cgroup overhaul

[BarryLhm]

This closes #45.
I'm inexperienced in programming,
so please do ensure all changes are tested to work correctly before merging!!!

[Moe-hacker]

Should we add the file w to .gitignore, or give it a clear name?

[Moe-hacker]

Should we add the file w to .gitignore, or give it a clear name?

[BarryLhm]

I haven't come up with a clear name.
"w" is for "wrapper" like in "gradlew" or "wand"
what about "r" or "rr"? 🤔
or a formal long name?

[Moe-hacker]

Maybe we can use make/just command instead it.

[Moe-hacker]

On debian 13 aarch64, clang got error like

../src/cgroup.c:57:33: error: use of undeclared identifier 'LONG_MAX'                                          57 |                 case 'g': case 'G': if (ret > SSIZE_MAX / 1024 / 1024 / 1024) return -2; ret *= 1024 * 1024 * 1024; break;                                      |                                               ^                                                     /usr/include/aarch64-linux-gnu/bits/posix1_lim.h:169:21: note: expanded from macro 'SSIZE_MAX'                169 | #  define SSIZE_MAX     LONG_MAX                    |                         ^

So we need a #include <limits.h>

[Moe-hacker]

If we re-run the container, seems that cgroup limit will not work properly, did you enforced to attach the process to correct cgroup?

[BarryLhm]

On debian 13 aarch64, clang got error like

../src/cgroup.c:57:33: error: use of undeclared identifier 'LONG_MAX'                                          57 |                 case 'g': case 'G': if (ret > SSIZE_MAX / 1024 / 1024 / 1024) return -2; ret *= 1024 * 1024 * 1024; break;                                      |                                               ^                                                     /usr/include/aarch64-linux-gnu/bits/posix1_lim.h:169:21: note: expanded from macro 'SSIZE_MAX'                169 | #  define SSIZE_MAX     LONG_MAX                    |                         ^

So we need a #include <limits.h>

iirc i added it, let me check..

maybe it disappeared after a checkout
fixed

[BarryLhm]

If we re-run the container, seems that cgroup limit will not work properly, did you enforced to attach the process to correct cgroup?

the original logic is not changed, which means i have to find the bug.

[BarryLhm]

Maybe we can use make/just command instead it.

they are hard to pass arguments to

[BarryLhm]

Maybe we can use make/just command instead it.

removed after i found make format

[Moe-hacker]

Cgroup limit should only be set once when init the container, after that, even no limitations set, we still need to attach process to the existing cgroup.

And for unshare container, maybe we should setup a new cgroup on host before unshare(), so that we can just use cgroupfs on host to kill it.