Add kern.racct.enable tunable and RACCT_DISABLED config option.

[trasz]

Summary

Add kern.racct.enable tunable and RACCT_DISABLED config option. The point of this is to be able to add RACCT (with RACCT_DISABLED) to GENERIC, to avoid having to rebuild the kernel to use rctl(8).

[kib]

sys/kern/kern_racct.c

323 ↗

(On Diff #5002)

(Random place). I do not see much value in these asserts, are you trying to catch missed #ifdef RACCT braces ?

If keeping them around, I suggest making the whole statement a macros, like ASSERT_RACCT_ENABLED();

sys/vm/vm_pageout.c

1792 ↗

(On Diff #5002)

Better put only the last argument under the #ifdef, otherwise it is quite confusing.

[trasz]

Add ASSERT_RACCT_ENABLED() and reword the conditional in vm_pageout.c.

[op]

head/sys/kern/kern_racct.c

461

This is useless, if my theory not wrong.

489

Same here.

984

This does not cause memory leak?

The steps in theory:

1. enable the resource accounting
2. start a random process
3. disable resource accounting before the process exit
4. wait to the process exit or kill the process

Before this change, the racct_destroy() was always gets running.

[op]

head/sys/kern/kern_racct.c

489

This expression should be against "*racct == NULL" instead of "racct_enable".

[trasz]

head/sys/kern/kern_racct.c

984

The step #3 cannot ever happen; racct_enable is a tunable, not a sysctl - for this and several others, quite fundamental, reasons.

[op]

head/sys/kern/kern_racct.c

81

Ahh, I see now. That's a RO sysctl.
Could you please add a comment here, to explain, that this never should be RW?

984

Fine. I missed that the sysctl is RO only.