[v2] syscalls/keyctl02: wait for last key to be garbage collected

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>
---
 include/lapi/keyctl.h                       |  4 +++
 testcases/kernel/syscalls/keyctl/keyctl02.c | 39 ++++++++++++++++++++++++++---
 2 files changed, 40 insertions(+), 3 deletions(-)

Changes in v2:
  - drop KEYCTL_INVALIDATE ifdef
  - add KEYCTL_INVALIDATE to lapi/...
  - check for EOPNOTSUPP when using KEYCTL_INVALIDATE
  - tested as far back as 2.6.32

Hi!
>  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 +95,45 @@ 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 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");
> +
> +	/*
> +	 * If we have invalidate, we can drop extra key immediately as well,
> +	 * which also schedules gc.
> +	 */
> +	if (keyctl(KEYCTL_INVALIDATE, key_inv) == -1 && errno != EOPNOTSUPP)
> +		tst_brk(TBROK | TERRNO, "Failed to invalidate key");
> +
> +	/*
> +	 * 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);

Minor enhancement may be:

	if (i)
		tst_res(TINFO, "waiting for key gc took: %d ms", i);


Anyways, the patch looks good, acked.

----- Original Message -----
> Hi!
> >  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 +95,45 @@ 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 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");
> > +
> > +	/*
> > +	 * If we have invalidate, we can drop extra key immediately as well,
> > +	 * which also schedules gc.
> > +	 */
> > +	if (keyctl(KEYCTL_INVALIDATE, key_inv) == -1 && errno != EOPNOTSUPP)
> > +		tst_brk(TBROK | TERRNO, "Failed to invalidate key");
> > +
> > +	/*
> > +	 * 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);
> 
> Minor enhancement may be:
> 
> 	if (i)
> 		tst_res(TINFO, "waiting for key gc took: %d ms", i);
> 
> 
> Anyways, the patch looks good, acked.

Pushed with "if (i)" above.

Regards,
Jan

> 
> --
> Cyril Hrubis
> chrubis@suse.cz
>