Btrfs: Online btree defragmentation fixes

The btree defragger wasn't making forward progress because the new key wasn't
being saved by the btrfs_search_forward function.

This also disables the automatic btree defrag, it wasn't scaling well to
huge filesystems.  The auto-defrag needs to be done differently.

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

9 files changed, 190 insertions, 129 deletions

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 7f4cc2b88d09..0cb80f32a9c7 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -63,10 +63,9 @@ void btrfs_free_path(struct btrfs_path *p)
 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 {
 	int i;
-	int keep = p->keep_locks;
-	int skip = p->skip_locking;
 
 	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+		p->slots[i] = 0;
 		if (!p->nodes[i])
 			continue;
 		if (p->locks[i]) {
@@ -74,10 +73,8 @@ void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 			p->locks[i] = 0;
 		}
 		free_extent_buffer(p->nodes[i]);
+		p->nodes[i] = NULL;
 	}
-	memset(p, 0, sizeof(*p));
-	p->keep_locks = keep;
-	p->skip_locking = skip;
 }
 
 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
@@ -463,8 +460,6 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
 		search_start = cur->start;
 		last_block = cur->start;
 		*last_ret = search_start;
-		if (parent_level == 1)
-			btrfs_clear_buffer_defrag(cur);
 		btrfs_tree_unlock(cur);
 		free_extent_buffer(cur);
 	}
@@ -2969,8 +2964,138 @@ int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
 	return 1;
 }
 
+/*
+ * A helper function to walk down the tree starting at min_key, and looking
+ * for nodes or leaves that are either in cache or have a minimum
+ * transaction id.  This is used by the btree defrag code, but could
+ * also be used to search for blocks that have changed since a given
+ * transaction id.
+ *
+ * This does not cow, but it does stuff the starting key it finds back
+ * into min_key, so you can call btrfs_search_slot with cow=1 on the
+ * key and get a writable path.
+ *
+ * This does lock as it descends, and path->keep_locks should be set
+ * to 1 by the caller.
+ *
+ * This honors path->lowest_level to prevent descent past a given level
+ * of the tree.
+ *
+ * returns zero if something useful was found, < 0 on error and 1 if there
+ * was nothing in the tree that matched the search criteria.
+ */
+int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
+			 struct btrfs_path *path, int cache_only,
+			 u64 min_trans)
+{
+	struct extent_buffer *cur;
+	struct btrfs_key found_key;
+	int slot;
+	u32 nritems;
+	int level;
+	int ret = 1;
+
+again:
+	cur = btrfs_lock_root_node(root);
+	level = btrfs_header_level(cur);
+	path->nodes[level] = cur;
+	path->locks[level] = 1;
+
+	if (btrfs_header_generation(cur) < min_trans) {
+		ret = 1;
+		goto out;
+	}
+	while(1) {
+		nritems = btrfs_header_nritems(cur);
+		level = btrfs_header_level(cur);
+		bin_search(cur, min_key, level, &slot);
+
+		/* at level = 0, we're done, setup the path and exit */
+		if (level == 0) {
+			ret = 0;
+			path->slots[level] = slot;
+			btrfs_item_key_to_cpu(cur, &found_key, slot);
+			goto out;
+		}
+		/*
+		 * check this node pointer against the cache_only and
+		 * min_trans parameters.  If it isn't in cache or is too
+		 * old, skip to the next one.
+		 */
+		while(slot < nritems) {
+			u64 blockptr;
+			u64 gen;
+			struct extent_buffer *tmp;
+			blockptr = btrfs_node_blockptr(cur, slot);
+			gen = btrfs_node_ptr_generation(cur, slot);
+			if (gen < min_trans) {
+				slot++;
+				continue;
+			}
+			if (!cache_only)
+				break;
+
+			tmp = btrfs_find_tree_block(root, blockptr,
+					    btrfs_level_size(root, level - 1));
+
+			if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
+				free_extent_buffer(tmp);
+				break;
+			}
+			if (tmp)
+				free_extent_buffer(tmp);
+			slot++;
+		}
+		/*
+		 * we didn't find a candidate key in this node, walk forward
+		 * and find another one
+		 */
+		if (slot >= nritems) {
+			ret = btrfs_find_next_key(root, path, min_key, level,
+						  cache_only, min_trans);
+			if (ret == 0) {
+				btrfs_release_path(root, path);
+				goto again;
+			} else {
+				goto out;
+			}
+		}
+		/* save our key for returning back */
+		btrfs_node_key_to_cpu(cur, &found_key, slot);
+		path->slots[level] = slot;
+		if (level == path->lowest_level) {
+			ret = 0;
+			unlock_up(path, level, 1);
+			goto out;
+		}
+		cur = read_node_slot(root, cur, slot);
+
+		btrfs_tree_lock(cur);
+		path->locks[level - 1] = 1;
+		path->nodes[level - 1] = cur;
+		unlock_up(path, level, 1);
+	}
+out:
+	if (ret == 0)
+		memcpy(min_key, &found_key, sizeof(found_key));
+	return ret;
+}
+
+/*
+ * this is similar to btrfs_next_leaf, but does not try to preserve
+ * and fixup the path.  It looks for and returns the next key in the
+ * tree based on the current path and the cache_only and min_trans
+ * parameters.
+ *
+ * 0 is returned if another key is found, < 0 if there are any errors
+ * and 1 is returned if there are no higher keys in the tree
+ *
+ * path->keep_locks should be set to 1 on the search made before
+ * calling this function.
+ */
 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
-			struct btrfs_key *key, int lowest_level)
+			struct btrfs_key *key, int lowest_level,
+			int cache_only, u64 min_trans)
 {
 	int level = lowest_level;
 	int slot;
@@ -2982,6 +3107,7 @@ int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
 
 		slot = path->slots[level] + 1;
 		c = path->nodes[level];
+next:
 		if (slot >= btrfs_header_nritems(c)) {
 			level++;
 			if (level == BTRFS_MAX_LEVEL) {
@@ -2991,8 +3117,28 @@ int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
 		}
 		if (level == 0)
 			btrfs_item_key_to_cpu(c, key, slot);
-		else
+		else {
+			u64 blockptr = btrfs_node_blockptr(c, slot);
+			u64 gen = btrfs_node_ptr_generation(c, slot);
+
+			if (cache_only) {
+				struct extent_buffer *cur;
+				cur = btrfs_find_tree_block(root, blockptr,
+					    btrfs_level_size(root, level - 1));
+				if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
+					slot++;
+					if (cur)
+						free_extent_buffer(cur);
+					goto next;
+				}
+				free_extent_buffer(cur);
+			}
+			if (gen < min_trans) {
+				slot++;
+				goto next;
+			}
 			btrfs_node_key_to_cpu(c, key, slot);
+		}
 		return 0;
 	}
 	return 1;
@@ -3095,6 +3241,12 @@ done:
 	return 0;
 }
 
+/*
+ * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
+ * searching until it gets past min_objectid or finds an item of 'type'
+ *
+ * returns 0 if something is found, 1 if nothing was found and < 0 on error
+ */
 int btrfs_previous_item(struct btrfs_root *root,
 			struct btrfs_path *path, u64 min_objectid,
 			int type)
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index ca8e6f15859e..a28796482b4a 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -609,6 +609,7 @@ struct btrfs_root {
 	u64 last_inode_alloc;
 	int ref_cows;
 	int track_dirty;
+	u64 defrag_trans_start;
 	struct btrfs_key defrag_progress;
 	struct btrfs_key defrag_max;
 	int defrag_running;
@@ -1412,7 +1413,11 @@ int btrfs_previous_item(struct btrfs_root *root,
 struct extent_buffer *btrfs_root_node(struct btrfs_root *root);
 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root);
 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
-			struct btrfs_key *key, int lowest_level);
+			struct btrfs_key *key, int lowest_level,
+			int cache_only, u64 min_trans);
+int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
+			 struct btrfs_path *path, int cache_only,
+			 u64 min_trans);
 
 int btrfs_cow_block(struct btrfs_trans_handle *trans,
 		    struct btrfs_root *root, struct extent_buffer *buf,
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 31ca9f89388d..4cdc0b6a2672 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -295,7 +295,6 @@ int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 	ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 					     btrfs_header_generation(eb));
 	BUG_ON(ret);
-	btrfs_clear_buffer_defrag(eb);
 	found_start = btrfs_header_bytenr(eb);
 	if (found_start != start) {
 		printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
@@ -355,7 +354,6 @@ int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 	}
 	eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 
-	btrfs_clear_buffer_defrag(eb);
 	found_start = btrfs_header_bytenr(eb);
 	if (found_start != start) {
 		ret = -EIO;
@@ -736,6 +734,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 	memset(&root->root_item, 0, sizeof(root->root_item));
 	memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 	memset(&root->root_kobj, 0, sizeof(root->root_kobj));
+	root->defrag_trans_start = fs_info->generation;
 	init_completion(&root->kobj_unregister);
 	root->defrag_running = 0;
 	root->defrag_level = 0;
@@ -1168,7 +1167,6 @@ static int transaction_kthread(void *arg)
 			goto sleep;
 		}
 		mutex_unlock(&root->fs_info->trans_mutex);
-		btrfs_defrag_dirty_roots(root->fs_info);
 		trans = btrfs_start_transaction(root, 1);
 		ret = btrfs_commit_transaction(trans, root);
 sleep:
@@ -1434,12 +1432,12 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 						   tree_root,
 						   "btrfs-transaction");
 	if (!fs_info->transaction_kthread)
-		goto fail_trans_kthread;
+		goto fail_cleaner;
 
 
 	return tree_root;
 
-fail_trans_kthread:
+fail_cleaner:
 	kthread_stop(fs_info->cleaner_kthread);
 fail_extent_root:
 	free_extent_buffer(extent_root->node);
@@ -1662,7 +1660,6 @@ int close_ctree(struct btrfs_root *root)
 	kthread_stop(root->fs_info->transaction_kthread);
 	kthread_stop(root->fs_info->cleaner_kthread);
 
-	btrfs_defrag_dirty_roots(root->fs_info);
 	btrfs_clean_old_snapshots(root);
 	trans = btrfs_start_transaction(root, 1);
 	ret = btrfs_commit_transaction(trans, root);
@@ -1794,58 +1791,6 @@ void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
 	return;
 }
 
-void btrfs_set_buffer_defrag(struct extent_buffer *buf)
-{
-	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
-	struct inode *btree_inode = root->fs_info->btree_inode;
-	set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
-			buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
-}
-
-void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
-{
-	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
-	struct inode *btree_inode = root->fs_info->btree_inode;
-	set_extent_bits(&BTRFS_I(btree_inode)->io_tree, buf->start,
-			buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
-			GFP_NOFS);
-}
-
-int btrfs_buffer_defrag(struct extent_buffer *buf)
-{
-	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
-	struct inode *btree_inode = root->fs_info->btree_inode;
-	return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
-		     buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
-}
-
-int btrfs_buffer_defrag_done(struct extent_buffer *buf)
-{
-	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
-	struct inode *btree_inode = root->fs_info->btree_inode;
-	return test_range_bit(&BTRFS_I(btree_inode)->io_tree,
-		     buf->start, buf->start + buf->len - 1,
-		     EXTENT_DEFRAG_DONE, 0);
-}
-
-int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
-{
-	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
-	struct inode *btree_inode = root->fs_info->btree_inode;
-	return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
-		     buf->start, buf->start + buf->len - 1,
-		     EXTENT_DEFRAG_DONE, GFP_NOFS);
-}
-
-int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
-{
-	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
-	struct inode *btree_inode = root->fs_info->btree_inode;
-	return clear_extent_bits(&BTRFS_I(btree_inode)->io_tree,
-		     buf->start, buf->start + buf->len - 1,
-		     EXTENT_DEFRAG, GFP_NOFS);
-}
-
 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
 {
 	struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
diff --git a/fs/btrfs/disk-io.h b/fs/btrfs/disk-io.h
index deff6b4815a7..353c3c50c957 100644
--- a/fs/btrfs/disk-io.h
+++ b/fs/btrfs/disk-io.h
@@ -61,12 +61,6 @@ int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid);
 int btrfs_set_buffer_uptodate(struct extent_buffer *buf);
 int wait_on_tree_block_writeback(struct btrfs_root *root,
 				 struct extent_buffer *buf);
-void btrfs_set_buffer_defrag(struct extent_buffer *buf);
-void btrfs_set_buffer_defrag_done(struct extent_buffer *buf);
-int btrfs_buffer_defrag(struct extent_buffer *buf);
-int btrfs_buffer_defrag_done(struct extent_buffer *buf);
-int btrfs_clear_buffer_defrag(struct extent_buffer *buf);
-int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf);
 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid);
 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len);
 void btrfs_csum_final(u32 crc, char *result);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index dc3c03c6612d..5e0857ffbc35 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2095,8 +2095,6 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 
 	set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
 			 buf->start + buf->len - 1, GFP_NOFS);
-	if (!btrfs_test_opt(root, SSD))
-		btrfs_set_buffer_defrag(buf);
 	trans->blocks_used++;
 	return buf;
 }
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 726d6871fa13..5e28cf5c2e85 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -365,7 +365,6 @@ int btrfs_sync_fs(struct super_block *sb, int wait)
 		return 0;
 	}
 	btrfs_clean_old_snapshots(root);
-	btrfs_defrag_dirty_roots(root->fs_info);
 	trans = btrfs_start_transaction(root, 1);
 	ret = btrfs_commit_transaction(trans, root);
 	sb->s_dirt = 0;
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 8e909cb97c6d..98f422d9ab07 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -30,7 +30,6 @@ extern struct kmem_cache *btrfs_trans_handle_cachep;
 extern struct kmem_cache *btrfs_transaction_cachep;
 
 #define BTRFS_ROOT_TRANS_TAG 0
-#define BTRFS_ROOT_DEFRAG_TAG 1
 
 static noinline void put_transaction(struct btrfs_transaction *transaction)
 {
@@ -92,9 +91,6 @@ static noinline int record_root_in_trans(struct btrfs_root *root)
 			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
 				   (unsigned long)root->root_key.objectid,
 				   BTRFS_ROOT_TRANS_TAG);
-			radix_tree_tag_set(&root->fs_info->fs_roots_radix,
-				   (unsigned long)root->root_key.objectid,
-				   BTRFS_ROOT_DEFRAG_TAG);
 			root->commit_root = btrfs_root_node(root);
 		} else {
 			WARN_ON(1);
@@ -403,44 +399,15 @@ int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
 		cond_resched();
 
 		trans = btrfs_start_transaction(root, 1);
-		if (ret != -EAGAIN)
+		if (root->fs_info->closing || ret != -EAGAIN)
 			break;
 	}
 	root->defrag_running = 0;
 	smp_mb();
-	radix_tree_tag_clear(&info->fs_roots_radix,
-		     (unsigned long)root->root_key.objectid,
-		     BTRFS_ROOT_DEFRAG_TAG);
 	btrfs_end_transaction(trans, root);
 	return 0;
 }
 
-int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
-{
-	struct btrfs_root *gang[1];
-	struct btrfs_root *root;
-	int i;
-	int ret;
-	int err = 0;
-	u64 last = 0;
-
-	while(1) {
-		ret = radix_tree_gang_lookup_tag(&info->fs_roots_radix,
-						 (void **)gang, last,
-						 ARRAY_SIZE(gang),
-						 BTRFS_ROOT_DEFRAG_TAG);
-		if (ret == 0)
-			break;
-		for (i = 0; i < ret; i++) {
-			root = gang[i];
-			last = root->root_key.objectid + 1;
-			btrfs_defrag_root(root, 1);
-		}
-	}
-	btrfs_defrag_root(info->extent_root, 1);
-	return err;
-}
-
 static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
 				     struct list_head *list)
 {
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index e1e5a06b65f4..9ccd5a5b170f 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -84,7 +84,6 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 
 int btrfs_add_dead_root(struct btrfs_root *root, struct btrfs_root *latest,
 			struct list_head *dead_list);
-int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info);
 int btrfs_defrag_root(struct btrfs_root *root, int cacheonly);
 int btrfs_clean_old_snapshots(struct btrfs_root *root);
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
diff --git a/fs/btrfs/tree-defrag.c b/fs/btrfs/tree-defrag.c
index b17693f61fbc..cc2650b06952 100644
--- a/fs/btrfs/tree-defrag.c
+++ b/fs/btrfs/tree-defrag.c
@@ -32,10 +32,13 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 	int wret;
 	int level;
 	int orig_level;
-	int i;
 	int is_extent = 0;
 	int next_key_ret = 0;
 	u64 last_ret = 0;
+	u64 min_trans = 0;
+
+	if (cache_only)
+		goto out;
 
 	if (root->fs_info->extent_root == root) {
 		/*
@@ -43,10 +46,6 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		 * we can't defrag the extent root without deadlock
 		 */
 		goto out;
-#if 0
-		mutex_lock(&root->fs_info->alloc_mutex);
-		is_extent = 1;
-#endif
 	}
 
 	if (root->ref_cows == 0 && !is_extent)
@@ -84,6 +83,17 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 
 	path->lowest_level = 1;
 	path->keep_locks = 1;
+	if (cache_only)
+		min_trans = root->defrag_trans_start;
+
+	ret = btrfs_search_forward(root, &key, path, cache_only, min_trans);
+	if (ret < 0)
+		goto out;
+	if (ret > 0) {
+		ret = 0;
+		goto out;
+	}
+	btrfs_release_path(root, path);
 	wret = btrfs_search_slot(trans, root, &key, path, 0, 1);
 
 	if (wret < 0) {
@@ -95,7 +105,8 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		goto out;
 	}
 	path->slots[1] = btrfs_header_nritems(path->nodes[1]);
-	next_key_ret = btrfs_find_next_key(root, path, &key, 1);
+	next_key_ret = btrfs_find_next_key(root, path, &key, 1, cache_only,
+					   min_trans);
 	ret = btrfs_realloc_node(trans, root,
 				 path->nodes[1], 0,
 				 cache_only, &last_ret,
@@ -106,19 +117,9 @@ int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
 		ret = -EAGAIN;
 	}
 
-	for (i = 1; i < BTRFS_MAX_LEVEL; i++) {
-		if (path->locks[i]) {
-			btrfs_tree_unlock(path->nodes[i]);
-			path->locks[i] = 0;
-		}
-		if (path->nodes[i]) {
-			free_extent_buffer(path->nodes[i]);
-			path->nodes[i] = NULL;
-		}
-	}
+	btrfs_release_path(root, path);
 	if (is_extent)
 		btrfs_extent_post_op(trans, root);
-
 out:
 	if (is_extent)
 		mutex_unlock(&root->fs_info->alloc_mutex);
@@ -138,6 +139,7 @@ done:
 	if (ret != -EAGAIN) {
 		memset(&root->defrag_progress, 0,
 		       sizeof(root->defrag_progress));
+		root->defrag_trans_start = trans->transid;
 	}
 	return ret;
 }