Skip to contentSkip to conversation

landlock: fix minimal required size for landlock_ruleset_attr copying

RFC PATCH In review Li Wang 3 participants 4 messages Jul 2, 2024

Subsystems (2)
lore
Share & tools
lore

Li Wang

author
#0 lore raw

RFC PATCH 1/1 landlock: fix minimal required size for landlock_ruleset_attr copying

As kernel commit fff69fb03dde ("landlock: Support network rules with TCP bind and connect") introducing a new field 'handled_access_net' in the structure landlock_ruleset_attr, but in the landlock_create_ruleset() it still uses the first field 'handled_access_fs' to calculate minimal size, so that made decrease 1 is useless in LTP landlock01.c to test the too-small-size.

Test code:
rule_small_size = sizeof(struct landlock_ruleset_attr) - 1; tst_syscall(__NR_landlock_create_ruleset, ..., rule_small_size, 0)

Result:
landlock01.c:49: TFAIL: Size is too small expected EINVAL: ENOMSG (42)

Cc Mickaël Salaün <[email protected]>
Cc Konstantin Meskhidze <[email protected]>
Cc Paul Moore <[email protected]>

---

Notes:
Hi Mickael,
I'm not quite sure if that is on purpose to use the first field or kernel bug, can you take a look?

 security/landlock/syscalls.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
security/landlock/syscalls.c +1 -1
index 03b470f5a85a..f3cd7def7624 100644
@@ -198,7 +198,7 @@ SYSCALL_DEFINE3(landlock_create_ruleset,
 	/* Copies raw user space buffer. */
 	err = copy_min_struct_from_user(&ruleset_attr, sizeof(ruleset_attr),
 					offsetofend(typeof(ruleset_attr),
-						    handled_access_fs),
+						    handled_access_net),
 					attr, size);
 	if (err)
 		return err;
Signature · 1 line
2.45.2
maintainer
#1 lore raw
Quoted 6 lines · Li Wang

As kernel commit fff69fb03dde ("landlock: Support network rules with TCP bind and connect") introducing a new field 'handled_access_net' in the structure landlock_ruleset_attr, but in the landlock_create_ruleset() it still uses the first field 'handled_access_fs' to calculate minimal size, so that made decrease 1 is useless in LTP landlock01.c to test the too-small-size.

Quoted 6 lines

Test code:

   rule_small_size = sizeof(struct landlock_ruleset_attr) - 1;
   tst_syscall(__NR_landlock_create_ruleset, ..., rule_small_size, 0)

Result:
landlock01.c:49: TFAIL: Size is too small expected EINVAL: ENOMSG (42)

Interesting, this looks like a bug in these LTP tests.

Quoted 11 lines
Cc Mickaël Salaün <[email protected]>
Cc Konstantin Meskhidze <[email protected]>
Cc Paul Moore <[email protected]>

---

Notes:
Hi Mickael,
I'm not quite sure if that is on purpose to use the first field or kernel bug, can you take a look?

Hi Li,

Yes this is on purpose. The handled_access_fs minimal size check should never change for backward compatibility reason. User space built with old headers must still work with new kernels. This is tested with the "inconsistent_attr" test in tools/testing/selftests/landlock/base_test.c

Quoted 20 lines
 security/landlock/syscalls.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
security/landlock/syscalls.c +1 -1
index 03b470f5a85a..f3cd7def7624 100644
@@ -198,7 +198,7 @@ SYSCALL_DEFINE3(landlock_create_ruleset,
 	/* Copies raw user space buffer. */
 	err = copy_min_struct_from_user(&ruleset_attr, sizeof(ruleset_attr),
 					offsetofend(typeof(ruleset_attr),
-						    handled_access_fs),
+						    handled_access_net),
 					attr, size);
 	if (err)
 		return err;
Signature · 1 line
2.45.2
Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp
author
#2 lore raw
Quoted 11 lines · Mickaël Salaün
Quoted 1 line · Li Wang

As kernel commit fff69fb03dde ("landlock: Support network rules with TCP

bind and connect")

Quoted 1 line

introducing a new field 'handled_access_net' in the structure

landlock_ruleset_attr,

Quoted 1 line

but in the landlock_create_ruleset() it still uses the first field

'handled_access_fs'

Quoted 1 line

to calculate minimal size, so that made decrease 1 is useless in LTP

landlock01.c to

Quoted 2 lines

test the too-small-size.

Quoted 9 lines
Quoted 6 lines

Test code:

   rule_small_size = sizeof(struct landlock_ruleset_attr) - 1;
   tst_syscall(__NR_landlock_create_ruleset, ..., rule_small_size, 0)

Result:
landlock01.c:49: TFAIL: Size is too small expected EINVAL: ENOMSG (42)

Interesting, this looks like a bug in these LTP tests.

Yes, Andrea drafted a series of landlock tests for LTP.

FYI -
https://patchwork.ozlabs.org/project/ltp/list/?series=&submitter=&state=&q=landlock&archive=&delegate=

Quoted 12 lines
Quoted 11 lines
Cc Mickaël Salaün <[email protected]>
Cc Konstantin Meskhidze <[email protected]>
Cc Paul Moore <[email protected]>

---

Notes:
Hi Mickael,
I'm not quite sure if that is on purpose to use the first field

or kernel

Quoted 9 lines
Quoted 1 line

bug, can you take a look?

Hi Li,

Yes this is on purpose. The handled_access_fs minimal size check should never change for backward compatibility reason. User space built with old headers must still work with new kernels. This is tested with the "inconsistent_attr" test in tools/testing/selftests/landlock/base_test.c

Thanks for the confirmation, very helpful.

@Andrea Cervesato <[email protected]>, seems we have to adjust the test to use a real small
rule size which is at least small than sizeof(__u64).

Signature · 5 lines
Regards,
Li Wang

-- 
Mailing list info: https://lists.linux.it/listinfo/ltp
#3 lore raw
Quoted 55 lines · Li Wang

On Tue, Jul 2, 2024 at 9:00 PM Mickaël Salaün <[email protected]> wrote:

On Tue, Jul 02, 2024 at 05:47:45PM +0800, Li Wang wrote:
> As kernel commit fff69fb03dde ("landlock: Support network rules with TCP bind and connect")
> introducing a new field 'handled_access_net' in the structure landlock_ruleset_attr,
> but in the landlock_create_ruleset() it still uses the first field 'handled_access_fs'
> to calculate minimal size, so that made decrease 1 is useless in LTP landlock01.c to
> test the too-small-size.
>
> Test code:
>    rule_small_size = sizeof(struct landlock_ruleset_attr) - 1; >    tst_syscall(__NR_landlock_create_ruleset, ..., rule_small_size, 0)
>
> Result:
>   landlock01.c:49: TFAIL: Size is too small expected EINVAL:
ENOMSG (42)

Interesting, this looks like a bug in these LTP tests.

Yes, Andrea drafted a series of landlock tests for LTP.

FYI - https://patchwork.ozlabs.org/project/ltp/list/?series=&submitter=&state=&q=landlock&archive=&delegate= <https://patchwork.ozlabs.org/project/ltp/list/?series=&submitter=&state=&q=landlock&archive=&delegate=>

Quoted 10 lines
Cc Mickaël Salaün <[email protected]>
Cc Konstantin Meskhidze <[email protected]>
Cc Paul Moore <[email protected]>

---

Notes:
     Hi Mickael,
        I'm not quite sure if that is on purpose to use the first

field or kernel
>        bug, can you take a look?

Hi Li,

Yes this is on purpose.  The handled_access_fs minimal size check should
never change for backward compatibility reason.  User space built with old headers must still work with new kernels.  This is tested with the "inconsistent_attr" test in tools/testing/selftests/landlock/base_test.c

Quoted 6 lines

Thanks for the confirmation, very helpful.

@Andrea Cervesato <mailto:[email protected]>, seems we have to adjust the test to use a real small
rule size which is at least small than sizeof(__u64).

Sure, thanks for checking.

Quoted 4 lines
Signature · 3 lines
-- 
Regards,
Li Wang

Andrea

Signature · 1 line
Mailing list info: https://lists.linux.it/listinfo/ltp