f2fs: fallocate data blocks in single locked node page
authorJaegeuk Kim <jaegeuk@kernel.org>
Fri, 6 May 2016 22:30:38 +0000 (15:30 -0700)
committerJaegeuk Kim <jaegeuk@kernel.org>
Sat, 7 May 2016 17:44:55 +0000 (10:44 -0700)
This patch is to improve the expand_inode speed in fallocate by allocating
data blocks as many as possible in single locked node page.

In SSD,
 # time fallocate -l 500G $MNT/testfile

Before : 1m 33.410 s
After  : 24.758 s

Reported-by: Stephen Bates <stephen.bates@microsemi.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/file.c

index dd50f30a2f5711aac9d6fe95eafed608f13d8cf9..fb8cc27d4cc6f0806e41cb233b92368d977cd898 100644 (file)
@@ -1158,10 +1158,11 @@ static int expand_inode_data(struct inode *inode, loff_t offset,
                                        loff_t len, int mode)
 {
        struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
-       pgoff_t index, pg_start, pg_end;
+       struct f2fs_map_blocks map = { .m_next_pgofs = NULL };
+       pgoff_t pg_end;
        loff_t new_size = i_size_read(inode);
-       loff_t off_start, off_end;
-       int ret = 0;
+       loff_t off_end;
+       int ret;
 
        ret = inode_newsize_ok(inode, (len + offset));
        if (ret)
@@ -1173,43 +1174,35 @@ static int expand_inode_data(struct inode *inode, loff_t offset,
 
        f2fs_balance_fs(sbi, true);
 
-       pg_start = ((unsigned long long) offset) >> PAGE_SHIFT;
-       pg_end = ((unsigned long long) offset + len) >> PAGE_SHIFT;
-
-       off_start = offset & (PAGE_SIZE - 1);
+       pg_end = ((unsigned long long)offset + len) >> PAGE_SHIFT;
        off_end = (offset + len) & (PAGE_SIZE - 1);
 
-       f2fs_lock_op(sbi);
+       map.m_lblk = ((unsigned long long)offset) >> PAGE_SHIFT;
+       map.m_len = pg_end - map.m_lblk;
+       if (off_end)
+               map.m_len++;
 
-       for (index = pg_start; index <= pg_end; index++) {
-               struct dnode_of_data dn;
+       ret = f2fs_map_blocks(inode, &map, 1, F2FS_GET_BLOCK_PRE_AIO);
+       if (ret) {
+               pgoff_t last_off;
 
-               if (index == pg_end && !off_end)
-                       goto noalloc;
+               if (!map.m_len)
+                       return ret;
 
-               set_new_dnode(&dn, inode, NULL, NULL, 0);
-               ret = f2fs_reserve_block(&dn, index);
-               if (ret)
-                       break;
-noalloc:
-               if (pg_start == pg_end)
-                       new_size = offset + len;
-               else if (index == pg_start && off_start)
-                       new_size = (loff_t)(index + 1) << PAGE_SHIFT;
-               else if (index == pg_end)
-                       new_size = ((loff_t)index << PAGE_SHIFT) +
-                                                               off_end;
-               else
-                       new_size += PAGE_SIZE;
+               last_off = map.m_lblk + map.m_len - 1;
+
+               /* update new size to the failed position */
+               new_size = (last_off == pg_end) ? offset + len:
+                                       (loff_t)(last_off + 1) << PAGE_SHIFT;
+       } else {
+               new_size = ((loff_t)pg_end << PAGE_SHIFT) + off_end;
        }
 
-       if (!(mode & FALLOC_FL_KEEP_SIZE) &&
-               i_size_read(inode) < new_size) {
+       if (!(mode & FALLOC_FL_KEEP_SIZE) && i_size_read(inode) < new_size) {
                i_size_write(inode, new_size);
                mark_inode_dirty(inode);
                update_inode_page(inode);
        }
-       f2fs_unlock_op(sbi);
 
        return ret;
 }