[svn] Try to reuse connections that return error codes.

[wget] / src / retr.c

/* File retrieval.
   Copyright (C) 1995, 1996, 1997, 1998, 2000, 2001 Free Software Foundation, Inc.

This file is part of GNU Wget.

GNU Wget is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or (at
your option) any later version.

GNU Wget is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with Wget; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

In addition, as a special exception, the Free Software Foundation
gives permission to link the code of its release of Wget with the
OpenSSL project's "OpenSSL" library (or with modified versions of it
that use the same license as the "OpenSSL" library), and distribute
the linked executables.  You must obey the GNU General Public License
in all respects for all of the code used other than "OpenSSL".  If you
modify this file, you may extend this exception to your version of the
file, but you are not obligated to do so.  If you do not wish to do
so, delete this exception statement from your version.  */

#include <config.h>

#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#ifdef HAVE_UNISTD_H
# include <unistd.h>
#endif /* HAVE_UNISTD_H */
#include <errno.h>
#ifdef HAVE_STRING_H
# include <string.h>
#else
# include <strings.h>
#endif /* HAVE_STRING_H */
#include <assert.h>

#include "wget.h"
#include "utils.h"
#include "retr.h"
#include "progress.h"
#include "url.h"
#include "recur.h"
#include "ftp.h"
#include "host.h"
#include "connect.h"
#include "hash.h"
#include "convert.h"

#ifdef HAVE_SSL
# include "gen_sslfunc.h"       /* for ssl_iread */
#endif

#ifndef errno
extern int errno;
#endif

/* See the comment in gethttp() why this is needed. */
int global_download_count;

/* Total size of downloaded files.  Used to enforce quota.  */
LARGE_INT total_downloaded_bytes;

\f
static struct {
  long chunk_bytes;
  double chunk_start;
  double sleep_adjust;
} limit_data;

static void
limit_bandwidth_reset (void)
{
  limit_data.chunk_bytes = 0;
  limit_data.chunk_start = 0;
}

/* Limit the bandwidth by pausing the download for an amount of time.
   BYTES is the number of bytes received from the network, and TIMER
   is the timer that started at the beginning of download.  */

static void
limit_bandwidth (long bytes, struct wget_timer *timer)
{
  double delta_t = wtimer_read (timer) - limit_data.chunk_start;
  double expected;

  limit_data.chunk_bytes += bytes;

  /* Calculate the amount of time we expect downloading the chunk
     should take.  If in reality it took less time, sleep to
     compensate for the difference.  */
  expected = 1000.0 * limit_data.chunk_bytes / opt.limit_rate;

  if (expected > delta_t)
    {
      double slp = expected - delta_t + limit_data.sleep_adjust;
      double t0, t1;
      if (slp < 200)
        {
          DEBUGP (("deferring a %.2f ms sleep (%ld/%.2f).\n",
                   slp, limit_data.chunk_bytes, delta_t));
          return;
        }
      DEBUGP (("\nsleeping %.2f ms for %ld bytes, adjust %.2f ms\n",
               slp, limit_data.chunk_bytes, limit_data.sleep_adjust));

      t0 = wtimer_read (timer);
      xsleep (slp / 1000);
      wtimer_update (timer);
      t1 = wtimer_read (timer);

      /* Due to scheduling, we probably slept slightly longer (or
         shorter) than desired.  Calculate the difference between the
         desired and the actual sleep, and adjust the next sleep by
         that amount.  */
      limit_data.sleep_adjust = slp - (t1 - t0);
    }

  limit_data.chunk_bytes = 0;
  limit_data.chunk_start = wtimer_read (timer);
}

#ifndef MIN
# define MIN(i, j) ((i) <= (j) ? (i) : (j))
#endif

/* Reads the contents of file descriptor FD, until it is closed, or a
   read error occurs.  The data is read in 8K chunks, and stored to
   stream fp, which should have been open for writing.

   The EXPECTED argument is passed to show_progress() unchanged, but
   otherwise ignored.

   If opt.verbose is set, the progress is also shown.  RESTVAL
   (RESTart VALue) is the position from which the download starts,
   needed for progress display.

   The function exits and returns codes of 0, -1 and -2 if the
   connection was closed, there was a read error, or if it could not
   write to the output stream, respectively.  */

int
fd_read_body (int fd, FILE *out, long *len, long restval, long expected,
              int use_expected, double *elapsed)
{
  int res = 0;

  static char dlbuf[16384];
  int dlbufsize = sizeof (dlbuf);

  struct wget_timer *timer = wtimer_allocate ();
  double last_successful_read_tm;

  /* The progress gauge, set according to the user preferences. */
  void *progress = NULL;

  /* Non-zero if the progress gauge is interactive, i.e. if it can
     continually update the display.  When true, smaller timeout
     values are used so that the gauge can update the display when
     data arrives slowly. */
  int progress_interactive = 0;

  *len = restval;

  if (opt.verbose)
    {
      progress = progress_create (restval, expected);
      progress_interactive = progress_interactive_p (progress);
    }

  if (opt.limit_rate)
    limit_bandwidth_reset ();
  wtimer_reset (timer);
  last_successful_read_tm = 0;

  /* Use a smaller buffer for low requested bandwidths.  For example,
     with --limit-rate=2k, it doesn't make sense to slurp in 16K of
     data and then sleep for 8s.  With buffer size equal to the limit,
     we never have to sleep for more than one second.  */
  if (opt.limit_rate && opt.limit_rate < dlbufsize)
    dlbufsize = opt.limit_rate;

  /* Read from FD while there is available data.

     Normally, if expected is 0, it means that it is not known how
     much data is expected.  However, if use_expected is specified,
     then expected being zero means exactly that.  */
  while (!use_expected || (*len < expected))
    {
      int amount_to_read = (use_expected
                            ? MIN (expected - *len, dlbufsize) : dlbufsize);
      double tmout = opt.read_timeout;
      if (progress_interactive)
        {
          double waittm;
          /* For interactive progress gauges, always specify a ~1s
             timeout, so that the gauge can be updated regularly even
             when the data arrives very slowly or stalls.  */
          tmout = 0.95;
          waittm = (wtimer_read (timer) - last_successful_read_tm) / 1000;
          if (waittm + tmout > opt.read_timeout)
            {
              /* Don't allow waiting time to exceed read timeout. */
              tmout = opt.read_timeout - waittm;
              if (tmout < 0)
                {
                  /* We've already exceeded the timeout. */
                  res = -1, errno = ETIMEDOUT;
                  break;
                }
            }
        }
      res = fd_read (fd, dlbuf, amount_to_read, tmout);

      if (res == 0 || (res < 0 && errno != ETIMEDOUT))
        break;
      else if (res < 0)
        res = 0;                /* timeout */

      wtimer_update (timer);
      if (res > 0)
        {
          if (out)
            {
              fwrite (dlbuf, 1, res, out);
              /* Always flush the contents of the network packet.
                 This should not hinder performance: fast downloads
                 will be received in 16K chunks (which stdio would
                 write out anyway), and slow downloads won't be
                 limited by disk performance.  */
              fflush (out);
              if (ferror (out))
                {
                  res = -2;
                  goto out;
                }
            }
          last_successful_read_tm = wtimer_read (timer);
        }

      if (opt.limit_rate)
        limit_bandwidth (res, timer);

      *len += res;
      if (progress)
        progress_update (progress, res, wtimer_read (timer));
#ifdef WINDOWS
      if (use_expected && expected > 0)
        ws_percenttitle (100.0 * (double)(*len) / (double)expected);
#endif
    }
  if (res < -1)
    res = -1;

 out:
  if (progress)
    progress_finish (progress, wtimer_read (timer));
  if (elapsed)
    *elapsed = wtimer_read (timer);
  wtimer_delete (timer);

  return res;
}
\f
/* Read a hunk of data from FD, up until a terminator.  The terminator
   is whatever the TERMINATOR function determines it to be; for
   example, it can be a line of data, or the head of an HTTP response.
   The function returns the data read allocated with malloc.

   In case of error, NULL is returned.  In case of EOF and no data
   read, NULL is returned and errno set to 0.  In case of EOF with
   data having been read, the data is returned, but it will
   (obviously) not contain the terminator.

   The idea is to be able to read a line of input, or otherwise a hunk
   of text, such as the head of an HTTP request, without crossing the
   boundary, so that the next call to fd_read etc. reads the data
   after the hunk.  To achieve that, this function does the following:

   1. Peek at available data.

   2. Determine whether the peeked data, along with the previously
      read data, includes the terminator.

      2a. If yes, read the data until the end of the terminator, and
          exit.

      2b. If no, read the peeked data and goto 1.

   The function is careful to assume as little as possible about the
   implementation of peeking.  For example, every peek is followed by
   a read.  If the read returns a different amount of data, the
   process is retried until all data arrives safely.

   BUFSIZE is the size of the initial buffer expected to read all the
   data in the typical case.

   This function should be used as a building block for other
   functions -- see fd_read_line as a simple example.  */

char *
fd_read_hunk (int fd, hunk_terminator_t hunk_terminator, int bufsize)
{
  char *hunk = xmalloc (bufsize);
  int tail = 0;                 /* tail position in HUNK */

  while (1)
    {
      const char *end;
      int pklen, rdlen, remain;

      /* First, peek at the available data. */

      pklen = fd_peek (fd, hunk + tail, bufsize - 1 - tail, -1);
      if (pklen < 0)
        {
          xfree (hunk);
          return NULL;
        }
      end = hunk_terminator (hunk, tail, pklen);
      if (end)
        {
          /* The data contains the terminator: we'll drain the data up
             to the end of the terminator.  */
          remain = end - (hunk + tail);
          if (remain == 0)
            {
              /* No more data needs to be read. */
              hunk[tail] = '\0';
              return hunk;
            }
          if (bufsize - 1 < tail + remain)
            {
              bufsize = tail + remain + 1;
              hunk = xrealloc (hunk, bufsize);
            }
        }
      else
        /* No terminator: simply read the data we know is (or should
           be) available.  */
        remain = pklen;

      /* Now, read the data.  Note that we make no assumptions about
         how much data we'll get.  (Some TCP stacks are notorious for
         read returning less data than the previous MSG_PEEK.)  */

      rdlen = fd_read (fd, hunk + tail, remain, 0);
      if (rdlen < 0)
        {
          xfree_null (hunk);
          return NULL;
        }
      tail += rdlen;
      hunk[tail] = '\0';

      if (rdlen == 0)
        {
          if (tail == 0)
            {
              /* EOF without anything having been read */
              xfree (hunk);
              errno = 0;
              return NULL;
            }
          else
            /* EOF seen: return the data we've read. */
            return hunk;
        }
      if (end && rdlen == remain)
        /* The terminator was seen and the remaining data drained --
           we got what we came for.  */
        return hunk;

      /* Keep looping until all the data arrives. */

      if (tail == bufsize - 1)
        {
          bufsize <<= 1;
          hunk = xrealloc (hunk, bufsize);
        }
    }
}

static const char *
line_terminator (const char *hunk, int oldlen, int peeklen)
{
  const char *p = memchr (hunk + oldlen, '\n', peeklen);
  if (p)
    /* p+1 because we want the line to include '\n' */
    return p + 1;
  return NULL;
}

/* Read one line from FD and return it.  The line is allocated using
   malloc.

   If an error occurs, or if no data can be read, NULL is returned.
   In the former case errno indicates the error condition, and in the
   latter case, errno is NULL.  */

char *
fd_read_line (int fd)
{
  return fd_read_hunk (fd, line_terminator, 128);
}
\f
/* Return a printed representation of the download rate, as
   appropriate for the speed.  If PAD is non-zero, strings will be
   padded to the width of 7 characters (xxxx.xx).  */
char *
retr_rate (long bytes, double msecs, int pad)
{
  static char res[20];
  static char *rate_names[] = {"B/s", "KB/s", "MB/s", "GB/s" };
  int units = 0;

  double dlrate = calc_rate (bytes, msecs, &units);
  sprintf (res, pad ? "%7.2f %s" : "%.2f %s", dlrate, rate_names[units]);

  return res;
}

/* Calculate the download rate and trim it as appropriate for the
   speed.  Appropriate means that if rate is greater than 1K/s,
   kilobytes are used, and if rate is greater than 1MB/s, megabytes
   are used.

   UNITS is zero for B/s, one for KB/s, two for MB/s, and three for
   GB/s.  */
double
calc_rate (long bytes, double msecs, int *units)
{
  double dlrate;

  assert (msecs >= 0);
  assert (bytes >= 0);

  if (msecs == 0)
    /* If elapsed time is exactly zero, it means we're under the
       granularity of the timer.  This often happens on systems that
       use time() for the timer.  */
    msecs = wtimer_granularity ();

  dlrate = (double)1000 * bytes / msecs;
  if (dlrate < 1024.0)
    *units = 0;
  else if (dlrate < 1024.0 * 1024.0)
    *units = 1, dlrate /= 1024.0;
  else if (dlrate < 1024.0 * 1024.0 * 1024.0)
    *units = 2, dlrate /= (1024.0 * 1024.0);
  else
    /* Maybe someone will need this, one day. */
    *units = 3, dlrate /= (1024.0 * 1024.0 * 1024.0);

  return dlrate;
}
\f
/* Maximum number of allowed redirections.  20 was chosen as a
   "reasonable" value, which is low enough to not cause havoc, yet
   high enough to guarantee that normal retrievals will not be hurt by
   the check.  */

#define MAX_REDIRECTIONS 20

#define SUSPEND_POST_DATA do {                  \
  post_data_suspended = 1;                      \
  saved_post_data = opt.post_data;              \
  saved_post_file_name = opt.post_file_name;    \
  opt.post_data = NULL;                         \
  opt.post_file_name = NULL;                    \
} while (0)

#define RESTORE_POST_DATA do {                          \
  if (post_data_suspended)                              \
    {                                                   \
      opt.post_data = saved_post_data;                  \
      opt.post_file_name = saved_post_file_name;        \
      post_data_suspended = 0;                          \
    }                                                   \
} while (0)

static char *getproxy PARAMS ((struct url *));

/* Retrieve the given URL.  Decides which loop to call -- HTTP, FTP,
   FTP, proxy, etc.  */

/* #### This function should be rewritten so it doesn't return from
   multiple points. */

uerr_t
retrieve_url (const char *origurl, char **file, char **newloc,
              const char *refurl, int *dt)
{
  uerr_t result;
  char *url;
  int location_changed, dummy;
  char *mynewloc, *proxy;
  struct url *u, *proxy_url;
  int up_error_code;            /* url parse error code */
  char *local_file;
  int redirection_count = 0;

  int post_data_suspended = 0;
  char *saved_post_data = NULL;
  char *saved_post_file_name = NULL;

  /* If dt is NULL, use local storage.  */
  if (!dt)
    {
      dt = &dummy;
      dummy = 0;
    }
  url = xstrdup (origurl);
  if (newloc)
    *newloc = NULL;
  if (file)
    *file = NULL;

  u = url_parse (url, &up_error_code);
  if (!u)
    {
      logprintf (LOG_NOTQUIET, "%s: %s.\n", url, url_error (up_error_code));
      xfree (url);
      return URLERROR;
    }

  if (!refurl)
    refurl = opt.referer;

 redirected:

  result = NOCONERROR;
  mynewloc = NULL;
  local_file = NULL;
  proxy_url = NULL;

  proxy = getproxy (u);
  if (proxy)
    {
      /* Parse the proxy URL.  */
      proxy_url = url_parse (proxy, &up_error_code);
      if (!proxy_url)
        {
          logprintf (LOG_NOTQUIET, _("Error parsing proxy URL %s: %s.\n"),
                     proxy, url_error (up_error_code));
          xfree (url);
          RESTORE_POST_DATA;
          return PROXERR;
        }
      if (proxy_url->scheme != SCHEME_HTTP && proxy_url->scheme != u->scheme)
        {
          logprintf (LOG_NOTQUIET, _("Error in proxy URL %s: Must be HTTP.\n"), proxy);
          url_free (proxy_url);
          xfree (url);
          RESTORE_POST_DATA;
          return PROXERR;
        }
    }

  if (u->scheme == SCHEME_HTTP
#ifdef HAVE_SSL
      || u->scheme == SCHEME_HTTPS
#endif
      || (proxy_url && proxy_url->scheme == SCHEME_HTTP))
    {
      result = http_loop (u, &mynewloc, &local_file, refurl, dt, proxy_url);
    }
  else if (u->scheme == SCHEME_FTP)
    {
      /* If this is a redirection, we must not allow recursive FTP
         retrieval, so we save recursion to oldrec, and restore it
         later.  */
      int oldrec = opt.recursive;
      if (redirection_count)
        opt.recursive = 0;
      result = ftp_loop (u, dt, proxy_url);
      opt.recursive = oldrec;

      /* There is a possibility of having HTTP being redirected to
         FTP.  In these cases we must decide whether the text is HTML
         according to the suffix.  The HTML suffixes are `.html',
         `.htm' and a few others, case-insensitive.  */
      if (redirection_count && local_file && u->scheme == SCHEME_FTP)
        {
          if (has_html_suffix_p (local_file))
            *dt |= TEXTHTML;
        }
    }

  if (proxy_url)
    {
      url_free (proxy_url);
      proxy_url = NULL;
    }

  location_changed = (result == NEWLOCATION);
  if (location_changed)
    {
      char *construced_newloc;
      struct url *newloc_parsed;

      assert (mynewloc != NULL);

      if (local_file)
        xfree (local_file);

      /* The HTTP specs only allow absolute URLs to appear in
         redirects, but a ton of boneheaded webservers and CGIs out
         there break the rules and use relative URLs, and popular
         browsers are lenient about this, so wget should be too. */
      construced_newloc = uri_merge (url, mynewloc);
      xfree (mynewloc);
      mynewloc = construced_newloc;

      /* Now, see if this new location makes sense. */
      newloc_parsed = url_parse (mynewloc, &up_error_code);
      if (!newloc_parsed)
        {
          logprintf (LOG_NOTQUIET, "%s: %s.\n", mynewloc,
                     url_error (up_error_code));
          url_free (u);
          xfree (url);
          xfree (mynewloc);
          RESTORE_POST_DATA;
          return result;
        }

      /* Now mynewloc will become newloc_parsed->url, because if the
         Location contained relative paths like .././something, we
         don't want that propagating as url.  */
      xfree (mynewloc);
      mynewloc = xstrdup (newloc_parsed->url);

      /* Check for max. number of redirections.  */
      if (++redirection_count > MAX_REDIRECTIONS)
        {
          logprintf (LOG_NOTQUIET, _("%d redirections exceeded.\n"),
                     MAX_REDIRECTIONS);
          url_free (newloc_parsed);
          url_free (u);
          xfree (url);
          xfree (mynewloc);
          RESTORE_POST_DATA;
          return WRONGCODE;
        }

      xfree (url);
      url = mynewloc;
      url_free (u);
      u = newloc_parsed;

      /* If we're being redirected from POST, we don't want to POST
         again.  Many requests answer POST with a redirection to an
         index page; that redirection is clearly a GET.  We "suspend"
         POST data for the duration of the redirections, and restore
         it when we're done. */
      if (!post_data_suspended)
        SUSPEND_POST_DATA;

      goto redirected;
    }

  if (local_file)
    {
      if (*dt & RETROKF)
        {
          register_download (u->url, local_file);
          if (redirection_count && 0 != strcmp (origurl, u->url))
            register_redirection (origurl, u->url);
          if (*dt & TEXTHTML)
            register_html (u->url, local_file);
        }
    }

  if (file)
    *file = local_file ? local_file : NULL;
  else
    xfree_null (local_file);

  url_free (u);

  if (redirection_count)
    {
      if (newloc)
        *newloc = url;
      else
        xfree (url);
    }
  else
    {
      if (newloc)
        *newloc = NULL;
      xfree (url);
    }

  ++global_download_count;
  RESTORE_POST_DATA;

  return result;
}

/* Find the URLs in the file and call retrieve_url() for each of
   them.  If HTML is non-zero, treat the file as HTML, and construct
   the URLs accordingly.

   If opt.recursive is set, call retrieve_tree() for each file.  */

uerr_t
retrieve_from_file (const char *file, int html, int *count)
{
  uerr_t status;
  struct urlpos *url_list, *cur_url;

  url_list = (html ? get_urls_html (file, NULL, NULL)
              : get_urls_file (file));
  status = RETROK;             /* Suppose everything is OK.  */
  *count = 0;                  /* Reset the URL count.  */

  for (cur_url = url_list; cur_url; cur_url = cur_url->next, ++*count)
    {
      char *filename = NULL, *new_file = NULL;
      int dt;

      if (cur_url->ignore_when_downloading)
        continue;

      if (opt.quota && total_downloaded_bytes > opt.quota)
        {
          status = QUOTEXC;
          break;
        }
      if ((opt.recursive || opt.page_requisites)
          && cur_url->url->scheme != SCHEME_FTP)
        status = retrieve_tree (cur_url->url->url);
      else
        status = retrieve_url (cur_url->url->url, &filename, &new_file, NULL, &dt);

      if (filename && opt.delete_after && file_exists_p (filename))
        {
          DEBUGP (("Removing file due to --delete-after in"
                   " retrieve_from_file():\n"));
          logprintf (LOG_VERBOSE, _("Removing %s.\n"), filename);
          if (unlink (filename))
            logprintf (LOG_NOTQUIET, "unlink: %s\n", strerror (errno));
          dt &= ~RETROKF;
        }

      xfree_null (new_file);
      xfree_null (filename);
    }

  /* Free the linked list of URL-s.  */
  free_urlpos (url_list);

  return status;
}

/* Print `giving up', or `retrying', depending on the impending
   action.  N1 and N2 are the attempt number and the attempt limit.  */
void
printwhat (int n1, int n2)
{
  logputs (LOG_VERBOSE, (n1 == n2) ? _("Giving up.\n\n") : _("Retrying.\n\n"));
}

/* If opt.wait or opt.waitretry are specified, and if certain
   conditions are met, sleep the appropriate number of seconds.  See
   the documentation of --wait and --waitretry for more information.

   COUNT is the count of current retrieval, beginning with 1. */

void
sleep_between_retrievals (int count)
{
  static int first_retrieval = 1;

  if (first_retrieval)
    {
      /* Don't sleep before the very first retrieval. */
      first_retrieval = 0;
      return;
    }

  if (opt.waitretry && count > 1)
    {
      /* If opt.waitretry is specified and this is a retry, wait for
         COUNT-1 number of seconds, or for opt.waitretry seconds.  */
      if (count <= opt.waitretry)
        xsleep (count - 1);
      else
        xsleep (opt.waitretry);
    }
  else if (opt.wait)
    {
      if (!opt.random_wait || count > 1)
        /* If random-wait is not specified, or if we are sleeping
           between retries of the same download, sleep the fixed
           interval.  */
        xsleep (opt.wait);
      else
        {
          /* Sleep a random amount of time averaging in opt.wait
             seconds.  The sleeping amount ranges from 0 to
             opt.wait*2, inclusive.  */
          double waitsecs = 2 * opt.wait * random_float ();
          DEBUGP (("sleep_between_retrievals: avg=%f,sleep=%f\n",
                   opt.wait, waitsecs));
          xsleep (waitsecs);
        }
    }
}

/* Free the linked list of urlpos.  */
void
free_urlpos (struct urlpos *l)
{
  while (l)
    {
      struct urlpos *next = l->next;
      if (l->url)
        url_free (l->url);
      xfree_null (l->local_name);
      xfree (l);
      l = next;
    }
}

/* Rotate FNAME opt.backups times */
void
rotate_backups(const char *fname)
{
  int maxlen = strlen (fname) + 1 + numdigit (opt.backups) + 1;
  char *from = (char *)alloca (maxlen);
  char *to = (char *)alloca (maxlen);
  struct stat sb;
  int i;

  if (stat (fname, &sb) == 0)
    if (S_ISREG (sb.st_mode) == 0)
      return;

  for (i = opt.backups; i > 1; i--)
    {
      sprintf (from, "%s.%d", fname, i - 1);
      sprintf (to, "%s.%d", fname, i);
      rename (from, to);
    }

  sprintf (to, "%s.%d", fname, 1);
  rename(fname, to);
}

static int no_proxy_match PARAMS ((const char *, const char **));

/* Return the URL of the proxy appropriate for url U.  */

static char *
getproxy (struct url *u)
{
  char *proxy = NULL;
  char *rewritten_url;
  static char rewritten_storage[1024];

  if (!opt.use_proxy)
    return NULL;
  if (!no_proxy_match (u->host, (const char **)opt.no_proxy))
    return NULL;

  switch (u->scheme)
    {
    case SCHEME_HTTP:
      proxy = opt.http_proxy ? opt.http_proxy : getenv ("http_proxy");
      break;
#ifdef HAVE_SSL
    case SCHEME_HTTPS:
      proxy = opt.https_proxy ? opt.https_proxy : getenv ("https_proxy");
      break;
#endif
    case SCHEME_FTP:
      proxy = opt.ftp_proxy ? opt.ftp_proxy : getenv ("ftp_proxy");
      break;
    case SCHEME_INVALID:
      break;
    }
  if (!proxy || !*proxy)
    return NULL;

  /* Handle shorthands.  `rewritten_storage' is a kludge to allow
     getproxy() to return static storage. */
  rewritten_url = rewrite_shorthand_url (proxy);
  if (rewritten_url)
    {
      strncpy (rewritten_storage, rewritten_url, sizeof (rewritten_storage));
      rewritten_storage[sizeof (rewritten_storage) - 1] = '\0';
      proxy = rewritten_storage;
    }

  return proxy;
}

/* Should a host be accessed through proxy, concerning no_proxy?  */
int
no_proxy_match (const char *host, const char **no_proxy)
{
  if (!no_proxy)
    return 1;
  else
    return !sufmatch (no_proxy, host);
}