[PATCH] ipc: Add refcount to ipc_rcu_alloc

The lifetime of the ipc objects (sem array, msg queue, shm mapping) is
controlled by kern_ipc_perms->lock - a spinlock.  There is no simple way to
reacquire this spinlock after it was dropped to
schedule()/kmalloc/copy_{to,from}_user/whatever.

The attached patch adds a reference count as a preparation to get rid of
sem_revalidate().

Signed-Off-By: Manfred Spraul <manfred@colorfullife.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

5 files changed, 85 insertions, 39 deletions

diff --git a/ipc/msg.c b/ipc/msg.c
index 63777b6e9ea03a..64e4bf32863c40 100644
--- a/ipc/msg.c
+++ b/ipc/msg.c
@@ -100,14 +100,14 @@ static int newque (key_t key, int msgflg)
 	msq->q_perm.security = NULL;
 	retval = security_msg_queue_alloc(msq);
 	if (retval) {
-		ipc_rcu_free(msq, sizeof(*msq));
+		ipc_rcu_putref(msq);
 		return retval;
 	}
 
 	id = ipc_addid(&msg_ids, &msq->q_perm, msg_ctlmni);
 	if(id == -1) {
 		security_msg_queue_free(msq);
-		ipc_rcu_free(msq, sizeof(*msq));
+		ipc_rcu_putref(msq);
 		return -ENOSPC;
 	}
 
@@ -193,7 +193,7 @@ static void freeque (struct msg_queue *msq, int id)
 	}
 	atomic_sub(msq->q_cbytes, &msg_bytes);
 	security_msg_queue_free(msq);
-	ipc_rcu_free(msq, sizeof(struct msg_queue));
+	ipc_rcu_putref(msq);
 }
 
 asmlinkage long sys_msgget (key_t key, int msgflg)
diff --git a/ipc/sem.c b/ipc/sem.c
index 6ed70962c94f29..e7af623f63171f 100644
--- a/ipc/sem.c
+++ b/ipc/sem.c
@@ -179,14 +179,14 @@ static int newary (key_t key, int nsems, int semflg)
 	sma->sem_perm.security = NULL;
 	retval = security_sem_alloc(sma);
 	if (retval) {
-		ipc_rcu_free(sma, size);
+		ipc_rcu_putref(sma);
 		return retval;
 	}
 
 	id = ipc_addid(&sem_ids, &sma->sem_perm, sc_semmni);
 	if(id == -1) {
 		security_sem_free(sma);
-		ipc_rcu_free(sma, size);
+		ipc_rcu_putref(sma);
 		return -ENOSPC;
 	}
 	used_sems += nsems;
@@ -473,7 +473,7 @@ static void freeary (struct sem_array *sma, int id)
 	used_sems -= sma->sem_nsems;
 	size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem);
 	security_sem_free(sma);
-	ipc_rcu_free(sma, size);
+	ipc_rcu_putref(sma);
 }
 
 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
diff --git a/ipc/shm.c b/ipc/shm.c
index d519c9bb2b256f..de76c1961367aa 100644
--- a/ipc/shm.c
+++ b/ipc/shm.c
@@ -117,7 +117,7 @@ static void shm_destroy (struct shmid_kernel *shp)
 		shmem_lock(shp->shm_file, 0);
 	fput (shp->shm_file);
 	security_shm_free(shp);
-	ipc_rcu_free(shp, sizeof(struct shmid_kernel));
+	ipc_rcu_putref(shp);
 }
 
 /*
@@ -194,7 +194,7 @@ static int newseg (key_t key, int shmflg, size_t size)
 	shp->shm_perm.security = NULL;
 	error = security_shm_alloc(shp);
 	if (error) {
-		ipc_rcu_free(shp, sizeof(*shp));
+		ipc_rcu_putref(shp);
 		return error;
 	}
 
@@ -234,7 +234,7 @@ no_id:
 	fput(file);
 no_file:
 	security_shm_free(shp);
-	ipc_rcu_free(shp, sizeof(*shp));
+	ipc_rcu_putref(shp);
 	return error;
 }
 
diff --git a/ipc/util.c b/ipc/util.c
index 727c33233f5f46..5fb0e451c39849 100644
--- a/ipc/util.c
+++ b/ipc/util.c
@@ -135,7 +135,6 @@ static int grow_ary(struct ipc_ids* ids, int newsize)
 		new[i].p = NULL;
 	}
 	old = ids->entries;
-	i = ids->size;
 
 	/*
 	 * before setting the ids->entries to the new array, there must be a
@@ -147,7 +146,7 @@ static int grow_ary(struct ipc_ids* ids, int newsize)
 	smp_wmb();	/* prevent indexing into old array based on new size. */
 	ids->size = newsize;
 
-	ipc_rcu_free(old, sizeof(struct ipc_id)*i);
+	ipc_rcu_putref(old);
 	return ids->size;
 }
 
@@ -277,25 +276,47 @@ void ipc_free(void* ptr, int size)
 		kfree(ptr);
 }
 
-struct ipc_rcu_kmalloc
+/*
+ * rcu allocations:
+ * There are three headers that are prepended to the actual allocation:
+ * - during use: ipc_rcu_hdr.
+ * - during the rcu grace period: ipc_rcu_grace.
+ * - [only if vmalloc]: ipc_rcu_sched.
+ * Their lifetime doesn't overlap, thus the headers share the same memory.
+ * Unlike a normal union, they are right-aligned, thus some container_of
+ * forward/backward casting is necessary:
+ */
+struct ipc_rcu_hdr
+{
+	int refcount;
+	int is_vmalloc;
+	void *data[0];
+};
+
+
+struct ipc_rcu_grace
 {
 	struct rcu_head rcu;
 	/* "void *" makes sure alignment of following data is sane. */
 	void *data[0];
 };
 
-struct ipc_rcu_vmalloc
+struct ipc_rcu_sched
 {
-	struct rcu_head rcu;
 	struct work_struct work;
 	/* "void *" makes sure alignment of following data is sane. */
 	void *data[0];
 };
 
+#define HDRLEN_KMALLOC		(sizeof(struct ipc_rcu_grace) > sizeof(struct ipc_rcu_hdr) ? \
+					sizeof(struct ipc_rcu_grace) : sizeof(struct ipc_rcu_hdr))
+#define HDRLEN_VMALLOC		(sizeof(struct ipc_rcu_sched) > HDRLEN_KMALLOC ? \
+					sizeof(struct ipc_rcu_sched) : HDRLEN_KMALLOC)
+
 static inline int rcu_use_vmalloc(int size)
 {
 	/* Too big for a single page? */
-	if (sizeof(struct ipc_rcu_kmalloc) + size > PAGE_SIZE)
+	if (HDRLEN_KMALLOC + size > PAGE_SIZE)
 		return 1;
 	return 0;
 }
@@ -317,16 +338,29 @@ void* ipc_rcu_alloc(int size)
 	 * workqueue if necessary (for vmalloc). 
 	 */
 	if (rcu_use_vmalloc(size)) {
-		out = vmalloc(sizeof(struct ipc_rcu_vmalloc) + size);
-		if (out) out += sizeof(struct ipc_rcu_vmalloc);
+		out = vmalloc(HDRLEN_VMALLOC + size);
+		if (out) {
+			out += HDRLEN_VMALLOC;
+			container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 1;
+			container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
+		}
 	} else {
-		out = kmalloc(sizeof(struct ipc_rcu_kmalloc)+size, GFP_KERNEL);
-		if (out) out += sizeof(struct ipc_rcu_kmalloc);
+		out = kmalloc(HDRLEN_KMALLOC + size, GFP_KERNEL);
+		if (out) {
+			out += HDRLEN_KMALLOC;
+			container_of(out, struct ipc_rcu_hdr, data)->is_vmalloc = 0;
+			container_of(out, struct ipc_rcu_hdr, data)->refcount = 1;
+		}
 	}
 
 	return out;
 }
 
+void ipc_rcu_getref(void *ptr)
+{
+	container_of(ptr, struct ipc_rcu_hdr, data)->refcount++;
+}
+
 /**
  *	ipc_schedule_free	- free ipc + rcu space
  * 
@@ -335,11 +369,13 @@ void* ipc_rcu_alloc(int size)
  */
 static void ipc_schedule_free(struct rcu_head *head)
 {
-	struct ipc_rcu_vmalloc *free =
-		container_of(head, struct ipc_rcu_vmalloc, rcu);
+	struct ipc_rcu_grace *grace =
+		container_of(head, struct ipc_rcu_grace, rcu);
+	struct ipc_rcu_sched *sched =
+			container_of(&(grace->data[0]), struct ipc_rcu_sched, data[0]);
 
-	INIT_WORK(&free->work, vfree, free);
-	schedule_work(&free->work);
+	INIT_WORK(&sched->work, vfree, sched);
+	schedule_work(&sched->work);
 }
 
 /**
@@ -350,25 +386,23 @@ static void ipc_schedule_free(struct rcu_head *head)
  */
 static void ipc_immediate_free(struct rcu_head *head)
 {
-	struct ipc_rcu_kmalloc *free =
-		container_of(head, struct ipc_rcu_kmalloc, rcu);
+	struct ipc_rcu_grace *free =
+		container_of(head, struct ipc_rcu_grace, rcu);
 	kfree(free);
 }
 
-
-
-void ipc_rcu_free(void* ptr, int size)
+void ipc_rcu_putref(void *ptr)
 {
-	if (rcu_use_vmalloc(size)) {
-		struct ipc_rcu_vmalloc *free;
-		free = ptr - sizeof(*free);
-		call_rcu(&free->rcu, ipc_schedule_free);
+	if (--container_of(ptr, struct ipc_rcu_hdr, data)->refcount > 0)
+		return;
+
+	if (container_of(ptr, struct ipc_rcu_hdr, data)->is_vmalloc) {
+		call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
+				ipc_schedule_free);
 	} else {
-		struct ipc_rcu_kmalloc *free;
-		free = ptr - sizeof(*free);
-		call_rcu(&free->rcu, ipc_immediate_free);
+		call_rcu(&container_of(ptr, struct ipc_rcu_grace, data)->rcu,
+				ipc_immediate_free);
 	}
-
 }
 
 /**
@@ -506,6 +540,12 @@ struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id)
 	return out;
 }
 
+void ipc_lock_by_ptr(struct kern_ipc_perm *perm)
+{
+	rcu_read_lock();
+	spin_lock(&perm->lock);
+}
+
 void ipc_unlock(struct kern_ipc_perm* perm)
 {
 	spin_unlock(&perm->lock);
diff --git a/ipc/util.h b/ipc/util.h
index 8697df67b592d7..3b3adb2d54b431 100644
--- a/ipc/util.h
+++ b/ipc/util.h
@@ -45,14 +45,20 @@ int ipcperms (struct kern_ipc_perm *ipcp, short flg);
  */
 void* ipc_alloc(int size);
 void ipc_free(void* ptr, int size);
-/* for allocation that need to be freed by RCU
- * both function can sleep
+
+/*
+ * For allocation that need to be freed by RCU.
+ * Objects are reference counted, they start with reference count 1.
+ * getref increases the refcount, the putref call that reduces the recount
+ * to 0 schedules the rcu destruction. Caller must guarantee locking.
  */
 void* ipc_rcu_alloc(int size);
-void ipc_rcu_free(void* arg, int size);
+void ipc_rcu_getref(void *ptr);
+void ipc_rcu_putref(void *ptr);
 
 struct kern_ipc_perm* ipc_get(struct ipc_ids* ids, int id);
 struct kern_ipc_perm* ipc_lock(struct ipc_ids* ids, int id);
+void ipc_lock_by_ptr(struct kern_ipc_perm *ipcp);
 void ipc_unlock(struct kern_ipc_perm* perm);
 int ipc_buildid(struct ipc_ids* ids, int id, int seq);
 int ipc_checkid(struct ipc_ids* ids, struct kern_ipc_perm* ipcp, int uid);