io context sharing: preliminary support

Detach task state from ioc, instead keep track of how many processes
are accessing the ioc.

Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

5 files changed, 49 insertions, 23 deletions

diff --git a/block/ll_rw_blk.c b/block/ll_rw_blk.c
index b9bb02e..d4550ec 100644
--- a/block/ll_rw_blk.c
+++ b/block/ll_rw_blk.c
@@ -3854,12 +3854,13 @@ int __init blk_dev_init(void)
 }
 
 /*
- * IO Context helper functions
+ * IO Context helper functions. put_io_context() returns 1 if there are no
+ * more users of this io context, 0 otherwise.
  */
-void put_io_context(struct io_context *ioc)
+int put_io_context(struct io_context *ioc)
 {
 	if (ioc == NULL)
-		return;
+		return 1;
 
 	BUG_ON(atomic_read(&ioc->refcount) == 0);
 
@@ -3878,7 +3879,9 @@ void put_io_context(struct io_context *ioc)
 		rcu_read_unlock();
 
 		kmem_cache_free(iocontext_cachep, ioc);
+		return 1;
 	}
+	return 0;
 }
 EXPORT_SYMBOL(put_io_context);
 
@@ -3893,15 +3896,17 @@ void exit_io_context(void)
 	current->io_context = NULL;
 	task_unlock(current);
 
-	ioc->task = NULL;
-	if (ioc->aic && ioc->aic->exit)
-		ioc->aic->exit(ioc->aic);
-	if (ioc->cic_root.rb_node != NULL) {
-		cic = rb_entry(rb_first(&ioc->cic_root), struct cfq_io_context, rb_node);
-		cic->exit(ioc);
-	}
+	if (atomic_dec_and_test(&ioc->nr_tasks)) {
+		if (ioc->aic && ioc->aic->exit)
+			ioc->aic->exit(ioc->aic);
+		if (ioc->cic_root.rb_node != NULL) {
+			cic = rb_entry(rb_first(&ioc->cic_root),
+				struct cfq_io_context, rb_node);
+			cic->exit(ioc);
+		}
 
-	put_io_context(ioc);
+		put_io_context(ioc);
+	}
 }
 
 struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
@@ -3911,7 +3916,8 @@ struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
 	ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
 	if (ret) {
 		atomic_set(&ret->refcount, 1);
-		ret->task = current;
+		atomic_set(&ret->nr_tasks, 1);
+		spin_lock_init(&ret->lock);
 		ret->ioprio_changed = 0;
 		ret->ioprio = 0;
 		ret->last_waited = jiffies; /* doesn't matter... */
@@ -3959,10 +3965,18 @@ static struct io_context *current_io_context(gfp_t gfp_flags, int node)
  */
 struct io_context *get_io_context(gfp_t gfp_flags, int node)
 {
-	struct io_context *ret;
-	ret = current_io_context(gfp_flags, node);
-	if (likely(ret))
-		atomic_inc(&ret->refcount);
+	struct io_context *ret = NULL;
+
+	/*
+	 * Check for unlikely race with exiting task. ioc ref count is
+	 * zero when ioc is being detached.
+	 */
+	do {
+		ret = current_io_context(gfp_flags, node);
+		if (unlikely(!ret))
+			break;
+	} while (!atomic_inc_not_zero(&ret->refcount));
+
 	return ret;
 }
 EXPORT_SYMBOL(get_io_context);
diff --git a/fs/ioprio.c b/fs/ioprio.c
index a760040..06b5d97 100644
--- a/fs/ioprio.c
+++ b/fs/ioprio.c
@@ -54,7 +54,6 @@ static int set_task_ioprio(struct task_struct *task, int ioprio)
 			break;
 		}
 		task->io_context = ioc;
-		ioc->task = task;
 	} while (1);
 
 	if (!err) {
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 510a18b..2483a05 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -34,7 +34,7 @@ struct sg_io_hdr;
 #define BLKDEV_MIN_RQ	4
 #define BLKDEV_MAX_RQ	128	/* Default maximum */
 
-void put_io_context(struct io_context *ioc);
+int put_io_context(struct io_context *ioc);
 void exit_io_context(void);
 struct io_context *get_io_context(gfp_t gfp_flags, int node);
 struct io_context *alloc_io_context(gfp_t gfp_flags, int node);
diff --git a/include/linux/iocontext.h b/include/linux/iocontext.h
index 186807e..cd44d45 100644
--- a/include/linux/iocontext.h
+++ b/include/linux/iocontext.h
@@ -54,13 +54,15 @@ struct cfq_io_context {
 };
 
 /*
- * This is the per-process I/O subsystem state.  It is refcounted and
- * kmalloc'ed. Currently all fields are modified in process io context
- * (apart from the atomic refcount), so require no locking.
+ * I/O subsystem state of the associated processes.  It is refcounted
+ * and kmalloc'ed. These could be shared between processes.
  */
 struct io_context {
 	atomic_t refcount;
-	struct task_struct *task;
+	atomic_t nr_tasks;
+
+	/* all the fields below are protected by this lock */
+	spinlock_t lock;
 
 	unsigned short ioprio;
 	unsigned short ioprio_changed;
@@ -76,4 +78,16 @@ struct io_context {
 	void *ioc_data;
 };
 
+static inline struct io_context *ioc_task_link(struct io_context *ioc)
+{
+	/*
+	 * if ref count is zero, don't allow sharing (ioc is going away, it's
+	 * a race).
+	 */
+	if (ioc && atomic_inc_not_zero(&ioc->refcount))
+		return ioc;
+
+	return NULL;
+}
+
 #endif
diff --git a/kernel/fork.c b/kernel/fork.c
index 2a86c9d..1987c57 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -805,7 +805,6 @@ static int copy_io(struct task_struct *tsk)
 		if (unlikely(!tsk->io_context))
 			return -ENOMEM;
 
-		tsk->io_context->task = tsk;
 		tsk->io_context->ioprio = ioc->ioprio;
 	}
 #endif