use the netmsg initial SID for NetLabel connections

RFC In review Paul Moore 2 participants 5 messages Jun 7, 2007 – Jun 12, 2007 SELinux security module

Share & tools

Paul Moore

author

2007-06-07 21:11 UTC #0 lore raw

RFC 1/1 use the netmsg initial SID for NetLabel connections

This is a simple RFC patch for the Reference Policy to go along with the proposed NetLabel kernel changes to use the netmsg initial SID, the kernel change thread can be found here:

* http://marc.info/?t=117394234100037&r=1&w=2

The basic idea is that NetLabel would use the netmsg initial SID as the base/TE context for MLS-only connections and the unlabeled SID for connections that did not have have security attributes at all, true unlabeled connections. The reason for doing this would be to enable us to allow/disallow unlabeled NetLabel traffic on a per-domain basis; something that is not easily possible with the current approach of using the unlabeled SID as the base/TE context for MLS-only connections.

I don't consider the changes below to be complete, there would still need to be changes to all of the user and application domains, but I want to start a conversation on the "right" way to do this. Chris, Dan, I'm especially interested in what the two of you have to say.

FYI: This _really_ is a RFC patch, I haven't even compile tested it, I'm just using it as a starting point for the conversation.

---
 policy/mls                              |    5 
 policy/modules/kernel/corenetwork.if.in |   88 ++++++++++++++
 policy/modules/kernel/kernel.if         |  194 +++++++++++++++++++++++++++++---
 policy/modules/kernel/kernel.te         |    8 +
 4 files changed, 276 insertions(+), 19 deletions(-)

policy/mls +3 -2

Index: refpolicy_svn_repo/policy/mls
===================================================================
@@ -182,11 +182,12 @@ mlsconstrain { socket tcp_socket udp_soc
 	 (( t1 == mlsnetwritetoclr ) and ( h1 dom l2 ) and ( l1 domby l2 )) or
 	 ( t1 == mlsnetwrite ));
 
-# used by netlabel to restrict normal domains to same level connections
+# used by netlabel to restrict normal domains to same level connections unless the connection is unlabeled
 mlsconstrain { tcp_socket udp_socket rawip_socket } recvfrom
 	(( l1 eq l2 ) or
 	 (( t1 == mlsnetreadtoclr ) and ( h1 dom l2 )) or
-	 ( t1 == mlsnetread ));
+	 ( t1 == mlsnetread ) or
+	 ( t2 == unlabeled_t ));
 
 # these access vectors have no MLS restrictions
 # { socket tcp_socket udp_socket rawip_socket netlink_socket packet_socket key_socket unix_stream_socket unix_dgram_socket netlink_route_socket netlink_firewall_socket netlink_tcpdiag_socket netlink_nflog_socket netlink_xfrm_socket netlink_selinux_socket netlink_audit_socket netlink_ip6fw_socket netlink_dnrt_socket } { ioctl create lock append bind sendto send_msg name_bind }

policy/modules/kernel/corenetwork.if.in +88 -0

127 lines not on this page · raw message

policy/modules/kernel/kernel.if +178 ~~-16~~

236 lines not on this page · raw message

policy/modules/kernel/kernel.te +7 -1

Index: refpolicy_svn_repo/policy/modules/kernel/kernel.te
===================================================================
@@ -139,6 +139,13 @@ type sysctl_dev_t, sysctl_type;
 genfscon proc /sys/dev gen_context(system_u:object_r:sysctl_dev_t,s0)
 
 #
+# The netmsg inital SID is used by the kernel's NetLabel subsystem for network
+# connections which do not carry full SELinux contexts.
+#
+type netlabel_peer_t;
+sid netmsg		gen_context(system_u:object_r:netlabel_peer_t,mls_systemhigh)
+
+#
 # unlabeled_t is the type of unlabeled objects.
 # Objects that have no known labeling information or that
 # have labels that are no longer valid are treated as having this type.

@@ -153,7 +160,6 @@ sid icmp_socket		gen_context(system_u:ob
 sid igmp_packet		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid init		gen_context(system_u:object_r:unlabeled_t,s0)
 sid kmod		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
-sid netmsg		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid policy		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid scmp_packet		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid sysctl_modprobe 	gen_context(system_u:object_r:unlabeled_t,s0)

Signature · 8 lines

paul moore
linux security @ hp


--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to [email protected] with
the words "unsubscribe selinux" without quotes as the message.

Christopher J. PeBenito

2007-06-08 13:38 UTC #1 lore raw

Quoted 18 lines · Paul Moore

This is a simple RFC patch for the Reference Policy to go along with the proposed NetLabel kernel changes to use the netmsg initial SID, the kernel change thread can be found here:
* http://marc.info/?t=117394234100037&r=1&w=2
The basic idea is that NetLabel would use the netmsg initial SID as the base/TE context for MLS-only connections and the unlabeled SID for connections that did not have have security attributes at all, true unlabeled connections. The reason for doing this would be to enable us to allow/disallow unlabeled NetLabel traffic on a per-domain basis; something that is not easily possible with the current approach of using the unlabeled SID as the base/TE context for MLS-only connections.
I don't consider the changes below to be complete, there would still need to be changes to all of the user and application domains, but I want to start a conversation on the "right" way to do this. Chris, Dan, I'm especially interested in what the two of you have to say.

[...]

Quoted 17 lines

policy/modules/kernel/kernel.te +7 -0

@@ -139,6 +139,13 @@ type sysctl_dev_t, sysctl_type;
 genfscon proc /sys/dev gen_context(system_u:object_r:sysctl_dev_t,s0)
 
 #
+# The netmsg inital SID is used by the kernel's NetLabel subsystem for network
+# connections which do not carry full SELinux contexts.
+#
+type netlabel_peer_t;
+sid netmsg		gen_context(system_u:object_r:netlabel_peer_t,mls_systemhigh)
+
+#
 # unlabeled_t is the type of unlabeled objects.
 # Objects that have no known labeling information or that
 # have labels that are no longer valid are treated as having this type.

policy/modules/kernel/kernel.te sid icmp_socket gen_context(system_u:ob · line 165

Quoted 7 lines

policy/modules/kernel/kernel.te +0 -1

@@ -153,7 +160,6 @@ sid icmp_socket		gen_context(system_u:ob
 sid igmp_packet		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid init		gen_context(system_u:object_r:unlabeled_t,s0)
 sid kmod		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
-sid netmsg		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid policy		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid scmp_packet		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid sysctl_modprobe 	gen_context(system_u:object_r:unlabeled_t,s0)

The type declaration and initial sid line should move over to corenetwork (with corresponding interface changes). The only reason there were netlabel interfaces in the kernel module were because they were using unlabeled_t.

Signature · 9 lines

Chris PeBenito
Tresys Technology, LLC
(410) 290-1411 x150


--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to [email protected] with
the words "unsubscribe selinux" without quotes as the message.

Paul Moore

author

2007-06-08 13:58 UTC #2 lore raw

Quoted 24 lines · Christopher J. PeBenito

Quoted 23 lines · Paul Moore

policy/modules/kernel/kernel.te +1 -0

@@ -139,6 +139,13 @@ type sysctl_dev_t, sysctl_type;
 genfscon proc /sys/dev gen_context(system_u:object_r:sysctl_dev_t,s0)
 
 #
+# The netmsg inital SID is used by the kernel's NetLabel subsystem for

network +# connections which do not carry full SELinux contexts.

policy/modules/kernel/kernel.te +3 -0

+#
+type netlabel_peer_t;
+sid

netmsg gen_context(system_u:object_r:netlabel_peer_t,mls_systemhigh) +

policy/modules/kernel/kernel.te +1 -1

+#
 # unlabeled_t is the type of unlabeled objects.
 # Objects that have no known labeling information or that
 # have labels that are no longer valid are treated as having this type.

@@ -153,7 +160,6 @@ sid icmp_socket		gen_context(system_u:ob
 sid
 igmp_packet		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid init		gen_context(system_u:object_r:unlabeled_t,s0)
 sid kmod		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
-sid netmsg		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid policy		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)

Quoted 8 lines

Quoted 3 lines
policy/modules/kernel/kernel.te +0 -0
 sid
scmp_packet gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
sid sysctl_modprobe gen_context(system_u:object_r:unlabeled_t,s0)
The type declaration and initial sid line should move over to corenetwork (with corresponding interface changes). The only reason there were netlabel interfaces in the kernel module were because they were using unlabeled_t.

Okay, that makes sense.

I still have a question about the best way to provide both labeled and unlabeled NetLabel support to all of the user/application domains in the policy. I don't have a problem going through all the individual domains and adding calls like these:

corenet_{tcp,udp,raw}_recv_unlabeled(<domain>)
corenet_{tcp,udp,raw}_recv_netlabel(<domain>)

... but I was wondering if there was another way I should go about making the change?

Signature · 7 lines

paul moore
linux security @ hp

--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to [email protected] with
the words "unsubscribe selinux" without quotes as the message.

Christopher J. PeBenito

2007-06-12 12:13 UTC #3 lore raw

Quoted 41 lines · Paul Moore

Quoted 24 lines · Christopher J. PeBenito
Quoted 23 lines · Paul Moore
policy/modules/kernel/kernel.te +1 -0
@@ -139,6 +139,13 @@ type sysctl_dev_t, sysctl_type;
 genfscon proc /sys/dev gen_context(system_u:object_r:sysctl_dev_t,s0)
 
 #
+# The netmsg inital SID is used by the kernel's NetLabel subsystem for
network +# connections which do not carry full SELinux contexts.
policy/modules/kernel/kernel.te +3 -0
+#
+type netlabel_peer_t;
+sid
netmsg gen_context(system_u:object_r:netlabel_peer_t,mls_systemhigh) +
policy/modules/kernel/kernel.te +1 -1
+#
 # unlabeled_t is the type of unlabeled objects.
 # Objects that have no known labeling information or that
 # have labels that are no longer valid are treated as having this type.
@@ -153,7 +160,6 @@ sid icmp_socket		gen_context(system_u:ob
 sid
 igmp_packet		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid init		gen_context(system_u:object_r:unlabeled_t,s0)
 sid kmod		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
-sid netmsg		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
 sid policy		gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
Quoted 8 lines
Quoted 3 lines
policy/modules/kernel/kernel.te +0 -0
 sid
scmp_packet gen_context(system_u:object_r:unlabeled_t,mls_systemhigh)
sid sysctl_modprobe gen_context(system_u:object_r:unlabeled_t,s0)
The type declaration and initial sid line should move over to corenetwork (with corresponding interface changes). The only reason there were netlabel interfaces in the kernel module were because they were using unlabeled_t.
Okay, that makes sense.
I still have a question about the best way to provide both labeled and unlabeled NetLabel support to all of the user/application domains in the policy. I don't have a problem going through all the individual domains and adding calls like these:

Quoted 5 lines

corenet_{tcp,udp,raw}_recv_unlabeled(<domain>)
corenet_{tcp,udp,raw}_recv_netlabel(<domain>)
... but I was wondering if there was another way I should go about making the change?

No, that's the right method. I've been thinking about if this is the right vocabulary for the interfaces; I'd like to make it workable for labeled ipsec too. The problem is that ipsec is a little different, where you might have interfaces like apache_tcp_recvfrom_user_script().

Signature · 9 lines

Chris PeBenito
Tresys Technology, LLC
(410) 290-1411 x150


--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to [email protected] with
the words "unsubscribe selinux" without quotes as the message.

Paul Moore

author

2007-06-12 16:46 UTC #4 lore raw

Quoted 9 lines · Christopher J. PeBenito

Quoted 8 lines · Paul Moore
I still have a question about the best way to provide both labeled and unlabeled NetLabel support to all of the user/application domains in the policy. I don't have a problem going through all the individual domains and adding calls like these:
corenet_{tcp,udp,raw}_recv_unlabeled(<domain>)
corenet_{tcp,udp,raw}_recv_netlabel(<domain>)

Quoted 6 lines

Quoted 2 lines
... but I was wondering if there was another way I should go about making the change?
No, that's the right method. I've been thinking about if this is the right vocabulary for the interfaces; I'd like to make it workable for labeled ipsec too.

Yes, especially for the unlabeled case. I would think having a single interface to allow a domain to receive unlabeled traffic would be much better than two.

Quoted 2 lines

The problem is that ipsec is a little different,
where you might have interfaces like apache_tcp_recvfrom_user_script().

Yes, you also have all the SPD matching issues to deal with. Hopefully later this year I'll have NetLabel supporting full SELinux contexts which should make NetLabel and IPsec very similar in regards to packet subject/object labels; the class/permissions will still be different but that is more of an interface implementation issue than an actual interface/API issue.

Thanks for you help, I'll hope to have a patch for you to review within a week or two that adds the corenet/NetLabel interfaces to the policy in SVN. I'd like to get this accepted and merged before I push the final kernel patch up to James/Stephen as the kernel changes require a new/updated policy to work correctly.

Signature · 7 lines

paul moore
linux security @ hp

--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to [email protected] with
the words "unsubscribe selinux" without quotes as the message.