Skip to contentSkip to conversation

futex_wake04: avoid tst_ts_from_ns overflow on 32-bit platforms

PATCH New Thadeu Lima de Souza Cascardo 2 participants 6 messages Jun 8, 2021 – Jun 9, 2021

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 futex_wake04: avoid tst_ts_from_ns overflow on 32-bit platforms

When multiplying 30 with NSEC_PER_SEC, the result would overflow on 32-bit platforms, unless cast to long long, which is what tst_ts_from_ns expects.

Though we could change NSEC_PER_SEC to be a long long constant, it was considered a risk for regressions, as it would affect every use of it.

After this change, futex_wake04 passes on i386.

---
 testcases/kernel/syscalls/futex/futex_wake04.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
testcases/kernel/syscalls/futex/futex_wake04.c +1 -1
index 2260a3936d6e..479001e2e59a 100644
@@ -56,7 +56,7 @@ static void setup(void)
 	tst_res(TINFO, "Testing variant: %s", tv->desc);
 	futex_supported_by_kernel(tv->fntype);
 
-	to = tst_ts_from_ns(tv->tstype, 30 * NSEC_PER_SEC);
+	to = tst_ts_from_ns(tv->tstype, (long long) 30 * NSEC_PER_SEC);
 }
 
 static void *wait_thread1(void *arg LTP_ATTRIBUTE_UNUSED)
Signature · 1 line
2.30.2
#1 lore raw

PATCH 1/1 futex_wake04: avoid tst_ts_from_ns overflow on 32-bit platforms

Hi!
Good catch, I guess that it would be slightly cleaner to fix the NSEC_PER_SEC size instead with:

include/lapi/common_timers.h +1 -1
index b783beff4..e50f698d6 100644
@@ -12,7 +12,7 @@
 #include "lapi/posix_clocks.h"
 
 #ifndef NSEC_PER_SEC
-#define NSEC_PER_SEC (1000000000L)
+#define NSEC_PER_SEC (1000000000LL)
 #endif
 
 static const clock_t clock_list[] = {

What do you think?

Signature · 2 lines
Cyril Hrubis
[email protected]
#2 lore raw

PATCH 1/1 futex_wake04: avoid tst_ts_from_ns overflow on 32-bit platforms

Quoted 19 lines · Cyril Hrubis

Hi!
Good catch, I guess that it would be slightly cleaner to fix the NSEC_PER_SEC size instead with:

include/lapi/common_timers.h +1 -1
index b783beff4..e50f698d6 100644
@@ -12,7 +12,7 @@
 #include "lapi/posix_clocks.h"
 
 #ifndef NSEC_PER_SEC
-#define NSEC_PER_SEC (1000000000L)
+#define NSEC_PER_SEC (1000000000LL)
 #endif
 
 static const clock_t clock_list[] = {
Quoted 5 lines

What do you think?

Signature · 2 lines
Cyril Hrubis
[email protected]

Yeah, I even wrote and tested that it would build, but I wasn't able to go through any build logs or run complete tests.

But maybe I am just overthinking this, and any promotions due to this would only fix issues and not introduce any new ones.

Well, on second thought, on 32-bit, in cases like variadic functions, it could cause a problem. So, I am just afraid that applying this fix without looking for every use of NSEC_PER_SEC might lead to new bugs.

Cascardo.

#3 lore raw

PATCH 1/1 futex_wake04: avoid tst_ts_from_ns overflow on 32-bit platforms

Hi!

Quoted 27 lines
Quoted 18 lines

Good catch, I guess that it would be slightly cleaner to fix the NSEC_PER_SEC size instead with:

include/lapi/common_timers.h +1 -1
index b783beff4..e50f698d6 100644
@@ -12,7 +12,7 @@
 #include "lapi/posix_clocks.h"
 
 #ifndef NSEC_PER_SEC
-#define NSEC_PER_SEC (1000000000L)
+#define NSEC_PER_SEC (1000000000LL)
 #endif
 
 static const clock_t clock_list[] = {
Quoted 5 lines

What do you think?

Signature · 2 lines
Cyril Hrubis
[email protected]

Yeah, I even wrote and tested that it would build, but I wasn't able to go through any build logs or run complete tests.

Quoted 2 lines

But maybe I am just overthinking this, and any promotions due to this would only fix issues and not introduce any new ones.

That would be my expectation.

Quoted 3 lines

Well, on second thought, on 32-bit, in cases like variadic functions, it could cause a problem. So, I am just afraid that applying this fix without looking for every use of NSEC_PER_SEC might lead to new bugs.

Having a look at 'git grep NSEC_PER_SEC' there are five uses of NSEC_PER_SEC and all of these should be, as far as I can tell, safe.

Signature · 2 lines
Cyril Hrubis
[email protected]
#4 lore raw

PATCH 1/1 futex_wake04: avoid tst_ts_from_ns overflow on 32-bit platforms

Quoted 31 lines · Cyril Hrubis

Hi!

Quoted 27 lines
Quoted 18 lines

Good catch, I guess that it would be slightly cleaner to fix the NSEC_PER_SEC size instead with:

include/lapi/common_timers.h +1 -1
index b783beff4..e50f698d6 100644
@@ -12,7 +12,7 @@
 #include "lapi/posix_clocks.h"
 
 #ifndef NSEC_PER_SEC
-#define NSEC_PER_SEC (1000000000L)
+#define NSEC_PER_SEC (1000000000LL)
 #endif
 
 static const clock_t clock_list[] = {
Quoted 5 lines

What do you think?

Signature · 2 lines
Cyril Hrubis
[email protected]

Yeah, I even wrote and tested that it would build, but I wasn't able to go through any build logs or run complete tests.

Quoted 2 lines

But maybe I am just overthinking this, and any promotions due to this would only fix issues and not introduce any new ones.

Quoted 9 lines

That would be my expectation.

Quoted 3 lines

Well, on second thought, on 32-bit, in cases like variadic functions, it could cause a problem. So, I am just afraid that applying this fix without looking for every use of NSEC_PER_SEC might lead to new bugs.

Having a look at 'git grep NSEC_PER_SEC' there are five uses of NSEC_PER_SEC and all of these should be, as far as I can tell, safe.

Yes, there are even uses where NSEC_PER_SEC is redefined.

I am fine with this other change. I had already tested that it fixes the problem just the same.

How about the commit below?

Cascardo.

Quoted 3 lines
Signature · 2 lines
Cyril Hrubis
[email protected]

From 3c248cd5e6fc1dd967787249857fd310e688d89f Mon Sep 17 00:00:00 2001
From: Thadeu Lima de Souza Cascardo <[email protected]>
Date: Tue, 8 Jun 2021 10:21:48 -0300
Subject: [PATCH] common_timers: define NSEC_PER_SEC as long long to avoid
overflow on 32-bit

When multiplying 30 with NSEC_PER_SEC, the result would overflow on 32-bit platforms, unless there was promotion to long long, which is what tst_ts_from_ns expects.

futex_wake04, which uses that, would end up getting EINVAL when calling futex, because timespec_valid would fail, as tv_nsec would have an invalid value.

After this change, futex_wake04 passes on i386.

---
 include/lapi/common_timers.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/lapi/common_timers.h +1 -1
index 8da3abec24b2..953741d73810 100644
@@ -12,7 +12,7 @@
 #include "lapi/posix_clocks.h"
 
 #ifndef NSEC_PER_SEC
-#define NSEC_PER_SEC (1000000000L)
+#define NSEC_PER_SEC (1000000000LL)
 #endif
 
 static const clock_t clock_list[] = {
Signature · 1 line
2.30.2
#5 lore raw

PATCH 1/1 futex_wake04: avoid tst_ts_from_ns overflow on 32-bit platforms

Hi!

Quoted 6 lines

Yes, there are even uses where NSEC_PER_SEC is redefined.

I am fine with this other change. I had already tested that it fixes the problem just the same.

How about the commit below?

Looks good.

Signature · 2 lines
Cyril Hrubis
[email protected]