Btrfs: Add the ability to find and remove dead roots after a crash.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

7 files changed, 164 insertions, 32 deletions

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 77071f2..fb6fffb 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -122,12 +122,12 @@ struct btrfs_super_block {
 	u8 fsid[16];    /* FS specific uuid */
 	__le64 blocknr; /* this block number */
 	__le64 magic;
-	__le32 blocksize;
 	__le64 generation;
 	__le64 root;
 	__le64 total_blocks;
 	__le64 blocks_used;
 	__le64 root_dir_objectid;
+	__le32 blocksize;
 } __attribute__ ((__packed__));
 
 /*
@@ -226,10 +226,12 @@ struct btrfs_root_item {
 	struct btrfs_inode_item inode;
 	__le64 root_dirid;
 	__le64 blocknr;
-	__le32 flags;
 	__le64 block_limit;
 	__le64 blocks_used;
+	__le32 flags;
 	__le32 refs;
+	struct btrfs_disk_key drop_progress;
+	u8 drop_level;
 } __attribute__ ((__packed__));
 
 #define BTRFS_FILE_EXTENT_REG 0
@@ -800,6 +802,16 @@ static inline void btrfs_set_root_refs(struct btrfs_root_item *item, u32 val)
 	item->refs = cpu_to_le32(val);
 }
 
+static inline u32 btrfs_root_flags(struct btrfs_root_item *item)
+{
+	return le32_to_cpu(item->flags);
+}
+
+static inline void btrfs_set_root_flags(struct btrfs_root_item *item, u32 val)
+{
+	item->flags = cpu_to_le32(val);
+}
+
 static inline u64 btrfs_super_blocknr(struct btrfs_super_block *s)
 {
 	return le64_to_cpu(s->blocknr);
@@ -1076,6 +1088,7 @@ int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *item);
 int btrfs_find_last_root(struct btrfs_root *root, u64 objectid, struct
 			 btrfs_root_item *item, struct btrfs_key *key);
+int btrfs_find_dead_roots(struct btrfs_root *root);
 /* dir-item.c */
 int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
 			  *root, const char *name, int name_len, u64 dir,
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 865a284..d7615e1 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -326,8 +326,8 @@ static int find_and_setup_root(int blocksize,
 	return 0;
 }
 
-struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
-				      struct btrfs_key *location)
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
+					       struct btrfs_key *location)
 {
 	struct btrfs_root *root;
 	struct btrfs_root *tree_root = fs_info->tree_root;
@@ -336,11 +336,7 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 	u64 highest_inode;
 	int ret = 0;
 
-	root = radix_tree_lookup(&fs_info->fs_roots_radix,
-				 (unsigned long)location->objectid);
-	if (root)
-		return root;
-	root = kmalloc(sizeof(*root), GFP_NOFS);
+	root = kzalloc(sizeof(*root), GFP_NOFS);
 	if (!root)
 		return ERR_PTR(-ENOMEM);
 	if (location->offset == (u64)-1) {
@@ -383,6 +379,28 @@ out:
 	BUG_ON(!root->node);
 insert:
 	root->ref_cows = 1;
+	ret = btrfs_find_highest_inode(root, &highest_inode);
+	if (ret == 0) {
+		root->highest_inode = highest_inode;
+		root->last_inode_alloc = highest_inode;
+	}
+	return root;
+}
+
+struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
+				      struct btrfs_key *location)
+{
+	struct btrfs_root *root;
+	int ret;
+
+	root = radix_tree_lookup(&fs_info->fs_roots_radix,
+				 (unsigned long)location->objectid);
+	if (root)
+		return root;
+
+	root = btrfs_read_fs_root_no_radix(fs_info, location);
+	if (IS_ERR(root))
+		return root;
 	ret = radix_tree_insert(&fs_info->fs_roots_radix,
 				(unsigned long)root->root_key.objectid,
 				root);
@@ -391,11 +409,6 @@ insert:
 		kfree(root);
 		return ERR_PTR(ret);
 	}
-	ret = btrfs_find_highest_inode(root, &highest_inode);
-	if (ret == 0) {
-		root->highest_inode = highest_inode;
-		root->last_inode_alloc = highest_inode;
-	}
 	return root;
 }
 
@@ -489,6 +502,9 @@ struct btrfs_root *open_ctree(struct super_block *sb)
 	btrfs_read_block_groups(extent_root);
 
 	fs_info->generation = btrfs_super_generation(disk_super) + 1;
+	ret = btrfs_find_dead_roots(tree_root);
+	if (ret)
+		goto fail_tree_root;
 	mutex_unlock(&fs_info->fs_mutex);
 	return tree_root;
 
@@ -538,7 +554,7 @@ int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 	return 0;
 }
 
-static int free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
+int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 {
 	radix_tree_delete(&fs_info->fs_roots_radix,
 			  (unsigned long)root->root_key.objectid);
@@ -565,7 +581,7 @@ static int del_fs_roots(struct btrfs_fs_info *fs_info)
 		if (!ret)
 			break;
 		for (i = 0; i < ret; i++)
-			free_fs_root(fs_info, gang[i]);
+			btrfs_free_fs_root(fs_info, gang[i]);
 	}
 	return 0;
 }
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index 7b76ccc..c4a695a 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -65,6 +65,8 @@ int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 		    char *result);
 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 				      struct btrfs_key *location);
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
+					       struct btrfs_key *location);
 u64 bh_blocknr(struct buffer_head *bh);
 int btrfs_insert_dev_radix(struct btrfs_root *root,
 			   struct block_device *bdev,
@@ -75,4 +77,5 @@ int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
 			     u64 logical);
 int btrfs_releasepage(struct page *page, gfp_t flags);
 void btrfs_btree_balance_dirty(struct btrfs_root *root);
+int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root);
 #endif
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 6d031da..9d2a0a3 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -2028,6 +2028,8 @@ static int create_subvol(struct btrfs_root *root, char *name, int namelen)
 
 	btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
 	btrfs_set_root_refs(&root_item, 1);
+	memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
+	root_item.drop_level = 0;
 	brelse(subvol);
 	subvol = NULL;
 
diff --git a/fs/btrfs/root-tree.c b/fs/btrfs/root-tree.c
index ac0fae7..737e5a3 100644
--- a/fs/btrfs/root-tree.c
+++ b/fs/btrfs/root-tree.c
@@ -18,6 +18,7 @@
 
 #include <linux/module.h>
 #include "ctree.h"
+#include "transaction.h"
 #include "disk-io.h"
 #include "print-tree.h"
 
@@ -32,7 +33,7 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid,
 
 	search_key.objectid = objectid;
 	search_key.flags = (u32)-1;
-	search_key.offset = (u32)-1;
+	search_key.offset = (u64)-1;
 
 	path = btrfs_alloc_path();
 	BUG_ON(!path);
@@ -50,6 +51,7 @@ int btrfs_find_last_root(struct btrfs_root *root, u64 objectid,
 	memcpy(item, btrfs_item_ptr(l, slot, struct btrfs_root_item),
 		sizeof(*item));
 	btrfs_disk_key_to_cpu(key, &l->items[slot].key);
+printk("find last finds key %Lu %u %Lu slot %d search for obj %Lu\n", key->objectid, key->flags, key->offset, slot, objectid);
 	ret = 0;
 out:
 	btrfs_release_path(root, path);
@@ -93,6 +95,67 @@ int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root
 	return ret;
 }
 
+int btrfs_find_dead_roots(struct btrfs_root *root)
+{
+	struct btrfs_root *dead_root;
+	struct btrfs_item *item;
+	struct btrfs_root_item *ri;
+	struct btrfs_key key;
+	struct btrfs_path *path;
+	int ret;
+	u32 nritems;
+	struct btrfs_leaf *leaf;
+	int slot;
+
+	key.objectid = 0;
+	key.flags = 0;
+	btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
+	key.offset = 0;
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+	if (ret < 0)
+		goto err;
+	while(1) {
+		leaf = btrfs_buffer_leaf(path->nodes[0]);
+		nritems = btrfs_header_nritems(&leaf->header);
+		slot = path->slots[0];
+		if (slot >= nritems) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret)
+				break;
+			leaf = btrfs_buffer_leaf(path->nodes[0]);
+			nritems = btrfs_header_nritems(&leaf->header);
+			slot = path->slots[0];
+		}
+		item = leaf->items + slot;
+		btrfs_disk_key_to_cpu(&key, &item->key);
+		if (btrfs_key_type(&key) != BTRFS_ROOT_ITEM_KEY)
+			goto next;
+		ri = btrfs_item_ptr(leaf, slot, struct btrfs_root_item);
+		if (btrfs_root_refs(ri) != 0)
+			goto next;
+		dead_root = btrfs_read_fs_root_no_radix(root->fs_info, &key);
+		if (IS_ERR(root)) {
+			ret = PTR_ERR(root);
+			goto err;
+		}
+printk("found dead root %Lu %u %Lu\n", key.objectid, key.flags, key.offset);
+		ret = btrfs_add_dead_root(dead_root,
+					  &root->fs_info->dead_roots);
+		if (ret)
+			goto err;
+next:
+		slot++;
+		path->slots[0]++;
+	}
+	ret = 0;
+err:
+	btrfs_free_path(path);
+	return ret;
+}
+
 int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 		   struct btrfs_key *key)
 {
@@ -111,14 +174,8 @@ int btrfs_del_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 			    path->slots[0], struct btrfs_root_item);
 
 	refs = btrfs_root_refs(ri);
-	BUG_ON(refs == 0);
-	if (refs == 1) {
-		ret = btrfs_del_item(trans, root, path);
-	} else {
-		btrfs_set_root_refs(ri, refs - 1);
-		WARN_ON(1);
-		mark_buffer_dirty(path->nodes[0]);
-	}
+	BUG_ON(refs != 0);
+	ret = btrfs_del_item(trans, root, path);
 out:
 	btrfs_release_path(root, path);
 	btrfs_free_path(path);
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 321f885..85a2a5e 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -85,11 +85,15 @@ struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
 
 	if (root != root->fs_info->tree_root && root->last_trans <
 	    running_trans_id) {
-		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
-				   (unsigned long)root->root_key.objectid,
-				   BTRFS_ROOT_TRANS_TAG);
-		root->commit_root = root->node;
-		get_bh(root->node);
+		if (root->root_item.refs != 0) {
+			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+					   (unsigned long)root->root_key.objectid,
+					   BTRFS_ROOT_TRANS_TAG);
+			root->commit_root = root->node;
+			get_bh(root->node);
+		} else {
+			WARN_ON(1);
+		}
 	}
 	root->last_trans = running_trans_id;
 	h->transid = running_trans_id;
@@ -208,8 +212,24 @@ struct dirty_root {
 	struct btrfs_key snap_key;
 	struct buffer_head *commit_root;
 	struct btrfs_root *root;
+	int free_on_drop;
 };
 
+int btrfs_add_dead_root(struct btrfs_root *root, struct list_head *dead_list)
+{
+	struct dirty_root *dirty;
+
+	dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
+	if (!dirty)
+		return -ENOMEM;
+	memcpy(&dirty->snap_key, &root->root_key, sizeof(root->root_key));
+	dirty->commit_root = root->node;
+	dirty->root = root;
+	dirty->free_on_drop = 1;
+	list_add(&dirty->list, dead_list);
+	return 0;
+}
+
 static int add_dirty_roots(struct btrfs_trans_handle *trans,
 			   struct radix_tree_root *radix,
 			   struct list_head *list)
@@ -217,9 +237,11 @@ static int add_dirty_roots(struct btrfs_trans_handle *trans,
 	struct dirty_root *dirty;
 	struct btrfs_root *gang[8];
 	struct btrfs_root *root;
+	struct btrfs_root_item tmp_item;
 	int i;
 	int ret;
 	int err = 0;
+	u32 refs;
 
 	while(1) {
 		ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
@@ -246,6 +268,9 @@ static int add_dirty_roots(struct btrfs_trans_handle *trans,
 			dirty->commit_root = root->commit_root;
 			root->commit_root = NULL;
 			dirty->root = root;
+			dirty->free_on_drop = 0;
+			memcpy(&tmp_item, &root->root_item, sizeof(tmp_item));
+
 			root->root_key.offset = root->fs_info->generation;
 			btrfs_set_root_blocknr(&root->root_item,
 					       bh_blocknr(root->node));
@@ -254,7 +279,18 @@ static int add_dirty_roots(struct btrfs_trans_handle *trans,
 						&root->root_item);
 			if (err)
 				break;
-			list_add(&dirty->list, list);
+
+			refs = btrfs_root_refs(&tmp_item);
+			btrfs_set_root_refs(&tmp_item, refs - 1);
+			err = btrfs_update_root(trans, root->fs_info->tree_root,
+						&dirty->snap_key,
+						&tmp_item);
+
+			BUG_ON(err);
+			if (refs == 1)
+				list_add(&dirty->list, list);
+			else
+				kfree(dirty);
 		}
 	}
 	return err;
@@ -270,16 +306,20 @@ static int drop_dirty_roots(struct btrfs_root *tree_root,
 		mutex_lock(&tree_root->fs_info->fs_mutex);
 		dirty = list_entry(list->next, struct dirty_root, list);
 		list_del_init(&dirty->list);
+
 		trans = btrfs_start_transaction(tree_root, 1);
+printk("deleting root %Lu %u %Lu\n", dirty->snap_key.objectid, dirty->snap_key.flags, dirty->snap_key.offset);
 		ret = btrfs_drop_snapshot(trans, dirty->root,
 					  dirty->commit_root);
 		BUG_ON(ret);
-
 		ret = btrfs_del_root(trans, tree_root, &dirty->snap_key);
 		if (ret)
 			break;
 		ret = btrfs_end_transaction(trans, tree_root);
 		BUG_ON(ret);
+
+		if (dirty->free_on_drop)
+			kfree(dirty->root);
 		kfree(dirty);
 		mutex_unlock(&tree_root->fs_info->fs_mutex);
 		btrfs_btree_balance_dirty(tree_root);
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index 5fb1d32..ebf44f3 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -69,5 +69,6 @@ void btrfs_transaction_flush_work(struct btrfs_root *root);
 void btrfs_transaction_queue_work(struct btrfs_root *root, int delay);
 void btrfs_init_transaction_sys(void);
 void btrfs_exit_transaction_sys(void);
+int btrfs_add_dead_root(struct btrfs_root *root, struct list_head *dead_list);
 
 #endif