[CRIU] Dumping a 32-bit process on aarch64 kernel

Rayson Ho raysonlogin at gmail.com
Fri Dec 8 16:13:44 MSK 2017


On Fri, Dec 8, 2017 at 7:44 AM, Dmitry Safonov <0x7f454c46 at gmail.com> wrote:
> Well, you can try.. quite sure no one did it before you.
> But for production, you'll need 32-bit support in criu.

Yup, for my use case I just need to checkpoint & restore 32-bit
processes, so a 32-bit criu is good enough for me!

However, I am still debugging the following failure -- as soon as criu
hits the compel code, then it hits the error:

(00.274647) ----------------------------------------
(00.274650)
(00.274653) Collecting fds (pid: 26575)
(00.274656) ----------------------------------------
(00.274678) Found 3 file descriptors
(00.274681) ----------------------------------------
(00.274692) Dump private signals of 26575
(00.274702) Dump shared signals of 26575
(00.274712) Parasite syscall_ip at 0x8000
(00.274839) Error (compel/src/lib/infect.c:773): Can't rmap memfd for
parasite blob
(00.274890) Error (criu/cr-dump.c:1265): Can't infect (pid: 26575) with parasite
(00.275005) Unlock network
(00.275012) Unfreezing tasks into 1
(00.275015)     Unseizing 26575 into 1
(00.275026) Error (criu/cr-dump.c:1654): Dumping FAILED.

So the dumpee fails to execute the mmap that is injected by
remote_mmap, and I am still trying to find an easier way to debug this
than doing debug printfs inside the compel code... Kind of hard to
debug the dumpee as ptrace can't be used (so no strace or gdb for
me!).

Rayson

==================================================
Open Grid Scheduler - The Official Open Source Grid Engine
http://gridscheduler.sourceforge.net/
http://gridscheduler.sourceforge.net/GridEngine/GridEngineCloud.html





> Imagine, you have 32-bit bash that calls some helpers
> like sed/awk/tail/anything. That way you'll have this
> tree of:
>
> 64-bit parent
>        |
>     bash (32-bit)
>      /           \
>  sed (64)   ...
>
> So, you might found that you anyway need 32-bit C/R
> support in criu in case of such mixed-bitness trees.
>
>>
>> BTW, thanks for the replies, Dmitry & Andrei!
>>
>
> --
>              Dmitry


More information about the CRIU mailing list