linux-kernel - [PATCH] lib: proportion: fix underflow in prop_norm

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 14 Dec 2007 17:01:26 +0100
From:	Peter Zijlstra <a.p.zijlstra@...llo.nl>
To:	zhejiang <zhe.jiang@...el.com>
Cc:	linux-kernel@...r.kernel.org, yanmin_zhang@...ux.intel.com,
	Andrew Morton <akpm@...ux-foundation.org>
Subject: [PATCH] lib: proportion: fix underflow in prop_norm_percpu()

Subject: lib: proportion: fix underflow in prop_norm_percpu()

Zhe Jiang noticed that its possible to underflow pl->events in
prop_norm_percpu() when the value returned by percpu_counter_read() is less
than the error on that read and the period delay > 1. In that case half might
not trigger the batch increment and the value will be identical on the next
iteration, causing the same half to be subtracted again and again.

Fix this by rewriting the division as a single subtraction instead of a
subtraction loop and using percpu_counter_sum() when the value returned
by percpu_counter_read() is smaller than the error.

The latter is still needed if we want pl->events to shrink properly in the
error region.

Jiang, can I get a Reviewed-by from you? - if you agree that is :-)

Signed-off-by: Peter Zijlstra <a.p.zijlstra@...llo.nl>
Cc: zhejiang <zhe.jiang@...el.com>
---
 lib/proportions.c |   36 +++++++++++++++---------------------
 1 file changed, 15 insertions(+), 21 deletions(-)

Index: linux-2.6/lib/proportions.c
===================================================================
--- linux-2.6.orig/lib/proportions.c
+++ linux-2.6/lib/proportions.c
@@ -190,6 +190,8 @@ prop_adjust_shift(int *pl_shift, unsigne
  * PERCPU
  */
 
+#define PROP_BATCH (8*(1+ilog2(nr_cpu_ids)))
+
 int prop_local_init_percpu(struct prop_local_percpu *pl)
 {
 	spin_lock_init(&pl->lock);
@@ -230,31 +232,23 @@ void prop_norm_percpu(struct prop_global
 
 	spin_lock_irqsave(&pl->lock, flags);
 	prop_adjust_shift(&pl->shift, &pl->period, pg->shift);
+
 	/*
 	 * For each missed period, we half the local counter.
 	 * basically:
 	 *   pl->events >> (global_period - pl->period);
-	 *
-	 * but since the distributed nature of percpu counters make division
-	 * rather hard, use a regular subtraction loop. This is safe, because
-	 * the events will only every be incremented, hence the subtraction
-	 * can never result in a negative number.
 	 */
-	while (pl->period != global_period) {
-		unsigned long val = percpu_counter_read(&pl->events);
-		unsigned long half = (val + 1) >> 1;
-
-		/*
-		 * Half of zero won't be much less, break out.
-		 * This limits the loop to shift iterations, even
-		 * if we missed a million.
-		 */
-		if (!val)
-			break;
-
-		percpu_counter_add(&pl->events, -half);
-		pl->period += period;
-	}
+	period = (global_period - pl->period) >> (pg->shift - 1);
+	if (period < BITS_PER_LONG) {
+		s64 val = percpu_counter_read(&pl->events);
+
+		if (val < (nr_cpu_ids * PROP_BATCH))
+			val = percpu_counter_sum(&pl->events);
+
+		__percpu_counter_add(&pl->events, -val + (val >> period), PROP_BATCH);
+	} else
+		percpu_counter_set(&pl->events, 0);
+
 	pl->period = global_period;
 	spin_unlock_irqrestore(&pl->lock, flags);
 }
@@ -267,7 +261,7 @@ void __prop_inc_percpu(struct prop_descr
 	struct prop_global *pg = prop_get_global(pd);
 
 	prop_norm_percpu(pg, pl);
-	percpu_counter_add(&pl->events, 1);
+	__percpu_counter_add(&pl->events, 1, PROP_BATCH);
 	percpu_counter_add(&pg->events, 1);
 	prop_put_global(pd, pg);
 }


Download attachment "signature.asc" of type "application/pgp-signature" (190 bytes)