snapshot: replace global sync with a namespace sync

snapshot_create() called did a global 'sync' after freeze()
which syncs everything including all other containers and
the host. So if you want to snapshot container A while
container B tries to write to a broken NFS mount the
snapshot will hang in that sync call.
Instead we now enter the container's mount namespace and do
a syncfs() on all of its mountpoints.

diff --git a/src/PVE/LXC.pm b/src/PVE/LXC.pm
index 0a6b451c38171826ea5dd35cd785bd3bc2109aa1..6545df57543a4e8f8a3978a4eefbd09aaff3ea98 100644 (file)
--- a/src/PVE/LXC.pm
+++ b/src/PVE/LXC.pm
@@ -4,10 +4,12 @@ use strict;
 use warnings;
 use POSIX qw(EINTR);
 
+use Socket;
+
 use File::Path;
 use File::Spec;
 use Cwd qw();
-use Fcntl ':flock';
+use Fcntl qw(O_RDONLY :flock);
 
 use PVE::Cluster qw(cfs_register_file cfs_read_file);
 use PVE::Storage;
@@ -1797,6 +1799,75 @@ sub has_feature {
     return $err ? 0 : 1;
 }
 
+my $enter_namespace = sub {
+    my ($vmid, $pid, $which, $type) = @_;
+    sysopen my $fd, "/proc/$pid/ns/$which", O_RDONLY
+       or die "failed to open $which namespace of container $vmid: $!\n";
+    PVE::Tools::setns(fileno($fd), $type)
+       or die "failed to enter $which namespace of container $vmid: $!\n";
+    close $fd;
+};
+
+my $do_syncfs = sub {
+    my ($vmid, $pid, $socket) = @_;
+
+    &$enter_namespace($vmid, $pid, 'mnt', PVE::Tools::CLONE_NEWNS);
+
+    # Tell the parent process to start reading our /proc/mounts
+    print {$socket} "go\n";
+    $socket->flush();
+
+    # Receive /proc/self/mounts
+    my $mountdata = do { local $/ = undef; <$socket> };
+    close $socket;
+
+    # Now sync all mountpoints...
+    my $mounts = PVE::ProcFSTools::parse_mounts($mountdata);
+    foreach my $mp (@$mounts) {
+       my ($what, $dir, $fs) = @$mp;
+       next if $fs eq 'fuse.lxcfs';
+       eval { PVE::Tools::sync_mountpoint($dir); };
+       warn $@ if $@;
+    }
+};
+
+sub sync_container_namespace {
+    my ($vmid) = @_;
+    my $pid = find_lxc_pid($vmid);
+
+    # SOCK_DGRAM is nicer for barriers but cannot be slurped
+    socketpair my $pfd, my $cfd, AF_UNIX, SOCK_STREAM, PF_UNSPEC
+       or die "failed to create socketpair: $!\n";
+
+    my $child = fork();
+    die "fork failed: $!\n" if !defined($child);
+
+    if (!$child) {
+       eval {
+           close $pfd;
+           &$do_syncfs($vmid, $pid, $cfd);
+       };
+       if (my $err = $@) {
+           warn $err;
+           POSIX::_exit(1);
+       }
+       POSIX::_exit(0);
+    }
+    close $cfd;
+    my $go = <$pfd>;
+    die "failed to enter container namespace\n" if $go ne "go\n";
+
+    open my $mounts, '<', "/proc/$child/mounts"
+       or die "failed to open container's /proc/mounts: $!\n";
+    my $mountdata = do { local $/ = undef; <$mounts> };
+    close $mounts;
+    print {$pfd} $mountdata;
+    close $pfd;
+
+    while (waitpid($child, 0) != $child) {}
+    die "failed to sync container namespace\n" if $? != 0;
+}
+
 sub snapshot_create {
     my ($vmid, $snapname, $comment) = @_;
 
@@ -1814,7 +1885,7 @@ sub snapshot_create {
        if ($running) {
            $unfreeze = 1;
            PVE::Tools::run_command(['/usr/bin/lxc-freeze', '-n', $vmid]);
-           PVE::Tools::run_command(['/bin/sync']);
+           sync_container_namespace($vmid);
        };
 
        my $storecfg = PVE::Storage::config();