bcachefs: Add a workqueue for btree io completions

Also, clean up workqueue usage - we shouldn't be using system
workqueues, pretty much everything we do needs to be on our own
WQ_MEM_RECLAIM workqueues.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

fs/bcachefs/bcachefs.h

@@ -627,6 +627,7 @@ struct bch_fs {
 
 	/* BTREE CACHE */
 	struct bio_set		btree_bio;
+	struct workqueue_struct	*io_complete_wq;
 
 	struct btree_root	btree_roots[BTREE_ID_NR];
 	struct mutex		btree_root_lock;
@@ -664,7 +665,7 @@ struct bch_fs {
 
 	struct btree_key_cache	btree_key_cache;
 
-	struct workqueue_struct	*wq;
+	struct workqueue_struct	*btree_update_wq;
 	/* copygc needs its own workqueue for index updates.. */
 	struct workqueue_struct	*copygc_wq;
 

fs/bcachefs/btree_io.c

@@ -1109,7 +1109,7 @@ static void btree_node_read_endio(struct bio *bio)
 		bch2_latency_acct(ca, rb->start_time, READ);
 	}
 
-	queue_work(system_unbound_wq, &rb->work);
+	queue_work(c->io_complete_wq, &rb->work);
 }
 
 struct btree_node_read_all {
@@ -1362,7 +1362,8 @@ static int btree_node_read_all_replicas(struct bch_fs *c, struct btree *b, bool
 		closure_sync(&ra->cl);
 		btree_node_read_all_replicas_done(&ra->cl);
 	} else {
-		continue_at(&ra->cl, btree_node_read_all_replicas_done, system_unbound_wq);
+		continue_at(&ra->cl, btree_node_read_all_replicas_done,
+			    c->io_complete_wq);
 	}
 
 	return 0;
@@ -1433,7 +1434,7 @@ void bch2_btree_node_read(struct bch_fs *c, struct btree *b,
 		if (sync)
 			btree_node_read_work(&rb->work);
 		else
-			queue_work(system_unbound_wq, &rb->work);
+			queue_work(c->io_complete_wq, &rb->work);
 	}
 }
 
@@ -1600,7 +1601,7 @@ static void btree_node_write_work(struct work_struct *work)
 		bio_list_add(&c->btree_write_error_list, &wbio->wbio.bio);
 		spin_unlock_irqrestore(&c->btree_write_error_lock, flags);
 
-		queue_work(c->wq, &c->btree_write_error_work);
+		queue_work(c->btree_update_wq, &c->btree_write_error_work);
 		return;
 	}
 
@@ -1639,7 +1640,7 @@ static void btree_node_write_endio(struct bio *bio)
 			container_of(orig, struct btree_write_bio, wbio);
 
 		INIT_WORK(&wb->work, btree_node_write_work);
-		queue_work(system_unbound_wq, &wb->work);
+		queue_work(c->io_complete_wq, &wb->work);
 	}
 }
 
@@ -1900,7 +1901,7 @@ void __bch2_btree_node_write(struct bch_fs *c, struct btree *b)
 	atomic64_add(sectors_to_write, &c->btree_writes_sectors);
 
 	INIT_WORK(&wbio->work, btree_write_submit);
-	schedule_work(&wbio->work);
+	queue_work(c->io_complete_wq, &wbio->work);
 	return;
 err:
 	set_btree_node_noevict(b);

fs/bcachefs/btree_update_interior.c

@@ -908,7 +908,8 @@ void bch2_btree_update_done(struct btree_update *as)
 
 	bch2_btree_reserve_put(as);
 
-	continue_at(&as->cl, btree_update_set_nodes_written, system_freezable_wq);
+	continue_at(&as->cl, btree_update_set_nodes_written,
+		    as->c->btree_interior_update_worker);
 }
 
 struct btree_update *
@@ -1847,7 +1848,7 @@ void bch2_btree_node_rewrite_async(struct bch_fs *c, struct btree *b)
 	a->seq		= b->data->keys.seq;
 
 	INIT_WORK(&a->work, async_btree_node_rewrite_work);
-	queue_work(system_long_wq, &a->work);
+	queue_work(c->btree_interior_update_worker, &a->work);
 }
 
 static void __bch2_btree_node_update_key(struct bch_fs *c,

fs/bcachefs/io.c

@@ -1440,7 +1440,7 @@ static void promote_start(struct promote_op *op, struct bch_read_bio *rbio)
 	bch2_migrate_read_done(&op->write, rbio);
 
 	closure_init(cl, NULL);
-	closure_call(&op->write.op.cl, bch2_write, c->wq, cl);
+	closure_call(&op->write.op.cl, bch2_write, c->btree_update_wq, cl);
 	closure_return_with_destructor(cl, promote_done);
 }
 
@@ -1823,6 +1823,13 @@ static void __bch2_read_endio(struct work_struct *work)
 	if (bch2_crc_cmp(csum, rbio->pick.crc.csum) && !c->opts.no_data_io)
 		goto csum_err;
 
+	/*
+	 * XXX
+	 * We need to rework the narrow_crcs path to deliver the read completion
+	 * first, and then punt to a different workqueue, otherwise we're
+	 * holding up reads while doing btree updates which is bad for memory
+	 * reclaim.
+	 */
 	if (unlikely(rbio->narrow_crcs))
 		bch2_rbio_narrow_crcs(rbio);
 

fs/bcachefs/io.h

@@ -62,7 +62,7 @@ static inline struct workqueue_struct *index_update_wq(struct bch_write_op *op)
 {
 	return op->alloc_reserve == RESERVE_MOVINGGC
 		? op->c->copygc_wq
-		: op->c->wq;
+		: op->c->btree_update_wq;
 }
 
 int bch2_sum_sector_overwrites(struct btree_trans *, struct btree_iter *,

fs/bcachefs/journal.c

@@ -117,7 +117,9 @@ void bch2_journal_halt(struct journal *j)
 
 void __bch2_journal_buf_put(struct journal *j)
 {
-	closure_call(&j->io, bch2_journal_write, system_highpri_wq, NULL);
+	struct bch_fs *c = container_of(j, struct bch_fs, journal);
+
+	closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 }
 
 /*
@@ -303,7 +305,7 @@ static int journal_entry_open(struct journal *j)
 				       j->res_get_blocked_start);
 	j->res_get_blocked_start = 0;
 
-	mod_delayed_work(system_freezable_wq,
+	mod_delayed_work(c->io_complete_wq,
 			 &j->write_work,
 			 msecs_to_jiffies(j->write_delay_ms));
 	journal_wake(j);

fs/bcachefs/journal_io.c

@@ -1295,12 +1295,12 @@ static void journal_write_done(struct closure *cl)
 	journal_wake(j);
 
 	if (test_bit(JOURNAL_NEED_WRITE, &j->flags))
-		mod_delayed_work(system_freezable_wq, &j->write_work, 0);
+		mod_delayed_work(c->io_complete_wq, &j->write_work, 0);
 	spin_unlock(&j->lock);
 
 	if (new.unwritten_idx != new.idx &&
 	    !journal_state_count(new, new.unwritten_idx))
-		closure_call(&j->io, bch2_journal_write, system_highpri_wq, NULL);
+		closure_call(&j->io, bch2_journal_write, c->io_complete_wq, NULL);
 }
 
 static void journal_write_endio(struct bio *bio)
@@ -1367,7 +1367,7 @@ static void do_journal_write(struct closure *cl)
 			le64_to_cpu(w->data->seq);
 	}
 
-	continue_at(cl, journal_write_done, system_highpri_wq);
+	continue_at(cl, journal_write_done, c->io_complete_wq);
 	return;
 }
 
@@ -1506,7 +1506,7 @@ void bch2_journal_write(struct closure *cl)
 			journal_debug_buf);
 		kfree(journal_debug_buf);
 		bch2_fatal_error(c);
-		continue_at(cl, journal_write_done, system_highpri_wq);
+		continue_at(cl, journal_write_done, c->io_complete_wq);
 		return;
 	}
 
@@ -1537,14 +1537,14 @@ void bch2_journal_write(struct closure *cl)
 
 	bch2_bucket_seq_cleanup(c);
 
-	continue_at(cl, do_journal_write, system_highpri_wq);
+	continue_at(cl, do_journal_write, c->io_complete_wq);
 	return;
 no_io:
 	bch2_bucket_seq_cleanup(c);
 
-	continue_at(cl, journal_write_done, system_highpri_wq);
+	continue_at(cl, journal_write_done, c->io_complete_wq);
 	return;
 err:
 	bch2_inconsistent_error(c);
-	continue_at(cl, journal_write_done, system_highpri_wq);
+	continue_at(cl, journal_write_done, c->io_complete_wq);
 }

fs/bcachefs/super.c

@@ -510,10 +510,12 @@ static void __bch2_fs_free(struct bch_fs *c)
 	kfree(c->unused_inode_hints);
 	free_heap(&c->copygc_heap);
 
+	if (c->io_complete_wq )
+		destroy_workqueue(c->io_complete_wq );
 	if (c->copygc_wq)
 		destroy_workqueue(c->copygc_wq);
-	if (c->wq)
-		destroy_workqueue(c->wq);
+	if (c->btree_update_wq)
+		destroy_workqueue(c->btree_update_wq);
 
 	bch2_free_super(&c->disk_sb);
 	kvpfree(c, sizeof(*c));
@@ -762,10 +764,12 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
 
 	c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
 
-	if (!(c->wq = alloc_workqueue("bcachefs",
+	if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
 				WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
 				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
+	    !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
+				WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 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) ||