Re: [ckrm-tech] [patch 07/38] CKRM e18: Numtasks Controller

[Naoaki Maeda <maeda.naoaki@jp.fujitsu.com>]

Gerrit Huizenga wrote:

> Index: linux-2.6.12-ckrm1/kernel/fork.c
> ===================================================================
> --- linux-2.6.12-ckrm1.orig/kernel/fork.c	2005-06-20 13:08:27.000000000 -0700
> +++ linux-2.6.12-ckrm1/kernel/fork.c	2005-06-20 13:08:34.000000000 -0700
> @@ -42,6 +42,8 @@
>  #include <linux/rmap.h>
>  #include <linux/acct.h>
>  #include <linux/ckrm_events.h>
> +#include <linux/ckrm_tsk.h>
> +#include <linux/ckrm_tc.h>
>  
>  #include <asm/pgtable.h>
>  #include <asm/pgalloc.h>
> @@ -1211,6 +1213,9 @@ long do_fork(unsigned long clone_flags,
>  			clone_flags |= CLONE_PTRACE;
>  	}
>  
> +	if (numtasks_get_ref(&current->taskclass->core, 0) == 0) {
> +		return -ENOMEM;
> +	}
>  	p = copy_process(clone_flags, stack_start, regs, stack_size, parent_tidptr, child_tidptr, pid);
>  	/*
>  	 * Do this prior waking up the new thread - the thread pointer
> @@ -1250,6 +1255,7 @@ long do_fork(unsigned long clone_flags,
>  				ptrace_notify ((PTRACE_EVENT_VFORK_DONE << 8) | SIGTRAP);
>  		}
>  	} else {
> +		numtasks_put_ref(&current->taskclass->core);
>  		free_pidmap(pid);
>  		pid = PTR_ERR(p);
>  	}

Instead of returning -ENOMEM on fork fail due to numtask or forkrate
limit, I'd rather prefer returning -EAGAIN.

Because, according to IEEE Std 1003.1, if fork() fails due to
reaching limit, it shall set the errno to EAGAIN.

Thanks,
MAEDA Naoaki