]> git.karo-electronics.de Git - mv-sheeva.git/blobdiff - fs/btrfs/transaction.c
Btrfs: Let some locks go during defrag and snapshot dropping
[mv-sheeva.git] / fs / btrfs / transaction.c
index b5b99a85763f91e76723d8c05d0b8ee94bd264c6..88b0752550b63d0a84cebc2580cce4340f6b9c16 100644 (file)
@@ -16,7 +16,6 @@
  * Boston, MA 021110-1307, USA.
  */
 
-#include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/sched.h>
 #include "ctree.h"
@@ -30,6 +29,7 @@ extern struct kmem_cache *btrfs_transaction_cachep;
 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)
 {
@@ -70,29 +70,41 @@ static int join_transaction(struct btrfs_root *root)
        return 0;
 }
 
+static 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) {
+               WARN_ON(root == root->fs_info->extent_root);
+               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);
+                       radix_tree_tag_set(&root->fs_info->fs_roots_radix,
+                                  (unsigned long)root->root_key.objectid,
+                                  BTRFS_ROOT_DEFRAG_TAG);
+                       root->commit_root = root->node;
+                       get_bh(root->node);
+               } else {
+                       WARN_ON(1);
+               }
+               root->last_trans = running_trans_id;
+       }
+       return 0;
+}
+
 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
                                                   int num_blocks)
 {
        struct btrfs_trans_handle *h =
                kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
        int ret;
-       u64 running_trans_id;
 
        mutex_lock(&root->fs_info->trans_mutex);
        ret = join_transaction(root);
        BUG_ON(ret);
-       running_trans_id = root->fs_info->running_transaction->transid;
 
-       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);
-       }
-       root->last_trans = running_trans_id;
-       h->transid = running_trans_id;
+       record_root_in_trans(root);
+       h->transid = root->fs_info->running_transaction->transid;
        h->transaction = root->fs_info->running_transaction;
        h->blocks_reserved = num_blocks;
        h->blocks_used = 0;
@@ -109,10 +121,11 @@ int btrfs_end_transaction(struct btrfs_trans_handle *trans,
 
        mutex_lock(&root->fs_info->trans_mutex);
        cur_trans = root->fs_info->running_transaction;
+       WARN_ON(cur_trans != trans->transaction);
        WARN_ON(cur_trans->num_writers < 1);
+       cur_trans->num_writers--;
        if (waitqueue_active(&cur_trans->writer_wait))
                wake_up(&cur_trans->writer_wait);
-       cur_trans->num_writers--;
        put_transaction(cur_trans);
        mutex_unlock(&root->fs_info->trans_mutex);
        memset(trans, 0, sizeof(*trans));
@@ -149,6 +162,15 @@ int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
                                              gang[i]);
                        if (!page)
                                continue;
+                       if (PageWriteback(page)) {
+                               if (PageDirty(page))
+                                       wait_on_page_writeback(page);
+                               else {
+                                       unlock_page(page);
+                                       page_cache_release(page);
+                                       continue;
+                               }
+                       }
                        err = write_one_page(page, 0);
                        if (err)
                                werr = err;
@@ -190,6 +212,7 @@ static int wait_for_commit(struct btrfs_root *root,
                           struct btrfs_transaction *commit)
 {
        DEFINE_WAIT(wait);
+       mutex_lock(&root->fs_info->trans_mutex);
        while(!commit->commit_done) {
                prepare_to_wait(&commit->commit_wait, &wait,
                                TASK_UNINTERRUPTIBLE);
@@ -199,17 +222,28 @@ static int wait_for_commit(struct btrfs_root *root,
                schedule();
                mutex_lock(&root->fs_info->trans_mutex);
        }
+       mutex_unlock(&root->fs_info->trans_mutex);
        finish_wait(&commit->commit_wait, &wait);
        return 0;
 }
 
 struct dirty_root {
        struct list_head list;
-       struct btrfs_key snap_key;
-       struct buffer_head *commit_root;
        struct btrfs_root *root;
 };
 
+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;
+       dirty->root = root;
+       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)
@@ -219,7 +253,9 @@ static int add_dirty_roots(struct btrfs_trans_handle *trans,
        struct btrfs_root *root;
        int i;
        int ret;
-       int err;
+       int err = 0;
+       u32 refs;
+
        while(1) {
                ret = radix_tree_gang_lookup_tag(radix, (void **)gang, 0,
                                                 ARRAY_SIZE(gang),
@@ -240,22 +276,96 @@ static int add_dirty_roots(struct btrfs_trans_handle *trans,
                        }
                        dirty = kmalloc(sizeof(*dirty), GFP_NOFS);
                        BUG_ON(!dirty);
-                       memcpy(&dirty->snap_key, &root->root_key,
-                              sizeof(root->root_key));
-                       dirty->commit_root = root->commit_root;
+                       dirty->root = kmalloc(sizeof(*dirty->root), GFP_NOFS);
+                       BUG_ON(!dirty->root);
+
+                       memset(&root->root_item.drop_progress, 0,
+                              sizeof(struct btrfs_disk_key));
+                       root->root_item.drop_level = 0;
+
+                       memcpy(dirty->root, root, sizeof(*root));
+                       dirty->root->node = root->commit_root;
                        root->commit_root = NULL;
-                       dirty->root = root;
+
                        root->root_key.offset = root->fs_info->generation;
                        btrfs_set_root_blocknr(&root->root_item,
                                               bh_blocknr(root->node));
                        err = btrfs_insert_root(trans, root->fs_info->tree_root,
                                                &root->root_key,
                                                &root->root_item);
+                       if (err)
+                               break;
+
+                       refs = btrfs_root_refs(&dirty->root->root_item);
+                       btrfs_set_root_refs(&dirty->root->root_item, refs - 1);
+                       err = btrfs_update_root(trans, root->fs_info->tree_root,
+                                               &dirty->root->root_key,
+                                               &dirty->root->root_item);
+
                        BUG_ON(err);
-                       list_add(&dirty->list, list);
+                       if (refs == 1) {
+                               list_add(&dirty->list, list);
+                       } else {
+                               WARN_ON(1);
+                               kfree(dirty->root);
+                               kfree(dirty);
+                       }
                }
        }
-       return 0;
+       return err;
+}
+
+int btrfs_defrag_dirty_roots(struct btrfs_fs_info *info)
+{
+       struct btrfs_root *gang[1];
+       struct btrfs_root *root;
+       struct btrfs_root *tree_root = info->tree_root;
+       struct btrfs_trans_handle *trans;
+       int i;
+       int ret;
+       int err = 0;
+       u64 last = 0;
+
+       trans = btrfs_start_transaction(tree_root, 1);
+       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;
+                       radix_tree_tag_clear(&info->fs_roots_radix,
+                                    (unsigned long)root->root_key.objectid,
+                                    BTRFS_ROOT_DEFRAG_TAG);
+                       if (root->defrag_running)
+                               continue;
+
+                       while (1) {
+                               mutex_lock(&root->fs_info->trans_mutex);
+                               record_root_in_trans(root);
+                               mutex_unlock(&root->fs_info->trans_mutex);
+
+                               root->defrag_running = 1;
+                               err = btrfs_defrag_leaves(trans, root, 1);
+                               btrfs_end_transaction(trans, tree_root);
+                               mutex_unlock(&info->fs_mutex);
+
+                               btrfs_btree_balance_dirty(root);
+                               cond_resched();
+
+                               mutex_lock(&info->fs_mutex);
+                               trans = btrfs_start_transaction(tree_root, 1);
+                               if (err != -EAGAIN)
+                                       break;
+                       }
+                       root->defrag_running = 0;
+               }
+       }
+       btrfs_end_transaction(trans, tree_root);
+       return err;
 }
 
 static int drop_dirty_roots(struct btrfs_root *tree_root,
@@ -263,25 +373,49 @@ static int drop_dirty_roots(struct btrfs_root *tree_root,
 {
        struct dirty_root *dirty;
        struct btrfs_trans_handle *trans;
-       int ret;
+       int ret = 0;
+       int err;
+
        while(!list_empty(list)) {
                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);
-               ret = btrfs_drop_snapshot(trans, dirty->root,
-                                         dirty->commit_root);
-               BUG_ON(ret);
 
-               ret = btrfs_del_root(trans, tree_root, &dirty->snap_key);
+               while(1) {
+                       trans = btrfs_start_transaction(tree_root, 1);
+                       ret = btrfs_drop_snapshot(trans, dirty->root);
+                       if (ret != -EAGAIN) {
+                               break;
+                       }
+                       err = btrfs_update_root(trans,
+                                       tree_root,
+                                       &dirty->root->root_key,
+                                       &dirty->root->root_item);
+                       if (err)
+                               ret = err;
+                       ret = btrfs_end_transaction(trans, tree_root);
+                       BUG_ON(ret);
+                       mutex_unlock(&tree_root->fs_info->fs_mutex);
+
+                       btrfs_btree_balance_dirty(tree_root);
+                       schedule();
+
+                       mutex_lock(&tree_root->fs_info->fs_mutex);
+               }
                BUG_ON(ret);
+               ret = btrfs_del_root(trans, tree_root, &dirty->root->root_key);
+               if (ret)
+                       break;
                ret = btrfs_end_transaction(trans, tree_root);
                BUG_ON(ret);
+
+               kfree(dirty->root);
                kfree(dirty);
                mutex_unlock(&tree_root->fs_info->fs_mutex);
                btrfs_btree_balance_dirty(tree_root);
+               schedule();
        }
-       return 0;
+       return ret;
 }
 
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
@@ -291,65 +425,83 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
        struct btrfs_transaction *cur_trans;
        struct btrfs_transaction *prev_trans = NULL;
        struct list_head dirty_fs_roots;
+       struct radix_tree_root pinned_copy;
        DEFINE_WAIT(wait);
 
+       init_bit_radix(&pinned_copy);
        INIT_LIST_HEAD(&dirty_fs_roots);
 
        mutex_lock(&root->fs_info->trans_mutex);
        if (trans->transaction->in_commit) {
                cur_trans = trans->transaction;
                trans->transaction->use_count++;
+               mutex_unlock(&root->fs_info->trans_mutex);
                btrfs_end_transaction(trans, root);
+
+               mutex_unlock(&root->fs_info->fs_mutex);
                ret = wait_for_commit(root, cur_trans);
                BUG_ON(ret);
                put_transaction(cur_trans);
-               mutex_unlock(&root->fs_info->trans_mutex);
+               mutex_lock(&root->fs_info->fs_mutex);
                return 0;
        }
-       cur_trans = trans->transaction;
        trans->transaction->in_commit = 1;
+       cur_trans = trans->transaction;
+       if (cur_trans->list.prev != &root->fs_info->trans_list) {
+               prev_trans = list_entry(cur_trans->list.prev,
+                                       struct btrfs_transaction, list);
+               if (!prev_trans->commit_done) {
+                       prev_trans->use_count++;
+                       mutex_unlock(&root->fs_info->fs_mutex);
+                       mutex_unlock(&root->fs_info->trans_mutex);
+
+                       wait_for_commit(root, prev_trans);
+                       put_transaction(prev_trans);
+
+                       mutex_lock(&root->fs_info->fs_mutex);
+                       mutex_lock(&root->fs_info->trans_mutex);
+               }
+       }
        while (trans->transaction->num_writers > 1) {
                WARN_ON(cur_trans != trans->transaction);
                prepare_to_wait(&trans->transaction->writer_wait, &wait,
                                TASK_UNINTERRUPTIBLE);
                if (trans->transaction->num_writers <= 1)
                        break;
+               mutex_unlock(&root->fs_info->fs_mutex);
                mutex_unlock(&root->fs_info->trans_mutex);
                schedule();
+               mutex_lock(&root->fs_info->fs_mutex);
                mutex_lock(&root->fs_info->trans_mutex);
                finish_wait(&trans->transaction->writer_wait, &wait);
        }
        finish_wait(&trans->transaction->writer_wait, &wait);
        WARN_ON(cur_trans != trans->transaction);
-       add_dirty_roots(trans, &root->fs_info->fs_roots_radix, &dirty_fs_roots);
+       ret = add_dirty_roots(trans, &root->fs_info->fs_roots_radix,
+                             &dirty_fs_roots);
+       BUG_ON(ret);
+
        ret = btrfs_commit_tree_roots(trans, root);
        BUG_ON(ret);
+
        cur_trans = root->fs_info->running_transaction;
        root->fs_info->running_transaction = NULL;
-       if (cur_trans->list.prev != &root->fs_info->trans_list) {
-               prev_trans = list_entry(cur_trans->list.prev,
-                                       struct btrfs_transaction, list);
-               if (prev_trans->commit_done)
-                       prev_trans = NULL;
-               else
-                       prev_trans->use_count++;
-       }
+       btrfs_set_super_generation(&root->fs_info->super_copy,
+                                  cur_trans->transid);
+       btrfs_set_super_root(&root->fs_info->super_copy,
+                            bh_blocknr(root->fs_info->tree_root->node));
+       memcpy(root->fs_info->disk_super, &root->fs_info->super_copy,
+              sizeof(root->fs_info->super_copy));
+
+       btrfs_copy_pinned(root, &pinned_copy);
+
        mutex_unlock(&root->fs_info->trans_mutex);
        mutex_unlock(&root->fs_info->fs_mutex);
        ret = btrfs_write_and_wait_transaction(trans, root);
-       if (prev_trans) {
-               mutex_lock(&root->fs_info->trans_mutex);
-               wait_for_commit(root, prev_trans);
-               put_transaction(prev_trans);
-               mutex_unlock(&root->fs_info->trans_mutex);
-       }
-       btrfs_set_super_generation(root->fs_info->disk_super,
-                                  cur_trans->transid);
        BUG_ON(ret);
        write_ctree_super(trans, root);
-
        mutex_lock(&root->fs_info->fs_mutex);
-       btrfs_finish_extent_commit(trans, root);
+       btrfs_finish_extent_commit(trans, root, &pinned_copy);
        mutex_lock(&root->fs_info->trans_mutex);
        cur_trans->commit_done = 1;
        wake_up(&cur_trans->commit_wait);
@@ -399,6 +551,7 @@ void btrfs_transaction_cleaner(struct work_struct *work)
                goto out;
        }
        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);
 out: