bcachefs: fix initial page state after falloc

On a new page reservation check if the backing extent was already
allocated before adding to the number of sectors we will allocate.
It is possible to write to a page that is beyond the inode i_size
that is also backed by an allocated extent. This can happen when
fallocate is used with FALLOC_FL_KEEP_SIZE.

Signed-off-by: Dan Robertson <[email protected]>

fs/bcachefs/fs-io.c

@@ -387,23 +387,45 @@ static void bch2_page_reservation_put(struct bch_fs *c,
 static int bch2_page_reservation_get(struct bch_fs *c,
 			struct bch_inode_info *inode, struct page *page,
 			struct bch2_page_reservation *res,
-			unsigned offset, unsigned len, bool check_enospc)
+			loff_t raw_offset, unsigned len, bool check_enospc)
 {
 	struct bch_page_state *s = bch2_page_state_create(page, 0);
 	unsigned i, disk_sectors = 0, quota_sectors = 0;
 	int ret;
+	struct btree_iter *iter;
+	struct btree_trans trans;
+	struct bkey_s_c k;
+	unsigned offset = raw_offset & (PAGE_SIZE - 1);
 
 	if (!s)
 		return -ENOMEM;
 
+	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
+	iter = bch2_trans_get_iter(&trans, BTREE_ID_extents,
+			POS(inode->v.i_ino, raw_offset >> 9),
+			BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
+
 	for (i = round_down(offset, block_bytes(c)) >> 9;
 	     i < round_up(offset + len, block_bytes(c)) >> 9;
 	     i++) {
-		disk_sectors += sectors_to_reserve(&s->s[i],
-						res->disk.nr_replicas);
-		quota_sectors += s->s[i].state == SECTOR_UNALLOCATED;
+		k = bch2_btree_iter_peek_slot(iter);
+		ret = bkey_err(k);
+		if (ret) {
+			break;
+		} else if (bkey_extent_is_allocation(k.k)) {
+			s->s[i].state = SECTOR_ALLOCATED;
+		} else {
+			quota_sectors += s->s[i].state == SECTOR_UNALLOCATED;
+			disk_sectors += sectors_to_reserve(&s->s[i], res->disk.nr_replicas);
+		}
 	}
 
+	bch2_trans_iter_put(&trans, iter);
+	bch2_trans_exit(&trans);
+
+	if (ret)
+		return ret;
+
 	if (disk_sectors) {
 		ret = bch2_disk_reservation_add(c, &res->disk,
 						disk_sectors,
@@ -581,7 +603,8 @@ vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf)
 
 	len = min_t(loff_t, PAGE_SIZE, isize - page_offset(page));
 
-	if (bch2_page_reservation_get(c, inode, page, &res, 0, len, true)) {
+	if (bch2_page_reservation_get(c, inode, page, &res,
+				      page_offset(page), len, true)) {
 		unlock_page(page);
 		ret = VM_FAULT_SIGBUS;
 		goto out;
@@ -1338,7 +1361,7 @@ int bch2_write_begin(struct file *file, struct address_space *mapping,
 		goto err;
 out:
 	ret = bch2_page_reservation_get(c, inode, page, res,
-					offset, len, true);
+					pos, len, true);
 	if (ret) {
 		if (!PageUptodate(page)) {
 			/*
@@ -1472,7 +1495,7 @@ static int __bch2_buffered_write(struct bch_inode_info *inode,
 					PAGE_SIZE - pg_offset);
 retry_reservation:
 		ret = bch2_page_reservation_get(c, inode, page, &res,
-						pg_offset, pg_len, true);
+						pos + reserved, pg_len, true);
 
 		if (ret && !PageUptodate(page)) {
 			ret = bch2_read_single_page(page, mapping);