X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fsubvolume.c;h=ce7aed12194238071f8fbf37aa111160ced286c9;hb=d1900b637ee57c2e523e308cc335deee0164b882;hp=0214a98deb4ff2a566eaea9915457182af04bd34;hpb=63c8f14756921c1d1d6a99082a679b92aef288c1;p=bcachefs-tools-debian

diff --git a/libbcachefs/subvolume.c b/libbcachefs/subvolume.c
index 0214a98..ce7aed1 100644
--- a/libbcachefs/subvolume.c
+++ b/libbcachefs/subvolume.c
@@ -13,13 +13,26 @@
 
 static int bch2_subvolume_delete(struct btree_trans *, u32);
 
+static struct bpos subvolume_children_pos(struct bkey_s_c k)
+{
+	if (k.k->type != KEY_TYPE_subvolume)
+		return POS_MIN;
+
+	struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
+	if (!s.v->fs_path_parent)
+		return POS_MIN;
+	return POS(le32_to_cpu(s.v->fs_path_parent), s.k->p.offset);
+}
+
 static int check_subvol(struct btree_trans *trans,
 			struct btree_iter *iter,
 			struct bkey_s_c k)
 {
 	struct bch_fs *c = trans->c;
 	struct bkey_s_c_subvolume subvol;
+	struct btree_iter subvol_children_iter = {};
 	struct bch_snapshot snapshot;
+	struct printbuf buf = PRINTBUF;
 	unsigned snapid;
 	int ret = 0;
 
@@ -37,15 +50,77 @@ static int check_subvol(struct btree_trans *trans,
 		return ret;
 
 	if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
-		bch2_fs_lazy_rw(c);
-
 		ret = bch2_subvolume_delete(trans, iter->pos.offset);
+		bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
+		return ret ?: -BCH_ERR_transaction_restart_nested;
+	}
+
+	if (fsck_err_on(subvol.k->p.offset == BCACHEFS_ROOT_SUBVOL &&
+			subvol.v->fs_path_parent,
+			c, subvol_root_fs_path_parent_nonzero,
+			"root subvolume has nonzero fs_path_parent\n%s",
+			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+		struct bkey_i_subvolume *n =
+			bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
+		ret = PTR_ERR_OR_ZERO(n);
 		if (ret)
-			bch_err(c, "error deleting subvolume %llu: %s",
-				iter->pos.offset, bch2_err_str(ret));
+			goto err;
+
+		n->v.fs_path_parent = 0;
+	}
+
+	if (subvol.v->fs_path_parent) {
+		struct bpos pos = subvolume_children_pos(k);
+
+		struct bkey_s_c subvol_children_k =
+			bch2_bkey_get_iter(trans, &subvol_children_iter,
+					   BTREE_ID_subvolume_children, pos, 0);
+		ret = bkey_err(subvol_children_k);
+		if (ret)
+			goto err;
+
+		if (fsck_err_on(subvol_children_k.k->type != KEY_TYPE_set,
+				c, subvol_children_not_set,
+				"subvolume not set in subvolume_children btree at %llu:%llu\n%s",
+				pos.inode, pos.offset,
+				(printbuf_reset(&buf),
+				 bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+			ret = bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, true);
+			if (ret)
+				goto err;
+		}
+	}
+
+	struct bch_inode_unpacked inode;
+	struct btree_iter inode_iter = {};
+	ret = bch2_inode_peek_nowarn(trans, &inode_iter, &inode,
+				    (subvol_inum) { k.k->p.offset, le64_to_cpu(subvol.v->inode) },
+				    0);
+	bch2_trans_iter_exit(trans, &inode_iter);
+
+	if (ret && !bch2_err_matches(ret, ENOENT))
+		return ret;
+
+	if (fsck_err_on(ret, c, subvol_to_missing_root,
+			"subvolume %llu points to missing subvolume root %llu:%u",
+			k.k->p.offset, le64_to_cpu(subvol.v->inode),
+			le32_to_cpu(subvol.v->snapshot))) {
+		ret = bch2_subvolume_delete(trans, iter->pos.offset);
+		bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
 		return ret ?: -BCH_ERR_transaction_restart_nested;
 	}
 
+	if (fsck_err_on(inode.bi_subvol != subvol.k->p.offset,
+			c, subvol_root_wrong_bi_subvol,
+			"subvol root %llu:%u has wrong bi_subvol field: got %u, should be %llu",
+			inode.bi_inum, inode_iter.k.p.snapshot,
+			inode.bi_subvol, subvol.k->p.offset)) {
+		inode.bi_subvol = subvol.k->p.offset;
+		ret = __bch2_fsck_write_inode(trans, &inode, le32_to_cpu(subvol.v->snapshot));
+		if (ret)
+			goto err;
+	}
+
 	if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
 		u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
 		u32 snapshot_tree;
@@ -63,7 +138,8 @@ static int check_subvol(struct btree_trans *trans,
 		if (ret)
 			return ret;
 
-		if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset, c,
+		if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset,
+				c, subvol_not_master_and_not_snapshot,
 				"subvolume %llu is not set as snapshot but is not master subvolume",
 				k.k->p.offset)) {
 			struct bkey_i_subvolume *s =
@@ -75,39 +151,73 @@ static int check_subvol(struct btree_trans *trans,
 			SET_BCH_SUBVOLUME_SNAP(&s->v, true);
 		}
 	}
-
+err:
 fsck_err:
+	bch2_trans_iter_exit(trans, &subvol_children_iter);
+	printbuf_exit(&buf);
 	return ret;
 }
 
 int bch2_check_subvols(struct bch_fs *c)
 {
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	int ret;
+	int ret = bch2_trans_run(c,
+		for_each_btree_key_commit(trans, iter,
+				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+			check_subvol(trans, &iter, k)));
+	bch_err_fn(c, ret);
+	return ret;
+}
 
-	ret = bch2_trans_run(c,
-		for_each_btree_key_commit(&trans, iter,
-			BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
-			NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
-		check_subvol(&trans, &iter, k)));
-	if (ret)
-		bch_err_fn(c, ret);
+static int check_subvol_child(struct btree_trans *trans,
+			      struct btree_iter *child_iter,
+			      struct bkey_s_c child_k)
+{
+	struct bch_fs *c = trans->c;
+	struct bch_subvolume s;
+	int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, child_k.k->p.offset),
+					  0, subvolume, &s);
+	if (ret && !bch2_err_matches(ret, ENOENT))
+		return ret;
+
+	if (fsck_err_on(ret ||
+			le32_to_cpu(s.fs_path_parent) != child_k.k->p.inode,
+			c, subvol_children_bad,
+			"incorrect entry in subvolume_children btree %llu:%llu",
+			child_k.k->p.inode, child_k.k->p.offset)) {
+		ret = bch2_btree_delete_at(trans, child_iter, 0);
+		if (ret)
+			goto err;
+	}
+err:
+fsck_err:
 	return ret;
 }
 
+int bch2_check_subvol_children(struct bch_fs *c)
+{
+	int ret = bch2_trans_run(c,
+		for_each_btree_key_commit(trans, iter,
+				BTREE_ID_subvolume_children, POS_MIN, BTREE_ITER_PREFETCH, k,
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+			check_subvol_child(trans, &iter, k)));
+	bch_err_fn(c, ret);
+	return 0;
+}
+
 /* Subvolumes: */
 
-int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
-			   unsigned flags, struct printbuf *err)
+int bch2_subvolume_invalid(struct bch_fs *c, struct bkey_s_c k,
+			   enum bkey_invalid_flags flags, struct printbuf *err)
 {
-	if (bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
-	    bkey_gt(k.k->p, SUBVOL_POS_MAX)) {
-		prt_printf(err, "invalid pos");
-		return -BCH_ERR_invalid_bkey;
-	}
+	int ret = 0;
 
-	return 0;
+	bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
+			 bkey_gt(k.k->p, SUBVOL_POS_MAX), c, err,
+			 subvol_pos_bad,
+			 "invalid pos");
+fsck_err:
+	return ret;
 }
 
 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
@@ -119,8 +229,50 @@ void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
 		   le64_to_cpu(s.v->inode),
 		   le32_to_cpu(s.v->snapshot));
 
-	if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, parent))
-		prt_printf(out, " parent %u", le32_to_cpu(s.v->parent));
+	if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, creation_parent)) {
+		prt_printf(out, " creation_parent %u", le32_to_cpu(s.v->creation_parent));
+		prt_printf(out, " fs_parent %u", le32_to_cpu(s.v->fs_path_parent));
+	}
+}
+
+static int subvolume_children_mod(struct btree_trans *trans, struct bpos pos, bool set)
+{
+	return !bpos_eq(pos, POS_MIN)
+		? bch2_btree_bit_mod(trans, BTREE_ID_subvolume_children, pos, set)
+		: 0;
+}
+
+int bch2_subvolume_trigger(struct btree_trans *trans,
+			   enum btree_id btree_id, unsigned level,
+			   struct bkey_s_c old, struct bkey_s new,
+			   unsigned flags)
+{
+	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
+		struct bpos children_pos_old = subvolume_children_pos(old);
+		struct bpos children_pos_new = subvolume_children_pos(new.s_c);
+
+		if (!bpos_eq(children_pos_old, children_pos_new)) {
+			int ret = subvolume_children_mod(trans, children_pos_old, false) ?:
+				  subvolume_children_mod(trans, children_pos_new, true);
+			if (ret)
+				return ret;
+		}
+	}
+
+	return 0;
+}
+
+int bch2_subvol_has_children(struct btree_trans *trans, u32 subvol)
+{
+	struct btree_iter iter;
+
+	bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolume_children, POS(subvol, 0), 0);
+	struct bkey_s_c k = bch2_btree_iter_peek(&iter);
+	bch2_trans_iter_exit(trans, &iter);
+
+	return bkey_err(k) ?: k.k && k.k->p.inode == subvol
+		? -BCH_ERR_ENOTEMPTY_subvol_not_empty
+		: 0;
 }
 
 static __always_inline int
@@ -145,6 +297,24 @@ int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
 	return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
 }
 
+int bch2_subvol_is_ro_trans(struct btree_trans *trans, u32 subvol)
+{
+	struct bch_subvolume s;
+	int ret = bch2_subvolume_get_inlined(trans, subvol, true, 0, &s);
+	if (ret)
+		return ret;
+
+	if (BCH_SUBVOLUME_RO(&s))
+		return -EROFS;
+	return 0;
+}
+
+int bch2_subvol_is_ro(struct bch_fs *c, u32 subvol)
+{
+	return bch2_trans_do(c, NULL, NULL, 0,
+		bch2_subvol_is_ro_trans(trans, subvol));
+}
+
 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
 			     struct bch_subvolume *subvol)
 {
@@ -186,8 +356,8 @@ static int bch2_subvolume_reparent(struct btree_trans *trans,
 	if (k.k->type != KEY_TYPE_subvolume)
 		return 0;
 
-	if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, parent) &&
-	    le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent)
+	if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, creation_parent) &&
+	    le32_to_cpu(bkey_s_c_to_subvolume(k).v->creation_parent) != old_parent)
 		return 0;
 
 	s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
@@ -195,7 +365,7 @@ static int bch2_subvolume_reparent(struct btree_trans *trans,
 	if (ret)
 		return ret;
 
-	s->v.parent = cpu_to_le32(new_parent);
+	s->v.creation_parent = cpu_to_le32(new_parent);
 	return 0;
 }
 
@@ -209,8 +379,6 @@ static int bch2_subvolume_reparent(struct btree_trans *trans,
  */
 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
 {
-	struct btree_iter iter;
-	struct bkey_s_c k;
 	struct bch_subvolume s;
 
 	return lockrestart_do(trans,
@@ -218,9 +386,9 @@ static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_d
 				   BTREE_ITER_CACHED, &s)) ?:
 		for_each_btree_key_commit(trans, iter,
 				BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
-				NULL, NULL, BTREE_INSERT_NOFAIL,
+				NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 			bch2_subvolume_reparent(trans, &iter, k,
-					subvolid_to_delete, le32_to_cpu(s.parent)));
+					subvolid_to_delete, le32_to_cpu(s.creation_parent)));
 }
 
 /*
@@ -231,7 +399,6 @@ static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 {
 	struct btree_iter iter;
 	struct bkey_s_c_subvolume subvol;
-	struct btree_trans_commit_hook *h;
 	u32 snapid;
 	int ret = 0;
 
@@ -247,22 +414,8 @@ static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 
 	snapid = le32_to_cpu(subvol.v->snapshot);
 
-	ret = bch2_btree_delete_at(trans, &iter, 0);
-	if (ret)
-		goto err;
-
-	ret = bch2_snapshot_node_set_deleted(trans, snapid);
-	if (ret)
-		goto err;
-
-	h = bch2_trans_kmalloc(trans, sizeof(*h));
-	ret = PTR_ERR_OR_ZERO(h);
-	if (ret)
-		goto err;
-
-	h->fn = bch2_delete_dead_snapshots_hook;
-	bch2_trans_commit_hook(trans, h);
-err:
+	ret =   bch2_btree_delete_at(trans, &iter, 0) ?:
+		bch2_snapshot_node_set_deleted(trans, snapid);
 	bch2_trans_iter_exit(trans, &iter);
 	return ret;
 }
@@ -270,7 +423,7 @@ err:
 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
 {
 	return bch2_subvolumes_reparent(trans, subvolid) ?:
-		commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
+		commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
 			  __bch2_subvolume_delete(trans, subvolid));
 }
 
@@ -294,11 +447,10 @@ static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *wor
 		bch2_evict_subvolume_inodes(c, &s);
 
 		for (id = s.data; id < s.data + s.nr; id++) {
-			ret = bch2_trans_run(c, bch2_subvolume_delete(&trans, *id));
-			if (ret) {
-				bch_err(c, "error deleting subvolume %u: %s", *id, bch2_err_str(ret));
+			ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id));
+			bch_err_msg(c, ret, "deleting subvolume %u", *id);
+			if (ret)
 				break;
-			}
 		}
 
 		darray_exit(&s);
@@ -367,6 +519,7 @@ int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
 }
 
 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
+			  u32 parent_subvolid,
 			  u32 src_subvolid,
 			  u32 *new_subvolid,
 			  u32 *new_snapshotid,
@@ -423,12 +576,13 @@ int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
 	if (ret)
 		goto err;
 
-	new_subvol->v.flags	= 0;
-	new_subvol->v.snapshot	= cpu_to_le32(new_nodes[0]);
-	new_subvol->v.inode	= cpu_to_le64(inode);
-	new_subvol->v.parent	= cpu_to_le32(src_subvolid);
-	new_subvol->v.otime.lo	= cpu_to_le64(bch2_current_time(c));
-	new_subvol->v.otime.hi	= 0;
+	new_subvol->v.flags		= 0;
+	new_subvol->v.snapshot		= cpu_to_le32(new_nodes[0]);
+	new_subvol->v.inode		= cpu_to_le64(inode);
+	new_subvol->v.creation_parent	= cpu_to_le32(src_subvolid);
+	new_subvol->v.fs_path_parent	= cpu_to_le32(parent_subvolid);
+	new_subvol->v.otime.lo		= cpu_to_le64(bch2_current_time(c));
+	new_subvol->v.otime.hi		= 0;
 
 	SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
 	SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);