Btrfs: Update on disk i_size only after pending ordered extents are done

This changes the ordered data code to update i_size after the extent
is on disk.  An on disk i_size is maintained in the in-memory btrfs inode
structures, and this is updated as extents finish.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

5 files changed, 119 insertions, 11 deletions

diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index 8d03687510e0..81c0444f37b6 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -40,11 +40,20 @@ struct btrfs_inode {
          */
         u64 last_trans;
         u64 delalloc_bytes;
+        u64 disk_i_size;
         u32 flags;
 };
+
 static inline struct btrfs_inode *BTRFS_I(struct inode *inode)
 {
         return container_of(inode, struct btrfs_inode, vfs_inode);
 }
 
+static inline void btrfs_i_size_write(struct inode *inode, u64 size)
+{
+        inode->i_size = size;
+        BTRFS_I(inode)->disk_i_size = size;
+}
+
+
 #endif
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 20928639d173..3e4e5c227c0c 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -338,7 +338,7 @@ static int noinline dirty_and_release_pages(struct btrfs_trans_handle *trans,
                 btrfs_update_inode(trans, root, inode);
         }
 failed:
-        err = btrfs_end_transaction(trans, root);
+        err = btrfs_end_transaction_throttle(trans, root);
 out_unlock:
         unlock_extent(io_tree, start_pos, end_of_last_block, GFP_NOFS);
         return err;
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 47a008c19308..baf46017d0d3 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -542,6 +542,7 @@ int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
         add_pending_csums(trans, inode, ordered_extent->file_offset,
                           &ordered_extent->list);
 
+        btrfs_ordered_update_i_size(inode, ordered_extent);
         btrfs_remove_ordered_extent(inode, ordered_extent);
         /* once for us */
         btrfs_put_ordered_extent(ordered_extent);
@@ -792,7 +793,7 @@ void btrfs_read_locked_inode(struct inode *inode)
         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
-        inode->i_size = btrfs_inode_size(leaf, inode_item);
+        btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
 
         tspec = btrfs_inode_atime(inode_item);
         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
@@ -860,7 +861,7 @@ static void fill_inode_item(struct extent_buffer *leaf,
 {
         btrfs_set_inode_uid(leaf, item, inode->i_uid);
         btrfs_set_inode_gid(leaf, item, inode->i_gid);
-        btrfs_set_inode_size(leaf, item, inode->i_size);
+        btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
         btrfs_set_inode_mode(leaf, item, inode->i_mode);
         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 
@@ -982,7 +983,7 @@ static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 err:
         btrfs_free_path(path);
         if (!ret) {
-                dir->i_size -= name_len * 2;
+                btrfs_i_size_write(dir, dir->i_size - name_len * 2);
                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
                 btrfs_update_inode(trans, root, dir);
 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
@@ -1044,7 +1045,7 @@ static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
         /* now the directory is empty */
         err = btrfs_unlink_trans(trans, root, dir, dentry);
         if (!err) {
-                inode->i_size = 0;
+                btrfs_i_size_write(inode, 0);
         }
 
         nr = trans->blocks_used;
@@ -1089,7 +1090,6 @@ static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
         int extent_type = -1;
         u64 mask = root->sectorsize - 1;
 
-        btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
         path = btrfs_alloc_path();
         path->reada = -1;
@@ -1427,7 +1427,7 @@ void btrfs_delete_inode(struct inode *inode)
                 goto no_delete;
         }
 
-        inode->i_size = 0;
+        btrfs_i_size_write(inode, 0);
         trans = btrfs_start_transaction(root, 1);
 
         btrfs_set_trans_block_group(trans, inode);
@@ -1561,6 +1561,7 @@ static int btrfs_init_locked_inode(struct inode *inode, void *p)
         inode->i_ino = args->ino;
         BTRFS_I(inode)->root = args->root;
         BTRFS_I(inode)->delalloc_bytes = 0;
+        BTRFS_I(inode)->disk_i_size = 0;
         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
                              inode->i_mapping, GFP_NOFS);
@@ -1869,6 +1870,7 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
                              inode->i_mapping, GFP_NOFS);
         mutex_init(&BTRFS_I(inode)->csum_mutex);
         BTRFS_I(inode)->delalloc_bytes = 0;
+        BTRFS_I(inode)->disk_i_size = 0;
         BTRFS_I(inode)->root = root;
 
         if (mode & S_IFDIR)
@@ -1964,7 +1966,8 @@ static int btrfs_add_link(struct btrfs_trans_handle *trans,
                                              dentry->d_parent->d_inode->i_ino);
                 }
                 parent_inode = dentry->d_parent->d_inode;
-                parent_inode->i_size += dentry->d_name.len * 2;
+                btrfs_i_size_write(parent_inode, parent_inode->i_size +
+                                   dentry->d_name.len * 2);
                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
                 ret = btrfs_update_inode(trans, root,
                                          dentry->d_parent->d_inode);
@@ -2092,6 +2095,7 @@ static int btrfs_create(struct inode *dir, struct dentry *dentry,
                                      inode->i_mapping, GFP_NOFS);
                 mutex_init(&BTRFS_I(inode)->csum_mutex);
                 BTRFS_I(inode)->delalloc_bytes = 0;
+                BTRFS_I(inode)->disk_i_size = 0;
                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
         }
         dir->i_sb->s_dirt = 1;
@@ -2199,7 +2203,7 @@ static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
         inode->i_fop = &btrfs_dir_file_operations;
         btrfs_set_trans_block_group(trans, inode);
 
-        inode->i_size = 0;
+        btrfs_i_size_write(inode, 0);
         err = btrfs_update_inode(trans, root, inode);
         if (err)
                 goto out_fail;
@@ -2756,6 +2760,7 @@ static void btrfs_truncate(struct inode *inode)
         int ret;
         struct btrfs_trans_handle *trans;
         unsigned long nr;
+        u64 mask = root->sectorsize - 1;
 
         if (!S_ISREG(inode->i_mode))
                 return;
@@ -2766,6 +2771,8 @@ static void btrfs_truncate(struct inode *inode)
 
         trans = btrfs_start_transaction(root, 1);
         btrfs_set_trans_block_group(trans, inode);
+        btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
+        btrfs_i_size_write(inode, inode->i_size);
 
         /* FIXME, add redo link to tree so we don't leak on crash */
         ret = btrfs_truncate_in_trans(trans, root, inode,
@@ -2821,7 +2828,7 @@ int btrfs_create_subvol_root(struct btrfs_root *new_root,
         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
                                      new_dirid);
         inode->i_nlink = 1;
-        inode->i_size = 0;
+        btrfs_i_size_write(inode, 0);
 
         return btrfs_update_inode(trans, new_root, inode);
 }
@@ -3069,6 +3076,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
                                      inode->i_mapping, GFP_NOFS);
                 mutex_init(&BTRFS_I(inode)->csum_mutex);
                 BTRFS_I(inode)->delalloc_bytes = 0;
+                BTRFS_I(inode)->disk_i_size = 0;
                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
         }
         dir->i_sb->s_dirt = 1;
@@ -3103,7 +3111,7 @@ static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
         inode->i_op = &btrfs_symlink_inode_operations;
         inode->i_mapping->a_ops = &btrfs_symlink_aops;
         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
-        inode->i_size = name_len - 1;
+        btrfs_i_size_write(inode, name_len - 1);
         err = btrfs_update_inode(trans, root, inode);
         if (err)
                 drop_inode = 1;
diff --git a/fs/btrfs/ordered-data.c b/fs/btrfs/ordered-data.c
index 6513270f054c..d86a953ae51d 100644
--- a/fs/btrfs/ordered-data.c
+++ b/fs/btrfs/ordered-data.c
@@ -406,3 +406,92 @@ out:
         mutex_unlock(&tree->mutex);
         return entry;
 }
+
+int btrfs_ordered_update_i_size(struct inode *inode,
+                                struct btrfs_ordered_extent *ordered)
+{
+        struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree;
+        struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+        u64 disk_i_size;
+        u64 new_i_size;
+        u64 i_size_test;
+        struct rb_node *node;
+        struct btrfs_ordered_extent *test;
+
+        mutex_lock(&tree->mutex);
+        disk_i_size = BTRFS_I(inode)->disk_i_size;
+
+        /*
+         * if the disk i_size is already at the inode->i_size, or
+         * this ordered extent is inside the disk i_size, we're done
+         */
+        if (disk_i_size >= inode->i_size ||
+            ordered->file_offset + ordered->len <= disk_i_size) {
+                goto out;
+        }
+
+        /*
+         * we can't update the disk_isize if there are delalloc bytes
+         * between disk_i_size and  this ordered extent
+         */
+        if (test_range_bit(io_tree, disk_i_size,
+                           ordered->file_offset + ordered->len - 1,
+                           EXTENT_DELALLOC, 0)) {
+                goto out;
+        }
+        /*
+         * walk backward from this ordered extent to disk_i_size.
+         * if we find an ordered extent then we can't update disk i_size
+         * yet
+         */
+        while(1) {
+                node = rb_prev(&ordered->rb_node);
+                if (!node)
+                        break;
+                test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+                if (test->file_offset + test->len <= disk_i_size)
+                        break;
+                if (test->file_offset >= inode->i_size)
+                        break;
+                if (test->file_offset >= disk_i_size)
+                        goto out;
+        }
+        new_i_size = min_t(u64, entry_end(ordered), i_size_read(inode));
+
+        /*
+         * at this point, we know we can safely update i_size to at least
+         * the offset from this ordered extent.  But, we need to
+         * walk forward and see if ios from higher up in the file have
+         * finished.
+         */
+        node = rb_next(&ordered->rb_node);
+        i_size_test = 0;
+        if (node) {
+                /*
+                 * do we have an area where IO might have finished
+                 * between our ordered extent and the next one.
+                 */
+                test = rb_entry(node, struct btrfs_ordered_extent, rb_node);
+                if (test->file_offset > entry_end(ordered)) {
+                        i_size_test = test->file_offset - 1;
+                }
+        } else {
+                i_size_test = i_size_read(inode);
+        }
+
+        /*
+         * i_size_test is the end of a region after this ordered
+         * extent where there are no ordered extents.  As long as there
+         * are no delalloc bytes in this area, it is safe to update
+         * disk_i_size to the end of the region.
+         */
+        if (i_size_test > entry_end(ordered) &&
+            !test_range_bit(io_tree, entry_end(ordered), i_size_test,
+                           EXTENT_DELALLOC, 0)) {
+                new_i_size = min_t(u64, i_size_test, i_size_read(inode));
+        }
+        BTRFS_I(inode)->disk_i_size = new_i_size;
+out:
+        mutex_unlock(&tree->mutex);
+        return 0;
+}
diff --git a/fs/btrfs/ordered-data.h b/fs/btrfs/ordered-data.h
index 33292c5fe90c..40e9126ad954 100644
--- a/fs/btrfs/ordered-data.h
+++ b/fs/btrfs/ordered-data.h
@@ -89,4 +89,6 @@ btrfs_lookup_first_ordered_extent(struct inode * inode, u64 file_offset);
 int btrfs_add_ordered_pending(struct inode *inode,
                               struct btrfs_ordered_extent *ordered,
                               u64 start, u64 len);
+int btrfs_ordered_update_i_size(struct inode *inode,
+                                struct btrfs_ordered_extent *ordered);
 #endif