]> git.karo-electronics.de Git - mv-sheeva.git/commitdiff
NFSv4.1: pnfs: filelayout: add driver's LAYOUTGET and GETDEVICEINFO infrastructure
authorAndy Adamson <andros@netapp.com>
Wed, 20 Oct 2010 04:18:04 +0000 (00:18 -0400)
committerTrond Myklebust <Trond.Myklebust@netapp.com>
Sun, 24 Oct 2010 22:07:11 +0000 (18:07 -0400)
Implement the driver's io_ops->alloc_lseg and free_lseg functions,
which integrate into the deviceid cache and calls out to
nfs4_proc_getdeviceinfo when necessary.

Signed-off-by: Andy Adamson <andros@netapp.com>
Signed-off-by: Dean Hildebrand <dhildebz@umich.edu>
Signed-off-by: Marc Eshel <eshel@almaden.ibm.com>
Signed-off-by: Mike Sager <sager@netapp.com>
Signed-off-by: Oleg Drokin <green@linuxhacker.ru>
Signed-off-by: Ricardo Labiaga <ricardo.labiaga@netapp.com>
Signed-off-by: Tao Guo <guotao@nrchpc.ac.cn>
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Signed-off-by: Fred Isaman <iisaman@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
fs/nfs/Makefile
fs/nfs/client.c
fs/nfs/nfs4filelayout.c
fs/nfs/nfs4filelayout.h [new file with mode: 0644]
fs/nfs/nfs4filelayoutdev.c [new file with mode: 0644]

index 08a8889c514924bde9fd0bbdf939e2efd958bd54..4776ff9e38143d5f3c01744d8cd7261c2a62553c 100644 (file)
@@ -20,4 +20,4 @@ nfs-$(CONFIG_SYSCTL) += sysctl.o
 nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
 
 obj-$(CONFIG_PNFS_FILE_LAYOUT) += nfs_layout_nfsv41_files.o
-nfs_layout_nfsv41_files-y := nfs4filelayout.o
+nfs_layout_nfsv41_files-y := nfs4filelayout.o nfs4filelayoutdev.o
index e55ad211dbed3ec4f0564823097f936913dbfc9a..8934b2e339df5acfe39f04bfaa213c168e71d652 100644 (file)
@@ -255,6 +255,7 @@ void nfs_put_client(struct nfs_client *clp)
                nfs_free_client(clp);
        }
 }
+EXPORT_SYMBOL_GPL(nfs_put_client);
 
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 /*
index 696e283c26324f086f7d6e05c4b29852c9385915..9e82b51cb515ccda778088f921df87b520cd6200 100644 (file)
@@ -30,7 +30,9 @@
  */
 
 #include <linux/nfs_fs.h>
-#include "pnfs.h"
+
+#include "internal.h"
+#include "nfs4filelayout.h"
 
 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
 
@@ -41,23 +43,223 @@ MODULE_DESCRIPTION("The NFSv4 file layout driver");
 int
 filelayout_initialize_mountpoint(struct nfs_server *nfss)
 {
+       int status = pnfs_alloc_init_deviceid_cache(nfss->nfs_client,
+                                               nfs4_fl_free_deviceid_callback);
+       if (status) {
+               printk(KERN_WARNING "%s: deviceid cache could not be "
+                       "initialized\n", __func__);
+               return status;
+       }
+       dprintk("%s: deviceid cache has been initialized successfully\n",
+               __func__);
        return 0;
 }
 
+/* Uninitialize a mountpoint by destroying its device list */
 int
 filelayout_uninitialize_mountpoint(struct nfs_server *nfss)
 {
        dprintk("--> %s\n", __func__);
 
+       if (nfss->nfs_client->cl_devid_cache)
+               pnfs_put_deviceid_cache(nfss->nfs_client);
        return 0;
 }
 
+/*
+ * filelayout_check_layout()
+ *
+ * Make sure layout segment parameters are sane WRT the device.
+ * At this point no generic layer initialization of the lseg has occurred,
+ * and nothing has been added to the layout_hdr cache.
+ *
+ */
+static int
+filelayout_check_layout(struct pnfs_layout_hdr *lo,
+                       struct nfs4_filelayout_segment *fl,
+                       struct nfs4_layoutget_res *lgr,
+                       struct nfs4_deviceid *id)
+{
+       struct nfs4_file_layout_dsaddr *dsaddr;
+       int status = -EINVAL;
+       struct nfs_server *nfss = NFS_SERVER(lo->inode);
+
+       dprintk("--> %s\n", __func__);
+
+       if (fl->pattern_offset > lgr->range.offset) {
+               dprintk("%s pattern_offset %lld to large\n",
+                               __func__, fl->pattern_offset);
+               goto out;
+       }
+
+       if (fl->stripe_unit % PAGE_SIZE) {
+               dprintk("%s Stripe unit (%u) not page aligned\n",
+                       __func__, fl->stripe_unit);
+               goto out;
+       }
+
+       /* find and reference the deviceid */
+       dsaddr = nfs4_fl_find_get_deviceid(nfss->nfs_client, id);
+       if (dsaddr == NULL) {
+               dsaddr = get_device_info(lo->inode, id);
+               if (dsaddr == NULL)
+                       goto out;
+       }
+       fl->dsaddr = dsaddr;
+
+       if (fl->first_stripe_index < 0 ||
+           fl->first_stripe_index >= dsaddr->stripe_count) {
+               dprintk("%s Bad first_stripe_index %d\n",
+                               __func__, fl->first_stripe_index);
+               goto out_put;
+       }
+
+       if ((fl->stripe_type == STRIPE_SPARSE &&
+           fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
+           (fl->stripe_type == STRIPE_DENSE &&
+           fl->num_fh != dsaddr->stripe_count)) {
+               dprintk("%s num_fh %u not valid for given packing\n",
+                       __func__, fl->num_fh);
+               goto out_put;
+       }
+
+       if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
+               dprintk("%s Stripe unit (%u) not aligned with rsize %u "
+                       "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
+                       nfss->wsize);
+       }
+
+       status = 0;
+out:
+       dprintk("--> %s returns %d\n", __func__, status);
+       return status;
+out_put:
+       pnfs_put_deviceid(nfss->nfs_client->cl_devid_cache, &dsaddr->deviceid);
+       goto out;
+}
+
+static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
+{
+       int i;
+
+       for (i = 0; i < fl->num_fh; i++) {
+               if (!fl->fh_array[i])
+                       break;
+               kfree(fl->fh_array[i]);
+       }
+       kfree(fl->fh_array);
+       fl->fh_array = NULL;
+}
+
+static void
+_filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
+{
+       filelayout_free_fh_array(fl);
+       kfree(fl);
+}
+
+static int
+filelayout_decode_layout(struct pnfs_layout_hdr *flo,
+                        struct nfs4_filelayout_segment *fl,
+                        struct nfs4_layoutget_res *lgr,
+                        struct nfs4_deviceid *id)
+{
+       uint32_t *p = (uint32_t *)lgr->layout.buf;
+       uint32_t nfl_util;
+       int i;
+
+       dprintk("%s: set_layout_map Begin\n", __func__);
+
+       memcpy(id, p, sizeof(*id));
+       p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
+       print_deviceid(id);
+
+       nfl_util = be32_to_cpup(p++);
+       if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
+               fl->commit_through_mds = 1;
+       if (nfl_util & NFL4_UFLG_DENSE)
+               fl->stripe_type = STRIPE_DENSE;
+       else
+               fl->stripe_type = STRIPE_SPARSE;
+       fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
+
+       fl->first_stripe_index = be32_to_cpup(p++);
+       p = xdr_decode_hyper(p, &fl->pattern_offset);
+       fl->num_fh = be32_to_cpup(p++);
+
+       dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
+               __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
+               fl->pattern_offset);
+
+       fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
+                              GFP_KERNEL);
+       if (!fl->fh_array)
+               return -ENOMEM;
+
+       for (i = 0; i < fl->num_fh; i++) {
+               /* Do we want to use a mempool here? */
+               fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
+               if (!fl->fh_array[i]) {
+                       filelayout_free_fh_array(fl);
+                       return -ENOMEM;
+               }
+               fl->fh_array[i]->size = be32_to_cpup(p++);
+               if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
+                       printk(KERN_ERR "Too big fh %d received %d\n",
+                              i, fl->fh_array[i]->size);
+                       filelayout_free_fh_array(fl);
+                       return -EIO;
+               }
+               memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
+               p += XDR_QUADLEN(fl->fh_array[i]->size);
+               dprintk("DEBUG: %s: fh len %d\n", __func__,
+                       fl->fh_array[i]->size);
+       }
+
+       return 0;
+}
+
+static struct pnfs_layout_segment *
+filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
+                     struct nfs4_layoutget_res *lgr)
+{
+       struct nfs4_filelayout_segment *fl;
+       int rc;
+       struct nfs4_deviceid id;
+
+       dprintk("--> %s\n", __func__);
+       fl = kzalloc(sizeof(*fl), GFP_KERNEL);
+       if (!fl)
+               return NULL;
+
+       rc = filelayout_decode_layout(layoutid, fl, lgr, &id);
+       if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id)) {
+               _filelayout_free_lseg(fl);
+               return NULL;
+       }
+       return &fl->generic_hdr;
+}
+
+static void
+filelayout_free_lseg(struct pnfs_layout_segment *lseg)
+{
+       struct nfs_server *nfss = NFS_SERVER(lseg->layout->inode);
+       struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
+
+       dprintk("--> %s\n", __func__);
+       pnfs_put_deviceid(nfss->nfs_client->cl_devid_cache,
+                         &fl->dsaddr->deviceid);
+       _filelayout_free_lseg(fl);
+}
+
 static struct pnfs_layoutdriver_type filelayout_type = {
        .id = LAYOUT_NFSV4_1_FILES,
        .name = "LAYOUT_NFSV4_1_FILES",
        .owner = THIS_MODULE,
        .initialize_mountpoint   = filelayout_initialize_mountpoint,
        .uninitialize_mountpoint = filelayout_uninitialize_mountpoint,
+       .alloc_lseg              = filelayout_alloc_lseg,
+       .free_lseg               = filelayout_free_lseg,
 };
 
 static int __init nfs4filelayout_init(void)
diff --git a/fs/nfs/nfs4filelayout.h b/fs/nfs/nfs4filelayout.h
new file mode 100644 (file)
index 0000000..bbf60dd
--- /dev/null
@@ -0,0 +1,94 @@
+/*
+ *  NFSv4 file layout driver data structures.
+ *
+ *  Copyright (c) 2002
+ *  The Regents of the University of Michigan
+ *  All Rights Reserved
+ *
+ *  Dean Hildebrand <dhildebz@umich.edu>
+ *
+ *  Permission is granted to use, copy, create derivative works, and
+ *  redistribute this software and such derivative works for any purpose,
+ *  so long as the name of the University of Michigan is not used in
+ *  any advertising or publicity pertaining to the use or distribution
+ *  of this software without specific, written prior authorization. If
+ *  the above copyright notice or any other identification of the
+ *  University of Michigan is included in any copy of any portion of
+ *  this software, then the disclaimer below must also be included.
+ *
+ *  This software is provided as is, without representation or warranty
+ *  of any kind either express or implied, including without limitation
+ *  the implied warranties of merchantability, fitness for a particular
+ *  purpose, or noninfringement.  The Regents of the University of
+ *  Michigan shall not be liable for any damages, including special,
+ *  indirect, incidental, or consequential damages, with respect to any
+ *  claim arising out of or in connection with the use of the software,
+ *  even if it has been or is hereafter advised of the possibility of
+ *  such damages.
+ */
+
+#ifndef FS_NFS_NFS4FILELAYOUT_H
+#define FS_NFS_NFS4FILELAYOUT_H
+
+#include "pnfs.h"
+
+/*
+ * Field testing shows we need to support upto 4096 stripe indices.
+ * We store each index as a u8 (u32 on the wire) to keep the memory footprint
+ * reasonable. This in turn means we support a maximum of 256
+ * RFC 5661 multipath_list4 structures.
+ */
+#define NFS4_PNFS_MAX_STRIPE_CNT 4096
+#define NFS4_PNFS_MAX_MULTI_CNT  256 /* 256 fit into a u8 stripe_index */
+
+enum stripetype4 {
+       STRIPE_SPARSE = 1,
+       STRIPE_DENSE = 2
+};
+
+/* Individual ip address */
+struct nfs4_pnfs_ds {
+       struct list_head        ds_node;  /* nfs4_pnfs_dev_hlist dev_dslist */
+       u32                     ds_ip_addr;
+       u32                     ds_port;
+       struct nfs_client       *ds_clp;
+       atomic_t                ds_count;
+};
+
+struct nfs4_file_layout_dsaddr {
+       struct pnfs_deviceid_node       deviceid;
+       u32                             stripe_count;
+       u8                              *stripe_indices;
+       u32                             ds_num;
+       struct nfs4_pnfs_ds             *ds_list[1];
+};
+
+struct nfs4_filelayout_segment {
+       struct pnfs_layout_segment generic_hdr;
+       u32 stripe_type;
+       u32 commit_through_mds;
+       u32 stripe_unit;
+       u32 first_stripe_index;
+       u64 pattern_offset;
+       struct nfs4_file_layout_dsaddr *dsaddr; /* Point to GETDEVINFO data */
+       unsigned int num_fh;
+       struct nfs_fh **fh_array;
+};
+
+static inline struct nfs4_filelayout_segment *
+FILELAYOUT_LSEG(struct pnfs_layout_segment *lseg)
+{
+       return container_of(lseg,
+                           struct nfs4_filelayout_segment,
+                           generic_hdr);
+}
+
+extern void nfs4_fl_free_deviceid_callback(struct pnfs_deviceid_node *);
+extern void print_ds(struct nfs4_pnfs_ds *ds);
+extern void print_deviceid(struct nfs4_deviceid *dev_id);
+extern struct nfs4_file_layout_dsaddr *
+nfs4_fl_find_get_deviceid(struct nfs_client *, struct nfs4_deviceid *dev_id);
+struct nfs4_file_layout_dsaddr *
+get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id);
+
+#endif /* FS_NFS_NFS4FILELAYOUT_H */
diff --git a/fs/nfs/nfs4filelayoutdev.c b/fs/nfs/nfs4filelayoutdev.c
new file mode 100644 (file)
index 0000000..51fe64a
--- /dev/null
@@ -0,0 +1,448 @@
+/*
+ *  Device operations for the pnfs nfs4 file layout driver.
+ *
+ *  Copyright (c) 2002
+ *  The Regents of the University of Michigan
+ *  All Rights Reserved
+ *
+ *  Dean Hildebrand <dhildebz@umich.edu>
+ *  Garth Goodson   <Garth.Goodson@netapp.com>
+ *
+ *  Permission is granted to use, copy, create derivative works, and
+ *  redistribute this software and such derivative works for any purpose,
+ *  so long as the name of the University of Michigan is not used in
+ *  any advertising or publicity pertaining to the use or distribution
+ *  of this software without specific, written prior authorization. If
+ *  the above copyright notice or any other identification of the
+ *  University of Michigan is included in any copy of any portion of
+ *  this software, then the disclaimer below must also be included.
+ *
+ *  This software is provided as is, without representation or warranty
+ *  of any kind either express or implied, including without limitation
+ *  the implied warranties of merchantability, fitness for a particular
+ *  purpose, or noninfringement.  The Regents of the University of
+ *  Michigan shall not be liable for any damages, including special,
+ *  indirect, incidental, or consequential damages, with respect to any
+ *  claim arising out of or in connection with the use of the software,
+ *  even if it has been or is hereafter advised of the possibility of
+ *  such damages.
+ */
+
+#include <linux/nfs_fs.h>
+#include <linux/vmalloc.h>
+
+#include "internal.h"
+#include "nfs4filelayout.h"
+
+#define NFSDBG_FACILITY                NFSDBG_PNFS_LD
+
+/*
+ * Data server cache
+ *
+ * Data servers can be mapped to different device ids.
+ * nfs4_pnfs_ds reference counting
+ *   - set to 1 on allocation
+ *   - incremented when a device id maps a data server already in the cache.
+ *   - decremented when deviceid is removed from the cache.
+ */
+DEFINE_SPINLOCK(nfs4_ds_cache_lock);
+static LIST_HEAD(nfs4_data_server_cache);
+
+/* Debug routines */
+void
+print_ds(struct nfs4_pnfs_ds *ds)
+{
+       if (ds == NULL) {
+               printk("%s NULL device\n", __func__);
+               return;
+       }
+       printk("        ip_addr %x port %hu\n"
+               "        ref count %d\n"
+               "        client %p\n"
+               "        cl_exchange_flags %x\n",
+               ntohl(ds->ds_ip_addr), ntohs(ds->ds_port),
+               atomic_read(&ds->ds_count), ds->ds_clp,
+               ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
+}
+
+void
+print_ds_list(struct nfs4_file_layout_dsaddr *dsaddr)
+{
+       int i;
+
+       ifdebug(FACILITY) {
+               printk("%s dsaddr->ds_num %d\n", __func__,
+                      dsaddr->ds_num);
+               for (i = 0; i < dsaddr->ds_num; i++)
+                       print_ds(dsaddr->ds_list[i]);
+       }
+}
+
+void print_deviceid(struct nfs4_deviceid *id)
+{
+       u32 *p = (u32 *)id;
+
+       dprintk("%s: device id= [%x%x%x%x]\n", __func__,
+               p[0], p[1], p[2], p[3]);
+}
+
+/* nfs4_ds_cache_lock is held */
+static struct nfs4_pnfs_ds *
+_data_server_lookup_locked(u32 ip_addr, u32 port)
+{
+       struct nfs4_pnfs_ds *ds;
+
+       dprintk("_data_server_lookup: ip_addr=%x port=%hu\n",
+                       ntohl(ip_addr), ntohs(port));
+
+       list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) {
+               if (ds->ds_ip_addr == ip_addr &&
+                   ds->ds_port == port) {
+                       return ds;
+               }
+       }
+       return NULL;
+}
+
+static void
+destroy_ds(struct nfs4_pnfs_ds *ds)
+{
+       dprintk("--> %s\n", __func__);
+       ifdebug(FACILITY)
+               print_ds(ds);
+
+       if (ds->ds_clp)
+               nfs_put_client(ds->ds_clp);
+       kfree(ds);
+}
+
+static void
+nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
+{
+       struct nfs4_pnfs_ds *ds;
+       int i;
+
+       print_deviceid(&dsaddr->deviceid.de_id);
+
+       for (i = 0; i < dsaddr->ds_num; i++) {
+               ds = dsaddr->ds_list[i];
+               if (ds != NULL) {
+                       if (atomic_dec_and_lock(&ds->ds_count,
+                                               &nfs4_ds_cache_lock)) {
+                               list_del_init(&ds->ds_node);
+                               spin_unlock(&nfs4_ds_cache_lock);
+                               destroy_ds(ds);
+                       }
+               }
+       }
+       kfree(dsaddr->stripe_indices);
+       kfree(dsaddr);
+}
+
+void
+nfs4_fl_free_deviceid_callback(struct pnfs_deviceid_node *device)
+{
+       struct nfs4_file_layout_dsaddr *dsaddr =
+               container_of(device, struct nfs4_file_layout_dsaddr, deviceid);
+
+       nfs4_fl_free_deviceid(dsaddr);
+}
+
+static struct nfs4_pnfs_ds *
+nfs4_pnfs_ds_add(struct inode *inode, u32 ip_addr, u32 port)
+{
+       struct nfs4_pnfs_ds *tmp_ds, *ds;
+
+       ds = kzalloc(sizeof(*tmp_ds), GFP_KERNEL);
+       if (!ds)
+               goto out;
+
+       spin_lock(&nfs4_ds_cache_lock);
+       tmp_ds = _data_server_lookup_locked(ip_addr, port);
+       if (tmp_ds == NULL) {
+               ds->ds_ip_addr = ip_addr;
+               ds->ds_port = port;
+               atomic_set(&ds->ds_count, 1);
+               INIT_LIST_HEAD(&ds->ds_node);
+               ds->ds_clp = NULL;
+               list_add(&ds->ds_node, &nfs4_data_server_cache);
+               dprintk("%s add new data server ip 0x%x\n", __func__,
+                       ds->ds_ip_addr);
+       } else {
+               kfree(ds);
+               atomic_inc(&tmp_ds->ds_count);
+               dprintk("%s data server found ip 0x%x, inc'ed ds_count to %d\n",
+                       __func__, tmp_ds->ds_ip_addr,
+                       atomic_read(&tmp_ds->ds_count));
+               ds = tmp_ds;
+       }
+       spin_unlock(&nfs4_ds_cache_lock);
+out:
+       return ds;
+}
+
+/*
+ * Currently only support ipv4, and one multi-path address.
+ */
+static struct nfs4_pnfs_ds *
+decode_and_add_ds(__be32 **pp, struct inode *inode)
+{
+       struct nfs4_pnfs_ds *ds = NULL;
+       char *buf;
+       const char *ipend, *pstr;
+       u32 ip_addr, port;
+       int nlen, rlen, i;
+       int tmp[2];
+       __be32 *r_netid, *r_addr, *p = *pp;
+
+       /* r_netid */
+       nlen = be32_to_cpup(p++);
+       r_netid = p;
+       p += XDR_QUADLEN(nlen);
+
+       /* r_addr */
+       rlen = be32_to_cpup(p++);
+       r_addr = p;
+       p += XDR_QUADLEN(rlen);
+       *pp = p;
+
+       /* Check that netid is "tcp" */
+       if (nlen != 3 ||  memcmp((char *)r_netid, "tcp", 3)) {
+               dprintk("%s: ERROR: non ipv4 TCP r_netid\n", __func__);
+               goto out_err;
+       }
+
+       /* ipv6 length plus port is legal */
+       if (rlen > INET6_ADDRSTRLEN + 8) {
+               dprintk("%s Invalid address, length %d\n", __func__,
+                       rlen);
+               goto out_err;
+       }
+       buf = kmalloc(rlen + 1, GFP_KERNEL);
+       buf[rlen] = '\0';
+       memcpy(buf, r_addr, rlen);
+
+       /* replace the port dots with dashes for the in4_pton() delimiter*/
+       for (i = 0; i < 2; i++) {
+               char *res = strrchr(buf, '.');
+               *res = '-';
+       }
+
+       /* Currently only support ipv4 address */
+       if (in4_pton(buf, rlen, (u8 *)&ip_addr, '-', &ipend) == 0) {
+               dprintk("%s: Only ipv4 addresses supported\n", __func__);
+               goto out_free;
+       }
+
+       /* port */
+       pstr = ipend;
+       sscanf(pstr, "-%d-%d", &tmp[0], &tmp[1]);
+       port = htons((tmp[0] << 8) | (tmp[1]));
+
+       ds = nfs4_pnfs_ds_add(inode, ip_addr, port);
+       dprintk("%s Decoded address and port %s\n", __func__, buf);
+out_free:
+       kfree(buf);
+out_err:
+       return ds;
+}
+
+/* Decode opaque device data and return the result */
+static struct nfs4_file_layout_dsaddr*
+decode_device(struct inode *ino, struct pnfs_device *pdev)
+{
+       int i, dummy;
+       u32 cnt, num;
+       u8 *indexp;
+       __be32 *p = (__be32 *)pdev->area, *indicesp;
+       struct nfs4_file_layout_dsaddr *dsaddr;
+
+       /* Get the stripe count (number of stripe index) */
+       cnt = be32_to_cpup(p++);
+       dprintk("%s stripe count  %d\n", __func__, cnt);
+       if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
+               printk(KERN_WARNING "%s: stripe count %d greater than "
+                      "supported maximum %d\n", __func__,
+                       cnt, NFS4_PNFS_MAX_STRIPE_CNT);
+               goto out_err;
+       }
+
+       /* Check the multipath list count */
+       indicesp = p;
+       p += XDR_QUADLEN(cnt << 2);
+       num = be32_to_cpup(p++);
+       dprintk("%s ds_num %u\n", __func__, num);
+       if (num > NFS4_PNFS_MAX_MULTI_CNT) {
+               printk(KERN_WARNING "%s: multipath count %d greater than "
+                       "supported maximum %d\n", __func__,
+                       num, NFS4_PNFS_MAX_MULTI_CNT);
+               goto out_err;
+       }
+       dsaddr = kzalloc(sizeof(*dsaddr) +
+                       (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
+                       GFP_KERNEL);
+       if (!dsaddr)
+               goto out_err;
+
+       dsaddr->stripe_indices = kzalloc(sizeof(u8) * cnt, GFP_KERNEL);
+       if (!dsaddr->stripe_indices)
+               goto out_err_free;
+
+       dsaddr->stripe_count = cnt;
+       dsaddr->ds_num = num;
+
+       memcpy(&dsaddr->deviceid.de_id, &pdev->dev_id, sizeof(pdev->dev_id));
+
+       /* Go back an read stripe indices */
+       p = indicesp;
+       indexp = &dsaddr->stripe_indices[0];
+       for (i = 0; i < dsaddr->stripe_count; i++) {
+               *indexp = be32_to_cpup(p++);
+               if (*indexp >= num)
+                       goto out_err_free;
+               indexp++;
+       }
+       /* Skip already read multipath list count */
+       p++;
+
+       for (i = 0; i < dsaddr->ds_num; i++) {
+               int j;
+
+               dummy = be32_to_cpup(p++); /* multipath count */
+               if (dummy > 1) {
+                       printk(KERN_WARNING
+                              "%s: Multipath count %d not supported, "
+                              "skipping all greater than 1\n", __func__,
+                               dummy);
+               }
+               for (j = 0; j < dummy; j++) {
+                       if (j == 0) {
+                               dsaddr->ds_list[i] = decode_and_add_ds(&p, ino);
+                               if (dsaddr->ds_list[i] == NULL)
+                                       goto out_err_free;
+                       } else {
+                               u32 len;
+                               /* skip extra multipath */
+                               len = be32_to_cpup(p++);
+                               p += XDR_QUADLEN(len);
+                               len = be32_to_cpup(p++);
+                               p += XDR_QUADLEN(len);
+                               continue;
+                       }
+               }
+       }
+       return dsaddr;
+
+out_err_free:
+       nfs4_fl_free_deviceid(dsaddr);
+out_err:
+       dprintk("%s ERROR: returning NULL\n", __func__);
+       return NULL;
+}
+
+/*
+ * Decode the opaque device specified in 'dev'
+ * and add it to the list of available devices.
+ * If the deviceid is already cached, nfs4_add_deviceid will return
+ * a pointer to the cached struct and throw away the new.
+ */
+static struct nfs4_file_layout_dsaddr*
+decode_and_add_device(struct inode *inode, struct pnfs_device *dev)
+{
+       struct nfs4_file_layout_dsaddr *dsaddr;
+       struct pnfs_deviceid_node *d;
+
+       dsaddr = decode_device(inode, dev);
+       if (!dsaddr) {
+               printk(KERN_WARNING "%s: Could not decode or add device\n",
+                       __func__);
+               return NULL;
+       }
+
+       d = pnfs_add_deviceid(NFS_SERVER(inode)->nfs_client->cl_devid_cache,
+                             &dsaddr->deviceid);
+
+       return container_of(d, struct nfs4_file_layout_dsaddr, deviceid);
+}
+
+/*
+ * Retrieve the information for dev_id, add it to the list
+ * of available devices, and return it.
+ */
+struct nfs4_file_layout_dsaddr *
+get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id)
+{
+       struct pnfs_device *pdev = NULL;
+       u32 max_resp_sz;
+       int max_pages;
+       struct page **pages = NULL;
+       struct nfs4_file_layout_dsaddr *dsaddr = NULL;
+       int rc, i;
+       struct nfs_server *server = NFS_SERVER(inode);
+
+       /*
+        * Use the session max response size as the basis for setting
+        * GETDEVICEINFO's maxcount
+        */
+       max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
+       max_pages = max_resp_sz >> PAGE_SHIFT;
+       dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
+               __func__, inode, max_resp_sz, max_pages);
+
+       pdev = kzalloc(sizeof(struct pnfs_device), GFP_KERNEL);
+       if (pdev == NULL)
+               return NULL;
+
+       pages = kzalloc(max_pages * sizeof(struct page *), GFP_KERNEL);
+       if (pages == NULL) {
+               kfree(pdev);
+               return NULL;
+       }
+       for (i = 0; i < max_pages; i++) {
+               pages[i] = alloc_page(GFP_KERNEL);
+               if (!pages[i])
+                       goto out_free;
+       }
+
+       /* set pdev->area */
+       pdev->area = vmap(pages, max_pages, VM_MAP, PAGE_KERNEL);
+       if (!pdev->area)
+               goto out_free;
+
+       memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
+       pdev->layout_type = LAYOUT_NFSV4_1_FILES;
+       pdev->pages = pages;
+       pdev->pgbase = 0;
+       pdev->pglen = PAGE_SIZE * max_pages;
+       pdev->mincount = 0;
+
+       rc = nfs4_proc_getdeviceinfo(server, pdev);
+       dprintk("%s getdevice info returns %d\n", __func__, rc);
+       if (rc)
+               goto out_free;
+
+       /*
+        * Found new device, need to decode it and then add it to the
+        * list of known devices for this mountpoint.
+        */
+       dsaddr = decode_and_add_device(inode, pdev);
+out_free:
+       if (pdev->area != NULL)
+               vunmap(pdev->area);
+       for (i = 0; i < max_pages; i++)
+               __free_page(pages[i]);
+       kfree(pages);
+       kfree(pdev);
+       dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
+       return dsaddr;
+}
+
+struct nfs4_file_layout_dsaddr *
+nfs4_fl_find_get_deviceid(struct nfs_client *clp, struct nfs4_deviceid *id)
+{
+       struct pnfs_deviceid_node *d;
+
+       d = pnfs_find_get_deviceid(clp->cl_devid_cache, id);
+       return (d == NULL) ? NULL :
+               container_of(d, struct nfs4_file_layout_dsaddr, deviceid);
+}