Skip to contentSkip to conversation

sys_setrlimit() in 2.6.16

RFC Cliff Wickman 3 participants 3 messages Feb 14, 2006 – Feb 16, 2006

lore
Share & tools
lore
#0 lore raw

A test suite uncovered a possible bug in setrlimit(2), in 2.6.16-rc3.

A code that does

        mylimits.rlim_cur = 0;
        setrlimit(RLIMIT_CPU, &mylimits);

does not set a cpu time limit.

No signal is sent to this code when its "limit" of 0 seconds expires. Whereas, under the 2.6.5 kernel (SuSE SLESS9) a signal was sent.

I don't see any obvious difference in sys_setrlimit() or set_process_cpu_timer() between 2.6.5 and 2.6.16.

Is this a correction, or a bug?

Is a cpu time limit of 0 supposed to limit a task to 0 seconds, or leave it unlimited?

Signature · 4 lines
Cliff Wickman
Silicon Graphics, Inc.
[email protected]
(651) 683-3824
#1 lore raw
Quoted 11 lines · Cliff Wickman

A test suite uncovered a possible bug in setrlimit(2), in 2.6.16-rc3.

A code that does

        mylimits.rlim_cur = 0;
        setrlimit(RLIMIT_CPU, &mylimits);

does not set a cpu time limit.

No signal is sent to this code when its "limit" of 0 seconds expires. Whereas, under the 2.6.5 kernel (SuSE SLESS9) a signal was sent.

Quoted 8 lines

I don't see any obvious difference in sys_setrlimit() or set_process_cpu_timer() between 2.6.5 and 2.6.16.

Is this a correction, or a bug?

Is a cpu time limit of 0 supposed to limit a task to 0 seconds, or leave it unlimited?

This has to be considered a bug. The spec certainly implies that a limit of zero should be honoured and, probably more importantly, that's how it works in 2.4.

Problem is, the code in there all assumes that an it_prof_expires of zero means "it was never set", and changing that (add a yes-it-has flag?) would be less than trivial.

So I think the path of least resistance here is to just convert the caller's zero seconds into one second. That in fact gives the same behaviour as 2.4: you get whacked after one second or more CPU time.

(This is not a final patch - that revolting expression in sys_setrlimit() needs help first).

kernel/sys.c renamed from kernel/sys.c~a +13 -1
diff -puN kernel/sys.c~a kernel/sys.c
@@ -1657,7 +1657,19 @@ asmlinkage long sys_setrlimit(unsigned i
 	    (cputime_eq(current->signal->it_prof_expires, cputime_zero) ||
 	     new_rlim.rlim_cur <= cputime_to_secs(
 		     current->signal->it_prof_expires))) {
-		cputime_t cputime = secs_to_cputime(new_rlim.rlim_cur);
+		unsigned long rlim_cur = new_rlim.rlim_cur;
+		cputime_t cputime;
+
+		if (rlim_cur == 0) {
+			/*
+			 * The caller is asking for an immediate RLIMIT_CPU
+			 * expiry.  But we use the zero value to mean "it was
+			 * never set".  So let's cheat and make it one second
+			 * instead
+			 */
+			rlim_cur = 1;
+		}
+		cputime = secs_to_cputime(rlim_cur);
 		read_lock(&tasklist_lock);
 		spin_lock_irq(&current->sighand->siglock);
 		set_process_cpu_timer(current, CPUCLOCK_PROF,

_

Acked-by
#2 lore raw
Quoted 8 lines · Andrew Morton

This has to be considered a bug. The spec certainly implies that a limit of zero should be honoured and, probably more importantly, that's how it works in 2.4.

Problem is, the code in there all assumes that an it_prof_expires of zero means "it was never set", and changing that (add a yes-it-has flag?) would be less than trivial.

Quoted 6 lines

So I think the path of least resistance here is to just convert the caller's zero seconds into one second. That in fact gives the same behaviour as 2.4: you get whacked after one second or more CPU time.

(This is not a final patch - that revolting expression in sys_setrlimit() needs help first).

your approach looks good to me. It doesnt make much sense anyway to have a task whacked right after startup ... so adding a common-sense "the user must have meant some really small value" thing doesnt look all that wrong.

Ingo