[PATCH 2.6.24-rc8-mm1 12/15] (RFC) IPC/semaphores: make use of RCU to free the sem_undo_list

[pierre.peiffer@bull.net]

[-- Attachment #1: ipc_exit_sem_use_rcu.patch --]
[-- Type: text/plain, Size: 5080 bytes --]

From: Pierre Peiffer <pierre.peiffer@bull.net>

Today, the sem_undo_list is freed when the last task using it exits.
There is no mechanism in place, that allows a safe concurrent access to
the sem_undo_list of a target task and protects efficiently against a
task-exit.

That is okay for now as we don't need this.

As I would like to provide a /proc interface to access this data, I need
such a safe access, without blocking the target task if possible. 

This patch proposes to introduce the use of RCU to delay the real free of
these sem_undo_list structures. They can then be accessed in a safe manner
by any tasks inside read critical section, this way:

	struct sem_undo_list *undo_list;
	int ret;
	...
	rcu_read_lock();
	undo_list = rcu_dereference(task->sysvsem.undo_list);
	if (undo_list)
		ret = atomic_inc_not_zero(&undo_list->refcnt);
	rcu_read_unlock();
	...
	if (undo_list && ret) {
		/* section where undo_list can be used quietly */
		...
	}
	...

Signed-off-by: Pierre Peiffer <pierre.peiffer@bull.net>
---

 include/linux/sem.h |    7 +++++--
 ipc/sem.c           |   42 ++++++++++++++++++++++++++----------------
 2 files changed, 31 insertions(+), 18 deletions(-)

Index: b/include/linux/sem.h
===================================================================
--- a/include/linux/sem.h
+++ b/include/linux/sem.h
@@ -115,7 +115,8 @@ struct sem_queue {
 };
 
 /* Each task has a list of undo requests. They are executed automatically
- * when the process exits.
+ * when the last refcnt of sem_undo_list is released (ie when the process exits
+ * in the general case)
  */
 struct sem_undo {
 	struct sem_undo *	proc_next;	/* next entry on this process */
@@ -125,12 +126,14 @@ struct sem_undo {
 };
 
 /* sem_undo_list controls shared access to the list of sem_undo structures
- * that may be shared among all a CLONE_SYSVSEM task group.
+ * that may be shared among all a CLONE_SYSVSEM task group or with an external
+ * process which changes the list through procfs.
  */ 
 struct sem_undo_list {
 	atomic_t	refcnt;
 	spinlock_t	lock;
 	struct sem_undo	*proc_list;
+	struct ipc_namespace *ns;
 };
 
 struct sysv_sem {
Index: b/ipc/sem.c
===================================================================
--- a/ipc/sem.c
+++ b/ipc/sem.c
@@ -1038,6 +1038,7 @@ static inline int get_undo_list(struct s
 			return -ENOMEM;
 		spin_lock_init(&undo_list->lock);
 		atomic_set(&undo_list->refcnt, 1);
+		undo_list->ns = get_ipc_ns(current->nsproxy->ipc_ns);
 		current->sysvsem.undo_list = undo_list;
 	}
 	*undo_listp = undo_list;
@@ -1316,7 +1317,8 @@ int copy_semundo(unsigned long clone_fla
 }
 
 /*
- * add semadj values to semaphores, free undo structures.
+ * add semadj values to semaphores, free undo structures, if there is no
+ * more user.
  * undo structures are not freed when semaphore arrays are destroyed
  * so some of them may be out of date.
  * IMPLEMENTATION NOTE: There is some confusion over whether the
@@ -1326,23 +1328,17 @@ int copy_semundo(unsigned long clone_fla
  * The original implementation attempted to do this (queue and wait).
  * The current implementation does not do so. The POSIX standard
  * and SVID should be consulted to determine what behavior is mandated.
+ *
+ * Note:
+ * A concurrent task is only allowed to access and go through the list
+ * of sem_undo if it successfully grabs a refcnt.
  */
-void exit_sem(struct task_struct *tsk)
+static void free_semundo_list(struct sem_undo_list *undo_list)
 {
-	struct sem_undo_list *undo_list;
 	struct sem_undo *u, **up;
-	struct ipc_namespace *ns;
 
-	undo_list = tsk->sysvsem.undo_list;
-	if (!undo_list)
-		return;
-
-	if (!atomic_dec_and_test(&undo_list->refcnt))
-		return;
-
-	ns = tsk->nsproxy->ipc_ns;
-	/* There's no need to hold the semundo list lock, as current
-         * is the last task exiting for this undo list.
+	/* There's no need to hold the semundo list lock, as there are
+	 * no more tasks or possible users for this undo list.
 	 */
 	for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
 		struct sem_array *sma;
@@ -1354,7 +1350,7 @@ void exit_sem(struct task_struct *tsk)
 
 		if(semid == -1)
 			continue;
-		sma = sem_lock(ns, semid);
+		sma = sem_lock(undo_list->ns, semid);
 		if (IS_ERR(sma))
 			continue;
 
@@ -1368,7 +1364,8 @@ void exit_sem(struct task_struct *tsk)
 			if (u == un)
 				goto found;
 		}
-		printk ("exit_sem undo list error id=%d\n", u->semid);
+		printk(KERN_ERR "free_semundo_list error id=%d\n",
+		       u->semid);
 		goto next_entry;
 found:
 		*unp = un->id_next;
@@ -1404,9 +1401,22 @@ found:
 next_entry:
 		sem_unlock(sma);
 	}
+	put_ipc_ns(undo_list->ns);
 	kfree(undo_list);
 }
 
+/* called from do_exit() */
+void exit_sem(struct task_struct *tsk)
+{
+	struct sem_undo_list *ul = tsk->sysvsem.undo_list;
+	if (ul) {
+		rcu_assign_pointer(tsk->sysvsem.undo_list, NULL);
+		synchronize_rcu();
+		if (atomic_dec_and_test(&ul->refcnt))
+			free_semundo_list(ul);
+	}
+}
+
 #ifdef CONFIG_PROC_FS
 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
 {

-- 
Pierre Peiffer