pnfs/flexfiles: Add the FlexFile Layout Driver
authorTom Haynes <loghyr@primarydata.com>
Thu, 11 Dec 2014 22:02:04 +0000 (17:02 -0500)
committerTom Haynes <loghyr@primarydata.com>
Tue, 3 Feb 2015 19:06:52 +0000 (11:06 -0800)
The flexfile layout is a new layout that extends the
file layout. It is currently being drafted as a specification at
https://datatracker.ietf.org/doc/draft-ietf-nfsv4-layout-types/

Signed-off-by: Weston Andros Adamson <dros@primarydata.com>
Signed-off-by: Tom Haynes <loghyr@primarydata.com>
Signed-off-by: Tao Peng <bergwolf@primarydata.com>
13 files changed:
fs/nfs/Kconfig
fs/nfs/Makefile
fs/nfs/flexfilelayout/Makefile [new file with mode: 0644]
fs/nfs/flexfilelayout/flexfilelayout.c [new file with mode: 0644]
fs/nfs/flexfilelayout/flexfilelayout.h [new file with mode: 0644]
fs/nfs/flexfilelayout/flexfilelayoutdev.c [new file with mode: 0644]
fs/nfs/idmap.c
fs/nfs/nfs4proc.c
fs/nfs/pnfs.c
fs/nfs/pnfs.h
include/linux/nfs4.h
include/linux/nfs_idmap.h
include/linux/sunrpc/metrics.h

index 3dece03f2fc8a4031db1bca96b9a8365bafd0312..c7abc10279af3382eda38e2f2fa2591a834b35be 100644 (file)
@@ -128,6 +128,11 @@ config PNFS_OBJLAYOUT
        depends on NFS_V4_1 && SCSI_OSD_ULD
        default NFS_V4
 
+config PNFS_FLEXFILE_LAYOUT
+       tristate
+       depends on NFS_V4_1 && NFS_V3
+       default m
+
 config NFS_V4_1_IMPLEMENTATION_ID_DOMAIN
        string "NFSv4.1 Implementation ID Domain"
        depends on NFS_V4_1
index 23abffa8a4cef355d4733f915d78ec1b2ffbe8bd..1e987acf20c941312d31cd22bd153222f92e8d71 100644 (file)
@@ -33,3 +33,4 @@ nfsv4-$(CONFIG_NFS_V4_2)      += nfs42proc.o
 obj-$(CONFIG_PNFS_FILE_LAYOUT) += filelayout/
 obj-$(CONFIG_PNFS_OBJLAYOUT) += objlayout/
 obj-$(CONFIG_PNFS_BLOCK) += blocklayout/
+obj-$(CONFIG_PNFS_FLEXFILE_LAYOUT) += flexfilelayout/
diff --git a/fs/nfs/flexfilelayout/Makefile b/fs/nfs/flexfilelayout/Makefile
new file mode 100644 (file)
index 0000000..1d2c9f6
--- /dev/null
@@ -0,0 +1,5 @@
+#
+# Makefile for the pNFS Flexfile Layout Driver kernel module
+#
+obj-$(CONFIG_PNFS_FLEXFILE_LAYOUT) += nfs_layout_flexfiles.o
+nfs_layout_flexfiles-y := flexfilelayout.o flexfilelayoutdev.o
diff --git a/fs/nfs/flexfilelayout/flexfilelayout.c b/fs/nfs/flexfilelayout/flexfilelayout.c
new file mode 100644 (file)
index 0000000..f29fb7d
--- /dev/null
@@ -0,0 +1,1574 @@
+/*
+ * Module for pnfs flexfile layout driver.
+ *
+ * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
+ *
+ * Tao Peng <bergwolf@primarydata.com>
+ */
+
+#include <linux/nfs_fs.h>
+#include <linux/nfs_page.h>
+#include <linux/module.h>
+
+#include <linux/sunrpc/metrics.h>
+#include <linux/nfs_idmap.h>
+
+#include "flexfilelayout.h"
+#include "../nfs4session.h"
+#include "../internal.h"
+#include "../delegation.h"
+#include "../nfs4trace.h"
+#include "../iostat.h"
+#include "../nfs.h"
+
+#define NFSDBG_FACILITY         NFSDBG_PNFS_LD
+
+#define FF_LAYOUT_POLL_RETRY_MAX     (15*HZ)
+
+static struct pnfs_layout_hdr *
+ff_layout_alloc_layout_hdr(struct inode *inode, gfp_t gfp_flags)
+{
+       struct nfs4_flexfile_layout *ffl;
+
+       ffl = kzalloc(sizeof(*ffl), gfp_flags);
+       if (ffl) {
+               INIT_LIST_HEAD(&ffl->error_list);
+               return &ffl->generic_hdr;
+       } else
+               return NULL;
+}
+
+static void
+ff_layout_free_layout_hdr(struct pnfs_layout_hdr *lo)
+{
+       struct nfs4_ff_layout_ds_err *err, *n;
+
+       list_for_each_entry_safe(err, n, &FF_LAYOUT_FROM_HDR(lo)->error_list,
+                                list) {
+               list_del(&err->list);
+               kfree(err);
+       }
+       kfree(FF_LAYOUT_FROM_HDR(lo));
+}
+
+static int decode_stateid(struct xdr_stream *xdr, nfs4_stateid *stateid)
+{
+       __be32 *p;
+
+       p = xdr_inline_decode(xdr, NFS4_STATEID_SIZE);
+       if (unlikely(p == NULL))
+               return -ENOBUFS;
+       memcpy(stateid, p, NFS4_STATEID_SIZE);
+       dprintk("%s: stateid id= [%x%x%x%x]\n", __func__,
+               p[0], p[1], p[2], p[3]);
+       return 0;
+}
+
+static int decode_deviceid(struct xdr_stream *xdr, struct nfs4_deviceid *devid)
+{
+       __be32 *p;
+
+       p = xdr_inline_decode(xdr, NFS4_DEVICEID4_SIZE);
+       if (unlikely(!p))
+               return -ENOBUFS;
+       memcpy(devid, p, NFS4_DEVICEID4_SIZE);
+       nfs4_print_deviceid(devid);
+       return 0;
+}
+
+static int decode_nfs_fh(struct xdr_stream *xdr, struct nfs_fh *fh)
+{
+       __be32 *p;
+
+       p = xdr_inline_decode(xdr, 4);
+       if (unlikely(!p))
+               return -ENOBUFS;
+       fh->size = be32_to_cpup(p++);
+       if (fh->size > sizeof(struct nfs_fh)) {
+               printk(KERN_ERR "NFS flexfiles: Too big fh received %d\n",
+                      fh->size);
+               return -EOVERFLOW;
+       }
+       /* fh.data */
+       p = xdr_inline_decode(xdr, fh->size);
+       if (unlikely(!p))
+               return -ENOBUFS;
+       memcpy(&fh->data, p, fh->size);
+       dprintk("%s: fh len %d\n", __func__, fh->size);
+
+       return 0;
+}
+
+/*
+ * Currently only stringified uids and gids are accepted.
+ * I.e., kerberos is not supported to the DSes, so no pricipals.
+ *
+ * That means that one common function will suffice, but when
+ * principals are added, this should be split to accomodate
+ * calls to both nfs_map_name_to_uid() and nfs_map_group_to_gid().
+ */
+static int
+decode_name(struct xdr_stream *xdr, u32 *id)
+{
+       __be32 *p;
+       int len;
+
+       /* opaque_length(4)*/
+       p = xdr_inline_decode(xdr, 4);
+       if (unlikely(!p))
+               return -ENOBUFS;
+       len = be32_to_cpup(p++);
+       if (len < 0)
+               return -EINVAL;
+
+       dprintk("%s: len %u\n", __func__, len);
+
+       /* opaque body */
+       p = xdr_inline_decode(xdr, len);
+       if (unlikely(!p))
+               return -ENOBUFS;
+
+       if (!nfs_map_string_to_numeric((char *)p, len, id))
+               return -EINVAL;
+
+       return 0;
+}
+
+static void ff_layout_free_mirror_array(struct nfs4_ff_layout_segment *fls)
+{
+       int i;
+
+       if (fls->mirror_array) {
+               for (i = 0; i < fls->mirror_array_cnt; i++) {
+                       /* normally mirror_ds is freed in
+                        * .free_deviceid_node but we still do it here
+                        * for .alloc_lseg error path */
+                       if (fls->mirror_array[i]) {
+                               kfree(fls->mirror_array[i]->fh_versions);
+                               nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
+                               kfree(fls->mirror_array[i]);
+                       }
+               }
+               kfree(fls->mirror_array);
+               fls->mirror_array = NULL;
+       }
+}
+
+static int ff_layout_check_layout(struct nfs4_layoutget_res *lgr)
+{
+       int ret = 0;
+
+       dprintk("--> %s\n", __func__);
+
+       /* FIXME: remove this check when layout segment support is added */
+       if (lgr->range.offset != 0 ||
+           lgr->range.length != NFS4_MAX_UINT64) {
+               dprintk("%s Only whole file layouts supported. Use MDS i/o\n",
+                       __func__);
+               ret = -EINVAL;
+       }
+
+       dprintk("--> %s returns %d\n", __func__, ret);
+       return ret;
+}
+
+static void _ff_layout_free_lseg(struct nfs4_ff_layout_segment *fls)
+{
+       if (fls) {
+               ff_layout_free_mirror_array(fls);
+               kfree(fls);
+       }
+}
+
+static void ff_layout_sort_mirrors(struct nfs4_ff_layout_segment *fls)
+{
+       struct nfs4_ff_layout_mirror *tmp;
+       int i, j;
+
+       for (i = 0; i < fls->mirror_array_cnt - 1; i++) {
+               for (j = i + 1; j < fls->mirror_array_cnt; j++)
+                       if (fls->mirror_array[i]->efficiency <
+                           fls->mirror_array[j]->efficiency) {
+                               tmp = fls->mirror_array[i];
+                               fls->mirror_array[i] = fls->mirror_array[j];
+                               fls->mirror_array[j] = tmp;
+                       }
+       }
+}
+
+static struct pnfs_layout_segment *
+ff_layout_alloc_lseg(struct pnfs_layout_hdr *lh,
+                    struct nfs4_layoutget_res *lgr,
+                    gfp_t gfp_flags)
+{
+       struct pnfs_layout_segment *ret;
+       struct nfs4_ff_layout_segment *fls = NULL;
+       struct xdr_stream stream;
+       struct xdr_buf buf;
+       struct page *scratch;
+       u64 stripe_unit;
+       u32 mirror_array_cnt;
+       __be32 *p;
+       int i, rc;
+
+       dprintk("--> %s\n", __func__);
+       scratch = alloc_page(gfp_flags);
+       if (!scratch)
+               return ERR_PTR(-ENOMEM);
+
+       xdr_init_decode_pages(&stream, &buf, lgr->layoutp->pages,
+                             lgr->layoutp->len);
+       xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
+
+       /* stripe unit and mirror_array_cnt */
+       rc = -EIO;
+       p = xdr_inline_decode(&stream, 8 + 4);
+       if (!p)
+               goto out_err_free;
+
+       p = xdr_decode_hyper(p, &stripe_unit);
+       mirror_array_cnt = be32_to_cpup(p++);
+       dprintk("%s: stripe_unit=%llu mirror_array_cnt=%u\n", __func__,
+               stripe_unit, mirror_array_cnt);
+
+       if (mirror_array_cnt > NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT ||
+           mirror_array_cnt == 0)
+               goto out_err_free;
+
+       rc = -ENOMEM;
+       fls = kzalloc(sizeof(*fls), gfp_flags);
+       if (!fls)
+               goto out_err_free;
+
+       fls->mirror_array_cnt = mirror_array_cnt;
+       fls->stripe_unit = stripe_unit;
+       fls->mirror_array = kcalloc(fls->mirror_array_cnt,
+                                   sizeof(fls->mirror_array[0]), gfp_flags);
+       if (fls->mirror_array == NULL)
+               goto out_err_free;
+
+       for (i = 0; i < fls->mirror_array_cnt; i++) {
+               struct nfs4_deviceid devid;
+               struct nfs4_deviceid_node *idnode;
+               u32 ds_count;
+               u32 fh_count;
+               int j;
+
+               rc = -EIO;
+               p = xdr_inline_decode(&stream, 4);
+               if (!p)
+                       goto out_err_free;
+               ds_count = be32_to_cpup(p);
+
+               /* FIXME: allow for striping? */
+               if (ds_count != 1)
+                       goto out_err_free;
+
+               fls->mirror_array[i] =
+                       kzalloc(sizeof(struct nfs4_ff_layout_mirror),
+                               gfp_flags);
+               if (fls->mirror_array[i] == NULL) {
+                       rc = -ENOMEM;
+                       goto out_err_free;
+               }
+
+               spin_lock_init(&fls->mirror_array[i]->lock);
+               fls->mirror_array[i]->ds_count = ds_count;
+
+               /* deviceid */
+               rc = decode_deviceid(&stream, &devid);
+               if (rc)
+                       goto out_err_free;
+
+               idnode = nfs4_find_get_deviceid(NFS_SERVER(lh->plh_inode),
+                                               &devid, lh->plh_lc_cred,
+                                               gfp_flags);
+               /*
+                * upon success, mirror_ds is allocated by previous
+                * getdeviceinfo, or newly by .alloc_deviceid_node
+                * nfs4_find_get_deviceid failure is indeed getdeviceinfo falure
+                */
+               if (idnode)
+                       fls->mirror_array[i]->mirror_ds =
+                               FF_LAYOUT_MIRROR_DS(idnode);
+               else
+                       goto out_err_free;
+
+               /* efficiency */
+               rc = -EIO;
+               p = xdr_inline_decode(&stream, 4);
+               if (!p)
+                       goto out_err_free;
+               fls->mirror_array[i]->efficiency = be32_to_cpup(p);
+
+               /* stateid */
+               rc = decode_stateid(&stream, &fls->mirror_array[i]->stateid);
+               if (rc)
+                       goto out_err_free;
+
+               /* fh */
+               p = xdr_inline_decode(&stream, 4);
+               if (!p)
+                       goto out_err_free;
+               fh_count = be32_to_cpup(p);
+
+               fls->mirror_array[i]->fh_versions =
+                       kzalloc(fh_count * sizeof(struct nfs_fh),
+                               gfp_flags);
+               if (fls->mirror_array[i]->fh_versions == NULL) {
+                       rc = -ENOMEM;
+                       goto out_err_free;
+               }
+
+               for (j = 0; j < fh_count; j++) {
+                       rc = decode_nfs_fh(&stream,
+                                          &fls->mirror_array[i]->fh_versions[j]);
+                       if (rc)
+                               goto out_err_free;
+               }
+
+               fls->mirror_array[i]->fh_versions_cnt = fh_count;
+
+               /* user */
+               rc = decode_name(&stream, &fls->mirror_array[i]->uid);
+               if (rc)
+                       goto out_err_free;
+
+               /* group */
+               rc = decode_name(&stream, &fls->mirror_array[i]->gid);
+               if (rc)
+                       goto out_err_free;
+
+               dprintk("%s: uid %d gid %d\n", __func__,
+                       fls->mirror_array[i]->uid,
+                       fls->mirror_array[i]->gid);
+       }
+
+       ff_layout_sort_mirrors(fls);
+       rc = ff_layout_check_layout(lgr);
+       if (rc)
+               goto out_err_free;
+
+       ret = &fls->generic_hdr;
+       dprintk("<-- %s (success)\n", __func__);
+out_free_page:
+       __free_page(scratch);
+       return ret;
+out_err_free:
+       _ff_layout_free_lseg(fls);
+       ret = ERR_PTR(rc);
+       dprintk("<-- %s (%d)\n", __func__, rc);
+       goto out_free_page;
+}
+
+static bool ff_layout_has_rw_segments(struct pnfs_layout_hdr *layout)
+{
+       struct pnfs_layout_segment *lseg;
+
+       list_for_each_entry(lseg, &layout->plh_segs, pls_list)
+               if (lseg->pls_range.iomode == IOMODE_RW)
+                       return true;
+
+       return false;
+}
+
+static void
+ff_layout_free_lseg(struct pnfs_layout_segment *lseg)
+{
+       struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
+       int i;
+
+       dprintk("--> %s\n", __func__);
+
+       for (i = 0; i < fls->mirror_array_cnt; i++) {
+               if (fls->mirror_array[i]) {
+                       nfs4_ff_layout_put_deviceid(fls->mirror_array[i]->mirror_ds);
+                       fls->mirror_array[i]->mirror_ds = NULL;
+                       if (fls->mirror_array[i]->cred) {
+                               put_rpccred(fls->mirror_array[i]->cred);
+                               fls->mirror_array[i]->cred = NULL;
+                       }
+               }
+       }
+
+       if (lseg->pls_range.iomode == IOMODE_RW) {
+               struct nfs4_flexfile_layout *ffl;
+               struct inode *inode;
+
+               ffl = FF_LAYOUT_FROM_HDR(lseg->pls_layout);
+               inode = ffl->generic_hdr.plh_inode;
+               spin_lock(&inode->i_lock);
+               if (!ff_layout_has_rw_segments(lseg->pls_layout)) {
+                       ffl->commit_info.nbuckets = 0;
+                       kfree(ffl->commit_info.buckets);
+                       ffl->commit_info.buckets = NULL;
+               }
+               spin_unlock(&inode->i_lock);
+       }
+       _ff_layout_free_lseg(fls);
+}
+
+/* Return 1 until we have multiple lsegs support */
+static int
+ff_layout_get_lseg_count(struct nfs4_ff_layout_segment *fls)
+{
+       return 1;
+}
+
+static int
+ff_layout_alloc_commit_info(struct pnfs_layout_segment *lseg,
+                           struct nfs_commit_info *cinfo,
+                           gfp_t gfp_flags)
+{
+       struct nfs4_ff_layout_segment *fls = FF_LAYOUT_LSEG(lseg);
+       struct pnfs_commit_bucket *buckets;
+       int size;
+
+       if (cinfo->ds->nbuckets != 0) {
+               /* This assumes there is only one RW lseg per file.
+                * To support multiple lseg per file, we need to
+                * change struct pnfs_commit_bucket to allow dynamic
+                * increasing nbuckets.
+                */
+               return 0;
+       }
+
+       size = ff_layout_get_lseg_count(fls) * FF_LAYOUT_MIRROR_COUNT(lseg);
+
+       buckets = kcalloc(size, sizeof(struct pnfs_commit_bucket),
+                         gfp_flags);
+       if (!buckets)
+               return -ENOMEM;
+       else {
+               int i;
+
+               spin_lock(cinfo->lock);
+               if (cinfo->ds->nbuckets != 0)
+                       kfree(buckets);
+               else {
+                       cinfo->ds->buckets = buckets;
+                       cinfo->ds->nbuckets = size;
+                       for (i = 0; i < size; i++) {
+                               INIT_LIST_HEAD(&buckets[i].written);
+                               INIT_LIST_HEAD(&buckets[i].committing);
+                               /* mark direct verifier as unset */
+                               buckets[i].direct_verf.committed =
+                                       NFS_INVALID_STABLE_HOW;
+                       }
+               }
+               spin_unlock(cinfo->lock);
+               return 0;
+       }
+}
+
+static struct nfs4_pnfs_ds *
+ff_layout_choose_best_ds_for_read(struct nfs_pageio_descriptor *pgio,
+                                 int *best_idx)
+{
+       struct nfs4_ff_layout_segment *fls;
+       struct nfs4_pnfs_ds *ds;
+       int idx;
+
+       fls = FF_LAYOUT_LSEG(pgio->pg_lseg);
+       /* mirrors are sorted by efficiency */
+       for (idx = 0; idx < fls->mirror_array_cnt; idx++) {
+               ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, idx, false);
+               if (ds) {
+                       *best_idx = idx;
+                       return ds;
+               }
+       }
+
+       return NULL;
+}
+
+static void
+ff_layout_pg_init_read(struct nfs_pageio_descriptor *pgio,
+                       struct nfs_page *req)
+{
+       struct nfs_pgio_mirror *pgm;
+       struct nfs4_ff_layout_mirror *mirror;
+       struct nfs4_pnfs_ds *ds;
+       int ds_idx;
+
+       /* Use full layout for now */
+       if (!pgio->pg_lseg)
+               pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
+                                                  req->wb_context,
+                                                  0,
+                                                  NFS4_MAX_UINT64,
+                                                  IOMODE_READ,
+                                                  GFP_KERNEL);
+       /* If no lseg, fall back to read through mds */
+       if (pgio->pg_lseg == NULL)
+               goto out_mds;
+
+       ds = ff_layout_choose_best_ds_for_read(pgio, &ds_idx);
+       if (!ds)
+               goto out_mds;
+       mirror = FF_LAYOUT_COMP(pgio->pg_lseg, ds_idx);
+
+       pgio->pg_mirror_idx = ds_idx;
+
+       /* read always uses only one mirror - idx 0 for pgio layer */
+       pgm = &pgio->pg_mirrors[0];
+       pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].rsize;
+
+       return;
+out_mds:
+       pnfs_put_lseg(pgio->pg_lseg);
+       pgio->pg_lseg = NULL;
+       nfs_pageio_reset_read_mds(pgio);
+}
+
+static void
+ff_layout_pg_init_write(struct nfs_pageio_descriptor *pgio,
+                       struct nfs_page *req)
+{
+       struct nfs4_ff_layout_mirror *mirror;
+       struct nfs_pgio_mirror *pgm;
+       struct nfs_commit_info cinfo;
+       struct nfs4_pnfs_ds *ds;
+       int i;
+       int status;
+
+       if (!pgio->pg_lseg)
+               pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
+                                                  req->wb_context,
+                                                  0,
+                                                  NFS4_MAX_UINT64,
+                                                  IOMODE_RW,
+                                                  GFP_NOFS);
+       /* If no lseg, fall back to write through mds */
+       if (pgio->pg_lseg == NULL)
+               goto out_mds;
+
+       nfs_init_cinfo(&cinfo, pgio->pg_inode, pgio->pg_dreq);
+       status = ff_layout_alloc_commit_info(pgio->pg_lseg, &cinfo, GFP_NOFS);
+       if (status < 0)
+               goto out_mds;
+
+       /* Use a direct mapping of ds_idx to pgio mirror_idx */
+       if (WARN_ON_ONCE(pgio->pg_mirror_count !=
+           FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg)))
+               goto out_mds;
+
+       for (i = 0; i < pgio->pg_mirror_count; i++) {
+               ds = nfs4_ff_layout_prepare_ds(pgio->pg_lseg, i, true);
+               if (!ds)
+                       goto out_mds;
+               pgm = &pgio->pg_mirrors[i];
+               mirror = FF_LAYOUT_COMP(pgio->pg_lseg, i);
+               pgm->pg_bsize = mirror->mirror_ds->ds_versions[0].wsize;
+       }
+
+       return;
+
+out_mds:
+       pnfs_put_lseg(pgio->pg_lseg);
+       pgio->pg_lseg = NULL;
+       nfs_pageio_reset_write_mds(pgio);
+}
+
+static unsigned int
+ff_layout_pg_get_mirror_count_write(struct nfs_pageio_descriptor *pgio,
+                                   struct nfs_page *req)
+{
+       if (!pgio->pg_lseg)
+               pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
+                                                  req->wb_context,
+                                                  0,
+                                                  NFS4_MAX_UINT64,
+                                                  IOMODE_RW,
+                                                  GFP_NOFS);
+       if (pgio->pg_lseg)
+               return FF_LAYOUT_MIRROR_COUNT(pgio->pg_lseg);
+
+       /* no lseg means that pnfs is not in use, so no mirroring here */
+       pnfs_put_lseg(pgio->pg_lseg);
+       pgio->pg_lseg = NULL;
+       nfs_pageio_reset_write_mds(pgio);
+       return 1;
+}
+
+static const struct nfs_pageio_ops ff_layout_pg_read_ops = {
+       .pg_init = ff_layout_pg_init_read,
+       .pg_test = pnfs_generic_pg_test,
+       .pg_doio = pnfs_generic_pg_readpages,
+       .pg_cleanup = pnfs_generic_pg_cleanup,
+};
+
+static const struct nfs_pageio_ops ff_layout_pg_write_ops = {
+       .pg_init = ff_layout_pg_init_write,
+       .pg_test = pnfs_generic_pg_test,
+       .pg_doio = pnfs_generic_pg_writepages,
+       .pg_get_mirror_count = ff_layout_pg_get_mirror_count_write,
+       .pg_cleanup = pnfs_generic_pg_cleanup,
+};
+
+static void ff_layout_reset_write(struct nfs_pgio_header *hdr, bool retry_pnfs)
+{
+       struct rpc_task *task = &hdr->task;
+
+       pnfs_layoutcommit_inode(hdr->inode, false);
+
+       if (retry_pnfs) {
+               dprintk("%s Reset task %5u for i/o through pNFS "
+                       "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
+                       hdr->task.tk_pid,
+                       hdr->inode->i_sb->s_id,
+                       (unsigned long long)NFS_FILEID(hdr->inode),
+                       hdr->args.count,
+                       (unsigned long long)hdr->args.offset);
+
+               if (!hdr->dreq) {
+                       struct nfs_open_context *ctx;
+
+                       ctx = nfs_list_entry(hdr->pages.next)->wb_context;
+                       set_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
+                       hdr->completion_ops->error_cleanup(&hdr->pages);
+               } else {
+                       nfs_direct_set_resched_writes(hdr->dreq);
+                       /* fake unstable write to let common nfs resend pages */
+                       hdr->verf.committed = NFS_UNSTABLE;
+                       hdr->good_bytes = 0;
+               }
+               return;
+       }
+
+       if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
+               dprintk("%s Reset task %5u for i/o through MDS "
+                       "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
+                       hdr->task.tk_pid,
+                       hdr->inode->i_sb->s_id,
+                       (unsigned long long)NFS_FILEID(hdr->inode),
+                       hdr->args.count,
+                       (unsigned long long)hdr->args.offset);
+
+               task->tk_status = pnfs_write_done_resend_to_mds(hdr);
+       }
+}
+
+static void ff_layout_reset_read(struct nfs_pgio_header *hdr)
+{
+       struct rpc_task *task = &hdr->task;
+
+       pnfs_layoutcommit_inode(hdr->inode, false);
+
+       if (!test_and_set_bit(NFS_IOHDR_REDO, &hdr->flags)) {
+               dprintk("%s Reset task %5u for i/o through MDS "
+                       "(req %s/%llu, %u bytes @ offset %llu)\n", __func__,
+                       hdr->task.tk_pid,
+                       hdr->inode->i_sb->s_id,
+                       (unsigned long long)NFS_FILEID(hdr->inode),
+                       hdr->args.count,
+                       (unsigned long long)hdr->args.offset);
+
+               task->tk_status = pnfs_read_done_resend_to_mds(hdr);
+       }
+}
+
+static int ff_layout_async_handle_error_v4(struct rpc_task *task,
+                                          struct nfs4_state *state,
+                                          struct nfs_client *clp,
+                                          struct pnfs_layout_segment *lseg,
+                                          int idx)
+{
+       struct pnfs_layout_hdr *lo = lseg->pls_layout;
+       struct inode *inode = lo->plh_inode;
+       struct nfs_server *mds_server = NFS_SERVER(inode);
+
+       struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
+       struct nfs_client *mds_client = mds_server->nfs_client;
+       struct nfs4_slot_table *tbl = &clp->cl_session->fc_slot_table;
+
+       if (task->tk_status >= 0)
+               return 0;
+
+       switch (task->tk_status) {
+       /* MDS state errors */
+       case -NFS4ERR_DELEG_REVOKED:
+       case -NFS4ERR_ADMIN_REVOKED:
+       case -NFS4ERR_BAD_STATEID:
+               if (state == NULL)
+                       break;
+               nfs_remove_bad_delegation(state->inode);
+       case -NFS4ERR_OPENMODE:
+               if (state == NULL)
+                       break;
+               if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
+                       goto out_bad_stateid;
+               goto wait_on_recovery;
+       case -NFS4ERR_EXPIRED:
+               if (state != NULL) {
+                       if (nfs4_schedule_stateid_recovery(mds_server, state) < 0)
+                               goto out_bad_stateid;
+               }
+               nfs4_schedule_lease_recovery(mds_client);
+               goto wait_on_recovery;
+       /* DS session errors */
+       case -NFS4ERR_BADSESSION:
+       case -NFS4ERR_BADSLOT:
+       case -NFS4ERR_BAD_HIGH_SLOT:
+       case -NFS4ERR_DEADSESSION:
+       case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
+       case -NFS4ERR_SEQ_FALSE_RETRY:
+       case -NFS4ERR_SEQ_MISORDERED:
+               dprintk("%s ERROR %d, Reset session. Exchangeid "
+                       "flags 0x%x\n", __func__, task->tk_status,
+                       clp->cl_exchange_flags);
+               nfs4_schedule_session_recovery(clp->cl_session, task->tk_status);
+               break;
+       case -NFS4ERR_DELAY:
+       case -NFS4ERR_GRACE:
+               rpc_delay(task, FF_LAYOUT_POLL_RETRY_MAX);
+               break;
+       case -NFS4ERR_RETRY_UNCACHED_REP:
+               break;
+       /* Invalidate Layout errors */
+       case -NFS4ERR_PNFS_NO_LAYOUT:
+       case -ESTALE:           /* mapped NFS4ERR_STALE */
+       case -EBADHANDLE:       /* mapped NFS4ERR_BADHANDLE */
+       case -EISDIR:           /* mapped NFS4ERR_ISDIR */
+       case -NFS4ERR_FHEXPIRED:
+       case -NFS4ERR_WRONG_TYPE:
+               dprintk("%s Invalid layout error %d\n", __func__,
+                       task->tk_status);
+               /*
+                * Destroy layout so new i/o will get a new layout.
+                * Layout will not be destroyed until all current lseg
+                * references are put. Mark layout as invalid to resend failed
+                * i/o and all i/o waiting on the slot table to the MDS until
+                * layout is destroyed and a new valid layout is obtained.
+                */
+               pnfs_destroy_layout(NFS_I(inode));
+               rpc_wake_up(&tbl->slot_tbl_waitq);
+               goto reset;
+       /* RPC connection errors */
+       case -ECONNREFUSED:
+       case -EHOSTDOWN:
+       case -EHOSTUNREACH:
+       case -ENETUNREACH:
+       case -EIO:
+       case -ETIMEDOUT:
+       case -EPIPE:
+               dprintk("%s DS connection error %d\n", __func__,
+                       task->tk_status);
+               nfs4_mark_deviceid_unavailable(devid);
+               rpc_wake_up(&tbl->slot_tbl_waitq);
+               /* fall through */
+       default:
+               if (ff_layout_has_available_ds(lseg))
+                       return -NFS4ERR_RESET_TO_PNFS;
+reset:
+               dprintk("%s Retry through MDS. Error %d\n", __func__,
+                       task->tk_status);
+               return -NFS4ERR_RESET_TO_MDS;
+       }
+out:
+       task->tk_status = 0;
+       return -EAGAIN;
+out_bad_stateid:
+       task->tk_status = -EIO;
+       return 0;
+wait_on_recovery:
+       rpc_sleep_on(&mds_client->cl_rpcwaitq, task, NULL);
+       if (test_bit(NFS4CLNT_MANAGER_RUNNING, &mds_client->cl_state) == 0)
+               rpc_wake_up_queued_task(&mds_client->cl_rpcwaitq, task);
+       goto out;
+}
+
+/* Retry all errors through either pNFS or MDS except for -EJUKEBOX */
+static int ff_layout_async_handle_error_v3(struct rpc_task *task,
+                                          struct pnfs_layout_segment *lseg,
+                                          int idx)
+{
+       struct nfs4_deviceid_node *devid = FF_LAYOUT_DEVID_NODE(lseg, idx);
+
+       if (task->tk_status >= 0)
+               return 0;
+
+       if (task->tk_status != -EJUKEBOX) {
+               dprintk("%s DS connection error %d\n", __func__,
+                       task->tk_status);
+               nfs4_mark_deviceid_unavailable(devid);
+               if (ff_layout_has_available_ds(lseg))
+                       return -NFS4ERR_RESET_TO_PNFS;
+               else
+                       return -NFS4ERR_RESET_TO_MDS;
+       }
+
+       if (task->tk_status == -EJUKEBOX)
+               nfs_inc_stats(lseg->pls_layout->plh_inode, NFSIOS_DELAY);
+       task->tk_status = 0;
+       rpc_restart_call(task);
+       rpc_delay(task, NFS_JUKEBOX_RETRY_TIME);
+       return -EAGAIN;
+}
+
+static int ff_layout_async_handle_error(struct rpc_task *task,
+                                       struct nfs4_state *state,
+                                       struct nfs_client *clp,
+                                       struct pnfs_layout_segment *lseg,
+                                       int idx)
+{
+       int vers = clp->cl_nfs_mod->rpc_vers->number;
+
+       switch (vers) {
+       case 3:
+               return ff_layout_async_handle_error_v3(task, lseg, idx);
+       case 4:
+               return ff_layout_async_handle_error_v4(task, state, clp,
+                                                      lseg, idx);
+       default:
+               /* should never happen */
+               WARN_ON_ONCE(1);
+               return 0;
+       }
+}
+
+static void ff_layout_io_track_ds_error(struct pnfs_layout_segment *lseg,
+                                       int idx, u64 offset, u64 length,
+                                       u32 status, int opnum)
+{
+       struct nfs4_ff_layout_mirror *mirror;
+       int err;
+
+       mirror = FF_LAYOUT_COMP(lseg, idx);
+       err = ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
+                                      mirror, offset, length, status, opnum,
+                                      GFP_NOIO);
+       dprintk("%s: err %d op %d status %u\n", __func__, err, opnum, status);
+}
+
+/* NFS_PROTO call done callback routines */
+
+static int ff_layout_read_done_cb(struct rpc_task *task,
+                               struct nfs_pgio_header *hdr)
+{
+       struct inode *inode;
+       int err;
+
+       trace_nfs4_pnfs_read(hdr, task->tk_status);
+       if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
+               hdr->res.op_status = NFS4ERR_NXIO;
+       if (task->tk_status < 0 && hdr->res.op_status)
+               ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
+                                           hdr->args.offset, hdr->args.count,
+                                           hdr->res.op_status, OP_READ);
+       err = ff_layout_async_handle_error(task, hdr->args.context->state,
+                                          hdr->ds_clp, hdr->lseg,
+                                          hdr->pgio_mirror_idx);
+
+       switch (err) {
+       case -NFS4ERR_RESET_TO_PNFS:
+               set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
+                       &hdr->lseg->pls_layout->plh_flags);
+               pnfs_read_resend_pnfs(hdr);
+               return task->tk_status;
+       case -NFS4ERR_RESET_TO_MDS:
+               inode = hdr->lseg->pls_layout->plh_inode;
+               pnfs_error_mark_layout_for_return(inode, hdr->lseg);
+               ff_layout_reset_read(hdr);
+               return task->tk_status;
+       case -EAGAIN:
+               rpc_restart_call_prepare(task);
+               return -EAGAIN;
+       }
+
+       return 0;
+}
+
+/*
+ * We reference the rpc_cred of the first WRITE that triggers the need for
+ * a LAYOUTCOMMIT, and use it to send the layoutcommit compound.
+ * rfc5661 is not clear about which credential should be used.
+ *
+ * Flexlayout client should treat DS replied FILE_SYNC as DATA_SYNC, so
+ * to follow http://www.rfc-editor.org/errata_search.php?rfc=5661&eid=2751
+ * we always send layoutcommit after DS writes.
+ */
+static void
+ff_layout_set_layoutcommit(struct nfs_pgio_header *hdr)
+{
+       pnfs_set_layoutcommit(hdr);
+       dprintk("%s inode %lu pls_end_pos %lu\n", __func__, hdr->inode->i_ino,
+               (unsigned long) NFS_I(hdr->inode)->layout->plh_lwb);
+}
+
+static bool
+ff_layout_reset_to_mds(struct pnfs_layout_segment *lseg, int idx)
+{
+       /* No mirroring for now */
+       struct nfs4_deviceid_node *node = FF_LAYOUT_DEVID_NODE(lseg, idx);
+
+       return ff_layout_test_devid_unavailable(node);
+}
+
+static int ff_layout_read_prepare_common(struct rpc_task *task,
+                                        struct nfs_pgio_header *hdr)
+{
+       if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
+               rpc_exit(task, -EIO);
+               return -EIO;
+       }
+       if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
+               dprintk("%s task %u reset io to MDS\n", __func__, task->tk_pid);
+               if (ff_layout_has_available_ds(hdr->lseg))
+                       pnfs_read_resend_pnfs(hdr);
+               else
+                       ff_layout_reset_read(hdr);
+               rpc_exit(task, 0);
+               return -EAGAIN;
+       }
+       hdr->pgio_done_cb = ff_layout_read_done_cb;
+
+       return 0;
+}
+
+/*
+ * Call ops for the async read/write cases
+ * In the case of dense layouts, the offset needs to be reset to its
+ * original value.
+ */
+static void ff_layout_read_prepare_v3(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       if (ff_layout_read_prepare_common(task, hdr))
+               return;
+
+       rpc_call_start(task);
+}
+
+static int ff_layout_setup_sequence(struct nfs_client *ds_clp,
+                                   struct nfs4_sequence_args *args,
+                                   struct nfs4_sequence_res *res,
+                                   struct rpc_task *task)
+{
+       if (ds_clp->cl_session)
+               return nfs41_setup_sequence(ds_clp->cl_session,
+                                          args,
+                                          res,
+                                          task);
+       return nfs40_setup_sequence(ds_clp->cl_slot_tbl,
+                                  args,
+                                  res,
+                                  task);
+}
+
+static void ff_layout_read_prepare_v4(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       if (ff_layout_read_prepare_common(task, hdr))
+               return;
+
+       if (ff_layout_setup_sequence(hdr->ds_clp,
+                                    &hdr->args.seq_args,
+                                    &hdr->res.seq_res,
+                                    task))
+               return;
+
+       if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
+                       hdr->args.lock_context, FMODE_READ) == -EIO)
+               rpc_exit(task, -EIO); /* lost lock, terminate I/O */
+}
+
+static void ff_layout_read_call_done(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       dprintk("--> %s task->tk_status %d\n", __func__, task->tk_status);
+
+       if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
+           task->tk_status == 0) {
+               nfs4_sequence_done(task, &hdr->res.seq_res);
+               return;
+       }
+
+       /* Note this may cause RPC to be resent */
+       hdr->mds_ops->rpc_call_done(task, hdr);
+}
+
+static void ff_layout_read_count_stats(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       rpc_count_iostats_metrics(task,
+           &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_READ]);
+}
+
+static int ff_layout_write_done_cb(struct rpc_task *task,
+                               struct nfs_pgio_header *hdr)
+{
+       struct inode *inode;
+       int err;
+
+       trace_nfs4_pnfs_write(hdr, task->tk_status);
+       if (task->tk_status == -ETIMEDOUT && !hdr->res.op_status)
+               hdr->res.op_status = NFS4ERR_NXIO;
+       if (task->tk_status < 0 && hdr->res.op_status)
+               ff_layout_io_track_ds_error(hdr->lseg, hdr->pgio_mirror_idx,
+                                           hdr->args.offset, hdr->args.count,
+                                           hdr->res.op_status, OP_WRITE);
+       err = ff_layout_async_handle_error(task, hdr->args.context->state,
+                                          hdr->ds_clp, hdr->lseg,
+                                          hdr->pgio_mirror_idx);
+
+       switch (err) {
+       case -NFS4ERR_RESET_TO_PNFS:
+       case -NFS4ERR_RESET_TO_MDS:
+               inode = hdr->lseg->pls_layout->plh_inode;
+               pnfs_error_mark_layout_for_return(inode, hdr->lseg);
+               if (err == -NFS4ERR_RESET_TO_PNFS) {
+                       pnfs_set_retry_layoutget(hdr->lseg->pls_layout);
+                       ff_layout_reset_write(hdr, true);
+               } else {
+                       pnfs_clear_retry_layoutget(hdr->lseg->pls_layout);
+                       ff_layout_reset_write(hdr, false);
+               }
+               return task->tk_status;
+       case -EAGAIN:
+               rpc_restart_call_prepare(task);
+               return -EAGAIN;
+       }
+
+       if (hdr->res.verf->committed == NFS_FILE_SYNC ||
+           hdr->res.verf->committed == NFS_DATA_SYNC)
+               ff_layout_set_layoutcommit(hdr);
+
+       return 0;
+}
+
+static int ff_layout_commit_done_cb(struct rpc_task *task,
+                                    struct nfs_commit_data *data)
+{
+       struct inode *inode;
+       int err;
+
+       trace_nfs4_pnfs_commit_ds(data, task->tk_status);
+       if (task->tk_status == -ETIMEDOUT && !data->res.op_status)
+               data->res.op_status = NFS4ERR_NXIO;
+       if (task->tk_status < 0 && data->res.op_status)
+               ff_layout_io_track_ds_error(data->lseg, data->ds_commit_index,
+                                           data->args.offset, data->args.count,
+                                           data->res.op_status, OP_COMMIT);
+       err = ff_layout_async_handle_error(task, NULL, data->ds_clp,
+                                          data->lseg, data->ds_commit_index);
+
+       switch (err) {
+       case -NFS4ERR_RESET_TO_PNFS:
+       case -NFS4ERR_RESET_TO_MDS:
+               inode = data->lseg->pls_layout->plh_inode;
+               pnfs_error_mark_layout_for_return(inode, data->lseg);
+               if (err == -NFS4ERR_RESET_TO_PNFS)
+                       pnfs_set_retry_layoutget(data->lseg->pls_layout);
+               else
+                       pnfs_clear_retry_layoutget(data->lseg->pls_layout);
+               pnfs_generic_prepare_to_resend_writes(data);
+               return -EAGAIN;
+       case -EAGAIN:
+               rpc_restart_call_prepare(task);
+               return -EAGAIN;
+       }
+
+       if (data->verf.committed == NFS_UNSTABLE)
+               pnfs_commit_set_layoutcommit(data);
+
+       return 0;
+}
+
+static int ff_layout_write_prepare_common(struct rpc_task *task,
+                                         struct nfs_pgio_header *hdr)
+{
+       if (unlikely(test_bit(NFS_CONTEXT_BAD, &hdr->args.context->flags))) {
+               rpc_exit(task, -EIO);
+               return -EIO;
+       }
+
+       if (ff_layout_reset_to_mds(hdr->lseg, hdr->pgio_mirror_idx)) {
+               bool retry_pnfs;
+
+               retry_pnfs = ff_layout_has_available_ds(hdr->lseg);
+               dprintk("%s task %u reset io to %s\n", __func__,
+                       task->tk_pid, retry_pnfs ? "pNFS" : "MDS");
+               ff_layout_reset_write(hdr, retry_pnfs);
+               rpc_exit(task, 0);
+               return -EAGAIN;
+       }
+
+       return 0;
+}
+
+static void ff_layout_write_prepare_v3(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       if (ff_layout_write_prepare_common(task, hdr))
+               return;
+
+       rpc_call_start(task);
+}
+
+static void ff_layout_write_prepare_v4(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       if (ff_layout_write_prepare_common(task, hdr))
+               return;
+
+       if (ff_layout_setup_sequence(hdr->ds_clp,
+                                    &hdr->args.seq_args,
+                                    &hdr->res.seq_res,
+                                    task))
+               return;
+
+       if (nfs4_set_rw_stateid(&hdr->args.stateid, hdr->args.context,
+                       hdr->args.lock_context, FMODE_WRITE) == -EIO)
+               rpc_exit(task, -EIO); /* lost lock, terminate I/O */
+}
+
+static void ff_layout_write_call_done(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       if (test_bit(NFS_IOHDR_REDO, &hdr->flags) &&
+           task->tk_status == 0) {
+               nfs4_sequence_done(task, &hdr->res.seq_res);
+               return;
+       }
+
+       /* Note this may cause RPC to be resent */
+       hdr->mds_ops->rpc_call_done(task, hdr);
+}
+
+static void ff_layout_write_count_stats(struct rpc_task *task, void *data)
+{
+       struct nfs_pgio_header *hdr = data;
+
+       rpc_count_iostats_metrics(task,
+           &NFS_CLIENT(hdr->inode)->cl_metrics[NFSPROC4_CLNT_WRITE]);
+}
+
+static void ff_layout_commit_prepare_v3(struct rpc_task *task, void *data)
+{
+       rpc_call_start(task);
+}
+
+static void ff_layout_commit_prepare_v4(struct rpc_task *task, void *data)
+{
+       struct nfs_commit_data *wdata = data;
+
+       ff_layout_setup_sequence(wdata->ds_clp,
+                                &wdata->args.seq_args,
+                                &wdata->res.seq_res,
+                                task);
+}
+
+static void ff_layout_commit_count_stats(struct rpc_task *task, void *data)
+{
+       struct nfs_commit_data *cdata = data;
+
+       rpc_count_iostats_metrics(task,
+           &NFS_CLIENT(cdata->inode)->cl_metrics[NFSPROC4_CLNT_COMMIT]);
+}
+
+static const struct rpc_call_ops ff_layout_read_call_ops_v3 = {
+       .rpc_call_prepare = ff_layout_read_prepare_v3,
+       .rpc_call_done = ff_layout_read_call_done,
+       .rpc_count_stats = ff_layout_read_count_stats,
+       .rpc_release = pnfs_generic_rw_release,
+};
+
+static const struct rpc_call_ops ff_layout_read_call_ops_v4 = {
+       .rpc_call_prepare = ff_layout_read_prepare_v4,
+       .rpc_call_done = ff_layout_read_call_done,
+       .rpc_count_stats = ff_layout_read_count_stats,
+       .rpc_release = pnfs_generic_rw_release,
+};
+
+static const struct rpc_call_ops ff_layout_write_call_ops_v3 = {
+       .rpc_call_prepare = ff_layout_write_prepare_v3,
+       .rpc_call_done = ff_layout_write_call_done,
+       .rpc_count_stats = ff_layout_write_count_stats,
+       .rpc_release = pnfs_generic_rw_release,
+};
+
+static const struct rpc_call_ops ff_layout_write_call_ops_v4 = {
+       .rpc_call_prepare = ff_layout_write_prepare_v4,
+       .rpc_call_done = ff_layout_write_call_done,
+       .rpc_count_stats = ff_layout_write_count_stats,
+       .rpc_release = pnfs_generic_rw_release,
+};
+
+static const struct rpc_call_ops ff_layout_commit_call_ops_v3 = {
+       .rpc_call_prepare = ff_layout_commit_prepare_v3,
+       .rpc_call_done = pnfs_generic_write_commit_done,
+       .rpc_count_stats = ff_layout_commit_count_stats,
+       .rpc_release = pnfs_generic_commit_release,
+};
+
+static const struct rpc_call_ops ff_layout_commit_call_ops_v4 = {
+       .rpc_call_prepare = ff_layout_commit_prepare_v4,
+       .rpc_call_done = pnfs_generic_write_commit_done,
+       .rpc_count_stats = ff_layout_commit_count_stats,
+       .rpc_release = pnfs_generic_commit_release,
+};
+
+static enum pnfs_try_status
+ff_layout_read_pagelist(struct nfs_pgio_header *hdr)
+{
+       struct pnfs_layout_segment *lseg = hdr->lseg;
+       struct nfs4_pnfs_ds *ds;
+       struct rpc_clnt *ds_clnt;
+       struct rpc_cred *ds_cred;
+       loff_t offset = hdr->args.offset;
+       u32 idx = hdr->pgio_mirror_idx;
+       int vers;
+       struct nfs_fh *fh;
+
+       dprintk("--> %s ino %lu pgbase %u req %Zu@%llu\n",
+               __func__, hdr->inode->i_ino,
+               hdr->args.pgbase, (size_t)hdr->args.count, offset);
+
+       ds = nfs4_ff_layout_prepare_ds(lseg, idx, false);
+       if (!ds)
+               goto out_failed;
+
+       ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
+                                                  hdr->inode);
+       if (IS_ERR(ds_clnt))
+               goto out_failed;
+
+       ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
+       if (IS_ERR(ds_cred))
+               goto out_failed;
+
+       vers = nfs4_ff_layout_ds_version(lseg, idx);
+
+       dprintk("%s USE DS: %s cl_count %d vers %d\n", __func__,
+               ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count), vers);
+
+       atomic_inc(&ds->ds_clp->cl_count);
+       hdr->ds_clp = ds->ds_clp;
+       fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
+       if (fh)
+               hdr->args.fh = fh;
+
+       /*
+        * Note that if we ever decide to split across DSes,
+        * then we may need to handle dense-like offsets.
+        */
+       hdr->args.offset = offset;
+       hdr->mds_offset = offset;
+
+       /* Perform an asynchronous read to ds */
+       nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
+                         vers == 3 ? &ff_layout_read_call_ops_v3 :
+                                     &ff_layout_read_call_ops_v4,
+                         0, RPC_TASK_SOFTCONN);
+
+       return PNFS_ATTEMPTED;
+
+out_failed:
+       if (ff_layout_has_available_ds(lseg))
+               return PNFS_TRY_AGAIN;
+       return PNFS_NOT_ATTEMPTED;
+}
+
+/* Perform async writes. */
+static enum pnfs_try_status
+ff_layout_write_pagelist(struct nfs_pgio_header *hdr, int sync)
+{
+       struct pnfs_layout_segment *lseg = hdr->lseg;
+       struct nfs4_pnfs_ds *ds;
+       struct rpc_clnt *ds_clnt;
+       struct rpc_cred *ds_cred;
+       loff_t offset = hdr->args.offset;
+       int vers;
+       struct nfs_fh *fh;
+       int idx = hdr->pgio_mirror_idx;
+
+       ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
+       if (!ds)
+               return PNFS_NOT_ATTEMPTED;
+
+       ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
+                                                  hdr->inode);
+       if (IS_ERR(ds_clnt))
+               return PNFS_NOT_ATTEMPTED;
+
+       ds_cred = ff_layout_get_ds_cred(lseg, idx, hdr->cred);
+       if (IS_ERR(ds_cred))
+               return PNFS_NOT_ATTEMPTED;
+
+       vers = nfs4_ff_layout_ds_version(lseg, idx);
+
+       dprintk("%s ino %lu sync %d req %Zu@%llu DS: %s cl_count %d vers %d\n",
+               __func__, hdr->inode->i_ino, sync, (size_t) hdr->args.count,
+               offset, ds->ds_remotestr, atomic_read(&ds->ds_clp->cl_count),
+               vers);
+
+       hdr->pgio_done_cb = ff_layout_write_done_cb;
+       atomic_inc(&ds->ds_clp->cl_count);
+       hdr->ds_clp = ds->ds_clp;
+       hdr->ds_commit_idx = idx;
+       fh = nfs4_ff_layout_select_ds_fh(lseg, idx);
+       if (fh)
+               hdr->args.fh = fh;
+
+       /*
+        * Note that if we ever decide to split across DSes,
+        * then we may need to handle dense-like offsets.
+        */
+       hdr->args.offset = offset;
+
+       /* Perform an asynchronous write */
+       nfs_initiate_pgio(ds_clnt, hdr, ds_cred, ds->ds_clp->rpc_ops,
+                         vers == 3 ? &ff_layout_write_call_ops_v3 :
+                                     &ff_layout_write_call_ops_v4,
+                         sync, RPC_TASK_SOFTCONN);
+       return PNFS_ATTEMPTED;
+}
+
+static void
+ff_layout_mark_request_commit(struct nfs_page *req,
+                             struct pnfs_layout_segment *lseg,
+                             struct nfs_commit_info *cinfo,
+                             u32 ds_commit_idx)
+{
+       struct list_head *list;
+       struct pnfs_commit_bucket *buckets;
+
+       spin_lock(cinfo->lock);
+       buckets = cinfo->ds->buckets;
+       list = &buckets[ds_commit_idx].written;
+       if (list_empty(list)) {
+               /* Non-empty buckets hold a reference on the lseg.  That ref
+                * is normally transferred to the COMMIT call and released
+                * there.  It could also be released if the last req is pulled
+                * off due to a rewrite, in which case it will be done in
+                * pnfs_common_clear_request_commit
+                */
+               WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL);
+               buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg);
+       }
+       set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
+       cinfo->ds->nwritten++;
+
+       /* nfs_request_add_commit_list(). We need to add req to list without
+        * dropping cinfo lock.
+        */
+       set_bit(PG_CLEAN, &(req)->wb_flags);
+       nfs_list_add_request(req, list);
+       cinfo->mds->ncommit++;
+       spin_unlock(cinfo->lock);
+       if (!cinfo->dreq) {
+               inc_zone_page_state(req->wb_page, NR_UNSTABLE_NFS);
+               inc_bdi_stat(page_file_mapping(req->wb_page)->backing_dev_info,
+                            BDI_RECLAIMABLE);
+               __mark_inode_dirty(req->wb_context->dentry->d_inode,
+                                  I_DIRTY_DATASYNC);
+       }
+}
+
+static u32 calc_ds_index_from_commit(struct pnfs_layout_segment *lseg, u32 i)
+{
+       return i;
+}
+
+static struct nfs_fh *
+select_ds_fh_from_commit(struct pnfs_layout_segment *lseg, u32 i)
+{
+       struct nfs4_ff_layout_segment *flseg = FF_LAYOUT_LSEG(lseg);
+
+       /* FIXME: Assume that there is only one NFS version available
+        * for the DS.
+        */
+       return &flseg->mirror_array[i]->fh_versions[0];
+}
+
+static int ff_layout_initiate_commit(struct nfs_commit_data *data, int how)
+{
+       struct pnfs_layout_segment *lseg = data->lseg;
+       struct nfs4_pnfs_ds *ds;
+       struct rpc_clnt *ds_clnt;
+       struct rpc_cred *ds_cred;
+       u32 idx;
+       int vers;
+       struct nfs_fh *fh;
+
+       idx = calc_ds_index_from_commit(lseg, data->ds_commit_index);
+       ds = nfs4_ff_layout_prepare_ds(lseg, idx, true);
+       if (!ds)
+               goto out_err;
+
+       ds_clnt = nfs4_ff_find_or_create_ds_client(lseg, idx, ds->ds_clp,
+                                                  data->inode);
+       if (IS_ERR(ds_clnt))
+               goto out_err;
+
+       ds_cred = ff_layout_get_ds_cred(lseg, idx, data->cred);
+       if (IS_ERR(ds_cred))
+               goto out_err;
+
+       vers = nfs4_ff_layout_ds_version(lseg, idx);
+
+       dprintk("%s ino %lu, how %d cl_count %d vers %d\n", __func__,
+               data->inode->i_ino, how, atomic_read(&ds->ds_clp->cl_count),
+               vers);
+       data->commit_done_cb = ff_layout_commit_done_cb;
+       data->cred = ds_cred;
+       atomic_inc(&ds->ds_clp->cl_count);
+       data->ds_clp = ds->ds_clp;
+       fh = select_ds_fh_from_commit(lseg, data->ds_commit_index);
+       if (fh)
+               data->args.fh = fh;
+       return nfs_initiate_commit(ds_clnt, data, ds->ds_clp->rpc_ops,
+                                  vers == 3 ? &ff_layout_commit_call_ops_v3 :
+                                              &ff_layout_commit_call_ops_v4,
+                                  how, RPC_TASK_SOFTCONN);
+out_err:
+       pnfs_generic_prepare_to_resend_writes(data);
+       pnfs_generic_commit_release(data);
+       return -EAGAIN;
+}
+
+static int
+ff_layout_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
+                          int how, struct nfs_commit_info *cinfo)
+{
+       return pnfs_generic_commit_pagelist(inode, mds_pages, how, cinfo,
+                                           ff_layout_initiate_commit);
+}
+
+static struct pnfs_ds_commit_info *
+ff_layout_get_ds_info(struct inode *inode)
+{
+       struct pnfs_layout_hdr *layout = NFS_I(inode)->layout;
+
+       if (layout == NULL)
+               return NULL;
+
+       return &FF_LAYOUT_FROM_HDR(layout)->commit_info;
+}
+
+static void
+ff_layout_free_deveiceid_node(struct nfs4_deviceid_node *d)
+{
+       nfs4_ff_layout_free_deviceid(container_of(d, struct nfs4_ff_layout_ds,
+                                                 id_node));
+}
+
+static int ff_layout_encode_ioerr(struct nfs4_flexfile_layout *flo,
+                                 struct xdr_stream *xdr,
+                                 const struct nfs4_layoutreturn_args *args)
+{
+       struct pnfs_layout_hdr *hdr = &flo->generic_hdr;
+       __be32 *start;
+       int count = 0, ret = 0;
+
+       start = xdr_reserve_space(xdr, 4);
+       if (unlikely(!start))
+               return -E2BIG;
+
+       /* This assume we always return _ALL_ layouts */
+       spin_lock(&hdr->plh_inode->i_lock);
+       ret = ff_layout_encode_ds_ioerr(flo, xdr, &count, &args->range);
+       spin_unlock(&hdr->plh_inode->i_lock);
+
+       *start = cpu_to_be32(count);
+
+       return ret;
+}
+
+/* report nothing for now */
+static void ff_layout_encode_iostats(struct nfs4_flexfile_layout *flo,
+                                    struct xdr_stream *xdr,
+                                    const struct nfs4_layoutreturn_args *args)
+{
+       __be32 *p;
+
+       p = xdr_reserve_space(xdr, 4);
+       if (likely(p))
+               *p = cpu_to_be32(0);
+}
+
+static struct nfs4_deviceid_node *
+ff_layout_alloc_deviceid_node(struct nfs_server *server,
+                             struct pnfs_device *pdev, gfp_t gfp_flags)
+{
+       struct nfs4_ff_layout_ds *dsaddr;
+
+       dsaddr = nfs4_ff_alloc_deviceid_node(server, pdev, gfp_flags);
+       if (!dsaddr)
+               return NULL;
+       return &dsaddr->id_node;
+}
+
+static void
+ff_layout_encode_layoutreturn(struct pnfs_layout_hdr *lo,
+                             struct xdr_stream *xdr,
+                             const struct nfs4_layoutreturn_args *args)
+{
+       struct nfs4_flexfile_layout *flo = FF_LAYOUT_FROM_HDR(lo);
+       __be32 *start;
+
+       dprintk("%s: Begin\n", __func__);
+       start = xdr_reserve_space(xdr, 4);
+       BUG_ON(!start);
+
+       if (ff_layout_encode_ioerr(flo, xdr, args))
+               goto out;
+
+       ff_layout_encode_iostats(flo, xdr, args);
+out:
+       *start = cpu_to_be32((xdr->p - start - 1) * 4);
+       dprintk("%s: Return\n", __func__);
+}
+
+static struct pnfs_layoutdriver_type flexfilelayout_type = {
+       .id                     = LAYOUT_FLEX_FILES,
+       .name                   = "LAYOUT_FLEX_FILES",
+       .owner                  = THIS_MODULE,
+       .alloc_layout_hdr       = ff_layout_alloc_layout_hdr,
+       .free_layout_hdr        = ff_layout_free_layout_hdr,
+       .alloc_lseg             = ff_layout_alloc_lseg,
+       .free_lseg              = ff_layout_free_lseg,
+       .pg_read_ops            = &ff_layout_pg_read_ops,
+       .pg_write_ops           = &ff_layout_pg_write_ops,
+       .get_ds_info            = ff_layout_get_ds_info,
+       .free_deviceid_node     = ff_layout_free_deveiceid_node,
+       .mark_request_commit    = ff_layout_mark_request_commit,
+       .clear_request_commit   = pnfs_generic_clear_request_commit,
+       .scan_commit_lists      = pnfs_generic_scan_commit_lists,
+       .recover_commit_reqs    = pnfs_generic_recover_commit_reqs,
+       .commit_pagelist        = ff_layout_commit_pagelist,
+       .read_pagelist          = ff_layout_read_pagelist,
+       .write_pagelist         = ff_layout_write_pagelist,
+       .alloc_deviceid_node    = ff_layout_alloc_deviceid_node,
+       .encode_layoutreturn    = ff_layout_encode_layoutreturn,
+};
+
+static int __init nfs4flexfilelayout_init(void)
+{
+       printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Registering...\n",
+              __func__);
+       return pnfs_register_layoutdriver(&flexfilelayout_type);
+}
+
+static void __exit nfs4flexfilelayout_exit(void)
+{
+       printk(KERN_INFO "%s: NFSv4 Flexfile Layout Driver Unregistering...\n",
+              __func__);
+       pnfs_unregister_layoutdriver(&flexfilelayout_type);
+}
+
+MODULE_ALIAS("nfs-layouttype4-4");
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("The NFSv4 flexfile layout driver");
+
+module_init(nfs4flexfilelayout_init);
+module_exit(nfs4flexfilelayout_exit);
diff --git a/fs/nfs/flexfilelayout/flexfilelayout.h b/fs/nfs/flexfilelayout/flexfilelayout.h
new file mode 100644 (file)
index 0000000..070f204
--- /dev/null
@@ -0,0 +1,155 @@
+/*
+ * NFSv4 flexfile layout driver data structures.
+ *
+ * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
+ *
+ * Tao Peng <bergwolf@primarydata.com>
+ */
+
+#ifndef FS_NFS_NFS4FLEXFILELAYOUT_H
+#define FS_NFS_NFS4FLEXFILELAYOUT_H
+
+#include "../pnfs.h"
+
+/* XXX: Let's filter out insanely large mirror count for now to avoid oom
+ * due to network error etc. */
+#define NFS4_FLEXFILE_LAYOUT_MAX_MIRROR_CNT 4096
+
+struct nfs4_ff_ds_version {
+       u32                             version;
+       u32                             minor_version;
+       u32                             rsize;
+       u32                             wsize;
+       bool                            tightly_coupled;
+};
+
+/* chained in global deviceid hlist */
+struct nfs4_ff_layout_ds {
+       struct nfs4_deviceid_node       id_node;
+       u32                             ds_versions_cnt;
+       struct nfs4_ff_ds_version       *ds_versions;
+       struct nfs4_pnfs_ds             *ds;
+};
+
+struct nfs4_ff_layout_ds_err {
+       struct list_head                list; /* linked in mirror error_list */
+       u64                             offset;
+       u64                             length;
+       int                             status;
+       enum nfs_opnum4                 opnum;
+       nfs4_stateid                    stateid;
+       struct nfs4_deviceid            deviceid;
+};
+
+struct nfs4_ff_layout_mirror {
+       u32                             ds_count;
+       u32                             efficiency;
+       struct nfs4_ff_layout_ds        *mirror_ds;
+       u32                             fh_versions_cnt;
+       struct nfs_fh                   *fh_versions;
+       nfs4_stateid                    stateid;
+       struct nfs4_string              user_name;
+       struct nfs4_string              group_name;
+       u32                             uid;
+       u32                             gid;
+       struct rpc_cred                 *cred;
+       spinlock_t                      lock;
+};
+
+struct nfs4_ff_layout_segment {
+       struct pnfs_layout_segment      generic_hdr;
+       u64                             stripe_unit;
+       u32                             mirror_array_cnt;
+       struct nfs4_ff_layout_mirror    **mirror_array;
+};
+
+struct nfs4_flexfile_layout {
+       struct pnfs_layout_hdr generic_hdr;
+       struct pnfs_ds_commit_info commit_info;
+       struct list_head        error_list; /* nfs4_ff_layout_ds_err */
+};
+
+static inline struct nfs4_flexfile_layout *
+FF_LAYOUT_FROM_HDR(struct pnfs_layout_hdr *lo)
+{
+       return container_of(lo, struct nfs4_flexfile_layout, generic_hdr);
+}
+
+static inline struct nfs4_ff_layout_segment *
+FF_LAYOUT_LSEG(struct pnfs_layout_segment *lseg)
+{
+       return container_of(lseg,
+                           struct nfs4_ff_layout_segment,
+                           generic_hdr);
+}
+
+static inline struct nfs4_deviceid_node *
+FF_LAYOUT_DEVID_NODE(struct pnfs_layout_segment *lseg, u32 idx)
+{
+       if (idx >= FF_LAYOUT_LSEG(lseg)->mirror_array_cnt ||
+           FF_LAYOUT_LSEG(lseg)->mirror_array[idx] == NULL ||
+           FF_LAYOUT_LSEG(lseg)->mirror_array[idx]->mirror_ds == NULL)
+               return NULL;
+       return &FF_LAYOUT_LSEG(lseg)->mirror_array[idx]->mirror_ds->id_node;
+}
+
+static inline struct nfs4_ff_layout_ds *
+FF_LAYOUT_MIRROR_DS(struct nfs4_deviceid_node *node)
+{
+       return container_of(node, struct nfs4_ff_layout_ds, id_node);
+}
+
+static inline struct nfs4_ff_layout_mirror *
+FF_LAYOUT_COMP(struct pnfs_layout_segment *lseg, u32 idx)
+{
+       if (idx >= FF_LAYOUT_LSEG(lseg)->mirror_array_cnt)
+               return NULL;
+       return FF_LAYOUT_LSEG(lseg)->mirror_array[idx];
+}
+
+static inline u32
+FF_LAYOUT_MIRROR_COUNT(struct pnfs_layout_segment *lseg)
+{
+       return FF_LAYOUT_LSEG(lseg)->mirror_array_cnt;
+}
+
+static inline bool
+ff_layout_test_devid_unavailable(struct nfs4_deviceid_node *node)
+{
+       return nfs4_test_deviceid_unavailable(node);
+}
+
+static inline int
+nfs4_ff_layout_ds_version(struct pnfs_layout_segment *lseg, u32 ds_idx)
+{
+       return FF_LAYOUT_COMP(lseg, ds_idx)->mirror_ds->ds_versions[0].version;
+}
+
+struct nfs4_ff_layout_ds *
+nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
+                           gfp_t gfp_flags);
+void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds);
+void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds);
+int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
+                            struct nfs4_ff_layout_mirror *mirror, u64 offset,
+                            u64 length, int status, enum nfs_opnum4 opnum,
+                            gfp_t gfp_flags);
+int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
+                             struct xdr_stream *xdr, int *count,
+                             const struct pnfs_layout_range *range);
+struct nfs_fh *
+nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx);
+
+struct nfs4_pnfs_ds *
+nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
+                         bool fail_return);
+
+struct rpc_clnt *
+nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg,
+                                u32 ds_idx,
+                                struct nfs_client *ds_clp,
+                                struct inode *inode);
+struct rpc_cred *ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg,
+                                      u32 ds_idx, struct rpc_cred *mdscred);
+bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg);
+#endif /* FS_NFS_NFS4FLEXFILELAYOUT_H */
diff --git a/fs/nfs/flexfilelayout/flexfilelayoutdev.c b/fs/nfs/flexfilelayout/flexfilelayoutdev.c
new file mode 100644 (file)
index 0000000..3bbb16b
--- /dev/null
@@ -0,0 +1,552 @@
+/*
+ * Device operations for the pnfs nfs4 file layout driver.
+ *
+ * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
+ *
+ * Tao Peng <bergwolf@primarydata.com>
+ */
+
+#include <linux/nfs_fs.h>
+#include <linux/vmalloc.h>
+#include <linux/module.h>
+#include <linux/sunrpc/addr.h>
+
+#include "../internal.h"
+#include "../nfs4session.h"
+#include "flexfilelayout.h"
+
+#define NFSDBG_FACILITY                NFSDBG_PNFS_LD
+
+static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
+static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
+
+void nfs4_ff_layout_put_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
+{
+       if (mirror_ds)
+               nfs4_put_deviceid_node(&mirror_ds->id_node);
+}
+
+void nfs4_ff_layout_free_deviceid(struct nfs4_ff_layout_ds *mirror_ds)
+{
+       nfs4_print_deviceid(&mirror_ds->id_node.deviceid);
+       nfs4_pnfs_ds_put(mirror_ds->ds);
+       kfree(mirror_ds);
+}
+
+/* Decode opaque device data and construct new_ds using it */
+struct nfs4_ff_layout_ds *
+nfs4_ff_alloc_deviceid_node(struct nfs_server *server, struct pnfs_device *pdev,
+                           gfp_t gfp_flags)
+{
+       struct xdr_stream stream;
+       struct xdr_buf buf;
+       struct page *scratch;
+       struct list_head dsaddrs;
+       struct nfs4_pnfs_ds_addr *da;
+       struct nfs4_ff_layout_ds *new_ds = NULL;
+       struct nfs4_ff_ds_version *ds_versions = NULL;
+       u32 mp_count;
+       u32 version_count;
+       __be32 *p;
+       int i, ret = -ENOMEM;
+
+       /* set up xdr stream */
+       scratch = alloc_page(gfp_flags);
+       if (!scratch)
+               goto out_err;
+
+       new_ds = kzalloc(sizeof(struct nfs4_ff_layout_ds), gfp_flags);
+       if (!new_ds)
+               goto out_scratch;
+
+       nfs4_init_deviceid_node(&new_ds->id_node,
+                               server,
+                               &pdev->dev_id);
+       INIT_LIST_HEAD(&dsaddrs);
+
+       xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
+       xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
+
+       /* multipath count */
+       p = xdr_inline_decode(&stream, 4);
+       if (unlikely(!p))
+               goto out_err_drain_dsaddrs;
+       mp_count = be32_to_cpup(p);
+       dprintk("%s: multipath ds count %d\n", __func__, mp_count);
+
+       for (i = 0; i < mp_count; i++) {
+               /* multipath ds */
+               da = nfs4_decode_mp_ds_addr(server->nfs_client->cl_net,
+                                           &stream, gfp_flags);
+               if (da)
+                       list_add_tail(&da->da_node, &dsaddrs);
+       }
+       if (list_empty(&dsaddrs)) {
+               dprintk("%s: no suitable DS addresses found\n",
+                       __func__);
+               ret = -ENOMEDIUM;
+               goto out_err_drain_dsaddrs;
+       }
+
+       /* version count */
+       p = xdr_inline_decode(&stream, 4);
+       if (unlikely(!p))
+               goto out_err_drain_dsaddrs;
+       version_count = be32_to_cpup(p);
+       dprintk("%s: version count %d\n", __func__, version_count);
+
+       ds_versions = kzalloc(version_count * sizeof(struct nfs4_ff_ds_version),
+                             gfp_flags);
+       if (!ds_versions)
+               goto out_scratch;
+
+       for (i = 0; i < version_count; i++) {
+               /* 20 = version(4) + minor_version(4) + rsize(4) + wsize(4) +
+                * tightly_coupled(4) */
+               p = xdr_inline_decode(&stream, 20);
+               if (unlikely(!p))
+                       goto out_err_drain_dsaddrs;
+               ds_versions[i].version = be32_to_cpup(p++);
+               ds_versions[i].minor_version = be32_to_cpup(p++);
+               ds_versions[i].rsize = nfs_block_size(be32_to_cpup(p++), NULL);
+               ds_versions[i].wsize = nfs_block_size(be32_to_cpup(p++), NULL);
+               ds_versions[i].tightly_coupled = be32_to_cpup(p);
+
+               if (ds_versions[i].rsize > NFS_MAX_FILE_IO_SIZE)
+                       ds_versions[i].rsize = NFS_MAX_FILE_IO_SIZE;
+               if (ds_versions[i].wsize > NFS_MAX_FILE_IO_SIZE)
+                       ds_versions[i].wsize = NFS_MAX_FILE_IO_SIZE;
+
+               if (ds_versions[i].version != 3 || ds_versions[i].minor_version != 0) {
+                       dprintk("%s: [%d] unsupported ds version %d-%d\n", __func__,
+                               i, ds_versions[i].version,
+                               ds_versions[i].minor_version);
+                       ret = -EPROTONOSUPPORT;
+                       goto out_err_drain_dsaddrs;
+               }
+
+               dprintk("%s: [%d] vers %u minor_ver %u rsize %u wsize %u coupled %d\n",
+                       __func__, i, ds_versions[i].version,
+                       ds_versions[i].minor_version,
+                       ds_versions[i].rsize,
+                       ds_versions[i].wsize,
+                       ds_versions[i].tightly_coupled);
+       }
+
+       new_ds->ds_versions = ds_versions;
+       new_ds->ds_versions_cnt = version_count;
+
+       new_ds->ds = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
+       if (!new_ds->ds)
+               goto out_err_drain_dsaddrs;
+
+       /* If DS was already in cache, free ds addrs */
+       while (!list_empty(&dsaddrs)) {
+               da = list_first_entry(&dsaddrs,
+                                     struct nfs4_pnfs_ds_addr,
+                                     da_node);
+               list_del_init(&da->da_node);
+               kfree(da->da_remotestr);
+               kfree(da);
+       }
+
+       __free_page(scratch);
+       return new_ds;
+
+out_err_drain_dsaddrs:
+       while (!list_empty(&dsaddrs)) {
+               da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
+                                     da_node);
+               list_del_init(&da->da_node);
+               kfree(da->da_remotestr);
+               kfree(da);
+       }
+
+       kfree(ds_versions);
+out_scratch:
+       __free_page(scratch);
+out_err:
+       kfree(new_ds);
+
+       dprintk("%s ERROR: returning %d\n", __func__, ret);
+       return NULL;
+}
+
+static u64
+end_offset(u64 start, u64 len)
+{
+       u64 end;
+
+       end = start + len;
+       return end >= start ? end : NFS4_MAX_UINT64;
+}
+
+static void extend_ds_error(struct nfs4_ff_layout_ds_err *err,
+                           u64 offset, u64 length)
+{
+       u64 end;
+
+       end = max_t(u64, end_offset(err->offset, err->length),
+                   end_offset(offset, length));
+       err->offset = min_t(u64, err->offset, offset);
+       err->length = end - err->offset;
+}
+
+static bool ds_error_can_merge(struct nfs4_ff_layout_ds_err *err,  u64 offset,
+                              u64 length, int status, enum nfs_opnum4 opnum,
+                              nfs4_stateid *stateid,
+                              struct nfs4_deviceid *deviceid)
+{
+       return err->status == status && err->opnum == opnum &&
+              nfs4_stateid_match(&err->stateid, stateid) &&
+              !memcmp(&err->deviceid, deviceid, sizeof(*deviceid)) &&
+              end_offset(err->offset, err->length) >= offset &&
+              err->offset <= end_offset(offset, length);
+}
+
+static bool merge_ds_error(struct nfs4_ff_layout_ds_err *old,
+                          struct nfs4_ff_layout_ds_err *new)
+{
+       if (!ds_error_can_merge(old, new->offset, new->length, new->status,
+                               new->opnum, &new->stateid, &new->deviceid))
+               return false;
+
+       extend_ds_error(old, new->offset, new->length);
+       return true;
+}
+
+static bool
+ff_layout_add_ds_error_locked(struct nfs4_flexfile_layout *flo,
+                             struct nfs4_ff_layout_ds_err *dserr)
+{
+       struct nfs4_ff_layout_ds_err *err;
+
+       list_for_each_entry(err, &flo->error_list, list) {
+               if (merge_ds_error(err, dserr)) {
+                       return true;
+               }
+       }
+
+       list_add(&dserr->list, &flo->error_list);
+       return false;
+}
+
+static bool
+ff_layout_update_ds_error(struct nfs4_flexfile_layout *flo, u64 offset,
+                         u64 length, int status, enum nfs_opnum4 opnum,
+                         nfs4_stateid *stateid, struct nfs4_deviceid *deviceid)
+{
+       bool found = false;
+       struct nfs4_ff_layout_ds_err *err;
+
+       list_for_each_entry(err, &flo->error_list, list) {
+               if (ds_error_can_merge(err, offset, length, status, opnum,
+                                      stateid, deviceid)) {
+                       found = true;
+                       extend_ds_error(err, offset, length);
+                       break;
+               }
+       }
+
+       return found;
+}
+
+int ff_layout_track_ds_error(struct nfs4_flexfile_layout *flo,
+                            struct nfs4_ff_layout_mirror *mirror, u64 offset,
+                            u64 length, int status, enum nfs_opnum4 opnum,
+                            gfp_t gfp_flags)
+{
+       struct nfs4_ff_layout_ds_err *dserr;
+       bool needfree;
+
+       if (status == 0)
+               return 0;
+
+       if (mirror->mirror_ds == NULL)
+               return -EINVAL;
+
+       spin_lock(&flo->generic_hdr.plh_inode->i_lock);
+       if (ff_layout_update_ds_error(flo, offset, length, status, opnum,
+                                     &mirror->stateid,
+                                     &mirror->mirror_ds->id_node.deviceid)) {
+               spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
+               return 0;
+       }
+       spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
+       dserr = kmalloc(sizeof(*dserr), gfp_flags);
+       if (!dserr)
+               return -ENOMEM;
+
+       INIT_LIST_HEAD(&dserr->list);
+       dserr->offset = offset;
+       dserr->length = length;
+       dserr->status = status;
+       dserr->opnum = opnum;
+       nfs4_stateid_copy(&dserr->stateid, &mirror->stateid);
+       memcpy(&dserr->deviceid, &mirror->mirror_ds->id_node.deviceid,
+              NFS4_DEVICEID4_SIZE);
+
+       spin_lock(&flo->generic_hdr.plh_inode->i_lock);
+       needfree = ff_layout_add_ds_error_locked(flo, dserr);
+       spin_unlock(&flo->generic_hdr.plh_inode->i_lock);
+       if (needfree)
+               kfree(dserr);
+
+       return 0;
+}
+
+/* currently we only support AUTH_NONE and AUTH_SYS */
+static rpc_authflavor_t
+nfs4_ff_layout_choose_authflavor(struct nfs4_ff_layout_mirror *mirror)
+{
+       if (mirror->uid == (u32)-1)
+               return RPC_AUTH_NULL;
+       return RPC_AUTH_UNIX;
+}
+
+/* fetch cred for NFSv3 DS */
+static int ff_layout_update_mirror_cred(struct nfs4_ff_layout_mirror *mirror,
+                                     struct nfs4_pnfs_ds *ds)
+{
+       if (ds->ds_clp && !mirror->cred &&
+           mirror->mirror_ds->ds_versions[0].version == 3) {
+               struct rpc_auth *auth = ds->ds_clp->cl_rpcclient->cl_auth;
+               struct rpc_cred *cred;
+               struct auth_cred acred = {
+                       .uid = make_kuid(&init_user_ns, mirror->uid),
+                       .gid = make_kgid(&init_user_ns, mirror->gid),
+               };
+
+               /* AUTH_NULL ignores acred */
+               cred = auth->au_ops->lookup_cred(auth, &acred, 0);
+               if (IS_ERR(cred)) {
+                       dprintk("%s: lookup_cred failed with %ld\n",
+                               __func__, PTR_ERR(cred));
+                       return PTR_ERR(cred);
+               } else {
+                       mirror->cred = cred;
+               }
+       }
+       return 0;
+}
+
+struct nfs_fh *
+nfs4_ff_layout_select_ds_fh(struct pnfs_layout_segment *lseg, u32 mirror_idx)
+{
+       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, mirror_idx);
+       struct nfs_fh *fh = NULL;
+       struct nfs4_deviceid_node *devid;
+
+       if (mirror == NULL || mirror->mirror_ds == NULL ||
+           mirror->mirror_ds->ds == NULL) {
+               printk(KERN_ERR "NFS: %s: No data server for mirror offset index %d\n",
+                       __func__, mirror_idx);
+               if (mirror && mirror->mirror_ds) {
+                       devid = &mirror->mirror_ds->id_node;
+                       pnfs_generic_mark_devid_invalid(devid);
+               }
+               goto out;
+       }
+
+       /* FIXME: For now assume there is only 1 version available for the DS */
+       fh = &mirror->fh_versions[0];
+out:
+       return fh;
+}
+
+/* Upon return, either ds is connected, or ds is NULL */
+struct nfs4_pnfs_ds *
+nfs4_ff_layout_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx,
+                         bool fail_return)
+{
+       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
+       struct nfs4_pnfs_ds *ds = NULL;
+       struct nfs4_deviceid_node *devid;
+       struct inode *ino = lseg->pls_layout->plh_inode;
+       struct nfs_server *s = NFS_SERVER(ino);
+       unsigned int max_payload;
+       rpc_authflavor_t flavor;
+
+       if (mirror == NULL || mirror->mirror_ds == NULL ||
+           mirror->mirror_ds->ds == NULL) {
+               printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
+                       __func__, ds_idx);
+               if (mirror && mirror->mirror_ds) {
+                       devid = &mirror->mirror_ds->id_node;
+                       pnfs_generic_mark_devid_invalid(devid);
+               }
+               goto out;
+       }
+
+       devid = &mirror->mirror_ds->id_node;
+       if (ff_layout_test_devid_unavailable(devid))
+               goto out;
+
+       ds = mirror->mirror_ds->ds;
+       /* matching smp_wmb() in _nfs4_pnfs_v3/4_ds_connect */
+       smp_rmb();
+       if (ds->ds_clp)
+               goto out;
+
+       flavor = nfs4_ff_layout_choose_authflavor(mirror);
+
+       /* FIXME: For now we assume the server sent only one version of NFS
+        * to use for the DS.
+        */
+       nfs4_pnfs_ds_connect(s, ds, devid, dataserver_timeo,
+                            dataserver_retrans,
+                            mirror->mirror_ds->ds_versions[0].version,
+                            mirror->mirror_ds->ds_versions[0].minor_version,
+                            flavor);
+
+       /* connect success, check rsize/wsize limit */
+       if (ds->ds_clp) {
+               max_payload =
+                       nfs_block_size(rpc_max_payload(ds->ds_clp->cl_rpcclient),
+                                      NULL);
+               if (mirror->mirror_ds->ds_versions[0].rsize > max_payload)
+                       mirror->mirror_ds->ds_versions[0].rsize = max_payload;
+               if (mirror->mirror_ds->ds_versions[0].wsize > max_payload)
+                       mirror->mirror_ds->ds_versions[0].wsize = max_payload;
+       } else {
+               ff_layout_track_ds_error(FF_LAYOUT_FROM_HDR(lseg->pls_layout),
+                                        mirror, lseg->pls_range.offset,
+                                        lseg->pls_range.length, NFS4ERR_NXIO,
+                                        OP_ILLEGAL, GFP_NOIO);
+               if (fail_return) {
+                       pnfs_error_mark_layout_for_return(ino, lseg);
+                       if (ff_layout_has_available_ds(lseg))
+                               pnfs_set_retry_layoutget(lseg->pls_layout);
+                       else
+                               pnfs_clear_retry_layoutget(lseg->pls_layout);
+
+               } else {
+                       if (ff_layout_has_available_ds(lseg))
+                               set_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE,
+                                       &lseg->pls_layout->plh_flags);
+                       else {
+                               pnfs_error_mark_layout_for_return(ino, lseg);
+                               pnfs_clear_retry_layoutget(lseg->pls_layout);
+                       }
+               }
+       }
+
+       if (ff_layout_update_mirror_cred(mirror, ds))
+               ds = NULL;
+out:
+       return ds;
+}
+
+struct rpc_cred *
+ff_layout_get_ds_cred(struct pnfs_layout_segment *lseg, u32 ds_idx,
+                     struct rpc_cred *mdscred)
+{
+       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
+       struct rpc_cred *cred = ERR_PTR(-EINVAL);
+
+       if (!nfs4_ff_layout_prepare_ds(lseg, ds_idx, true))
+               goto out;
+
+       if (mirror && mirror->cred)
+               cred = mirror->cred;
+       else
+               cred = mdscred;
+out:
+       return cred;
+}
+
+/**
+* Find or create a DS rpc client with th MDS server rpc client auth flavor
+* in the nfs_client cl_ds_clients list.
+*/
+struct rpc_clnt *
+nfs4_ff_find_or_create_ds_client(struct pnfs_layout_segment *lseg, u32 ds_idx,
+                                struct nfs_client *ds_clp, struct inode *inode)
+{
+       struct nfs4_ff_layout_mirror *mirror = FF_LAYOUT_COMP(lseg, ds_idx);
+
+       switch (mirror->mirror_ds->ds_versions[0].version) {
+       case 3:
+               /* For NFSv3 DS, flavor is set when creating DS connections */
+               return ds_clp->cl_rpcclient;
+       case 4:
+               return nfs4_find_or_create_ds_client(ds_clp, inode);
+       default:
+               BUG();
+       }
+}
+
+static bool is_range_intersecting(u64 offset1, u64 length1,
+                                 u64 offset2, u64 length2)
+{
+       u64 end1 = end_offset(offset1, length1);
+       u64 end2 = end_offset(offset2, length2);
+
+       return (end1 == NFS4_MAX_UINT64 || end1 > offset2) &&
+              (end2 == NFS4_MAX_UINT64 || end2 > offset1);
+}
+
+/* called with inode i_lock held */
+int ff_layout_encode_ds_ioerr(struct nfs4_flexfile_layout *flo,
+                             struct xdr_stream *xdr, int *count,
+                             const struct pnfs_layout_range *range)
+{
+       struct nfs4_ff_layout_ds_err *err, *n;
+       __be32 *p;
+
+       list_for_each_entry_safe(err, n, &flo->error_list, list) {
+               if (!is_range_intersecting(err->offset, err->length,
+                                          range->offset, range->length))
+                       continue;
+               /* offset(8) + length(8) + stateid(NFS4_STATEID_SIZE)
+                * + deviceid(NFS4_DEVICEID4_SIZE) + status(4) + opnum(4)
+                */
+               p = xdr_reserve_space(xdr,
+                               24 + NFS4_STATEID_SIZE + NFS4_DEVICEID4_SIZE);
+               if (unlikely(!p))
+                       return -ENOBUFS;
+               p = xdr_encode_hyper(p, err->offset);
+               p = xdr_encode_hyper(p, err->length);
+               p = xdr_encode_opaque_fixed(p, &err->stateid,
+                                           NFS4_STATEID_SIZE);
+               p = xdr_encode_opaque_fixed(p, &err->deviceid,
+                                           NFS4_DEVICEID4_SIZE);
+               *p++ = cpu_to_be32(err->status);
+               *p++ = cpu_to_be32(err->opnum);
+               *count += 1;
+               list_del(&err->list);
+               kfree(err);
+               dprintk("%s: offset %llu length %llu status %d op %d count %d\n",
+                       __func__, err->offset, err->length, err->status,
+                       err->opnum, *count);
+       }
+
+       return 0;
+}
+
+bool ff_layout_has_available_ds(struct pnfs_layout_segment *lseg)
+{
+       struct nfs4_ff_layout_mirror *mirror;
+       struct nfs4_deviceid_node *devid;
+       int idx;
+
+       for (idx = 0; idx < FF_LAYOUT_MIRROR_COUNT(lseg); idx++) {
+               mirror = FF_LAYOUT_COMP(lseg, idx);
+               if (mirror && mirror->mirror_ds) {
+                       devid = &mirror->mirror_ds->id_node;
+                       if (!ff_layout_test_devid_unavailable(devid))
+                               return true;
+               }
+       }
+
+       return false;
+}
+
+module_param(dataserver_retrans, uint, 0644);
+MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
+                       "retries a request before it attempts further "
+                       " recovery  action.");
+module_param(dataserver_timeo, uint, 0644);
+MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
+                       "NFSv4.1  client  waits for a response from a "
+                       " data server before it retries an NFS request.");
index 2f5db844c172534a229d86b011be1f6ec2c29aff..857e2a99acc8d30e6bcc85a7b7eb9e0b193ae5bc 100644 (file)
@@ -152,7 +152,7 @@ void nfs_fattr_map_and_free_names(struct nfs_server *server, struct nfs_fattr *f
                nfs_fattr_free_group_name(fattr);
 }
 
-static int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *res)
+int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *res)
 {
        unsigned long val;
        char buf[16];
@@ -166,6 +166,7 @@ static int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *re
        *res = val;
        return 1;
 }
+EXPORT_SYMBOL_GPL(nfs_map_string_to_numeric);
 
 static int nfs_map_numeric_to_string(__u32 id, char *buf, size_t buflen)
 {
index 44c600aac90752fe9867b3671ef00f07997a00be..ca6dda0f68bb5ec612e700d5e4d6c8481bd13634 100644 (file)
@@ -7796,9 +7796,7 @@ static void nfs4_layoutreturn_release(void *calldata)
        spin_lock(&lo->plh_inode->i_lock);
        if (lrp->res.lrs_present)
                pnfs_set_layout_stateid(lo, &lrp->res.stateid, true);
-       clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
-       smp_mb__after_atomic();
-       wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
+       pnfs_clear_layoutreturn_waitbit(lo);
        clear_bit(NFS_LAYOUT_RETURN_BEFORE_CLOSE, &lo->plh_flags);
        rpc_wake_up(&NFS_SERVER(lo->plh_inode)->roc_rpcwaitq);
        lo->plh_block_lgets--;
index c4c9fe606ae66ea3f9896992c80c74937e301511..0fb0f1920a1f5b9048397155038351c83be7aba5 100644 (file)
@@ -910,7 +910,9 @@ send_layoutget(struct pnfs_layout_hdr *lo,
                        pnfs_layout_io_set_failed(lo, range->iomode);
                }
                return NULL;
-       }
+       } else
+               pnfs_layout_clear_fail_bit(lo,
+                               pnfs_iomode_to_fail_bit(range->iomode));
 
        return lseg;
 }
@@ -930,6 +932,13 @@ static void pnfs_clear_layoutcommit(struct inode *inode,
        }
 }
 
+void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo)
+{
+       clear_bit_unlock(NFS_LAYOUT_RETURN, &lo->plh_flags);
+       smp_mb__after_atomic();
+       wake_up_bit(&lo->plh_flags, NFS_LAYOUT_RETURN);
+}
+
 static int
 pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
                       enum pnfs_iomode iomode, bool sync)
@@ -943,6 +952,7 @@ pnfs_send_layoutreturn(struct pnfs_layout_hdr *lo, nfs4_stateid stateid,
                status = -ENOMEM;
                spin_lock(&ino->i_lock);
                lo->plh_block_lgets--;
+               pnfs_clear_layoutreturn_waitbit(lo);
                rpc_wake_up(&NFS_SERVER(ino)->roc_rpcwaitq);
                spin_unlock(&ino->i_lock);
                pnfs_put_layout_hdr(lo);
@@ -1418,6 +1428,15 @@ static bool pnfs_prepare_to_retry_layoutget(struct pnfs_layout_hdr *lo)
                                   TASK_UNINTERRUPTIBLE);
 }
 
+static void pnfs_clear_first_layoutget(struct pnfs_layout_hdr *lo)
+{
+       unsigned long *bitlock = &lo->plh_flags;
+
+       clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
+       smp_mb__after_atomic();
+       wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
+}
+
 /*
  * Layout segment is retreived from the server if not cached.
  * The appropriate layout segment is referenced and returned to the caller.
@@ -1499,6 +1518,8 @@ lookup_again:
                spin_unlock(&ino->i_lock);
                dprintk("%s wait for layoutreturn\n", __func__);
                if (pnfs_prepare_to_retry_layoutget(lo)) {
+                       if (first)
+                               pnfs_clear_first_layoutget(lo);
                        pnfs_put_layout_hdr(lo);
                        dprintk("%s retrying\n", __func__);
                        goto lookup_again;
@@ -1533,13 +1554,8 @@ lookup_again:
        pnfs_clear_retry_layoutget(lo);
        atomic_dec(&lo->plh_outstanding);
 out_put_layout_hdr:
-       if (first) {
-               unsigned long *bitlock = &lo->plh_flags;
-
-               clear_bit_unlock(NFS_LAYOUT_FIRST_LAYOUTGET, bitlock);
-               smp_mb__after_atomic();
-               wake_up_bit(bitlock, NFS_LAYOUT_FIRST_LAYOUTGET);
-       }
+       if (first)
+               pnfs_clear_first_layoutget(lo);
        pnfs_put_layout_hdr(lo);
 out:
        dprintk("%s: inode %s/%llu pNFS layout segment %s for "
index 49a4667084001cc8bde8a1af01bc9b2dce697021..7642021484bfe38803716e69346e8ae8ddc4dd5d 100644 (file)
@@ -278,6 +278,7 @@ struct pnfs_layout_segment *pnfs_update_layout(struct inode *ino,
                                               u64 count,
                                               enum pnfs_iomode iomode,
                                               gfp_t gfp_flags);
+void pnfs_clear_layoutreturn_waitbit(struct pnfs_layout_hdr *lo);
 
 void nfs4_deviceid_mark_client_invalid(struct nfs_client *clp);
 int pnfs_read_done_resend_to_mds(struct nfs_pgio_header *);
index 022b761dbf0a187478680fe3cc35198e145c11a5..de7c91ca427e6e078afa6e0cd45b2d4424214c82 100644 (file)
@@ -516,6 +516,7 @@ enum pnfs_layouttype {
        LAYOUT_NFSV4_1_FILES  = 1,
        LAYOUT_OSD2_OBJECTS = 2,
        LAYOUT_BLOCK_VOLUME = 3,
+       LAYOUT_FLEX_FILES = 4,
 };
 
 /* used for both layout return and recall */
index 0f4b79da65845504c0ffca60a9731b041e21d917..333844e38f66c4918d3693a96cfdb0b50f9b031d 100644 (file)
@@ -73,5 +73,7 @@ int nfs_map_group_to_gid(const struct nfs_server *, const char *, size_t, kgid_t
 int nfs_map_uid_to_name(const struct nfs_server *, kuid_t, char *, size_t);
 int nfs_map_gid_to_group(const struct nfs_server *, kgid_t, char *, size_t);
 
+int nfs_map_string_to_numeric(const char *name, size_t namelen, __u32 *res);
+
 extern unsigned int nfs_idmap_cache_timeout;
 #endif /* NFS_IDMAP_H */
index 89f2ca178873483b79944eb43e0dfb55628ff8a5..7e61a17030a4843bf0f80fda4dc2ae9f379409ef 100644 (file)
@@ -89,6 +89,8 @@ void                  rpc_free_iostats(struct rpc_iostats *);
 static inline struct rpc_iostats *rpc_alloc_iostats(struct rpc_clnt *clnt) { return NULL; }
 static inline void rpc_count_iostats(const struct rpc_task *task,
                                     struct rpc_iostats *stats) {}
+static inline void rpc_count_iostats_metrics(const struct rpc_task *,
+                                            struct rpc_iostats *) {}
 static inline void rpc_print_iostats(struct seq_file *seq, struct rpc_clnt *clnt) {}
 static inline void rpc_free_iostats(struct rpc_iostats *stats) {}