Btrfs: Add delayed allocation to the extent based page tree code
[safe/jmp/linux-2.6] / fs / btrfs / file.c
index 83836fb..d3d39e4 100644 (file)
@@ -16,7 +16,6 @@
  * Boston, MA 021110-1307, USA.
  */
 
-#include <linux/module.h>
 #include <linux/buffer_head.h>
 #include <linux/fs.h>
 #include <linux/pagemap.h>
@@ -31,6 +30,7 @@
 #include <linux/writeback.h>
 #include <linux/statfs.h>
 #include <linux/compat.h>
+#include <linux/version.h>
 #include "ctree.h"
 #include "disk-io.h"
 #include "transaction.h"
@@ -81,6 +81,55 @@ static void btrfs_drop_pages(struct page **pages, size_t num_pages)
        }
 }
 
+static int insert_inline_extent(struct btrfs_trans_handle *trans,
+                               struct btrfs_root *root, struct inode *inode,
+                               u64 offset, ssize_t size,
+                               struct page *page, size_t page_offset)
+{
+       struct btrfs_key key;
+       struct btrfs_path *path;
+       char *ptr, *kaddr;
+       struct btrfs_file_extent_item *ei;
+       u32 datasize;
+       int err = 0;
+       int ret;
+
+       path = btrfs_alloc_path();
+       if (!path)
+               return -ENOMEM;
+
+       btrfs_set_trans_block_group(trans, inode);
+
+       key.objectid = inode->i_ino;
+       key.offset = offset;
+       key.flags = 0;
+       btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
+       BUG_ON(size >= PAGE_CACHE_SIZE);
+       datasize = btrfs_file_extent_calc_inline_size(size);
+
+       ret = btrfs_insert_empty_item(trans, root, path, &key,
+                                     datasize);
+       if (ret) {
+               err = ret;
+               goto fail;
+       }
+       ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
+              path->slots[0], struct btrfs_file_extent_item);
+       btrfs_set_file_extent_generation(ei, trans->transid);
+       btrfs_set_file_extent_type(ei,
+                                  BTRFS_FILE_EXTENT_INLINE);
+       ptr = btrfs_file_extent_inline_start(ei);
+
+       kaddr = kmap_atomic(page, KM_USER0);
+       btrfs_memcpy(root, path->nodes[0]->b_data,
+                    ptr, kaddr + page_offset, size);
+       kunmap_atomic(kaddr, KM_USER0);
+       btrfs_mark_buffer_dirty(path->nodes[0]);
+fail:
+       btrfs_free_path(path);
+       return err;
+}
+
 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
                                   struct btrfs_root *root,
                                   struct file *file,
@@ -89,85 +138,133 @@ static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
                                   loff_t pos,
                                   size_t write_bytes)
 {
-       int i;
-       int offset;
        int err = 0;
-       int ret;
-       int this_write;
+       int i;
        struct inode *inode = file->f_path.dentry->d_inode;
-       struct buffer_head *bh;
-       struct btrfs_file_extent_item *ei;
+       struct extent_map *em;
+       struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+       u64 hint_block;
+       u64 num_blocks;
+       u64 start_pos;
+       u64 end_of_last_block;
+       u64 end_pos = pos + write_bytes;
+       loff_t isize = i_size_read(inode);
 
-       for (i = 0; i < num_pages; i++) {
-               offset = pos & (PAGE_CACHE_SIZE -1);
-               this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
-               /* FIXME, one block at a time */
-
-               mutex_lock(&root->fs_info->fs_mutex);
-               trans = btrfs_start_transaction(root, 1);
-               btrfs_set_trans_block_group(trans, inode);
-
-               bh = page_buffers(pages[i]);
-
-               if (buffer_mapped(bh) && bh->b_blocknr == 0) {
-                       struct btrfs_key key;
-                       struct btrfs_path *path;
-                       char *ptr, *kaddr;
-                       u32 datasize;
-
-                       /* create an inline extent, and copy the data in */
-                       path = btrfs_alloc_path();
-                       BUG_ON(!path);
-                       key.objectid = inode->i_ino;
-                       key.offset = pages[i]->index << PAGE_CACHE_SHIFT;
-                       key.flags = 0;
-                       btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
-                       BUG_ON(write_bytes >= PAGE_CACHE_SIZE);
-                       datasize = offset +
-                               btrfs_file_extent_calc_inline_size(write_bytes);
-
-                       ret = btrfs_insert_empty_item(trans, root, path, &key,
-                                                     datasize);
-                       BUG_ON(ret);
-                       ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
-                              path->slots[0], struct btrfs_file_extent_item);
-                       btrfs_set_file_extent_generation(ei, trans->transid);
-                       btrfs_set_file_extent_type(ei,
-                                                  BTRFS_FILE_EXTENT_INLINE);
-                       ptr = btrfs_file_extent_inline_start(ei);
-                       kaddr = kmap_atomic(bh->b_page, KM_USER0);
-                       btrfs_memcpy(root, path->nodes[0]->b_data,
-                                    ptr, kaddr + bh_offset(bh),
-                                    offset + write_bytes);
-                       kunmap_atomic(kaddr, KM_USER0);
-                       mark_buffer_dirty(path->nodes[0]);
-                       btrfs_free_path(path);
-               } else if (buffer_mapped(bh)) {
-                       /* csum the file data */
-                       btrfs_csum_file_block(trans, root, inode->i_ino,
-                                     pages[i]->index << PAGE_CACHE_SHIFT,
-                                     kmap(pages[i]), PAGE_CACHE_SIZE);
-                       kunmap(pages[i]);
+       em = alloc_extent_map(GFP_NOFS);
+       if (!em)
+               return -ENOMEM;
+
+       em->bdev = inode->i_sb->s_bdev;
+
+       start_pos = pos & ~((u64)root->blocksize - 1);
+       num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
+                       inode->i_blkbits;
+
+       end_of_last_block = start_pos + (num_blocks << inode->i_blkbits) - 1;
+       lock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
+       mutex_lock(&root->fs_info->fs_mutex);
+       trans = btrfs_start_transaction(root, 1);
+       if (!trans) {
+               err = -ENOMEM;
+               goto out_unlock;
+       }
+       btrfs_set_trans_block_group(trans, inode);
+       inode->i_blocks += num_blocks << 3;
+       hint_block = 0;
+
+       if ((end_of_last_block & 4095) == 0) {
+               printk("strange end of last %Lu %lu %Lu\n", start_pos, write_bytes, end_of_last_block);
+       }
+       set_extent_uptodate(em_tree, start_pos, end_of_last_block, GFP_NOFS);
+
+       /* FIXME...EIEIO, ENOSPC and more */
+
+       /* insert any holes we need to create */
+       if (inode->i_size < start_pos) {
+               u64 last_pos_in_file;
+               u64 hole_size;
+               u64 mask = root->blocksize - 1;
+               last_pos_in_file = (isize + mask) & ~mask;
+               hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
+               hole_size >>= inode->i_blkbits;
+               if (last_pos_in_file < start_pos) {
+                       err = btrfs_insert_file_extent(trans, root,
+                                                      inode->i_ino,
+                                                      last_pos_in_file,
+                                                      0, 0, hole_size);
                }
-               SetPageChecked(pages[i]);
-               ret = btrfs_end_transaction(trans, root);
-               BUG_ON(ret);
-               mutex_unlock(&root->fs_info->fs_mutex);
-
-               ret = btrfs_commit_write(file, pages[i], offset,
-                                        offset + this_write);
-               pos += this_write;
-               if (ret) {
-                       err = ret;
+               if (err)
                        goto failed;
+       }
+
+       /*
+        * either allocate an extent for the new bytes or setup the key
+        * to show we are doing inline data in the extent
+        */
+       if (isize >= PAGE_CACHE_SIZE || pos + write_bytes < inode->i_size ||
+           pos + write_bytes - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
+               u64 last_end;
+               for (i = 0; i < num_pages; i++) {
+                       struct page *p = pages[i];
+                       SetPageUptodate(p);
+                       set_page_dirty(p);
                }
-               WARN_ON(this_write > write_bytes);
-               write_bytes -= this_write;
+               last_end = pages[num_pages -1]->index << PAGE_CACHE_SHIFT;
+               last_end += PAGE_CACHE_SIZE - 1;
+               set_extent_delalloc(em_tree, start_pos, end_of_last_block,
+                                GFP_NOFS);
+       } else {
+               struct page *p = pages[0];
+               /* step one, delete the existing extents in this range */
+               /* FIXME blocksize != pagesize */
+               if (start_pos < inode->i_size) {
+                       err = btrfs_drop_extents(trans, root, inode, start_pos,
+                                (pos + write_bytes + root->blocksize -1) &
+                                ~((u64)root->blocksize - 1), &hint_block);
+                       if (err)
+                               goto failed;
+               }
+
+               err = insert_inline_extent(trans, root, inode, start_pos,
+                                          end_pos - start_pos, p, 0);
+               BUG_ON(err);
+               em->start = start_pos;
+               em->end = end_pos;
+               em->block_start = EXTENT_MAP_INLINE;
+               em->block_end = EXTENT_MAP_INLINE;
+               add_extent_mapping(em_tree, em);
+       }
+       if (end_pos > isize) {
+               i_size_write(inode, end_pos);
+               btrfs_update_inode(trans, root, inode);
        }
 failed:
+       err = btrfs_end_transaction(trans, root);
+out_unlock:
+       mutex_unlock(&root->fs_info->fs_mutex);
+       unlock_extent(em_tree, start_pos, end_of_last_block, GFP_NOFS);
+       free_extent_map(em);
        return err;
 }
 
+int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end)
+{
+       struct extent_map *em;
+       struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
+
+       while(1) {
+               em = lookup_extent_mapping(em_tree, start, end);
+               if (!em)
+                       break;
+               remove_extent_mapping(em_tree, em);
+               /* once for us */
+               free_extent_map(em);
+               /* once for the tree*/
+               free_extent_map(em);
+       }
+       return 0;
+}
+
 /*
  * this is very complex, but the basic idea is to drop all extents
  * in the range start - end.  hint_block is filled in with a block number
@@ -195,11 +292,15 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
        int found_type;
        int found_extent;
        int found_inline;
+       int recow;
+
+       btrfs_drop_extent_cache(inode, start, end - 1);
 
        path = btrfs_alloc_path();
        if (!path)
                return -ENOMEM;
        while(1) {
+               recow = 0;
                btrfs_release_path(root, path);
                ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
                                               search_start, -1);
@@ -212,6 +313,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
                        }
                        path->slots[0]--;
                }
+next_slot:
                keep = 0;
                bookend = 0;
                found_extent = 0;
@@ -219,39 +321,53 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
                extent = NULL;
                leaf = btrfs_buffer_leaf(path->nodes[0]);
                slot = path->slots[0];
+               ret = 0;
                btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
                if (key.offset >= end || key.objectid != inode->i_ino) {
-                       ret = 0;
                        goto out;
                }
-               if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) {
-                       ret = 0;
+               if (btrfs_key_type(&key) > BTRFS_EXTENT_DATA_KEY) {
                        goto out;
                }
-               extent = btrfs_item_ptr(leaf, slot,
-                                       struct btrfs_file_extent_item);
-               found_type = btrfs_file_extent_type(extent);
-               if (found_type == BTRFS_FILE_EXTENT_REG) {
-                       extent_end = key.offset +
-                               (btrfs_file_extent_num_blocks(extent) <<
-                                inode->i_blkbits);
-                       found_extent = 1;
-               } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
-                       found_inline = 1;
-                       extent_end = key.offset +
-                            btrfs_file_extent_inline_len(leaf->items + slot);
+               if (recow) {
+                       search_start = key.offset;
+                       continue;
                }
-
-               /* we found nothing we can drop */
-               if (!found_extent && !found_inline) {
-                       ret = 0;
-                       goto out;
+               if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
+                       extent = btrfs_item_ptr(leaf, slot,
+                                               struct btrfs_file_extent_item);
+                       found_type = btrfs_file_extent_type(extent);
+                       if (found_type == BTRFS_FILE_EXTENT_REG) {
+                               extent_end = key.offset +
+                                       (btrfs_file_extent_num_blocks(extent) <<
+                                        inode->i_blkbits);
+                               found_extent = 1;
+                       } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
+                               found_inline = 1;
+                               extent_end = key.offset +
+                                    btrfs_file_extent_inline_len(leaf->items +
+                                                                 slot);
+                       }
+               } else {
+                       extent_end = search_start;
                }
 
-               /* we found nothing inside the range */
-               if (search_start >= extent_end) {
-                       ret = 0;
-                       goto out;
+               /* we found nothing we can drop */
+               if ((!found_extent && !found_inline) ||
+                   search_start >= extent_end) {
+                       int nextret;
+                       u32 nritems;
+                       nritems = btrfs_header_nritems(
+                                       btrfs_buffer_header(path->nodes[0]));
+                       if (slot >= nritems - 1) {
+                               nextret = btrfs_next_leaf(root, path);
+                               if (nextret)
+                                       goto out;
+                               recow = 1;
+                       } else {
+                               path->slots[0]++;
+                       }
+                       goto next_slot;
                }
 
                /* FIXME, there's only one inline extent allowed right now */
@@ -277,7 +393,6 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
                        WARN_ON(found_inline);
                        bookend = 1;
                }
-
                /* truncate existing extent */
                if (start > key.offset) {
                        u64 new_num;
@@ -296,7 +411,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
                                }
                                btrfs_set_file_extent_num_blocks(extent,
                                                                 new_num);
-                               mark_buffer_dirty(path->nodes[0]);
+                               btrfs_mark_buffer_dirty(path->nodes[0]);
                        } else {
                                WARN_ON(1);
                        }
@@ -317,6 +432,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
                                        btrfs_file_extent_disk_blocknr(extent);
                        }
                        ret = btrfs_del_item(trans, root, path);
+                       /* TODO update progress marker and return */
                        BUG_ON(ret);
                        btrfs_release_path(root, path);
                        extent = NULL;
@@ -342,10 +458,14 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,
                        ins.offset = end;
                        ins.flags = 0;
                        btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
-
                        btrfs_release_path(root, path);
                        ret = btrfs_insert_empty_item(trans, root, path, &ins,
                                                      sizeof(*extent));
+
+                       if (ret) {
+                               btrfs_print_leaf(root, btrfs_buffer_leaf(path->nodes[0]));
+                               printk("got %d on inserting %Lu %u %Lu start %Lu end %Lu found %Lu %Lu\n", ret , ins.objectid, ins.flags, ins.offset, start, end, key.offset, extent_end);
+                       }
                        BUG_ON(ret);
                        extent = btrfs_item_ptr(
                                    btrfs_buffer_leaf(path->nodes[0]),
@@ -392,18 +512,18 @@ static int prepare_pages(struct btrfs_root *root,
                         loff_t pos,
                         unsigned long first_index,
                         unsigned long last_index,
-                        size_t write_bytes,
-                        u64 alloc_extent_start)
+                        size_t write_bytes)
 {
        int i;
        unsigned long index = pos >> PAGE_CACHE_SHIFT;
        struct inode *inode = file->f_path.dentry->d_inode;
-       int offset;
        int err = 0;
-       int this_write;
-       struct buffer_head *bh;
-       struct buffer_head *head;
-       loff_t isize = i_size_read(inode);
+       u64 num_blocks;
+       u64 start_pos;
+
+       start_pos = pos & ~((u64)root->blocksize - 1);
+       num_blocks = (write_bytes + pos - start_pos + root->blocksize - 1) >>
+                       inode->i_blkbits;
 
        memset(pages, 0, num_pages * sizeof(struct page *));
 
@@ -411,44 +531,19 @@ static int prepare_pages(struct btrfs_root *root,
                pages[i] = grab_cache_page(inode->i_mapping, index + i);
                if (!pages[i]) {
                        err = -ENOMEM;
-                       goto failed_release;
+                       BUG_ON(1);
                }
                cancel_dirty_page(pages[i], PAGE_CACHE_SIZE);
                wait_on_page_writeback(pages[i]);
-               offset = pos & (PAGE_CACHE_SIZE -1);
-               this_write = min((size_t)PAGE_CACHE_SIZE - offset, write_bytes);
-               if (!page_has_buffers(pages[i])) {
-                       create_empty_buffers(pages[i],
-                                            root->fs_info->sb->s_blocksize,
-                                            (1 << BH_Uptodate));
+               if (!PagePrivate(pages[i])) {
+                       SetPagePrivate(pages[i]);
+                       set_page_private(pages[i], 1);
+                       WARN_ON(!pages[i]->mapping->a_ops->invalidatepage);
+                       page_cache_get(pages[i]);
                }
-               head = page_buffers(pages[i]);
-               bh = head;
-               do {
-                       err = btrfs_map_bh_to_logical(root, bh,
-                                                     alloc_extent_start);
-                       BUG_ON(err);
-                       if (err)
-                               goto failed_truncate;
-                       bh = bh->b_this_page;
-                       if (alloc_extent_start)
-                               alloc_extent_start++;
-               } while (bh != head);
-               pos += this_write;
-               WARN_ON(this_write > write_bytes);
-               write_bytes -= this_write;
+               WARN_ON(!PageLocked(pages[i]));
        }
        return 0;
-
-failed_release:
-       btrfs_drop_pages(pages, num_pages);
-       return err;
-
-failed_truncate:
-       btrfs_drop_pages(pages, num_pages);
-       if (pos > isize)
-               vmtruncate(inode, isize);
-       return err;
 }
 
 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
@@ -460,16 +555,14 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
        int ret = 0;
        struct inode *inode = file->f_path.dentry->d_inode;
        struct btrfs_root *root = BTRFS_I(inode)->root;
-       struct page *pages[8];
+       struct page **pages = NULL;
+       int nrptrs;
        struct page *pinned[2];
        unsigned long first_index;
        unsigned long last_index;
-       u64 start_pos;
-       u64 num_blocks;
-       u64 alloc_extent_start;
-       u64 hint_block;
-       struct btrfs_trans_handle *trans;
-       struct btrfs_key ins;
+
+       nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
+                    PAGE_CACHE_SIZE / (sizeof(struct page *)));
        pinned[0] = NULL;
        pinned[1] = NULL;
        if (file->f_flags & O_DIRECT)
@@ -487,9 +580,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
                goto out;
        file_update_time(file);
 
-       start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
-       num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
-                       inode->i_blkbits;
+       pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 
        mutex_lock(&inode->i_mutex);
        first_index = pos >> PAGE_CACHE_SHIFT;
@@ -503,7 +594,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
        if ((pos & (PAGE_CACHE_SIZE - 1))) {
                pinned[0] = grab_cache_page(inode->i_mapping, first_index);
                if (!PageUptodate(pinned[0])) {
-                       ret = mpage_readpage(pinned[0], btrfs_get_block);
+                       ret = btrfs_readpage(NULL, pinned[0]);
                        BUG_ON(ret);
                        wait_on_page_locked(pinned[0]);
                } else {
@@ -513,7 +604,7 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
        if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
                pinned[1] = grab_cache_page(inode->i_mapping, last_index);
                if (!PageUptodate(pinned[1])) {
-                       ret = mpage_readpage(pinned[1], btrfs_get_block);
+                       ret = btrfs_readpage(NULL, pinned[1]);
                        BUG_ON(ret);
                        wait_on_page_locked(pinned[1]);
                } else {
@@ -521,280 +612,117 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
                }
        }
 
-       mutex_lock(&root->fs_info->fs_mutex);
-       trans = btrfs_start_transaction(root, 1);
-       if (!trans) {
-               err = -ENOMEM;
-               mutex_unlock(&root->fs_info->fs_mutex);
-               goto out_unlock;
-       }
-       btrfs_set_trans_block_group(trans, inode);
-       /* FIXME blocksize != 4096 */
-       inode->i_blocks += num_blocks << 3;
-       hint_block = 0;
-
-       /* FIXME...EIEIO, ENOSPC and more */
-
-       /* step one, delete the existing extents in this range */
-       if (start_pos < inode->i_size) {
-               /* FIXME blocksize != pagesize */
-               ret = btrfs_drop_extents(trans, root, inode,
-                                        start_pos,
-                                        (pos + count + root->blocksize -1) &
-                                        ~((u64)root->blocksize - 1),
-                                        &hint_block);
-               BUG_ON(ret);
-       }
-
-       /* insert any holes we need to create */
-       if (inode->i_size < start_pos) {
-               u64 last_pos_in_file;
-               u64 hole_size;
-               u64 mask = root->blocksize - 1;
-               last_pos_in_file = (inode->i_size + mask) & ~mask;
-               hole_size = (start_pos - last_pos_in_file + mask) & ~mask;
-               hole_size >>= inode->i_blkbits;
-               if (last_pos_in_file < start_pos) {
-                       ret = btrfs_insert_file_extent(trans, root,
-                                                      inode->i_ino,
-                                                      last_pos_in_file,
-                                                      0, 0, hole_size);
-               }
-               BUG_ON(ret);
-       }
-
-       /*
-        * either allocate an extent for the new bytes or setup the key
-        * to show we are doing inline data in the extent
-        */
-       if (inode->i_size >= PAGE_CACHE_SIZE || pos + count < inode->i_size ||
-           pos + count - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
-               ret = btrfs_alloc_extent(trans, root, inode->i_ino,
-                                        num_blocks, hint_block, (u64)-1,
-                                        &ins, 1);
-               BUG_ON(ret);
-               ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
-                                      start_pos, ins.objectid, ins.offset,
-                                      ins.offset);
-               BUG_ON(ret);
-       } else {
-               ins.offset = 0;
-               ins.objectid = 0;
-       }
-       BUG_ON(ret);
-       alloc_extent_start = ins.objectid;
-       ret = btrfs_end_transaction(trans, root);
-       mutex_unlock(&root->fs_info->fs_mutex);
-
        while(count > 0) {
                size_t offset = pos & (PAGE_CACHE_SIZE - 1);
-               size_t write_bytes = min(count,
-                                        (size_t)PAGE_CACHE_SIZE - offset);
+               size_t write_bytes = min(count, nrptrs *
+                                       (size_t)PAGE_CACHE_SIZE -
+                                        offset);
                size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
                                        PAGE_CACHE_SHIFT;
 
+               WARN_ON(num_pages > nrptrs);
                memset(pages, 0, sizeof(pages));
                ret = prepare_pages(root, file, pages, num_pages,
                                    pos, first_index, last_index,
-                                   write_bytes, alloc_extent_start);
-               BUG_ON(ret);
+                                   write_bytes);
+               if (ret)
+                       goto out;
 
-               /* FIXME blocks != pagesize */
-               if (alloc_extent_start)
-                       alloc_extent_start += num_pages;
                ret = btrfs_copy_from_user(pos, num_pages,
                                           write_bytes, pages, buf);
-               BUG_ON(ret);
+               if (ret) {
+                       btrfs_drop_pages(pages, num_pages);
+                       goto out;
+               }
 
                ret = dirty_and_release_pages(NULL, root, file, pages,
                                              num_pages, pos, write_bytes);
-               BUG_ON(ret);
                btrfs_drop_pages(pages, num_pages);
+               if (ret)
+                       goto out;
 
                buf += write_bytes;
                count -= write_bytes;
                pos += write_bytes;
                num_written += write_bytes;
 
-               balance_dirty_pages_ratelimited(inode->i_mapping);
+               balance_dirty_pages_ratelimited_nr(inode->i_mapping, num_pages);
                btrfs_btree_balance_dirty(root);
                cond_resched();
        }
-out_unlock:
        mutex_unlock(&inode->i_mutex);
 out:
+       kfree(pages);
        if (pinned[0])
                page_cache_release(pinned[0]);
        if (pinned[1])
                page_cache_release(pinned[1]);
        *ppos = pos;
        current->backing_dev_info = NULL;
-       mark_inode_dirty(inode);
        return num_written ? num_written : err;
 }
 
-/*
- * FIXME, do this by stuffing the csum we want in the info hanging off
- * page->private.  For now, verify file csums on read
- */
-static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
-                       unsigned long offset, unsigned long size)
-{
-       char *kaddr;
-       unsigned long left, count = desc->count;
-       struct inode *inode = page->mapping->host;
-
-       if (size > count)
-               size = count;
-
-       if (!PageChecked(page)) {
-               /* FIXME, do it per block */
-               struct btrfs_root *root = BTRFS_I(inode)->root;
-               int ret;
-               struct buffer_head *bh;
-
-               if (page_has_buffers(page)) {
-                       bh = page_buffers(page);
-                       if (!buffer_mapped(bh)) {
-                               SetPageChecked(page);
-                               goto checked;
-                       }
-               }
-
-               ret = btrfs_csum_verify_file_block(root,
-                                 page->mapping->host->i_ino,
-                                 page->index << PAGE_CACHE_SHIFT,
-                                 kmap(page), PAGE_CACHE_SIZE);
-               if (ret) {
-                       if (ret != -ENOENT) {
-                               printk("failed to verify ino %lu page %lu ret %d\n",
-                                      page->mapping->host->i_ino,
-                                      page->index, ret);
-                               memset(page_address(page), 1, PAGE_CACHE_SIZE);
-                               flush_dcache_page(page);
-                       }
-               }
-               SetPageChecked(page);
-               kunmap(page);
-       }
-checked:
-       /*
-        * Faults on the destination of a read are common, so do it before
-        * taking the kmap.
-        */
-       if (!fault_in_pages_writeable(desc->arg.buf, size)) {
-               kaddr = kmap_atomic(page, KM_USER0);
-               left = __copy_to_user_inatomic(desc->arg.buf,
-                                               kaddr + offset, size);
-               kunmap_atomic(kaddr, KM_USER0);
-               if (left == 0)
-                       goto success;
-       }
-
-       /* Do it the slow way */
-       kaddr = kmap(page);
-       left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
-       kunmap(page);
-
-       if (left) {
-               size -= left;
-               desc->error = -EFAULT;
-       }
-success:
-       desc->count = count - size;
-       desc->written += size;
-       desc->arg.buf += size;
-       return size;
-}
-
-/**
- * btrfs_file_aio_read - filesystem read routine, with a mod to csum verify
- * @iocb:      kernel I/O control block
- * @iov:       io vector request
- * @nr_segs:   number of segments in the iovec
- * @pos:       current file position
- */
-static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
-                                  unsigned long nr_segs, loff_t pos)
-{
-       struct file *filp = iocb->ki_filp;
-       ssize_t retval;
-       unsigned long seg;
-       size_t count;
-       loff_t *ppos = &iocb->ki_pos;
-
-       count = 0;
-       for (seg = 0; seg < nr_segs; seg++) {
-               const struct iovec *iv = &iov[seg];
-
-               /*
-                * If any segment has a negative length, or the cumulative
-                * length ever wraps negative then return -EINVAL.
-                */
-               count += iv->iov_len;
-               if (unlikely((ssize_t)(count|iv->iov_len) < 0))
-                       return -EINVAL;
-               if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
-                       continue;
-               if (seg == 0)
-                       return -EFAULT;
-               nr_segs = seg;
-               count -= iv->iov_len;   /* This segment is no good */
-               break;
-       }
-       retval = 0;
-       if (count) {
-               for (seg = 0; seg < nr_segs; seg++) {
-                       read_descriptor_t desc;
-
-                       desc.written = 0;
-                       desc.arg.buf = iov[seg].iov_base;
-                       desc.count = iov[seg].iov_len;
-                       if (desc.count == 0)
-                               continue;
-                       desc.error = 0;
-                       do_generic_file_read(filp, ppos, &desc,
-                                            btrfs_read_actor);
-                       retval += desc.written;
-                       if (desc.error) {
-                               retval = retval ?: desc.error;
-                               break;
-                       }
-               }
-       }
-       return retval;
-}
-
 static int btrfs_sync_file(struct file *file,
                           struct dentry *dentry, int datasync)
 {
        struct inode *inode = dentry->d_inode;
        struct btrfs_root *root = BTRFS_I(inode)->root;
-       int ret;
+       int ret = 0;
        struct btrfs_trans_handle *trans;
 
        /*
-        * FIXME, use inode generation number to check if we can skip the
-        * commit
+        * check the transaction that last modified this inode
+        * and see if its already been committed
         */
        mutex_lock(&root->fs_info->fs_mutex);
+       if (!BTRFS_I(inode)->last_trans)
+               goto out;
+       mutex_lock(&root->fs_info->trans_mutex);
+       if (BTRFS_I(inode)->last_trans <=
+           root->fs_info->last_trans_committed) {
+               BTRFS_I(inode)->last_trans = 0;
+               mutex_unlock(&root->fs_info->trans_mutex);
+               goto out;
+       }
+       mutex_unlock(&root->fs_info->trans_mutex);
+
+       /*
+        * ok we haven't committed the transaction yet, lets do a commit
+        */
        trans = btrfs_start_transaction(root, 1);
        if (!trans) {
                ret = -ENOMEM;
                goto out;
        }
        ret = btrfs_commit_transaction(trans, root);
-       mutex_unlock(&root->fs_info->fs_mutex);
 out:
+       mutex_unlock(&root->fs_info->fs_mutex);
        return ret > 0 ? EIO : ret;
 }
 
+static struct vm_operations_struct btrfs_file_vm_ops = {
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
+       .nopage         = filemap_nopage,
+       .populate       = filemap_populate,
+#else
+       .fault          = filemap_fault,
+#endif
+       .page_mkwrite   = btrfs_page_mkwrite,
+};
+
+static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+       vma->vm_ops = &btrfs_file_vm_ops;
+       file_accessed(filp);
+       return 0;
+}
+
 struct file_operations btrfs_file_operations = {
        .llseek         = generic_file_llseek,
        .read           = do_sync_read,
-       .aio_read       = btrfs_file_aio_read,
+       .aio_read       = generic_file_aio_read,
        .write          = btrfs_file_write,
-       .mmap           = generic_file_mmap,
+       .mmap           = btrfs_file_mmap,
        .open           = generic_file_open,
        .ioctl          = btrfs_ioctl,
        .fsync          = btrfs_sync_file,