On Fri, Mar 29, 2019 at 6:03 PM Trond Myklebust <trondmy@xxxxxxxxx> wrote:
>
> When the client is reading or writing using pNFS, and hits an error
> on the DS,
Doesn't the client retry IO against the MDS when IO to the DS fails? I
find the commit message confusing. What re-tries are we talking about?
I recall after a while the client will try to get a layout again and
if it succeeds it will send the IO to the DS. So are you trying to
prevent these new retries to the DS that will fail (as you say if DS
is in unrecoverable state)? Then why would there be a fatal error
since writing thru the MDS should (hopefully) always succeed?
> then it typically sends a LAYOUTERROR and/or LAYOUTRETURN
> to the MDS, before redirtying the failed pages, and going for a new
> round of reads/writebacks. The problem is that if the server has no
> way to fix the DS, then we may need a way to interrupt this loop
> after a set number of attempts have been made.
> This patch adds an optional module parameter that allows the admin
> to specify how many times to retry the read/writeback process before
> failing with a fatal error.
> The default behaviour is to retry forever.
>
> Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> ---
> fs/nfs/direct.c | 7 +++++++
> fs/nfs/flexfilelayout/flexfilelayout.c | 8 ++++++++
> fs/nfs/pagelist.c | 14 +++++++++++++-
> fs/nfs/write.c | 5 +++++
> include/linux/nfs_page.h | 4 +++-
> 5 files changed, 36 insertions(+), 2 deletions(-)
>
> diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c
> index 2d301a1a73e2..2436bd92bc00 100644
> --- a/fs/nfs/direct.c
> +++ b/fs/nfs/direct.c
> @@ -663,6 +663,8 @@ static void nfs_direct_write_reschedule(struct nfs_direct_req *dreq)
> }
>
> list_for_each_entry_safe(req, tmp, &reqs, wb_list) {
> + /* Bump the transmission count */
> + req->wb_nio++;
> if (!nfs_pageio_add_request(&desc, req)) {
> nfs_list_move_request(req, &failed);
> spin_lock(&cinfo.inode->i_lock);
> @@ -703,6 +705,11 @@ static void nfs_direct_commit_complete(struct nfs_commit_data *data)
> req = nfs_list_entry(data->pages.next);
> nfs_list_remove_request(req);
> if (dreq->flags == NFS_ODIRECT_RESCHED_WRITES) {
> + /*
> + * Despite the reboot, the write was successful,
> + * so reset wb_nio.
> + */
> + req->wb_nio = 0;
> /* Note the rewrite will go through mds */
> nfs_mark_request_commit(req, NULL, &cinfo, 0);
> } else
> diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
> index 6673d4ff5a2a..9fdbcfd3e39d 100644
> --- a/fs/nfs/flexfilelayout/flexfilelayout.c
> +++ b/fs/nfs/flexfilelayout/flexfilelayout.c
> @@ -28,6 +28,8 @@
> #define FF_LAYOUT_POLL_RETRY_MAX (15*HZ)
> #define FF_LAYOUTRETURN_MAXERR 20
>
> +static unsigned short io_maxretrans;
> +
> static void ff_layout_read_record_layoutstats_done(struct rpc_task *task,
> struct nfs_pgio_header *hdr);
> static int ff_layout_mirror_prepare_stats(struct pnfs_layout_hdr *lo,
> @@ -925,6 +927,7 @@ ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
> pgm = &pgio->pg_mirrors[0];
> pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
>
> + pgio->pg_maxretrans = io_maxretrans;
> return;
> out_nolseg:
> if (pgio->pg_error < 0)
> @@ -992,6 +995,7 @@ ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
> pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
> }
>
> + pgio->pg_maxretrans = io_maxretrans;
> return;
>
> out_mds:
> @@ -2515,3 +2519,7 @@ MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
>
> module_init(nfs4flexfilelayout_init);
> module_exit(nfs4flexfilelayout_exit);
> +
> +module_param(io_maxretrans, ushort, 0644);
> +MODULE_PARM_DESC(io_maxretrans, "The number of times the NFSv4.1 client "
> + "retries an I/O request before returning an error. ");
> diff --git a/fs/nfs/pagelist.c b/fs/nfs/pagelist.c
> index b8301c40dd78..4a31284f411e 100644
> --- a/fs/nfs/pagelist.c
> +++ b/fs/nfs/pagelist.c
> @@ -16,8 +16,8 @@
> #include <linux/nfs.h>
> #include <linux/nfs3.h>
> #include <linux/nfs4.h>
> -#include <linux/nfs_page.h>
> #include <linux/nfs_fs.h>
> +#include <linux/nfs_page.h>
> #include <linux/nfs_mount.h>
> #include <linux/export.h>
>
> @@ -327,6 +327,7 @@ __nfs_create_request(struct nfs_lock_context *l_ctx, struct page *page,
> req->wb_bytes = count;
> req->wb_context = get_nfs_open_context(ctx);
> kref_init(&req->wb_kref);
> + req->wb_nio = 0;
> return req;
> }
>
> @@ -370,6 +371,7 @@ nfs_create_subreq(struct nfs_page *req, struct nfs_page *last,
> nfs_lock_request(ret);
> ret->wb_index = req->wb_index;
> nfs_page_group_init(ret, last);
> + ret->wb_nio = req->wb_nio;
> }
> return ret;
> }
> @@ -724,6 +726,7 @@ void nfs_pageio_init(struct nfs_pageio_descriptor *desc,
> desc->pg_mirrors_dynamic = NULL;
> desc->pg_mirrors = desc->pg_mirrors_static;
> nfs_pageio_mirror_init(&desc->pg_mirrors[0], bsize);
> + desc->pg_maxretrans = 0;
> }
>
> /**
> @@ -983,6 +986,15 @@ static int nfs_pageio_do_add_request(struct nfs_pageio_descriptor *desc,
> return 0;
> mirror->pg_base = req->wb_pgbase;
> }
> +
> + if (desc->pg_maxretrans && req->wb_nio > desc->pg_maxretrans) {
> + if (NFS_SERVER(desc->pg_inode)->flags & NFS_MOUNT_SOFTERR)
> + desc->pg_error = -ETIMEDOUT;
> + else
> + desc->pg_error = -EIO;
> + return 0;
> + }
> +
> if (!nfs_can_coalesce_requests(prev, req, desc))
> return 0;
> nfs_list_move_request(req, &mirror->pg_list);
> diff --git a/fs/nfs/write.c b/fs/nfs/write.c
> index 0712d886ff08..908b166d635d 100644
> --- a/fs/nfs/write.c
> +++ b/fs/nfs/write.c
> @@ -1009,6 +1009,8 @@ static void nfs_write_completion(struct nfs_pgio_header *hdr)
> goto remove_req;
> }
> if (nfs_write_need_commit(hdr)) {
> + /* Reset wb_nio, since the write was successful. */
> + req->wb_nio = 0;
> memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
> nfs_mark_request_commit(req, hdr->lseg, &cinfo,
> hdr->pgio_mirror_idx);
> @@ -1142,6 +1144,7 @@ static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
> req->wb_bytes = end - req->wb_offset;
> else
> req->wb_bytes = rqend - req->wb_offset;
> + req->wb_nio = 0;
> return req;
> out_flushme:
> /*
> @@ -1416,6 +1419,8 @@ static void nfs_initiate_write(struct nfs_pgio_header *hdr,
> */
> static void nfs_redirty_request(struct nfs_page *req)
> {
> + /* Bump the transmission count */
> + req->wb_nio++;
> nfs_mark_request_dirty(req);
> set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
> nfs_end_page_writeback(req);
> diff --git a/include/linux/nfs_page.h b/include/linux/nfs_page.h
> index b7d0f15615c2..8b36800d342d 100644
> --- a/include/linux/nfs_page.h
> +++ b/include/linux/nfs_page.h
> @@ -53,6 +53,7 @@ struct nfs_page {
> struct nfs_write_verifier wb_verf; /* Commit cookie */
> struct nfs_page *wb_this_page; /* list of reqs for this page */
> struct nfs_page *wb_head; /* head pointer for req list */
> + unsigned short wb_nio; /* Number of I/O attempts */
> };
>
> struct nfs_pageio_descriptor;
> @@ -87,7 +88,6 @@ struct nfs_pgio_mirror {
> };
>
> struct nfs_pageio_descriptor {
> - unsigned char pg_moreio : 1;
> struct inode *pg_inode;
> const struct nfs_pageio_ops *pg_ops;
> const struct nfs_rw_ops *pg_rw_ops;
> @@ -105,6 +105,8 @@ struct nfs_pageio_descriptor {
> struct nfs_pgio_mirror pg_mirrors_static[1];
> struct nfs_pgio_mirror *pg_mirrors_dynamic;
> u32 pg_mirror_idx; /* current mirror */
> + unsigned short pg_maxretrans;
> + unsigned char pg_moreio : 1;
> };
>
> /* arbitrarily selected limit to number of mirrors */
> --
> 2.20.1
>
