On Mon, 2017-05-22 at 05:59 -0400, Martin Brandenburg wrote:
> With this and the last commit, OrangeFS is capable of writing through
> the page cache. This should significantly increase performance of very
> small writes, since writeback will not be done for every write call.
>
> However it is not appropriate for use with multiple clients due to the
> long writeback delay.
>
> Signed-off-by: Martin Brandenburg <martin@xxxxxxxxxxxx>
> ---
> fs/orangefs/file.c | 128 +++++++++++++++++++++++------------------------------
> 1 file changed, 56 insertions(+), 72 deletions(-)
>
> diff --git a/fs/orangefs/file.c b/fs/orangefs/file.c
> index c03deea..3ab0a1f 100644
> --- a/fs/orangefs/file.c
> +++ b/fs/orangefs/file.c
> @@ -448,69 +448,11 @@ static ssize_t orangefs_file_read_iter(struct kiocb *iocb,
> return generic_file_read_iter(iocb, iter);
> }
>
> -static ssize_t orangefs_file_write_iter(struct kiocb *iocb, struct iov_iter *iter)
> +static ssize_t orangefs_file_write_iter(struct kiocb *iocb,
> + struct iov_iter *iter)
> {
> - struct file *file = iocb->ki_filp;
> - loff_t pos;
> - ssize_t rc;
> -
> - BUG_ON(iocb->private);
> -
> - gossip_debug(GOSSIP_FILE_DEBUG, "orangefs_file_write_iter\n");
> -
> - inode_lock(file->f_mapping->host);
> -
> - /* Make sure generic_write_checks sees an up to date inode size. */
> - if (file->f_flags & O_APPEND) {
> - rc = orangefs_inode_getattr(file->f_mapping->host, 0, 1,
> - STATX_SIZE);
> - if (rc == -ESTALE)
> - rc = -EIO;
> - if (rc) {
> - gossip_err("%s: orangefs_inode_getattr failed, "
> - "rc:%zd:.\n", __func__, rc);
> - goto out;
> - }
> - }
> -
> - if (file->f_pos > i_size_read(file->f_mapping->host))
> - orangefs_i_size_write(file->f_mapping->host, file->f_pos);
> -
> - rc = generic_write_checks(iocb, iter);
> -
> - if (rc <= 0) {
> - gossip_err("%s: generic_write_checks failed, rc:%zd:.\n",
> - __func__, rc);
> - goto out;
> - }
> -
> - /*
> - * if we are appending, generic_write_checks would have updated
> - * pos to the end of the file, so we will wait till now to set
> - * pos...
> - */
> - pos = *(&iocb->ki_pos);
> -
> - rc = do_readv_writev(ORANGEFS_IO_WRITE,
> - file,
> - &pos,
> - iter);
> - if (rc < 0) {
> - gossip_err("%s: do_readv_writev failed, rc:%zd:.\n",
> - __func__, rc);
> - goto out;
> - }
> -
> - iocb->ki_pos = pos;
> orangefs_stats.writes++;
> -
> - if (pos > i_size_read(file->f_mapping->host))
> - orangefs_i_size_write(file->f_mapping->host, pos);
> -
> -out:
> -
> - inode_unlock(file->f_mapping->host);
> - return rc;
> + return generic_file_write_iter(iocb, iter);
> }
>
> /*
> @@ -606,9 +548,8 @@ static int orangefs_file_release(struct inode *inode, struct file *file)
> orangefs_flush_inode(inode);
>
> /*
> - * remove all associated inode pages from the page cache and
> - * readahead cache (if any); this forces an expensive refresh of
> - * data for the next caller of mmap (or 'get_block' accesses)
> + * remove all associated inode pages from the readahead cache
> + * (if any)
> */
> if (file_inode(file) &&
> file_inode(file)->i_mapping &&
> @@ -621,8 +562,6 @@ static int orangefs_file_release(struct inode *inode, struct file *file)
> gossip_debug(GOSSIP_INODE_DEBUG,
> "flush_racache finished\n");
> }
> - truncate_inode_pages(file_inode(file)->i_mapping,
> - 0);
> }
> return 0;
> }
> @@ -741,6 +680,40 @@ const struct file_operations orangefs_file_operations = {
> .fsync = orangefs_fsync,
> };
>
> +static int orangefs_writepage(struct page *page,
> + struct writeback_control *wbc)
> +{
> + struct inode *inode = page->mapping->host;
> + struct iov_iter iter;
> + struct iovec iov;
> + loff_t off;
> + size_t len;
> + ssize_t r;
> + void *map;
> +
> + off = page_offset(page);
> + len = i_size_read(inode);
> + if (off + PAGE_SIZE > len)
> + len = len - off;
> + else
> + len = PAGE_SIZE;
> +
> + map = kmap_atomic(page);
> + iov.iov_base = map;
> + iov.iov_len = len;
> + iov_iter_init(&iter, WRITE, &iov, 1, len);
> +
> + set_page_writeback(page);
> +
wait_for_direct_io can sleep, so you can't use kmap_atomic there.
Regular old kmap should be ok there though.
Also, you probably really don't want to use iovecs there, as those are
expected to deal in userland addresses and the kmap address won't be
one.
It may be cleaner to use a bio_vec there instead. You most likely
wouldn't need to kmap at all if you did that.
> + r = wait_for_direct_io(ORANGEFS_IO_WRITE, inode, &off, &iter,
> + len, 0);
> + kunmap_atomic(map);
> +
When writeback fails for some reason, you'll also want to call
mapping_set_error to help ensure that those errors get reported (unless
you're tracking them on your own somehow). I don't see where that's
being done in a cursory glance at these patches, but I could have missed
it.
> + end_page_writeback(page);
> + unlock_page(page);
> + return 0;
> +}
> +
> static int orangefs_readpage(struct file *file, struct page *page)
> {
> int ret;
> @@ -786,6 +759,17 @@ static int orangefs_readpage(struct file *file, struct page *page)
> return ret;
> }
>
> +static int orangefs_write_end(struct file *file,
> + struct address_space *mapping, loff_t pos, unsigned len,
> + unsigned copied, struct page *page, void *fsdata)
> +{
> + int r;
> + r = simple_write_end(file, mapping, pos, len, copied, page,
> + fsdata);
> + mark_inode_dirty_sync(file_inode(file));
> + return r;
> +}
> +
> static void orangefs_invalidatepage(struct page *page,
> unsigned int offset,
> unsigned int length)
> @@ -815,17 +799,17 @@ static ssize_t orangefs_direct_IO(struct kiocb *iocb,
> {
> struct file *file = iocb->ki_filp;
> loff_t pos = *(&iocb->ki_pos);
> - /*
> - * This cannot happen until write_iter becomes
> - * generic_file_write_iter.
> - */
> - BUG_ON(iov_iter_rw(iter) != READ);
> - return do_readv_writev(ORANGEFS_IO_READ, file, &pos, iter);
> + return do_readv_writev(iocb->ki_flags & IOCB_WRITE ?
> + ORANGEFS_IO_WRITE : ORANGEFS_IO_READ, file, &pos, iter);
> }
>
> /** ORANGEFS2 implementation of address space operations */
> const struct address_space_operations orangefs_address_operations = {
> + .writepage = orangefs_writepage,
> .readpage = orangefs_readpage,
> + .set_page_dirty = __set_page_dirty_nobuffers,
> + .write_begin = simple_write_begin,
> + .write_end = orangefs_write_end,
> .invalidatepage = orangefs_invalidatepage,
> .releasepage = orangefs_releasepage,
> .direct_IO = orangefs_direct_IO,
--
Jeff Layton <jlayton@xxxxxxxxxx>
