2 Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or (at
9 your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
36 #endif /* HAVE_UNISTD_H */
54 /* Total size of downloaded files. Used to enforce quota. */
55 SUM_SIZE_INT total_downloaded_bytes;
57 /* If non-NULL, the stream to which output should be written. This
58 stream is initialized when `-O' is used. */
61 /* Whether output_document is a regular file we can manipulate,
62 i.e. not `-' or a device file. */
63 bool output_stream_regular;
72 limit_bandwidth_reset (void)
74 limit_data.chunk_bytes = 0;
75 limit_data.chunk_start = 0;
76 limit_data.sleep_adjust = 0;
79 /* Limit the bandwidth by pausing the download for an amount of time.
80 BYTES is the number of bytes received from the network, and TIMER
81 is the timer that started at the beginning of download. */
84 limit_bandwidth (wgint bytes, struct ptimer *timer)
86 double delta_t = ptimer_read (timer) - limit_data.chunk_start;
89 limit_data.chunk_bytes += bytes;
91 /* Calculate the amount of time we expect downloading the chunk
92 should take. If in reality it took less time, sleep to
93 compensate for the difference. */
94 expected = 1000.0 * limit_data.chunk_bytes / opt.limit_rate;
96 if (expected > delta_t)
98 double slp = expected - delta_t + limit_data.sleep_adjust;
102 DEBUGP (("deferring a %.2f ms sleep (%s/%.2f).\n",
103 slp, number_to_static_string (limit_data.chunk_bytes),
107 DEBUGP (("\nsleeping %.2f ms for %s bytes, adjust %.2f ms\n",
108 slp, number_to_static_string (limit_data.chunk_bytes),
109 limit_data.sleep_adjust));
111 t0 = ptimer_read (timer);
113 t1 = ptimer_measure (timer);
115 /* Due to scheduling, we probably slept slightly longer (or
116 shorter) than desired. Calculate the difference between the
117 desired and the actual sleep, and adjust the next sleep by
119 limit_data.sleep_adjust = slp - (t1 - t0);
120 /* If sleep_adjust is very large, it's likely due to suspension
121 and not clock inaccuracy. Don't enforce those. */
122 if (limit_data.sleep_adjust > 500)
123 limit_data.sleep_adjust = 500;
124 else if (limit_data.sleep_adjust < -500)
125 limit_data.sleep_adjust = -500;
128 limit_data.chunk_bytes = 0;
129 limit_data.chunk_start = ptimer_read (timer);
133 # define MIN(i, j) ((i) <= (j) ? (i) : (j))
136 /* Write data in BUF to OUT. However, if *SKIP is non-zero, skip that
137 amount of data and decrease SKIP. Increment *TOTAL by the amount
141 write_data (FILE *out, const char *buf, int bufsize, wgint *skip,
160 fwrite (buf, 1, bufsize, out);
163 /* Immediately flush the downloaded data. This should not hinder
164 performance: fast downloads will arrive in large 16K chunks
165 (which stdio would write out immediately anyway), and slow
166 downloads wouldn't be limited by disk speed. */
168 return !ferror (out);
171 /* Read the contents of file descriptor FD until it the connection
172 terminates or a read error occurs. The data is read in portions of
173 up to 16K and written to OUT as it arrives. If opt.verbose is set,
174 the progress is shown.
176 TOREAD is the amount of data expected to arrive, normally only used
177 by the progress gauge.
179 STARTPOS is the position from which the download starts, used by
180 the progress gauge. If QTYREAD is non-NULL, the value it points to
181 is incremented by the amount of data read from the network. If
182 QTYWRITTEN is non-NULL, the value it points to is incremented by
183 the amount of data written to disk. The time it took to download
184 the data (in milliseconds) is stored to ELAPSED.
186 The function exits and returns the amount of data read. In case of
187 error while reading data, -1 is returned. In case of error while
188 writing data, -2 is returned. */
191 fd_read_body (int fd, FILE *out, wgint toread, wgint startpos,
192 wgint *qtyread, wgint *qtywritten, double *elapsed, int flags)
196 static char dlbuf[16384];
197 int dlbufsize = sizeof (dlbuf);
199 struct ptimer *timer = NULL;
200 double last_successful_read_tm = 0;
202 /* The progress gauge, set according to the user preferences. */
203 void *progress = NULL;
205 /* Non-zero if the progress gauge is interactive, i.e. if it can
206 continually update the display. When true, smaller timeout
207 values are used so that the gauge can update the display when
208 data arrives slowly. */
209 bool progress_interactive = false;
211 bool exact = !!(flags & rb_read_exactly);
214 /* How much data we've read/written. */
216 wgint sum_written = 0;
218 if (flags & rb_skip_startpos)
223 /* If we're skipping STARTPOS bytes, pass 0 as the INITIAL
224 argument to progress_create because the indicator doesn't
225 (yet) know about "skipping" data. */
226 progress = progress_create (skip ? 0 : startpos, startpos + toread);
227 progress_interactive = progress_interactive_p (progress);
231 limit_bandwidth_reset ();
233 /* A timer is needed for tracking progress, for throttling, and for
234 tracking elapsed time. If either of these are requested, start
236 if (progress || opt.limit_rate || elapsed)
238 timer = ptimer_new ();
239 last_successful_read_tm = 0;
242 /* Use a smaller buffer for low requested bandwidths. For example,
243 with --limit-rate=2k, it doesn't make sense to slurp in 16K of
244 data and then sleep for 8s. With buffer size equal to the limit,
245 we never have to sleep for more than one second. */
246 if (opt.limit_rate && opt.limit_rate < dlbufsize)
247 dlbufsize = opt.limit_rate;
249 /* Read from FD while there is data to read. Normally toread==0
250 means that it is unknown how much data is to arrive. However, if
251 EXACT is set, then toread==0 means what it says: that no data
253 while (!exact || (sum_read < toread))
255 int rdsize = exact ? MIN (toread - sum_read, dlbufsize) : dlbufsize;
256 double tmout = opt.read_timeout;
257 if (progress_interactive)
259 /* For interactive progress gauges, always specify a ~1s
260 timeout, so that the gauge can be updated regularly even
261 when the data arrives very slowly or stalls. */
263 if (opt.read_timeout)
266 waittm = (ptimer_read (timer) - last_successful_read_tm) / 1000;
267 if (waittm + tmout > opt.read_timeout)
269 /* Don't let total idle time exceed read timeout. */
270 tmout = opt.read_timeout - waittm;
273 /* We've already exceeded the timeout. */
274 ret = -1, errno = ETIMEDOUT;
280 ret = fd_read (fd, dlbuf, rdsize, tmout);
282 if (progress_interactive && ret < 0 && errno == ETIMEDOUT)
283 ret = 0; /* interactive timeout, handled above */
285 break; /* EOF or read error */
287 if (progress || opt.limit_rate)
289 ptimer_measure (timer);
291 last_successful_read_tm = ptimer_read (timer);
297 if (!write_data (out, dlbuf, ret, &skip, &sum_written))
305 limit_bandwidth (ret, timer);
308 progress_update (progress, ret, ptimer_read (timer));
310 if (toread > 0 && !opt.quiet)
311 ws_percenttitle (100.0 *
312 (startpos + sum_read) / (startpos + toread));
320 progress_finish (progress, ptimer_read (timer));
323 *elapsed = ptimer_read (timer);
325 ptimer_destroy (timer);
328 *qtyread += sum_read;
330 *qtywritten += sum_written;
335 /* Read a hunk of data from FD, up until a terminator. The terminator
336 is whatever the TERMINATOR function determines it to be; for
337 example, it can be a line of data, or the head of an HTTP response.
338 The function returns the data read allocated with malloc.
340 In case of error, NULL is returned. In case of EOF and no data
341 read, NULL is returned and errno set to 0. In case of EOF with
342 data having been read, the data is returned, but it will
343 (obviously) not contain the terminator.
345 The idea is to be able to read a line of input, or otherwise a hunk
346 of text, such as the head of an HTTP request, without crossing the
347 boundary, so that the next call to fd_read etc. reads the data
348 after the hunk. To achieve that, this function does the following:
350 1. Peek at available data.
352 2. Determine whether the peeked data, along with the previously
353 read data, includes the terminator.
355 2a. If yes, read the data until the end of the terminator, and
358 2b. If no, read the peeked data and goto 1.
360 The function is careful to assume as little as possible about the
361 implementation of peeking. For example, every peek is followed by
362 a read. If the read returns a different amount of data, the
363 process is retried until all data arrives safely.
365 SIZEHINT is the buffer size sufficient to hold all the data in the
366 typical case (it is used as the initial buffer size). MAXSIZE is
367 the maximum amount of memory this function is allowed to allocate,
368 or 0 if no upper limit is to be enforced.
370 This function should be used as a building block for other
371 functions -- see fd_read_line as a simple example. */
374 fd_read_hunk (int fd, hunk_terminator_t terminator, long sizehint, long maxsize)
376 long bufsize = sizehint;
377 char *hunk = xmalloc (bufsize);
378 int tail = 0; /* tail position in HUNK */
380 assert (maxsize >= bufsize);
385 int pklen, rdlen, remain;
387 /* First, peek at the available data. */
389 pklen = fd_peek (fd, hunk + tail, bufsize - 1 - tail, -1);
395 end = terminator (hunk, tail, pklen);
398 /* The data contains the terminator: we'll drain the data up
399 to the end of the terminator. */
400 remain = end - (hunk + tail);
403 /* No more data needs to be read. */
407 if (bufsize - 1 < tail + remain)
409 bufsize = tail + remain + 1;
410 hunk = xrealloc (hunk, bufsize);
414 /* No terminator: simply read the data we know is (or should
418 /* Now, read the data. Note that we make no assumptions about
419 how much data we'll get. (Some TCP stacks are notorious for
420 read returning less data than the previous MSG_PEEK.) */
422 rdlen = fd_read (fd, hunk + tail, remain, 0);
435 /* EOF without anything having been read */
441 /* EOF seen: return the data we've read. */
444 if (end && rdlen == remain)
445 /* The terminator was seen and the remaining data drained --
446 we got what we came for. */
449 /* Keep looping until all the data arrives. */
451 if (tail == bufsize - 1)
453 /* Double the buffer size, but refuse to allocate more than
455 if (maxsize && bufsize >= maxsize)
462 if (maxsize && bufsize > maxsize)
464 hunk = xrealloc (hunk, bufsize);
470 line_terminator (const char *hunk, int oldlen, int peeklen)
472 const char *p = memchr (hunk + oldlen, '\n', peeklen);
474 /* p+1 because we want the line to include '\n' */
479 /* The maximum size of the single line we agree to accept. This is
480 not meant to impose an arbitrary limit, but to protect the user
481 from Wget slurping up available memory upon encountering malicious
482 or buggy server output. Define it to 0 to remove the limit. */
483 #define FD_READ_LINE_MAX 4096
485 /* Read one line from FD and return it. The line is allocated using
486 malloc, but is never larger than FD_READ_LINE_MAX.
488 If an error occurs, or if no data can be read, NULL is returned.
489 In the former case errno indicates the error condition, and in the
490 latter case, errno is NULL. */
493 fd_read_line (int fd)
495 return fd_read_hunk (fd, line_terminator, 128, FD_READ_LINE_MAX);
498 /* Return a printed representation of the download rate, as
499 appropriate for the speed. If PAD is true, strings will be padded
500 to the width of 7 characters (xxxx.xx). */
502 retr_rate (wgint bytes, double msecs, bool pad)
505 static const char *rate_names[] = {"B/s", "KB/s", "MB/s", "GB/s" };
508 double dlrate = calc_rate (bytes, msecs, &units);
509 sprintf (res, pad ? "%7.2f %s" : "%.2f %s", dlrate, rate_names[units]);
514 /* Calculate the download rate and trim it as appropriate for the
515 speed. Appropriate means that if rate is greater than 1K/s,
516 kilobytes are used, and if rate is greater than 1MB/s, megabytes
519 UNITS is zero for B/s, one for KB/s, two for MB/s, and three for
522 calc_rate (wgint bytes, double msecs, int *units)
530 /* If elapsed time is exactly zero, it means we're under the
531 resolution of the timer. This can easily happen on systems
532 that use time() for the timer. Since the interval lies between
533 0 and the timer's resolution, assume half the resolution. */
534 msecs = ptimer_resolution () / 2.0;
536 dlrate = 1000.0 * bytes / msecs;
539 else if (dlrate < 1024.0 * 1024.0)
540 *units = 1, dlrate /= 1024.0;
541 else if (dlrate < 1024.0 * 1024.0 * 1024.0)
542 *units = 2, dlrate /= (1024.0 * 1024.0);
544 /* Maybe someone will need this, one day. */
545 *units = 3, dlrate /= (1024.0 * 1024.0 * 1024.0);
550 /* Maximum number of allowed redirections. 20 was chosen as a
551 "reasonable" value, which is low enough to not cause havoc, yet
552 high enough to guarantee that normal retrievals will not be hurt by
555 #define MAX_REDIRECTIONS 20
557 #define SUSPEND_POST_DATA do { \
558 post_data_suspended = true; \
559 saved_post_data = opt.post_data; \
560 saved_post_file_name = opt.post_file_name; \
561 opt.post_data = NULL; \
562 opt.post_file_name = NULL; \
565 #define RESTORE_POST_DATA do { \
566 if (post_data_suspended) \
568 opt.post_data = saved_post_data; \
569 opt.post_file_name = saved_post_file_name; \
570 post_data_suspended = false; \
574 static char *getproxy (struct url *);
576 /* Retrieve the given URL. Decides which loop to call -- HTTP, FTP,
579 /* #### This function should be rewritten so it doesn't return from
583 retrieve_url (const char *origurl, char **file, char **newloc,
584 const char *refurl, int *dt)
588 bool location_changed;
590 char *mynewloc, *proxy;
591 struct url *u, *proxy_url;
592 int up_error_code; /* url parse error code */
594 int redirection_count = 0;
596 bool post_data_suspended = false;
597 char *saved_post_data = NULL;
598 char *saved_post_file_name = NULL;
600 /* If dt is NULL, use local storage. */
606 url = xstrdup (origurl);
612 u = url_parse (url, &up_error_code);
615 logprintf (LOG_NOTQUIET, "%s: %s.\n", url, url_error (up_error_code));
621 refurl = opt.referer;
630 proxy = getproxy (u);
633 /* Parse the proxy URL. */
634 proxy_url = url_parse (proxy, &up_error_code);
637 logprintf (LOG_NOTQUIET, _("Error parsing proxy URL %s: %s.\n"),
638 proxy, url_error (up_error_code));
643 if (proxy_url->scheme != SCHEME_HTTP && proxy_url->scheme != u->scheme)
645 logprintf (LOG_NOTQUIET, _("Error in proxy URL %s: Must be HTTP.\n"), proxy);
646 url_free (proxy_url);
653 if (u->scheme == SCHEME_HTTP
655 || u->scheme == SCHEME_HTTPS
657 || (proxy_url && proxy_url->scheme == SCHEME_HTTP))
659 result = http_loop (u, &mynewloc, &local_file, refurl, dt, proxy_url);
661 else if (u->scheme == SCHEME_FTP)
663 /* If this is a redirection, temporarily turn off opt.ftp_glob
664 and opt.recursive, both being undesirable when following
666 bool oldrec = opt.recursive, oldglob = opt.ftp_glob;
667 if (redirection_count)
668 opt.recursive = opt.ftp_glob = false;
670 result = ftp_loop (u, dt, proxy_url);
671 opt.recursive = oldrec;
672 opt.ftp_glob = oldglob;
674 /* There is a possibility of having HTTP being redirected to
675 FTP. In these cases we must decide whether the text is HTML
676 according to the suffix. The HTML suffixes are `.html',
677 `.htm' and a few others, case-insensitive. */
678 if (redirection_count && local_file && u->scheme == SCHEME_FTP)
680 if (has_html_suffix_p (local_file))
687 url_free (proxy_url);
691 location_changed = (result == NEWLOCATION);
692 if (location_changed)
694 char *construced_newloc;
695 struct url *newloc_parsed;
697 assert (mynewloc != NULL);
702 /* The HTTP specs only allow absolute URLs to appear in
703 redirects, but a ton of boneheaded webservers and CGIs out
704 there break the rules and use relative URLs, and popular
705 browsers are lenient about this, so wget should be too. */
706 construced_newloc = uri_merge (url, mynewloc);
708 mynewloc = construced_newloc;
710 /* Now, see if this new location makes sense. */
711 newloc_parsed = url_parse (mynewloc, &up_error_code);
714 logprintf (LOG_NOTQUIET, "%s: %s.\n", escnonprint_uri (mynewloc),
715 url_error (up_error_code));
723 /* Now mynewloc will become newloc_parsed->url, because if the
724 Location contained relative paths like .././something, we
725 don't want that propagating as url. */
727 mynewloc = xstrdup (newloc_parsed->url);
729 /* Check for max. number of redirections. */
730 if (++redirection_count > MAX_REDIRECTIONS)
732 logprintf (LOG_NOTQUIET, _("%d redirections exceeded.\n"),
734 url_free (newloc_parsed);
747 /* If we're being redirected from POST, we don't want to POST
748 again. Many requests answer POST with a redirection to an
749 index page; that redirection is clearly a GET. We "suspend"
750 POST data for the duration of the redirections, and restore
751 it when we're done. */
752 if (!post_data_suspended)
762 register_download (u->url, local_file);
763 if (redirection_count && 0 != strcmp (origurl, u->url))
764 register_redirection (origurl, u->url);
766 register_html (u->url, local_file);
771 *file = local_file ? local_file : NULL;
773 xfree_null (local_file);
777 if (redirection_count)
796 /* Find the URLs in the file and call retrieve_url() for each of them.
797 If HTML is true, treat the file as HTML, and construct the URLs
800 If opt.recursive is set, call retrieve_tree() for each file. */
803 retrieve_from_file (const char *file, bool html, int *count)
806 struct urlpos *url_list, *cur_url;
808 url_list = (html ? get_urls_html (file, NULL, NULL)
809 : get_urls_file (file));
810 status = RETROK; /* Suppose everything is OK. */
811 *count = 0; /* Reset the URL count. */
813 for (cur_url = url_list; cur_url; cur_url = cur_url->next, ++*count)
815 char *filename = NULL, *new_file = NULL;
818 if (cur_url->ignore_when_downloading)
821 if (opt.quota && total_downloaded_bytes > opt.quota)
826 if ((opt.recursive || opt.page_requisites)
827 && cur_url->url->scheme != SCHEME_FTP)
828 status = retrieve_tree (cur_url->url->url);
830 status = retrieve_url (cur_url->url->url, &filename, &new_file, NULL, &dt);
832 if (filename && opt.delete_after && file_exists_p (filename))
835 Removing file due to --delete-after in retrieve_from_file():\n"));
836 logprintf (LOG_VERBOSE, _("Removing %s.\n"), filename);
837 if (unlink (filename))
838 logprintf (LOG_NOTQUIET, "unlink: %s\n", strerror (errno));
842 xfree_null (new_file);
843 xfree_null (filename);
846 /* Free the linked list of URL-s. */
847 free_urlpos (url_list);
852 /* Print `giving up', or `retrying', depending on the impending
853 action. N1 and N2 are the attempt number and the attempt limit. */
855 printwhat (int n1, int n2)
857 logputs (LOG_VERBOSE, (n1 == n2) ? _("Giving up.\n\n") : _("Retrying.\n\n"));
860 /* If opt.wait or opt.waitretry are specified, and if certain
861 conditions are met, sleep the appropriate number of seconds. See
862 the documentation of --wait and --waitretry for more information.
864 COUNT is the count of current retrieval, beginning with 1. */
867 sleep_between_retrievals (int count)
869 static bool first_retrieval = true;
873 /* Don't sleep before the very first retrieval. */
874 first_retrieval = false;
878 if (opt.waitretry && count > 1)
880 /* If opt.waitretry is specified and this is a retry, wait for
881 COUNT-1 number of seconds, or for opt.waitretry seconds. */
882 if (count <= opt.waitretry)
885 xsleep (opt.waitretry);
889 if (!opt.random_wait || count > 1)
890 /* If random-wait is not specified, or if we are sleeping
891 between retries of the same download, sleep the fixed
896 /* Sleep a random amount of time averaging in opt.wait
897 seconds. The sleeping amount ranges from 0 to
898 opt.wait*2, inclusive. */
899 double waitsecs = 2 * opt.wait * random_float ();
900 DEBUGP (("sleep_between_retrievals: avg=%f,sleep=%f\n",
901 opt.wait, waitsecs));
907 /* Free the linked list of urlpos. */
909 free_urlpos (struct urlpos *l)
913 struct urlpos *next = l->next;
916 xfree_null (l->local_name);
922 /* Rotate FNAME opt.backups times */
924 rotate_backups(const char *fname)
926 int maxlen = strlen (fname) + 1 + numdigit (opt.backups) + 1;
927 char *from = (char *)alloca (maxlen);
928 char *to = (char *)alloca (maxlen);
932 if (stat (fname, &sb) == 0)
933 if (S_ISREG (sb.st_mode) == 0)
936 for (i = opt.backups; i > 1; i--)
938 sprintf (from, "%s.%d", fname, i - 1);
939 sprintf (to, "%s.%d", fname, i);
943 sprintf (to, "%s.%d", fname, 1);
947 static bool no_proxy_match (const char *, const char **);
949 /* Return the URL of the proxy appropriate for url U. */
952 getproxy (struct url *u)
956 static char rewritten_storage[1024];
960 if (!no_proxy_match (u->host, (const char **)opt.no_proxy))
966 proxy = opt.http_proxy ? opt.http_proxy : getenv ("http_proxy");
970 proxy = opt.https_proxy ? opt.https_proxy : getenv ("https_proxy");
974 proxy = opt.ftp_proxy ? opt.ftp_proxy : getenv ("ftp_proxy");
979 if (!proxy || !*proxy)
982 /* Handle shorthands. `rewritten_storage' is a kludge to allow
983 getproxy() to return static storage. */
984 rewritten_url = rewrite_shorthand_url (proxy);
987 strncpy (rewritten_storage, rewritten_url, sizeof (rewritten_storage));
988 rewritten_storage[sizeof (rewritten_storage) - 1] = '\0';
989 proxy = rewritten_storage;
995 /* Should a host be accessed through proxy, concerning no_proxy? */
997 no_proxy_match (const char *host, const char **no_proxy)
1002 return !sufmatch (no_proxy, host);