Btrfs: do less aggressive btree readahead
[safe/jmp/linux-2.6] / fs / btrfs / transaction.c
index e72a013..919172d 100644 (file)
@@ -20,6 +20,7 @@
 #include <linux/sched.h>
 #include <linux/writeback.h>
 #include <linux/pagemap.h>
+#include <linux/blkdev.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
 #include "ref-cache.h"
 #include "tree-log.h"
 
-static int total_trans = 0;
-extern struct kmem_cache *btrfs_trans_handle_cachep;
-extern struct kmem_cache *btrfs_transaction_cachep;
-
 #define BTRFS_ROOT_TRANS_TAG 0
 
 static noinline void put_transaction(struct btrfs_transaction *transaction)
@@ -38,8 +35,6 @@ static noinline void put_transaction(struct btrfs_transaction *transaction)
        WARN_ON(transaction->use_count == 0);
        transaction->use_count--;
        if (transaction->use_count == 0) {
-               WARN_ON(total_trans == 0);
-               total_trans--;
                list_del_init(&transaction->list);
                memset(transaction, 0, sizeof(*transaction));
                kmem_cache_free(btrfs_transaction_cachep, transaction);
@@ -56,7 +51,6 @@ static noinline int join_transaction(struct btrfs_root *root)
        if (!cur_trans) {
                cur_trans = kmem_cache_alloc(btrfs_transaction_cachep,
                                             GFP_NOFS);
-               total_trans++;
                BUG_ON(!cur_trans);
                root->fs_info->generation++;
                root->fs_info->last_alloc = 0;
@@ -88,10 +82,10 @@ static noinline int join_transaction(struct btrfs_root *root)
 }
 
 /*
- * this does all the record keeping required to make sure that a
- * reference counted root is properly recorded in a given transaction.
- * This is required to make sure the old root from before we joined the transaction
- * is deleted when the transaction commits
+ * this does all the record keeping required to make sure that a reference
+ * counted root is properly recorded in a given transaction.  This is required
+ * to make sure the old root from before we joined the transaction is deleted
+ * when the transaction commits
  */
 noinline int btrfs_record_root_in_trans(struct btrfs_root *root)
 {
@@ -147,7 +141,7 @@ static void wait_current_trans(struct btrfs_root *root)
        if (cur_trans && cur_trans->blocked) {
                DEFINE_WAIT(wait);
                cur_trans->use_count++;
-               while(1) {
+               while (1) {
                        prepare_to_wait(&root->fs_info->transaction_wait, &wait,
                                        TASK_UNINTERRUPTIBLE);
                        if (cur_trans->blocked) {
@@ -185,7 +179,7 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
        h->transaction = root->fs_info->running_transaction;
        h->blocks_reserved = num_blocks;
        h->blocks_used = 0;
-       h->block_group = NULL;
+       h->block_group = 0;
        h->alloc_exclude_nr = 0;
        h->alloc_exclude_start = 0;
        root->fs_info->running_transaction->use_count++;
@@ -216,7 +210,7 @@ static noinline int wait_for_commit(struct btrfs_root *root,
 {
        DEFINE_WAIT(wait);
        mutex_lock(&root->fs_info->trans_mutex);
-       while(!commit->commit_done) {
+       while (!commit->commit_done) {
                prepare_to_wait(&commit->commit_wait, &wait,
                                TASK_UNINTERRUPTIBLE);
                if (commit->commit_done)
@@ -231,8 +225,8 @@ static noinline int wait_for_commit(struct btrfs_root *root,
 }
 
 /*
- * rate limit against the drop_snapshot code.  This helps to slow down new operations
- * if the drop_snapshot code isn't able to keep up.
+ * rate limit against the drop_snapshot code.  This helps to slow down new
+ * operations if the drop_snapshot code isn't able to keep up.
  */
 static void throttle_on_drops(struct btrfs_root *root)
 {
@@ -335,12 +329,12 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
        u64 end;
        unsigned long index;
 
-       while(1) {
+       while (1) {
                ret = find_first_extent_bit(dirty_pages, start, &start, &end,
                                            EXTENT_DIRTY);
                if (ret)
                        break;
-               while(start <= end) {
+               while (start <= end) {
                        cond_resched();
 
                        index = start >> PAGE_CACHE_SHIFT;
@@ -371,14 +365,14 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
                        page_cache_release(page);
                }
        }
-       while(1) {
+       while (1) {
                ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
                                            EXTENT_DIRTY);
                if (ret)
                        break;
 
                clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
-               while(start <= end) {
+               while (start <= end) {
                        index = start >> PAGE_CACHE_SHIFT;
                        start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
                        page = find_get_page(btree_inode->i_mapping, index);
@@ -434,7 +428,7 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
        btrfs_write_dirty_block_groups(trans, root);
        btrfs_extent_post_op(trans, root);
 
-       while(1) {
+       while (1) {
                old_root_bytenr = btrfs_root_bytenr(&root->root_item);
                if (old_root_bytenr == root->node->start)
                        break;
@@ -475,7 +469,7 @@ int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
 
        btrfs_extent_post_op(trans, fs_info->tree_root);
 
-       while(!list_empty(&fs_info->dirty_cowonly_roots)) {
+       while (!list_empty(&fs_info->dirty_cowonly_roots)) {
                next = fs_info->dirty_cowonly_roots.next;
                list_del_init(next);
                root = list_entry(next, struct btrfs_root, dirty_list);
@@ -524,7 +518,7 @@ static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
        int err = 0;
        u32 refs;
 
-       while(1) {
+       while (1) {
                ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
                                                 ARRAY_SIZE(gang),
                                                 BTRFS_ROOT_TRANS_TAG);
@@ -656,7 +650,7 @@ static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
        int ret = 0;
        int err;
 
-       while(!list_empty(list)) {
+       while (!list_empty(list)) {
                struct btrfs_root *root;
 
                dirty = list_entry(list->prev, struct btrfs_dirty_root, list);
@@ -666,13 +660,12 @@ static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
                root = dirty->latest_root;
                atomic_inc(&root->fs_info->throttles);
 
-               while(1) {
+               while (1) {
                        trans = btrfs_start_transaction(tree_root, 1);
                        mutex_lock(&root->fs_info->drop_mutex);
                        ret = btrfs_drop_snapshot(trans, dirty->root);
-                       if (ret != -EAGAIN) {
+                       if (ret != -EAGAIN)
                                break;
-                       }
                        mutex_unlock(&root->fs_info->drop_mutex);
 
                        err = btrfs_update_root(trans,
@@ -751,7 +744,6 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
        struct extent_buffer *tmp;
        struct extent_buffer *old;
        int ret;
-       int namelen;
        u64 objectid;
 
        new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
@@ -788,28 +780,67 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
        if (ret)
                goto fail;
 
+       key.offset = (u64)-1;
+       memcpy(&pending->root_key, &key, sizeof(key));
+fail:
+       kfree(new_root_item);
+       return ret;
+}
+
+static noinline int finish_pending_snapshot(struct btrfs_fs_info *fs_info,
+                                  struct btrfs_pending_snapshot *pending)
+{
+       int ret;
+       int namelen;
+       u64 index = 0;
+       struct btrfs_trans_handle *trans;
+       struct inode *parent_inode;
+       struct inode *inode;
+       struct btrfs_root *parent_root;
+
+       parent_inode = pending->dentry->d_parent->d_inode;
+       parent_root = BTRFS_I(parent_inode)->root;
+       trans = btrfs_join_transaction(parent_root, 1);
+
        /*
         * insert the directory item
         */
-       key.offset = (u64)-1;
        namelen = strlen(pending->name);
-       ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
-                                   pending->name, namelen,
-                                   root->fs_info->sb->s_root->d_inode->i_ino,
-                                   &key, BTRFS_FT_DIR, 0);
+       ret = btrfs_set_inode_index(parent_inode, &index);
+       ret = btrfs_insert_dir_item(trans, parent_root,
+                           pending->name, namelen,
+                           parent_inode->i_ino,
+                           &pending->root_key, BTRFS_FT_DIR, index);
 
        if (ret)
                goto fail;
 
-       ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
-                            pending->name, strlen(pending->name), objectid,
-                            root->fs_info->sb->s_root->d_inode->i_ino, 0);
+       btrfs_i_size_write(parent_inode, parent_inode->i_size + namelen * 2);
+       ret = btrfs_update_inode(trans, parent_root, parent_inode);
+       BUG_ON(ret);
+
+       /* add the backref first */
+       ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
+                                pending->root_key.objectid,
+                                BTRFS_ROOT_BACKREF_KEY,
+                                parent_root->root_key.objectid,
+                                parent_inode->i_ino, index, pending->name,
+                                namelen);
+
+       BUG_ON(ret);
 
-       /* Invalidate existing dcache entry for new snapshot. */
-       btrfs_invalidate_dcache_root(root, pending->name, namelen);
+       /* now add the forward ref */
+       ret = btrfs_add_root_ref(trans, parent_root->fs_info->tree_root,
+                                parent_root->root_key.objectid,
+                                BTRFS_ROOT_REF_KEY,
+                                pending->root_key.objectid,
+                                parent_inode->i_ino, index, pending->name,
+                                namelen);
 
+       inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
+       d_instantiate(pending->dentry, inode);
 fail:
-       kfree(new_root_item);
+       btrfs_end_transaction(trans, fs_info->fs_root);
        return ret;
 }
 
@@ -823,10 +854,24 @@ static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
        struct list_head *head = &trans->transaction->pending_snapshots;
        int ret;
 
-       while(!list_empty(head)) {
+       list_for_each_entry(pending, head, list) {
+               ret = create_pending_snapshot(trans, fs_info, pending);
+               BUG_ON(ret);
+       }
+       return 0;
+}
+
+static noinline int finish_pending_snapshots(struct btrfs_trans_handle *trans,
+                                            struct btrfs_fs_info *fs_info)
+{
+       struct btrfs_pending_snapshot *pending;
+       struct list_head *head = &trans->transaction->pending_snapshots;
+       int ret;
+
+       while (!list_empty(head)) {
                pending = list_entry(head->next,
                                     struct btrfs_pending_snapshot, list);
-               ret = create_pending_snapshot(trans, fs_info, pending);
+               ret = finish_pending_snapshot(fs_info, pending);
                BUG_ON(ret);
                list_del(&pending->list);
                kfree(pending->name);
@@ -991,7 +1036,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        mutex_unlock(&root->fs_info->trans_mutex);
        ret = btrfs_write_and_wait_transaction(trans, root);
        BUG_ON(ret);
-       write_ctree_super(trans, root);
+       write_ctree_super(trans, root, 0);
 
        /*
         * the super is written, we can safely allow the tree-loggers
@@ -1005,11 +1050,15 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        btrfs_drop_dead_reloc_roots(root);
        mutex_unlock(&root->fs_info->tree_reloc_mutex);
 
+       /* do the directory inserts of any pending snapshot creations */
+       finish_pending_snapshots(trans, root->fs_info);
+
        mutex_lock(&root->fs_info->trans_mutex);
 
        cur_trans->commit_done = 1;
        root->fs_info->last_trans_committed = cur_trans->transid;
        wake_up(&cur_trans->commit_wait);
+
        put_transaction(cur_trans);
        put_transaction(cur_trans);
 
@@ -1018,11 +1067,11 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
                list_splice_init(&root->fs_info->dead_roots, &dirty_fs_roots);
 
        mutex_unlock(&root->fs_info->trans_mutex);
+
        kmem_cache_free(btrfs_trans_handle_cachep, trans);
 
-       if (root->fs_info->closing) {
+       if (root->fs_info->closing)
                drop_dirty_roots(root->fs_info->tree_root, &dirty_fs_roots);
-       }
        return ret;
 }