selinux: Measure state and hash of policy using IMA

Critical data structures of security modules are currently not measured. Therefore an attestation service, for instance, would not be able to attest whether the security modules are always operating with the policies and configurations that the system administrator had setup. The policies and configurations for the security modules could be tampered by rogue user mode agents or modified through some inadvertent actions on the system. Measuring such critical data would enable an attestation service to reliably assess the security configuration of the system.

SELinux configuration and policy are some of the critical data for this security module that need to be measured. This measurement can be used by an attestation service, for instance, to verify if the configurations and policies have been setup correctly and that they haven't been tampered at run-time.

Measure SELinux configurations, policy capabilities settings, and the hash of the loaded policy by calling the IMA hook ima_measure_critical_data(). Since the size of the loaded policy can be large (several MB), measure the hash of the policy instead of the entire policy to avoid bloating the IMA log entry.

Add "selinux" to the list of supported data sources maintained by IMA to enable measuring SELinux data.

Since SELinux calls the IMA hook to measure data before
a custom IMA policy is loaded, enable queuing if CONFIG_SECURITY_SELINUX is enabled, to defer processing SELinux data until a custom IMA policy is loaded.

Sample measurement of SELinux state and hash of the policy:

10 e32e...5ac3 ima-buf sha256:86e8...4594 selinux-state-1595389364:287899386 696e697469616c697a65643d313b656e61626c65643d313b656e666f7263696e673d303b636865636b72657170726f743d313b6e6574776f726b5f706565725f636f6e74726f6c733d313b6f70656e5f7065726d733d313b657874656e6465645f736f636b65745f636c6173733d313b616c776179735f636865636b5f6e6574776f726b3d303b6367726f75705f7365636c6162656c3d313b6e6e705f6e6f737569645f7472616e736974696f6e3d313b67656e66735f7365636c6162656c5f73796d6c696e6b733d303 10 9e81...0857 ima-buf sha256:4941...68fc selinux-policy-hash-1597335667:462051628 8d1d...1834

To verify the measurement check the following:

Execute the following command to extract the measured data from the IMA log for SELinux configuration (selinux-state).

grep -m 1 "selinux-state" /sys/kernel/security/integrity/ima/ascii_runtime_measurements | cut -d' ' -f 6 | xxd -r -p

The output should be the list of key-value pairs. For example,
initialized=1;enabled=1;enforcing=0;checkreqprot=1;network_peer_controls=1;open_perms=1;extended_socket_class=1;always_check_network=0;cgroup_seclabel=1;nnp_nosuid_transition=1;genfs_seclabel_symlinks=0;

To verify the measured data with the current SELinux state:

=> enabled should be set to 1 if /sys/fs/selinux folder exists,
0 otherwise

For other entries, compare the integer value in the files
=> /sys/fs/selinux/enforce
=> /sys/fs/selinux/checkreqprot
And, each of the policy capabilities files under
=> /sys/fs/selinux/policy_capabilities

For selinux-policy-hash, the hash of SELinux policy is included in the IMA log entry.

To verify the measured data with the current SELinux policy run the following commands and verify the output hash values match.

sha256sum /sys/fs/selinux/policy | cut -d' ' -f 1

grep -m 1 "selinux-policy-hash" /sys/kernel/security/integrity/ima/ascii_runtime_measurements | cut -d' ' -f 6

Signed-off-by Lakshmi Ramasubramanian <[email protected]>

Suggested-by Stephen Smalley <[email protected]>

---

security/selinux/measure.c new file +67 -0

new file mode 100644
index 000000000000..b29baaa271f0
@@ -0,0 +1,154 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Measure SELinux state using IMA subsystem.
+ */
+#include <linux/vmalloc.h>
+#include <linux/ktime.h>
+#include <linux/ima.h>
+#include "security.h"
+
+/*
+ * This function creates a unique name by appending the timestamp to
+ * the given string. This string is passed as "event_name" to the IMA
+ * hook to measure the given SELinux data.
+ *
+ * The data provided by SELinux to the IMA subsystem for measuring may have
+ * already been measured (for instance the same state existed earlier).
+ * But for SELinux the current data represents a state change and hence
+ * needs to be measured again. To enable this, pass a unique "event_name"
+ * to the IMA hook so that IMA subsystem will always measure the given data.
+ *
+ * For example,
+ * At time T0 SELinux data to be measured is "foo". IMA measures it.
+ * At time T1 the data is changed to "bar". IMA measures it.
+ * At time T2 the data is changed to "foo" again. IMA will not measure it
+ * (since it was already measured) unless the event_name, for instance,
+ * is different in this call.
+ */
+static char *selinux_event_name(const char *name_prefix)
+{
+       char *event_name = NULL;
+       struct timespec64 cur_time;
+
+       ktime_get_real_ts64(&cur_time);
+       event_name = kasprintf(GFP_KERNEL, "%s-%lld:%09ld", name_prefix,
+                              cur_time.tv_sec, cur_time.tv_nsec);
+       if (!event_name) {
+               pr_err("%s: event name not allocated.\n", __func__);
+               return NULL;
+       }
+
+       return event_name;
+}
+
+static int read_selinux_state(char **state_str, int *state_str_len,
+                             struct selinux_state *state)
+{
+       char *buf, *str_fmt = "%s=%d;";
+       int i, buf_len, curr;
+       bool initialized = selinux_initialized(state);
+       bool enabled = !selinux_disabled(state);
+       bool enforcing = enforcing_enabled(state);
+       bool checkreqprot = checkreqprot_get(state);
+
+       buf_len = snprintf(NULL, 0, str_fmt, "initialized", initialized);
+       buf_len += snprintf(NULL, 0, str_fmt, "enabled", enabled);
+       buf_len += snprintf(NULL, 0, str_fmt, "enforcing", enforcing);
+       buf_len += snprintf(NULL, 0, str_fmt, "checkreqprot", checkreqprot);
+
+       for (i = 0; i < __POLICYDB_CAPABILITY_MAX; i++) {
+               buf_len += snprintf(NULL, 0, str_fmt,
+                                   selinux_policycap_names[i],
+                                   state->policycap[i]);
+       }
+       ++buf_len;
+
+       buf = kzalloc(buf_len, GFP_KERNEL);
+       if (!buf)

security/selinux/measure.c new file +31 -0

+       }
+
+       *state_str = buf;
+       *state_str_len = curr;
+
+       return 0;
+}
+
+/*
+ * selinux_measure_state - Measure SELinux state configuration and hash of
+ *                        the SELinux policy.
+ * @state: selinux state struct
+ *
+ * NOTE: This function must be called with policy_mutex held.
+ */
+void selinux_measure_state(struct selinux_state *state)
+{
+       void *policy = NULL;
+       char *state_event_name = NULL;
+       char *policy_event_name = NULL;
+       char *state_str = NULL;
+       size_t policy_len;
+       int state_str_len, rc = 0;
+       bool initialized = selinux_initialized(state);
+
+       rc = read_selinux_state(&state_str, &state_str_len, state);
+       if (rc) {
+               pr_err("%s: Failed to read selinux state.\n", __func__);
+               return;
+       }
+

security/selinux/measure.c new file +11 -0

+
+       ima_measure_critical_data(state_event_name, "selinux",
+                                 state_str, state_str_len, false);
+
+       /*
+        * Measure SELinux policy only after initialization is completed.
+        */
+       if (!initialized)
+               goto out;
+
+       rc = security_read_policy_kernel(state, &policy, &policy_len);

Quoted 22 lines · Lakshmi Ramasubramanian

Critical data structures of security modules are currently not measured. Therefore an attestation service, for instance, would not be able to attest whether the security modules are always operating with the policies and configurations that the system administrator had setup. The policies and configurations for the security modules could be tampered by rogue user mode agents or modified through some inadvertent actions on the system. Measuring such critical data would enable an attestation service to reliably assess the security configuration of the system.

SELinux configuration and policy are some of the critical data for this security module that need to be measured. This measurement can be used by an attestation service, for instance, to verify if the configurations and policies have been setup correctly and that they haven't been tampered at run-time.

Measure SELinux configurations, policy capabilities settings, and the hash of the loaded policy by calling the IMA hook ima_measure_critical_data(). Since the size of the loaded policy can be large (several MB), measure the hash of the policy instead of the entire policy to avoid bloating the IMA log entry.

Quoted 2 lines

Add "selinux" to the list of supported data sources maintained by IMA to enable measuring SELinux data.

Please provide an example /etc/ima/ima-policy snippet for enabling this support, e.g.
measure func=CRITICAL_DATA data_sources=selinux template=ima-buf

Quoted 16 lines

Since SELinux calls the IMA hook to measure data before
a custom IMA policy is loaded, enable queuing if CONFIG_SECURITY_SELINUX is enabled, to defer processing SELinux data until a custom IMA policy is loaded.

Sample measurement of SELinux state and hash of the policy:

10 e32e...5ac3 ima-buf sha256:86e8...4594 selinux-state-1595389364:287899386 696e697469616c697a65643d313b656e61626c65643d313b656e666f7263696e673d303b636865636b72657170726f743d313b6e6574776f726b5f706565725f636f6e74726f6c733d313b6f70656e5f7065726d733d313b657874656e6465645f736f636b65745f636c6173733d313b616c776179735f636865636b5f6e6574776f726b3d303b6367726f75705f7365636c6162656c3d313b6e6e705f6e6f737569645f7472616e736974696f6e3d313b67656e66735f7365636c6162656c5f73796d6c696e6b733d303 10 9e81...0857 ima-buf sha256:4941...68fc selinux-policy-hash-1597335667:462051628 8d1d...1834

To verify the measurement check the following:

Execute the following command to extract the measured data from the IMA log for SELinux configuration (selinux-state).

Quoted 1 line

grep -m 1 "selinux-state" /sys/kernel/security/integrity/ima/ascii_runtime_measurements | cut -d' ' -f 6 | xxd -r -p

NB This will only extract the first such record, which will always be prior to policy load (initialized=0) so that won't be terribly useful. For real verification, they would want to check all such records or at least the last/latest one (depending on their goal).

Quoted 6 lines

The output should be the list of key-value pairs. For example,
initialized=1;enabled=1;enforcing=0;checkreqprot=1;network_peer_controls=1;open_perms=1;extended_socket_class=1;always_check_network=0;cgroup_seclabel=1;nnp_nosuid_transition=1;genfs_seclabel_symlinks=0;

To verify the measured data with the current SELinux state:

Quoted 8 lines

=> enabled should be set to 1 if /sys/fs/selinux folder exists,
0 otherwise

For other entries, compare the integer value in the files
=> /sys/fs/selinux/enforce
=> /sys/fs/selinux/checkreqprot
And, each of the policy capabilities files under
=> /sys/fs/selinux/policy_capabilities

To be clear, actual verification would be against an expected state and done on a system other than the measured system, typically requiring "initialized=1; enabled=1;enforcing=1;checkreqprot=0;" for a secure state and then whatever policy capabilities are actually set in the expected policy (which can be extracted from the policy itself via seinfo or the like).

Quoted 9 lines

For selinux-policy-hash, the hash of SELinux policy is included in the IMA log entry.

To verify the measured data with the current SELinux policy run the following commands and verify the output hash values match.

   sha256sum /sys/fs/selinux/policy | cut -d' ' -f 1

Quoted 1 line

grep -m 1 "selinux-policy-hash" /sys/kernel/security/integrity/ima/ascii_runtime_measurements | cut -d' ' -f 6

As above, this will only extract the first policy load, whereas for real verification they will want to check either all of them or at least the latest/last one. For actual verification, they would need to load the expected policy into an identical kernel on a pristine/known-safe system and run the sha256sum /sys/kernel/selinux/policy there to get the expected hash.

Quoted 86 lines

Signed-off-by Lakshmi Ramasubramanian <[email protected]>

Suggested-by Stephen Smalley <[email protected]>

---

security/selinux/measure.c new file +75 -0

new file mode 100644
index 000000000000..b29baaa271f0
@@ -0,0 +1,154 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Measure SELinux state using IMA subsystem.
+ */
+#include <linux/vmalloc.h>
+#include <linux/ktime.h>
+#include <linux/ima.h>
+#include "security.h"
+
+/*
+ * This function creates a unique name by appending the timestamp to
+ * the given string. This string is passed as "event_name" to the IMA
+ * hook to measure the given SELinux data.
+ *
+ * The data provided by SELinux to the IMA subsystem for measuring may have
+ * already been measured (for instance the same state existed earlier).
+ * But for SELinux the current data represents a state change and hence
+ * needs to be measured again. To enable this, pass a unique "event_name"
+ * to the IMA hook so that IMA subsystem will always measure the given data.
+ *
+ * For example,
+ * At time T0 SELinux data to be measured is "foo". IMA measures it.
+ * At time T1 the data is changed to "bar". IMA measures it.
+ * At time T2 the data is changed to "foo" again. IMA will not measure it
+ * (since it was already measured) unless the event_name, for instance,
+ * is different in this call.
+ */
+static char *selinux_event_name(const char *name_prefix)
+{
+       char *event_name = NULL;
+       struct timespec64 cur_time;
+
+       ktime_get_real_ts64(&cur_time);
+       event_name = kasprintf(GFP_KERNEL, "%s-%lld:%09ld", name_prefix,
+                              cur_time.tv_sec, cur_time.tv_nsec);
+       if (!event_name) {
+               pr_err("%s: event name not allocated.\n", __func__);
+               return NULL;
+       }
+
+       return event_name;
+}
+
+static int read_selinux_state(char **state_str, int *state_str_len,
+                             struct selinux_state *state)
+{
+       char *buf, *str_fmt = "%s=%d;";
+       int i, buf_len, curr;
+       bool initialized = selinux_initialized(state);
+       bool enabled = !selinux_disabled(state);
+       bool enforcing = enforcing_enabled(state);
+       bool checkreqprot = checkreqprot_get(state);
+
+       buf_len = snprintf(NULL, 0, str_fmt, "initialized", initialized);
+       buf_len += snprintf(NULL, 0, str_fmt, "enabled", enabled);
+       buf_len += snprintf(NULL, 0, str_fmt, "enforcing", enforcing);
+       buf_len += snprintf(NULL, 0, str_fmt, "checkreqprot", checkreqprot);
+
+       for (i = 0; i < __POLICYDB_CAPABILITY_MAX; i++) {
+               buf_len += snprintf(NULL, 0, str_fmt,
+                                   selinux_policycap_names[i],
+                                   state->policycap[i]);
+       }
+       ++buf_len;
+
+       buf = kzalloc(buf_len, GFP_KERNEL);
+       if (!buf)
+               return -ENOMEM;
+
+       curr = snprintf(buf, buf_len, str_fmt,
+                       "initialized", initialized);
+       curr += snprintf((buf + curr), (buf_len - curr), str_fmt,
+                        "enabled", enabled);
+       curr += snprintf((buf + curr), (buf_len - curr), str_fmt,
+                        "enforcing", enforcing);

security/selinux/measure.c line 77

Quoted 2 lines

security/selinux/measure.c new file +2 -0

+       curr += snprintf((buf + curr), (buf_len - curr), str_fmt,
+                        "checkreqprot", checkreqprot);

Wondering if we should be using scnprintf() when writing to the buffer instead of snprintf() to ensure it returns the actual length written. Technically shouldn't be an issue since we just computed the length above and allocated to that size but might be less prone to error in the future.

security/selinux/measure.c line 82

Quoted 6 lines

security/selinux/measure.c new file +5 -0

+
+       for (i = 0; i < __POLICYDB_CAPABILITY_MAX; i++) {
+               curr += snprintf((buf + curr), (buf_len - curr), str_fmt,
+                                selinux_policycap_names[i],
+                                state->policycap[i]);

Ditto

Quoted 39 lines

security/selinux/measure.c new file +38 -0

+       }
+
+       *state_str = buf;
+       *state_str_len = curr;
+
+       return 0;
+}
+
+/*
+ * selinux_measure_state - Measure SELinux state configuration and hash of
+ *                        the SELinux policy.
+ * @state: selinux state struct
+ *
+ * NOTE: This function must be called with policy_mutex held.
+ */
+void selinux_measure_state(struct selinux_state *state)
+{
+       void *policy = NULL;
+       char *state_event_name = NULL;
+       char *policy_event_name = NULL;
+       char *state_str = NULL;
+       size_t policy_len;
+       int state_str_len, rc = 0;
+       bool initialized = selinux_initialized(state);
+
+       rc = read_selinux_state(&state_str, &state_str_len, state);
+       if (rc) {
+               pr_err("%s: Failed to read selinux state.\n", __func__);
+               return;
+       }
+
+       /*
+        * Get a unique string for measuring the current SELinux state.
+        */
+       state_event_name = selinux_event_name("selinux-state");
+       if (!state_event_name) {
+               pr_err("%s: Event name for state not allocated.\n",
+                      __func__);

security/selinux/measure.c line 123

Quoted 3 lines

security/selinux/measure.c new file +3 -0

+               rc = -ENOMEM;
+               goto out;
+       }

Why get the event name after creating the state string? If memory is under sufficient pressure to cause the event name allocation to fail, then we're going to fail on allocating the state string too and no point in doing the work in that case.

Quoted 16 lines

security/selinux/measure.c new file +15 -0

+
+       ima_measure_critical_data(state_event_name, "selinux",
+                                 state_str, state_str_len, false);
+
+       /*
+        * Measure SELinux policy only after initialization is completed.
+        */
+       if (!initialized)
+               goto out;
+
+       rc = security_read_policy_kernel(state, &policy, &policy_len);
+       if (rc)
+               goto out;
+
+       policy_event_name = selinux_event_name("selinux-policy-hash");

security/selinux/measure.c line 144

Quoted 6 lines

security/selinux/measure.c new file +6 -0

+       if (!policy_event_name) {
+               pr_err("%s: Event name for policy not allocated.\n",
+                      __func__);
+               rc = -ENOMEM;
+               goto out;
+       }

Ditto.

security/selinux/measure.c line 153

Quoted 10 lines

security/selinux/measure.c new file +9 -0

+
+       ima_measure_critical_data(policy_event_name, "selinux",
+                                 policy, policy_len, true);
+
+out:
+       kfree(state_event_name);
+       kfree(policy_event_name);
+       kfree(state_str);
+       vfree(policy);

Can you free them in the reverse order in which they were allocated? And the vfree() likely can get moved before the out:
label if you reverse the order in which the event name and policy get allocated. For some related discussion around error handling and freeing of things, see https://lore.kernel.org/selinux/20200825125130.GA304650@mwanda/
and https://lore.kernel.org/selinux/20200826113148.GA393664@mwanda/.

Quoted 12 lines

security/selinux/measure.c new file +1 -0

+}

security/selinux/selinuxfs.c +2 -0

index 4bde570d56a2..a4f1282f7178 100644
@@ -182,6 +182,10 @@ static ssize_t sel_write_enforce(struct file *file, const char __user *buf,
                 selinux_status_update_setenforce(state, new_value);
                 if (!new_value)
                         call_blocking_lsm_notifier(LSM_POLICY_CHANGE, NULL);
+
+               mutex_lock(&state->policy_mutex);

security/selinux/selinuxfs.c static ssize_t sel_write_enforce(struct file *file, const char __user *buf, · line 188

Quoted 2 lines

security/selinux/selinuxfs.c +2 -0

+               selinux_measure_state(state);
+               mutex_unlock(&state->policy_mutex);

Side bar question for selinux maintainers: should we extend the scope of this mutex to cover the entire operation (or at least everything from reading the old value to setting the new value)? It isn't strictly necessary but might be nicer. Ditto for disable and checkreqprot, although both of those are being deprecated.