non-leaf because we can overstep the
538 maximum depth to get the requisites): */
540 opt.reclevel == INFINITE_RECURSION
542 || depth < opt.reclevel - 1
543 /* -p, which implies non-leaf (see above) */
544 || opt.page_requisites)))
546 if (!acceptable (u->file))
548 DEBUGP (("%s (%s) does not match acc/rej rules.\n",
555 if (schemes_are_similar_p (u->scheme, parent->scheme))
556 if (!opt.spanhost && 0 != strcasecmp (parent->host, u->host))
558 DEBUGP (("This is not the same hostname as the parent's (%s and %s).\n",
559 u->host, parent->host));
564 if (opt.use_robots && u_scheme_like_http)
566 struct robot_specs *specs = res_get_specs (u->host, u->port);
570 if (res_retrieve_file (url, &rfile))
572 specs = res_parse_from_file (rfile);
574 /* Delete the robots.txt file if we chose to either delete the
575 files after downloading or we're just running a spider. */
576 if (opt.delete_after || opt.spider)
578 logprintf (LOG_VERBOSE, "Removing %s.\n", rfile);
580 logprintf (LOG_NOTQUIET, "unlink: %s\n",
588 /* If we cannot get real specs, at least produce
589 dummy ones so that we can register them and stop
590 trying to retrieve them. */
591 specs = res_parse ("", 0);
593 res_register_specs (u->host, u->port, specs);
596 /* Now that we have (or don't have) robots.txt specs, we can
597 check what they say. */
598 if (!res_match_path (specs, u->path))
600 DEBUGP (("Not following %s because robots.txt forbids it.\n", url));
601 string_set_add (blacklist, url);
606 /* The URL has passed all the tests. It can be placed in the
608 DEBUGP (("Decided to load it.\n"));
613 DEBUGP (("Decided NOT to load it.\n"));
618 /* This function determines whether we will consider downloading the
619 children of a URL whose download resulted in a redirection,
620 possibly to another host, etc. It is needed very rarely, and thus
621 it is merely a simple-minded wrapper around download_child_p. */
624 descend_redirect_p (const char *redirected, const char *original, int depth,
625 struct url *start_url_parsed, struct hash_table *blacklist)
627 struct url *orig_parsed, *new_parsed;
631 orig_parsed = url_parse (original, NULL);
632 assert (orig_parsed != NULL);
634 new_parsed = url_parse (redirected, NULL);
635 assert (new_parsed != NULL);
637 upos = xnew0 (struct urlpos);
638 upos->url = new_parsed;
640 success = download_child_p (upos, orig_parsed, depth,
641 start_url_parsed, blacklist);
643 url_free (orig_parsed);
644 url_free (new_parsed);
648 DEBUGP (("Redirection \"%s\" failed the test.\n", redirected));
653 /* vim:set sts=2 sw=2 cino+={s: */