[LTP] [PATCH] syscalls/keyctl02: wait for last key to be garbage collected

[LTP] [PATCH] syscalls/keyctl02: wait for last key to be garbage collected

[Jan Stancek]

Fixes: #409

This tests creates a lot of keys, which are revoked, but they may
not all be destroyed by the time test finishes. This can create
problems for subsequent tests which will hit quota limit and fail.

This patch adds loop, that periodically sleeps and checks if last
test key has been garbage collected yet.

Signed-off-by: Jan Stancek <jstancek@redhat.com>
---
 testcases/kernel/syscalls/keyctl/keyctl02.c | 40 ++++++++++++++++++++++++++---
 1 file changed, 37 insertions(+), 3 deletions(-)

diff --git a/testcases/kernel/syscalls/keyctl/keyctl02.c b/testcases/kernel/syscalls/keyctl/keyctl02.c
index 515da96e3b2c..477601c1090f 100644
--- a/testcases/kernel/syscalls/keyctl/keyctl02.c
+++ b/testcases/kernel/syscalls/keyctl/keyctl02.c
@@ -44,6 +44,7 @@
 #include "lapi/keyctl.h"
 
 #define LOOPS	20000
+#define MAX_WAIT_FOR_GC_MS 5000
 #define PATH_KEY_COUNT_QUOTA	"/proc/sys/kernel/keys/root_maxkeys"
 
 static int orig_maxkeys;
@@ -69,8 +70,8 @@ static void *do_revoke(void *arg)
 
 static void do_test(void)
 {
-	int i;
-	key_serial_t key;
+	int i, ret;
+	key_serial_t key, key_inv;
 	pthread_t pth[4];
 
 	for (i = 0; i < LOOPS; i++) {
@@ -94,13 +95,46 @@ static void do_test(void)
 		SAFE_PTHREAD_JOIN(pth[3], NULL);
 	}
 
+	/*
+	 * Kernel should start garbage collect when last reference to key
+	 * is removed (see key_put()). Since we are adding keys with identical
+	 * description and type, each replacement should schedule a gc run,
+	 * see comment@__key_link().
+	 *
+	 * We create extra key here, to remove reference to last revoked key.
+	 */
+	key_inv = add_key("user", "ltptestkey", "foo", 3,
+		KEY_SPEC_PROCESS_KEYRING);
+	if (key_inv == -1)
+		tst_brk(TBROK | TERRNO, "Failed to add key");
+#ifdef KEYCTL_INVALIDATE
+	/*
+	 * If we have invalidate, we can drop extra key immediately as well,
+	 * which also schedules gc.
+	 */
+	if (keyctl(KEYCTL_INVALIDATE, key_inv))
+		tst_brk(TBROK | TERRNO, "Failed to invalidate key");
+#endif
+
+	/*
+	 * At this point we are quite confident that gc has been scheduled,
+	 * so we wait and periodically check for last test key to be removed.
+	 */
+	for (i = 0; i < MAX_WAIT_FOR_GC_MS; i += 100) {
+		ret = keyctl(KEYCTL_REVOKE, key);
+		if (ret == -1 && errno == ENOKEY)
+			break;
+		usleep(100*1000);
+	}
+
+	tst_res(TINFO, "waiting for key gc took: %d ms", i);
 	tst_res(TPASS, "Bug not reproduced");
 }
 
 static void setup(void)
 {
 	SAFE_FILE_SCANF(PATH_KEY_COUNT_QUOTA, "%d", &orig_maxkeys);
-	SAFE_FILE_PRINTF(PATH_KEY_COUNT_QUOTA, "%d", orig_maxkeys + LOOPS);
+	SAFE_FILE_PRINTF(PATH_KEY_COUNT_QUOTA, "%d", orig_maxkeys + LOOPS + 1);
 }
 
 static void cleanup(void)
-- 
1.8.3.1

[LTP] [PATCH] syscalls/keyctl02: wait for last key to be garbage collected

[Cyril Hrubis]

Hi!
> +	/*
> +	 * Kernel should start garbage collect when last reference to key
> +	 * is removed (see key_put()). Since we are adding keys with identical
> +	 * description and type, each replacement should schedule a gc run,
> +	 * see comment at __key_link().
> +	 *
> +	 * We create extra key here, to remove reference to last revoked key.
> +	 */
> +	key_inv = add_key("user", "ltptestkey", "foo", 3,
> +		KEY_SPEC_PROCESS_KEYRING);
> +	if (key_inv == -1)
> +		tst_brk(TBROK | TERRNO, "Failed to add key");
> +#ifdef KEYCTL_INVALIDATE
> +	/*
> +	 * If we have invalidate, we can drop extra key immediately as well,
> +	 * which also schedules gc.
> +	 */
> +	if (keyctl(KEYCTL_INVALIDATE, key_inv))
> +		tst_brk(TBROK | TERRNO, "Failed to invalidate key");

We should probably handle the situation where someone compiles this code
on newer userspace and runs it on older kernel than 3.5, I suppose that
we will get -1 with EINVAL or EOPNOTSUPP here.

And if that is handled we can also drop the ifdef in favor of fallback
definition.


Otherwise this looks good.

-- 
Cyril Hrubis
chrubis@suse.cz

[LTP] [PATCH] syscalls/keyctl02: wait for last key to be garbage collected

[Jan Stancek]

----- Original Message -----
> Hi!
> > +	/*
> > +	 * Kernel should start garbage collect when last reference to key
> > +	 * is removed (see key_put()). Since we are adding keys with identical
> > +	 * description and type, each replacement should schedule a gc run,
> > +	 * see comment at __key_link().
> > +	 *
> > +	 * We create extra key here, to remove reference to last revoked key.
> > +	 */
> > +	key_inv = add_key("user", "ltptestkey", "foo", 3,
> > +		KEY_SPEC_PROCESS_KEYRING);
> > +	if (key_inv == -1)
> > +		tst_brk(TBROK | TERRNO, "Failed to add key");
> > +#ifdef KEYCTL_INVALIDATE
> > +	/*
> > +	 * If we have invalidate, we can drop extra key immediately as well,
> > +	 * which also schedules gc.
> > +	 */
> > +	if (keyctl(KEYCTL_INVALIDATE, key_inv))
> > +		tst_brk(TBROK | TERRNO, "Failed to invalidate key");
> 
> We should probably handle the situation where someone compiles this code
> on newer userspace and runs it on older kernel than 3.5, I suppose that
> we will get -1 with EINVAL or EOPNOTSUPP here.
> 
> And if that is handled we can also drop the ifdef in favor of fallback
> definition.

Agreed, I'll test on some ancient kernel and send v2.

Regards,
Jan

> 
> 
> Otherwise this looks good.
> 
> --
> Cyril Hrubis
> chrubis@suse.cz
>