bcachefs: check_allocations_done_lock

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

10 files changed, 65 insertions, 10 deletions

diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index 494967290e72..7670a2302f9c 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -1023,6 +1023,8 @@ struct bch_fs {
 	 */
 	seqcount_t		gc_pos_lock;
 	struct gc_pos		gc_pos;
+	struct percpu_rw_semaphore
+				check_allocations_done_lock;
 
 	struct mutex		gc_gens_lock;
 
diff --git a/fs/bcachefs/btree_gc.c b/fs/bcachefs/btree_gc.c
index 64259f3b1988..411fc28a862c 100644
--- a/fs/bcachefs/btree_gc.c
+++ b/fs/bcachefs/btree_gc.c
@@ -973,7 +973,9 @@ static int bch2_gc_alloc_done(struct bch_fs *c)
 					POS(ca->dev_idx, ca->mi.first_bucket),
 					POS(ca->dev_idx, ca->mi.nbuckets - 1),
 					BTREE_ITER_slots|BTREE_ITER_prefetch, k,
-					NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+					NULL, NULL,
+					BCH_TRANS_COMMIT_no_enospc|
+					BCH_TRANS_COMMIT_check_allocations_lock_held,
 				bch2_alloc_write_key(trans, &iter, ca, k)));
 		if (ret) {
 			bch2_dev_put(ca);
@@ -1062,7 +1064,9 @@ static int bch2_gc_stripes_done(struct bch_fs *c)
 		for_each_btree_key_commit(trans, iter,
 				BTREE_ID_stripes, POS_MIN,
 				BTREE_ITER_prefetch, k,
-				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+				NULL, NULL,
+				BCH_TRANS_COMMIT_no_enospc|
+				BCH_TRANS_COMMIT_check_allocations_lock_held,
 			bch2_gc_write_stripes_key(trans, &iter, k)));
 }
 
@@ -1114,10 +1118,14 @@ int bch2_check_allocations(struct bch_fs *c)
 
 	c->gc_count++;
 
+	percpu_down_write(&c->check_allocations_done_lock);
+	bch2_btree_interior_updates_flush(c);
+
 	ret   = bch2_gc_alloc_done(c) ?:
 		bch2_gc_accounting_done(c) ?:
 		bch2_gc_stripes_done(c) ?:
 		bch2_gc_reflink_done(c);
+	percpu_up_write(&c->check_allocations_done_lock);
 out:
 	percpu_down_write(&c->mark_lock);
 	/* Indicates that gc is no longer in progress: */
@@ -1306,10 +1314,24 @@ void bch2_gc_gens_async(struct bch_fs *c)
 		enumerated_ref_put(&c->writes, BCH_WRITE_REF_gc_gens);
 }
 
+void bch2_fs_btree_gc_exit(struct bch_fs *c)
+{
+	percpu_free_rwsem(&c->check_allocations_done_lock);
+}
+
 void bch2_fs_btree_gc_init_early(struct bch_fs *c)
 {
 	seqcount_init(&c->gc_pos_lock);
 	INIT_WORK(&c->gc_gens_work, bch2_gc_gens_work);
 
+	int ret = percpu_init_rwsem(&c->check_allocations_done_lock);
+	if (ret)
+		return ret;
+
 	mutex_init(&c->gc_gens_lock);
 }
+
+int bch2_fs_btree_gc_init(struct bch_fs *c)
+{
+	return percpu_init_rwsem(&c->check_allocations_done_lock);
+}
diff --git a/fs/bcachefs/btree_gc.h b/fs/bcachefs/btree_gc.h
index 109ac5c67961..40b22052f844 100644
--- a/fs/bcachefs/btree_gc.h
+++ b/fs/bcachefs/btree_gc.h
@@ -84,6 +84,8 @@ void bch2_gc_pos_to_text(struct printbuf *, struct gc_pos *);
 int bch2_gc_gens(struct bch_fs *);
 void bch2_gc_gens_async(struct bch_fs *);
 
+void bch2_fs_btree_gc_exit(struct bch_fs *);
 void bch2_fs_btree_gc_init_early(struct bch_fs *);
+int bch2_fs_btree_gc_init(struct bch_fs *);
 
 #endif /* _BCACHEFS_BTREE_GC_H */
diff --git a/fs/bcachefs/btree_io.c b/fs/bcachefs/btree_io.c
index 28b53310912c..d189daee6d64 100644
--- a/fs/bcachefs/btree_io.c
+++ b/fs/bcachefs/btree_io.c
@@ -2146,7 +2146,8 @@ static void btree_node_write_work(struct work_struct *work)
 		BCH_WATERMARK_interior_updates|
 		BCH_TRANS_COMMIT_journal_reclaim|
 		BCH_TRANS_COMMIT_no_enospc|
-		BCH_TRANS_COMMIT_no_check_rw;
+		BCH_TRANS_COMMIT_no_check_rw|
+		BCH_TRANS_COMMIT_check_allocations_lock_held;
 	u64 start_time		= wbio->start_time;
 	int ret = 0;
 
diff --git a/fs/bcachefs/btree_trans_commit.c b/fs/bcachefs/btree_trans_commit.c
index cdde769e7da3..885fb030c85e 100644
--- a/fs/bcachefs/btree_trans_commit.c
+++ b/fs/bcachefs/btree_trans_commit.c
@@ -990,9 +990,18 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 	    !trans->journal_entries_u64s)
 		goto out_reset;
 
+	if (likely(!(flags & BCH_TRANS_COMMIT_check_allocations_lock_held))) {
+		if (unlikely(!percpu_down_read_trylock(&c->check_allocations_done_lock))) {
+			ret = drop_locks_do(trans,
+				(percpu_down_read(&c->check_allocations_done_lock), 0));
+			if (ret)
+				goto out_reset_unlock;
+		}
+	}
+
 	ret = bch2_trans_commit_run_triggers(trans);
 	if (ret)
-		goto out_reset;
+		goto out_reset_unlock;
 
 	if (!(flags & BCH_TRANS_COMMIT_no_check_rw) &&
 	    unlikely(!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_trans))) {
@@ -1000,7 +1009,7 @@ int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
 			ret = do_bch2_trans_commit_to_journal_replay(trans);
 		else
 			ret = -BCH_ERR_erofs_trans_commit;
-		goto out_reset;
+		goto out_reset_unlock;
 	}
 
 	EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags));
@@ -1062,6 +1071,9 @@ retry:
 out:
 	if (likely(!(flags & BCH_TRANS_COMMIT_no_check_rw)))
 		enumerated_ref_put(&c->writes, BCH_WRITE_REF_trans);
+out_reset_unlock:
+	if (likely(!(flags & BCH_TRANS_COMMIT_check_allocations_lock_held)))
+		percpu_up_read(&c->check_allocations_done_lock);
 out_reset:
 	if (!ret)
 		bch2_trans_downgrade(trans);
diff --git a/fs/bcachefs/btree_update.h b/fs/bcachefs/btree_update.h
index 62d24b081e27..f440136b631b 100644
--- a/fs/bcachefs/btree_update.h
+++ b/fs/bcachefs/btree_update.h
@@ -28,7 +28,8 @@ void bch2_btree_insert_key_leaf(struct btree_trans *, struct btree_path *,
 			"pin journal entry referred to by trans->journal_res.seq")	\
 	x(journal_reclaim, "operation required for journal reclaim; may return error"	\
 			"instead of deadlocking if BCH_WATERMARK_reclaim not specified")\
-	x(skip_accounting_apply, "we're in journal replay - accounting updates have already been applied")
+	x(skip_accounting_apply, "we're in journal replay - accounting updates have already been applied")\
+	x(check_allocations_lock_held, "check_allocations_lock_held")
 
 enum __bch_trans_commit_flags {
 	/* First bits for bch_watermark: */
diff --git a/fs/bcachefs/btree_update_interior.c b/fs/bcachefs/btree_update_interior.c
index f8a2d0b6ded2..76d20e06d736 100644
--- a/fs/bcachefs/btree_update_interior.c
+++ b/fs/bcachefs/btree_update_interior.c
@@ -741,7 +741,8 @@ static void btree_update_nodes_written(struct btree_update *as)
 			BCH_WATERMARK_interior_updates|
 			BCH_TRANS_COMMIT_no_enospc|
 			BCH_TRANS_COMMIT_no_check_rw|
-			BCH_TRANS_COMMIT_journal_reclaim,
+			BCH_TRANS_COMMIT_journal_reclaim|
+			BCH_TRANS_COMMIT_check_allocations_lock_held,
 			btree_update_nodes_written_trans(trans, as));
 	bch2_trans_unlock(trans);
 
@@ -2220,14 +2221,22 @@ int bch2_btree_node_rewrite(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct btree *parent;
 	btree_path_idx_t new_path = 0;
+	int ret = 0;
 
 	flags |= BCH_TRANS_COMMIT_no_enospc;
 
+	if (unlikely(!percpu_down_read_trylock(&c->check_allocations_done_lock))) {
+		ret = drop_locks_do(trans,
+				    (percpu_down_read(&c->check_allocations_done_lock), 0));
+		if (ret)
+			goto out;
+	}
+
 	struct btree_path *path = btree_iter_path(trans, iter);
 	parent = btree_node_parent(path, b);
 	struct btree_update *as =
 		bch2_btree_update_start(trans, path, b->c.level, false, target, flags);
-	int ret = PTR_ERR_OR_ZERO(as);
+	ret = PTR_ERR_OR_ZERO(as);
 	if (ret)
 		goto out;
 
@@ -2272,6 +2281,7 @@ out:
 	if (new_path)
 		bch2_path_put(trans, new_path, true);
 	bch2_trans_downgrade(trans);
+	percpu_up_read(&c->check_allocations_done_lock);
 	return ret;
 err_free_node:
 	bch2_btree_node_free_never_used(as, trans, n);
diff --git a/fs/bcachefs/disk_accounting.c b/fs/bcachefs/disk_accounting.c
index 7be71952425c..784a9b298350 100644
--- a/fs/bcachefs/disk_accounting.c
+++ b/fs/bcachefs/disk_accounting.c
@@ -570,7 +570,9 @@ int bch2_gc_accounting_done(struct bch_fs *c)
 
 			if (fsck_err(trans, accounting_mismatch, "%s", buf.buf)) {
 				percpu_up_write(&c->mark_lock);
-				ret = commit_do(trans, NULL, NULL, 0,
+				ret = commit_do(trans, NULL, NULL,
+						BCH_TRANS_COMMIT_no_enospc|
+						BCH_TRANS_COMMIT_check_allocations_lock_held,
 						bch2_disk_accounting_mod(trans, &acc_k, src_v, nr, false));
 				percpu_down_write(&c->mark_lock);
 				if (ret)
diff --git a/fs/bcachefs/reflink.c b/fs/bcachefs/reflink.c
index 3a13dbcab6ba..29179bd36b97 100644
--- a/fs/bcachefs/reflink.c
+++ b/fs/bcachefs/reflink.c
@@ -826,7 +826,9 @@ int bch2_gc_reflink_done(struct bch_fs *c)
 		for_each_btree_key_commit(trans, iter,
 				BTREE_ID_reflink, POS_MIN,
 				BTREE_ITER_prefetch, k,
-				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+				NULL, NULL,
+				BCH_TRANS_COMMIT_no_enospc|
+				BCH_TRANS_COMMIT_check_allocations_lock_held,
 			bch2_gc_write_reflink_key(trans, &iter, k, &idx)));
 	c->reflink_gc_nr = 0;
 	return ret;
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 16a9632390dc..c8d40e88e850 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -971,6 +971,7 @@ static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts *opts,
 	ret =
 	    bch2_fs_async_obj_init(c) ?:
 	    bch2_fs_btree_cache_init(c) ?:
+	    bch2_fs_btree_gc_init(c) ?:
 	    bch2_fs_btree_iter_init(c) ?:
 	    bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
 	    bch2_fs_buckets_waiting_for_journal_init(c) ?: