bcachefs: bch2_assert_pos_locked()

This adds a new assertion to be used by bch2_inode_update_after_write(),
which updates the VFS inode based on the update to the btree inode we
just did - we require that the btree inode still be locked when we do
that update.

Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>

7 files changed, 88 insertions, 31 deletions

diff --git a/fs/bcachefs/acl.c b/fs/bcachefs/acl.c
index 51a0b48a5313..00cd40a8d7fa 100644
--- a/fs/bcachefs/acl.c
+++ b/fs/bcachefs/acl.c
@@ -339,7 +339,7 @@ btree_err:
 	if (unlikely(ret))
 		goto err;
 
-	bch2_inode_update_after_write(c, inode, &inode_u,
+	bch2_inode_update_after_write(&trans, inode, &inode_u,
 				      ATTR_CTIME|ATTR_MODE);
 
 	set_cached_acl(&inode->v, type, acl);
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index 94ba43626cde..1ad81cad36f1 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -46,7 +46,7 @@ static inline int __btree_path_cmp(const struct btree_path *l,
 				   unsigned		r_level)
 {
 	return   cmp_int(l->btree_id,	r_btree_id) ?:
-		 cmp_int(l->cached,	r_cached) ?:
+		 cmp_int((int) l->cached,	(int) r_cached) ?:
 		 bpos_cmp(l->pos,	r_pos) ?:
 		-cmp_int(l->level,	r_level);
 }
@@ -762,6 +762,43 @@ out:
 	return ret;
 }
 
+void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
+			    struct bpos pos, bool key_cache)
+{
+	struct btree_path *path;
+	unsigned idx;
+	char buf[100];
+
+	trans_for_each_path_inorder(trans, path, idx) {
+		int cmp = cmp_int(path->btree_id, id) ?:
+			cmp_int(path->cached, key_cache);
+
+		if (cmp > 0)
+			break;
+		if (cmp < 0)
+			continue;
+
+		if (!(path->nodes_locked & 1) ||
+		    !path->should_be_locked)
+			continue;
+
+		if (!key_cache) {
+			if (bkey_cmp(pos, path->l[0].b->data->min_key) >= 0 &&
+			    bkey_cmp(pos, path->l[0].b->key.k.p) <= 0)
+				return;
+		} else {
+			if (!bkey_cmp(pos, path->pos))
+				return;
+		}
+	}
+
+	bch2_dump_trans_paths_updates(trans);
+	panic("not locked: %s %s%s\n",
+	      bch2_btree_ids[id],
+	      (bch2_bpos_to_text(&PBUF(buf), pos), buf),
+	      key_cache ? " cached" : "");
+}
+
 #else
 
 static inline void bch2_btree_path_verify_level(struct btree_trans *trans,
@@ -1720,11 +1757,13 @@ void bch2_dump_trans_paths_updates(struct btree_trans *trans)
 	btree_trans_sort_paths(trans);
 
 	trans_for_each_path_inorder(trans, path, idx)
-		printk(KERN_ERR "path: idx %u ref %u:%u%s btree %s pos %s %pS\n",
+		printk(KERN_ERR "path: idx %u ref %u:%u%s%s btree %s pos %s locks %u %pS\n",
 		       path->idx, path->ref, path->intent_ref,
-		       path->preserve ? " preserve" : "",
+		       path->should_be_locked ? " S" : "",
+		       path->preserve ? " P" : "",
 		       bch2_btree_ids[path->btree_id],
 		       (bch2_bpos_to_text(&PBUF(buf1), path->pos), buf1),
+		       path->nodes_locked,
 #ifdef CONFIG_BCACHEFS_DEBUG
 		       (void *) path->ip_allocated
 #else
diff --git a/fs/bcachefs/btree_iter.h b/fs/bcachefs/btree_iter.h
index c71e42a782d6..72b9605cf3e7 100644
--- a/fs/bcachefs/btree_iter.h
+++ b/fs/bcachefs/btree_iter.h
@@ -166,9 +166,13 @@ inline struct bkey_s_c bch2_btree_path_peek_slot(struct btree_path *, struct bke
 #ifdef CONFIG_BCACHEFS_DEBUG
 void bch2_trans_verify_paths(struct btree_trans *);
 void bch2_trans_verify_locks(struct btree_trans *);
+void bch2_assert_pos_locked(struct btree_trans *, enum btree_id,
+			    struct bpos, bool);
 #else
 static inline void bch2_trans_verify_paths(struct btree_trans *trans) {}
 static inline void bch2_trans_verify_locks(struct btree_trans *trans) {}
+static inline void bch2_assert_pos_locked(struct btree_trans *trans, enum btree_id id,
+					  struct bpos pos, bool key_cache) {}
 #endif
 
 void bch2_btree_path_fix_key_modified(struct btree_trans *trans,
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 92919b16f2f5..5596081b93c1 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -37,7 +37,7 @@
 
 static struct kmem_cache *bch2_inode_cache;
 
-static void bch2_vfs_inode_init(struct bch_fs *, subvol_inum,
+static void bch2_vfs_inode_init(struct btree_trans *, subvol_inum,
 				struct bch_inode_info *,
 				struct bch_inode_unpacked *);
 
@@ -93,11 +93,19 @@ void bch2_pagecache_block_get(struct pagecache_lock *lock)
 	__pagecache_lock_get(lock, -1);
 }
 
-void bch2_inode_update_after_write(struct bch_fs *c,
+void bch2_inode_update_after_write(struct btree_trans *trans,
 				   struct bch_inode_info *inode,
 				   struct bch_inode_unpacked *bi,
 				   unsigned fields)
 {
+	struct bch_fs *c = trans->c;
+
+	BUG_ON(bi->bi_inum != inode->v.i_ino);
+
+	bch2_assert_pos_locked(trans, BTREE_ID_inodes,
+			       POS(0, bi->bi_inum),
+			       0 && c->opts.inodes_use_key_cache);
+
 	set_nlink(&inode->v, bch2_inode_nlink_get(bi));
 	i_uid_write(&inode->v, bi->bi_uid);
 	i_gid_write(&inode->v, bi->bi_gid);
@@ -126,6 +134,7 @@ int __must_check bch2_write_inode(struct bch_fs *c,
 	int ret;
 
 	bch2_trans_init(&trans, c, 0, 512);
+	trans.ip = _RET_IP_;
 retry:
 	bch2_trans_begin(&trans);
 
@@ -140,7 +149,7 @@ retry:
 	 * this is important for inode updates via bchfs_write_index_update
 	 */
 	if (!ret)
-		bch2_inode_update_after_write(c, inode, &inode_u, fields);
+		bch2_inode_update_after_write(&trans, inode, &inode_u, fields);
 
 	bch2_trans_iter_exit(&trans, &iter);
 
@@ -215,6 +224,7 @@ struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
 {
 	struct bch_inode_unpacked inode_u;
 	struct bch_inode_info *inode;
+	struct btree_trans trans;
 	int ret;
 
 	inode = to_bch_ei(iget5_locked(c->vfs_sb,
@@ -227,14 +237,19 @@ struct inode *bch2_vfs_inode_get(struct bch_fs *c, subvol_inum inum)
 	if (!(inode->v.i_state & I_NEW))
 		return &inode->v;
 
-	ret = bch2_inode_find_by_inum(c, inum, &inode_u);
+	bch2_trans_init(&trans, c, 8, 0);
+	ret = lockrestart_do(&trans,
+		bch2_inode_find_by_inum_trans(&trans, inum, &inode_u));
+
+	if (!ret)
+		bch2_vfs_inode_init(&trans, inum, inode, &inode_u);
+	bch2_trans_exit(&trans);
+
 	if (ret) {
 		iget_failed(&inode->v);
 		return ERR_PTR(ret);
 	}
 
-	bch2_vfs_inode_init(c, inum, inode, &inode_u);
-
 	unlock_new_inode(&inode->v);
 
 	return &inode->v;
@@ -306,7 +321,7 @@ err_before_quota:
 	}
 
 	if (!(flags & BCH_CREATE_TMPFILE)) {
-		bch2_inode_update_after_write(c, dir, &dir_u,
+		bch2_inode_update_after_write(&trans, dir, &dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
 		mutex_unlock(&dir->ei_update_lock);
 	}
@@ -314,7 +329,8 @@ err_before_quota:
 	inum.subvol = inode_u.bi_subvol ?: dir->ei_subvol;
 	inum.inum = inode_u.bi_inum;
 
-	bch2_vfs_inode_init(c, inum, inode, &inode_u);
+	bch2_iget5_set(&inode->v, &inum);
+	bch2_vfs_inode_init(&trans, inum, inode, &inode_u);
 
 	set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 	set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
@@ -429,11 +445,9 @@ static int __bch2_link(struct bch_fs *c,
 					&dentry->d_name));
 
 	if (likely(!ret)) {
-		BUG_ON(inode_u.bi_inum != inode->v.i_ino);
-
-		bch2_inode_update_after_write(c, dir, &dir_u,
+		bch2_inode_update_after_write(&trans, dir, &dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
-		bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
+		bch2_inode_update_after_write(&trans, inode, &inode_u, ATTR_CTIME);
 	}
 
 	bch2_trans_exit(&trans);
@@ -481,11 +495,9 @@ int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
 					  deleting_snapshot));
 
 	if (likely(!ret)) {
-		BUG_ON(inode_u.bi_inum != inode->v.i_ino);
-
-		bch2_inode_update_after_write(c, dir, &dir_u,
+		bch2_inode_update_after_write(&trans, dir, &dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
-		bch2_inode_update_after_write(c, inode, &inode_u,
+		bch2_inode_update_after_write(&trans, inode, &inode_u,
 					      ATTR_MTIME);
 	}
 
@@ -613,18 +625,18 @@ static int bch2_rename2(struct mnt_idmap *idmap,
 	BUG_ON(dst_inode &&
 	       dst_inode->v.i_ino != dst_inode_u.bi_inum);
 
-	bch2_inode_update_after_write(c, src_dir, &src_dir_u,
+	bch2_inode_update_after_write(&trans, src_dir, &src_dir_u,
 				      ATTR_MTIME|ATTR_CTIME);
 
 	if (src_dir != dst_dir)
-		bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
+		bch2_inode_update_after_write(&trans, dst_dir, &dst_dir_u,
 					      ATTR_MTIME|ATTR_CTIME);
 
-	bch2_inode_update_after_write(c, src_inode, &src_inode_u,
+	bch2_inode_update_after_write(&trans, src_inode, &src_inode_u,
 				      ATTR_CTIME);
 
 	if (dst_inode)
-		bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
+		bch2_inode_update_after_write(&trans, dst_inode, &dst_inode_u,
 					      ATTR_CTIME);
 err:
 	bch2_trans_exit(&trans);
@@ -742,7 +754,7 @@ btree_err:
 	if (unlikely(ret))
 		goto err_trans;
 
-	bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
+	bch2_inode_update_after_write(&trans, inode, &inode_u, attr->ia_valid);
 
 	if (acl)
 		set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
@@ -1154,11 +1166,11 @@ static const struct export_operations bch_export_ops = {
 	//.get_parent	= bch2_get_parent,
 };
 
-static void bch2_vfs_inode_init(struct bch_fs *c, subvol_inum inum,
+static void bch2_vfs_inode_init(struct btree_trans *trans, subvol_inum inum,
 				struct bch_inode_info *inode,
 				struct bch_inode_unpacked *bi)
 {
-	bch2_inode_update_after_write(c, inode, bi, ~0);
+	bch2_inode_update_after_write(trans, inode, bi, ~0);
 
 	inode->v.i_blocks	= bi->bi_sectors;
 	inode->v.i_ino		= bi->bi_inum;
diff --git a/fs/bcachefs/fs.h b/fs/bcachefs/fs.h
index 1c8936df9fbb..530238780a88 100644
--- a/fs/bcachefs/fs.h
+++ b/fs/bcachefs/fs.h
@@ -172,7 +172,7 @@ struct inode *bch2_vfs_inode_get(struct bch_fs *, subvol_inum);
 typedef int (*inode_set_fn)(struct bch_inode_info *,
 			    struct bch_inode_unpacked *, void *);
 
-void bch2_inode_update_after_write(struct bch_fs *,
+void bch2_inode_update_after_write(struct btree_trans *,
 				   struct bch_inode_info *,
 				   struct bch_inode_unpacked *,
 				   unsigned);
diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index 728545141a39..a24bbc5228c1 100644
--- a/fs/bcachefs/inode.c
+++ b/fs/bcachefs/inode.c
@@ -722,9 +722,9 @@ err:
 	return ret;
 }
 
-static int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
-					 subvol_inum inum,
-					 struct bch_inode_unpacked *inode)
+int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
+				  subvol_inum inum,
+				  struct bch_inode_unpacked *inode)
 {
 	struct btree_iter iter;
 	int ret;
diff --git a/fs/bcachefs/inode.h b/fs/bcachefs/inode.h
index d433d48de4e0..723186d8afb6 100644
--- a/fs/bcachefs/inode.h
+++ b/fs/bcachefs/inode.h
@@ -89,6 +89,8 @@ int bch2_inode_create(struct btree_trans *, struct btree_iter *,
 
 int bch2_inode_rm(struct bch_fs *, subvol_inum, bool);
 
+int bch2_inode_find_by_inum_trans(struct btree_trans *, subvol_inum,
+				  struct bch_inode_unpacked *);
 int bch2_inode_find_by_inum(struct bch_fs *, subvol_inum,
 			    struct bch_inode_unpacked *);