On 30/10/2023 14:34, Wolfgang Bumiller wrote:
On Tue, Oct 24, 2023 at 02:55:53PM +0200, Filip Schauer wrote:
Add a dev[n] argument to the container config to pass devices through to
a container. A device can be passed by its path. Alternatively a mapped
USB device can be passed through with usbmapping=<name>.
Signed-off-by: Filip Schauer<f.scha...@proxmox.com>
---
src/PVE/LXC.pm | 34 +++++++++++++++++++++++-
src/PVE/LXC/Config.pm | 60 +++++++++++++++++++++++++++++++++++++++++++
2 files changed, 93 insertions(+), 1 deletion(-)
diff --git a/src/PVE/LXC.pm b/src/PVE/LXC.pm
index c9b5ba7..a3ddb62 100644
--- a/src/PVE/LXC.pm
+++ b/src/PVE/LXC.pm
@@ -5,7 +5,8 @@ use warnings;
use Cwd qw();
use Errno qw(ELOOP ENOTDIR EROFS ECONNREFUSED EEXIST);
-use Fcntl qw(O_RDONLY O_WRONLY O_NOFOLLOW O_DIRECTORY);
+use Fcntl qw(O_RDONLY O_WRONLY O_NOFOLLOW O_DIRECTORY :mode);
+use File::Basename;
use File::Path;
use File::Spec;
use IO::Poll qw(POLLIN POLLHUP);
@@ -639,6 +640,37 @@ sub update_lxc_config {
$raw .= "lxc.mount.auto = sys:mixed\n";
}
+ # Clear passthrough directory from previous run
+ my $passthrough_dir = "/var/lib/lxc/$vmid/passthrough";
+ File::Path::rmtree($passthrough_dir);
I think we need to make a few changes here.
First: we don't necessarily need this directory.
Having a device list would certainly be nice, but it makes more sense to
just have a file we can easily parse (possibly even just a json hash),
like the `devices` file we already create in the pre-start hook, except
prepared *for* the pre-start hook, which *should* be able to just
`mknod` the devices right into the container's `/dev` on startup.
Devices mknoded into the container's /dev directory in the pre-start
hook will not be visible in the container once it is fully started.
Meanwhile mknoding a device to a different path inside the container
works fine. It seems that LXC mounts over the /dev directory. This can
be solved by calling mknod in lxc-pve-autodev-hook, but this does not
work with unprivileged containers without the mknod capability.
So are bind mounts our only option without modifying LXC,
or am I overlooking something?
We'd also avoid "lingering" device nodes with potentially harmful
uid/permissions in /var, which is certainly better from a security POV.
But note that we do need the `lxc.cgroup2.*` entries before starting the
container in order to ensure the devices cgroup has the right
permissions.
+
+ PVE::LXC::Config->foreach_passthrough_device($conf, sub {
+ my ($key, $sanitized_path) = @_;
+
+ my $absolute_path = "/$sanitized_path";
+ my ($mode, $rdev) = (stat($absolute_path))[2, 6];
+ die "Could not find major and minor ids of device $absolute_path.\n"
+ unless ($mode && $rdev);
+
+ my $major = PVE::Tools::dev_t_major($rdev);
+ my $minor = PVE::Tools::dev_t_minor($rdev);
+ my $device_type_char = S_ISBLK($mode) ? 'b' : 'c';
+ my $passthrough_device_path = "$passthrough_dir/$sanitized_path";
+ File::Path::make_path(dirname($passthrough_device_path));
+ PVE::Tools::run_command([
+ '/usr/bin/mknod',
+ '-m', '0660',
+ $passthrough_device_path,
+ $device_type_char,
+ $major,
+ $minor
+ ]);
It's probably worth adding a helper for the mknod syscall to
`PVE::Tools`, there are a bunch of syscalls in there already.
+ chown 100000, 100000, $passthrough_device_path if ($unprivileged);
^ This isn't necessarily the correct id. Users may have custom id
mappings.
`PVE::LXC::parse_id_maps($conf)` returns the mapping alongside the root
uid and gid. (See for example `sub mount_all` for how it's used.
+
+ $raw .= "lxc.cgroup2.devices.allow = $device_type_char $major:$minor
rw\n";
+ $raw .= "lxc.mount.entry = $passthrough_device_path $sanitized_path none
bind,create=file\n";
+ });
+
# WARNING: DO NOT REMOVE this without making sure that loop device nodes
# cannot be exposed to the container with r/w access (cgroup perms).
# When this is enabled mounts will still remain in the monitor's namespace
diff --git a/src/PVE/LXC/Config.pm b/src/PVE/LXC/Config.pm
index 56e1f10..edd813e 100644
--- a/src/PVE/LXC/Config.pm
+++ b/src/PVE/LXC/Config.pm
@@ -29,6 +29,7 @@ mkdir $lockdir;
mkdir "/etc/pve/nodes/$nodename/lxc";
my $MAX_MOUNT_POINTS = 256;
my $MAX_UNUSED_DISKS = $MAX_MOUNT_POINTS;
+my $MAX_DEVICES = 256;
# BEGIN implemented abstract methods from PVE::AbstractConfig
@@ -908,6 +909,49 @@ for (my $i = 0; $i < $MAX_UNUSED_DISKS; $i++) {
}
}
+PVE::JSONSchema::register_format('pve-lxc-dev-string', \&verify_lxc_dev_string);
+sub verify_lxc_dev_string {
+ my ($dev, $noerr) = @_;
+
+ if (
+ $dev =~m@/\.\.?/@ ||
+ $dev =~m@/\.\.?$@ ||
+ $dev !~ m!^/dev/!
+ ) {
+ return undef if $noerr;
+ die "$dev is not a valid device path\n";
+ }
+
+ return $dev;
+}
+
+my $dev_desc = {
+ path => {
+ optional => 1,
+ type => 'string',
+ default_key => 1,
+ format => 'pve-lxc-dev-string',
+ format_description => 'Path',
+ description => 'Device to pass through to the container',
+ verbose_description => 'Path to the device to pass through to the
container'
+ },
+ usbmapping => {
+ optional => 1,
+ type => 'string',
+ format => 'pve-configid',
+ format_description => 'mapping-id',
+ description => 'The ID of a cluster wide USB mapping.'
+ }
+};
+
+for (my $i = 0; $i < $MAX_DEVICES; $i++) {
+ $confdesc->{"dev$i"} = {
+ optional => 1,
+ type => 'string', format => $dev_desc,
+ description => "Device to pass through to the container",
+ }
+}
+
sub parse_pct_config {
my ($filename, $raw, $strict) = @_;
@@ -1255,6 +1299,22 @@ sub parse_volume {
return;
}
+sub parse_device {
+ my ($class, $device_string, $noerr) = @_;
+
+ my $res;
+ eval { $res = PVE::JSONSchema::parse_property_string($dev_desc,
$device_string) };
+ if ($@) {
+ return undef if $noerr;
+ die $@;
+ }
+
+ die "Either path or usbmapping has to be defined"
+ unless (defined($res->{path}) || defined($res->{usbmapping}));
+
+ return $res;
+}
+
sub print_volume {
my ($class, $key, $volume) = @_;
--
2.39.2
_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
