Merge branch 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4
[safe/jmp/linux-2.6] / fs / btrfs / transaction.c
index 94f816c..2cb1160 100644 (file)
@@ -17,6 +17,7 @@
  */
 
 #include <linux/fs.h>
+#include <linux/slab.h>
 #include <linux/sched.h>
 #include <linux/writeback.h>
 #include <linux/pagemap.h>
@@ -69,7 +70,7 @@ static noinline int join_transaction(struct btrfs_root *root)
                cur_trans->commit_done = 0;
                cur_trans->start_time = get_seconds();
 
-               cur_trans->delayed_refs.root.rb_node = NULL;
+               cur_trans->delayed_refs.root = RB_ROOT;
                cur_trans->delayed_refs.num_entries = 0;
                cur_trans->delayed_refs.num_heads_ready = 0;
                cur_trans->delayed_refs.num_heads = 0;
@@ -104,7 +105,6 @@ static noinline int record_root_in_trans(struct btrfs_trans_handle *trans,
 {
        if (root->ref_cows && root->last_trans < trans->transid) {
                WARN_ON(root == root->fs_info->extent_root);
-               WARN_ON(root->root_item.refs == 0);
                WARN_ON(root->commit_root != root->node);
 
                radix_tree_tag_set(&root->fs_info->fs_roots_radix,
@@ -148,24 +148,25 @@ static void wait_current_trans(struct btrfs_root *root)
                while (1) {
                        prepare_to_wait(&root->fs_info->transaction_wait, &wait,
                                        TASK_UNINTERRUPTIBLE);
-                       if (cur_trans->blocked) {
-                               mutex_unlock(&root->fs_info->trans_mutex);
-                               schedule();
-                               mutex_lock(&root->fs_info->trans_mutex);
-                               finish_wait(&root->fs_info->transaction_wait,
-                                           &wait);
-                       } else {
-                               finish_wait(&root->fs_info->transaction_wait,
-                                           &wait);
+                       if (!cur_trans->blocked)
                                break;
-                       }
+                       mutex_unlock(&root->fs_info->trans_mutex);
+                       schedule();
+                       mutex_lock(&root->fs_info->trans_mutex);
                }
+               finish_wait(&root->fs_info->transaction_wait, &wait);
                put_transaction(cur_trans);
        }
 }
 
+enum btrfs_trans_type {
+       TRANS_START,
+       TRANS_JOIN,
+       TRANS_USERSPACE,
+};
+
 static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
-                                            int num_blocks, int wait)
+                                            int num_blocks, int type)
 {
        struct btrfs_trans_handle *h =
                kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS);
@@ -173,7 +174,8 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 
        mutex_lock(&root->fs_info->trans_mutex);
        if (!root->fs_info->log_root_recovering &&
-           ((wait == 1 && !root->fs_info->open_ioctl_trans) || wait == 2))
+           ((type == TRANS_START && !root->fs_info->open_ioctl_trans) ||
+            type == TRANS_USERSPACE))
                wait_current_trans(root);
        ret = join_transaction(root);
        BUG_ON(ret);
@@ -187,6 +189,9 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
        h->alloc_exclude_start = 0;
        h->delayed_ref_updates = 0;
 
+       if (!current->journal_info && type != TRANS_USERSPACE)
+               current->journal_info = h;
+
        root->fs_info->running_transaction->use_count++;
        record_root_in_trans(h, root);
        mutex_unlock(&root->fs_info->trans_mutex);
@@ -196,18 +201,18 @@ static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root,
 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
                                                   int num_blocks)
 {
-       return start_transaction(root, num_blocks, 1);
+       return start_transaction(root, num_blocks, TRANS_START);
 }
 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
                                                   int num_blocks)
 {
-       return start_transaction(root, num_blocks, 0);
+       return start_transaction(root, num_blocks, TRANS_JOIN);
 }
 
 struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
                                                         int num_blocks)
 {
-       return start_transaction(r, num_blocks, 2);
+       return start_transaction(r, num_blocks, TRANS_USERSPACE);
 }
 
 /* wait for a transaction commit to be fully complete */
@@ -318,9 +323,15 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
                wake_up(&cur_trans->writer_wait);
        put_transaction(cur_trans);
        mutex_unlock(&info->trans_mutex);
+
+       if (current->journal_info == trans)
+               current->journal_info = NULL;
        memset(trans, 0, sizeof(*trans));
        kmem_cache_free(btrfs_trans_handle_cachep, trans);
 
+       if (throttle)
+               btrfs_run_delayed_iputs(root);
+
        return 0;
 }
 
@@ -339,10 +350,10 @@ int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
 /*
  * when btree blocks are allocated, they have some corresponding bits set for
  * them in one of two extent_io trees.  This is used to make sure all of
- * those extents are on disk for transaction or log commit
+ * those extents are sent to disk but does not wait on them
  */
-int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
-                                       struct extent_io_tree *dirty_pages)
+int btrfs_write_marked_extents(struct btrfs_root *root,
+                              struct extent_io_tree *dirty_pages, int mark)
 {
        int ret;
        int err = 0;
@@ -355,7 +366,7 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
 
        while (1) {
                ret = find_first_extent_bit(dirty_pages, start, &start, &end,
-                                           EXTENT_DIRTY);
+                                           mark);
                if (ret)
                        break;
                while (start <= end) {
@@ -389,13 +400,36 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
                        page_cache_release(page);
                }
        }
+       if (err)
+               werr = err;
+       return werr;
+}
+
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees.  This is used to make sure all of
+ * those extents are on disk for transaction or log commit.  We wait
+ * on all the pages and clear them from the dirty pages state tree
+ */
+int btrfs_wait_marked_extents(struct btrfs_root *root,
+                             struct extent_io_tree *dirty_pages, int mark)
+{
+       int ret;
+       int err = 0;
+       int werr = 0;
+       struct page *page;
+       struct inode *btree_inode = root->fs_info->btree_inode;
+       u64 start = 0;
+       u64 end;
+       unsigned long index;
+
        while (1) {
-               ret = find_first_extent_bit(dirty_pages, 0, &start, &end,
-                                           EXTENT_DIRTY);
+               ret = find_first_extent_bit(dirty_pages, start, &start, &end,
+                                           mark);
                if (ret)
                        break;
 
-               clear_extent_dirty(dirty_pages, start, end, GFP_NOFS);
+               clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
                while (start <= end) {
                        index = start >> PAGE_CACHE_SHIFT;
                        start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
@@ -419,6 +453,22 @@ int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
        return werr;
 }
 
+/*
+ * when btree blocks are allocated, they have some corresponding bits set for
+ * them in one of two extent_io trees.  This is used to make sure all of
+ * those extents are on disk for transaction or log commit
+ */
+int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
+                               struct extent_io_tree *dirty_pages, int mark)
+{
+       int ret;
+       int ret2;
+
+       ret = btrfs_write_marked_extents(root, dirty_pages, mark);
+       ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark);
+       return ret || ret2;
+}
+
 int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
                                     struct btrfs_root *root)
 {
@@ -428,7 +478,8 @@ int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
                return filemap_write_and_wait(btree_inode->i_mapping);
        }
        return btrfs_write_and_wait_marked_extents(root,
-                                          &trans->transaction->dirty_pages);
+                                          &trans->transaction->dirty_pages,
+                                          EXTENT_DIRTY);
 }
 
 /*
@@ -446,13 +497,16 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
 {
        int ret;
        u64 old_root_bytenr;
+       u64 old_root_used;
        struct btrfs_root *tree_root = root->fs_info->tree_root;
 
+       old_root_used = btrfs_root_used(&root->root_item);
        btrfs_write_dirty_block_groups(trans, root);
 
        while (1) {
                old_root_bytenr = btrfs_root_bytenr(&root->root_item);
-               if (old_root_bytenr == root->node->start)
+               if (old_root_bytenr == root->node->start &&
+                   old_root_used == btrfs_root_used(&root->root_item))
                        break;
 
                btrfs_set_root_node(&root->root_item, root->node);
@@ -461,6 +515,7 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
                                        &root->root_item);
                BUG_ON(ret);
 
+               old_root_used = btrfs_root_used(&root->root_item);
                ret = btrfs_write_dirty_block_groups(trans, root);
                BUG_ON(ret);
        }
@@ -701,10 +756,17 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
        struct btrfs_root_item *new_root_item;
        struct btrfs_root *tree_root = fs_info->tree_root;
        struct btrfs_root *root = pending->root;
+       struct btrfs_root *parent_root;
+       struct inode *parent_inode;
        struct extent_buffer *tmp;
        struct extent_buffer *old;
        int ret;
        u64 objectid;
+       int namelen;
+       u64 index = 0;
+
+       parent_inode = pending->dentry->d_parent->d_inode;
+       parent_root = BTRFS_I(parent_inode)->root;
 
        new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS);
        if (!new_root_item) {
@@ -715,92 +777,59 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
        if (ret)
                goto fail;
 
-       record_root_in_trans(trans, root);
-       btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
-       memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
-
        key.objectid = objectid;
        /* record when the snapshot was created in key.offset */
        key.offset = trans->transid;
        btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
 
-       old = btrfs_lock_root_node(root);
-       btrfs_cow_block(trans, root, old, NULL, 0, &old);
-       btrfs_set_lock_blocking(old);
-
-       btrfs_copy_root(trans, root, old, &tmp, objectid);
-       btrfs_tree_unlock(old);
-       free_extent_buffer(old);
-
-       btrfs_set_root_node(new_root_item, tmp);
-       ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
-                               new_root_item);
-       btrfs_tree_unlock(tmp);
-       free_extent_buffer(tmp);
-       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);
+       pending->root_key.offset = (u64)-1;
 
+       record_root_in_trans(trans, parent_root);
        /*
         * insert the directory item
         */
        namelen = strlen(pending->name);
        ret = btrfs_set_inode_index(parent_inode, &index);
+       BUG_ON(ret);
        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;
+       BUG_ON(ret);
 
        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);
+       record_root_in_trans(trans, root);
+       btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
+       memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 
+       old = btrfs_lock_root_node(root);
+       btrfs_cow_block(trans, root, old, NULL, 0, &old);
+       btrfs_set_lock_blocking(old);
+
+       btrfs_copy_root(trans, root, old, &tmp, objectid);
+       btrfs_tree_unlock(old);
+       free_extent_buffer(old);
+
+       btrfs_set_root_node(new_root_item, tmp);
+       ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
+                               new_root_item);
        BUG_ON(ret);
+       btrfs_tree_unlock(tmp);
+       free_extent_buffer(tmp);
 
-       /* 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_root->root_key.objectid,
                                 parent_inode->i_ino, index, pending->name,
                                 namelen);
+       BUG_ON(ret);
 
-       inode = btrfs_lookup_dentry(parent_inode, pending->dentry);
-       d_instantiate(pending->dentry, inode);
 fail:
-       btrfs_end_transaction(trans, fs_info->fs_root);
+       kfree(new_root_item);
        return ret;
 }
 
@@ -821,25 +850,6 @@ static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans,
        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 = finish_pending_snapshot(fs_info, pending);
-               BUG_ON(ret);
-               list_del(&pending->list);
-               kfree(pending->name);
-               kfree(pending);
-       }
-       return 0;
-}
-
 static void update_super_roots(struct btrfs_root *root)
 {
        struct btrfs_root_item *root_item;
@@ -951,12 +961,9 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 
                mutex_unlock(&root->fs_info->trans_mutex);
 
-               if (flush_on_commit) {
-                       btrfs_start_delalloc_inodes(root);
-                       ret = btrfs_wait_ordered_extents(root, 0);
-                       BUG_ON(ret);
-               } else if (snap_pending) {
-                       ret = btrfs_wait_ordered_extents(root, 1);
+               if (flush_on_commit || snap_pending) {
+                       btrfs_start_delalloc_inodes(root, 1);
+                       ret = btrfs_wait_ordered_extents(root, 0, 1);
                        BUG_ON(ret);
                }
 
@@ -1054,9 +1061,6 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 
        btrfs_finish_extent_commit(trans, root);
 
-       /* 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;
@@ -1070,7 +1074,14 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 
        mutex_unlock(&root->fs_info->trans_mutex);
 
+       if (current->journal_info == trans)
+               current->journal_info = NULL;
+
        kmem_cache_free(btrfs_trans_handle_cachep, trans);
+
+       if (current != root->fs_info->transaction_kthread)
+               btrfs_run_delayed_iputs(root);
+
        return ret;
 }
 
@@ -1088,8 +1099,13 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
 
        while (!list_empty(&list)) {
                root = list_entry(list.next, struct btrfs_root, root_list);
-               list_del_init(&root->root_list);
-               btrfs_drop_snapshot(root, 0);
+               list_del(&root->root_list);
+
+               if (btrfs_header_backref_rev(root->node) <
+                   BTRFS_MIXED_BACKREF_REV)
+                       btrfs_drop_snapshot(root, 0);
+               else
+                       btrfs_drop_snapshot(root, 1);
        }
        return 0;
 }