Btrfs: Copy into the log tree in big batches
[deliverable/linux.git] / fs / btrfs / disk-io.c
index bbba14b629d217d2f96e19707bb6e9f3eb82c6f1..5edb7f885799b46920124d430661be21087e6a0a 100644 (file)
@@ -41,6 +41,7 @@
 #include "async-thread.h"
 #include "locking.h"
 #include "ref-cache.h"
+#include "tree-log.h"
 
 #if 0
 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
@@ -250,7 +251,7 @@ static int btree_read_extent_buffer_pages(struct btrfs_root *root,
                if (!ret &&
                    !verify_parent_transid(io_tree, eb, parent_transid))
                        return ret;
-
+printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
                num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
                                              eb->start, eb->len);
                if (num_copies == 1)
@@ -306,9 +307,7 @@ int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
                goto err;
        }
        found_level = btrfs_header_level(eb);
-       spin_lock(&root->fs_info->hash_lock);
-       btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
-       spin_unlock(&root->fs_info->hash_lock);
+
        csum_tree_block(root, eb, 0);
 err:
        free_extent_buffer(eb);
@@ -347,7 +346,10 @@ int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
        eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 
        found_start = btrfs_header_bytenr(eb);
-       if (found_start != start) {
+       if (0 && found_start != start) {
+               printk("bad tree block start %llu %llu\n",
+                      (unsigned long long)found_start,
+                      (unsigned long long)eb->start);
                ret = -EIO;
                goto err;
        }
@@ -487,9 +489,15 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
        atomic_inc(&fs_info->nr_async_submits);
        btrfs_queue_worker(&fs_info->workers, &async->work);
 
-       wait_event_timeout(fs_info->async_submit_wait,
+       if (atomic_read(&fs_info->nr_async_submits) > limit) {
+               wait_event_timeout(fs_info->async_submit_wait,
                           (atomic_read(&fs_info->nr_async_submits) < limit),
                           HZ/10);
+
+               wait_event_timeout(fs_info->async_submit_wait,
+                          (atomic_read(&fs_info->nr_async_bios) < limit),
+                          HZ/10);
+       }
        return 0;
 }
 
@@ -584,6 +592,9 @@ static int btree_releasepage(struct page *page, gfp_t gfp_flags)
        struct extent_map_tree *map;
        int ret;
 
+       if (PageWriteback(page) || PageDirty(page))
+           return 0;
+
        tree = &BTRFS_I(page->mapping->host)->io_tree;
        map = &BTRFS_I(page->mapping->host)->extent_tree;
 
@@ -685,6 +696,18 @@ struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 }
 
 
+int btrfs_write_tree_block(struct extent_buffer *buf)
+{
+       return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
+                                     buf->start + buf->len - 1, WB_SYNC_NONE);
+}
+
+int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
+{
+       return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
+                                 buf->start, buf->start + buf->len -1);
+}
+
 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
                                      u32 blocksize, u64 parent_transid)
 {
@@ -703,6 +726,8 @@ struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 
        if (ret == 0) {
                buf->flags |= EXTENT_UPTODATE;
+       } else {
+               WARN_ON(1);
        }
        return buf;
 
@@ -721,15 +746,6 @@ int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
        return 0;
 }
 
-int wait_on_tree_block_writeback(struct btrfs_root *root,
-                                struct extent_buffer *buf)
-{
-       struct inode *btree_inode = root->fs_info->btree_inode;
-       wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
-                                       buf);
-       return 0;
-}
-
 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
                        u32 stripesize, struct btrfs_root *root,
                        struct btrfs_fs_info *fs_info,
@@ -760,6 +776,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
        spin_lock_init(&root->node_lock);
        spin_lock_init(&root->list_lock);
        mutex_init(&root->objectid_mutex);
+       mutex_init(&root->log_mutex);
 
        btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
        root->ref_tree = &root->ref_tree_struct;
@@ -798,11 +815,73 @@ static int find_and_setup_root(struct btrfs_root *tree_root,
        return 0;
 }
 
-struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
-                                              struct btrfs_key *location)
+int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
+                            struct btrfs_fs_info *fs_info)
+{
+       struct extent_buffer *eb;
+       int ret;
+
+       if (!fs_info->log_root_tree)
+               return 0;
+
+       eb = fs_info->log_root_tree->node;
+
+       WARN_ON(btrfs_header_level(eb) != 0);
+       WARN_ON(btrfs_header_nritems(eb) != 0);
+
+       ret = btrfs_free_reserved_extent(fs_info->tree_root,
+                               eb->start, eb->len);
+       BUG_ON(ret);
+
+       free_extent_buffer(eb);
+       kfree(fs_info->log_root_tree);
+       fs_info->log_root_tree = NULL;
+       return 0;
+}
+
+int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
+                            struct btrfs_fs_info *fs_info)
 {
        struct btrfs_root *root;
        struct btrfs_root *tree_root = fs_info->tree_root;
+
+       root = kzalloc(sizeof(*root), GFP_NOFS);
+       if (!root)
+               return -ENOMEM;
+
+       __setup_root(tree_root->nodesize, tree_root->leafsize,
+                    tree_root->sectorsize, tree_root->stripesize,
+                    root, fs_info, BTRFS_TREE_LOG_OBJECTID);
+
+       root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
+       root->root_key.type = BTRFS_ROOT_ITEM_KEY;
+       root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
+       root->ref_cows = 0;
+
+       root->node = btrfs_alloc_free_block(trans, root, root->leafsize,
+                                           BTRFS_TREE_LOG_OBJECTID,
+                                           0, 0, 0, 0, 0);
+
+       btrfs_set_header_nritems(root->node, 0);
+       btrfs_set_header_level(root->node, 0);
+       btrfs_set_header_bytenr(root->node, root->node->start);
+       btrfs_set_header_generation(root->node, trans->transid);
+       btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
+
+       write_extent_buffer(root->node, root->fs_info->fsid,
+                           (unsigned long)btrfs_header_fsid(root->node),
+                           BTRFS_FSID_SIZE);
+       btrfs_mark_buffer_dirty(root->node);
+       btrfs_tree_unlock(root->node);
+       fs_info->log_root_tree = root;
+       return 0;
+}
+
+struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
+                                              struct btrfs_key *location)
+{
+       struct btrfs_root *root;
+       struct btrfs_fs_info *fs_info = tree_root->fs_info;
        struct btrfs_path *path;
        struct extent_buffer *l;
        u64 highest_inode;
@@ -852,11 +931,13 @@ out:
                                     blocksize, 0);
        BUG_ON(!root->node);
 insert:
-       root->ref_cows = 1;
-       ret = btrfs_find_highest_inode(root, &highest_inode);
-       if (ret == 0) {
-               root->highest_inode = highest_inode;
-               root->last_inode_alloc = highest_inode;
+       if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
+               root->ref_cows = 1;
+               ret = btrfs_find_highest_inode(root, &highest_inode);
+               if (ret == 0) {
+                       root->highest_inode = highest_inode;
+                       root->last_inode_alloc = highest_inode;
+               }
        }
        return root;
 }
@@ -896,7 +977,7 @@ struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
        if (root)
                return root;
 
-       root = btrfs_read_fs_root_no_radix(fs_info, location);
+       root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
        if (IS_ERR(root))
                return root;
        ret = radix_tree_insert(&fs_info->fs_roots_radix,
@@ -1239,16 +1320,18 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        u32 blocksize;
        u32 stripesize;
        struct buffer_head *bh;
-       struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
                                                 GFP_NOFS);
-       struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
                                               GFP_NOFS);
        struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
                                                GFP_NOFS);
-       struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
                                                GFP_NOFS);
-       struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
+       struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
                                              GFP_NOFS);
+       struct btrfs_root *log_tree_root;
+
        int ret;
        int err = -EINVAL;
 
@@ -1332,6 +1415,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 
        mutex_init(&fs_info->trans_mutex);
+       mutex_init(&fs_info->tree_log_mutex);
        mutex_init(&fs_info->drop_mutex);
        mutex_init(&fs_info->alloc_mutex);
        mutex_init(&fs_info->chunk_mutex);
@@ -1341,6 +1425,10 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        init_waitqueue_head(&fs_info->transaction_throttle);
        init_waitqueue_head(&fs_info->transaction_wait);
        init_waitqueue_head(&fs_info->async_submit_wait);
+       init_waitqueue_head(&fs_info->tree_log_wait);
+       atomic_set(&fs_info->tree_log_commit, 0);
+       atomic_set(&fs_info->tree_log_writers, 0);
+       fs_info->tree_log_transid = 0;
 
 #if 0
        ret = add_hasher(fs_info, "crc32c");
@@ -1408,7 +1496,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
         * low idle thresh
         */
        fs_info->endio_workers.idle_thresh = 4;
-       fs_info->endio_write_workers.idle_thresh = 4;
+       fs_info->endio_write_workers.idle_thresh = 64;
 
        btrfs_start_workers(&fs_info->workers, 1);
        btrfs_start_workers(&fs_info->submit_workers, 1);
@@ -1521,7 +1609,26 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        if (!fs_info->transaction_kthread)
                goto fail_cleaner;
 
+       if (btrfs_super_log_root(disk_super) != 0) {
+               u32 blocksize;
+               u64 bytenr = btrfs_super_log_root(disk_super);
+
+               blocksize =
+                    btrfs_level_size(tree_root,
+                                     btrfs_super_log_root_level(disk_super));
+
+               log_tree_root = kzalloc(sizeof(struct btrfs_root),
+                                                     GFP_NOFS);
+
+               __setup_root(nodesize, leafsize, sectorsize, stripesize,
+                            log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
 
+               log_tree_root->node = read_tree_block(tree_root, bytenr,
+                                                     blocksize, 0);
+               ret = btrfs_recover_log_trees(log_tree_root);
+               BUG_ON(ret);
+       }
+       fs_info->last_trans_committed = btrfs_super_generation(disk_super);
        return tree_root;
 
 fail_cleaner:
@@ -1891,7 +1998,36 @@ int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
        return ret;
 }
 
+int btree_lock_page_hook(struct page *page)
+{
+       struct inode *inode = page->mapping->host;
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+       struct extent_buffer *eb;
+       unsigned long len;
+       u64 bytenr = page_offset(page);
+
+       if (page->private == EXTENT_PAGE_PRIVATE)
+               goto out;
+
+       len = page->private >> 2;
+       eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
+       if (!eb)
+               goto out;
+
+       btrfs_tree_lock(eb);
+       spin_lock(&root->fs_info->hash_lock);
+       btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
+       spin_unlock(&root->fs_info->hash_lock);
+       btrfs_tree_unlock(eb);
+       free_extent_buffer(eb);
+out:
+       lock_page(page);
+       return 0;
+}
+
 static struct extent_io_ops btree_extent_io_ops = {
+       .write_cache_pages_lock_hook = btree_lock_page_hook,
        .writepage_io_hook = btree_writepage_io_hook,
        .readpage_end_io_hook = btree_readpage_end_io_hook,
        .submit_bio_hook = btree_submit_bio_hook,
This page took 0.028151 seconds and 5 git commands to generate.