Re: [pve-devel] [PATCH qemu-server 14/16] introduce QSD module for qemu-storage-daemon functionality

[Fiona Ebner <f.ebner@proxmox.com>]

Am 17.10.25 um 3:08 PM schrieb Daniel Kral:
> On Tue Oct 14, 2025 at 4:39 PM CEST, Fiona Ebner wrote:
>> diff --git a/src/PVE/QemuServer/Helpers.pm b/src/PVE/QemuServer/Helpers.pm
>> index 2c78a7b4..ab8aa389 100644
>> --- a/src/PVE/QemuServer/Helpers.pm
>> +++ b/src/PVE/QemuServer/Helpers.pm
>> @@ -89,6 +89,24 @@ sub qsd_pidfile_name {
>>      return "${var_run_tmpdir}/qsd-${vmid}.pid";
>>  }
>>  
>> +sub qsd_fuse_export_cleanup_files {
>> +    my ($vmid) = @_;
>> +
>> +    PVE::Tools::dir_glob_foreach(
>> +        $var_run_tmpdir,
>> +        "qsd-${vmid}-.*.fuse",
>> +        sub {
>> +            my ($file) = @_;
>> +            unlink "${var_run_tmpdir}/${file}";
> 
> This could fail if the exported fuse fs wasn't cleaned up by qsd
> correctly, see more below.
> 
>> +        },
>> +    );
>> +}
>> +
>> +sub qsd_fuse_export_path {
>> +    my ($vmid, $export_name) = @_;
>> +    return "${var_run_tmpdir}/qsd-${vmid}-${export_name}.fuse";
>> +}
>> +
> 
> --- [ snip ] ---
> 
>> +=head3 quit
>> +
>> +    PVE::QemuServer::QSD::quit($vmid);
>> +
>> +Shut down the QEMU storage daemon associated to VM C<$vmid> and cleans up its PID file and socket.
>> +Waits for 60 seconds for clean shutdown, then sends SIGTERM and waits an additional 10 seconds
>> +before sending SIGKILL.
>> +
>> +=cut
>> +
>> +sub quit($vmid) {
>> +    eval { PVE::QemuServer::Monitor::qsd_cmd($vmid, 'quit'); };
>> +    my $qmp_err = $@;
>> +    warn "QEMU storage daemon for $vmid failed to handle 'quit' - $qmp_err" if $qmp_err;
>> +
>> +    my $count = $qmp_err ? 60 : 0; # can't wait for QMP 'quit' to terminate the process if it failed
>> +    my $pid = PVE::QemuServer::Helpers::qsd_running_locally($vmid);
>> +    while ($pid) {
>> +        if ($count == 60) {
>> +            warn "QEMU storage daemon for $vmid still running with PID $pid"
>> +                . " - terminating now with SIGTERM\n";
>> +            kill 15, $pid;
>> +        } elsif ($count == 70) {
>> +            warn "QEMU storage daemon for $vmid still running with PID $pid"
>> +                . " - terminating now with SIGKILL\n";
>> +            kill 9, $pid;
>> +            last;
> 
> SIGKILL will leave the exported fuse export in an unremovable state
> since the qsd couldn't clean it up. I could only reproduce this by doing
> it manually with `kill -SIGKILL ...`, which resulted in the VM not being
> startable anymore.
> 
> Might be worth to add an `umount /run/qemu-server/qsd-....fuse` if
> unlink fails or if we can stat -f that the file is still the
> FUSE-exported image?

Good catch! Yeah, I think trying an unmount is a good approach.


_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel