On Tue, 2012-05-29 at 12:38 +0800, Peng Tao wrote:
> NFS DIO bypasses generic DIO code path. So we need to plug/unplug
> on our own.
>
> Signed-off-by: Peng Tao <tao.peng@xxxxxxx>
> ---
> fs/nfs/blocklayout/blocklayout.c | 47 +++++++++++++++++++++++++++++++++++++-
> 1 files changed, 46 insertions(+), 1 deletions(-)
>
> diff --git a/fs/nfs/blocklayout/blocklayout.c b/fs/nfs/blocklayout/blocklayout.c
> index bab432f..b48b08b 100644
> --- a/fs/nfs/blocklayout/blocklayout.c
> +++ b/fs/nfs/blocklayout/blocklayout.c
> @@ -251,6 +251,7 @@ bl_read_pagelist(struct nfs_read_data *rdata)
>
> dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
> rdata->pages.npages, f_offset, (unsigned int)rdata->args.count);
> + BUG_ON(header->dreq && header->layout_private == NULL);
>
> if (!bl_check_alignment(f_offset, rdata->args.count, PAGE_CACHE_MASK))
> goto use_mds;
> @@ -320,9 +321,17 @@ out:
> bl_put_extent(cow_read);
> bl_submit_bio(READ, bio);
> put_parallel(par);
> + if (header->dreq && !header->moreio) {
> + blk_finish_plug((struct blk_plug *)header->layout_private);
> + kfree(header->layout_private);
Wait... Why can't this just go in the block-specific ->pg_doio()
callback? Why does it need to go to your
read_pagelist()/write_pagelist()?
> + }
> return PNFS_ATTEMPTED;
>
> - use_mds:
> +use_mds:
> + if (header->dreq) {
> + blk_finish_plug((struct blk_plug *)header->layout_private);
> + kfree(header->layout_private);
> + }
> dprintk("Giving up and using normal NFS\n");
> return PNFS_NOT_ATTEMPTED;
> }
> @@ -575,6 +584,8 @@ bl_write_pagelist(struct nfs_write_data *wdata, int sync)
> unsigned long blkmask = PAGE_CACHE_MASK;
>
> dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
> + BUG_ON(header->dreq && header->layout_private == NULL);
> +
> /* Check for alignment first */
> if (header->dreq)
> blkmask = NFS_SERVER(header->inode)->pnfs_blksize - 1;
> @@ -740,8 +751,16 @@ out:
> bl_put_extent(be);
> bl_submit_bio(WRITE, bio);
> put_parallel(par);
> + if (header->dreq && !header->moreio) {
> + blk_finish_plug((struct blk_plug *)header->layout_private);
> + kfree(header->layout_private);
> + }
> return PNFS_ATTEMPTED;
> out_mds:
> + if (header->dreq) {
> + blk_finish_plug((struct blk_plug *)header->layout_private);
> + kfree(header->layout_private);
> + }
> bl_put_extent(be);
> kfree(par);
> return PNFS_NOT_ATTEMPTED;
> @@ -1017,15 +1036,30 @@ bl_clear_layoutdriver(struct nfs_server *server)
> static void
> bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
> {
> + BUG_ON(pgio->pg_layout_private != NULL);
> +
> if (!bl_check_alignment(req->wb_offset, req->wb_bytes, PAGE_CACHE_MASK))
> nfs_pageio_reset_read_mds(pgio);
> else
> pnfs_generic_pg_init_read(pgio, req);
> +
> + if (pgio->pg_dreq && pgio->pg_lseg) {
> + struct blk_plug *plug;
> + /* Freed in bl_read_pagelist */
> + plug = kmalloc(sizeof(struct blk_plug), GFP_KERNEL);
> + if (plug != NULL) {
> + blk_start_plug(plug);
> + pgio->pg_layout_private = plug;
> + } else
> + nfs_pageio_reset_read_mds(pgio);
> + }
> }
>
> static void
> bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
> {
> + BUG_ON(pgio->pg_layout_private != NULL);
> +
> if (pgio->pg_dreq) {
> /* Direct write should be blksize aligned */
> unsigned blkmask = NFS_SERVER(pgio->pg_inode)->pnfs_blksize - 1;
> @@ -1038,6 +1072,17 @@ bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
> nfs_pageio_reset_write_mds(pgio);
> else
> pnfs_generic_pg_init_write(pgio, req);
> +
> + if (pgio->pg_dreq && pgio->pg_lseg) {
> + struct blk_plug *plug;
> + /* Freed in bl_write_pagelist */
> + plug = kmalloc(sizeof(struct blk_plug), GFP_NOFS);
> + if (plug != NULL) {
> + blk_start_plug(plug);
> + pgio->pg_layout_private = plug;
> + } else
> + nfs_pageio_reset_write_mds(pgio);
> + }
> }
>
> static const struct nfs_pageio_ops bl_pg_read_ops = {
--
Trond Myklebust
Linux NFS client maintainer
NetApp
Trond.Myklebust@xxxxxxxxxx
www.netapp.com
��.n��������+%������w��{.n�����{��w��� jg��� �����ݢj����G���� ���j:+v���w�m������w�������h�����٥
