Btrfs: Throttle file_write when data=ordered is flushing the inode

[karo-tx-linux.git] / fs / btrfs / transaction.c

diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 3f64d0c7ddb901cb886305c74adde4d8fad66ab2..c85cb48d95ee3dcd2aa1c8dbb1ad46971202563a 100644 (file)
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -33,7 +33,7 @@ static struct workqueue_struct *trans_wq;
 #define BTRFS_ROOT_TRANS_TAG 0
 #define BTRFS_ROOT_DEFRAG_TAG 1
 
-static void put_transaction(struct btrfs_transaction *transaction)
+static noinline void put_transaction(struct btrfs_transaction *transaction)
 {
        WARN_ON(transaction->use_count == 0);
        transaction->use_count--;
@@ -46,7 +46,7 @@ static void put_transaction(struct btrfs_transaction *transaction)
        }
 }
 
-static int join_transaction(struct btrfs_root *root)
+static noinline int join_transaction(struct btrfs_root *root)
 {
        struct btrfs_transaction *cur_trans;
        cur_trans = root->fs_info->running_transaction;
@@ -82,7 +82,7 @@ static int join_transaction(struct btrfs_root *root)
        return 0;
 }
 
-static int record_root_in_trans(struct btrfs_root *root)
+static noinline int record_root_in_trans(struct btrfs_root *root)
 {
        u64 running_trans_id = root->fs_info->running_transaction->transid;
        if (root->ref_cows && root->last_trans < running_trans_id) {
@@ -198,35 +198,48 @@ int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
        return werr;
 }
 
-int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
-                           struct btrfs_root *root)
+static int update_cowonly_root(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root)
 {
        int ret;
-       u64 old_extent_block;
-       struct btrfs_fs_info *fs_info = root->fs_info;
-       struct btrfs_root *tree_root = fs_info->tree_root;
-       struct btrfs_root *extent_root = fs_info->extent_root;
+       u64 old_root_bytenr;
+       struct btrfs_root *tree_root = root->fs_info->tree_root;
 
-       btrfs_write_dirty_block_groups(trans, extent_root);
+       btrfs_write_dirty_block_groups(trans, root);
        while(1) {
-               old_extent_block = btrfs_root_bytenr(&extent_root->root_item);
-               if (old_extent_block == extent_root->node->start)
+               old_root_bytenr = btrfs_root_bytenr(&root->root_item);
+               if (old_root_bytenr == root->node->start)
                        break;
-               btrfs_set_root_bytenr(&extent_root->root_item,
-                                     extent_root->node->start);
-               btrfs_set_root_level(&extent_root->root_item,
-                                    btrfs_header_level(extent_root->node));
+               btrfs_set_root_bytenr(&root->root_item,
+                                      root->node->start);
+               btrfs_set_root_level(&root->root_item,
+                                    btrfs_header_level(root->node));
                ret = btrfs_update_root(trans, tree_root,
-                                       &extent_root->root_key,
-                                       &extent_root->root_item);
+                                       &root->root_key,
+                                       &root->root_item);
                BUG_ON(ret);
-               btrfs_write_dirty_block_groups(trans, extent_root);
+               btrfs_write_dirty_block_groups(trans, root);
        }
        return 0;
 }
 
-static int wait_for_commit(struct btrfs_root *root,
-                          struct btrfs_transaction *commit)
+int btrfs_commit_tree_roots(struct btrfs_trans_handle *trans,
+                           struct btrfs_root *root)
+{
+       struct btrfs_fs_info *fs_info = root->fs_info;
+       struct list_head *next;
+
+       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);
+               update_cowonly_root(trans, root);
+       }
+       return 0;
+}
+
+static noinline int wait_for_commit(struct btrfs_root *root,
+                                   struct btrfs_transaction *commit)
 {
        DEFINE_WAIT(wait);
        mutex_lock(&root->fs_info->trans_mutex);
@@ -265,9 +278,9 @@ int btrfs_add_dead_root(struct btrfs_root *root,
        return 0;
 }
 
-static int add_dirty_roots(struct btrfs_trans_handle *trans,
-                          struct radix_tree_root *radix,
-                          struct list_head *list)
+static noinline int add_dirty_roots(struct btrfs_trans_handle *trans,
+                                   struct radix_tree_root *radix,
+                                   struct list_head *list)
 {
        struct dirty_root *dirty;
        struct btrfs_root *gang[8];
@@ -406,8 +419,8 @@ int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
        return err;
 }
 
-static int drop_dirty_roots(struct btrfs_root *tree_root,
-                           struct list_head *list)
+static noinline int drop_dirty_roots(struct btrfs_root *tree_root,
+                                    struct list_head *list)
 {
        struct dirty_root *dirty;
        struct btrfs_trans_handle *trans;
@@ -499,8 +512,11 @@ int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
                mutex_unlock(&root->fs_info->trans_mutex);
                mutex_unlock(&root->fs_info->fs_mutex);
 
-               if (S_ISREG(inode->i_mode))
+               if (S_ISREG(inode->i_mode)) {
+                       atomic_inc(&BTRFS_I(inode)->ordered_writeback);
                        filemap_fdatawrite(inode->i_mapping);
+                       atomic_dec(&BTRFS_I(inode)->ordered_writeback);
+               }
                iput(inode);
 
                mutex_lock(&root->fs_info->fs_mutex);
@@ -517,8 +533,11 @@ int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
                mutex_unlock(&root->fs_info->trans_mutex);
                mutex_unlock(&root->fs_info->fs_mutex);
 
-               if (S_ISREG(inode->i_mode))
+               if (S_ISREG(inode->i_mode)) {
+                       atomic_inc(&BTRFS_I(inode)->ordered_writeback);
                        filemap_write_and_wait(inode->i_mapping);
+                       atomic_dec(&BTRFS_I(inode)->ordered_writeback);
+               }
                atomic_dec(&inode->i_count);
                iput(inode);
 
@@ -529,23 +548,28 @@ int btrfs_write_ordered_inodes(struct btrfs_trans_handle *trans,
        return 0;
 }
 
-static int create_pending_snapshot(struct btrfs_trans_handle *trans,
+static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
                                   struct btrfs_fs_info *fs_info,
                                   struct btrfs_pending_snapshot *pending)
 {
        struct btrfs_key key;
-       struct btrfs_root_item new_root_item;
+       struct btrfs_root_item *new_root_item;
        struct btrfs_root *tree_root = fs_info->tree_root;
        struct btrfs_root *root = pending->root;
        struct extent_buffer *tmp;
        int ret;
        u64 objectid;
 
+       new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
+       if (!new_root_item) {
+               ret = -ENOMEM;
+               goto fail;
+       }
        ret = btrfs_find_free_objectid(trans, tree_root, 0, &objectid);
        if (ret)
                goto fail;
 
-       memcpy(&new_root_item, &root->root_item, sizeof(new_root_item));
+       memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 
        key.objectid = objectid;
        key.offset = 1;
@@ -557,10 +581,10 @@ static int create_pending_snapshot(struct btrfs_trans_handle *trans,
 
        btrfs_copy_root(trans, root, root->node, &tmp, objectid);
 
-       btrfs_set_root_bytenr(&new_root_item, tmp->start);
-       btrfs_set_root_level(&new_root_item, btrfs_header_level(tmp));
+       btrfs_set_root_bytenr(new_root_item, tmp->start);
+       btrfs_set_root_level(new_root_item, btrfs_header_level(tmp));
        ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
-                               &new_root_item);
+                               new_root_item);
        free_extent_buffer(tmp);
        if (ret)
                goto fail;
@@ -581,11 +605,12 @@ static int create_pending_snapshot(struct btrfs_trans_handle *trans,
                             pending->name, strlen(pending->name), objectid,
                             root->fs_info->sb->s_root->d_inode->i_ino);
 fail:
+       kfree(new_root_item);
        return ret;
 }
 
-static int create_pending_snapshots(struct btrfs_trans_handle *trans,
-                                  struct btrfs_fs_info *fs_info)
+static noinline int create_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;
@@ -610,6 +635,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        unsigned long timeout = 1;
        struct btrfs_transaction *cur_trans;
        struct btrfs_transaction *prev_trans = NULL;
+       struct btrfs_root *chunk_root = root->fs_info->chunk_root;
        struct list_head dirty_fs_roots;
        struct extent_io_tree *pinned_copy;
        DEFINE_WAIT(wait);
@@ -708,6 +734,10 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        btrfs_set_super_root_level(&root->fs_info->super_copy,
                           btrfs_header_level(root->fs_info->tree_root->node));
 
+       btrfs_set_super_chunk_root(&root->fs_info->super_copy,
+                                  chunk_root->node->start);
+       btrfs_set_super_chunk_root_level(&root->fs_info->super_copy,
+                                        btrfs_header_level(chunk_root->node));
        write_extent_buffer(root->fs_info->sb_buffer,
                            &root->fs_info->super_copy, 0,
                            sizeof(root->fs_info->super_copy));
@@ -819,7 +849,7 @@ void btrfs_transaction_flush_work(struct btrfs_root *root)
 
 void __init btrfs_init_transaction_sys(void)
 {
-       trans_wq = create_workqueue("btrfs");
+       trans_wq = create_workqueue("btrfs-transaction");
 }
 
 void btrfs_exit_transaction_sys(void)