writeback: switch to per-bdi threads for flushing data

This gets rid of pdflush for bdi writeout and kupdated style cleaning.
pdflush writeout suffers from lack of locality and also requires more
threads to handle the same workload, since it has to work in a
non-blocking fashion against each queue. This also introduces lumpy
behaviour and potential request starvation, since pdflush can be starved
for queue access if others are accessing it. A sample ffsb workload that
does random writes to files is about 8% faster here on a simple SATA drive
during the benchmark phase. File layout also seems a LOT more smooth in
vmstat:

 r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
 0  1      0 608848   2652 375372    0    0     0 71024  604    24  1 10 48 42
 0  1      0 549644   2712 433736    0    0     0 60692  505    27  1  8 48 44
 1  0      0 476928   2784 505192    0    0     4 29540  553    24  0  9 53 37
 0  1      0 457972   2808 524008    0    0     0 54876  331    16  0  4 38 58
 0  1      0 366128   2928 614284    0    0     4 92168  710    58  0 13 53 34
 0  1      0 295092   3000 684140    0    0     0 62924  572    23  0  9 53 37
 0  1      0 236592   3064 741704    0    0     4 58256  523    17  0  8 48 44
 0  1      0 165608   3132 811464    0    0     0 57460  560    21  0  8 54 38
 0  1      0 102952   3200 873164    0    0     4 74748  540    29  1 10 48 41
 0  1      0  48604   3252 926472    0    0     0 53248  469    29  0  7 47 45

where vanilla tends to fluctuate a lot in the creation phase:

 r  b   swpd   free   buff  cache   si   so    bi    bo   in    cs us sy id wa
 1  1      0 678716   5792 303380    0    0     0 74064  565    50  1 11 52 36
 1  0      0 662488   5864 319396    0    0     4   352  302   329  0  2 47 51
 0  1      0 599312   5924 381468    0    0     0 78164  516    55  0  9 51 40
 0  1      0 519952   6008 459516    0    0     4 78156  622    56  1 11 52 37
 1  1      0 436640   6092 541632    0    0     0 82244  622    54  0 11 48 41
 0  1      0 436640   6092 541660    0    0     0     8  152    39  0  0 51 49
 0  1      0 332224   6200 644252    0    0     4 102800  728    46  1 13 49 36
 1  0      0 274492   6260 701056    0    0     4 12328  459    49  0  7 50 43
 0  1      0 211220   6324 763356    0    0     0 106940  515    37  1 10 51 39
 1  0      0 160412   6376 813468    0    0     0  8224  415    43  0  6 49 45
 1  1      0  85980   6452 886556    0    0     4 113516  575    39  1 11 54 34
 0  2      0  85968   6452 886620    0    0     0  1640  158   211  0  0 46 54

A 10 disk test with btrfs performs 26% faster with per-bdi flushing. A
SSD based writeback test on XFS performs over 20% better as well, with
the throughput being very stable around 1GB/sec, where pdflush only
manages 750MB/sec and fluctuates wildly while doing so. Random buffered
writes to many files behave a lot better as well, as does random mmap'ed
writes.

A separate thread is added to sync the super blocks. In the long term,
adding sync_supers_bdi() functionality could get rid of this thread again.

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

1 files changed, 323 insertions, 18 deletions

diff --git a/mm/backing-dev.c b/mm/backing-dev.c
index 6f163e0f0509..7f3fa79f25c0 100644
--- a/mm/backing-dev.c
+++ b/mm/backing-dev.c
@@ -1,8 +1,11 @@
 
 #include <linux/wait.h>
 #include <linux/backing-dev.h>
+#include <linux/kthread.h>
+#include <linux/freezer.h>
 #include <linux/fs.h>
 #include <linux/pagemap.h>
+#include <linux/mm.h>
 #include <linux/sched.h>
 #include <linux/module.h>
 #include <linux/writeback.h>
@@ -22,8 +25,18 @@ struct backing_dev_info default_backing_dev_info = {
 EXPORT_SYMBOL_GPL(default_backing_dev_info);
 
 static struct class *bdi_class;
-DEFINE_MUTEX(bdi_lock);
+DEFINE_SPINLOCK(bdi_lock);
 LIST_HEAD(bdi_list);
+LIST_HEAD(bdi_pending_list);
+
+static struct task_struct *sync_supers_tsk;
+static struct timer_list sync_supers_timer;
+
+static int bdi_sync_supers(void *);
+static void sync_supers_timer_fn(unsigned long);
+static void arm_supers_timer(void);
+
+static void bdi_add_default_flusher_task(struct backing_dev_info *bdi);
 
 #ifdef CONFIG_DEBUG_FS
 #include <linux/debugfs.h>
@@ -187,6 +200,13 @@ static int __init default_bdi_init(void)
 {
 	int err;
 
+	sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
+	BUG_ON(IS_ERR(sync_supers_tsk));
+
+	init_timer(&sync_supers_timer);
+	setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
+	arm_supers_timer();
+
 	err = bdi_init(&default_backing_dev_info);
 	if (!err)
 		bdi_register(&default_backing_dev_info, NULL, "default");
@@ -195,6 +215,242 @@ static int __init default_bdi_init(void)
 }
 subsys_initcall(default_bdi_init);
 
+static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
+{
+	memset(wb, 0, sizeof(*wb));
+
+	wb->bdi = bdi;
+	wb->last_old_flush = jiffies;
+	INIT_LIST_HEAD(&wb->b_dirty);
+	INIT_LIST_HEAD(&wb->b_io);
+	INIT_LIST_HEAD(&wb->b_more_io);
+}
+
+static void bdi_task_init(struct backing_dev_info *bdi,
+			  struct bdi_writeback *wb)
+{
+	struct task_struct *tsk = current;
+
+	spin_lock(&bdi->wb_lock);
+	list_add_tail_rcu(&wb->list, &bdi->wb_list);
+	spin_unlock(&bdi->wb_lock);
+
+	tsk->flags |= PF_FLUSHER | PF_SWAPWRITE;
+	set_freezable();
+
+	/*
+	 * Our parent may run at a different priority, just set us to normal
+	 */
+	set_user_nice(tsk, 0);
+}
+
+static int bdi_start_fn(void *ptr)
+{
+	struct bdi_writeback *wb = ptr;
+	struct backing_dev_info *bdi = wb->bdi;
+	int ret;
+
+	/*
+	 * Add us to the active bdi_list
+	 */
+	spin_lock(&bdi_lock);
+	list_add(&bdi->bdi_list, &bdi_list);
+	spin_unlock(&bdi_lock);
+
+	bdi_task_init(bdi, wb);
+
+	/*
+	 * Clear pending bit and wakeup anybody waiting to tear us down
+	 */
+	clear_bit(BDI_pending, &bdi->state);
+	smp_mb__after_clear_bit();
+	wake_up_bit(&bdi->state, BDI_pending);
+
+	ret = bdi_writeback_task(wb);
+
+	/*
+	 * Remove us from the list
+	 */
+	spin_lock(&bdi->wb_lock);
+	list_del_rcu(&wb->list);
+	spin_unlock(&bdi->wb_lock);
+
+	/*
+	 * Flush any work that raced with us exiting. No new work
+	 * will be added, since this bdi isn't discoverable anymore.
+	 */
+	if (!list_empty(&bdi->work_list))
+		wb_do_writeback(wb, 1);
+
+	wb->task = NULL;
+	return ret;
+}
+
+int bdi_has_dirty_io(struct backing_dev_info *bdi)
+{
+	return wb_has_dirty_io(&bdi->wb);
+}
+
+static void bdi_flush_io(struct backing_dev_info *bdi)
+{
+	struct writeback_control wbc = {
+		.bdi			= bdi,
+		.sync_mode		= WB_SYNC_NONE,
+		.older_than_this	= NULL,
+		.range_cyclic		= 1,
+		.nr_to_write		= 1024,
+	};
+
+	writeback_inodes_wbc(&wbc);
+}
+
+/*
+ * kupdated() used to do this. We cannot do it from the bdi_forker_task()
+ * or we risk deadlocking on ->s_umount. The longer term solution would be
+ * to implement sync_supers_bdi() or similar and simply do it from the
+ * bdi writeback tasks individually.
+ */
+static int bdi_sync_supers(void *unused)
+{
+	set_user_nice(current, 0);
+
+	while (!kthread_should_stop()) {
+		set_current_state(TASK_INTERRUPTIBLE);
+		schedule();
+
+		/*
+		 * Do this periodically, like kupdated() did before.
+		 */
+		sync_supers();
+	}
+
+	return 0;
+}
+
+static void arm_supers_timer(void)
+{
+	unsigned long next;
+
+	next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
+	mod_timer(&sync_supers_timer, round_jiffies_up(next));
+}
+
+static void sync_supers_timer_fn(unsigned long unused)
+{
+	wake_up_process(sync_supers_tsk);
+	arm_supers_timer();
+}
+
+static int bdi_forker_task(void *ptr)
+{
+	struct bdi_writeback *me = ptr;
+
+	bdi_task_init(me->bdi, me);
+
+	for (;;) {
+		struct backing_dev_info *bdi, *tmp;
+		struct bdi_writeback *wb;
+
+		/*
+		 * Temporary measure, we want to make sure we don't see
+		 * dirty data on the default backing_dev_info
+		 */
+		if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
+			wb_do_writeback(me, 0);
+
+		spin_lock(&bdi_lock);
+
+		/*
+		 * Check if any existing bdi's have dirty data without
+		 * a thread registered. If so, set that up.
+		 */
+		list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) {
+			if (bdi->wb.task)
+				continue;
+			if (list_empty(&bdi->work_list) &&
+			    !bdi_has_dirty_io(bdi))
+				continue;
+
+			bdi_add_default_flusher_task(bdi);
+		}
+
+		set_current_state(TASK_INTERRUPTIBLE);
+
+		if (list_empty(&bdi_pending_list)) {
+			unsigned long wait;
+
+			spin_unlock(&bdi_lock);
+			wait = msecs_to_jiffies(dirty_writeback_interval * 10);
+			schedule_timeout(wait);
+			try_to_freeze();
+			continue;
+		}
+
+		__set_current_state(TASK_RUNNING);
+
+		/*
+		 * This is our real job - check for pending entries in
+		 * bdi_pending_list, and create the tasks that got added
+		 */
+		bdi = list_entry(bdi_pending_list.next, struct backing_dev_info,
+				 bdi_list);
+		list_del_init(&bdi->bdi_list);
+		spin_unlock(&bdi_lock);
+
+		wb = &bdi->wb;
+		wb->task = kthread_run(bdi_start_fn, wb, "flush-%s",
+					dev_name(bdi->dev));
+		/*
+		 * If task creation fails, then readd the bdi to
+		 * the pending list and force writeout of the bdi
+		 * from this forker thread. That will free some memory
+		 * and we can try again.
+		 */
+		if (IS_ERR(wb->task)) {
+			wb->task = NULL;
+
+			/*
+			 * Add this 'bdi' to the back, so we get
+			 * a chance to flush other bdi's to free
+			 * memory.
+			 */
+			spin_lock(&bdi_lock);
+			list_add_tail(&bdi->bdi_list, &bdi_pending_list);
+			spin_unlock(&bdi_lock);
+
+			bdi_flush_io(bdi);
+		}
+	}
+
+	return 0;
+}
+
+/*
+ * Add the default flusher task that gets created for any bdi
+ * that has dirty data pending writeout
+ */
+void static bdi_add_default_flusher_task(struct backing_dev_info *bdi)
+{
+	if (!bdi_cap_writeback_dirty(bdi))
+		return;
+
+	/*
+	 * Check with the helper whether to proceed adding a task. Will only
+	 * abort if we two or more simultanous calls to
+	 * bdi_add_default_flusher_task() occured, further additions will block
+	 * waiting for previous additions to finish.
+	 */
+	if (!test_and_set_bit(BDI_pending, &bdi->state)) {
+		list_move_tail(&bdi->bdi_list, &bdi_pending_list);
+
+		/*
+		 * We are now on the pending list, wake up bdi_forker_task()
+		 * to finish the job and add us back to the active bdi_list
+		 */
+		wake_up_process(default_backing_dev_info.wb.task);
+	}
+}
+
 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 		const char *fmt, ...)
 {
@@ -213,13 +469,34 @@ int bdi_register(struct backing_dev_info *bdi, struct device *parent,
 		goto exit;
 	}
 
-	mutex_lock(&bdi_lock);
+	spin_lock(&bdi_lock);
 	list_add_tail(&bdi->bdi_list, &bdi_list);
-	mutex_unlock(&bdi_lock);
+	spin_unlock(&bdi_lock);
 
 	bdi->dev = dev;
-	bdi_debug_register(bdi, dev_name(dev));
 
+	/*
+	 * Just start the forker thread for our default backing_dev_info,
+	 * and add other bdi's to the list. They will get a thread created
+	 * on-demand when they need it.
+	 */
+	if (bdi_cap_flush_forker(bdi)) {
+		struct bdi_writeback *wb = &bdi->wb;
+
+		wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s",
+						dev_name(dev));
+		if (IS_ERR(wb->task)) {
+			wb->task = NULL;
+			ret = -ENOMEM;
+
+			spin_lock(&bdi_lock);
+			list_del(&bdi->bdi_list);
+			spin_unlock(&bdi_lock);
+			goto exit;
+		}
+	}
+
+	bdi_debug_register(bdi, dev_name(dev));
 exit:
 	return ret;
 }
@@ -231,17 +508,42 @@ int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
 }
 EXPORT_SYMBOL(bdi_register_dev);
 
-static void bdi_remove_from_list(struct backing_dev_info *bdi)
+/*
+ * Remove bdi from the global list and shutdown any threads we have running
+ */
+static void bdi_wb_shutdown(struct backing_dev_info *bdi)
 {
-	mutex_lock(&bdi_lock);
+	struct bdi_writeback *wb;
+
+	if (!bdi_cap_writeback_dirty(bdi))
+		return;
+
+	/*
+	 * If setup is pending, wait for that to complete first
+	 */
+	wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
+			TASK_UNINTERRUPTIBLE);
+
+	/*
+	 * Make sure nobody finds us on the bdi_list anymore
+	 */
+	spin_lock(&bdi_lock);
 	list_del(&bdi->bdi_list);
-	mutex_unlock(&bdi_lock);
+	spin_unlock(&bdi_lock);
+
+	/*
+	 * Finally, kill the kernel threads. We don't need to be RCU
+	 * safe anymore, since the bdi is gone from visibility.
+	 */
+	list_for_each_entry(wb, &bdi->wb_list, list)
+		kthread_stop(wb->task);
 }
 
 void bdi_unregister(struct backing_dev_info *bdi)
 {
 	if (bdi->dev) {
-		bdi_remove_from_list(bdi);
+		if (!bdi_cap_flush_forker(bdi))
+			bdi_wb_shutdown(bdi);
 		bdi_debug_unregister(bdi);
 		device_unregister(bdi->dev);
 		bdi->dev = NULL;
@@ -251,18 +553,25 @@ EXPORT_SYMBOL(bdi_unregister);
 
 int bdi_init(struct backing_dev_info *bdi)
 {
-	int i;
-	int err;
+	int i, err;
 
 	bdi->dev = NULL;
 
 	bdi->min_ratio = 0;
 	bdi->max_ratio = 100;
 	bdi->max_prop_frac = PROP_FRAC_BASE;
+	spin_lock_init(&bdi->wb_lock);
 	INIT_LIST_HEAD(&bdi->bdi_list);
-	INIT_LIST_HEAD(&bdi->b_io);
-	INIT_LIST_HEAD(&bdi->b_dirty);
-	INIT_LIST_HEAD(&bdi->b_more_io);
+	INIT_LIST_HEAD(&bdi->wb_list);
+	INIT_LIST_HEAD(&bdi->work_list);
+
+	bdi_wb_init(&bdi->wb, bdi);
+
+	/*
+	 * Just one thread support for now, hard code mask and count
+	 */
+	bdi->wb_mask = 1;
+	bdi->wb_cnt = 1;
 
 	for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
 		err = percpu_counter_init(&bdi->bdi_stat[i], 0);
@@ -277,8 +586,6 @@ int bdi_init(struct backing_dev_info *bdi)
 err:
 		while (i--)
 			percpu_counter_destroy(&bdi->bdi_stat[i]);
-
-		bdi_remove_from_list(bdi);
 	}
 
 	return err;
@@ -289,9 +596,7 @@ void bdi_destroy(struct backing_dev_info *bdi)
 {
 	int i;
 
-	WARN_ON(!list_empty(&bdi->b_dirty));
-	WARN_ON(!list_empty(&bdi->b_io));
-	WARN_ON(!list_empty(&bdi->b_more_io));
+	WARN_ON(bdi_has_dirty_io(bdi));
 
 	bdi_unregister(bdi);