2008-01-22 22:36:11 -05:00
|
|
|
/*
|
|
|
|
* download.c
|
|
|
|
*
|
2011-01-05 23:45:15 -05:00
|
|
|
* Copyright (c) 2006-2011 Pacman Development Team <pacman-dev@archlinux.org>
|
2009-07-01 03:08:33 -04:00
|
|
|
* Copyright (c) 2002-2006 by Judd Vinet <jvinet@zeroflux.org>
|
2008-01-22 22:36:11 -05:00
|
|
|
*
|
|
|
|
* This program 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.
|
|
|
|
*
|
|
|
|
* This program 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 this program. If not, see <http://www.gnu.org/licenses/>.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include "config.h"
|
|
|
|
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <errno.h>
|
|
|
|
#include <string.h>
|
|
|
|
#include <unistd.h>
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
#include <sys/time.h>
|
|
|
|
#include <sys/types.h>
|
|
|
|
#include <sys/stat.h>
|
2008-10-31 20:40:55 -04:00
|
|
|
#include <signal.h>
|
2010-12-17 14:11:29 -05:00
|
|
|
/* the following two are needed for FreeBSD's libfetch */
|
|
|
|
#include <limits.h> /* PATH_MAX */
|
2008-04-19 13:36:18 -04:00
|
|
|
#if defined(HAVE_SYS_PARAM_H)
|
|
|
|
#include <sys/param.h> /* MAXHOSTNAMELEN */
|
|
|
|
#endif
|
|
|
|
|
2010-12-15 19:57:31 -05:00
|
|
|
#ifdef HAVE_LIBCURL
|
|
|
|
#include <curl/curl.h>
|
|
|
|
#endif
|
|
|
|
|
2010-09-06 12:20:51 -04:00
|
|
|
#ifdef HAVE_LIBFETCH
|
2008-04-19 13:36:18 -04:00
|
|
|
#include <fetch.h>
|
2008-04-06 22:00:11 -04:00
|
|
|
#endif
|
2008-01-22 22:36:11 -05:00
|
|
|
|
|
|
|
/* libalpm */
|
|
|
|
#include "dload.h"
|
|
|
|
#include "alpm_list.h"
|
|
|
|
#include "alpm.h"
|
|
|
|
#include "log.h"
|
|
|
|
#include "util.h"
|
|
|
|
#include "handle.h"
|
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
static char *get_filename(const char *url) {
|
|
|
|
char *filename = strrchr(url, '/');
|
|
|
|
if(filename != NULL) {
|
|
|
|
filename++;
|
|
|
|
}
|
|
|
|
return(filename);
|
|
|
|
}
|
2008-01-22 23:00:12 -05:00
|
|
|
|
2011-01-14 12:17:39 -05:00
|
|
|
#if defined(HAVE_LIBFETCH) || defined(HAVE_LIBCURL)
|
2008-02-27 16:36:53 -05:00
|
|
|
static char *get_destfile(const char *path, const char *filename) {
|
|
|
|
char *destfile;
|
|
|
|
/* len = localpath len + filename len + null */
|
2009-10-11 15:02:20 -04:00
|
|
|
size_t len = strlen(path) + strlen(filename) + 1;
|
2008-02-27 16:36:53 -05:00
|
|
|
CALLOC(destfile, len, sizeof(char), RET_ERR(PM_ERR_MEMORY, NULL));
|
|
|
|
snprintf(destfile, len, "%s%s", path, filename);
|
2008-01-22 23:00:12 -05:00
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
return(destfile);
|
|
|
|
}
|
|
|
|
|
|
|
|
static char *get_tempfile(const char *path, const char *filename) {
|
|
|
|
char *tempfile;
|
|
|
|
/* len = localpath len + filename len + '.part' len + null */
|
2009-10-11 15:02:20 -04:00
|
|
|
size_t len = strlen(path) + strlen(filename) + 6;
|
2008-02-27 16:36:53 -05:00
|
|
|
CALLOC(tempfile, len, sizeof(char), RET_ERR(PM_ERR_MEMORY, NULL));
|
|
|
|
snprintf(tempfile, len, "%s%s.part", path, filename);
|
|
|
|
|
|
|
|
return(tempfile);
|
|
|
|
}
|
|
|
|
|
2011-01-14 12:17:39 -05:00
|
|
|
#define check_stop() if(dload_interrupted) { ret = -1; goto cleanup; }
|
|
|
|
enum sighandlers { OLD = 0, NEW = 1 };
|
|
|
|
|
|
|
|
int dload_interrupted;
|
|
|
|
static void inthandler(int signum)
|
|
|
|
{
|
|
|
|
dload_interrupted = 1;
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
#ifdef HAVE_LIBFETCH
|
2011-01-12 09:51:06 -05:00
|
|
|
static const char *fetch_gethost(struct url *fileurl)
|
2009-09-29 13:19:54 -04:00
|
|
|
{
|
|
|
|
const char *host = _("disk");
|
|
|
|
if(strcmp(SCHEME_FILE, fileurl->scheme) != 0) {
|
|
|
|
host = fileurl->host;
|
|
|
|
}
|
|
|
|
return(host);
|
|
|
|
}
|
|
|
|
|
2011-01-12 09:51:06 -05:00
|
|
|
static int fetch_download_internal(const char *url, const char *localpath,
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
int force) {
|
2008-11-15 04:10:56 -05:00
|
|
|
FILE *localf = NULL;
|
2008-02-27 16:36:53 -05:00
|
|
|
struct stat st;
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
int ret = 0;
|
2009-10-11 15:02:20 -04:00
|
|
|
off_t dl_thisfile = 0;
|
2009-09-29 13:19:54 -04:00
|
|
|
ssize_t nread = 0;
|
2008-02-27 16:36:53 -05:00
|
|
|
char *tempfile, *destfile, *filename;
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
struct sigaction sig_pipe[2], sig_int[2];
|
|
|
|
|
|
|
|
off_t local_size = 0;
|
|
|
|
time_t local_time = 0;
|
|
|
|
|
2009-08-08 13:59:02 -04:00
|
|
|
struct url *fileurl;
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
struct url_stat ust;
|
|
|
|
fetchIO *dlf = NULL;
|
|
|
|
|
2008-11-17 22:47:55 -05:00
|
|
|
char buffer[PM_DLBUF_LEN];
|
2008-02-27 16:36:53 -05:00
|
|
|
|
|
|
|
filename = get_filename(url);
|
|
|
|
if(!filename) {
|
2009-12-09 17:20:29 -05:00
|
|
|
_alpm_log(PM_LOG_ERROR, _("url '%s' is invalid\n"), url);
|
|
|
|
RET_ERR(PM_ERR_SERVER_BAD_URL, -1);
|
2008-02-27 16:36:53 -05:00
|
|
|
}
|
2009-08-08 13:59:02 -04:00
|
|
|
|
|
|
|
fileurl = fetchParseURL(url);
|
|
|
|
if(!fileurl) {
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("url '%s' is invalid\n"), url);
|
2009-12-09 17:20:29 -05:00
|
|
|
RET_ERR(PM_ERR_LIBFETCH, -1);
|
2009-08-08 13:59:02 -04:00
|
|
|
}
|
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
destfile = get_destfile(localpath, filename);
|
|
|
|
tempfile = get_tempfile(localpath, filename);
|
|
|
|
|
2011-02-04 09:42:52 -05:00
|
|
|
if(stat(tempfile, &st) == 0 && S_ISREG(st.st_mode) && st.st_size > 0) {
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
_alpm_log(PM_LOG_DEBUG, "tempfile found, attempting continuation\n");
|
|
|
|
local_time = fileurl->last_modified = st.st_mtime;
|
|
|
|
local_size = fileurl->offset = (off_t)st.st_size;
|
2008-02-27 16:36:53 -05:00
|
|
|
dl_thisfile = st.st_size;
|
|
|
|
localf = fopen(tempfile, "ab");
|
2011-02-04 09:42:52 -05:00
|
|
|
} else if(!force && stat(destfile, &st) == 0 && S_ISREG(st.st_mode) && st.st_size > 0) {
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
_alpm_log(PM_LOG_DEBUG, "destfile found, using mtime only\n");
|
|
|
|
local_time = fileurl->last_modified = st.st_mtime;
|
|
|
|
local_size = /* no fu->off here */ (off_t)st.st_size;
|
2008-02-27 16:36:53 -05:00
|
|
|
} else {
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
_alpm_log(PM_LOG_DEBUG, "no file found matching criteria, starting from scratch\n");
|
2008-02-27 16:36:53 -05:00
|
|
|
}
|
|
|
|
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
/* pass the raw filename for passing to the callback function */
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "using '%s' for download progress\n", filename);
|
|
|
|
|
2008-12-10 20:45:15 -05:00
|
|
|
/* print proxy info for debug purposes */
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "HTTP_PROXY: %s\n", getenv("HTTP_PROXY"));
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "http_proxy: %s\n", getenv("http_proxy"));
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "FTP_PROXY: %s\n", getenv("FTP_PROXY"));
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "ftp_proxy: %s\n", getenv("ftp_proxy"));
|
|
|
|
|
2009-08-13 22:01:40 -04:00
|
|
|
/* 10s timeout */
|
|
|
|
fetchTimeout = 10;
|
2008-02-27 16:36:53 -05:00
|
|
|
|
2008-10-31 20:40:55 -04:00
|
|
|
/* ignore any SIGPIPE signals- these may occur if our FTP socket dies or
|
|
|
|
* something along those lines. Store the old signal handler first. */
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
sig_pipe[NEW].sa_handler = SIG_IGN;
|
|
|
|
sigemptyset(&sig_pipe[NEW].sa_mask);
|
|
|
|
sig_pipe[NEW].sa_flags = 0;
|
|
|
|
sigaction(SIGPIPE, NULL, &sig_pipe[OLD]);
|
|
|
|
sigaction(SIGPIPE, &sig_pipe[NEW], NULL);
|
|
|
|
|
|
|
|
dload_interrupted = 0;
|
|
|
|
sig_int[NEW].sa_handler = &inthandler;
|
|
|
|
sigemptyset(&sig_int[NEW].sa_mask);
|
|
|
|
sig_int[NEW].sa_flags = 0;
|
|
|
|
sigaction(SIGINT, NULL, &sig_int[OLD]);
|
|
|
|
sigaction(SIGINT, &sig_int[NEW], NULL);
|
|
|
|
|
|
|
|
/* NOTE: libfetch does not reset the error code, be sure to do it before
|
|
|
|
* calls into the library */
|
|
|
|
|
2011-02-04 10:15:22 -05:00
|
|
|
/* TODO: if we call fetchStat() and get a redirect (disabling automagic
|
|
|
|
* redirect following), we should repeat the file locator stuff and get a new
|
|
|
|
* filename rather than only base if off the first URL, and then verify
|
|
|
|
* get_filename() didn't return ''. Of course, libfetch might not even allow
|
|
|
|
* us to even get that URL...FS#22645. This would allow us to download things
|
|
|
|
* without totally puking like
|
|
|
|
* http://www.archlinux.org/packages/community/x86_64/exim/download/ */
|
|
|
|
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
/* find out the remote size *and* mtime in one go. there is a lot of
|
|
|
|
* trouble in trying to do both size and "if-modified-since" logic in a
|
|
|
|
* non-stat request, so avoid it. */
|
|
|
|
fetchLastErrCode = 0;
|
|
|
|
if(fetchStat(fileurl, &ust, "") == -1) {
|
2009-12-09 17:20:29 -05:00
|
|
|
pm_errno = PM_ERR_LIBFETCH;
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("failed retrieving file '%s' from %s : %s\n"),
|
2011-01-12 09:51:06 -05:00
|
|
|
filename, fetch_gethost(fileurl), fetchLastErrString);
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
ret = -1;
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
check_stop();
|
|
|
|
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "ust.mtime: %ld local_time: %ld compare: %ld\n",
|
|
|
|
ust.mtime, local_time, local_time - ust.mtime);
|
2009-11-14 14:26:37 -05:00
|
|
|
_alpm_log(PM_LOG_DEBUG, "ust.size: %jd local_size: %jd compare: %jd\n",
|
|
|
|
(intmax_t)ust.size, (intmax_t)local_size, (intmax_t)(local_size - ust.size));
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
if(!force && ust.mtime && ust.mtime == local_time
|
|
|
|
&& ust.size && ust.size == local_size) {
|
|
|
|
/* the remote time and size values agreed with what we have, so move on
|
|
|
|
* because there is nothing more to do. */
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "files are identical, skipping %s\n", filename);
|
2009-08-07 10:58:01 -04:00
|
|
|
ret = 1;
|
|
|
|
goto cleanup;
|
|
|
|
}
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
if(!ust.mtime || ust.mtime != local_time) {
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "mtimes were different or unavailable, downloading %s from beginning\n", filename);
|
|
|
|
fileurl->offset = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
fetchLastErrCode = 0;
|
|
|
|
dlf = fetchGet(fileurl, "");
|
|
|
|
check_stop();
|
2008-02-27 16:36:53 -05:00
|
|
|
|
2008-10-31 20:11:38 -04:00
|
|
|
if(fetchLastErrCode != 0 || dlf == NULL) {
|
|
|
|
pm_errno = PM_ERR_LIBFETCH;
|
2008-02-27 16:36:53 -05:00
|
|
|
_alpm_log(PM_LOG_ERROR, _("failed retrieving file '%s' from %s : %s\n"),
|
2011-01-12 09:51:06 -05:00
|
|
|
filename, fetch_gethost(fileurl), fetchLastErrString);
|
2008-02-27 16:36:53 -05:00
|
|
|
ret = -1;
|
|
|
|
goto cleanup;
|
|
|
|
} else {
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "connected to %s successfully\n", fileurl->host);
|
|
|
|
}
|
2008-01-22 23:00:12 -05:00
|
|
|
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
if(localf && fileurl->offset == 0) {
|
|
|
|
_alpm_log(PM_LOG_WARNING, _("resuming download of %s not possible; starting over\n"), filename);
|
|
|
|
fclose(localf);
|
|
|
|
localf = NULL;
|
|
|
|
} else if(fileurl->offset) {
|
2009-11-14 14:26:37 -05:00
|
|
|
_alpm_log(PM_LOG_DEBUG, "resuming download at position %jd\n", (intmax_t)fileurl->offset);
|
2008-02-27 16:36:53 -05:00
|
|
|
}
|
2008-01-22 23:00:12 -05:00
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
|
|
|
|
if(localf == NULL) {
|
|
|
|
_alpm_rmrf(tempfile);
|
|
|
|
fileurl->offset = (off_t)0;
|
|
|
|
dl_thisfile = 0;
|
|
|
|
localf = fopen(tempfile, "wb");
|
|
|
|
if(localf == NULL) { /* still null? */
|
2009-12-09 17:20:29 -05:00
|
|
|
pm_errno = PM_ERR_RETRIEVE;
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
_alpm_log(PM_LOG_ERROR, _("error writing to file '%s': %s\n"),
|
|
|
|
tempfile, strerror(errno));
|
2008-02-27 16:36:53 -05:00
|
|
|
ret = -1;
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Progress 0 - initialize */
|
|
|
|
if(handle->dlcb) {
|
|
|
|
handle->dlcb(filename, 0, ust.size);
|
|
|
|
}
|
|
|
|
|
2008-11-15 04:10:56 -05:00
|
|
|
while((nread = fetchIO_read(dlf, buffer, PM_DLBUF_LEN)) > 0) {
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
check_stop();
|
2008-11-17 22:47:55 -05:00
|
|
|
size_t nwritten = 0;
|
2011-01-07 22:06:06 -05:00
|
|
|
nwritten = fwrite(buffer, 1, (size_t)nread, localf);
|
2010-06-28 09:44:31 -04:00
|
|
|
if((nwritten != (size_t)nread) || ferror(localf)) {
|
2009-12-09 17:20:29 -05:00
|
|
|
pm_errno = PM_ERR_RETRIEVE;
|
2009-08-08 09:20:24 -04:00
|
|
|
_alpm_log(PM_LOG_ERROR, _("error writing to file '%s': %s\n"),
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
tempfile, strerror(errno));
|
2009-08-08 09:20:24 -04:00
|
|
|
ret = -1;
|
|
|
|
goto cleanup;
|
2008-02-27 16:36:53 -05:00
|
|
|
}
|
|
|
|
dl_thisfile += nread;
|
|
|
|
|
|
|
|
if(handle->dlcb) {
|
|
|
|
handle->dlcb(filename, dl_thisfile, ust.size);
|
|
|
|
}
|
|
|
|
}
|
2008-11-15 04:10:56 -05:00
|
|
|
|
|
|
|
/* did the transfer complete normally? */
|
2009-09-29 13:19:54 -04:00
|
|
|
if (nread == -1) {
|
|
|
|
/* not PM_ERR_LIBFETCH here because libfetch error string might be empty */
|
|
|
|
pm_errno = PM_ERR_RETRIEVE;
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("failed retrieving file '%s' from %s\n"),
|
2011-01-12 09:51:06 -05:00
|
|
|
filename, fetch_gethost(fileurl));
|
2009-09-29 13:19:54 -04:00
|
|
|
ret = -1;
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
|
2008-11-15 04:10:56 -05:00
|
|
|
if (ust.size != -1 && dl_thisfile < ust.size) {
|
2009-09-29 13:19:54 -04:00
|
|
|
pm_errno = PM_ERR_RETRIEVE;
|
2008-11-15 04:10:56 -05:00
|
|
|
_alpm_log(PM_LOG_ERROR, _("%s appears to be truncated: %jd/%jd bytes\n"),
|
|
|
|
filename, (intmax_t)dl_thisfile, (intmax_t)ust.size);
|
|
|
|
ret = -1;
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
/* probably safer to close the file descriptors now before renaming the file,
|
|
|
|
* for example to make sure the buffers are flushed.
|
|
|
|
*/
|
|
|
|
fclose(localf);
|
|
|
|
localf = NULL;
|
2008-11-15 04:10:56 -05:00
|
|
|
fetchIO_close(dlf);
|
2008-02-27 16:36:53 -05:00
|
|
|
dlf = NULL;
|
|
|
|
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
/* set the times on the file to the same as that of the remote file */
|
|
|
|
if(ust.mtime) {
|
|
|
|
struct timeval tv[2];
|
|
|
|
memset(&tv, 0, sizeof(tv));
|
|
|
|
tv[0].tv_sec = ust.atime;
|
|
|
|
tv[1].tv_sec = ust.mtime;
|
|
|
|
utimes(tempfile, tv);
|
|
|
|
}
|
2011-02-04 10:00:47 -05:00
|
|
|
if(rename(tempfile, destfile)) {
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("could not rename %s to %s (%s)\n"),
|
|
|
|
tempfile, destfile, strerror(errno));
|
|
|
|
ret = -1;
|
|
|
|
}
|
2008-02-27 16:36:53 -05:00
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
cleanup:
|
|
|
|
FREE(tempfile);
|
|
|
|
FREE(destfile);
|
|
|
|
if(localf != NULL) {
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
/* if we still had a local file open, we got interrupted. set the mtimes on
|
|
|
|
* the file accordingly. */
|
|
|
|
fflush(localf);
|
|
|
|
if(ust.mtime) {
|
|
|
|
struct timeval tv[2];
|
|
|
|
memset(&tv, 0, sizeof(tv));
|
|
|
|
tv[0].tv_sec = ust.atime;
|
|
|
|
tv[1].tv_sec = ust.mtime;
|
|
|
|
futimes(fileno(localf), tv);
|
|
|
|
}
|
2008-02-27 16:36:53 -05:00
|
|
|
fclose(localf);
|
|
|
|
}
|
|
|
|
if(dlf != NULL) {
|
2008-11-15 04:10:56 -05:00
|
|
|
fetchIO_close(dlf);
|
2008-02-27 16:36:53 -05:00
|
|
|
}
|
2008-10-31 20:11:38 -04:00
|
|
|
fetchFreeURL(fileurl);
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
|
|
|
|
/* restore the old signal handlers */
|
|
|
|
sigaction(SIGINT, &sig_int[OLD], NULL);
|
|
|
|
sigaction(SIGPIPE, &sig_pipe[OLD], NULL);
|
|
|
|
/* if we were interrupted, trip the old handler */
|
|
|
|
if(dload_interrupted) {
|
|
|
|
raise(SIGINT);
|
|
|
|
}
|
|
|
|
|
2008-01-22 23:00:12 -05:00
|
|
|
return(ret);
|
|
|
|
}
|
2008-04-06 22:00:11 -04:00
|
|
|
#endif
|
2008-01-22 23:00:12 -05:00
|
|
|
|
2011-02-01 15:38:37 -05:00
|
|
|
#ifdef HAVE_LIBCURL
|
|
|
|
static int curl_progress(void *filename, double dltotal, double dlnow,
|
|
|
|
double ultotal, double ulnow) {
|
|
|
|
|
|
|
|
/* unused parameters */
|
|
|
|
(void)ultotal;
|
|
|
|
(void)ulnow;
|
|
|
|
|
|
|
|
if(dltotal == 0) {
|
|
|
|
return(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
if(dload_interrupted) {
|
|
|
|
return(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
handle->dlcb((const char*)filename, (long)dlnow, (long)dltotal);
|
|
|
|
|
|
|
|
return(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int curl_gethost(const char *url, char *buffer) {
|
|
|
|
int hostlen;
|
|
|
|
char *p;
|
|
|
|
|
|
|
|
if(strncmp(url, "file://", 7) == 0) {
|
|
|
|
strcpy(buffer, _("disk"));
|
|
|
|
} else {
|
|
|
|
p = strstr(url, "//");
|
|
|
|
if(!p) {
|
|
|
|
return(1);
|
|
|
|
}
|
|
|
|
p += 2; /* jump over the found // */
|
|
|
|
hostlen = strcspn(p, "/");
|
|
|
|
if(hostlen > 255) {
|
|
|
|
/* buffer overflow imminent */
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("buffer overflow detected"));
|
|
|
|
return(1);
|
|
|
|
}
|
|
|
|
snprintf(buffer, hostlen + 1, "%s", p);
|
|
|
|
}
|
|
|
|
|
|
|
|
return(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int curl_download_internal(const char *url, const char *localpath,
|
|
|
|
int force) {
|
|
|
|
int ret = -1;
|
|
|
|
FILE *localf = NULL;
|
|
|
|
char *destfile, *filename, *tempfile;
|
|
|
|
char hostname[256]; /* RFC1123 states applications should support this length */
|
|
|
|
struct stat st;
|
|
|
|
long httpresp, timecond, remote_time, local_time;
|
|
|
|
double remote_size, bytes_dl;
|
|
|
|
struct sigaction sig_pipe[2], sig_int[2];
|
|
|
|
|
|
|
|
filename = get_filename(url);
|
|
|
|
if(!filename || curl_gethost(url, hostname) != 0) {
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("url '%s' is invalid\n"), url);
|
|
|
|
RET_ERR(PM_ERR_SERVER_BAD_URL, -1);
|
|
|
|
}
|
|
|
|
|
|
|
|
destfile = get_destfile(localpath, filename);
|
|
|
|
tempfile = get_tempfile(localpath, filename);
|
|
|
|
|
|
|
|
/* the curl_easy handle is initialized with the alpm handle, so we only need
|
|
|
|
* to reset the curl handle set parameters for each time it's used. */
|
|
|
|
curl_easy_reset(handle->curl);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_URL, url);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_FAILONERROR, 1L);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_ENCODING, "deflate, gzip");
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_CONNECTTIMEOUT, 10L);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_FILETIME, 1L);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_NOPROGRESS, 0L);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_FOLLOWLOCATION, 1L);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_PROGRESSFUNCTION, curl_progress);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_PROGRESSDATA, filename);
|
|
|
|
|
|
|
|
if(!force && stat(destfile, &st) == 0) {
|
|
|
|
/* assume its a sync, so we're starting from scratch. but, only download
|
|
|
|
* our local is out of date. */
|
|
|
|
local_time = (long)st.st_mtime;
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_TIMECONDITION, CURL_TIMECOND_IFMODSINCE);
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_TIMEVALUE, local_time);
|
|
|
|
} else if(stat(tempfile, &st) == 0 && st.st_size > 0) {
|
|
|
|
/* assume its a partial package download. we do not support resuming of
|
|
|
|
* transfers on partially downloaded sync DBs. */
|
|
|
|
localf = fopen(tempfile, "ab");
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_RESUME_FROM, (long)st.st_size);
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "tempfile found, attempting continuation");
|
|
|
|
}
|
|
|
|
|
|
|
|
/* no destfile and no tempfile. start from scratch */
|
|
|
|
if(localf == NULL) {
|
|
|
|
localf = fopen(tempfile, "wb");
|
|
|
|
if(localf == NULL) {
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* this has to be set _after_ figuring out which file we're opening */
|
|
|
|
curl_easy_setopt(handle->curl, CURLOPT_WRITEDATA, localf);
|
|
|
|
|
|
|
|
/* print proxy info for debug purposes */
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "HTTP_PROXY: %s\n", getenv("HTTP_PROXY"));
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "http_proxy: %s\n", getenv("http_proxy"));
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "FTP_PROXY: %s\n", getenv("FTP_PROXY"));
|
|
|
|
_alpm_log(PM_LOG_DEBUG, "ftp_proxy: %s\n", getenv("ftp_proxy"));
|
|
|
|
|
|
|
|
/* ignore any SIGPIPE signals- these may occur if our FTP socket dies or
|
|
|
|
* something along those lines. Store the old signal handler first. */
|
|
|
|
sig_pipe[NEW].sa_handler = SIG_IGN;
|
|
|
|
sigemptyset(&sig_pipe[NEW].sa_mask);
|
|
|
|
sig_pipe[NEW].sa_flags = 0;
|
|
|
|
sigaction(SIGPIPE, NULL, &sig_pipe[OLD]);
|
|
|
|
sigaction(SIGPIPE, &sig_pipe[NEW], NULL);
|
|
|
|
|
|
|
|
dload_interrupted = 0;
|
|
|
|
sig_int[NEW].sa_handler = &inthandler;
|
|
|
|
sigemptyset(&sig_int[NEW].sa_mask);
|
|
|
|
sig_int[NEW].sa_flags = 0;
|
|
|
|
sigaction(SIGINT, NULL, &sig_int[OLD]);
|
|
|
|
sigaction(SIGINT, &sig_int[NEW], NULL);
|
|
|
|
|
|
|
|
/* Progress 0 - initialize */
|
|
|
|
if(handle->dlcb) {
|
|
|
|
handle->dlcb(filename, 0, 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* perform transfer */
|
|
|
|
handle->curlerr = curl_easy_perform(handle->curl);
|
|
|
|
|
|
|
|
/* retrieve info about the state of the transfer */
|
|
|
|
curl_easy_getinfo(handle->curl, CURLINFO_HTTP_CODE, &httpresp);
|
|
|
|
curl_easy_getinfo(handle->curl, CURLINFO_FILETIME, &remote_time);
|
|
|
|
curl_easy_getinfo(handle->curl, CURLINFO_CONTENT_LENGTH_DOWNLOAD, &remote_size);
|
|
|
|
curl_easy_getinfo(handle->curl, CURLINFO_SIZE_DOWNLOAD, &bytes_dl);
|
|
|
|
curl_easy_getinfo(handle->curl, CURLINFO_CONDITION_UNMET, &timecond);
|
|
|
|
|
|
|
|
/* time condition was met and we didn't download anything. we need to
|
|
|
|
* clean up the 0 byte .part file that's left behind. */
|
|
|
|
if(bytes_dl == 0 && timecond == 1) {
|
|
|
|
ret = 1;
|
|
|
|
unlink(tempfile);
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(handle->curlerr == CURLE_ABORTED_BY_CALLBACK) {
|
|
|
|
goto cleanup;
|
|
|
|
} else if(handle->curlerr != CURLE_OK) {
|
|
|
|
pm_errno = PM_ERR_LIBCURL;
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("failed retrieving file '%s' from %s : %s\n"),
|
|
|
|
filename, hostname, curl_easy_strerror(handle->curlerr));
|
|
|
|
unlink(tempfile);
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* remote_size isn't necessarily the full size of the file, just what the
|
|
|
|
* server reported as remaining to download. compare it to what curl reported
|
|
|
|
* as actually being transferred during curl_easy_perform() */
|
|
|
|
if((remote_size != -1 && bytes_dl != -1) && bytes_dl != remote_size) {
|
|
|
|
pm_errno = PM_ERR_RETRIEVE;
|
|
|
|
_alpm_log(PM_LOG_ERROR, _("%s appears to be truncated: %jd/%jd bytes\n"),
|
|
|
|
filename, (intmax_t)bytes_dl, (intmax_t)remote_size);
|
|
|
|
goto cleanup;
|
|
|
|
}
|
|
|
|
|
|
|
|
fclose(localf);
|
|
|
|
localf = NULL;
|
|
|
|
|
|
|
|
/* set the times on the file to the same as that of the remote file */
|
|
|
|
if(remote_time != -1) {
|
|
|
|
struct timeval tv[2];
|
|
|
|
memset(&tv, 0, sizeof(tv));
|
|
|
|
tv[0].tv_sec = tv[1].tv_sec = remote_time;
|
|
|
|
utimes(tempfile, tv);
|
|
|
|
}
|
|
|
|
rename(tempfile, destfile);
|
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
cleanup:
|
|
|
|
FREE(tempfile);
|
|
|
|
FREE(destfile);
|
|
|
|
if(localf != NULL) {
|
|
|
|
/* if we still had a local file open, we got interrupted. set the mtimes on
|
|
|
|
* the file accordingly. */
|
|
|
|
fflush(localf);
|
|
|
|
if(remote_time != -1) {
|
|
|
|
struct timeval tv[2];
|
|
|
|
memset(&tv, 0, sizeof(tv));
|
|
|
|
tv[0].tv_sec = tv[1].tv_sec = remote_time;
|
|
|
|
futimes(fileno(localf), tv);
|
|
|
|
}
|
|
|
|
fclose(localf);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* restore the old signal handlers */
|
|
|
|
sigaction(SIGINT, &sig_int[OLD], NULL);
|
|
|
|
sigaction(SIGPIPE, &sig_pipe[OLD], NULL);
|
|
|
|
/* if we were interrupted, trip the old handler */
|
|
|
|
if(dload_interrupted) {
|
|
|
|
raise(SIGINT);
|
|
|
|
}
|
|
|
|
|
|
|
|
return(ret);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
static int download(const char *url, const char *localpath,
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
int force) {
|
2009-04-04 04:17:30 -04:00
|
|
|
if(handle->fetchcb == NULL) {
|
2010-09-06 12:20:51 -04:00
|
|
|
#ifdef HAVE_LIBFETCH
|
2011-01-12 09:51:06 -05:00
|
|
|
return(fetch_download_internal(url, localpath, force));
|
2008-04-06 22:00:11 -04:00
|
|
|
#else
|
|
|
|
RET_ERR(PM_ERR_EXTERNAL_DOWNLOAD, -1);
|
|
|
|
#endif
|
2008-02-27 16:36:53 -05:00
|
|
|
} else {
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
int ret = handle->fetchcb(url, localpath, force);
|
2009-04-04 04:17:30 -04:00
|
|
|
if(ret == -1) {
|
|
|
|
RET_ERR(PM_ERR_EXTERNAL_DOWNLOAD, -1);
|
|
|
|
}
|
|
|
|
return(ret);
|
2008-02-27 16:36:53 -05:00
|
|
|
}
|
|
|
|
}
|
2008-01-22 23:00:12 -05:00
|
|
|
|
2008-01-22 22:36:11 -05:00
|
|
|
/*
|
2008-02-27 16:36:53 -05:00
|
|
|
* Download a single file
|
2008-08-03 19:16:08 -04:00
|
|
|
* - servers must be a list of urls WITHOUT trailing slashes.
|
2008-01-22 22:36:11 -05:00
|
|
|
*
|
|
|
|
* RETURN: 0 for successful download
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
* 1 if the files are identical
|
2008-01-22 22:36:11 -05:00
|
|
|
* -1 on error
|
|
|
|
*/
|
2008-02-27 16:36:53 -05:00
|
|
|
int _alpm_download_single_file(const char *filename,
|
|
|
|
alpm_list_t *servers, const char *localpath,
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
int force)
|
2008-01-22 22:36:11 -05:00
|
|
|
{
|
|
|
|
alpm_list_t *i;
|
2008-02-27 11:45:01 -05:00
|
|
|
int ret = -1;
|
2008-01-22 22:36:11 -05:00
|
|
|
|
2009-06-06 12:03:29 -04:00
|
|
|
ASSERT(servers != NULL, RET_ERR(PM_ERR_SERVER_NONE, -1));
|
|
|
|
|
2008-02-27 11:45:01 -05:00
|
|
|
for(i = servers; i; i = i->next) {
|
2008-01-22 23:49:45 -05:00
|
|
|
const char *server = i->data;
|
2008-02-27 16:36:53 -05:00
|
|
|
char *fileurl = NULL;
|
2009-10-11 15:02:20 -04:00
|
|
|
size_t len;
|
2008-02-27 16:36:53 -05:00
|
|
|
|
|
|
|
/* print server + filename into a buffer */
|
|
|
|
len = strlen(server) + strlen(filename) + 2;
|
|
|
|
CALLOC(fileurl, len, sizeof(char), RET_ERR(PM_ERR_MEMORY, -1));
|
|
|
|
snprintf(fileurl, len, "%s/%s", server, filename);
|
|
|
|
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
ret = download(fileurl, localpath, force);
|
2008-02-27 16:36:53 -05:00
|
|
|
FREE(fileurl);
|
|
|
|
if(ret != -1) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
2008-01-22 22:36:11 -05:00
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
return(ret);
|
|
|
|
}
|
2008-01-22 22:36:11 -05:00
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
int _alpm_download_files(alpm_list_t *files,
|
|
|
|
alpm_list_t *servers, const char *localpath)
|
|
|
|
{
|
|
|
|
int ret = 0;
|
|
|
|
alpm_list_t *lp;
|
2008-01-22 22:36:11 -05:00
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
for(lp = files; lp; lp = lp->next) {
|
|
|
|
char *filename = lp->data;
|
|
|
|
if(_alpm_download_single_file(filename, servers,
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
localpath, 0) == -1) {
|
2008-02-27 16:36:53 -05:00
|
|
|
ret++;
|
2008-01-22 22:36:11 -05:00
|
|
|
}
|
|
|
|
}
|
2008-02-27 11:45:01 -05:00
|
|
|
|
|
|
|
return(ret);
|
2008-01-22 22:36:11 -05:00
|
|
|
}
|
|
|
|
|
|
|
|
/** Fetch a remote pkg.
|
|
|
|
* @param url URL of the package to download
|
|
|
|
* @return the downloaded filepath on success, NULL on error
|
|
|
|
* @addtogroup alpm_misc
|
|
|
|
*/
|
|
|
|
char SYMEXPORT *alpm_fetch_pkgurl(const char *url)
|
|
|
|
{
|
|
|
|
char *filename, *filepath;
|
|
|
|
const char *cachedir;
|
2008-02-27 16:36:53 -05:00
|
|
|
int ret;
|
2008-01-22 22:36:11 -05:00
|
|
|
|
|
|
|
ALPM_LOG_FUNC;
|
|
|
|
|
2008-02-27 16:36:53 -05:00
|
|
|
filename = get_filename(url);
|
2008-01-22 22:36:11 -05:00
|
|
|
|
|
|
|
/* find a valid cache dir to download to */
|
|
|
|
cachedir = _alpm_filecache_setup();
|
|
|
|
|
|
|
|
/* download the file */
|
download: major refactor to address lingering issues
Sorry for this being such a huge patch, but I believe it is necessary for
quite a few reasons which I will attempt to explain herein. I've been
mulling this over for a while, but wasn't super happy with making the
download interface more complex. Instead, if we carefully order things in
the internal download code, we can actually make the interface simpler.
1. FS#15657 - This involves `name.db.tar.gz.part` files being left around the
filesystem, and then causing all sorts of issues when someone attempts to
rerun the operation they canceled. We need to ensure that if we resume a
download, we are resuming it on exactly the same file; if we cannot be
almost postive of that then we need to start over.
2. http://www.mail-archive.com/pacman-dev@archlinux.org/msg03536.html - Here
we have a lighttpd bug to ruin the day. If we send both a Range: header and
If-Modified-Since: header across the wire in a GET request, lighttpd doesn't
do what we want in several cases. If the file hadn't been modified, it
returns a '304 Not Modified' instead of a '206 Partial Content'. We need to
do a stat (e.g. HEAD in HTTP terms) operation here, and the proceed
accordingly based off the values we get back from it.
3. The mtime stuff was rather ugly, and relied on the called function to
write back to a passed in reference, which isn't the greatest. Instead, use
the power of the filesystem to contain this info. Every file downloaded
internally is now carefully timestamped with the remote file time. This
should allow the resume logic to work. In order to guarantee this, we need
to implement a signal handler that catches interrupts, notifies the running
code, and causes it to set the mtimes on the file. It then rethrows the
signal so the pacman signal handler (or any frontend) works as expected.
4. We did a lot of funky stuff in trying to track the DB last modified time.
It is a lot easier to just keep the downloaded DB file around and track the
time on that rather than in a funky dot file. It also kills a lot of code.
5. For GPG verification of the databases down the road, we are going to need
the DB file around for at least a short bit of time anyway, so this gets us
closer to that.
Signed-off-by: Dan McGee <dan@archlinux.org>
[Xav: fixed printf with off_t]
Signed-off-by: Xavier Chantry <shiningxc@gmail.com>
2009-11-12 00:39:26 -05:00
|
|
|
ret = download(url, cachedir, 0);
|
2008-02-27 16:36:53 -05:00
|
|
|
if(ret == -1) {
|
2008-01-22 22:36:11 -05:00
|
|
|
_alpm_log(PM_LOG_WARNING, _("failed to download %s\n"), url);
|
|
|
|
return(NULL);
|
|
|
|
}
|
2008-01-22 23:49:45 -05:00
|
|
|
_alpm_log(PM_LOG_DEBUG, "successfully downloaded %s\n", url);
|
2008-01-22 22:36:11 -05:00
|
|
|
|
|
|
|
/* we should be able to find the file the second time around */
|
|
|
|
filepath = _alpm_filecache_find(filename);
|
|
|
|
return(filepath);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* vim: set ts=2 sw=2 noet: */
|