Merge branch 'cfq-ioc-share' of git://git.kernel.dk/linux-2.6-block

* 'cfq-ioc-share' of git://git.kernel.dk/linux-2.6-block:
  cfq-iosched: kill some big inlines
  cfq-iosched: relax IOPRIO_CLASS_IDLE restrictions
  kernel: add CLONE_IO to specifically request sharing of IO contexts
  io_context sharing - anticipatory changes
  block: cfq: make the io contect sharing lockless
  io_context sharing - cfq changes
  io context sharing: preliminary support
  ioprio: move io priority from task_struct to io_context

block/as-iosched.c

@@ -170,9 +170,11 @@ static void free_as_io_context(struct as_io_context *aic)
 
 static void as_trim(struct io_context *ioc)
 {
+	spin_lock(&ioc->lock);
 	if (ioc->aic)
 		free_as_io_context(ioc->aic);
 	ioc->aic = NULL;
+	spin_unlock(&ioc->lock);
 }
 
 /* Called when the task exits */
@@ -462,7 +464,9 @@ static void as_antic_timeout(unsigned long data)
 	spin_lock_irqsave(q->queue_lock, flags);
 	if (ad->antic_status == ANTIC_WAIT_REQ
 			|| ad->antic_status == ANTIC_WAIT_NEXT) {
-		struct as_io_context *aic = ad->io_context->aic;
+		struct as_io_context *aic;
+		spin_lock(&ad->io_context->lock);
+		aic = ad->io_context->aic;
 
 		ad->antic_status = ANTIC_FINISHED;
 		kblockd_schedule_work(&ad->antic_work);
@@ -475,6 +479,7 @@ static void as_antic_timeout(unsigned long data)
 			/* process not "saved" by a cooperating request */
 			ad->exit_no_coop = (7*ad->exit_no_coop + 256)/8;
 		}
+		spin_unlock(&ad->io_context->lock);
 	}
 	spin_unlock_irqrestore(q->queue_lock, flags);
 }
@@ -635,9 +640,11 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 
 	ioc = ad->io_context;
 	BUG_ON(!ioc);
+	spin_lock(&ioc->lock);
 
 	if (rq && ioc == RQ_IOC(rq)) {
 		/* request from same process */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
@@ -646,20 +653,25 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 		 * In this situation status should really be FINISHED,
 		 * however the timer hasn't had the chance to run yet.
 		 */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
 	aic = ioc->aic;
-	if (!aic)
+	if (!aic) {
+		spin_unlock(&ioc->lock);
 		return 0;
+	}
 
 	if (atomic_read(&aic->nr_queued) > 0) {
 		/* process has more requests queued */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
 	if (atomic_read(&aic->nr_dispatched) > 0) {
 		/* process has more requests dispatched */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
@@ -680,6 +692,7 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 		}
 
 		as_update_iohist(ad, aic, rq);
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
 
@@ -688,20 +701,27 @@ static int as_can_break_anticipation(struct as_data *ad, struct request *rq)
 		if (aic->ttime_samples == 0)
 			ad->exit_prob = (7*ad->exit_prob + 256)/8;
 
-		if (ad->exit_no_coop > 128)
+		if (ad->exit_no_coop > 128) {
+			spin_unlock(&ioc->lock);
 			return 1;
+		}
 	}
 
 	if (aic->ttime_samples == 0) {
-		if (ad->new_ttime_mean > ad->antic_expire)
+		if (ad->new_ttime_mean > ad->antic_expire) {
+			spin_unlock(&ioc->lock);
 			return 1;
-		if (ad->exit_prob * ad->exit_no_coop > 128*256)
+		}
+		if (ad->exit_prob * ad->exit_no_coop > 128*256) {
+			spin_unlock(&ioc->lock);
 			return 1;
+		}
 	} else if (aic->ttime_mean > ad->antic_expire) {
 		/* the process thinks too much between requests */
+		spin_unlock(&ioc->lock);
 		return 1;
 	}
-
+	spin_unlock(&ioc->lock);
 	return 0;
 }
 
@@ -1255,7 +1275,9 @@ static void as_merged_requests(struct request_queue *q, struct request *req,
 			 * Don't copy here but swap, because when anext is
 			 * removed below, it must contain the unused context
 			 */
+			double_spin_lock(&rioc->lock, &nioc->lock, rioc < nioc);
 			swap_io_context(&rioc, &nioc);
+			double_spin_unlock(&rioc->lock, &nioc->lock, rioc < nioc);
 		}
 	}
 

block/cfq-iosched.c

@@ -26,9 +26,9 @@ static const int cfq_slice_async_rq = 2;
 static int cfq_slice_idle = HZ / 125;
 
 /*
- * grace period before allowing idle class to get disk access
+ * offset from end of service tree
  */
-#define CFQ_IDLE_GRACE		(HZ / 10)
+#define CFQ_IDLE_DELAY		(HZ / 5)
 
 /*
  * below this threshold, we consider thinktime immediate
@@ -98,8 +98,6 @@ struct cfq_data {
 	struct cfq_queue *async_cfqq[2][IOPRIO_BE_NR];
 	struct cfq_queue *async_idle_cfqq;
 
-	struct timer_list idle_class_timer;
-
 	sector_t last_position;
 	unsigned long last_end_request;
 
@@ -199,8 +197,8 @@ CFQ_CFQQ_FNS(sync);
 
 static void cfq_dispatch_insert(struct request_queue *, struct request *);
 static struct cfq_queue *cfq_get_queue(struct cfq_data *, int,
-				       struct task_struct *, gfp_t);
-static struct cfq_io_context *cfq_cic_rb_lookup(struct cfq_data *,
+				       struct io_context *, gfp_t);
+static struct cfq_io_context *cfq_cic_lookup(struct cfq_data *,
 						struct io_context *);
 
 static inline struct cfq_queue *cic_to_cfqq(struct cfq_io_context *cic,
@@ -384,12 +382,15 @@ cfq_choose_req(struct cfq_data *cfqd, struct request *rq1, struct request *rq2)
 /*
  * The below is leftmost cache rbtree addon
  */
-static struct rb_node *cfq_rb_first(struct cfq_rb_root *root)
+static struct cfq_queue *cfq_rb_first(struct cfq_rb_root *root)
 {
 	if (!root->left)
 		root->left = rb_first(&root->rb);
 
-	return root->left;
+	if (root->left)
+		return rb_entry(root->left, struct cfq_queue, rb_node);
+
+	return NULL;
 }
 
 static void cfq_rb_erase(struct rb_node *n, struct cfq_rb_root *root)
@@ -446,12 +447,20 @@ static unsigned long cfq_slice_offset(struct cfq_data *cfqd,
 static void cfq_service_tree_add(struct cfq_data *cfqd,
 				    struct cfq_queue *cfqq, int add_front)
 {
-	struct rb_node **p = &cfqd->service_tree.rb.rb_node;
-	struct rb_node *parent = NULL;
+	struct rb_node **p, *parent;
+	struct cfq_queue *__cfqq;
 	unsigned long rb_key;
 	int left;
 
-	if (!add_front) {
+	if (cfq_class_idle(cfqq)) {
+		rb_key = CFQ_IDLE_DELAY;
+		parent = rb_last(&cfqd->service_tree.rb);
+		if (parent && parent != &cfqq->rb_node) {
+			__cfqq = rb_entry(parent, struct cfq_queue, rb_node);
+			rb_key += __cfqq->rb_key;
+		} else
+			rb_key += jiffies;
+	} else if (!add_front) {
 		rb_key = cfq_slice_offset(cfqd, cfqq) + jiffies;
 		rb_key += cfqq->slice_resid;
 		cfqq->slice_resid = 0;
@@ -469,8 +478,9 @@ static void cfq_service_tree_add(struct cfq_data *cfqd,
 	}
 
 	left = 1;
+	parent = NULL;
+	p = &cfqd->service_tree.rb.rb_node;
 	while (*p) {
-		struct cfq_queue *__cfqq;
 		struct rb_node **n;
 
 		parent = *p;
@@ -524,8 +534,7 @@ static void cfq_resort_rr_list(struct cfq_data *cfqd, struct cfq_queue *cfqq)
  * add to busy list of queues for service, trying to be fair in ordering
  * the pending list according to last request service
  */
-static inline void
-cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
+static void cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
 	BUG_ON(cfq_cfqq_on_rr(cfqq));
 	cfq_mark_cfqq_on_rr(cfqq);
@@ -538,8 +547,7 @@ cfq_add_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
  * Called when the cfqq no longer has requests pending, remove it from
  * the service tree.
  */
-static inline void
-cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
+static void cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 {
 	BUG_ON(!cfq_cfqq_on_rr(cfqq));
 	cfq_clear_cfqq_on_rr(cfqq);
@@ -554,7 +562,7 @@ cfq_del_cfqq_rr(struct cfq_data *cfqd, struct cfq_queue *cfqq)
 /*
  * rb tree support functions
  */
-static inline void cfq_del_rq_rb(struct request *rq)
+static void cfq_del_rq_rb(struct request *rq)
 {
 	struct cfq_queue *cfqq = RQ_CFQQ(rq);
 	struct cfq_data *cfqd = cfqq->cfqd;
@@ -594,8 +602,7 @@ static void cfq_add_rq_rb(struct request *rq)
 	BUG_ON(!cfqq->next_rq);
 }
 
-static inline void
-cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
+static void cfq_reposition_rq_rb(struct cfq_queue *cfqq, struct request *rq)
 {
 	elv_rb_del(&cfqq->sort_list, rq);
 	cfqq->queued[rq_is_sync(rq)]--;
@@ -609,7 +616,7 @@ cfq_find_rq_fmerge(struct cfq_data *cfqd, struct bio *bio)
 	struct cfq_io_context *cic;
 	struct cfq_queue *cfqq;
 
-	cic = cfq_cic_rb_lookup(cfqd, tsk->io_context);
+	cic = cfq_cic_lookup(cfqd, tsk->io_context);
 	if (!cic)
 		return NULL;
 
@@ -721,7 +728,7 @@ static int cfq_allow_merge(struct request_queue *q, struct request *rq,
 	 * Lookup the cfqq that this bio will be queued with. Allow
 	 * merge only if rq is queued there.
 	 */
-	cic = cfq_cic_rb_lookup(cfqd, current->io_context);
+	cic = cfq_cic_lookup(cfqd, current->io_context);
 	if (!cic)
 		return 0;
 
@@ -732,15 +739,10 @@ static int cfq_allow_merge(struct request_queue *q, struct request *rq,
 	return 0;
 }
 
-static inline void
-__cfq_set_active_queue(struct cfq_data *cfqd, struct cfq_queue *cfqq)
+static void __cfq_set_active_queue(struct cfq_data *cfqd,
+				   struct cfq_queue *cfqq)
 {
 	if (cfqq) {
-		/*
-		 * stop potential idle class queues waiting service
-		 */
-		del_timer(&cfqd->idle_class_timer);
-
 		cfqq->slice_end = 0;
 		cfq_clear_cfqq_must_alloc_slice(cfqq);
 		cfq_clear_cfqq_fifo_expire(cfqq);
@@ -789,47 +791,16 @@ static inline void cfq_slice_expired(struct cfq_data *cfqd, int timed_out)
 		__cfq_slice_expired(cfqd, cfqq, timed_out);
 }
 
-static int start_idle_class_timer(struct cfq_data *cfqd)
-{
-	unsigned long end = cfqd->last_end_request + CFQ_IDLE_GRACE;
-	unsigned long now = jiffies;
-
-	if (time_before(now, end) &&
-	    time_after_eq(now, cfqd->last_end_request)) {
-		mod_timer(&cfqd->idle_class_timer, end);
-		return 1;
-	}
-
-	return 0;
-}
-
 /*
  * Get next queue for service. Unless we have a queue preemption,
  * we'll simply select the first cfqq in the service tree.
  */
 static struct cfq_queue *cfq_get_next_queue(struct cfq_data *cfqd)
 {
-	struct cfq_queue *cfqq;
-	struct rb_node *n;
-
 	if (RB_EMPTY_ROOT(&cfqd->service_tree.rb))
 		return NULL;
 
-	n = cfq_rb_first(&cfqd->service_tree);
-	cfqq = rb_entry(n, struct cfq_queue, rb_node);
-
-	if (cfq_class_idle(cfqq)) {
-		/*
-		 * if we have idle queues and no rt or be queues had
-		 * pending requests, either allow immediate service if
-		 * the grace period has passed or arm the idle grace
-		 * timer
-		 */
-		if (start_idle_class_timer(cfqd))
-			cfqq = NULL;
-	}
-
-	return cfqq;
+	return cfq_rb_first(&cfqd->service_tree);
 }
 
 /*
@@ -895,7 +866,7 @@ static void cfq_arm_slice_timer(struct cfq_data *cfqd)
 	 * task has exited, don't wait
 	 */
 	cic = cfqd->active_cic;
-	if (!cic || !cic->ioc->task)
+	if (!cic || !atomic_read(&cic->ioc->nr_tasks))
 		return;
 
 	/*
@@ -939,7 +910,7 @@ static void cfq_dispatch_insert(struct request_queue *q, struct request *rq)
 /*
  * return expired entry, or NULL to just start from scratch in rbtree
  */
-static inline struct request *cfq_check_fifo(struct cfq_queue *cfqq)
+static struct request *cfq_check_fifo(struct cfq_queue *cfqq)
 {
 	struct cfq_data *cfqd = cfqq->cfqd;
 	struct request *rq;
@@ -1068,7 +1039,7 @@ __cfq_dispatch_requests(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 	return dispatched;
 }
 
-static inline int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
+static int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
 {
 	int dispatched = 0;
 
@@ -1087,14 +1058,11 @@ static inline int __cfq_forced_dispatch_cfqq(struct cfq_queue *cfqq)
  */
 static int cfq_forced_dispatch(struct cfq_data *cfqd)
 {
+	struct cfq_queue *cfqq;
 	int dispatched = 0;
-	struct rb_node *n;
-
-	while ((n = cfq_rb_first(&cfqd->service_tree)) != NULL) {
-		struct cfq_queue *cfqq = rb_entry(n, struct cfq_queue, rb_node);
 
+	while ((cfqq = cfq_rb_first(&cfqd->service_tree)) != NULL)
 		dispatched += __cfq_forced_dispatch_cfqq(cfqq);
-	}
 
 	cfq_slice_expired(cfqd, 0);
 
@@ -1170,20 +1138,69 @@ static void cfq_put_queue(struct cfq_queue *cfqq)
 	kmem_cache_free(cfq_pool, cfqq);
 }
 
-static void cfq_free_io_context(struct io_context *ioc)
+/*
+ * Call func for each cic attached to this ioc. Returns number of cic's seen.
+ */
+#define CIC_GANG_NR	16
+static unsigned int
+call_for_each_cic(struct io_context *ioc,
+		  void (*func)(struct io_context *, struct cfq_io_context *))
 {
-	struct cfq_io_context *__cic;
-	struct rb_node *n;
-	int freed = 0;
+	struct cfq_io_context *cics[CIC_GANG_NR];
+	unsigned long index = 0;
+	unsigned int called = 0;
+	int nr;
 
-	ioc->ioc_data = NULL;
+	rcu_read_lock();
 
-	while ((n = rb_first(&ioc->cic_root)) != NULL) {
-		__cic = rb_entry(n, struct cfq_io_context, rb_node);
-		rb_erase(&__cic->rb_node, &ioc->cic_root);
-		kmem_cache_free(cfq_ioc_pool, __cic);
-		freed++;
-	}
+	do {
+		int i;
+
+		/*
+		 * Perhaps there's a better way - this just gang lookups from
+		 * 0 to the end, restarting after each CIC_GANG_NR from the
+		 * last key + 1.
+		 */
+		nr = radix_tree_gang_lookup(&ioc->radix_root, (void **) cics,
+						index, CIC_GANG_NR);
+		if (!nr)
+			break;
+
+		called += nr;
+		index = 1 + (unsigned long) cics[nr - 1]->key;
+
+		for (i = 0; i < nr; i++)
+			func(ioc, cics[i]);
+	} while (nr == CIC_GANG_NR);
+
+	rcu_read_unlock();
+
+	return called;
+}
+
+static void cic_free_func(struct io_context *ioc, struct cfq_io_context *cic)
+{
+	unsigned long flags;
+
+	BUG_ON(!cic->dead_key);
+
+	spin_lock_irqsave(&ioc->lock, flags);
+	radix_tree_delete(&ioc->radix_root, cic->dead_key);
+	spin_unlock_irqrestore(&ioc->lock, flags);
+
+	kmem_cache_free(cfq_ioc_pool, cic);
+}
+
+static void cfq_free_io_context(struct io_context *ioc)
+{
+	int freed;
+
+	/*
+	 * ioc->refcount is zero here, so no more cic's are allowed to be
+	 * linked into this ioc. So it should be ok to iterate over the known
+	 * list, we will see all cic's since no new ones are added.
+	 */
+	freed = call_for_each_cic(ioc, cic_free_func);
 
 	elv_ioc_count_mod(ioc_count, -freed);
 
@@ -1205,7 +1222,12 @@ static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
 					 struct cfq_io_context *cic)
 {
 	list_del_init(&cic->queue_list);
+
+	/*
+	 * Make sure key == NULL is seen for dead queues
+	 */
 	smp_wmb();
+	cic->dead_key = (unsigned long) cic->key;
 	cic->key = NULL;
 
 	if (cic->cfqq[ASYNC]) {
@@ -1219,16 +1241,18 @@ static void __cfq_exit_single_io_context(struct cfq_data *cfqd,
 	}
 }
 
-static void cfq_exit_single_io_context(struct cfq_io_context *cic)
+static void cfq_exit_single_io_context(struct io_context *ioc,
+				       struct cfq_io_context *cic)
 {
 	struct cfq_data *cfqd = cic->key;
 
 	if (cfqd) {
 		struct request_queue *q = cfqd->queue;
+		unsigned long flags;
 
-		spin_lock_irq(q->queue_lock);
+		spin_lock_irqsave(q->queue_lock, flags);
 		__cfq_exit_single_io_context(cfqd, cic);
-		spin_unlock_irq(q->queue_lock);
+		spin_unlock_irqrestore(q->queue_lock, flags);
 	}
 }
 
@@ -1238,21 +1262,8 @@ static void cfq_exit_single_io_context(struct cfq_io_context *cic)
  */
 static void cfq_exit_io_context(struct io_context *ioc)
 {
-	struct cfq_io_context *__cic;
-	struct rb_node *n;
-
-	ioc->ioc_data = NULL;
-
-	/*
-	 * put the reference this task is holding to the various queues
-	 */
-	n = rb_first(&ioc->cic_root);
-	while (n != NULL) {
-		__cic = rb_entry(n, struct cfq_io_context, rb_node);
-
-		cfq_exit_single_io_context(__cic);
-		n = rb_next(n);
-	}
+	rcu_assign_pointer(ioc->ioc_data, NULL);
+	call_for_each_cic(ioc, cfq_exit_single_io_context);
 }
 
 static struct cfq_io_context *
@@ -1273,7 +1284,7 @@ cfq_alloc_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
 	return cic;
 }
 
-static void cfq_init_prio_data(struct cfq_queue *cfqq)
+static void cfq_init_prio_data(struct cfq_queue *cfqq, struct io_context *ioc)
 {
 	struct task_struct *tsk = current;
 	int ioprio_class;
@@ -1281,7 +1292,7 @@ static void cfq_init_prio_data(struct cfq_queue *cfqq)
 	if (!cfq_cfqq_prio_changed(cfqq))
 		return;
 
-	ioprio_class = IOPRIO_PRIO_CLASS(tsk->ioprio);
+	ioprio_class = IOPRIO_PRIO_CLASS(ioc->ioprio);
 	switch (ioprio_class) {
 		default:
 			printk(KERN_ERR "cfq: bad prio %x\n", ioprio_class);
@@ -1293,11 +1304,11 @@ static void cfq_init_prio_data(struct cfq_queue *cfqq)
 			cfqq->ioprio_class = IOPRIO_CLASS_BE;
 			break;
 		case IOPRIO_CLASS_RT:
-			cfqq->ioprio = task_ioprio(tsk);
+			cfqq->ioprio = task_ioprio(ioc);
 			cfqq->ioprio_class = IOPRIO_CLASS_RT;
 			break;
 		case IOPRIO_CLASS_BE:
-			cfqq->ioprio = task_ioprio(tsk);
+			cfqq->ioprio = task_ioprio(ioc);
 			cfqq->ioprio_class = IOPRIO_CLASS_BE;
 			break;
 		case IOPRIO_CLASS_IDLE:
@@ -1316,7 +1327,7 @@ static void cfq_init_prio_data(struct cfq_queue *cfqq)
 	cfq_clear_cfqq_prio_changed(cfqq);
 }
 
-static inline void changed_ioprio(struct cfq_io_context *cic)
+static void changed_ioprio(struct io_context *ioc, struct cfq_io_context *cic)
 {
 	struct cfq_data *cfqd = cic->key;
 	struct cfq_queue *cfqq;
@@ -1330,8 +1341,7 @@ static inline void changed_ioprio(struct cfq_io_context *cic)
 	cfqq = cic->cfqq[ASYNC];
 	if (cfqq) {
 		struct cfq_queue *new_cfqq;
-		new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc->task,
-					 GFP_ATOMIC);
+		new_cfqq = cfq_get_queue(cfqd, ASYNC, cic->ioc, GFP_ATOMIC);
 		if (new_cfqq) {
 			cic->cfqq[ASYNC] = new_cfqq;
 			cfq_put_queue(cfqq);
@@ -1347,29 +1357,19 @@ static inline void changed_ioprio(struct cfq_io_context *cic)
 
 static void cfq_ioc_set_ioprio(struct io_context *ioc)
 {
-	struct cfq_io_context *cic;
-	struct rb_node *n;
-
+	call_for_each_cic(ioc, changed_ioprio);
 	ioc->ioprio_changed = 0;
-
-	n = rb_first(&ioc->cic_root);
-	while (n != NULL) {
-		cic = rb_entry(n, struct cfq_io_context, rb_node);
-
-		changed_ioprio(cic);
-		n = rb_next(n);
-	}
 }
 
 static struct cfq_queue *
 cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync,
-		     struct task_struct *tsk, gfp_t gfp_mask)
+		     struct io_context *ioc, gfp_t gfp_mask)
 {
 	struct cfq_queue *cfqq, *new_cfqq = NULL;
 	struct cfq_io_context *cic;
 
 retry:
-	cic = cfq_cic_rb_lookup(cfqd, tsk->io_context);
+	cic = cfq_cic_lookup(cfqd, ioc);
 	/* cic always exists here */
 	cfqq = cic_to_cfqq(cic, is_sync);
 
@@ -1404,15 +1404,16 @@ cfq_find_alloc_queue(struct cfq_data *cfqd, int is_sync,
 		atomic_set(&cfqq->ref, 0);
 		cfqq->cfqd = cfqd;
 
-		if (is_sync) {
-			cfq_mark_cfqq_idle_window(cfqq);
-			cfq_mark_cfqq_sync(cfqq);
-		}
-
 		cfq_mark_cfqq_prio_changed(cfqq);
 		cfq_mark_cfqq_queue_new(cfqq);
 
-		cfq_init_prio_data(cfqq);
+		cfq_init_prio_data(cfqq, ioc);
+
+		if (is_sync) {
+			if (!cfq_class_idle(cfqq))
+				cfq_mark_cfqq_idle_window(cfqq);
+			cfq_mark_cfqq_sync(cfqq);
+		}
 	}
 
 	if (new_cfqq)
@@ -1439,11 +1440,11 @@ cfq_async_queue_prio(struct cfq_data *cfqd, int ioprio_class, int ioprio)
 }
 
 static struct cfq_queue *
-cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct task_struct *tsk,
+cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct io_context *ioc,
 	      gfp_t gfp_mask)
 {
-	const int ioprio = task_ioprio(tsk);
-	const int ioprio_class = task_ioprio_class(tsk);
+	const int ioprio = task_ioprio(ioc);
+	const int ioprio_class = task_ioprio_class(ioc);
 	struct cfq_queue **async_cfqq = NULL;
 	struct cfq_queue *cfqq = NULL;
 
@@ -1453,7 +1454,7 @@ cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct task_struct *tsk,
 	}
 
 	if (!cfqq) {
-		cfqq = cfq_find_alloc_queue(cfqd, is_sync, tsk, gfp_mask);
+		cfqq = cfq_find_alloc_queue(cfqd, is_sync, ioc, gfp_mask);
 		if (!cfqq)
 			return NULL;
 	}
@@ -1470,28 +1471,42 @@ cfq_get_queue(struct cfq_data *cfqd, int is_sync, struct task_struct *tsk,
 	return cfqq;
 }
 
+static void cfq_cic_free(struct cfq_io_context *cic)
+{
+	kmem_cache_free(cfq_ioc_pool, cic);
+	elv_ioc_count_dec(ioc_count);
+
+	if (ioc_gone && !elv_ioc_count_read(ioc_count))
+		complete(ioc_gone);
+}
+
 /*
  * We drop cfq io contexts lazily, so we may find a dead one.
  */
 static void
-cfq_drop_dead_cic(struct io_context *ioc, struct cfq_io_context *cic)
+cfq_drop_dead_cic(struct cfq_data *cfqd, struct io_context *ioc,
+		  struct cfq_io_context *cic)
 {
+	unsigned long flags;
+
 	WARN_ON(!list_empty(&cic->queue_list));
 
+	spin_lock_irqsave(&ioc->lock, flags);
+
 	if (ioc->ioc_data == cic)
-		ioc->ioc_data = NULL;
+		rcu_assign_pointer(ioc->ioc_data, NULL);
 
-	rb_erase(&cic->rb_node, &ioc->cic_root);
-	kmem_cache_free(cfq_ioc_pool, cic);
-	elv_ioc_count_dec(ioc_count);
+	radix_tree_delete(&ioc->radix_root, (unsigned long) cfqd);
+	spin_unlock_irqrestore(&ioc->lock, flags);
+
+	cfq_cic_free(cic);
 }
 
 static struct cfq_io_context *
-cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc)
+cfq_cic_lookup(struct cfq_data *cfqd, struct io_context *ioc)
 {
-	struct rb_node *n;
 	struct cfq_io_context *cic;
-	void *k, *key = cfqd;
+	void *k;
 
 	if (unlikely(!ioc))
 		return NULL;
@@ -1499,74 +1514,64 @@ cfq_cic_rb_lookup(struct cfq_data *cfqd, struct io_context *ioc)
 	/*
 	 * we maintain a last-hit cache, to avoid browsing over the tree
 	 */
-	cic = ioc->ioc_data;
+	cic = rcu_dereference(ioc->ioc_data);
 	if (cic && cic->key == cfqd)
 		return cic;
 
-restart:
-	n = ioc->cic_root.rb_node;
-	while (n) {
-		cic = rb_entry(n, struct cfq_io_context, rb_node);
+	do {
+		rcu_read_lock();
+		cic = radix_tree_lookup(&ioc->radix_root, (unsigned long) cfqd);
+		rcu_read_unlock();
+		if (!cic)
+			break;
 		/* ->key must be copied to avoid race with cfq_exit_queue() */
 		k = cic->key;
 		if (unlikely(!k)) {
-			cfq_drop_dead_cic(ioc, cic);
-			goto restart;
+			cfq_drop_dead_cic(cfqd, ioc, cic);
+			continue;
 		}
 
-		if (key < k)
-			n = n->rb_left;
-		else if (key > k)
-			n = n->rb_right;
-		else {
-			ioc->ioc_data = cic;
-			return cic;
-		}
-	}
+		rcu_assign_pointer(ioc->ioc_data, cic);
+		break;
+	} while (1);
 
-	return NULL;
+	return cic;
 }
 
-static inline void
-cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
-	     struct cfq_io_context *cic)
+/*
+ * Add cic into ioc, using cfqd as the search key. This enables us to lookup
+ * the process specific cfq io context when entered from the block layer.
+ * Also adds the cic to a per-cfqd list, used when this queue is removed.
+ */
+static int cfq_cic_link(struct cfq_data *cfqd, struct io_context *ioc,
+			struct cfq_io_context *cic, gfp_t gfp_mask)
 {
-	struct rb_node **p;
-	struct rb_node *parent;
-	struct cfq_io_context *__cic;
 	unsigned long flags;
-	void *k;
+	int ret;
 
-	cic->ioc = ioc;
-	cic->key = cfqd;
+	ret = radix_tree_preload(gfp_mask);
+	if (!ret) {
+		cic->ioc = ioc;
+		cic->key = cfqd;
 
-restart:
-	parent = NULL;
-	p = &ioc->cic_root.rb_node;
-	while (*p) {
-		parent = *p;
-		__cic = rb_entry(parent, struct cfq_io_context, rb_node);
-		/* ->key must be copied to avoid race with cfq_exit_queue() */
-		k = __cic->key;
-		if (unlikely(!k)) {
-			cfq_drop_dead_cic(ioc, __cic);
-			goto restart;
-		}
+		spin_lock_irqsave(&ioc->lock, flags);
+		ret = radix_tree_insert(&ioc->radix_root,
+						(unsigned long) cfqd, cic);
+		spin_unlock_irqrestore(&ioc->lock, flags);
 
-		if (cic->key < k)
-			p = &(*p)->rb_left;
-		else if (cic->key > k)
-			p = &(*p)->rb_right;
-		else
-			BUG();
+		radix_tree_preload_end();
+
+		if (!ret) {
+			spin_lock_irqsave(cfqd->queue->queue_lock, flags);
+			list_add(&cic->queue_list, &cfqd->cic_list);
+			spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
+		}
 	}
 
-	rb_link_node(&cic->rb_node, parent, p);
-	rb_insert_color(&cic->rb_node, &ioc->cic_root);
+	if (ret)
+		printk(KERN_ERR "cfq: cic link failed!\n");
 
-	spin_lock_irqsave(cfqd->queue->queue_lock, flags);
-	list_add(&cic->queue_list, &cfqd->cic_list);
-	spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
+	return ret;
 }
 
 /*
@@ -1586,7 +1591,7 @@ cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
 	if (!ioc)
 		return NULL;
 
-	cic = cfq_cic_rb_lookup(cfqd, ioc);
+	cic = cfq_cic_lookup(cfqd, ioc);
 	if (cic)
 		goto out;
 
@@ -1594,13 +1599,17 @@ cfq_get_io_context(struct cfq_data *cfqd, gfp_t gfp_mask)
 	if (cic == NULL)
 		goto err;
 
-	cfq_cic_link(cfqd, ioc, cic);
+	if (cfq_cic_link(cfqd, ioc, cic, gfp_mask))
+		goto err_free;
+
 out:
 	smp_read_barrier_depends();
 	if (unlikely(ioc->ioprio_changed))
 		cfq_ioc_set_ioprio(ioc);
 
 	return cic;
+err_free:
+	cfq_cic_free(cic);
 err:
 	put_io_context(ioc);
 	return NULL;
@@ -1655,12 +1664,15 @@ cfq_update_idle_window(struct cfq_data *cfqd, struct cfq_queue *cfqq,
 {
 	int enable_idle;
 
-	if (!cfq_cfqq_sync(cfqq))
+	/*
+	 * Don't idle for async or idle io prio class
+	 */
+	if (!cfq_cfqq_sync(cfqq) || cfq_class_idle(cfqq))
 		return;
 
 	enable_idle = cfq_cfqq_idle_window(cfqq);
 
-	if (!cic->ioc->task || !cfqd->cfq_slice_idle ||
+	if (!atomic_read(&cic->ioc->nr_tasks) || !cfqd->cfq_slice_idle ||
 	    (cfqd->hw_tag && CIC_SEEKY(cic)))
 		enable_idle = 0;
 	else if (sample_valid(cic->ttime_samples)) {
@@ -1793,7 +1805,7 @@ static void cfq_insert_request(struct request_queue *q, struct request *rq)
 	struct cfq_data *cfqd = q->elevator->elevator_data;
 	struct cfq_queue *cfqq = RQ_CFQQ(rq);
 
-	cfq_init_prio_data(cfqq);
+	cfq_init_prio_data(cfqq, RQ_CIC(rq)->ioc);
 
 	cfq_add_rq_rb(rq);
 
@@ -1834,7 +1846,7 @@ static void cfq_completed_request(struct request_queue *q, struct request *rq)
 			cfq_set_prio_slice(cfqd, cfqq);
 			cfq_clear_cfqq_slice_new(cfqq);
 		}
-		if (cfq_slice_used(cfqq))
+		if (cfq_slice_used(cfqq) || cfq_class_idle(cfqq))
 			cfq_slice_expired(cfqd, 1);
 		else if (sync && RB_EMPTY_ROOT(&cfqq->sort_list))
 			cfq_arm_slice_timer(cfqd);
@@ -1894,13 +1906,13 @@ static int cfq_may_queue(struct request_queue *q, int rw)
 	 * so just lookup a possibly existing queue, or return 'may queue'
 	 * if that fails
 	 */
-	cic = cfq_cic_rb_lookup(cfqd, tsk->io_context);
+	cic = cfq_cic_lookup(cfqd, tsk->io_context);
 	if (!cic)
 		return ELV_MQUEUE_MAY;
 
 	cfqq = cic_to_cfqq(cic, rw & REQ_RW_SYNC);
 	if (cfqq) {
-		cfq_init_prio_data(cfqq);
+		cfq_init_prio_data(cfqq, cic->ioc);
 		cfq_prio_boost(cfqq);
 
 		return __cfq_may_queue(cfqq);
@@ -1938,7 +1950,6 @@ static int
 cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
 {
 	struct cfq_data *cfqd = q->elevator->elevator_data;
-	struct task_struct *tsk = current;
 	struct cfq_io_context *cic;
 	const int rw = rq_data_dir(rq);
 	const int is_sync = rq_is_sync(rq);
@@ -1956,7 +1967,7 @@ cfq_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
 
 	cfqq = cic_to_cfqq(cic, is_sync);
 	if (!cfqq) {
-		cfqq = cfq_get_queue(cfqd, is_sync, tsk, gfp_mask);
+		cfqq = cfq_get_queue(cfqd, is_sync, cic->ioc, gfp_mask);
 
 		if (!cfqq)
 			goto queue_fail;
@@ -2039,29 +2050,9 @@ static void cfq_idle_slice_timer(unsigned long data)
 	spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
 }
 
-/*
- * Timer running if an idle class queue is waiting for service
- */
-static void cfq_idle_class_timer(unsigned long data)
-{
-	struct cfq_data *cfqd = (struct cfq_data *) data;
-	unsigned long flags;
-
-	spin_lock_irqsave(cfqd->queue->queue_lock, flags);
-
-	/*
-	 * race with a non-idle queue, reset timer
-	 */
-	if (!start_idle_class_timer(cfqd))
-		cfq_schedule_dispatch(cfqd);
-
-	spin_unlock_irqrestore(cfqd->queue->queue_lock, flags);
-}
-
 static void cfq_shutdown_timer_wq(struct cfq_data *cfqd)
 {
 	del_timer_sync(&cfqd->idle_slice_timer);
-	del_timer_sync(&cfqd->idle_class_timer);
 	kblockd_flush_work(&cfqd->unplug_work);
 }
 
@@ -2126,10 +2117,6 @@ static void *cfq_init_queue(struct request_queue *q)
 	cfqd->idle_slice_timer.function = cfq_idle_slice_timer;
 	cfqd->idle_slice_timer.data = (unsigned long) cfqd;
 
-	init_timer(&cfqd->idle_class_timer);
-	cfqd->idle_class_timer.function = cfq_idle_class_timer;
-	cfqd->idle_class_timer.data = (unsigned long) cfqd;
-
 	INIT_WORK(&cfqd->unplug_work, cfq_kick_queue);
 
 	cfqd->last_end_request = jiffies;
@@ -2160,7 +2147,7 @@ static int __init cfq_slab_setup(void)
 	if (!cfq_pool)
 		goto fail;
 
-	cfq_ioc_pool = KMEM_CACHE(cfq_io_context, 0);
+	cfq_ioc_pool = KMEM_CACHE(cfq_io_context, SLAB_DESTROY_BY_RCU);
 	if (!cfq_ioc_pool)
 		goto fail;
 

block/ll_rw_blk.c

@@ -3853,55 +3853,100 @@ int __init blk_dev_init(void)
 	return 0;
 }
 
+static void cfq_dtor(struct io_context *ioc)
+{
+	struct cfq_io_context *cic[1];
+	int r;
+
+	/*
+	 * We don't have a specific key to lookup with, so use the gang
+	 * lookup to just retrieve the first item stored. The cfq exit
+	 * function will iterate the full tree, so any member will do.
+	 */
+	r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
+	if (r > 0)
+		cic[0]->dtor(ioc);
+}
+
 /*
- * 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);
 
 	if (atomic_dec_and_test(&ioc->refcount)) {
-		struct cfq_io_context *cic;
-
 		rcu_read_lock();
 		if (ioc->aic && ioc->aic->dtor)
 			ioc->aic->dtor(ioc->aic);
-		if (ioc->cic_root.rb_node != NULL) {
-			struct rb_node *n = rb_first(&ioc->cic_root);
-
-			cic = rb_entry(n, struct cfq_io_context, rb_node);
-			cic->dtor(ioc);
-		}
 		rcu_read_unlock();
+		cfq_dtor(ioc);
 
 		kmem_cache_free(iocontext_cachep, ioc);
+		return 1;
 	}
+	return 0;
 }
 EXPORT_SYMBOL(put_io_context);
 
+static void cfq_exit(struct io_context *ioc)
+{
+	struct cfq_io_context *cic[1];
+	int r;
+
+	rcu_read_lock();
+	/*
+	 * See comment for cfq_dtor()
+	 */
+	r = radix_tree_gang_lookup(&ioc->radix_root, (void **) cic, 0, 1);
+	rcu_read_unlock();
+
+	if (r > 0)
+		cic[0]->exit(ioc);
+}
+
 /* Called by the exitting task */
 void exit_io_context(void)
 {
 	struct io_context *ioc;
-	struct cfq_io_context *cic;
 
 	task_lock(current);
 	ioc = current->io_context;
 	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);
+		cfq_exit(ioc);
+
+		put_io_context(ioc);
 	}
+}
+
+struct io_context *alloc_io_context(gfp_t gfp_flags, int node)
+{
+	struct io_context *ret;
 
-	put_io_context(ioc);
+	ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
+	if (ret) {
+		atomic_set(&ret->refcount, 1);
+		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... */
+		ret->nr_batch_requests = 0; /* because this is 0 */
+		ret->aic = NULL;
+		INIT_RADIX_TREE(&ret->radix_root, GFP_ATOMIC | __GFP_HIGH);
+		ret->ioc_data = NULL;
+	}
+
+	return ret;
 }
 
 /*
@@ -3921,16 +3966,8 @@ static struct io_context *current_io_context(gfp_t gfp_flags, int node)
 	if (likely(ret))
 		return ret;
 
-	ret = kmem_cache_alloc_node(iocontext_cachep, gfp_flags, node);
+	ret = alloc_io_context(gfp_flags, node);
 	if (ret) {
-		atomic_set(&ret->refcount, 1);
-		ret->task = current;
-		ret->ioprio_changed = 0;
-		ret->last_waited = jiffies; /* doesn't matter... */
-		ret->nr_batch_requests = 0; /* because this is 0 */
-		ret->aic = NULL;
-		ret->cic_root.rb_node = NULL;
-		ret->ioc_data = NULL;
 		/* make sure set_task_ioprio() sees the settings above */
 		smp_wmb();
 		tsk->io_context = ret;
@@ -3947,10 +3984,18 @@ 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);

fs/ioprio.c

@@ -41,18 +41,28 @@ static int set_task_ioprio(struct task_struct *task, int ioprio)
 		return err;
 
 	task_lock(task);
+	do {
+		ioc = task->io_context;
+		/* see wmb() in current_io_context() */
+		smp_read_barrier_depends();
+		if (ioc)
+			break;
 
-	task->ioprio = ioprio;
-
-	ioc = task->io_context;
-	/* see wmb() in current_io_context() */
-	smp_read_barrier_depends();
+		ioc = alloc_io_context(GFP_ATOMIC, -1);
+		if (!ioc) {
+			err = -ENOMEM;
+			break;
+		}
+		task->io_context = ioc;
+	} while (1);
 
-	if (ioc)
+	if (!err) {
+		ioc->ioprio = ioprio;
 		ioc->ioprio_changed = 1;
+	}
 
 	task_unlock(task);
-	return 0;
+	return err;
 }
 
 asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
@@ -75,8 +85,6 @@ asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
 
 			break;
 		case IOPRIO_CLASS_IDLE:
-			if (!capable(CAP_SYS_ADMIN))
-				return -EPERM;
 			break;
 		case IOPRIO_CLASS_NONE:
 			if (data)
@@ -148,7 +156,9 @@ static int get_task_ioprio(struct task_struct *p)
 	ret = security_task_getioprio(p);
 	if (ret)
 		goto out;
-	ret = p->ioprio;
+	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
+	if (p->io_context)
+		ret = p->io_context->ioprio;
 out:
 	return ret;
 }

include/linux/blkdev.h

@@ -34,83 +34,10 @@ struct sg_io_hdr;
 #define BLKDEV_MIN_RQ	4
 #define BLKDEV_MAX_RQ	128	/* Default maximum */
 
-/*
- * This is the per-process anticipatory I/O scheduler state.
- */
-struct as_io_context {
-	spinlock_t lock;
-
-	void (*dtor)(struct as_io_context *aic); /* destructor */
-	void (*exit)(struct as_io_context *aic); /* called on task exit */
-
-	unsigned long state;
-	atomic_t nr_queued; /* queued reads & sync writes */
-	atomic_t nr_dispatched; /* number of requests gone to the drivers */
-
-	/* IO History tracking */
-	/* Thinktime */
-	unsigned long last_end_request;
-	unsigned long ttime_total;
-	unsigned long ttime_samples;
-	unsigned long ttime_mean;
-	/* Layout pattern */
-	unsigned int seek_samples;
-	sector_t last_request_pos;
-	u64 seek_total;
-	sector_t seek_mean;
-};
-
-struct cfq_queue;
-struct cfq_io_context {
-	struct rb_node rb_node;
-	void *key;
-
-	struct cfq_queue *cfqq[2];
-
-	struct io_context *ioc;
-
-	unsigned long last_end_request;
-	sector_t last_request_pos;
-
-	unsigned long ttime_total;
-	unsigned long ttime_samples;
-	unsigned long ttime_mean;
-
-	unsigned int seek_samples;
-	u64 seek_total;
-	sector_t seek_mean;
-
-	struct list_head queue_list;
-
-	void (*dtor)(struct io_context *); /* destructor */
-	void (*exit)(struct io_context *); /* called on task exit */
-};
-
-/*
- * 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.
- */
-struct io_context {
-	atomic_t refcount;
-	struct task_struct *task;
-
-	unsigned int ioprio_changed;
-
-	/*
-	 * For request batching
-	 */
-	unsigned long last_waited; /* Time last woken after wait for request */
-	int nr_batch_requests;     /* Number of requests left in the batch */
-
-	struct as_io_context *aic;
-	struct rb_root cic_root;
-	void *ioc_data;
-};
-
-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);
 void copy_io_context(struct io_context **pdst, struct io_context **psrc);
 void swap_io_context(struct io_context **ioc1, struct io_context **ioc2);
 
@@ -894,6 +821,12 @@ static inline void exit_io_context(void)
 {
 }
 
+static inline int put_io_context(struct io_context *ioc)
+{
+	return 1;
+}
+
+
 #endif /* CONFIG_BLOCK */
 
 #endif

include/linux/init_task.h

@@ -137,7 +137,6 @@ extern struct group_info init_groups;
 		.time_slice	= HZ, 					\
 		.nr_cpus_allowed = NR_CPUS,				\
 	},								\
-	.ioprio		= 0,						\
 	.tasks		= LIST_HEAD_INIT(tsk.tasks),			\
 	.ptrace_children= LIST_HEAD_INIT(tsk.ptrace_children),		\
 	.ptrace_list	= LIST_HEAD_INIT(tsk.ptrace_list),		\

include/linux/iocontext.h

+#ifndef IOCONTEXT_H
+#define IOCONTEXT_H
+
+#include <linux/radix-tree.h>
+
+/*
+ * This is the per-process anticipatory I/O scheduler state.
+ */
+struct as_io_context {
+	spinlock_t lock;
+
+	void (*dtor)(struct as_io_context *aic); /* destructor */
+	void (*exit)(struct as_io_context *aic); /* called on task exit */
+
+	unsigned long state;
+	atomic_t nr_queued; /* queued reads & sync writes */
+	atomic_t nr_dispatched; /* number of requests gone to the drivers */
+
+	/* IO History tracking */
+	/* Thinktime */
+	unsigned long last_end_request;
+	unsigned long ttime_total;
+	unsigned long ttime_samples;
+	unsigned long ttime_mean;
+	/* Layout pattern */
+	unsigned int seek_samples;
+	sector_t last_request_pos;
+	u64 seek_total;
+	sector_t seek_mean;
+};
+
+struct cfq_queue;
+struct cfq_io_context {
+	void *key;
+	unsigned long dead_key;
+
+	struct cfq_queue *cfqq[2];
+
+	struct io_context *ioc;
+
+	unsigned long last_end_request;
+	sector_t last_request_pos;
+
+	unsigned long ttime_total;
+	unsigned long ttime_samples;
+	unsigned long ttime_mean;
+
+	unsigned int seek_samples;
+	u64 seek_total;
+	sector_t seek_mean;
+
+	struct list_head queue_list;
+
+	void (*dtor)(struct io_context *); /* destructor */
+	void (*exit)(struct io_context *); /* called on task exit */
+};
+
+/*
+ * 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;
+	atomic_t nr_tasks;
+
+	/* all the fields below are protected by this lock */
+	spinlock_t lock;
+
+	unsigned short ioprio;
+	unsigned short ioprio_changed;
+
+	/*
+	 * For request batching
+	 */
+	unsigned long last_waited; /* Time last woken after wait for request */
+	int nr_batch_requests;     /* Number of requests left in the batch */
+
+	struct as_io_context *aic;
+	struct radix_tree_root radix_root;
+	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

include/linux/ioprio.h

@@ -2,6 +2,7 @@
 #define IOPRIO_H
 
 #include <linux/sched.h>
+#include <linux/iocontext.h>
 
 /*
  * Gives us 8 prio classes with 13-bits of data for each class
@@ -45,18 +46,18 @@ enum {
  * the cpu scheduler nice value to an io priority
  */
 #define IOPRIO_NORM	(4)
-static inline int task_ioprio(struct task_struct *task)
+static inline int task_ioprio(struct io_context *ioc)
 {
-	if (ioprio_valid(task->ioprio))
-		return IOPRIO_PRIO_DATA(task->ioprio);
+	if (ioprio_valid(ioc->ioprio))
+		return IOPRIO_PRIO_DATA(ioc->ioprio);
 
 	return IOPRIO_NORM;
 }
 
-static inline int task_ioprio_class(struct task_struct *task)
+static inline int task_ioprio_class(struct io_context *ioc)
 {
-	if (ioprio_valid(task->ioprio))
-		return IOPRIO_PRIO_CLASS(task->ioprio);
+	if (ioprio_valid(ioc->ioprio))
+		return IOPRIO_PRIO_CLASS(ioc->ioprio);
 
 	return IOPRIO_CLASS_BE;
 }

include/linux/sched.h

@@ -27,6 +27,7 @@
 #define CLONE_NEWUSER		0x10000000	/* New user namespace */
 #define CLONE_NEWPID		0x20000000	/* New pid namespace */
 #define CLONE_NEWNET		0x40000000	/* New network namespace */
+#define CLONE_IO		0x80000000	/* Clone io context */
 
 /*
  * Scheduling policies
@@ -975,7 +976,6 @@ struct task_struct {
 	struct hlist_head preempt_notifiers;
 #endif
 
-	unsigned short ioprio;
 	/*
 	 * fpu_counter contains the number of consecutive context switches
 	 * that the FPU is used. If this is over a threshold, the lazy fpu

kernel/fork.c

@@ -51,6 +51,7 @@
 #include <linux/random.h>
 #include <linux/tty.h>
 #include <linux/proc_fs.h>
+#include <linux/blkdev.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -791,6 +792,31 @@ static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 	return error;
 }
 
+static int copy_io(unsigned long clone_flags, struct task_struct *tsk)
+{
+#ifdef CONFIG_BLOCK
+	struct io_context *ioc = current->io_context;
+
+	if (!ioc)
+		return 0;
+	/*
+	 * Share io context with parent, if CLONE_IO is set
+	 */
+	if (clone_flags & CLONE_IO) {
+		tsk->io_context = ioc_task_link(ioc);
+		if (unlikely(!tsk->io_context))
+			return -ENOMEM;
+	} else if (ioprio_valid(ioc->ioprio)) {
+		tsk->io_context = alloc_io_context(GFP_KERNEL, -1);
+		if (unlikely(!tsk->io_context))
+			return -ENOMEM;
+
+		tsk->io_context->ioprio = ioc->ioprio;
+	}
+#endif
+	return 0;
+}
+
 /*
  *	Helper to unshare the files of the current task.
  *	We don't want to expose copy_files internals to
@@ -1156,15 +1182,17 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 		goto bad_fork_cleanup_mm;
 	if ((retval = copy_namespaces(clone_flags, p)))
 		goto bad_fork_cleanup_keys;
+	if ((retval = copy_io(clone_flags, p)))
+		goto bad_fork_cleanup_namespaces;
 	retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
 	if (retval)
-		goto bad_fork_cleanup_namespaces;
+		goto bad_fork_cleanup_io;
 
 	if (pid != &init_struct_pid) {
 		retval = -ENOMEM;
 		pid = alloc_pid(task_active_pid_ns(p));
 		if (!pid)
-			goto bad_fork_cleanup_namespaces;
+			goto bad_fork_cleanup_io;
 
 		if (clone_flags & CLONE_NEWPID) {
 			retval = pid_ns_prepare_proc(task_active_pid_ns(p));
@@ -1234,9 +1262,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	/* Need tasklist lock for parent etc handling! */
 	write_lock_irq(&tasklist_lock);
 
-	/* for sys_ioprio_set(IOPRIO_WHO_PGRP) */
-	p->ioprio = current->ioprio;
-
 	/*
 	 * The task hasn't been attached yet, so its cpus_allowed mask will
 	 * not be changed, nor will its assigned CPU.
@@ -1328,6 +1353,8 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 bad_fork_free_pid:
 	if (pid != &init_struct_pid)
 		free_pid(pid);
+bad_fork_cleanup_io:
+	put_io_context(p->io_context);
 bad_fork_cleanup_namespaces:
 	exit_task_namespaces(p);
 bad_fork_cleanup_keys: