[PATCH 2/2] sysctl: lockdep support for sysctl reference counting.

[ebiederm@xmission.com (Eric W. Biederman)]

It is possible for get lock ordering deadlocks between locks
and waiting for the sysctl used count to drop to zero.  We have
recently observed one of these in the networking code.

So teach the sysctl code how to speak lockdep so the kernel
can warn about these kinds of rare issues proactively.

Signed-off-by: Eric Biederman <ebiederm@aristanetworks.com>
---
 include/linux/sysctl.h |    4 ++
 kernel/sysctl.c        |  108 ++++++++++++++++++++++++++++++++++++++---------
 2 files changed, 91 insertions(+), 21 deletions(-)

diff --git a/include/linux/sysctl.h b/include/linux/sysctl.h
index 39d471d..ec9b1dd 100644
--- a/include/linux/sysctl.h
+++ b/include/linux/sysctl.h
@@ -28,6 +28,7 @@
 #include <linux/kernel.h>
 #include <linux/types.h>
 #include <linux/compiler.h>
+#include <linux/lockdep.h>
 
 struct file;
 struct completion;
@@ -1087,6 +1088,9 @@ struct ctl_table_header
 	struct ctl_table *attached_by;
 	struct ctl_table *attached_to;
 	struct ctl_table_header *parent;
+#ifdef CONFIG_PROVE_LOCKING
+	struct lockdep_map dep_map;
+#endif
 };
 
 /* struct ctl_path describes where in the hierarchy a table is added */
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index c5ef44f..ea8cc39 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1454,12 +1454,63 @@ static struct ctl_table dev_table[] = {
 
 static DEFINE_SPINLOCK(sysctl_lock);
 
+#ifndef CONFIG_PROVE_LOCKING
+
+# define lock_sysctl() spin_lock(&sysctl_lock)
+# define unlock_sysctl() spin_unlock(&sysctl_lock)
+
+static inline void table_acquire_use(struct ctl_table_header *hdr) { }
+static inline void table_release_use(struct ctl_table_header *hdr) { }
+static inline void table_acquire(struct ctl_table_header *hdr) { }
+static inline void table_contended(struct ctl_table_header *hdr) { }
+static inline void table_acquired(struct ctl_table_header *hdr) { }
+static inline void table_release(struct ctl_table_header *hdr) { }
+
+#else	/* CONFIG_PROVE_LOCKING */
+
+#  define lock_sysctl() __raw_spin_lock(&sysctl_lock.raw_lock)
+#  define unlock_sysctl() __raw_spin_unlock(&sysctl_lock.raw_lock)
+
+static inline void table_acquire_use(struct ctl_table_header *hdr)
+{
+	lock_acquire(&hdr->dep_map, 0, 0, 1, 2, NULL, _RET_IP_);
+	lock_acquired(&hdr->dep_map, _RET_IP_);
+}
+
+static inline void table_release_use(struct ctl_table_header *hdr)
+{
+	lock_release(&hdr->dep_map, 0, _RET_IP_);
+}
+
+static inline void table_acquire(struct ctl_table_header *hdr)
+{
+	lock_acquire(&hdr->dep_map, 0, 0, 0, 2, NULL, _RET_IP_);
+}
+
+static inline void table_contended(struct ctl_table_header *hdr)
+{
+	lock_contended(&hdr->dep_map, _RET_IP_);
+}
+
+static inline void table_acquired(struct ctl_table_header *hdr)
+{
+	lock_acquired(&hdr->dep_map, _RET_IP_);
+}
+
+static inline void table_release(struct ctl_table_header *hdr)
+{
+	lock_release(&hdr->dep_map, 0, _RET_IP_);
+}
+
+#endif /* CONFIG_PROVE_LOCKING */
+
 /* called under sysctl_lock */
 static int use_table(struct ctl_table_header *p)
 {
 	if (unlikely(p->unregistering))
 		return 0;
 	p->used++;
+	table_acquire_use(p);
 	return 1;
 }
 
@@ -1469,6 +1520,8 @@ static void unuse_table(struct ctl_table_header *p)
 	if (!--p->used)
 		if (unlikely(p->unregistering))
 			complete(p->unregistering);
+
+	table_release_use(p);
 }
 
 /* called under sysctl_lock, will reacquire if has to wait */
@@ -1478,47 +1531,54 @@ static void start_unregistering(struct ctl_table_header *p)
 	 * if p->used is 0, nobody will ever touch that entry again;
 	 * we'll eliminate all paths to it before dropping sysctl_lock
 	 */
+	table_acquire(p);
 	if (unlikely(p->used)) {
 		struct completion wait;
+		table_contended(p);
+
 		init_completion(&wait);
 		p->unregistering = &wait;
-		spin_unlock(&sysctl_lock);
+		unlock_sysctl();
 		wait_for_completion(&wait);
-		spin_lock(&sysctl_lock);
+		lock_sysctl();
 	} else {
 		/* anything non-NULL; we'll never dereference it */
 		p->unregistering = ERR_PTR(-EINVAL);
 	}
+	table_acquired(p);
+
 	/*
 	 * do not remove from the list until nobody holds it; walking the
 	 * list in do_sysctl() relies on that.
 	 */
 	list_del_init(&p->ctl_entry);
+
+	table_release(p);
 }
 
 void sysctl_head_get(struct ctl_table_header *head)
 {
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	head->count++;
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 }
 
 void sysctl_head_put(struct ctl_table_header *head)
 {
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	if (!--head->count)
 		kfree(head);
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 }
 
 struct ctl_table_header *sysctl_head_grab(struct ctl_table_header *head)
 {
 	if (!head)
 		BUG();
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	if (!use_table(head))
 		head = ERR_PTR(-ENOENT);
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 	return head;
 }
 
@@ -1526,9 +1586,9 @@ void sysctl_head_finish(struct ctl_table_header *head)
 {
 	if (!head)
 		return;
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	unuse_table(head);
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 }
 
 static struct ctl_table_set *
@@ -1555,7 +1615,7 @@ struct ctl_table_header *__sysctl_head_next(struct nsproxy *namespaces,
 	struct ctl_table_header *head;
 	struct list_head *tmp;
 
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	if (prev) {
 		head = prev;
 		tmp = &prev->ctl_entry;
@@ -1568,7 +1628,7 @@ struct ctl_table_header *__sysctl_head_next(struct nsproxy *namespaces,
 
 		if (!use_table(head))
 			goto next;
-		spin_unlock(&sysctl_lock);
+		unlock_sysctl();
 		return head;
 	next:
 		root = head->root;
@@ -1587,7 +1647,7 @@ struct ctl_table_header *__sysctl_head_next(struct nsproxy *namespaces,
 		tmp = header_list->next;
 	}
 out:
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 	return NULL;
 }
 
@@ -1598,9 +1658,9 @@ struct ctl_table_header *sysctl_head_next(struct ctl_table_header *prev)
 
 void register_sysctl_root(struct ctl_table_root *root)
 {
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	list_add_tail(&root->root_list, &sysctl_table_root.root_list);
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 }
 
 #ifdef CONFIG_SYSCTL_SYSCALL
@@ -1951,7 +2011,13 @@ struct ctl_table_header *__register_sysctl_paths(
 		return NULL;
 	}
 #endif
-	spin_lock(&sysctl_lock);
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	{
+		static struct lock_class_key __key;
+		lockdep_init_map(&header->dep_map, "sysctl_used", &__key, 0);
+	}
+#endif	
+	lock_sysctl();
 	header->set = lookup_header_set(root, namespaces);
 	header->attached_by = header->ctl_table;
 	header->attached_to = root_table;
@@ -1966,7 +2032,7 @@ struct ctl_table_header *__register_sysctl_paths(
 	}
 	header->parent->count++;
 	list_add_tail(&header->ctl_entry, &header->set->list);
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 
 	return header;
 }
@@ -2018,7 +2084,7 @@ void unregister_sysctl_table(struct ctl_table_header * header)
 	if (header == NULL)
 		return;
 
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	start_unregistering(header);
 	if (!--header->parent->count) {
 		WARN_ON(1);
@@ -2026,21 +2092,21 @@ void unregister_sysctl_table(struct ctl_table_header * header)
 	}
 	if (!--header->count)
 		kfree(header);
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 }
 
 int sysctl_is_seen(struct ctl_table_header *p)
 {
 	struct ctl_table_set *set = p->set;
 	int res;
-	spin_lock(&sysctl_lock);
+	lock_sysctl();
 	if (p->unregistering)
 		res = 0;
 	else if (!set->is_seen)
 		res = 1;
 	else
 		res = set->is_seen(set);
-	spin_unlock(&sysctl_lock);
+	unlock_sysctl();
 	return res;
 }
 
-- 
1.6.1.2.350.g88cc