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
  34 #include "internal.h"
  35 #include "nfs4filelayout.h"
  36
  37 #define NFSDBG_FACILITY         NFSDBG_PNFS_LD
  38
  39 /*
  40  * Data server cache
  41  *
  42  * Data servers can be mapped to different device ids.
  43  * nfs4_pnfs_ds reference counting
  44  *   - set to 1 on allocation
  45  *   - incremented when a device id maps a data server already in the cache.
  46  *   - decremented when deviceid is removed from the cache.
  47  */
  48 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
  49 static LIST_HEAD(nfs4_data_server_cache);
  50
  51 /* Debug routines */
  52 void
  53 print_ds(struct nfs4_pnfs_ds *ds)
  54 {
  55         if (ds == NULL) {
  56                 printk("%s NULL device\n", __func__);
  57                 return;
  58         }
  59         printk("        ds %s\n"
  60                 "        ref count %d\n"
  61                 "        client %p\n"
  62                 "        cl_exchange_flags %x\n",
  63                 ds->ds_remotestr,
  64                 atomic_read(&ds->ds_count), ds->ds_clp,
  65                 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
  66 }
  67
  68 static bool
  69 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
  70 {
  71         struct sockaddr_in *a, *b;
  72         struct sockaddr_in6 *a6, *b6;
  73
  74         if (addr1->sa_family != addr2->sa_family)
  75                 return false;
  76
  77         switch (addr1->sa_family) {
  78         case AF_INET:
  79                 a = (struct sockaddr_in *)addr1;
  80                 b = (struct sockaddr_in *)addr2;
  81
  82                 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
  83                     a->sin_port == b->sin_port)
  84                         return true;
  85                 break;
  86
  87         case AF_INET6:
  88                 a6 = (struct sockaddr_in6 *)addr1;
  89                 b6 = (struct sockaddr_in6 *)addr2;
  90
  91                 /* LINKLOCAL addresses must have matching scope_id */
  92                 if (ipv6_addr_scope(&a6->sin6_addr) ==
  93                     IPV6_ADDR_SCOPE_LINKLOCAL &&
  94                     a6->sin6_scope_id != b6->sin6_scope_id)
  95                         return false;
  96
  97                 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
  98                     a6->sin6_port == b6->sin6_port)
  99                         return true;
 100                 break;
 101
 102         default:
 103                 dprintk("%s: unhandled address family: %u\n",
 104                         __func__, addr1->sa_family);
 105                 return false;
 106         }
 107
 108         return false;
 109 }
 110
 111 static bool
 112 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 113                                const struct list_head *dsaddrs2)
 114 {
 115         struct nfs4_pnfs_ds_addr *da1, *da2;
 116
 117         /* step through both lists, comparing as we go */
 118         for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
 119              da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
 120              da1 != NULL && da2 != NULL;
 121              da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
 122              da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
 123                 if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
 124                                    (struct sockaddr *)&da2->da_addr))
 125                         return false;
 126         }
 127         if (da1 == NULL && da2 == NULL)
 128                 return true;
 129
 130         return false;
 131 }
 132
 133 /*
 134  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 135  */
 136 static struct nfs4_pnfs_ds *
 137 _data_server_lookup_locked(const struct list_head *dsaddrs)
 138 {
 139         struct nfs4_pnfs_ds *ds;
 140
 141         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 142                 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 143                         return ds;
 144         return NULL;
 145 }
 146
 147 /*
 148  * Create an rpc connection to the nfs4_pnfs_ds data server
 149  * Currently only supports IPv4 and IPv6 addresses
 150  */
 151 static int
 152 nfs4_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds)
 153 {
 154         struct nfs_client *clp = ERR_PTR(-EIO);
 155         struct nfs4_pnfs_ds_addr *da;
 156         int status = 0;
 157
 158         dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
 159                 mds_srv->nfs_client->cl_rpcclient->cl_auth->au_flavor);
 160
 161         BUG_ON(list_empty(&ds->ds_addrs));
 162
 163         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 164                 dprintk("%s: DS %s: trying address %s\n",
 165                         __func__, ds->ds_remotestr, da->da_remotestr);
 166
 167                 clp = nfs4_set_ds_client(mds_srv->nfs_client,
 168                                  (struct sockaddr *)&da->da_addr,
 169                                  da->da_addrlen, IPPROTO_TCP);
 170                 if (!IS_ERR(clp))
 171                         break;
 172         }
 173
 174         if (IS_ERR(clp)) {
 175                 status = PTR_ERR(clp);
 176                 goto out;
 177         }
 178
 179         if ((clp->cl_exchange_flags & EXCHGID4_FLAG_MASK_PNFS) != 0) {
 180                 if (!is_ds_client(clp)) {
 181                         status = -ENODEV;
 182                         goto out_put;
 183                 }
 184                 ds->ds_clp = clp;
 185                 dprintk("%s [existing] server=%s\n", __func__,
 186                         ds->ds_remotestr);
 187                 goto out;
 188         }
 189
 190         /*
 191          * Do not set NFS_CS_CHECK_LEASE_TIME instead set the DS lease to
 192          * be equal to the MDS lease. Renewal is scheduled in create_session.
 193          */
 194         spin_lock(&mds_srv->nfs_client->cl_lock);
 195         clp->cl_lease_time = mds_srv->nfs_client->cl_lease_time;
 196         spin_unlock(&mds_srv->nfs_client->cl_lock);
 197         clp->cl_last_renewal = jiffies;
 198
 199         /* New nfs_client */
 200         status = nfs4_init_ds_session(clp);
 201         if (status)
 202                 goto out_put;
 203
 204         ds->ds_clp = clp;
 205         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 206 out:
 207         return status;
 208 out_put:
 209         nfs_put_client(clp);
 210         goto out;
 211 }
 212
 213 static void
 214 destroy_ds(struct nfs4_pnfs_ds *ds)
 215 {
 216         struct nfs4_pnfs_ds_addr *da;
 217
 218         dprintk("--> %s\n", __func__);
 219         ifdebug(FACILITY)
 220                 print_ds(ds);
 221
 222         if (ds->ds_clp)
 223                 nfs_put_client(ds->ds_clp);
 224
 225         while (!list_empty(&ds->ds_addrs)) {
 226                 da = list_first_entry(&ds->ds_addrs,
 227                                       struct nfs4_pnfs_ds_addr,
 228                                       da_node);
 229                 list_del_init(&da->da_node);
 230                 kfree(da->da_remotestr);
 231                 kfree(da);
 232         }
 233
 234         kfree(ds->ds_remotestr);
 235         kfree(ds);
 236 }
 237
 238 void
 239 nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
 240 {
 241         struct nfs4_pnfs_ds *ds;
 242         int i;
 243
 244         nfs4_print_deviceid(&dsaddr->id_node.deviceid);
 245
 246         for (i = 0; i < dsaddr->ds_num; i++) {
 247                 ds = dsaddr->ds_list[i];
 248                 if (ds != NULL) {
 249                         if (atomic_dec_and_lock(&ds->ds_count,
 250                                                 &nfs4_ds_cache_lock)) {
 251                                 list_del_init(&ds->ds_node);
 252                                 spin_unlock(&nfs4_ds_cache_lock);
 253                                 destroy_ds(ds);
 254                         }
 255                 }
 256         }
 257         kfree(dsaddr->stripe_indices);
 258         kfree(dsaddr);
 259 }
 260
 261 /*
 262  * Create a string with a human readable address and port to avoid
 263  * complicated setup around many dprinks.
 264  */
 265 static char *
 266 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 267 {
 268         struct nfs4_pnfs_ds_addr *da;
 269         char *remotestr;
 270         size_t len;
 271         char *p;
 272
 273         len = 3;        /* '{', '}' and eol */
 274         list_for_each_entry(da, dsaddrs, da_node) {
 275                 len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 276         }
 277
 278         remotestr = kzalloc(len, gfp_flags);
 279         if (!remotestr)
 280                 return NULL;
 281
 282         p = remotestr;
 283         *(p++) = '{';
 284         len--;
 285         list_for_each_entry(da, dsaddrs, da_node) {
 286                 size_t ll = strlen(da->da_remotestr);
 287
 288                 if (ll > len)
 289                         goto out_err;
 290
 291                 memcpy(p, da->da_remotestr, ll);
 292                 p += ll;
 293                 len -= ll;
 294
 295                 if (len < 1)
 296                         goto out_err;
 297                 (*p++) = ',';
 298                 len--;
 299         }
 300         if (len < 2)
 301                 goto out_err;
 302         *(p++) = '}';
 303         *p = '\0';
 304         return remotestr;
 305 out_err:
 306         kfree(remotestr);
 307         return NULL;
 308 }
 309
 310 static struct nfs4_pnfs_ds *
 311 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
 312 {
 313         struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 314         char *remotestr;
 315
 316         if (list_empty(dsaddrs)) {
 317                 dprintk("%s: no addresses defined\n", __func__);
 318                 goto out;
 319         }
 320
 321         ds = kzalloc(sizeof(*ds), gfp_flags);
 322         if (!ds)
 323                 goto out;
 324
 325         /* this is only used for debugging, so it's ok if its NULL */
 326         remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 327
 328         spin_lock(&nfs4_ds_cache_lock);
 329         tmp_ds = _data_server_lookup_locked(dsaddrs);
 330         if (tmp_ds == NULL) {
 331                 INIT_LIST_HEAD(&ds->ds_addrs);
 332                 list_splice_init(dsaddrs, &ds->ds_addrs);
 333                 ds->ds_remotestr = remotestr;
 334                 atomic_set(&ds->ds_count, 1);
 335                 INIT_LIST_HEAD(&ds->ds_node);
 336                 ds->ds_clp = NULL;
 337                 list_add(&ds->ds_node, &nfs4_data_server_cache);
 338                 dprintk("%s add new data server %s\n", __func__,
 339                         ds->ds_remotestr);
 340         } else {
 341                 kfree(remotestr);
 342                 kfree(ds);
 343                 atomic_inc(&tmp_ds->ds_count);
 344                 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 345                         __func__, tmp_ds->ds_remotestr,
 346                         atomic_read(&tmp_ds->ds_count));
 347                 ds = tmp_ds;
 348         }
 349         spin_unlock(&nfs4_ds_cache_lock);
 350 out:
 351         return ds;
 352 }
 353
 354 /*
 355  * Currently only supports ipv4, ipv6 and one multi-path address.
 356  */
 357 static struct nfs4_pnfs_ds_addr *
 358 decode_ds_addr(struct net *net, struct xdr_stream *streamp, gfp_t gfp_flags)
 359 {
 360         struct nfs4_pnfs_ds_addr *da = NULL;
 361         char *buf, *portstr;
 362         __be16 port;
 363         int nlen, rlen;
 364         int tmp[2];
 365         __be32 *p;
 366         char *netid, *match_netid;
 367         size_t len, match_netid_len;
 368         char *startsep = "";
 369         char *endsep = "";
 370
 371
 372         /* r_netid */
 373         p = xdr_inline_decode(streamp, 4);
 374         if (unlikely(!p))
 375                 goto out_err;
 376         nlen = be32_to_cpup(p++);
 377
 378         p = xdr_inline_decode(streamp, nlen);
 379         if (unlikely(!p))
 380                 goto out_err;
 381
 382         netid = kmalloc(nlen+1, gfp_flags);
 383         if (unlikely(!netid))
 384                 goto out_err;
 385
 386         netid[nlen] = '\0';
 387         memcpy(netid, p, nlen);
 388
 389         /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
 390         p = xdr_inline_decode(streamp, 4);
 391         if (unlikely(!p))
 392                 goto out_free_netid;
 393         rlen = be32_to_cpup(p);
 394
 395         p = xdr_inline_decode(streamp, rlen);
 396         if (unlikely(!p))
 397                 goto out_free_netid;
 398
 399         /* port is ".ABC.DEF", 8 chars max */
 400         if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
 401                 dprintk("%s: Invalid address, length %d\n", __func__,
 402                         rlen);
 403                 goto out_free_netid;
 404         }
 405         buf = kmalloc(rlen + 1, gfp_flags);
 406         if (!buf) {
 407                 dprintk("%s: Not enough memory\n", __func__);
 408                 goto out_free_netid;
 409         }
 410         buf[rlen] = '\0';
 411         memcpy(buf, p, rlen);
 412
 413         /* replace port '.' with '-' */
 414         portstr = strrchr(buf, '.');
 415         if (!portstr) {
 416                 dprintk("%s: Failed finding expected dot in port\n",
 417                         __func__);
 418                 goto out_free_buf;
 419         }
 420         *portstr = '-';
 421
 422         /* find '.' between address and port */
 423         portstr = strrchr(buf, '.');
 424         if (!portstr) {
 425                 dprintk("%s: Failed finding expected dot between address and "
 426                         "port\n", __func__);
 427                 goto out_free_buf;
 428         }
 429         *portstr = '\0';
 430
 431         da = kzalloc(sizeof(*da), gfp_flags);
 432         if (unlikely(!da))
 433                 goto out_free_buf;
 434
 435         INIT_LIST_HEAD(&da->da_node);
 436
 437         if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
 438                       sizeof(da->da_addr))) {
 439                 dprintk("%s: error parsing address %s\n", __func__, buf);
 440                 goto out_free_da;
 441         }
 442
 443         portstr++;
 444         sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
 445         port = htons((tmp[0] << 8) | (tmp[1]));
 446
 447         switch (da->da_addr.ss_family) {
 448         case AF_INET:
 449                 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
 450                 da->da_addrlen = sizeof(struct sockaddr_in);
 451                 match_netid = "tcp";
 452                 match_netid_len = 3;
 453                 break;
 454
 455         case AF_INET6:
 456                 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
 457                 da->da_addrlen = sizeof(struct sockaddr_in6);
 458                 match_netid = "tcp6";
 459                 match_netid_len = 4;
 460                 startsep = "[";
 461                 endsep = "]";
 462                 break;
 463
 464         default:
 465                 dprintk("%s: unsupported address family: %u\n",
 466                         __func__, da->da_addr.ss_family);
 467                 goto out_free_da;
 468         }
 469
 470         if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
 471                 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
 472                         __func__, netid, match_netid);
 473                 goto out_free_da;
 474         }
 475
 476         /* save human readable address */
 477         len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
 478         da->da_remotestr = kzalloc(len, gfp_flags);
 479
 480         /* NULL is ok, only used for dprintk */
 481         if (da->da_remotestr)
 482                 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
 483                          buf, endsep, ntohs(port));
 484
 485         dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
 486         kfree(buf);
 487         kfree(netid);
 488         return da;
 489
 490 out_free_da:
 491         kfree(da);
 492 out_free_buf:
 493         dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
 494         kfree(buf);
 495 out_free_netid:
 496         kfree(netid);
 497 out_err:
 498         return NULL;
 499 }
 500
 501 /* Decode opaque device data and return the result */
 502 static struct nfs4_file_layout_dsaddr*
 503 decode_device(struct inode *ino, struct pnfs_device *pdev, gfp_t gfp_flags)
 504 {
 505         int i;
 506         u32 cnt, num;
 507         u8 *indexp;
 508         __be32 *p;
 509         u8 *stripe_indices;
 510         u8 max_stripe_index;
 511         struct nfs4_file_layout_dsaddr *dsaddr = NULL;
 512         struct xdr_stream stream;
 513         struct xdr_buf buf;
 514         struct page *scratch;
 515         struct list_head dsaddrs;
 516         struct nfs4_pnfs_ds_addr *da;
 517
 518         /* set up xdr stream */
 519         scratch = alloc_page(gfp_flags);
 520         if (!scratch)
 521                 goto out_err;
 522
 523         xdr_init_decode_pages(&stream, &buf, pdev->pages, pdev->pglen);
 524         xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
 525
 526         /* Get the stripe count (number of stripe index) */
 527         p = xdr_inline_decode(&stream, 4);
 528         if (unlikely(!p))
 529                 goto out_err_free_scratch;
 530
 531         cnt = be32_to_cpup(p);
 532         dprintk("%s stripe count  %d\n", __func__, cnt);
 533         if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
 534                 printk(KERN_WARNING "NFS: %s: stripe count %d greater than "
 535                        "supported maximum %d\n", __func__,
 536                         cnt, NFS4_PNFS_MAX_STRIPE_CNT);
 537                 goto out_err_free_scratch;
 538         }
 539
 540         /* read stripe indices */
 541         stripe_indices = kcalloc(cnt, sizeof(u8), gfp_flags);
 542         if (!stripe_indices)
 543                 goto out_err_free_scratch;
 544
 545         p = xdr_inline_decode(&stream, cnt << 2);
 546         if (unlikely(!p))
 547                 goto out_err_free_stripe_indices;
 548
 549         indexp = &stripe_indices[0];
 550         max_stripe_index = 0;
 551         for (i = 0; i < cnt; i++) {
 552                 *indexp = be32_to_cpup(p++);
 553                 max_stripe_index = max(max_stripe_index, *indexp);
 554                 indexp++;
 555         }
 556
 557         /* Check the multipath list count */
 558         p = xdr_inline_decode(&stream, 4);
 559         if (unlikely(!p))
 560                 goto out_err_free_stripe_indices;
 561
 562         num = be32_to_cpup(p);
 563         dprintk("%s ds_num %u\n", __func__, num);
 564         if (num > NFS4_PNFS_MAX_MULTI_CNT) {
 565                 printk(KERN_WARNING "NFS: %s: multipath count %d greater than "
 566                         "supported maximum %d\n", __func__,
 567                         num, NFS4_PNFS_MAX_MULTI_CNT);
 568                 goto out_err_free_stripe_indices;
 569         }
 570
 571         /* validate stripe indices are all < num */
 572         if (max_stripe_index >= num) {
 573                 printk(KERN_WARNING "NFS: %s: stripe index %u >= num ds %u\n",
 574                         __func__, max_stripe_index, num);
 575                 goto out_err_free_stripe_indices;
 576         }
 577
 578         dsaddr = kzalloc(sizeof(*dsaddr) +
 579                         (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
 580                         gfp_flags);
 581         if (!dsaddr)
 582                 goto out_err_free_stripe_indices;
 583
 584         dsaddr->stripe_count = cnt;
 585         dsaddr->stripe_indices = stripe_indices;
 586         stripe_indices = NULL;
 587         dsaddr->ds_num = num;
 588         nfs4_init_deviceid_node(&dsaddr->id_node,
 589                                 NFS_SERVER(ino)->pnfs_curr_ld,
 590                                 NFS_SERVER(ino)->nfs_client,
 591                                 &pdev->dev_id);
 592
 593         INIT_LIST_HEAD(&dsaddrs);
 594
 595         for (i = 0; i < dsaddr->ds_num; i++) {
 596                 int j;
 597                 u32 mp_count;
 598
 599                 p = xdr_inline_decode(&stream, 4);
 600                 if (unlikely(!p))
 601                         goto out_err_free_deviceid;
 602
 603                 mp_count = be32_to_cpup(p); /* multipath count */
 604                 for (j = 0; j < mp_count; j++) {
 605                         da = decode_ds_addr(NFS_SERVER(ino)->nfs_client->net,
 606                                             &stream, gfp_flags);
 607                         if (da)
 608                                 list_add_tail(&da->da_node, &dsaddrs);
 609                 }
 610                 if (list_empty(&dsaddrs)) {
 611                         dprintk("%s: no suitable DS addresses found\n",
 612                                 __func__);
 613                         goto out_err_free_deviceid;
 614                 }
 615
 616                 dsaddr->ds_list[i] = nfs4_pnfs_ds_add(&dsaddrs, gfp_flags);
 617                 if (!dsaddr->ds_list[i])
 618                         goto out_err_drain_dsaddrs;
 619
 620                 /* If DS was already in cache, free ds addrs */
 621                 while (!list_empty(&dsaddrs)) {
 622                         da = list_first_entry(&dsaddrs,
 623                                               struct nfs4_pnfs_ds_addr,
 624                                               da_node);
 625                         list_del_init(&da->da_node);
 626                         kfree(da->da_remotestr);
 627                         kfree(da);
 628                 }
 629         }
 630
 631         __free_page(scratch);
 632         return dsaddr;
 633
 634 out_err_drain_dsaddrs:
 635         while (!list_empty(&dsaddrs)) {
 636                 da = list_first_entry(&dsaddrs, struct nfs4_pnfs_ds_addr,
 637                                       da_node);
 638                 list_del_init(&da->da_node);
 639                 kfree(da->da_remotestr);
 640                 kfree(da);
 641         }
 642 out_err_free_deviceid:
 643         nfs4_fl_free_deviceid(dsaddr);
 644         /* stripe_indicies was part of dsaddr */
 645         goto out_err_free_scratch;
 646 out_err_free_stripe_indices:
 647         kfree(stripe_indices);
 648 out_err_free_scratch:
 649         __free_page(scratch);
 650 out_err:
 651         dprintk("%s ERROR: returning NULL\n", __func__);
 652         return NULL;
 653 }
 654
 655 /*
 656  * Decode the opaque device specified in 'dev' and add it to the cache of
 657  * available devices.
 658  */
 659 static struct nfs4_file_layout_dsaddr *
 660 decode_and_add_device(struct inode *inode, struct pnfs_device *dev, gfp_t gfp_flags)
 661 {
 662         struct nfs4_deviceid_node *d;
 663         struct nfs4_file_layout_dsaddr *n, *new;
 664
 665         new = decode_device(inode, dev, gfp_flags);
 666         if (!new) {
 667                 printk(KERN_WARNING "NFS: %s: Could not decode or add device\n",
 668                         __func__);
 669                 return NULL;
 670         }
 671
 672         d = nfs4_insert_deviceid_node(&new->id_node);
 673         n = container_of(d, struct nfs4_file_layout_dsaddr, id_node);
 674         if (n != new) {
 675                 nfs4_fl_free_deviceid(new);
 676                 return n;
 677         }
 678
 679         return new;
 680 }
 681
 682 /*
 683  * Retrieve the information for dev_id, add it to the list
 684  * of available devices, and return it.
 685  */
 686 struct nfs4_file_layout_dsaddr *
 687 get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id, gfp_t gfp_flags)
 688 {
 689         struct pnfs_device *pdev = NULL;
 690         u32 max_resp_sz;
 691         int max_pages;
 692         struct page **pages = NULL;
 693         struct nfs4_file_layout_dsaddr *dsaddr = NULL;
 694         int rc, i;
 695         struct nfs_server *server = NFS_SERVER(inode);
 696
 697         /*
 698          * Use the session max response size as the basis for setting
 699          * GETDEVICEINFO's maxcount
 700          */
 701         max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
 702         max_pages = max_resp_sz >> PAGE_SHIFT;
 703         dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
 704                 __func__, inode, max_resp_sz, max_pages);
 705
 706         pdev = kzalloc(sizeof(struct pnfs_device), gfp_flags);
 707         if (pdev == NULL)
 708                 return NULL;
 709
 710         pages = kzalloc(max_pages * sizeof(struct page *), gfp_flags);
 711         if (pages == NULL) {
 712                 kfree(pdev);
 713                 return NULL;
 714         }
 715         for (i = 0; i < max_pages; i++) {
 716                 pages[i] = alloc_page(gfp_flags);
 717                 if (!pages[i])
 718                         goto out_free;
 719         }
 720
 721         memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
 722         pdev->layout_type = LAYOUT_NFSV4_1_FILES;
 723         pdev->pages = pages;
 724         pdev->pgbase = 0;
 725         pdev->pglen = PAGE_SIZE * max_pages;
 726         pdev->mincount = 0;
 727
 728         rc = nfs4_proc_getdeviceinfo(server, pdev);
 729         dprintk("%s getdevice info returns %d\n", __func__, rc);
 730         if (rc)
 731                 goto out_free;
 732
 733         /*
 734          * Found new device, need to decode it and then add it to the
 735          * list of known devices for this mountpoint.
 736          */
 737         dsaddr = decode_and_add_device(inode, pdev, gfp_flags);
 738 out_free:
 739         for (i = 0; i < max_pages; i++)
 740                 __free_page(pages[i]);
 741         kfree(pages);
 742         kfree(pdev);
 743         dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
 744         return dsaddr;
 745 }
 746
 747 void
 748 nfs4_fl_put_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
 749 {
 750         nfs4_put_deviceid_node(&dsaddr->id_node);
 751 }
 752
 753 /*
 754  * Want res = (offset - layout->pattern_offset)/ layout->stripe_unit
 755  * Then: ((res + fsi) % dsaddr->stripe_count)
 756  */
 757 u32
 758 nfs4_fl_calc_j_index(struct pnfs_layout_segment *lseg, loff_t offset)
 759 {
 760         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
 761         u64 tmp;
 762
 763         tmp = offset - flseg->pattern_offset;
 764         do_div(tmp, flseg->stripe_unit);
 765         tmp += flseg->first_stripe_index;
 766         return do_div(tmp, flseg->dsaddr->stripe_count);
 767 }
 768
 769 u32
 770 nfs4_fl_calc_ds_index(struct pnfs_layout_segment *lseg, u32 j)
 771 {
 772         return FILELAYOUT_LSEG(lseg)->dsaddr->stripe_indices[j];
 773 }
 774
 775 struct nfs_fh *
 776 nfs4_fl_select_ds_fh(struct pnfs_layout_segment *lseg, u32 j)
 777 {
 778         struct nfs4_filelayout_segment *flseg = FILELAYOUT_LSEG(lseg);
 779         u32 i;
 780
 781         if (flseg->stripe_type == STRIPE_SPARSE) {
 782                 if (flseg->num_fh == 1)
 783                         i = 0;
 784                 else if (flseg->num_fh == 0)
 785                         /* Use the MDS OPEN fh set in nfs_read_rpcsetup */
 786                         return NULL;
 787                 else
 788                         i = nfs4_fl_calc_ds_index(lseg, j);
 789         } else
 790                 i = j;
 791         return flseg->fh_array[i];
 792 }
 793
 794 static void
 795 filelayout_mark_devid_negative(struct nfs4_file_layout_dsaddr *dsaddr,
 796                                int err, const char *ds_remotestr)
 797 {
 798         u32 *p = (u32 *)&dsaddr->id_node.deviceid;
 799
 800         printk(KERN_ERR "NFS: data server %s connection error %d."
 801                 " Deviceid [%x%x%x%x] marked out of use.\n",
 802                 ds_remotestr, err, p[0], p[1], p[2], p[3]);
 803
 804         spin_lock(&nfs4_ds_cache_lock);
 805         dsaddr->flags |= NFS4_DEVICE_ID_NEG_ENTRY;
 806         spin_unlock(&nfs4_ds_cache_lock);
 807 }
 808
 809 struct nfs4_pnfs_ds *
 810 nfs4_fl_prepare_ds(struct pnfs_layout_segment *lseg, u32 ds_idx)
 811 {
 812         struct nfs4_file_layout_dsaddr *dsaddr = FILELAYOUT_LSEG(lseg)->dsaddr;
 813         struct nfs4_pnfs_ds *ds = dsaddr->ds_list[ds_idx];
 814
 815         if (ds == NULL) {
 816                 printk(KERN_ERR "NFS: %s: No data server for offset index %d\n",
 817                         __func__, ds_idx);
 818                 return NULL;
 819         }
 820
 821         if (!ds->ds_clp) {
 822                 struct nfs_server *s = NFS_SERVER(lseg->pls_layout->plh_inode);
 823                 int err;
 824
 825                 if (dsaddr->flags & NFS4_DEVICE_ID_NEG_ENTRY) {
 826                         /* Already tried to connect, don't try again */
 827                         dprintk("%s Deviceid marked out of use\n", __func__);
 828                         return NULL;
 829                 }
 830                 err = nfs4_ds_connect(s, ds);
 831                 if (err) {
 832                         filelayout_mark_devid_negative(dsaddr, err,
 833                                                        ds->ds_remotestr);
 834                         return NULL;
 835                 }
 836         }
 837         return ds;
 838 }