[apparmor] [PATCH 19/27] apparmor: convert profile lists to RCU based locking

John Johansen john.johansen at canonical.com
Wed Nov 21 04:39:59 UTC 2012


signed-offby: John Johansen <john.johansen at canonical.com>
---
 security/apparmor/domain.c         |    8 +-
 security/apparmor/include/policy.h |   17 +--
 security/apparmor/policy.c         |  199 +++++++++++++++++++-----------------
 security/apparmor/policy_unpack.c  |    5 +
 4 files changed, 125 insertions(+), 104 deletions(-)

diff --git a/security/apparmor/domain.c b/security/apparmor/domain.c
index b5deff1..9ed217a 100644
--- a/security/apparmor/domain.c
+++ b/security/apparmor/domain.c
@@ -144,7 +144,7 @@ static struct aa_profile *__attach_match(const char *name,
 	int len = 0;
 	struct aa_profile *profile, *candidate = NULL;
 
-	list_for_each_entry(profile, head, base.list) {
+	list_for_each_entry_rcu(profile, head, base.list) {
 		if (profile->flags & PFLAG_NULL)
 			continue;
 		if (profile->xmatch && profile->xmatch_len > len) {
@@ -177,9 +177,9 @@ static struct aa_profile *find_attach(struct aa_namespace *ns,
 {
 	struct aa_profile *profile;
 
-	read_lock(&ns->lock);
+	rcu_read_lock();
 	profile = aa_get_profile(__attach_match(name, list));
-	read_unlock(&ns->lock);
+	rcu_read_unlock();
 
 	return profile;
 }
@@ -648,7 +648,7 @@ int aa_change_hat(const char *hats[], int count, u64 token, bool permtest)
 			hat = aa_find_child(root, hats[i]);
 		if (!hat) {
 			if (!COMPLAIN_MODE(root) || permtest) {
-				if (list_empty(&root->base.profiles))
+				if (list_empty_rcu(&root->base.profiles))
 					error = -ECHILD;
 				else
 					error = -ENOENT;
diff --git a/security/apparmor/include/policy.h b/security/apparmor/include/policy.h
index aadcbf8..0d16d91 100644
--- a/security/apparmor/include/policy.h
+++ b/security/apparmor/include/policy.h
@@ -42,6 +42,7 @@ extern const char *const profile_mode_names[];
 
 #define PROFILE_IS_HAT(_profile) ((_profile)->flags & PFLAG_HAT)
 
+#define list_empty_rcu(X) (list_empty(X) || (X)->prev == LIST_POISON2)
 /*
  * FIXME: currently need a clean way to replace and remove profiles as a
  * set.  It should be done at the namespace level.
@@ -75,6 +76,7 @@ struct aa_profile;
  * @hname - The hierarchical name
  * @count: reference count of the obj
  * @list: list policy object is on
+ * @rcu: rcu head used when removing from @list
  * @profiles: head of the profiles list contained in the object
  */
 struct aa_policy {
@@ -83,6 +85,7 @@ struct aa_policy {
 	struct kref count;
 	struct list_head list;
 	struct list_head profiles;
+	struct rcu_head rcu;
 };
 
 /* struct aa_ns_acct - accounting of profiles in namespace
@@ -123,9 +126,9 @@ struct aa_ns_acct {
 struct aa_namespace {
 	struct aa_policy base;
 	struct aa_namespace *parent;
-	rwlock_t lock;
+	struct mutex lock;
 	struct aa_ns_acct acct;
-	struct aa_profile *unconfined;
+	struct aa_profile __rcu *unconfined;
 	struct list_head sub_ns;
 
 	atomic_t uniq_null;
@@ -166,7 +169,7 @@ struct aa_policydb {
  * attachments are determined by profile X transition rules.
  *
  * The @replacedby field is write protected by the profile lock.  Reads
- * are assumed to be atomic, and are done without locking.
+ * are assumed to be atomic, and are done with rcu.
  *
  * Profiles have a hierarchy where hats and children profiles keep
  * a reference to their parent.
@@ -177,10 +180,10 @@ struct aa_policydb {
  */
 struct aa_profile {
 	struct aa_policy base;
-	struct aa_profile *parent;
+	struct aa_profile __rcu *parent;
 
 	struct aa_namespace *ns;
-	struct aa_profile *replacedby;
+	struct aa_profile __rcu *replacedby;
 	const char *rename;
 
 	struct aa_dfa *xmatch;
@@ -274,8 +277,8 @@ ssize_t aa_remove_profiles(char *name, size_t size);
  */
 static inline struct aa_profile *aa_newest_version(struct aa_profile *profile)
 {
-	while (profile->replacedby)
-		profile = profile->replacedby;
+	while (rcu_dereference(profile->replacedby))
+		profile = rcu_dereference(profile->replacedby);
 
 	return profile;
 }
diff --git a/security/apparmor/policy.c b/security/apparmor/policy.c
index ea4653d..4d3f8ba 100644
--- a/security/apparmor/policy.c
+++ b/security/apparmor/policy.c
@@ -153,13 +153,13 @@ static bool policy_init(struct aa_policy *policy, const char *prefix,
 static void policy_destroy(struct aa_policy *policy)
 {
 	/* still contains profiles -- invalid */
-	if (!list_empty(&policy->profiles)) {
+	if (!list_empty_rcu(&policy->profiles)) {
 		AA_ERROR("%s: internal error, "
 			 "policy '%s' still contains profiles\n",
 			 __func__, policy->name);
 		BUG();
 	}
-	if (!list_empty(&policy->list)) {
+	if (!list_empty_rcu(&policy->list)) {
 		AA_ERROR("%s: internal error, policy '%s' still on list\n",
 			 __func__, policy->name);
 		BUG();
@@ -174,7 +174,7 @@ static void policy_destroy(struct aa_policy *policy)
  * @head: list to search  (NOT NULL)
  * @name: name to search for  (NOT NULL)
  *
- * Requires: correct locks for the @head list be held
+ * Requires: rcu_read_lock be held
  *
  * Returns: unrefcounted policy that match @name or NULL if not found
  */
@@ -182,7 +182,7 @@ static struct aa_policy *__policy_find(struct list_head *head, const char *name)
 {
 	struct aa_policy *policy;
 
-	list_for_each_entry(policy, head, list) {
+	list_for_each_entry_rcu(policy, head, list) {
 		if (!strcmp(policy->name, name))
 			return policy;
 	}
@@ -195,7 +195,7 @@ static struct aa_policy *__policy_find(struct list_head *head, const char *name)
  * @str: string to search for  (NOT NULL)
  * @len: length of match required
  *
- * Requires: correct locks for the @head list be held
+ * Requires: rcu_read_lock be held
  *
  * Returns: unrefcounted policy that match @str or NULL if not found
  *
@@ -207,7 +207,7 @@ static struct aa_policy *__policy_strn_find(struct list_head *head,
 {
 	struct aa_policy *policy;
 
-	list_for_each_entry(policy, head, list) {
+	list_for_each_entry_rcu(policy, head, list) {
 		if (aa_strneq(policy->name, str, len))
 			return policy;
 	}
@@ -284,7 +284,7 @@ static struct aa_namespace *alloc_namespace(const char *prefix,
 		goto fail_ns;
 
 	INIT_LIST_HEAD(&ns->sub_ns);
-	rwlock_init(&ns->lock);
+	mutex_init(&ns->lock);
 
 	/* released by free_namespace */
 	ns->unconfined = aa_alloc_profile("unconfined");
@@ -350,7 +350,7 @@ void aa_free_namespace_kref(struct kref *kref)
  *
  * Returns: unrefcounted namespace
  *
- * Requires: ns lock be held
+ * Requires: rcu_read_lock be held
  */
 static struct aa_namespace *__aa_find_namespace(struct list_head *head,
 						const char *name)
@@ -373,9 +373,9 @@ struct aa_namespace *aa_find_namespace(struct aa_namespace *root,
 {
 	struct aa_namespace *ns = NULL;
 
-	read_lock(&root->lock);
+	rcu_read_lock();
 	ns = aa_get_namespace(__aa_find_namespace(&root->sub_ns, name));
-	read_unlock(&root->lock);
+	rcu_read_unlock();
 
 	return ns;
 }
@@ -392,7 +392,7 @@ static struct aa_namespace *aa_prepare_namespace(const char *name)
 
 	root = aa_current_profile()->ns;
 
-	write_lock(&root->lock);
+	mutex_lock(&root->lock);
 
 	/* if name isn't specified the profile is loaded to the current ns */
 	if (!name) {
@@ -404,32 +404,18 @@ static struct aa_namespace *aa_prepare_namespace(const char *name)
 	/* try and find the specified ns and if it doesn't exist create it */
 	/* released by caller */
 	ns = aa_get_namespace(__aa_find_namespace(&root->sub_ns, name));
-	if (!ns) {
-		/* namespace not found */
-		struct aa_namespace *new_ns;
-		write_unlock(&root->lock);
-		new_ns = alloc_namespace(root->base.hname, name);
-		if (!new_ns)
-			return NULL;
-		write_lock(&root->lock);
-		/* test for race when new_ns was allocated */
-		ns = __aa_find_namespace(&root->sub_ns, name);
-		if (!ns) {
-			/* add parent ref */
-			new_ns->parent = aa_get_namespace(root);
-
-			list_add(&new_ns->base.list, &root->sub_ns);
-			/* add list ref */
-			ns = aa_get_namespace(new_ns);
-		} else {
-			/* raced so free the new one */
-			free_namespace(new_ns);
-			/* get reference on namespace */
-			aa_get_namespace(ns);
-		}
+	if (!ns)
+		ns = alloc_namespace(root->base.hname, name);
+
+	if (ns) {
+		/* add parent ref */
+		ns->parent = aa_get_namespace(root);
+		list_add_rcu(&ns->base.list, &root->sub_ns);
+		/* add list ref */
+		aa_get_namespace(ns);
 	}
 out:
-	write_unlock(&root->lock);
+	mutex_unlock(&root->lock);
 
 	/* return ref */
 	return ns;
@@ -447,7 +433,7 @@ out:
 static void __list_add_profile(struct list_head *list,
 			       struct aa_profile *profile)
 {
-	list_add(&profile->base.list, list);
+	list_add_rcu(&profile->base.list, list);
 	/* get list reference */
 	aa_get_profile(profile);
 }
@@ -466,10 +452,8 @@ static void __list_add_profile(struct list_head *list,
  */
 static void __list_remove_profile(struct aa_profile *profile)
 {
-	list_del_init(&profile->base.list);
-	if (!(profile->flags & PFLAG_NO_LIST_REF))
-		/* release list reference */
-		aa_put_profile(profile);
+	list_del_rcu(&profile->base.list);
+	aa_put_profile(profile);
 }
 
 /**
@@ -494,18 +478,26 @@ static void __replace_profile(struct aa_profile *old, struct aa_profile *new)
 	else
 		policy = &old->ns->base;
 
-	__list_add_profile(&policy->profiles, new);
-	/* inherit children */
-	list_for_each_entry_safe(child, tmp, &old->base.profiles, base.list) {
-		aa_put_profile(child->parent);
-		child->parent = aa_get_profile(new);
-		/* list refcount transferred to @new*/
-		list_move(&child->base.list, &new->base.profiles);
+	if (!list_empty_rcu(&old->base.profiles)) {
+		/* inherit children */
+		list_for_each_entry_safe(child, tmp, &old->base.profiles,
+					 base.list) {
+			struct aa_profile *parent = child->parent;
+			rcu_assign_pointer(child->parent, aa_get_profile(new));
+			aa_put_profile(parent);
+		}
+		/* this really should be replaced with a something that doesn't
+		 * use rcu sync as synching at each profile load will slow
+		 * profile loads significantly
+		 */
+		list_splice_init_rcu(&old->base.profiles, &new->base.profiles, synchronize_rcu);
 	}
 
 	/* released by free_profile */
-	old->replacedby = aa_get_profile(new);
-	__list_remove_profile(old);
+	rcu_assign_pointer(old->replacedby, aa_get_profile(new));
+	list_replace_rcu(&old->base.list, &new->base.list);
+	aa_get_profile(new);
+	aa_put_profile(old);
 }
 
 static void __profile_list_release(struct list_head *head);
@@ -520,8 +512,9 @@ static void __remove_profile(struct aa_profile *profile)
 {
 	/* release any children lists first */
 	__profile_list_release(&profile->base.profiles);
-	/* released by free_profile */
-	profile->replacedby = aa_get_profile(profile->ns->unconfined);
+	/* replacedby released by free_profile */
+	rcu_assign_pointer(profile->replacedby,
+			   aa_get_profile(profile->ns->unconfined));
 	__list_remove_profile(profile);
 }
 
@@ -546,17 +539,40 @@ static void __ns_list_release(struct list_head *head);
  */
 static void destroy_namespace(struct aa_namespace *ns)
 {
+	struct aa_profile *unconfined;
+
 	if (!ns)
 		return;
 
-	write_lock(&ns->lock);
+	mutex_lock(&ns->lock);
 	/* release all profiles in this namespace */
 	__profile_list_release(&ns->base.profiles);
 
 	/* release all sub namespaces */
 	__ns_list_release(&ns->sub_ns);
 
-	write_unlock(&ns->lock);
+	unconfined = ns->unconfined;
+	/*
+	 * break the ns, unconfined profile cyclic reference and forward
+	 * all new unconfined profiles requests to the parent namespace
+	 * This will result in all confined tasks that have a profile
+	 * being removed, inheriting the parent->unconfined profile.
+	 */
+	if (ns->parent)
+		ns->unconfined = aa_get_profile(ns->parent->unconfined);
+
+	/* release original ns->unconfined ref */
+	aa_put_profile(unconfined);
+
+	mutex_unlock(&ns->lock);
+}
+
+void aa_put_ns_rcu(struct rcu_head *head)
+{
+	struct aa_namespace *ns = container_of(head, struct aa_namespace,
+					       base.rcu);
+	/* release ns->base.list ref */
+	aa_put_namespace(ns);
 }
 
 /**
@@ -567,26 +583,12 @@ static void destroy_namespace(struct aa_namespace *ns)
  */
 static void __remove_namespace(struct aa_namespace *ns)
 {
-	struct aa_profile *unconfined = ns->unconfined;
-
 	/* remove ns from namespace list */
-	list_del_init(&ns->base.list);
-
-	/*
-	 * break the ns, unconfined profile cyclic reference and forward
-	 * all new unconfined profiles requests to the parent namespace
-	 * This will result in all confined tasks that have a profile
-	 * being removed, inheriting the parent->unconfined profile.
-	 */
-	if (ns->parent)
-		ns->unconfined = aa_get_profile(ns->parent->unconfined);
+	list_del_rcu(&ns->base.list);
 
 	destroy_namespace(ns);
 
-	/* release original ns->unconfined ref */
-	aa_put_profile(unconfined);
-	/* release ns->base.list ref, from removal above */
-	aa_put_namespace(ns);
+	call_rcu(&ns->base.rcu, aa_put_ns_rcu);
 }
 
 /**
@@ -650,7 +652,7 @@ static void free_profile(struct aa_profile *profile)
 	if (!profile)
 		return;
 
-	if (!list_empty(&profile->base.list)) {
+	if (!list_empty_rcu(&profile->base.list)) {
 		AA_ERROR("%s: internal error, "
 			 "profile '%s' still on ns list\n",
 			 __func__, profile->base.name);
@@ -678,12 +680,12 @@ static void free_profile(struct aa_profile *profile)
 	 * call (it looks like recursion, with free_profile calling
 	 * free_profile) for each profile in the chain lp#1056078.
 	 */
-	for (p = profile->replacedby; p; ) {
+	for (p = rcu_dereference(profile->replacedby); p; ) {
 		if (atomic_dec_and_test(&p->base.count.refcount)) {
 			/* no more refs on p, grab its replacedby */
-			struct aa_profile *next = p->replacedby;
+			struct aa_profile *next = rcu_dereference(p->replacedby);
 			/* break the chain */
-			p->replacedby = NULL;
+			rcu_assign_pointer(p->replacedby, NULL);
 			/* now free p, chain is broken */
 			free_profile(p);
 
@@ -697,6 +699,16 @@ static void free_profile(struct aa_profile *profile)
 }
 
 /**
+ * aa_free_profile_rcu - free aa_profile by rcu (called by aa_free_profile_kref)
+ * @head: rcu_head callback for freeing of a profile  (NOT NULL)
+ */
+void aa_free_profile_rcu(struct rcu_head *head)
+{
+	struct aa_profile *p = container_of(head, struct aa_profile, base.rcu);
+	free_profile(p);
+}
+
+/**
  * aa_free_profile_kref - free aa_profile by kref (called by aa_put_profile)
  * @kr: kref callback for freeing of a profile  (NOT NULL)
  */
@@ -704,8 +716,7 @@ void aa_free_profile_kref(struct kref *kref)
 {
 	struct aa_profile *p = container_of(kref, struct aa_profile,
 					    base.count);
-
-	free_profile(p);
+	call_rcu(&p->base.rcu, aa_free_profile_rcu);
 }
 
 /**
@@ -772,9 +783,9 @@ struct aa_profile *aa_new_null_profile(struct aa_profile *parent, int hat)
 	profile->parent = aa_get_profile(parent);
 	profile->ns = aa_get_namespace(parent->ns);
 
-	write_lock(&profile->ns->lock);
+	mutex_lock(&profile->ns->lock);
 	__list_add_profile(&parent->base.profiles, profile);
-	write_unlock(&profile->ns->lock);
+	mutex_unlock(&profile->ns->lock);
 
 	/* refcount released by caller */
 	return profile;
@@ -790,7 +801,7 @@ fail:
  * @head: list to search  (NOT NULL)
  * @name: name of profile (NOT NULL)
  *
- * Requires: ns lock protecting list be held
+ * Requires: rcu_read_lock be held
  *
  * Returns: unrefcounted profile ptr, or NULL if not found
  */
@@ -805,7 +816,7 @@ static struct aa_profile *__find_child(struct list_head *head, const char *name)
  * @name: name of profile (NOT NULL)
  * @len: length of @name substring to match
  *
- * Requires: ns lock protecting list be held
+ * Requires: rcu_read_lock be held
  *
  * Returns: unrefcounted profile ptr, or NULL if not found
  */
@@ -826,9 +837,9 @@ struct aa_profile *aa_find_child(struct aa_profile *parent, const char *name)
 {
 	struct aa_profile *profile;
 
-	read_lock(&parent->ns->lock);
+	rcu_read_lock();
 	profile = aa_get_profile(__find_child(&parent->base.profiles, name));
-	read_unlock(&parent->ns->lock);
+	rcu_read_unlock();
 
 	/* refcount released by caller */
 	return profile;
@@ -843,7 +854,7 @@ struct aa_profile *aa_find_child(struct aa_profile *parent, const char *name)
  * that matches hname does not need to exist, in general this
  * is used to load a new profile.
  *
- * Requires: ns->lock be held
+ * Requires: rcu_read_lock be held
  *
  * Returns: unrefcounted policy or NULL if not found
  */
@@ -875,7 +886,7 @@ static struct aa_policy *__lookup_parent(struct aa_namespace *ns,
  * @base: base list to start looking up profile name from  (NOT NULL)
  * @hname: hierarchical profile name  (NOT NULL)
  *
- * Requires: ns->lock be held
+ * Requires: rcu_read_lock be held
  *
  * Returns: unrefcounted profile pointer or NULL if not found
  *
@@ -914,9 +925,9 @@ struct aa_profile *aa_lookup_profile(struct aa_namespace *ns, const char *hname)
 {
 	struct aa_profile *profile;
 
-	read_lock(&ns->lock);
+	rcu_read_lock();
 	profile = aa_get_profile(__lookup_profile(&ns->base, hname));
-	read_unlock(&ns->lock);
+	rcu_read_unlock();
 
 	/* the unconfined profile is not in the regular profile list */
 	if (!profile && strcmp(hname, "unconfined") == 0)
@@ -1091,13 +1102,13 @@ ssize_t aa_replace_profiles(void *udata, size_t size, bool noreplace)
 		goto fail;
 	}
 
-	write_lock(&ns->lock);
+	mutex_lock(&ns->lock);
 	/* test to see if replacement can be done without failure */
 	list_for_each_entry(new, &lh, base.list) {
 		policy = __lookup_replace(ns, new, noreplace, NULL, NULL,
 					  &info);
 		if (IS_ERR(policy)) {
-			write_unlock(&ns->lock);
+			mutex_unlock(&ns->lock);
 			error = PTR_ERR(policy);
 			name = new->base.hname;
 			goto fail;
@@ -1113,6 +1124,7 @@ ssize_t aa_replace_profiles(void *udata, size_t size, bool noreplace)
 	list_for_each_entry_safe(new, tmp, &lh, base.list) {
 		struct aa_profile *old, *rename;
 
+		/* &lh list is private and not rcu based */
 		list_del_init(&new->base.list);
 		policy = __lookup_replace(ns, new, noreplace, &old, &rename,
 					  &info);
@@ -1131,7 +1143,7 @@ ssize_t aa_replace_profiles(void *udata, size_t size, bool noreplace)
 		aa_put_profile(old);
 		aa_put_profile(new);
 	}
-	write_unlock(&ns->lock);
+	mutex_unlock(&ns->lock);
 
 out:
 	aa_put_namespace(ns);
@@ -1144,6 +1156,7 @@ fail:
 	error = audit_policy(op, GFP_KERNEL, name, info, error);
 
 	list_for_each_entry_safe(new, tmp, &lh, base.list) {
+		/* &lh list is private and not rcu based */
 		list_del_init(&new->base.list);
 		aa_put_profile(new);
 	}
@@ -1196,12 +1209,12 @@ ssize_t aa_remove_profiles(char *fqname, size_t size)
 
 	if (!name) {
 		/* remove namespace - can only happen if fqname[0] == ':' */
-		write_lock(&ns->parent->lock);
+		mutex_lock(&ns->parent->lock);
 		__remove_namespace(ns);
-		write_unlock(&ns->parent->lock);
+		mutex_unlock(&ns->parent->lock);
 	} else {
 		/* remove profile */
-		write_lock(&ns->lock);
+		mutex_lock(&ns->lock);
 		profile = aa_get_profile(__lookup_profile(&ns->base, name));
 		if (!profile) {
 			error = -ENOENT;
@@ -1210,7 +1223,7 @@ ssize_t aa_remove_profiles(char *fqname, size_t size)
 		}
 		name = profile->base.hname;
 		__remove_profile(profile);
-		write_unlock(&ns->lock);
+		mutex_unlock(&ns->lock);
 	}
 
 	/* don't fail removal if audit fails */
@@ -1220,7 +1233,7 @@ ssize_t aa_remove_profiles(char *fqname, size_t size)
 	return size;
 
 fail_ns_lock:
-	write_unlock(&ns->lock);
+	mutex_unlock(&ns->lock);
 	aa_put_namespace(ns);
 
 fail:
diff --git a/security/apparmor/policy_unpack.c b/security/apparmor/policy_unpack.c
index 3d2d0d7..9da2506 100644
--- a/security/apparmor/policy_unpack.c
+++ b/security/apparmor/policy_unpack.c
@@ -754,6 +754,11 @@ int aa_unpack(void *udata, size_t size, struct list_head *lh, const char **ns)
 	return 0;
 
 fail:
+	/*
+	 * outside references to the list of profiles are not possible
+	 * as they have not been added to the global list yet, so rcu
+	 * is not necessary
+	 */
 	list_for_each_entry_safe(profile, tmp, lh, base.list) {
 			list_del_init(&profile->base.list);
 			aa_put_profile(profile);
-- 
1.7.10.4




More information about the AppArmor mailing list