Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / security / apparmor / resource.c
diff --git a/kernel/security/apparmor/resource.c b/kernel/security/apparmor/resource.c
new file mode 100644 (file)
index 0000000..748bf0c
--- /dev/null
@@ -0,0 +1,154 @@
+/*
+ * AppArmor security module
+ *
+ * This file contains AppArmor resource mediation and attachment
+ *
+ * Copyright (C) 1998-2008 Novell/SUSE
+ * Copyright 2009-2010 Canonical Ltd.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation, version 2 of the
+ * License.
+ */
+
+#include <linux/audit.h>
+
+#include "include/audit.h"
+#include "include/context.h"
+#include "include/resource.h"
+#include "include/policy.h"
+
+/*
+ * Table of rlimit names: we generate it from resource.h.
+ */
+#include "rlim_names.h"
+
+struct aa_fs_entry aa_fs_entry_rlimit[] = {
+       AA_FS_FILE_STRING("mask", AA_FS_RLIMIT_MASK),
+       { }
+};
+
+/* audit callback for resource specific fields */
+static void audit_cb(struct audit_buffer *ab, void *va)
+{
+       struct common_audit_data *sa = va;
+
+       audit_log_format(ab, " rlimit=%s value=%lu",
+                        rlim_names[sa->aad->rlim.rlim], sa->aad->rlim.max);
+}
+
+/**
+ * audit_resource - audit setting resource limit
+ * @profile: profile being enforced  (NOT NULL)
+ * @resoure: rlimit being auditing
+ * @value: value being set
+ * @error: error value
+ *
+ * Returns: 0 or sa->error else other error code on failure
+ */
+static int audit_resource(struct aa_profile *profile, unsigned int resource,
+                         unsigned long value, int error)
+{
+       struct common_audit_data sa;
+       struct apparmor_audit_data aad = {0,};
+
+       sa.type = LSM_AUDIT_DATA_NONE;
+       sa.aad = &aad;
+       aad.op = OP_SETRLIMIT,
+       aad.rlim.rlim = resource;
+       aad.rlim.max = value;
+       aad.error = error;
+       return aa_audit(AUDIT_APPARMOR_AUTO, profile, GFP_KERNEL, &sa,
+                       audit_cb);
+}
+
+/**
+ * aa_map_resouce - map compiled policy resource to internal #
+ * @resource: flattened policy resource number
+ *
+ * Returns: resource # for the current architecture.
+ *
+ * rlimit resource can vary based on architecture, map the compiled policy
+ * resource # to the internal representation for the architecture.
+ */
+int aa_map_resource(int resource)
+{
+       return rlim_map[resource];
+}
+
+/**
+ * aa_task_setrlimit - test permission to set an rlimit
+ * @profile - profile confining the task  (NOT NULL)
+ * @task - task the resource is being set on
+ * @resource - the resource being set
+ * @new_rlim - the new resource limit  (NOT NULL)
+ *
+ * Control raising the processes hard limit.
+ *
+ * Returns: 0 or error code if setting resource failed
+ */
+int aa_task_setrlimit(struct aa_profile *profile, struct task_struct *task,
+                     unsigned int resource, struct rlimit *new_rlim)
+{
+       struct aa_profile *task_profile;
+       int error = 0;
+
+       rcu_read_lock();
+       task_profile = aa_get_profile(aa_cred_profile(__task_cred(task)));
+       rcu_read_unlock();
+
+       /* TODO: extend resource control to handle other (non current)
+        * profiles.  AppArmor rules currently have the implicit assumption
+        * that the task is setting the resource of a task confined with
+        * the same profile.
+        */
+       if (profile != task_profile ||
+           (profile->rlimits.mask & (1 << resource) &&
+            new_rlim->rlim_max > profile->rlimits.limits[resource].rlim_max))
+               error = -EACCES;
+
+       aa_put_profile(task_profile);
+
+       return audit_resource(profile, resource, new_rlim->rlim_max, error);
+}
+
+/**
+ * __aa_transition_rlimits - apply new profile rlimits
+ * @old: old profile on task  (NOT NULL)
+ * @new: new profile with rlimits to apply  (NOT NULL)
+ */
+void __aa_transition_rlimits(struct aa_profile *old, struct aa_profile *new)
+{
+       unsigned int mask = 0;
+       struct rlimit *rlim, *initrlim;
+       int i;
+
+       /* for any rlimits the profile controlled reset the soft limit
+        * to the less of the tasks hard limit and the init tasks soft limit
+        */
+       if (old->rlimits.mask) {
+               for (i = 0, mask = 1; i < RLIM_NLIMITS; i++, mask <<= 1) {
+                       if (old->rlimits.mask & mask) {
+                               rlim = current->signal->rlim + i;
+                               initrlim = init_task.signal->rlim + i;
+                               rlim->rlim_cur = min(rlim->rlim_max,
+                                                    initrlim->rlim_cur);
+                       }
+               }
+       }
+
+       /* set any new hard limits as dictated by the new profile */
+       if (!new->rlimits.mask)
+               return;
+       for (i = 0, mask = 1; i < RLIM_NLIMITS; i++, mask <<= 1) {
+               if (!(new->rlimits.mask & mask))
+                       continue;
+
+               rlim = current->signal->rlim + i;
+               rlim->rlim_max = min(rlim->rlim_max,
+                                    new->rlimits.limits[i].rlim_max);
+               /* soft limit should not exceed hard limit */
+               rlim->rlim_cur = min(rlim->rlim_cur, rlim->rlim_max);
+       }
+}