CryoPID - A Process Freezer for Linux

CryoPID allows you to capture the state of a running process in Linux and save it to a file. This file can then be used to resume the process later on, either after a reboot or even on another machine.

Status

CryoPID was spawned out of a discussion on the Software suspend mailing list about the complexities of suspending and resuming individual processes.

CryoPID consists of a program called freeze that captures the state of a running process and writes it into a file. The file is self-executing and self-extracting, so to resume a process, you simply run that file. See the table below for more details on what is supported.

Features

Current features are:

Can run as an ordinary user! (no root privileges needed)
Works on both 2.4 and 2.6.
Works on x86 and AMD64.
Can start & stop a process multiple times
Can migrate processes between machines and between kernel versions (tested between 2.4 to 2.6 and 2.6 to 2.4).

Yes, checkpointing software such as this is nothing new, but I was unable to find a tool that worked on Linux that did not require at least one of the following:

root privileges
modifications to the kernel
recompiling/relinking your software
using an LD_PRELOAD when you start your program.

CryoPID requires none of these and hence can be used as an ordinary user on any Linux machine running a 2.4 or 2.6 kernel (2.2 untested). This will remain one of the main design goals of CryoPID.

Things that do/don't work

Legend:

Working

Partial

Hacky

Planned

Wontfix

Full image of executable	Yes
Dynamically-linked Libraries	Yes. Can be included in image to resume on a machine with differing libraries
mmap'd memory	Yes. Both mmap'd files and anonymous mappings can be saved in the image.
Terminal status	Yes. termios state is captured and restored.
Signal handlers	Yes.
Thread-local storage	Yes. (Note this does not imply threads. It was required for working with recent glibc's on 2.6 kernels, which setup thread-local storage for standalone processes).
File descriptors	Open files and offsets are restored. Temporary files that have been unlinked and are not accessible on the filesystem are always saved in the image. Other files that do not exist on resume are not yet restored. Support for saving file contents for such situations is planned.
Sockets	Established TCP connections can be restored if the kernel supports tcpcp. In future versions, listening sockets will be rebound, and UDP and UNIX sockets will be reconnected.
X applications	Support is experimental. Enable USE_GTK=y in the Makefile. Most bugs with resuming Gtk+ applications require fixes from either Gtk+ or the application itself.
Multiple processes with hierarchy	Not yet. Planned.
PTYs other than the current one	Would be needed to suspend/resume a screen session with multiple PTYs inside.
Process name in ps (/proc/pid/cmdline)	Planned. Currently gets garbled upon resume. Without help from the kernel, it involves tweaking the environment and command-line arguments and re-exec'ing to make task->arg_start land at the same place.
Compression	GZIP compression is used by default. LZO compression is supported. bzip2 compression might be planned, and a nice way to switch between them.
Processes using 0x00100000	The resumer sits at 0x00100000. If another process uses this area of virtual memory, it can not be frozen. CryoPID's location could be chosen dynamically, but not until a need arises.

Use Case Examples

So when would you want to use CryoPID? Here are some programs and cases where it does work:

irssi - you run irssi under a screen on a shell server somewhere. You realise the server is going to reboot. You don't want to lose your scrollback. You suspend irssi before the server shuts down. Once it's back, you resume irssi and have all your channels and scrollback still. irssi is smart enough to automatically reconnect to your IRC servers.
vim/nano - (convoluted) you want to point somebody at at particular point in a file, or show off some macro you recorded, or something. Save your vim session, send it to another machine, resume.
mutt - You have a huge inbox with 30000 emails that you don't want to have to close mutt and reopen to check your email. You also don't want to leave it running in the interests of memory consumption. Ensure you sync your mutt before saving it to a file, and the just resume it later. Et voila.

Some cases where it would be nice, and will hopefully be supported sometime:

screen - your shell server is going down. You want to preserve your entire screen session, with child processes and all across a reboot. Support for hierarchies of processes needed.
X applications - you want to save your X application while you restart the X server (to change to a projector?), or reboot. This could be done with the help of xmove (needs to be integrated into CryoPID).

Other Known Issues

PAX must be turned off if you have it (chpax -xperms <target program>)
Very undocumented.

Downloading it

So by this point you're probably saying, "Yeah right, that's impossible". Show me.

Latest version:

Older versions:

Mailing List

A mailing list for the discussion of CryoPID has been setup. Click here to go to the subscription page.