开启方法

PVE创建虚拟机后默认只有novnc方法连接虚拟机，无法复制粘贴，很不方便，因此可以打开xterm.js的方式弥补此缺陷

关闭 VM，并且新增 serial port 給 VM 在PVE Host 中用 qm 命令建立 serial port，假設我的 VM ID 是 100

qm set 100 -serial0 socket

重新开启 VM，用 dmesg 验证是否有 ttyS 出现

dmesg | grep ttyS

如果出現如下代表有了

[    3.521757] 00:04: ttyS0 at I/O 0x3f8 (irq = 4, base_baud = 115200) is a 16550A

接下来配置VM的grub，进入VM的ssh

cd /etc/default/
vi grub

修改grub中的GRUB_CMDLINE_LINUX值

GRUB_CMDLINE_LINUX="quiet console=tty0 console=ttyS0,115200”

更新grub

debian based
update-grub
redhat based
grub2-mkconfig --output=/boot/grub2/grub.cfg

在虚拟机添加

/etc/init/ttyS0.conf

并添加如下

# ttyS0 - getty
#
# This service maintains a getty on ttyS0 from the point the system is
# started until it is shut down again.
start on stopped rc RUNLEVEL=[12345]
stop on runlevel [!12345]
respawn
exec /sbin/getty -L 115200 ttyS0 vt102

设置自启动

systemctl enable serial-getty@ttyS0.service
systemctl start serial-getty@ttyS0.service

重启VM，测试xterm，已经可以用

官方文档

Why would I need a serial terminal / serial console?

Note: This apply to qemu/kvm virtualization.

If you do a lot of work over ssh on your pve server you would maybe appreciate to connect to your running VMs like:

qm status 101
status: running
qm terminal 101
starting serial terminal on interface serial0 (press control-O to exit)
Debian GNU/Linux 8 debian8 ttyS0
debian8 login:

Other use cases:

  • you lost network access to the guest and VNC is either too slow for you or does not have the features you need (i.e. easy copy/paste between other terminals)
  • your guest freezes or kernel panics, you want to debug it, but it is impossible to capture all messages on VNC screen
  • your keyboard layout is borked on the guest
  • you are familiar with xm console <guest> from Xen and you want to use a similar feature here

Some background so you get the idea

When you start Unix/Linux on a PC, the default input device is the attached keyboard (PS/2 or USB), and the default output device is the available VGA /HDMI / Display port of the PC. This is called in Unix-speak the default console, or system console. During boot, the kernel sends its boot messages, like device detection, to this default console, and at the end of the boot processes, fires a “login: “ prompt on this console. (Actually it spawns multiple login prompts, you can switch between them with Ctrl-alt-F1, Ctrl-alt-F2, etc … )

Now it is perfectly fine to send the boot messages and start a login prompt on something else. For instance the Linux Kernel has a netconsole feature, to send the boot messages over the network to another Linux computer. It is also possible to send a copy of the boot messages, and start a login prompt on the serial port of the PC. We get then what is called a serial console If you connect to this serial port any computing device with a terminal emulation program, it will be possible to work on the Linux PC as if you were using the PC locally in text mode. The requirements of the terminal emulation are quite low, so you could use HyperTerminal on a Windows PC, Minicom on Linux or even an old Atari ST with the appropriate program.

How do this apply to Proxmox PVE ?

In proxmox things work exactly the same, but with emulated devices. NoVNC/ VNC Applet connects to your VM keyboard and VM VGA display and displays the stuff you would expect from a real PC on a VGA display with a locally attached keyboard. Now we can also configure our VM to have an emulated serial port, and instruct the OS, to send a copy of the boot messages / start a login prompt on the emulated serial port. Once this is configured it will be possible to connect from the host (ie the server running pve) using a terminal emulation program, which is handily builtin in the Proxmox Qemu Manager.

Configuration on the host

Add a virtual serial port to the VM
Provided your VM IS is 101

open  /etc/pve/qemu-server/101.conf and add the following parameter at the end of file
serial0: socket

or use

qm set 101 -serial0 socket

Configuration on the guest

Configure the terminal if not present

As detailed in https://help.ubuntu.com/community/SerialConsoleHowto#Server_Setup , check if you have the file /etc/init/ttyS0.conf. If not, create it:

# ttyS0 - getty**
# This service maintains a getty on ttyS0 from the point the system is**
# started until it is shut down again.
start on stopped rc RUNLEVEL=[12345]
stop on runlevel [!12345]
respawn
exec /sbin/getty -L 115200 ttyS0 vt102

then ask upstart to start terminal

sudo start ttyS0

You might consider creating the ttyS1.conf file as well, just as a backup in case you have a crash with ttyS0

Reboot the VM, verify that the emulated serial port is there
dmesg | grep ttyS
[ 0.457814] 00:0a: ttyS0 at I/O 0x3f8 (irq = 4) is a 16550A
If ttyS0 is not there, typesudo start ttyS0from the VM shell
Instruct grub2 to send the boot messages on the VGA display and on the serial port

#in /etc/default/grub change the GRUB_CMDLINE_LINUX parameter to
GRUB_CMDLINE_LINUX="quiet console=tty0 console=ttyS0,115200"

run

#debian based
#redhat based
grub2-mkconfig --output=/boot/grub2/grub.cfg

and do a final reboot

Connecting to the Serial Terminal

On the Host, just enter：qm terminal <VMiD>
and enter enter a second time you should get a login prompt
Note: if it seems this is not working, and if you have defined ttyS1, you can connect to itwith the command：qm terminal <VMiD> -iface serial1

Now if you reboot your system you should see that the kernel startup messages are send to both your serial terminal and NoVNC display.

Additionally (quoting from a private email, edited for clarity):
Qemu also supports binding the emulated serial port to a telnet server. Look for the -chardev parameter in the qemu-kvm manual page. (With proxmox you would then use the “args:” parameter in vm.conf to specify extra args.) This would allow you to telnet to your host on a specific port to get a serial console. As with every telnet traffic remember that nothing is encrypted, so I wouldn’t do this in a production setup.