bcachefs: Move journal reclaim to a kthread

This is to make tracing easier.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/alloc_background.c

@@ -1409,7 +1409,7 @@ int bch2_dev_allocator_start(struct bch_dev *ca)
 		return 0;
 
 	p = kthread_create(bch2_allocator_thread, ca,
-			   "bch_alloc[%s]", ca->name);
+			   "bch-alloc/%s", ca->name);
 	if (IS_ERR(p))
 		return PTR_ERR(p);
 

fs/bcachefs/bcachefs.h

@@ -650,7 +650,6 @@ struct bch_fs {
 	struct workqueue_struct	*wq;
 	/* copygc needs its own workqueue for index updates.. */
 	struct workqueue_struct	*copygc_wq;
-	struct workqueue_struct	*journal_reclaim_wq;
 
 	/* ALLOCATION */
 	struct delayed_work	pd_controllers_update;

fs/bcachefs/btree_gc.c

@@ -1427,7 +1427,7 @@ int bch2_gc_thread_start(struct bch_fs *c)
 
 	BUG_ON(c->gc_thread);
 
-	p = kthread_create(bch2_gc_thread, c, "bch_gc");
+	p = kthread_create(bch2_gc_thread, c, "bch-gc/%s", c->name);
 	if (IS_ERR(p))
 		return PTR_ERR(p);
 

fs/bcachefs/btree_key_cache.c

@@ -497,7 +497,7 @@ bool bch2_btree_insert_key_cached(struct btree_trans *trans,
 				&ck->journal, btree_key_cache_journal_flush);
 
 	if (kick_reclaim)
-		mod_delayed_work(c->journal_reclaim_wq, &c->journal.reclaim_work, 0);
+		journal_reclaim_kick(&c->journal);
 	return true;
 }
 

fs/bcachefs/chardev.c

@@ -341,7 +341,8 @@ static long bch2_ioctl_data(struct bch_fs *c,
 	ctx->c = c;
 	ctx->arg = arg;
 
-	ctx->thread = kthread_create(bch2_data_thread, ctx, "[bcachefs]");
+	ctx->thread = kthread_create(bch2_data_thread, ctx,
+				     "bch-data/%s", c->name);
 	if (IS_ERR(ctx->thread)) {
 		ret = PTR_ERR(ctx->thread);
 		goto err;

fs/bcachefs/journal.c

@@ -225,11 +225,14 @@ static bool journal_entry_close(struct journal *j)
  */
 static int journal_entry_open(struct journal *j)
 {
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
 	struct journal_buf *buf = journal_cur_buf(j);
 	union journal_res_state old, new;
 	int u64s;
 	u64 v;
 
+	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
+
 	lockdep_assert_held(&j->lock);
 	BUG_ON(journal_entry_is_open(j));
 
@@ -480,8 +483,10 @@ static bool journal_preres_available(struct journal *j,
 {
 	bool ret = bch2_journal_preres_get_fast(j, res, new_u64s, flags);
 
-	if (!ret)
-		bch2_journal_reclaim_work(&j->reclaim_work.work);
+	if (!ret && mutex_trylock(&j->reclaim_lock)) {
+		bch2_journal_reclaim(j);
+		mutex_unlock(&j->reclaim_lock);
+	}
 
 	return ret;
 }
@@ -888,7 +893,7 @@ void bch2_fs_journal_stop(struct journal *j)
 		j->last_empty_seq + 1 != journal_cur_seq(j)));
 
 	cancel_delayed_work_sync(&j->write_work);
-	cancel_delayed_work_sync(&j->reclaim_work);
+	bch2_journal_reclaim_stop(j);
 }
 
 int bch2_fs_journal_start(struct journal *j, u64 cur_seq,
@@ -1019,7 +1024,6 @@ int bch2_fs_journal_init(struct journal *j)
 	spin_lock_init(&j->err_lock);
 	init_waitqueue_head(&j->wait);
 	INIT_DELAYED_WORK(&j->write_work, journal_write_work);
-	INIT_DELAYED_WORK(&j->reclaim_work, bch2_journal_reclaim_work);
 	init_waitqueue_head(&j->pin_flush_wait);
 	mutex_init(&j->reclaim_lock);
 	mutex_init(&j->discard_lock);
@@ -1071,6 +1075,8 @@ void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 	       "last_seq:\t\t%llu\n"
 	       "last_seq_ondisk:\t%llu\n"
 	       "prereserved:\t\t%u/%u\n"
+	       "nr direct reclaim:\t%llu\n"
+	       "nr background reclaim:\t%llu\n"
 	       "current entry sectors:\t%u\n"
 	       "current entry error:\t%u\n"
 	       "current entry:\t\t",
@@ -1080,6 +1086,8 @@ void bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 	       j->last_seq_ondisk,
 	       j->prereserved.reserved,
 	       j->prereserved.remaining,
+	       j->nr_direct_reclaim,
+	       j->nr_background_reclaim,
 	       j->cur_entry_sectors,
 	       j->cur_entry_error);
 

fs/bcachefs/journal_io.c

@@ -993,7 +993,7 @@ static void journal_write_done(struct closure *cl)
 	 * Must come before signaling write completion, for
 	 * bch2_fs_journal_stop():
 	 */
-	mod_delayed_work(c->journal_reclaim_wq, &j->reclaim_work, 0);
+	journal_reclaim_kick(&c->journal);
 
 	/* also must come before signalling write completion: */
 	closure_debug_destroy(cl);
@@ -1044,6 +1044,8 @@ void bch2_journal_write(struct closure *cl)
 	unsigned i, sectors, bytes, u64s;
 	int ret;
 
+	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
+
 	bch2_journal_pin_put(j, le64_to_cpu(w->data->seq));
 
 	journal_buf_realloc(j, w);

fs/bcachefs/journal_reclaim.c

@@ -9,6 +9,7 @@
 #include "super.h"
 #include "trace.h"
 
+#include <linux/kthread.h>
 #include <linux/sched/mm.h>
 
 /* Free space calculations: */
@@ -534,9 +535,10 @@ static u64 journal_seq_to_flush(struct journal *j)
  * 512 journal entries or 25% of all journal buckets, then
  * journal_next_bucket() should not stall.
  */
-void bch2_journal_reclaim(struct journal *j)
+static void __bch2_journal_reclaim(struct journal *j, bool direct)
 {
 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	bool kthread = (current->flags & PF_KTHREAD) != 0;
 	u64 seq_to_flush, nr_flushed = 0;
 	size_t min_nr;
 	unsigned flags;
@@ -551,6 +553,9 @@ void bch2_journal_reclaim(struct journal *j)
 	flags = memalloc_noreclaim_save();
 
 	do {
+		if (kthread && kthread_should_stop())
+			break;
+
 		bch2_journal_do_discards(j);
 
 		seq_to_flush = journal_seq_to_flush(j);
@@ -582,26 +587,83 @@ void bch2_journal_reclaim(struct journal *j)
 				c->btree_key_cache.nr_dirty,
 				c->btree_key_cache.nr_keys);
 
-		nr_flushed += journal_flush_pins(j, seq_to_flush, min_nr);
+		nr_flushed = journal_flush_pins(j, seq_to_flush, min_nr);
+
+		if (direct)
+			j->nr_direct_reclaim += nr_flushed;
+		else
+			j->nr_background_reclaim += nr_flushed;
+		trace_journal_reclaim_finish(c, nr_flushed);
 	} while (min_nr);
 
 	memalloc_noreclaim_restore(flags);
+}
 
-	trace_journal_reclaim_finish(c, nr_flushed);
-
-	if (!bch2_journal_error(j))
-		queue_delayed_work(c->journal_reclaim_wq, &j->reclaim_work,
-				   msecs_to_jiffies(j->reclaim_delay_ms));
+void bch2_journal_reclaim(struct journal *j)
+{
+	__bch2_journal_reclaim(j, true);
 }
 
-void bch2_journal_reclaim_work(struct work_struct *work)
+static int bch2_journal_reclaim_thread(void *arg)
 {
-	struct journal *j = container_of(to_delayed_work(work),
-				struct journal, reclaim_work);
+	struct journal *j = arg;
+	unsigned long next;
+
+	while (!kthread_should_stop()) {
+		j->reclaim_kicked = false;
 
 		mutex_lock(&j->reclaim_lock);
-	bch2_journal_reclaim(j);
+		__bch2_journal_reclaim(j, false);
 		mutex_unlock(&j->reclaim_lock);
+
+		next = j->last_flushed + msecs_to_jiffies(j->reclaim_delay_ms);
+
+		while (1) {
+			set_current_state(TASK_INTERRUPTIBLE);
+			if (kthread_should_stop())
+				break;
+			if (j->reclaim_kicked)
+				break;
+			if (time_after_eq(jiffies, next))
+				break;
+			schedule_timeout(next - jiffies);
+
+		}
+		__set_current_state(TASK_RUNNING);
+	}
+
+	return 0;
+}
+
+void bch2_journal_reclaim_stop(struct journal *j)
+{
+	struct task_struct *p = j->reclaim_thread;
+
+	j->reclaim_thread = NULL;
+
+	if (p) {
+		kthread_stop(p);
+		put_task_struct(p);
+	}
+}
+
+int bch2_journal_reclaim_start(struct journal *j)
+{
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+	struct task_struct *p;
+
+	if (j->reclaim_thread)
+		return 0;
+
+	p = kthread_create(bch2_journal_reclaim_thread, j,
+			   "bch-reclaim/%s", c->name);
+	if (IS_ERR(p))
+		return PTR_ERR(p);
+
+	get_task_struct(p);
+	j->reclaim_thread = p;
+	wake_up_process(p);
+	return 0;
 }
 
 static int journal_flush_done(struct journal *j, u64 seq_to_flush,

fs/bcachefs/journal_reclaim.h

@@ -10,6 +10,17 @@ enum journal_space_from {
 	journal_space_clean,
 };
 
+static inline void journal_reclaim_kick(struct journal *j)
+{
+	struct task_struct *p = READ_ONCE(j->reclaim_thread);
+
+	if (p && !j->reclaim_kicked) {
+		j->reclaim_kicked = true;
+		if (p)
+			wake_up_process(p);
+	}
+}
+
 unsigned bch2_journal_dev_buckets_available(struct journal *,
 					    struct journal_device *,
 					    enum journal_space_from);
@@ -55,7 +66,9 @@ void bch2_journal_pin_flush(struct journal *, struct journal_entry_pin *);
 
 void bch2_journal_do_discards(struct journal *);
 void bch2_journal_reclaim(struct journal *);
-void bch2_journal_reclaim_work(struct work_struct *);
+
+void bch2_journal_reclaim_stop(struct journal *);
+int bch2_journal_reclaim_start(struct journal *);
 
 bool bch2_journal_flush_pins(struct journal *, u64);
 

fs/bcachefs/journal_types.h

@@ -216,8 +216,12 @@ struct journal {
 	struct write_point	wp;
 	spinlock_t		err_lock;
 
-	struct delayed_work	reclaim_work;
 	struct mutex		reclaim_lock;
+	struct task_struct	*reclaim_thread;
+	bool			reclaim_kicked;
+	u64			nr_direct_reclaim;
+	u64			nr_background_reclaim;
+
 	unsigned long		last_flushed;
 	struct journal_entry_pin *flush_in_progress;
 	wait_queue_head_t	pin_flush_wait;

fs/bcachefs/movinggc.c

@@ -345,7 +345,7 @@ int bch2_copygc_start(struct bch_fs *c)
 	if (bch2_fs_init_fault("copygc_start"))
 		return -ENOMEM;
 
-	t = kthread_create(bch2_copygc_thread, c, "bch_copygc");
+	t = kthread_create(bch2_copygc_thread, c, "bch-copygc/%s", c->name);
 	if (IS_ERR(t))
 		return PTR_ERR(t);
 

fs/bcachefs/rebalance.c

@@ -314,7 +314,7 @@ int bch2_rebalance_start(struct bch_fs *c)
 	if (c->opts.nochanges)
 		return 0;
 
-	p = kthread_create(bch2_rebalance_thread, c, "bch_rebalance");
+	p = kthread_create(bch2_rebalance_thread, c, "bch-rebalance/%s", c->name);
 	if (IS_ERR(p))
 		return PTR_ERR(p);
 

fs/bcachefs/super.c

@@ -49,7 +49,6 @@
 #include <linux/debugfs.h>
 #include <linux/device.h>
 #include <linux/idr.h>
-#include <linux/kthread.h>
 #include <linux/module.h>
 #include <linux/percpu.h>
 #include <linux/random.h>
@@ -266,7 +265,7 @@ static void bch2_writes_disabled(struct percpu_ref *writes)
 void bch2_fs_read_only(struct bch_fs *c)
 {
 	if (!test_bit(BCH_FS_RW, &c->flags)) {
-		cancel_delayed_work_sync(&c->journal.reclaim_work);
+		BUG_ON(c->journal.reclaim_thread);
 		return;
 	}
 
@@ -424,6 +423,12 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 
 	set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
 
+	ret = bch2_journal_reclaim_start(&c->journal);
+	if (ret) {
+		bch_err(c, "error starting journal reclaim: %i", ret);
+		return ret;
+	}
+
 	if (!early) {
 		ret = bch2_fs_read_write_late(c);
 		if (ret)
@@ -432,9 +437,6 @@ static int __bch2_fs_read_write(struct bch_fs *c, bool early)
 
 	percpu_ref_reinit(&c->writes);
 	set_bit(BCH_FS_RW, &c->flags);
-
-	queue_delayed_work(c->journal_reclaim_wq,
-			   &c->journal.reclaim_work, 0);
 	return 0;
 err:
 	__bch2_fs_read_only(c);
@@ -503,8 +505,6 @@ static void __bch2_fs_free(struct bch_fs *c)
 	kfree(c->unused_inode_hints);
 	free_heap(&c->copygc_heap);
 
-	if (c->journal_reclaim_wq)
-		destroy_workqueue(c->journal_reclaim_wq);
 	if (c->copygc_wq)
 		destroy_workqueue(c->copygc_wq);
 	if (c->wq)
@@ -758,8 +758,6 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 				WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
-	    !(c->journal_reclaim_wq = alloc_workqueue("bcachefs_journal_reclaim",
-				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
 	    percpu_ref_init(&c->writes, bch2_writes_disabled,
 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||