bcachefs: fix check for if extent update is allocating

5 files changed, 83 insertions, 21 deletions

diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index d7db4ce16f9f..abc920b856cc 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -207,13 +207,14 @@ static bool bucket_became_unavailable(struct bucket_mark old,
 	       !is_available_bucket(new);
 }
 
-void bch2_fs_usage_apply(struct bch_fs *c,
-			 struct bch_fs_usage *fs_usage,
-			 struct disk_reservation *disk_res,
-			 struct gc_pos gc_pos)
+int bch2_fs_usage_apply(struct bch_fs *c,
+			struct bch_fs_usage *fs_usage,
+			struct disk_reservation *disk_res,
+			struct gc_pos gc_pos)
 {
 	s64 added = fs_usage->s.data + fs_usage->s.reserved;
 	s64 should_not_have_added;
+	int ret = 0;
 
 	percpu_rwsem_assert_held(&c->mark_lock);
 
@@ -226,6 +227,7 @@ void bch2_fs_usage_apply(struct bch_fs *c,
 		      "disk usage increased without a reservation")) {
 		atomic64_sub(should_not_have_added, &c->sectors_available);
 		added -= should_not_have_added;
+		ret = -1;
 	}
 
 	if (added > 0) {
@@ -245,6 +247,8 @@ void bch2_fs_usage_apply(struct bch_fs *c,
 	}
 
 	memset(fs_usage, 0, sizeof(*fs_usage));
+
+	return ret;
 }
 
 static inline void account_bucket(struct bch_fs_usage *fs_usage,
@@ -833,6 +837,8 @@ void bch2_mark_update(struct btree_insert *trans,
 	struct bch_fs_usage	fs_usage = { 0 };
 	struct gc_pos		pos = gc_pos_btree_node(b);
 	struct bkey_packed	*_k;
+	u64 disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
+	static int warned_disk_usage = 0;
 
 	if (!btree_node_type_needs_gc(iter->btree_id))
 		return;
@@ -892,7 +898,37 @@ void bch2_mark_update(struct btree_insert *trans,
 		bch2_btree_node_iter_advance(&node_iter, b);
 	}
 
-	bch2_fs_usage_apply(c, &fs_usage, trans->disk_res, pos);
+	if (bch2_fs_usage_apply(c, &fs_usage, trans->disk_res, pos) &&
+	    !warned_disk_usage &&
+	    !xchg(&warned_disk_usage, 1)) {
+		char buf[200];
+
+		pr_err("disk usage increased more than %llu sectors reserved", disk_res_sectors);
+
+		pr_err("while inserting");
+		bch2_bkey_val_to_text(&PBUF(buf), c, bkey_i_to_s_c(insert->k));
+		pr_err("%s", buf);
+		pr_err("overlapping with");
+
+		node_iter = iter->l[0].iter;
+		while ((_k = bch2_btree_node_iter_peek_filter(&node_iter, b,
+							      KEY_TYPE_discard))) {
+			struct bkey		unpacked;
+			struct bkey_s_c		k;
+
+			k = bkey_disassemble(b, _k, &unpacked);
+
+			if (btree_node_is_extents(b)
+			    ? bkey_cmp(insert->k->k.p, bkey_start_pos(k.k)) <= 0
+			    : bkey_cmp(insert->k->k.p, k.k->p))
+				break;
+
+			bch2_bkey_val_to_text(&PBUF(buf), c, k);
+			pr_err("%s", buf);
+
+			bch2_btree_node_iter_advance(&node_iter, b);
+		}
+	}
 
 	percpu_up_read_preempt_enable(&c->mark_lock);
 }
diff --git a/fs/bcachefs/buckets.h b/fs/bcachefs/buckets.h
index bc9180ffd5f5..10cacf36d238 100644
--- a/fs/bcachefs/buckets.h
+++ b/fs/bcachefs/buckets.h
@@ -253,8 +253,8 @@ int bch2_mark_key(struct bch_fs *, struct bkey_s_c,
 		  bool, s64, struct gc_pos,
 		  struct bch_fs_usage *, u64, unsigned);
 void bch2_mark_update(struct btree_insert *, struct btree_insert_entry *);
-void bch2_fs_usage_apply(struct bch_fs *, struct bch_fs_usage *,
-			 struct disk_reservation *, struct gc_pos);
+int bch2_fs_usage_apply(struct bch_fs *, struct bch_fs_usage *,
+			struct disk_reservation *, struct gc_pos);
 
 /* disk reservations: */
 
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 2980416871d8..0f075fa1d360 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -1664,12 +1664,13 @@ static bool bch2_extent_merge_inline(struct bch_fs *c,
 	return ret == BCH_MERGE_MERGE;
 }
 
-int bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size)
+bool bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size,
+			       unsigned nr_replicas)
 {
 	struct btree_iter iter;
 	struct bpos end = pos;
 	struct bkey_s_c k;
-	int ret = 0;
+	bool ret = true;
 
 	end.offset += size;
 
@@ -1678,8 +1679,8 @@ int bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size)
 		if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 			break;
 
-		if (!bch2_extent_is_fully_allocated(k)) {
-			ret = -ENOSPC;
+		if (nr_replicas > bch2_bkey_nr_ptrs_allocated(k)) {
+			ret = false;
 			break;
 		}
 	}
@@ -1688,6 +1689,29 @@ int bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size)
 	return ret;
 }
 
+unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c k)
+{
+	unsigned ret = 0;
+
+	switch (k.k->type) {
+	case KEY_TYPE_extent: {
+		struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
+		const union bch_extent_entry *entry;
+		struct extent_ptr_decoded p;
+
+		extent_for_each_ptr_decode(e, p, entry)
+			ret += !p.ptr.cached &&
+				p.crc.compression_type == BCH_COMPRESSION_NONE;
+		break;
+	}
+	case KEY_TYPE_reservation:
+		ret = bkey_s_c_to_reservation(k).v->nr_replicas;
+		break;
+	}
+
+	return ret;
+}
+
 /* KEY_TYPE_reservation: */
 
 const char *bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k)
diff --git a/fs/bcachefs/extents.h b/fs/bcachefs/extents.h
index 0e6f4a0bbcab..698b25818afb 100644
--- a/fs/bcachefs/extents.h
+++ b/fs/bcachefs/extents.h
@@ -571,6 +571,7 @@ static inline void extent_save(struct btree *b, struct bkey_packed *dst,
 		BUG_ON(!bch2_bkey_pack_key(dst, src, f));
 }
 
-int bch2_check_range_allocated(struct bch_fs *, struct bpos, u64);
+bool bch2_check_range_allocated(struct bch_fs *, struct bpos, u64, unsigned);
+unsigned bch2_bkey_nr_ptrs_allocated(struct bkey_s_c);
 
 #endif /* _BCACHEFS_EXTENTS_H */
diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 64e36b1dc510..f1fb17bff877 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -253,7 +253,9 @@ static s64 sum_sector_overwrites(struct bkey_i *new, struct btree_iter *_iter,
 		BUG_ON(btree_iter_err(old));
 
 		if (allocating &&
-		    !bch2_extent_is_fully_allocated(old))
+		    !*allocating &&
+		    bch2_bkey_nr_ptrs_allocated(old) <
+		    bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(new)))
 			*allocating = true;
 
 		delta += (min(new->k.p.offset,
@@ -858,9 +860,7 @@ static void bch2_add_page_sectors(struct bio *bio, struct bkey_s_c k)
 {
 	struct bvec_iter iter;
 	struct bio_vec bv;
-	unsigned nr_ptrs = !bch2_extent_is_compressed(k)
-		? bch2_bkey_nr_dirty_ptrs(k)
-		: 0;
+	unsigned nr_ptrs = bch2_bkey_nr_ptrs_allocated(k);
 
 	bio_for_each_segment(bv, bio, iter) {
 		/* brand new pages, don't need to be locked: */
@@ -1918,19 +1918,20 @@ static int bch2_direct_IO_write(struct kiocb *req,
 	if (unlikely(ret))
 		goto err;
 
+	dio->iop.op.nr_replicas	= dio->iop.op.opts.data_replicas;
+
 	ret = bch2_disk_reservation_get(c, &dio->iop.op.res, iter->count >> 9,
 					dio->iop.op.opts.data_replicas, 0);
 	if (unlikely(ret)) {
-		if (bch2_check_range_allocated(c, POS(inode->v.i_ino,
-						      req->ki_pos >> 9),
-					       iter->count >> 9))
+		if (!bch2_check_range_allocated(c, POS(inode->v.i_ino,
+						       req->ki_pos >> 9),
+						iter->count >> 9,
+						dio->iop.op.opts.data_replicas))
 			goto err;
 
 		dio->iop.unalloc = true;
 	}
 
-	dio->iop.op.nr_replicas	= dio->iop.op.res.nr_replicas;
-
 	return bch2_dio_write_loop(dio);
 err:
 	bch2_disk_reservation_put(c, &dio->iop.op.res);