Btrfs: Wait for IO on the block device inodes of newly added devices
[linux-2.6-block.git] / fs / btrfs / disk-io.c
index 86e84a8579e32dd3dd7cfb708ac3d17a3d8ef077..45b4f7285275dd839c81cf7bf4ac0b97b39a9b4a 100644 (file)
 #include <linux/buffer_head.h> // for block_sync_page
 #include <linux/workqueue.h>
 #include <linux/kthread.h>
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
 # include <linux/freezer.h>
-#else
-# include <linux/sched.h>
-#endif
 #include "crc32c.h"
 #include "ctree.h"
 #include "disk-io.h"
@@ -40,6 +36,8 @@
 #include "print-tree.h"
 #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)
@@ -187,13 +185,6 @@ static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
        btrfs_csum_final(crc, result);
 
        if (verify) {
-               int from_this_trans = 0;
-
-               if (root->fs_info->running_transaction &&
-                   btrfs_header_generation(buf) ==
-                   root->fs_info->running_transaction->transid)
-                       from_this_trans = 1;
-
                /* FIXME, this is not good */
                if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
                        u32 val;
@@ -202,11 +193,9 @@ static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 
                        read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
                        printk("btrfs: %s checksum verify failed on %llu "
-                              "wanted %X found %X from_this_trans %d "
-                              "level %d\n",
+                              "wanted %X found %X level %d\n",
                               root->fs_info->sb->s_id,
-                              buf->start, val, found, from_this_trans,
-                              btrfs_header_level(buf));
+                              buf->start, val, found, btrfs_header_level(buf));
                        return 1;
                }
        } else {
@@ -234,8 +223,8 @@ static int verify_parent_transid(struct extent_io_tree *io_tree,
               (unsigned long long)parent_transid,
               (unsigned long long)btrfs_header_generation(eb));
        ret = 1;
-out:
        clear_extent_buffer_uptodate(io_tree, eb);
+out:
        unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
                      GFP_NOFS);
        return ret;
@@ -258,7 +247,7 @@ 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)
@@ -314,9 +303,7 @@ 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);
@@ -324,14 +311,6 @@ out:
        return 0;
 }
 
-static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
-{
-       struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
-
-       csum_dirty_buffer(root, page);
-       return 0;
-}
-
 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
                               struct extent_state *state)
 {
@@ -356,6 +335,9 @@ int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 
        found_start = btrfs_header_bytenr(eb);
        if (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;
        }
@@ -387,21 +369,11 @@ out:
        return ret;
 }
 
-#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
 static void end_workqueue_bio(struct bio *bio, int err)
-#else
-static int end_workqueue_bio(struct bio *bio,
-                                  unsigned int bytes_done, int err)
-#endif
 {
        struct end_io_wq *end_io_wq = bio->bi_private;
        struct btrfs_fs_info *fs_info;
 
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
-       if (bio->bi_size)
-               return 1;
-#endif
-
        fs_info = end_io_wq->info;
        end_io_wq->error = err;
        end_io_wq->work.func = end_workqueue_fn;
@@ -411,10 +383,6 @@ static int end_workqueue_bio(struct bio *bio,
                                   &end_io_wq->work);
        else
                btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
-
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
-       return 0;
-#endif
 }
 
 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
@@ -437,14 +405,38 @@ int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
        return 0;
 }
 
+unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
+{
+       unsigned long limit = min_t(unsigned long,
+                                   info->workers.max_workers,
+                                   info->fs_devices->open_devices);
+       return 256 * limit;
+}
+
+int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
+{
+       return atomic_read(&info->nr_async_bios) >
+               btrfs_async_submit_limit(info);
+}
+
 static void run_one_async_submit(struct btrfs_work *work)
 {
        struct btrfs_fs_info *fs_info;
        struct async_submit_bio *async;
+       int limit;
 
        async = container_of(work, struct  async_submit_bio, work);
        fs_info = BTRFS_I(async->inode)->root->fs_info;
+
+       limit = btrfs_async_submit_limit(fs_info);
+       limit = limit * 2 / 3;
+
        atomic_dec(&fs_info->nr_async_submits);
+
+       if (atomic_read(&fs_info->nr_async_submits) < limit &&
+           waitqueue_active(&fs_info->async_submit_wait))
+               wake_up(&fs_info->async_submit_wait);
+
        async->submit_bio_hook(async->inode, async->rw, async->bio,
                               async->mirror_num);
        kfree(async);
@@ -455,6 +447,7 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
                        extent_submit_bio_hook_t *submit_bio_hook)
 {
        struct async_submit_bio *async;
+       int limit = btrfs_async_submit_limit(fs_info);
 
        async = kmalloc(sizeof(*async), GFP_NOFS);
        if (!async)
@@ -467,8 +460,41 @@ int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
        async->submit_bio_hook = submit_bio_hook;
        async->work.func = run_one_async_submit;
        async->work.flags = 0;
+
+       while(atomic_read(&fs_info->async_submit_draining) &&
+             atomic_read(&fs_info->nr_async_submits)) {
+               wait_event(fs_info->async_submit_wait,
+                          (atomic_read(&fs_info->nr_async_submits) == 0));
+       }
+
        atomic_inc(&fs_info->nr_async_submits);
        btrfs_queue_worker(&fs_info->workers, &async->work);
+
+       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;
+}
+
+static int btree_csum_one_bio(struct bio *bio)
+{
+       struct bio_vec *bvec = bio->bi_io_vec;
+       int bio_index = 0;
+       struct btrfs_root *root;
+
+       WARN_ON(bio->bi_vcnt <= 0);
+       while(bio_index < bio->bi_vcnt) {
+               root = BTRFS_I(bvec->bv_page->mapping->host)->root;
+               csum_dirty_buffer(root, bvec->bv_page);
+               bio_index++;
+               bvec++;
+       }
        return 0;
 }
 
@@ -476,18 +502,16 @@ static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
                                 int mirror_num)
 {
        struct btrfs_root *root = BTRFS_I(inode)->root;
-       u64 offset;
        int ret;
 
-       offset = bio->bi_sector << 9;
-
        /*
         * when we're called for a write, we're already in the async
-        * submission context.  Just jump ingo btrfs_map_bio
+        * submission context.  Just jump into btrfs_map_bio
         */
        if (rw & (1 << BIO_RW)) {
+               btree_csum_one_bio(bio);
                return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
-                                    mirror_num, 0);
+                                    mirror_num, 1);
        }
 
        /*
@@ -519,6 +543,12 @@ static int btree_writepage(struct page *page, struct writeback_control *wbc)
 {
        struct extent_io_tree *tree;
        tree = &BTRFS_I(page->mapping->host)->io_tree;
+
+       if (current->flags & PF_MEMALLOC) {
+               redirty_page_for_writepage(wbc, page);
+               unlock_page(page);
+               return 0;
+       }
        return extent_write_full_page(tree, page, btree_get_extent, wbc);
 }
 
@@ -530,16 +560,11 @@ static int btree_writepages(struct address_space *mapping,
        if (wbc->sync_mode == WB_SYNC_NONE) {
                u64 num_dirty;
                u64 start = 0;
-               unsigned long thresh = 96 * 1024 * 1024;
+               unsigned long thresh = 32 * 1024 * 1024;
 
                if (wbc->for_kupdate)
                        return 0;
 
-               if (current_is_pdflush()) {
-                       thresh = 96 * 1024 * 1024;
-               } else {
-                       thresh = 8 * 1024 * 1024;
-               }
                num_dirty = count_range_bits(tree, &start, (u64)-1,
                                             thresh, EXTENT_DIRTY);
                if (num_dirty < thresh) {
@@ -562,6 +587,9 @@ 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;
 
@@ -663,6 +691,18 @@ 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_ALL);
+}
+
+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)
 {
@@ -681,6 +721,8 @@ 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;
 
@@ -699,15 +741,6 @@ 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,
@@ -716,6 +749,7 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
        root->node = NULL;
        root->inode = NULL;
        root->commit_root = NULL;
+       root->ref_tree = NULL;
        root->sectorsize = sectorsize;
        root->nodesize = nodesize;
        root->leafsize = leafsize;
@@ -732,8 +766,18 @@ static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
        root->in_sysfs = 0;
 
        INIT_LIST_HEAD(&root->dirty_list);
+       INIT_LIST_HEAD(&root->orphan_list);
+       INIT_LIST_HEAD(&root->dead_list);
        spin_lock_init(&root->node_lock);
+       spin_lock_init(&root->list_lock);
        mutex_init(&root->objectid_mutex);
+       mutex_init(&root->log_mutex);
+       extent_io_tree_init(&root->dirty_log_pages,
+                            fs_info->btree_inode->i_mapping, GFP_NOFS);
+
+       btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
+       root->ref_tree = &root->ref_tree_struct;
+
        memset(&root->root_key, 0, sizeof(root->root_key));
        memset(&root->root_item, 0, sizeof(root->root_item));
        memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
@@ -768,11 +812,85 @@ 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;
+       struct btrfs_root *log_root_tree = fs_info->log_root_tree;
+       u64 start = 0;
+       u64 end = 0;
+       int ret;
+
+       if (!log_root_tree)
+               return 0;
+
+       while(1) {
+               ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
+                                   0, &start, &end, EXTENT_DIRTY);
+               if (ret)
+                       break;
+
+               clear_extent_dirty(&log_root_tree->dirty_log_pages,
+                                  start, end, GFP_NOFS);
+       }
+       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,
+                                           0, BTRFS_TREE_LOG_OBJECTID,
+                                           trans->transid, 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;
@@ -822,11 +940,13 @@ 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;
 }
@@ -866,7 +986,7 @@ 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,
@@ -938,15 +1058,13 @@ static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 {
        struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
        int ret = 0;
-       int limit = 256 * info->fs_devices->open_devices;
        struct list_head *cur;
        struct btrfs_device *device;
        struct backing_dev_info *bdi;
 
        if ((bdi_bits & (1 << BDI_write_congested)) &&
-           atomic_read(&info->nr_async_submits) > limit) {
+           btrfs_congested_async(info, 0))
                return 1;
-       }
 
        list_for_each(cur, &info->fs_devices->devices) {
                device = list_entry(cur, struct btrfs_device, dev_list);
@@ -1011,9 +1129,16 @@ void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
        spin_lock(&em_tree->lock);
        em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
        spin_unlock(&em_tree->lock);
-       if (!em)
+       if (!em) {
+               __unplug_io_fn(bdi, page);
                return;
+       }
 
+       if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
+               free_extent_map(em);
+               __unplug_io_fn(bdi, page);
+               return;
+       }
        offset = offset - em->start;
        btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
                          em->block_start + offset, page);
@@ -1022,9 +1147,7 @@ void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
 
 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
 {
-#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
        bdi_init(bdi);
-#endif
        bdi->ra_pages   = default_backing_dev_info.ra_pages;
        bdi->state              = 0;
        bdi->capabilities       = default_backing_dev_info.capabilities;
@@ -1103,11 +1226,7 @@ static void end_workqueue_fn(struct btrfs_work *work)
        bio->bi_private = end_io_wq->private;
        bio->bi_end_io = end_io_wq->end_io;
        kfree(end_io_wq);
-#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
-       bio_endio(bio, bio->bi_size, error);
-#else
        bio_endio(bio, error);
-#endif
 }
 
 static int cleaner_kthread(void *arg)
@@ -1156,12 +1275,18 @@ static int transaction_kthread(void *arg)
                vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
                mutex_lock(&root->fs_info->transaction_kthread_mutex);
 
+               if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
+                       printk("btrfs: total reference cache size %Lu\n",
+                               root->fs_info->total_ref_cache_size);
+               }
+
                mutex_lock(&root->fs_info->trans_mutex);
                cur = root->fs_info->running_transaction;
                if (!cur) {
                        mutex_unlock(&root->fs_info->trans_mutex);
                        goto sleep;
                }
+
                now = get_seconds();
                if (now < cur->start_time || now - cur->start_time < 30) {
                        mutex_unlock(&root->fs_info->trans_mutex);
@@ -1198,16 +1323,18 @@ 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;
 
@@ -1221,9 +1348,11 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        INIT_LIST_HEAD(&fs_info->trans_list);
        INIT_LIST_HEAD(&fs_info->dead_roots);
        INIT_LIST_HEAD(&fs_info->hashers);
+       INIT_LIST_HEAD(&fs_info->delalloc_inodes);
        spin_lock_init(&fs_info->hash_lock);
        spin_lock_init(&fs_info->delalloc_lock);
        spin_lock_init(&fs_info->new_trans_lock);
+       spin_lock_init(&fs_info->ref_cache_lock);
 
        init_completion(&fs_info->kobj_unregister);
        fs_info->tree_root = tree_root;
@@ -1235,7 +1364,10 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        INIT_LIST_HEAD(&fs_info->space_info);
        btrfs_mapping_init(&fs_info->mapping_tree);
        atomic_set(&fs_info->nr_async_submits, 0);
+       atomic_set(&fs_info->async_submit_draining, 0);
+       atomic_set(&fs_info->nr_async_bios, 0);
        atomic_set(&fs_info->throttles, 0);
+       atomic_set(&fs_info->throttle_gen, 0);
        fs_info->sb = sb;
        fs_info->max_extent = (u64)-1;
        fs_info->max_inline = 8192 * 1024;
@@ -1245,6 +1377,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
        fs_info->btree_inode->i_nlink = 1;
        fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
 
+       INIT_LIST_HEAD(&fs_info->ordered_extents);
+       spin_lock_init(&fs_info->ordered_extent_lock);
+
        sb->s_blocksize = 4096;
        sb->s_blocksize_bits = blksize_bits(4096);
 
@@ -1265,10 +1400,9 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 
        BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
 
-       extent_io_tree_init(&fs_info->free_space_cache,
-                            fs_info->btree_inode->i_mapping, GFP_NOFS);
-       extent_io_tree_init(&fs_info->block_group_cache,
-                            fs_info->btree_inode->i_mapping, GFP_NOFS);
+       spin_lock_init(&fs_info->block_group_cache_lock);
+       fs_info->block_group_cache_tree.rb_node = NULL;
+
        extent_io_tree_init(&fs_info->pinned_extents,
                             fs_info->btree_inode->i_mapping, GFP_NOFS);
        extent_io_tree_init(&fs_info->pending_del,
@@ -1277,21 +1411,33 @@ struct btrfs_root *open_ctree(struct super_block *sb,
                             fs_info->btree_inode->i_mapping, GFP_NOFS);
        fs_info->do_barriers = 1;
 
+       extent_io_tree_init(&fs_info->reloc_mapping_tree,
+                           fs_info->btree_inode->i_mapping, GFP_NOFS);
+       INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
+       btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
+       btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
+
        BTRFS_I(fs_info->btree_inode)->root = tree_root;
        memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
               sizeof(struct btrfs_key));
        insert_inode_hash(fs_info->btree_inode);
-       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);
        mutex_init(&fs_info->transaction_kthread_mutex);
        mutex_init(&fs_info->cleaner_mutex);
        mutex_init(&fs_info->volume_mutex);
+       mutex_init(&fs_info->tree_reloc_mutex);
        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");
@@ -1328,12 +1474,39 @@ struct btrfs_root *open_ctree(struct super_block *sb,
         * queue work function gets called at interrupt time, and so it
         * cannot dynamically grow.
         */
-       btrfs_init_workers(&fs_info->workers, fs_info->thread_pool_size);
-       btrfs_init_workers(&fs_info->submit_workers, fs_info->thread_pool_size);
-       btrfs_init_workers(&fs_info->fixup_workers, 1);
-       btrfs_init_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
-       btrfs_init_workers(&fs_info->endio_write_workers,
+       btrfs_init_workers(&fs_info->workers, "worker",
                           fs_info->thread_pool_size);
+       btrfs_init_workers(&fs_info->submit_workers, "submit",
+                          min_t(u64, fs_devices->num_devices,
+                          fs_info->thread_pool_size));
+
+       /* a higher idle thresh on the submit workers makes it much more
+        * likely that bios will be send down in a sane order to the
+        * devices
+        */
+       fs_info->submit_workers.idle_thresh = 64;
+
+       /* fs_info->workers is responsible for checksumming file data
+        * blocks and metadata.  Using a larger idle thresh allows each
+        * worker thread to operate on things in roughly the order they
+        * were sent by the writeback daemons, improving overall locality
+        * of the IO going down the pipe.
+        */
+       fs_info->workers.idle_thresh = 128;
+
+       btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
+       btrfs_init_workers(&fs_info->endio_workers, "endio",
+                          fs_info->thread_pool_size);
+       btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
+                          fs_info->thread_pool_size);
+
+       /*
+        * endios are largely parallel and should have a very
+        * low idle thresh
+        */
+       fs_info->endio_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);
        btrfs_start_workers(&fs_info->fixup_workers, 1);
@@ -1445,7 +1618,30 @@ 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);
+       }
+
+       ret = btrfs_cleanup_reloc_trees(tree_root);
+       BUG_ON(ret);
 
+       fs_info->last_trans_committed = btrfs_super_generation(disk_super);
        return tree_root;
 
 fail_cleaner:
@@ -1469,9 +1665,7 @@ fail:
 
        kfree(extent_root);
        kfree(tree_root);
-#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
        bdi_destroy(&fs_info->bdi);
-#endif
        kfree(fs_info);
        return ERR_PTR(err);
 }
@@ -1687,6 +1881,11 @@ int close_ctree(struct btrfs_root *root)
                printk("btrfs: at unmount delalloc count %Lu\n",
                       fs_info->delalloc_bytes);
        }
+       if (fs_info->total_ref_cache_size) {
+               printk("btrfs: at umount reference cache size %Lu\n",
+                       fs_info->total_ref_cache_size);
+       }
+
        if (fs_info->extent_root->node)
                free_extent_buffer(fs_info->extent_root->node);
 
@@ -1700,6 +1899,7 @@ int close_ctree(struct btrfs_root *root)
                free_extent_buffer(root->fs_info->dev_root->node);
 
        btrfs_free_block_groups(root->fs_info);
+       fs_info->closing = 2;
        del_fs_roots(fs_info);
 
        filemap_write_and_wait(fs_info->btree_inode->i_mapping);
@@ -1726,9 +1926,7 @@ int close_ctree(struct btrfs_root *root)
        btrfs_close_devices(fs_info->fs_devices);
        btrfs_mapping_tree_free(&fs_info->mapping_tree);
 
-#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
        bdi_destroy(&fs_info->bdi);
-#endif
 
        kfree(fs_info->extent_root);
        kfree(fs_info->tree_root);
@@ -1783,10 +1981,10 @@ void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
        struct extent_io_tree *tree;
        u64 num_dirty;
        u64 start = 0;
-       unsigned long thresh = 16 * 1024 * 1024;
+       unsigned long thresh = 96 * 1024 * 1024;
        tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 
-       if (current_is_pdflush())
+       if (current_is_pdflush() || current->flags & PF_MEMALLOC)
                return;
 
        num_dirty = count_range_bits(tree, &start, (u64)-1,
@@ -1809,8 +2007,36 @@ 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 = {
-       .writepage_io_hook = btree_writepage_io_hook,
+       .write_cache_pages_lock_hook = btree_lock_page_hook,
        .readpage_end_io_hook = btree_readpage_end_io_hook,
        .submit_bio_hook = btree_submit_bio_hook,
        /* note we're sharing with inode.c for the merge bio hook */