Skip to contentSkip to conversation

tests/bpf: ask for unlimited RLIMIT_MEMLOCK

PATCH In review Ondrej Mosnacek 2 participants 4 messages Mar 12, 2020 – Mar 20, 2020 SELinux security module

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tests/bpf: ask for unlimited RLIMIT_MEMLOCK

Currently the code sets it to at most 128K, but this is not enough in some aarch64/ppc64le environments. Therefore, stop guessing the magic threshold and just set it straight to RLIM_INFINITY.

Fixes 8f0f980a4ad5 ("selinux-testsuite: Add BPF tests")
---
 tests/bpf/bpf_common.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)
tests/bpf/bpf_common.c +4 -12
index 681e2eb..f499034 100644
@@ -41,24 +41,16 @@ int create_bpf_prog(void)
 
 /*
  * The default RLIMIT_MEMLOCK is normally 64K, however BPF map/prog requires
- * more than this so double it unless RLIM_INFINITY is set.
+ * more than this (the actual threshold varying across arches) so set it to
+ * RLIM_INFINITY.
  */
 void bpf_setrlimit(void)
 {
 	int result;
 	struct rlimit r;
 
-	result = getrlimit(RLIMIT_MEMLOCK, &r);
-	if (result < 0) {
-		fprintf(stderr, "Failed to get resource limit: %s\n",
-			strerror(errno));
-		exit(-1);
-	}
-
-	if (r.rlim_cur != RLIM_INFINITY && r.rlim_cur <= (64 * 1024)) {
-		r.rlim_cur *= 2;
-		r.rlim_max *= 2;
-	}
+	r.rlim_cur = RLIM_INFINITY;
+	r.rlim_max = RLIM_INFINITY;
 
 	result = setrlimit(RLIMIT_MEMLOCK, &r);
 	if (result < 0) {
Signature · 1 line
2.24.1
maintainer Acked-by on v1 1/1
#1 lore raw
Quoted 9 lines · Ondrej Mosnacek

Currently the code sets it to at most 128K, but this is not enough in some aarch64/ppc64le environments. Therefore, stop guessing the magic threshold and just set it straight to RLIM_INFINITY.

Fixes 8f0f980a4ad5 ("selinux-testsuite: Add BPF tests")
---
 tests/bpf/bpf_common.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)

I have to make a similar fix to the audit-testsuite earlier this week, I didn't think to check the selinux-testsuite because the bpf tests were running just fine on my aarch64 test system. Sorry about that, I should have checked regardless.

One small style nit below, but otherwise ...

Quoted 20 lines
tests/bpf/bpf_common.c +2 -4
index 681e2eb..f499034 100644
@@ -41,24 +41,16 @@ int create_bpf_prog(void)
 
 /*
  * The default RLIMIT_MEMLOCK is normally 64K, however BPF map/prog requires
- * more than this so double it unless RLIM_INFINITY is set.
+ * more than this (the actual threshold varying across arches) so set it to
+ * RLIM_INFINITY.
  */
 void bpf_setrlimit(void)
 {
        int result;
        struct rlimit r;
 
-       result = getrlimit(RLIMIT_MEMLOCK, &r);
-       if (result < 0) {
-               fprintf(stderr, "Failed to get resource limit: %s\n",

tests/bpf/bpf_common.c int create_bpf_prog(void) · line 53

Quoted 10 lines
tests/bpf/bpf_common.c +2 -8
-                       strerror(errno));
-               exit(-1);
-       }
-
-       if (r.rlim_cur != RLIM_INFINITY && r.rlim_cur <= (64 * 1024)) {
-               r.rlim_cur *= 2;
-               r.rlim_max *= 2;
-       }
+       r.rlim_cur = RLIM_INFINITY;
+       r.rlim_max = RLIM_INFINITY;

If you really want to simplify things you could assign those values when you declare "r":

struct rlimit r = {
.rlim_cur = RLIM_INFINITY,
.rlim_max = RLIM_INFINITY };

Quoted 4 lines
tests/bpf/bpf_common.c +0 -1
        result = setrlimit(RLIMIT_MEMLOCK, &r);
        if (result < 0) {
--

2.24.1

Signature · 2 lines
paul moore
www.paul-moore.com
#2 lore raw
Quoted 11 lines · Paul Moore
Quoted 9 lines · Ondrej Mosnacek

Currently the code sets it to at most 128K, but this is not enough in some aarch64/ppc64le environments. Therefore, stop guessing the magic threshold and just set it straight to RLIM_INFINITY.

Fixes 8f0f980a4ad5 ("selinux-testsuite: Add BPF tests")
---
 tests/bpf/bpf_common.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)
Quoted 4 lines

I have to make a similar fix to the audit-testsuite earlier this week, I didn't think to check the selinux-testsuite because the bpf tests were running just fine on my aarch64 test system. Sorry about that, I should have checked regardless.

No problem, I had kept the bpf subtest disabled on non-x86_64 in our testing for some time, since it was failing there and I didn't know why. Once I saw your audit-testsuite patch I realized that it was probably the same issue so I did some testing and indeed the RLIMIT_MEMLOCK was the cause so I put together a patch.

Quoted 39 lines

One small style nit below, but otherwise ...

Quoted 20 lines
tests/bpf/bpf_common.c +2 -4
index 681e2eb..f499034 100644
@@ -41,24 +41,16 @@ int create_bpf_prog(void)
 
 /*
  * The default RLIMIT_MEMLOCK is normally 64K, however BPF map/prog requires
- * more than this so double it unless RLIM_INFINITY is set.
+ * more than this (the actual threshold varying across arches) so set it to
+ * RLIM_INFINITY.
  */
 void bpf_setrlimit(void)
 {
        int result;
        struct rlimit r;
 
-       result = getrlimit(RLIMIT_MEMLOCK, &r);
-       if (result < 0) {
-               fprintf(stderr, "Failed to get resource limit: %s\n",

tests/bpf/bpf_common.c int create_bpf_prog(void) · line 53

Quoted 10 lines
tests/bpf/bpf_common.c +2 -8
-                       strerror(errno));
-               exit(-1);
-       }
-
-       if (r.rlim_cur != RLIM_INFINITY && r.rlim_cur <= (64 * 1024)) {
-               r.rlim_cur *= 2;
-               r.rlim_max *= 2;
-       }
+       r.rlim_cur = RLIM_INFINITY;
+       r.rlim_max = RLIM_INFINITY;

If you really want to simplify things you could assign those values when you declare "r":

Quoted 3 lines

struct rlimit r = {
.rlim_cur = RLIM_INFINITY,
.rlim_max = RLIM_INFINITY };

True, but I don't think the difference is big enough to respin the patch.

Quoted 10 lines
Quoted 5 lines
tests/bpf/bpf_common.c +0 -1
        result = setrlimit(RLIMIT_MEMLOCK, &r);
        if (result < 0) {
--

2.24.1

--
paul moore
www.paul-moore.com

Signature · 3 lines
Ondrej Mosnacek <omosnace at redhat dot com>
Software Engineer, Security Technologies
Red Hat, Inc.
#3 lore raw
In reply to Paul Moore
Quoted 11 lines · Paul Moore
Quoted 9 lines · Ondrej Mosnacek

Currently the code sets it to at most 128K, but this is not enough in some aarch64/ppc64le environments. Therefore, stop guessing the magic threshold and just set it straight to RLIM_INFINITY.

Fixes 8f0f980a4ad5 ("selinux-testsuite: Add BPF tests")
---
 tests/bpf/bpf_common.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)
Quoted 8 lines

I have to make a similar fix to the audit-testsuite earlier this week, I didn't think to check the selinux-testsuite because the bpf tests were running just fine on my aarch64 test system. Sorry about that, I should have checked regardless.

One small style nit below, but otherwise ...

I went ahead and applied the patch, thanks!

Signature · 3 lines
Ondrej Mosnacek <omosnace at redhat dot com>
Software Engineer, Security Technologies
Red Hat, Inc.