CRIU unable to dump a process which is monitored by userfaultfd

[itayslepian]

Description

We are trying to do migrations with CRIU to a process which is also monitored by Userfaultfd. We are running on Ubuntu18.04, linux kernel 5.15 When dumping the pid, we receive the following error:

(00.026780) Error (criu/files-ext.c:94): Can't dump file 3 of that type [600] (anon anon_inode:[userfaultfd])

Steps to reproduce the issue:

1. Run userfaultfd_demo.c from https://man7.org/linux/man-pages/man2/userfaultfd.2.html#EXAMPLE
2. Dump the pid of by using CRIU through terminal: sudo criu  dump -vvvv -o dump.log -t --shell-job && echo OK

CRIU logs and information:

CRIU full dump/restore logs:

Attached is a file named LOG.txt ``` ```

Output of `criu --version`:

``` Version: 3.17 GitID: v3.17-117-g50db2be1a ```

Output of `criu check --all`:

``` Warn (criu/cr-check.c:1347): Nftables based locking requires libnftables and set concatenations support Looks good but some kernel features are missing which, depending on your process tree, may cause dump or restore failure. ```

LOG.txt

[adrianreber]

Thanks for opening. At this point I am not aware of anyone working on this, so right now we cannot really help you unless you have an idea how to implement this in CRIU.

[mihalicyn]

That's a pretty interesting but really hard-to-implement feature.

I'm not an expert in memory management and uffd but as I can see we need:

• to get a list of all VMAs which is watchable by the uffd (kernel modification is needed to implement such API?)
• some way to restore pending uffd events
  • simulate them during restore? (hard to implement, not reliable)
  • provide kernel API to push them directly? (kernel modification is needed)
  • wait until all pending events will be processed with the hope that no new ones will be added to a queue during waiting? (bad way, IMHO)