Btrfs: relocate file extents in clusters
authorYan, Zheng <zheng.yan@oracle.com>
Thu, 24 Sep 2009 13:17:31 +0000 (09:17 -0400)
committerChris Mason <chris.mason@oracle.com>
Thu, 24 Sep 2009 13:17:31 +0000 (09:17 -0400)
The extent relocation code copy file extents one by one when
relocating data block group. This is inefficient if file
extents are small. This patch makes the relocation code copy
file extents in clusters. So we can can make better use of
read-ahead.

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
fs/btrfs/relocation.c

index 48a5042606355f76d27ed71dd3649befd283032a..361ad323faaceb4ed3313bd1e2904ea29ea7d2f3 100644 (file)
@@ -121,6 +121,15 @@ struct inodevec {
        int nr;
 };
 
+#define MAX_EXTENTS 128
+
+struct file_extent_cluster {
+       u64 start;
+       u64 end;
+       u64 boundary[MAX_EXTENTS];
+       unsigned int nr;
+};
+
 struct reloc_control {
        /* block group to relocate */
        struct btrfs_block_group_cache *block_group;
@@ -2529,56 +2538,94 @@ out:
 }
 
 static noinline_for_stack
-int relocate_inode_pages(struct inode *inode, u64 start, u64 len)
+int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
+                        u64 block_start)
+{
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+       struct extent_map *em;
+       int ret = 0;
+
+       em = alloc_extent_map(GFP_NOFS);
+       if (!em)
+               return -ENOMEM;
+
+       em->start = start;
+       em->len = end + 1 - start;
+       em->block_len = em->len;
+       em->block_start = block_start;
+       em->bdev = root->fs_info->fs_devices->latest_bdev;
+       set_bit(EXTENT_FLAG_PINNED, &em->flags);
+
+       lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+       while (1) {
+               write_lock(&em_tree->lock);
+               ret = add_extent_mapping(em_tree, em);
+               write_unlock(&em_tree->lock);
+               if (ret != -EEXIST) {
+                       free_extent_map(em);
+                       break;
+               }
+               btrfs_drop_extent_cache(inode, start, end, 0);
+       }
+       unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
+       return ret;
+}
+
+static int relocate_file_extent_cluster(struct inode *inode,
+                                       struct file_extent_cluster *cluster)
 {
        u64 page_start;
        u64 page_end;
-       unsigned long i;
-       unsigned long first_index;
+       u64 offset = BTRFS_I(inode)->index_cnt;
+       unsigned long index;
        unsigned long last_index;
-       unsigned int total_read = 0;
-       unsigned int total_dirty = 0;
+       unsigned int dirty_page = 0;
        struct page *page;
        struct file_ra_state *ra;
-       struct btrfs_ordered_extent *ordered;
-       struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+       int nr = 0;
        int ret = 0;
 
+       if (!cluster->nr)
+               return 0;
+
        ra = kzalloc(sizeof(*ra), GFP_NOFS);
        if (!ra)
                return -ENOMEM;
 
+       index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
+       last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
+
        mutex_lock(&inode->i_mutex);
-       first_index = start >> PAGE_CACHE_SHIFT;
-       last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
 
-       /* make sure the dirty trick played by the caller work */
-       while (1) {
-               ret = invalidate_inode_pages2_range(inode->i_mapping,
-                                                   first_index, last_index);
-               if (ret != -EBUSY)
-                       break;
-               schedule_timeout(HZ/10);
-       }
+       i_size_write(inode, cluster->end + 1 - offset);
+       ret = setup_extent_mapping(inode, cluster->start - offset,
+                                  cluster->end - offset, cluster->start);
        if (ret)
                goto out_unlock;
 
        file_ra_state_init(ra, inode->i_mapping);
 
-       for (i = first_index ; i <= last_index; i++) {
-               if (total_read % ra->ra_pages == 0) {
-                       btrfs_force_ra(inode->i_mapping, ra, NULL, i,
-                               min(last_index, ra->ra_pages + i - 1));
-               }
-               total_read++;
-again:
-               if (((u64)i << PAGE_CACHE_SHIFT) > i_size_read(inode))
-                       BUG_ON(1);
-               page = grab_cache_page(inode->i_mapping, i);
+       WARN_ON(cluster->start != cluster->boundary[0]);
+       while (index <= last_index) {
+               page = find_lock_page(inode->i_mapping, index);
                if (!page) {
-                       ret = -ENOMEM;
-                       goto out_unlock;
+                       page_cache_sync_readahead(inode->i_mapping,
+                                                 ra, NULL, index,
+                                                 last_index + 1 - index);
+                       page = grab_cache_page(inode->i_mapping, index);
+                       if (!page) {
+                               ret = -ENOMEM;
+                               goto out_unlock;
+                       }
+               }
+
+               if (PageReadahead(page)) {
+                       page_cache_async_readahead(inode->i_mapping,
+                                                  ra, NULL, page, index,
+                                                  last_index + 1 - index);
                }
+
                if (!PageUptodate(page)) {
                        btrfs_readpage(NULL, page);
                        lock_page(page);
@@ -2589,75 +2636,79 @@ again:
                                goto out_unlock;
                        }
                }
-               wait_on_page_writeback(page);
 
                page_start = (u64)page->index << PAGE_CACHE_SHIFT;
                page_end = page_start + PAGE_CACHE_SIZE - 1;
-               lock_extent(io_tree, page_start, page_end, GFP_NOFS);
-
-               ordered = btrfs_lookup_ordered_extent(inode, page_start);
-               if (ordered) {
-                       unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
-                       unlock_page(page);
-                       page_cache_release(page);
-                       btrfs_start_ordered_extent(inode, ordered, 1);
-                       btrfs_put_ordered_extent(ordered);
-                       goto again;
-               }
+
+               lock_extent(&BTRFS_I(inode)->io_tree,
+                           page_start, page_end, GFP_NOFS);
+
                set_page_extent_mapped(page);
 
-               if (i == first_index)
-                       set_extent_bits(io_tree, page_start, page_end,
+               if (nr < cluster->nr &&
+                   page_start + offset == cluster->boundary[nr]) {
+                       set_extent_bits(&BTRFS_I(inode)->io_tree,
+                                       page_start, page_end,
                                        EXTENT_BOUNDARY, GFP_NOFS);
+                       nr++;
+               }
                btrfs_set_extent_delalloc(inode, page_start, page_end);
 
                set_page_dirty(page);
-               total_dirty++;
+               dirty_page++;
 
-               unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
+               unlock_extent(&BTRFS_I(inode)->io_tree,
+                             page_start, page_end, GFP_NOFS);
                unlock_page(page);
                page_cache_release(page);
+
+               index++;
+               if (nr < cluster->nr &&
+                   page_end + 1 + offset == cluster->boundary[nr]) {
+                       balance_dirty_pages_ratelimited_nr(inode->i_mapping,
+                                                          dirty_page);
+                       dirty_page = 0;
+               }
+       }
+       if (dirty_page) {
+               balance_dirty_pages_ratelimited_nr(inode->i_mapping,
+                                                  dirty_page);
        }
+       WARN_ON(nr != cluster->nr);
 out_unlock:
        mutex_unlock(&inode->i_mutex);
        kfree(ra);
-       balance_dirty_pages_ratelimited_nr(inode->i_mapping, total_dirty);
        return ret;
 }
 
 static noinline_for_stack
-int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key)
+int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
+                        struct file_extent_cluster *cluster)
 {
-       struct btrfs_root *root = BTRFS_I(inode)->root;
-       struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
-       struct extent_map *em;
-       u64 start = extent_key->objectid - BTRFS_I(inode)->index_cnt;
-       u64 end = start + extent_key->offset - 1;
-
-       em = alloc_extent_map(GFP_NOFS);
-       em->start = start;
-       em->len = extent_key->offset;
-       em->block_len = extent_key->offset;
-       em->block_start = extent_key->objectid;
-       em->bdev = root->fs_info->fs_devices->latest_bdev;
-       set_bit(EXTENT_FLAG_PINNED, &em->flags);
+       int ret;
 
-       /* setup extent map to cheat btrfs_readpage */
-       lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
-       while (1) {
-               int ret;
-               write_lock(&em_tree->lock);
-               ret = add_extent_mapping(em_tree, em);
-               write_unlock(&em_tree->lock);
-               if (ret != -EEXIST) {
-                       free_extent_map(em);
-                       break;
-               }
-               btrfs_drop_extent_cache(inode, start, end, 0);
+       if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
+               ret = relocate_file_extent_cluster(inode, cluster);
+               if (ret)
+                       return ret;
+               cluster->nr = 0;
        }
-       unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
 
-       return relocate_inode_pages(inode, start, extent_key->offset);
+       if (!cluster->nr)
+               cluster->start = extent_key->objectid;
+       else
+               BUG_ON(cluster->nr >= MAX_EXTENTS);
+       cluster->end = extent_key->objectid + extent_key->offset - 1;
+       cluster->boundary[cluster->nr] = extent_key->objectid;
+       cluster->nr++;
+
+       if (cluster->nr >= MAX_EXTENTS) {
+               ret = relocate_file_extent_cluster(inode, cluster);
+               if (ret)
+                       return ret;
+               cluster->nr = 0;
+       }
+       return 0;
 }
 
 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
@@ -3208,6 +3259,7 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 {
        struct rb_root blocks = RB_ROOT;
        struct btrfs_key key;
+       struct file_extent_cluster *cluster;
        struct btrfs_trans_handle *trans = NULL;
        struct btrfs_path *path;
        struct btrfs_extent_item *ei;
@@ -3217,6 +3269,10 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
        int ret;
        int err = 0;
 
+       cluster = kzalloc(sizeof(*cluster), GFP_NOFS);
+       if (!cluster)
+               return -ENOMEM;
+
        path = btrfs_alloc_path();
        if (!path)
                return -ENOMEM;
@@ -3310,14 +3366,15 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
                }
 
                nr = trans->blocks_used;
-               btrfs_end_transaction_throttle(trans, rc->extent_root);
+               btrfs_end_transaction(trans, rc->extent_root);
                trans = NULL;
                btrfs_btree_balance_dirty(rc->extent_root, nr);
 
                if (rc->stage == MOVE_DATA_EXTENTS &&
                    (flags & BTRFS_EXTENT_FLAG_DATA)) {
                        rc->found_file_extent = 1;
-                       ret = relocate_data_extent(rc->data_inode, &key);
+                       ret = relocate_data_extent(rc->data_inode,
+                                                  &key, cluster);
                        if (ret < 0) {
                                err = ret;
                                break;
@@ -3332,6 +3389,14 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
                btrfs_btree_balance_dirty(rc->extent_root, nr);
        }
 
+       if (!err) {
+               ret = relocate_file_extent_cluster(rc->data_inode, cluster);
+               if (ret < 0)
+                       err = ret;
+       }
+
+       kfree(cluster);
+
        rc->create_reloc_root = 0;
        smp_mb();
 
@@ -3352,8 +3417,7 @@ static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
 }
 
 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
-                                struct btrfs_root *root,
-                                u64 objectid, u64 size)
+                                struct btrfs_root *root, u64 objectid)
 {
        struct btrfs_path *path;
        struct btrfs_inode_item *item;
@@ -3372,7 +3436,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
        item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
        memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
        btrfs_set_inode_generation(leaf, item, 1);
-       btrfs_set_inode_size(leaf, item, size);
+       btrfs_set_inode_size(leaf, item, 0);
        btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
        btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS);
        btrfs_mark_buffer_dirty(leaf);
@@ -3408,12 +3472,7 @@ static struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
        if (err)
                goto out;
 
-       err = __insert_orphan_inode(trans, root, objectid, group->key.offset);
-       BUG_ON(err);
-
-       err = btrfs_insert_file_extent(trans, root, objectid, 0, 0, 0,
-                                      group->key.offset, 0, group->key.offset,
-                                      0, 0, 0);
+       err = __insert_orphan_inode(trans, root, objectid);
        BUG_ON(err);
 
        key.objectid = objectid;
@@ -3519,10 +3578,10 @@ int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
                }
        }
 
-       filemap_fdatawrite_range(fs_info->btree_inode->i_mapping,
-                                rc->block_group->key.objectid,
-                                rc->block_group->key.objectid +
-                                rc->block_group->key.offset - 1);
+       filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
+                                    rc->block_group->key.objectid,
+                                    rc->block_group->key.objectid +
+                                    rc->block_group->key.offset - 1);
 
        WARN_ON(rc->block_group->pinned > 0);
        WARN_ON(rc->block_group->reserved > 0);