fs/nfs/nfs4filelayoutdev.c

   1 /*
   2  *  Device operations for the pnfs nfs4 file layout driver.
   3  *
   4  *  Copyright (c) 2002
   5  *  The Regents of the University of Michigan
   6  *  All Rights Reserved
   7  *
   8  *  Dean Hildebrand <dhildebz@umich.edu>
   9  *  Garth Goodson   <Garth.Goodson@netapp.com>
  10  *
  11  *  Permission is granted to use, copy, create derivative works, and
  12  *  redistribute this software and such derivative works for any purpose,
  13  *  so long as the name of the University of Michigan is not used in
  14  *  any advertising or publicity pertaining to the use or distribution
  15  *  of this software without specific, written prior authorization. If
  16  *  the above copyright notice or any other identification of the
  17  *  University of Michigan is included in any copy of any portion of
  18  *  this software, then the disclaimer below must also be included.
  19  *
  20  *  This software is provided as is, without representation or warranty
  21  *  of any kind either express or implied, including without limitation
  22  *  the implied warranties of merchantability, fitness for a particular
  23  *  purpose, or noninfringement.  The Regents of the University of
  24  *  Michigan shall not be liable for any damages, including special,
  25  *  indirect, incidental, or consequential damages, with respect to any
  26  *  claim arising out of or in connection with the use of the software,
  27  *  even if it has been or is hereafter advised of the possibility of
  28  *  such damages.
  29  */
  30
  31 #include <linux/nfs_fs.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/module.h>
  34
  35 #include "internal.h"
  36 #include "nfs4filelayout.h"
  37
  38 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  39
  40 static unsigned int dataserver_timeo = NFS4_DEF_DS_TIMEO;
  41 static unsigned int dataserver_retrans = NFS4_DEF_DS_RETRANS;
  42
  43 /*
  44  * Data server cache
  45  *
  46  * Data servers can be mapped to different device ids.
  47  * nfs4_pnfs_ds reference counting
  48  *   - set to 1 on allocation
  49  *   - incremented when a device id maps a data server already in the cache.
  50  *   - decremented when deviceid is removed from the cache.
  51  */
  52 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
  53 static LIST_HEAD(nfs4_data_server_cache);
  54
  55 /* Debug routines */
  56 void
  57 print_ds(struct nfs4_pnfs_ds *ds)
  58 {
  59         if (ds == NULL) {
  60                 printk("%s NULL device\n", __func__);
  61                 return;
  62         }
  63         printk("        ds %s\n"
  64                 "        ref count %d\n"
  65                 "        client %p\n"
  66                 "        cl_exchange_flags %x\n",
  67                 ds->ds_remotestr,
  68                 atomic_read(&ds->ds_count), ds->ds_clp,
  69                 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
  70 }
  71
  72 static bool
  73 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
  74 {
  75         struct sockaddr_in *a, *b;
  76         struct sockaddr_in6 *a6, *b6;
  77
  78         if (addr1->sa_family != addr2->sa_family)
  79                 return false;
  80
  81         switch (addr1->sa_family) {
  82         case AF_INET:
  83                 a = (struct sockaddr_in *)addr1;
  84                 b = (struct sockaddr_in *)addr2;
  85
  86                 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
  87                     a->sin_port == b->sin_port)
  88                         return true;
  89                 break;
  90
  91         case AF_INET6:
  92                 a6 = (struct sockaddr_in6 *)addr1;
  93                 b6 = (struct sockaddr_in6 *)addr2;
  94
  95                 /* LINKLOCAL addresses must have matching scope_id */
  96                 if (ipv6_addr_scope(&a6->sin6_addr) ==
  97                     IPV6_ADDR_SCOPE_LINKLOCAL &&
  98                     a6->sin6_scope_id != b6->sin6_scope_id)
  99                         return false;
 100
 101                 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
 102                     a6->sin6_port == b6->sin6_port)
 103                         return true;
 104                 break;
 105
 106         default:
 107                 dprintk("%s: unhandled address family: %u\n",
 108                         __func__, addr1->sa_family);
 109                 return false;
 110         }
 111
 112         return false;
 113 }
 114
 115 static bool
 116 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 117                                const struct list_head *dsaddrs2)
 118 {
 119         struct nfs4_pnfs_ds_addr *da1, *da2;
 120
 121         /* step through both lists, comparing as we go */
 122         for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
 123              da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
 124              da1 != NULL && da2 != NULL;
 125              da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
 126              da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
 127                 if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
 128                                    (struct sockaddr *)&da2->da_addr))
 129                         return false;
 130         }
 131         if (da1 == NULL && da2 == NULL)
 132                 return true;
 133
 134         return false;
 135 }
 136
 137 /*
 138  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 139  */
 140 static struct nfs4_pnfs_ds *
 141 _data_server_lookup_locked(const struct list_head *dsaddrs)
 142 {
 143         struct nfs4_pnfs_ds *ds;
 144
 145         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 146                 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 147                         return ds;
 148         return NULL;
 149 }
 150
 151 /*
 152  * Lookup DS by nfs_client pointer. Zero data server client pointer
 153  */
 154 void nfs4_ds_disconnect(struct nfs_client *clp)
 155 {
 156         struct nfs4_pnfs_ds *ds;
 157         struct nfs_client *found = NULL;
 158
 159         dprintk("%s clp %p\n", __func__, clp);
 160         spin_lock(&nfs4_ds_cache_lock);
 161         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 162                 if (ds->ds_clp && ds->ds_clp == clp) {
 163                         found = ds->ds_clp;
 164                         ds->ds_clp = NULL;
 165                 }
 166         spin_unlock(&nfs4_ds_cache_lock);
 167         if (found) {
 168                 set_bit(NFS_CS_STOP_RENEW, &clp->cl_res_state);
 169                 nfs_put_client(clp);
 170         }
 171 }
 172
 173 /*
 174  * Create an rpc connection to the nfs4_pnfs_ds data server
 175  * Currently only supports IPv4 and IPv6 addresses
 176  */
 177 static int
 178 nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
 179 {
 180         struct nfs_client *clp = ERR_PTR(-EIO);
 181         struct nfs4_pnfs_ds_addr *da;
 182         int status = 0;
 183
 184         dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
 185                 mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor);
 186
 187         BUG_ON(list_empty(&ds->ds_addrs));
 188
 189         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 190                 dprintk("%s: DS %s: trying address %s\n",
 191                         __func__, ds->ds_remotestr, da->da_remotestr);
 192
 193                 clp = nfs4_set_ds_client(mds_srv->nfs_client,
 194                                         (struct sockaddr *)&da->da_addr,
 195                                         da->da_addrlen, IPPROTO_TCP,
 196                                         dataserver_timeo, dataserver_retrans);
 197                 if (!IS_ERR(clp))
 198                         break;
 199         }
 200
 201         if (IS_ERR(clp)) {
 202                 status = PTR_ERR(clp);
 203                 goto out;
 204         }
 205
 206         status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time);
 207         if (status)
 208                 goto out_put;
 209
 210         ds->ds_clp = clp;
 211         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 212 out:
 213         return status;
 214 out_put:
 215         nfs_put_client(clp);
 216         goto out;
 217 }
 218
 219 static void
 220 destroy_ds(struct nfs4_pnfs_ds *ds)
 221 {
 222         struct nfs4_pnfs_ds_addr *da;
 223
 224         dprintk("--> %s\n", __func__);
 225         ifdebug(FACILITY)
 226                 print_ds(ds);
 227
 228         if (ds->ds_clp)
 229                 nfs_put_client(ds->ds_clp);
 230
 231         while (!list_empty(&ds->ds_addrs)) {
 232                 da = list_first_entry(&ds->ds_addrs,
 233                                       struct nfs4_pnfs_ds_addr,
 234                                       da_node);
 235                 list_del_init(&da->da_node);
 236                 kfree(da->da_remotestr);
 237                 kfree(da);
 238         }
 239
 240         kfree(ds->ds_remotestr);
 241         kfree(ds);
 242 }
 243
 244 void
 245 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
 246 {
 247         struct nfs4_pnfs_ds *ds;
 248         int i;
 249
 250         nfs4_print_deviceid(&dsaddr->id_node.deviceid);
 251
 252         for (i = 0; i < dsaddr->ds_num; i++) {
 253                 ds = dsaddr->ds_list[i];
 254                 if (ds != NULL) {
 255                         if (atomic_dec_and_lock(&ds->ds_count,
 256                                                 &nfs4_ds_cache_lock)) {
 257                                 list_del_init(&ds->ds_node);
 258                                 spin_unlock(&nfs4_ds_cache_lock);
 259                                 destroy_ds(ds);
 260                         }
 261                 }
 262         }
 263         kfree(dsaddr->stripe_indices);
 264         kfree(dsaddr);
 265 }
 266
 267 /*
 268  * Create a string with a human readable address and port to avoid
 269  * complicated setup around many dprinks.
 270  */
 271 static char *
 272 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 273 {
 274         struct nfs4_pnfs_ds_addr *da;
 275         char *remotestr;
 276         size_t len;
 277         char *p;
 278
 279         len = 3;        /* '{', '}' and eol */
 280         list_for_each_entry(da, dsaddrs, da_node) {
 281                 len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 282         }
 283
 284         remotestr = kzalloc(len, gfp_flags);
 285         if (!remotestr)
 286                 return NULL;
 287
 288         p = remotestr;
 289         *(p++) = '{';
 290         len--;
 291         list_for_each_entry(da, dsaddrs, da_node) {
 292                 size_t ll = strlen(da->da_remotestr);
 293
 294                 if (ll > len)
 295                         goto out_err;
 296
 297                 memcpy(p, da->da_remotestr, ll);
 298                 p += ll;
 299                 len -= ll;
 300
 301                 if (len < 1)
 302                         goto out_err;
 303                 (*p++) = ',';
 304                 len--;
 305         }
 306         if (len < 2)
 307                 goto out_err;
 308         *(p++) = '}';
 309         *p = '\0';
 310         return remotestr;
 311 out_err:
 312         kfree(remotestr);
 313         return NULL;
 314 }
 315
 316 static struct nfs4_pnfs_ds *
 317 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
 318 {
 319         struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 320         char *remotestr;
 321
 322         if (list_empty(dsaddrs)) {
 323                 dprintk("%s: no addresses defined\n", __func__);
 324                 goto out;
 325         }
 326
 327         ds = kzalloc(sizeof(*ds), gfp_flags);
 328         if (!ds)
 329                 goto out;
 330
 331         /* this is only used for debugging, so it's ok if its NULL */
 332         remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 333
 334         spin_lock(&nfs4_ds_cache_lock);
 335         tmp_ds = _data_server_lookup_locked(dsaddrs);
 336         if (tmp_ds == NULL) {
 337                 INIT_LIST_HEAD(&ds->ds_addrs);
 338                 list_splice_init(dsaddrs, &ds->ds_addrs);
 339                 ds->ds_remotestr = remotestr;
 340                 atomic_set(&ds->ds_count, 1);
 341                 INIT_LIST_HEAD(&ds->ds_node);
 342                 ds->ds_clp = NULL;
 343                 list_add(&ds->ds_node, &nfs4_data_server_cache);
 344                 dprintk("%s add new data server %s\n", __func__,
 345                         ds->ds_remotestr);
 346         } else {
 347                 kfree(remotestr);
 348                 kfree(ds);
 349                 atomic_inc(&tmp_ds->ds_count);
 350                 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 351                         __func__, tmp_ds->ds_remotestr,
 352                         atomic_read(&tmp_ds->ds_count));
 353                 ds = tmp_ds;
 354         }
 355         spin_unlock(&nfs4_ds_cache_lock);
 356 out:
 357         return ds;
 358 }
 359
 360 /*
 361  * Currently only supports ipv4, ipv6 and one multi-path address.
 362  */
 363 static struct nfs4_pnfs_ds_addr *
 364 decode_ds_addr(struct net *net, struct xdr_stream *streamp, gfp_t gfp_flags)
 365 {
 366         struct nfs4_pnfs_ds_addr *da = NULL;
 367         char *buf, *portstr;
 368         __be16 port;
 369         int nlen, rlen;
 370         int tmp[2];
 371         __be32 *p;
 372         char *netid, *match_netid;
 373         size_t len, match_netid_len;
 374         char *startsep = "";
 375         char *endsep = "";
 376
 377
 378         /* r_netid */
 379         p = xdr_inline_decode(streamp, 4);
 380         if (unlikely(!p))
 381                 goto out_err;
 382         nlen = be32_to_cpup(p++);
 383
 384         p = xdr_inline_decode(streamp, nlen);
 385         if (unlikely(!p))
 386                 goto out_err;
 387
 388         netid = kmalloc(nlen+1, gfp_flags);
 389         if (unlikely(!netid))
 390                 goto out_err;
 391
 392         netid[nlen] = '\0';
 393         memcpy(netid, p, nlen);
 394
 395         /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
 396         p = xdr_inline_decode(streamp, 4);
 397         if (unlikely(!p))
 398                 goto out_free_netid;
 399         rlen = be32_to_cpup(p);
 400
 401         p = xdr_inline_decode(streamp, rlen);
 402         if (unlikely(!p))
 403                 goto out_free_netid;
 404
 405         /* port is ".ABC.DEF", 8 chars max */
 406         if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
 407                 dprintk("%s: Invalid address, length %d\n", __func__,
 408                         rlen);
 409                 goto out_free_netid;
 410         }
 411         buf = kmalloc(rlen + 1, gfp_flags);
 412         if (!buf) {
 413                 dprintk("%s: Not enough memory\n", __func__);
 414                 goto out_free_netid;
 415         }
 416         buf[rlen] = '\0';
 417         memcpy(buf, p, rlen);
 418
 419         /* replace port '.' with '-' */
 420         portstr = strrchr(buf, '.');
 421         if (!portstr) {
 422                 dprintk("%s: Failed finding expected dot in port\n",
 423                         __func__);
 424                 goto out_free_buf;
 425         }
 426         *portstr = '-';
 427
 428         /* find '.' between address and port */
 429         portstr = strrchr(buf, '.');
 430         if (!portstr) {
 431                 dprintk("%s: Failed finding expected dot between address and "
 432                         "port\n", __func__);
 433                 goto out_free_buf;
 434         }
 435         *portstr = '\0';
 436
 437         da = kzalloc(sizeof(*da), gfp_flags);
 438         if (unlikely(!da))
 439                 goto out_free_buf;
 440
 441         INIT_LIST_HEAD(&da->da_node);
 442
 443         if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
 444                       sizeof(da->da_addr))) {
 445                 dprintk("%s: error parsing address %s\n", __func__, buf);
 446                 goto out_free_da;
 447         }
 448
 449         portstr++;
 450         sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
 451         port = htons((tmp[0] << 8) | (tmp[1]));
 452
 453         switch (da->da_addr.ss_family) {
 454         case AF_INET:
 455                 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
 456                 da->da_addrlen = sizeof(struct sockaddr_in);
 457                 match_netid = "tcp";
 458                 match_netid_len = 3;
 459                 break;
 460
 461         case AF_INET6:
 462                 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
 463                 da->da_addrlen = sizeof(struct sockaddr_in6);
 464                 match_netid = "tcp6";
 465                 match_netid_len = 4;
 466                 startsep = "[";
 467                 endsep = "]";
 468                 break;
 469
 470         default:
 471                 dprintk("%s: unsupported address family: %u\n",
 472                         __func__, da->da_addr.ss_family);
 473                 goto out_free_da;
 474         }
 475
 476         if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
 477                 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
 478                         __func__, netid, match_netid);
 479                 goto out_free_da;
 480         }
 481
 482         /* save human readable address */
 483         len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
 484         da->da_remotestr = kzalloc(len, gfp_flags);
 485
 486         /* NULL is ok, only used for dprintk */
 487         if (da->da_remotestr)
 488                 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
 489                          buf, endsep, ntohs(port));
 490
 491         dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
 492         kfree(buf);
 493         kfree(netid);
 494         return da;
 495
 496 out_free_da:
 497         kfree(da);
 498 out_free_buf:
 499         dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
 500         kfree(buf);
 501 out_free_netid:
 502         kfree(netid);
 503 out_err:
 504         return NULL;
 505 }
 506
 507 /* Decode opaque device data and return the result */
 508 static struct nfs4_file_layout_dsaddr*
 509 decode_device(struct inode *ino, struct pnfs_device *pdev, gfp_t gfp_flags)
 510 {
 511         int i;
 512         u32 cnt, num;
 513         u8 *indexp;
 514         __be32 *p;
 515         u8 *stripe_indices;
 516         u8 max_stripe_index;
 517         struct nfs4_file_layout_dsaddr *dsaddr = NULL;
 518         struct xdr_stream stream;
 519         struct xdr_buf buf;
 520         struct page *scratch;
 521         struct list_head dsaddrs;
 522         struct nfs4_pnfs_ds_addr *da;
 523
 524         /* set up xdr stream */
 525         scratch = alloc_page(gfp_flags);
 526         if (!scratch)
 527                 goto out_err;
 528
 529         xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
 530         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 531
 532         /* Get the stripe count (number of stripe index) */
 533         p = xdr_inline_decode(&stream, 4);
 534         if (unlikely(!p))
 535                 goto out_err_free_scratch;
 536
 537         cnt = be32_to_cpup(p);
 538         dprintk("%s stripe count  %d\n", __func__, cnt);
 539         if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
 540                 printk(KERN_WARNING "NFS: %s: stripe count %d greater than "
 541                        "supported maximum %d\n", __func__,
 542                         cnt, NFS4_PNFS_MAX_STRIPE_CNT);
 543                 goto out_err_free_scratch;
 544         }
 545
 546         /* read stripe indices */
 547         stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
 548         if (!stripe_indices)
 549                 goto out_err_free_scratch;
 550
 551         p = xdr_inline_decode(&stream, cnt << 2);
 552         if (unlikely(!p))
 553                 goto out_err_free_stripe_indices;
 554
 555         indexp = &stripe_indices[0];
 556         max_stripe_index = 0;
 557         for (i = 0; i < cnt; i++) {
 558                 *indexp = be32_to_cpup(p++);
 559                 max_stripe_index = max(max_stripe_index, *indexp);
 560                 indexp++;
 561         }
 562
 563         /* Check the multipath list count */
 564         p = xdr_inline_decode(&stream, 4);
 565         if (unlikely(!p))
 566                 goto out_err_free_stripe_indices;
 567
 568         num = be32_to_cpup(p);
 569         dprintk("%s ds_num %u\n", __func__, num);
 570         if (num > NFS4_PNFS_MAX_MULTI_CNT) {
 571                 printk(KERN_WARNING "NFS: %s: multipath count %d greater than "
 572                         "supported maximum %d\n", __func__,
 573                         num, NFS4_PNFS_MAX_MULTI_CNT);
 574                 goto out_err_free_stripe_indices;
 575         }
 576
 577         /* validate stripe indices are all < num */
 578         if (max_stripe_index >= num) {
 579                 printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n",
 580                         __func__, max_stripe_index, num);
 581                 goto out_err_free_stripe_indices;
 582         }
 583
 584         dsaddr = kzalloc(sizeof(*dsaddr) +
 585                         (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
 586                         gfp_flags);
 587         if (!dsaddr)
 588                 goto out_err_free_stripe_indices;
 589
 590         dsaddr->stripe_count = cnt;
 591         dsaddr->stripe_indices = stripe_indices;
 592         stripe_indices = NULL;
 593         dsaddr->ds_num = num;
 594         nfs4_init_deviceid_node(&dsaddr->id_node,
 595                                 NFS_SERVER(ino)->pnfs_curr_ld,
 596                                 NFS_SERVER(ino)->nfs_client,
 597                                 &pdev->dev_id);
 598
 599         INIT_LIST_HEAD(&dsaddrs);
 600
 601         for (i = 0; i < dsaddr->ds_num; i++) {
 602                 int j;
 603                 u32 mp_count;
 604
 605                 p = xdr_inline_decode(&stream, 4);
 606                 if (unlikely(!p))
 607                         goto out_err_free_deviceid;
 608
 609                 mp_count = be32_to_cpup(p); /* multipath count */
 610                 for (j = 0; j < mp_count; j++) {
 611                         da = decode_ds_addr(NFS_SERVER(ino)->nfs_client->cl_net,
 612                                             &stream, gfp_flags);
 613                         if (da)
 614                                 list_add_tail(&da->da_node, &dsaddrs);
 615                 }
 616                 if (list_empty(&dsaddrs)) {
 617                         dprintk("%s: no suitable DS addresses found\n",
 618                                 __func__);
 619                         goto out_err_free_deviceid;
 620                 }
 621
 622                 dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
 623                 if (!dsaddr->ds_list[i])
 624                         goto out_err_drain_dsaddrs;
 625
 626                 /* If DS was already in cache, free ds addrs */
 627                 while (!list_empty(&dsaddrs)) {
 628                         da = list_first_entry(&dsaddrs,
 629                                               struct nfs4_pnfs_ds_addr,
 630                                               da_node);
 631                         list_del_init(&da->da_node);
 632                         kfree(da->da_remotestr);
 633                         kfree(da);
 634                 }
 635         }
 636
 637         __free_page(scratch);
 638         return dsaddr;
 639
 640 out_err_drain_dsaddrs:
 641         while (!list_empty(&dsaddrs)) {
 642                 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
 643                                       da_node);
 644                 list_del_init(&da->da_node);
 645                 kfree(da->da_remotestr);
 646                 kfree(da);
 647         }
 648 out_err_free_deviceid:
 649         nfs4_fl_free_deviceid(dsaddr);
 650         /* stripe_indicies was part of dsaddr */
 651         goto out_err_free_scratch;
 652 out_err_free_stripe_indices:
 653         kfree(stripe_indices);
 654 out_err_free_scratch:
 655         __free_page(scratch);
 656 out_err:
 657         dprintk("%s ERROR: returning NULL\n", __func__);
 658         return NULL;
 659 }
 660
 661 /*
 662  * Decode the opaque device specified in 'dev' and add it to the cache of
 663  * available devices.
 664  */
 665 static struct nfs4_file_layout_dsaddr *
 666 decode_and_add_device(struct inode *inode, struct pnfs_device *dev, gfp_t gfp_flags)
 667 {
 668         struct nfs4_deviceid_node *d;
 669         struct nfs4_file_layout_dsaddr *n, *new;
 670
 671         new = decode_device(inode, dev, gfp_flags);
 672         if (!new) {
 673                 printk(KERN_WARNING "NFS: %s: Could not decode or add device\n",
 674                         __func__);
 675                 return NULL;
 676         }
 677
 678         d = nfs4_insert_deviceid_node(&new->id_node);
 679         n = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
 680         if (n != new) {
 681                 nfs4_fl_free_deviceid(new);
 682                 return n;
 683         }
 684
 685         return new;
 686 }
 687
 688 /*
 689  * Retrieve the information for dev_id, add it to the list
 690  * of available devices, and return it.
 691  */
 692 struct nfs4_file_layout_dsaddr *
 693 get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id, gfp_t gfp_flags)
 694 {
 695         struct pnfs_device *pdev = NULL;
 696         u32 max_resp_sz;
 697         int max_pages;
 698         struct page **pages = NULL;
 699         struct nfs4_file_layout_dsaddr *dsaddr = NULL;
 700         int rc, i;
 701         struct nfs_server *server = NFS_SERVER(inode);
 702
 703         /*
 704          * Use the session max response size as the basis for setting
 705          * GETDEVICEINFO's maxcount
 706          */
 707         max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
 708         max_pages = nfs_page_array_len(0, max_resp_sz);
 709         dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
 710                 __func__, inode, max_resp_sz, max_pages);
 711
 712         pdev = kzalloc(sizeof(struct pnfs_device), gfp_flags);
 713         if (pdev == NULL)
 714                 return NULL;
 715
 716         pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
 717         if (pages == NULL) {
 718                 kfree(pdev);
 719                 return NULL;
 720         }
 721         for (i = 0; i < max_pages; i++) {
 722                 pages[i] = alloc_page(gfp_flags);
 723                 if (!pages[i])
 724                         goto out_free;
 725         }
 726
 727         memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
 728         pdev->layout_type = LAYOUT_NFSV4_1_FILES;
 729         pdev->pages = pages;
 730         pdev->pgbase = 0;
 731         pdev->pglen = PAGE_SIZE * max_pages;
 732         pdev->mincount = 0;
 733
 734         rc = nfs4_proc_getdeviceinfo(server, pdev);
 735         dprintk("%s getdevice info returns %d\n", __func__, rc);
 736         if (rc)
 737                 goto out_free;
 738
 739         /*
 740          * Found new device, need to decode it and then add it to the
 741          * list of known devices for this mountpoint.
 742          */
 743         dsaddr = decode_and_add_device(inode, pdev, gfp_flags);
 744 out_free:
 745         for (i = 0; i < max_pages; i++)
 746                 __free_page(pages[i]);
 747         kfree(pages);
 748         kfree(pdev);
 749         dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
 750         return dsaddr;
 751 }
 752
 753 void
 754 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
 755 {
 756         nfs4_put_deviceid_node(&dsaddr->id_node);
 757 }
 758
 759 /*
 760  * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
 761  * Then: ((res + fsi) % dsaddr->stripe_count)
 762  */
 763 u32
 764 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
 765 {
 766         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
 767         u64 tmp;
 768
 769         tmp = offset - flseg->pattern_offset;
 770         do_div(tmp, flseg->stripe_unit);
 771         tmp += flseg->first_stripe_index;
 772         return do_div(tmp, flseg->dsaddr->stripe_count);
 773 }
 774
 775 u32
 776 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
 777 {
 778         return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
 779 }
 780
 781 struct nfs_fh *
 782 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
 783 {
 784         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
 785         u32 i;
 786
 787         if (flseg->stripe_type == STRIPE_SPARSE) {
 788                 if (flseg->num_fh == 1)
 789                         i = 0;
 790                 else if (flseg->num_fh == 0)
 791                         /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
 792                         return NULL;
 793                 else
 794                         i = nfs4_fl_calc_ds_index(lseg, j);
 795         } else
 796                 i = j;
 797         return flseg->fh_array[i];
 798 }
 799
 800 struct nfs4_pnfs_ds *
 801 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
 802 {
 803         struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
 804         struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
 805         struct nfs4_deviceid_node *devid = FILELAYOUT_DEVID_NODE(lseg);
 806
 807         if (filelayout_test_devid_invalid(devid))
 808                 return NULL;
 809
 810         if (ds == NULL) {
 811                 printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
 812                         __func__, ds_idx);
 813                 goto mark_dev_invalid;
 814         }
 815
 816         if (!ds->ds_clp) {
 817                 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
 818                 int err;
 819
 820                 err = nfs4_ds_connect(s, ds);
 821                 if (err)
 822                         goto mark_dev_invalid;
 823         }
 824         return ds;
 825
 826 mark_dev_invalid:
 827         filelayout_mark_devid_invalid(devid);
 828         return NULL;
 829 }
 830
 831 module_param(dataserver_retrans, uint, 0644);
 832 MODULE_PARM_DESC(dataserver_retrans, "The  number of times the NFSv4.1 client "
 833                         "retries a request before it attempts further "
 834                         " recovery  action.");
 835 module_param(dataserver_timeo, uint, 0644);
 836 MODULE_PARM_DESC(dataserver_timeo, "The time (in tenths of a second) the "
 837                         "NFSv4.1  client  waits for a response from a "
 838                         " data server before it retries an NFS request.");