(2010/02/09 15:46), KaiGai Kohei wrote: > (2010/02/05 23:50), Stephen Smalley wrote: >> On Fri, 2010-02-05 at 14:42 +0900, KaiGai Kohei wrote: >>> What is the current status of this patch? >>> Its kernel side patch has been already merged into James's -next tree. >>> >>> http://git.kernel.org/?p=linux/kernel/git/jmorris/security-testing-2.6.git;a=commit;h=7d52a155e38d5a165759dbbee656455861bf7801 >> >> I had to use -l to get it to apply (whitespace mangled). >> >> Before applying it, when trying to install the test_policy.pp module >> from the selinux testsuite with expand-check=1 >> in /etc/selinux/semanage.conf, I get: >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_file_blue_t : file { ioctl read getattr lock open } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_file_red_t : file { write append } >> libsepol.hierarchy_check_constraints: 2 total errors found during hierarchy check >> >> And after applying it, I get the following: >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_file_blue_t : file { ioctl read getattr lock open } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : process { fork transition sigchld sigkill sigstop signull signal getsched setsched getsession getpgid setpgid getcap setcap share getattr setexec setfscreate noatsecure siginh rlimitinh setcurrent setkeycreate setsockcreate } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : capability { dac_override dac_read_search } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : file { ioctl read write getattr lock append open } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : dir { ioctl read getattr lock search open } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : fd { use } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : lnk_file { ioctl read getattr lock } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : fifo_file { ioctl read write getattr lock append open } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : unix_stream_socket { ioctl read write create getattr setattr append bind connect listen accept getopt setopt shutdown } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : unix_dgram_socket { ioctl read write create getattr setattr append bind connect getopt setopt shutdown sendto } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_child_t : association { sendto } >> libsepol.check_avtab_hierarchy_callback: hierarchy violation between types test_bounds_child_t and test_bounds_file_red_t : file { write append } >> libsepol.hierarchy_check_constraints: 12 total errors found during hierarchy check >> > > We might overlook something in the case when both of source and target are bounded. > > When test_bounds_child_t tries to access a pseudo file, such as /proc/<PID>, these > entries are labeled to domain of the corresponding process. > The test_bounds_child_t is bounded to test_bounds_parent_t, and it is also a case > when both of source and target are bounded to its super domain. > > This test policy indeed violated to the first check that ensures the permissions > between source (test_bounds_child_t) and target (test_bounds_child_t) are enclosed > by the permissions between parent of the source (test_bounds_parent_t) and the target, > because nothing are allowed between test_bounds_parent_t and test_bounds_child_t as > a filesystem object. > > However, the older implementation had also ensured the permissions between the source > and the target are enclosed by the permissions between parent of the source and parent > of the target, then, if these are enclosed, this check allows the permissions between > both of children, even if the first checks were violated. > > In other words, we have to allow the test_bounds_parent_t to performs something on > the test_bounds_child_t explicitly, in this case. > > At first, I replied to Jacques Thomas that I could not find any actual use cases in > the target side boundary, but it might be misjudge. > If I vote it again, I'll support an idea to revert my patch. What is your opinion for this topic? At first, I didn't find out the case when a bounded domain accesses /proc/self/* entries with its parent's security context. In this case, the policy writer has to allow parent domain to access all the pseudo file entries of possible child domains. The previous boundary code on the target side well handled it, but I removed it as a dead code. I think we have two options. The one is simple. It is to revert this commit: 7d52a155e38d5a165759dbbee656455861bf7801 The other is to have a convention when we write policies using bounds domain. For example, if all the domain bounded to httpd_t must have httpd_child_domain attribute, we can allow httpd_t domain to access file entries of possible child domains in apache.te. Right now, my preference is the later option. Thanks, -- OSS Platform Development Division, NEC KaiGai Kohei <kaigai@xxxxxxxxxxxxx> -- This message was distributed to subscribers of the selinux mailing list. If you no longer wish to subscribe, send mail to majordomo@xxxxxxxxxxxxx with the words "unsubscribe selinux" without quotes as the message.
