[PATCH] fix rusage semantics

This patch changes the rusage bookkeeping and the semantics of the
getrusage and times calls in a couple of ways.

The first change is in the c* fields counting dead child processes.  POSIX
requires that children that have died be counted in these fields when they
are reaped by a wait* call, and that if they are never reaped (e.g.
because of ignoring SIGCHLD, or exitting yourself first) then they are
never counted.  These were counted in release_task for all threads.  I've
changed it so they are counted in wait_task_zombie, i.e.  exactly when
being reaped.

POSIX also specifies for RUSAGE_CHILDREN that the report include the reaped
child processes of the calling process, i.e.  whole thread group in Linux,
not just ones forked by the calling thread.  POSIX specifies tms_c[us]time
fields in the times call the same way.  I've moved the c* fields that
contain this information into signal_struct, where the single set of
counters accumulates data from any thread in the group that calls wait*.

Finally, POSIX specifies getrusage and times as returning cumulative totals
for the whole process (aka thread group), not just the calling thread.
I've added fields in signal_struct to accumulate the stats of detached
threads as they die.  The process stats are the sums of these records plus
the stats of remaining each live/zombie thread.  The times and getrusage
calls, and the internal uses for filling in wait4 results and siginfo_t,
now iterate over the threads in the thread group and sum up their stats
along with the stats recorded for threads already dead and gone.

I added a new value RUSAGE_GROUP (-3) for the getrusage system call rather
than changing the behavior of the old RUSAGE_SELF (0).  POSIX specifies
RUSAGE_SELF to mean all threads, so the glibc getrusage call will just
translate it to RUSAGE_GROUP for new kernels.  I did this thinking that
someone somewhere might want the old behavior with an old glibc and a new
kernel (it is only different if they are using CLONE_THREAD anyway). 
However, I've changed the times system call to conform to POSIX as well and
did not provide any backward compatibility there.  In that case there is
nothing easy like a parameter value to use, it would have to be a new
system call number.  That seems pretty pointless.  Given that, I wonder if
it is worth bothering to preserve the compatible RUSAGE_SELF behavior by
introducing RUSAGE_GROUP instead of just changing RUSAGE_SELF's meaning.
Comments?

I've done some basic testing on x86 and x86-64, and all the numbers come
out right after these fixes.  (I have a test program that shows a few

Signed-off-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

2 files changed, 14 insertions, 3 deletions

diff --git a/include/linux/resource.h b/include/linux/resource.h
index 21a86cb6acdb..3c848324df44 100644
--- a/include/linux/resource.h
+++ b/include/linux/resource.h
@@ -17,6 +17,7 @@
 #define	RUSAGE_SELF	0
 #define	RUSAGE_CHILDREN	(-1)
 #define RUSAGE_BOTH	(-2)		/* sys_wait4() uses this */
+#define RUSAGE_GROUP	(-3)		/* thread group sum + dead threads */
 
 struct	rusage {
 	struct timeval ru_utime;	/* user time used */
diff --git a/include/linux/sched.h b/include/linux/sched.h
index f89a32a4bf48..3aaf8e5f2216 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -301,6 +301,16 @@ struct signal_struct {
 	int leader;
 
 	struct tty_struct *tty; /* NULL if no tty */
+
+	/*
+	 * Cumulative resource counters for dead threads in the group,
+	 * and for reaped dead child processes forked by this group.
+	 * Live threads maintain their own counters and add to these
+	 * in __exit_signal, except for the group leader.
+	 */
+	unsigned long utime, stime, cutime, cstime;
+	unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw;
+	unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
 };
 
 /*
@@ -495,11 +505,11 @@ struct task_struct {
 	unsigned long it_real_value, it_prof_value, it_virt_value;
 	unsigned long it_real_incr, it_prof_incr, it_virt_incr;
 	struct timer_list real_timer;
-	unsigned long utime, stime, cutime, cstime;
-	unsigned long nvcsw, nivcsw, cnvcsw, cnivcsw; /* context switch counts */
+	unsigned long utime, stime;
+	unsigned long nvcsw, nivcsw; /* context switch counts */
 	u64 start_time;
 /* mm fault and swap info: this can arguably be seen as either mm-specific or thread-specific */
-	unsigned long min_flt, maj_flt, cmin_flt, cmaj_flt;
+	unsigned long min_flt, maj_flt;
 /* process credentials */
 	uid_t uid,euid,suid,fsuid;
 	gid_t gid,egid,sgid,fsgid;