fs/nfs/pnfs_nfs.c

   1 /*
   2  * Common NFS I/O  operations for the pnfs file based
   3  * layout drivers.
   4  *
   5  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   6  *
   7  * Tom Haynes <loghyr@primarydata.com>
   8  */
   9
  10 #include <linux/nfs_fs.h>
  11 #include <linux/nfs_page.h>
  12 #include <linux/sunrpc/addr.h>
  13 #include <linux/module.h>
  14
  15 #include "nfs4session.h"
  16 #include "internal.h"
  17 #include "pnfs.h"
  18
  19 #define NFSDBG_FACILITY         NFSDBG_PNFS
  20
  21 static void pnfs_generic_fenceme(struct inode *inode,
  22                                  struct pnfs_layout_hdr *lo)
  23 {
  24         if (!test_and_clear_bit(NFS_LAYOUT_RETURN, &lo->plh_flags))
  25                 return;
  26         pnfs_return_layout(inode);
  27 }
  28
  29 void pnfs_generic_rw_release(void *data)
  30 {
  31         struct nfs_pgio_header *hdr = data;
  32         struct pnfs_layout_hdr *lo = hdr->lseg->pls_layout;
  33
  34         pnfs_generic_fenceme(lo->plh_inode, lo);
  35         nfs_put_client(hdr->ds_clp);
  36         hdr->mds_ops->rpc_release(data);
  37 }
  38 EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
  39
  40 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
  41 void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data)
  42 {
  43         struct nfs_page *first = nfs_list_entry(data->pages.next);
  44
  45         data->task.tk_status = 0;
  46         memcpy(&data->verf.verifier, &first->wb_verf,
  47                sizeof(data->verf.verifier));
  48         data->verf.verifier.data[0]++; /* ensure verifier mismatch */
  49 }
  50 EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes);
  51
  52 void pnfs_generic_write_commit_done(struct rpc_task *task, void *data)
  53 {
  54         struct nfs_commit_data *wdata = data;
  55
  56         /* Note this may cause RPC to be resent */
  57         wdata->mds_ops->rpc_call_done(task, data);
  58 }
  59 EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done);
  60
  61 void pnfs_generic_commit_release(void *calldata)
  62 {
  63         struct nfs_commit_data *data = calldata;
  64
  65         data->completion_ops->completion(data);
  66         pnfs_put_lseg(data->lseg);
  67         nfs_put_client(data->ds_clp);
  68         nfs_commitdata_release(data);
  69 }
  70 EXPORT_SYMBOL_GPL(pnfs_generic_commit_release);
  71
  72 /* The generic layer is about to remove the req from the commit list.
  73  * If this will make the bucket empty, it will need to put the lseg reference.
  74  * Note this must be called holding the inode (/cinfo) lock
  75  */
  76 void
  77 pnfs_generic_clear_request_commit(struct nfs_page *req,
  78                                   struct nfs_commit_info *cinfo)
  79 {
  80         struct pnfs_layout_segment *freeme = NULL;
  81
  82         if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
  83                 goto out;
  84         cinfo->ds->nwritten--;
  85         if (list_is_singular(&req->wb_list)) {
  86                 struct pnfs_commit_bucket *bucket;
  87
  88                 bucket = list_first_entry(&req->wb_list,
  89                                           struct pnfs_commit_bucket,
  90                                           written);
  91                 freeme = bucket->wlseg;
  92                 bucket->wlseg = NULL;
  93         }
  94 out:
  95         nfs_request_remove_commit_list(req, cinfo);
  96         pnfs_put_lseg_locked(freeme);
  97 }
  98 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
  99
 100 static int
 101 pnfs_generic_transfer_commit_list(struct list_head *src, struct list_head *dst,
 102                                   struct nfs_commit_info *cinfo, int max)
 103 {
 104         struct nfs_page *req, *tmp;
 105         int ret = 0;
 106
 107         list_for_each_entry_safe(req, tmp, src, wb_list) {
 108                 if (!nfs_lock_request(req))
 109                         continue;
 110                 kref_get(&req->wb_kref);
 111                 if (cond_resched_lock(cinfo->lock))
 112                         list_safe_reset_next(req, tmp, wb_list);
 113                 nfs_request_remove_commit_list(req, cinfo);
 114                 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
 115                 nfs_list_add_request(req, dst);
 116                 ret++;
 117                 if ((ret == max) && !cinfo->dreq)
 118                         break;
 119         }
 120         return ret;
 121 }
 122
 123 static int
 124 pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
 125                                  struct nfs_commit_info *cinfo,
 126                                  int max)
 127 {
 128         struct list_head *src = &bucket->written;
 129         struct list_head *dst = &bucket->committing;
 130         int ret;
 131
 132         lockdep_assert_held(cinfo->lock);
 133         ret = pnfs_generic_transfer_commit_list(src, dst, cinfo, max);
 134         if (ret) {
 135                 cinfo->ds->nwritten -= ret;
 136                 cinfo->ds->ncommitting += ret;
 137                 bucket->clseg = bucket->wlseg;
 138                 if (list_empty(src))
 139                         bucket->wlseg = NULL;
 140                 else
 141                         pnfs_get_lseg(bucket->clseg);
 142         }
 143         return ret;
 144 }
 145
 146 /* Move reqs from written to committing lists, returning count
 147  * of number moved.
 148  */
 149 int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo,
 150                                    int max)
 151 {
 152         int i, rv = 0, cnt;
 153
 154         lockdep_assert_held(cinfo->lock);
 155         for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
 156                 cnt = pnfs_generic_scan_ds_commit_list(&cinfo->ds->buckets[i],
 157                                                        cinfo, max);
 158                 max -= cnt;
 159                 rv += cnt;
 160         }
 161         return rv;
 162 }
 163 EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
 164
 165 /* Pull everything off the committing lists and dump into @dst.  */
 166 void pnfs_generic_recover_commit_reqs(struct list_head *dst,
 167                                       struct nfs_commit_info *cinfo)
 168 {
 169         struct pnfs_commit_bucket *b;
 170         struct pnfs_layout_segment *freeme;
 171         int i;
 172
 173         lockdep_assert_held(cinfo->lock);
 174 restart:
 175         for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
 176                 if (pnfs_generic_transfer_commit_list(&b->written, dst,
 177                                                       cinfo, 0)) {
 178                         freeme = b->wlseg;
 179                         b->wlseg = NULL;
 180                         spin_unlock(cinfo->lock);
 181                         pnfs_put_lseg(freeme);
 182                         spin_lock(cinfo->lock);
 183                         goto restart;
 184                 }
 185         }
 186         cinfo->ds->nwritten = 0;
 187 }
 188 EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
 189
 190 static void pnfs_generic_retry_commit(struct nfs_commit_info *cinfo, int idx)
 191 {
 192         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 193         struct pnfs_commit_bucket *bucket;
 194         struct pnfs_layout_segment *freeme;
 195         int i;
 196
 197         for (i = idx; i < fl_cinfo->nbuckets; i++) {
 198                 bucket = &fl_cinfo->buckets[i];
 199                 if (list_empty(&bucket->committing))
 200                         continue;
 201                 nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo);
 202                 spin_lock(cinfo->lock);
 203                 freeme = bucket->clseg;
 204                 bucket->clseg = NULL;
 205                 spin_unlock(cinfo->lock);
 206                 pnfs_put_lseg(freeme);
 207         }
 208 }
 209
 210 static unsigned int
 211 pnfs_generic_alloc_ds_commits(struct nfs_commit_info *cinfo,
 212                               struct list_head *list)
 213 {
 214         struct pnfs_ds_commit_info *fl_cinfo;
 215         struct pnfs_commit_bucket *bucket;
 216         struct nfs_commit_data *data;
 217         int i;
 218         unsigned int nreq = 0;
 219
 220         fl_cinfo = cinfo->ds;
 221         bucket = fl_cinfo->buckets;
 222         for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
 223                 if (list_empty(&bucket->committing))
 224                         continue;
 225                 data = nfs_commitdata_alloc();
 226                 if (!data)
 227                         break;
 228                 data->ds_commit_index = i;
 229                 spin_lock(cinfo->lock);
 230                 data->lseg = bucket->clseg;
 231                 bucket->clseg = NULL;
 232                 spin_unlock(cinfo->lock);
 233                 list_add(&data->pages, list);
 234                 nreq++;
 235         }
 236
 237         /* Clean up on error */
 238         pnfs_generic_retry_commit(cinfo, i);
 239         return nreq;
 240 }
 241
 242 /* This follows nfs_commit_list pretty closely */
 243 int
 244 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
 245                              int how, struct nfs_commit_info *cinfo,
 246                              int (*initiate_commit)(struct nfs_commit_data *data,
 247                                                     int how))
 248 {
 249         struct nfs_commit_data *data, *tmp;
 250         LIST_HEAD(list);
 251         unsigned int nreq = 0;
 252
 253         if (!list_empty(mds_pages)) {
 254                 data = nfs_commitdata_alloc();
 255                 if (data != NULL) {
 256                         data->lseg = NULL;
 257                         list_add(&data->pages, &list);
 258                         nreq++;
 259                 } else {
 260                         nfs_retry_commit(mds_pages, NULL, cinfo);
 261                         pnfs_generic_retry_commit(cinfo, 0);
 262                         cinfo->completion_ops->error_cleanup(NFS_I(inode));
 263                         return -ENOMEM;
 264                 }
 265         }
 266
 267         nreq += pnfs_generic_alloc_ds_commits(cinfo, &list);
 268
 269         if (nreq == 0) {
 270                 cinfo->completion_ops->error_cleanup(NFS_I(inode));
 271                 goto out;
 272         }
 273
 274         atomic_add(nreq, &cinfo->mds->rpcs_out);
 275
 276         list_for_each_entry_safe(data, tmp, &list, pages) {
 277                 list_del_init(&data->pages);
 278                 if (!data->lseg) {
 279                         nfs_init_commit(data, mds_pages, NULL, cinfo);
 280                         nfs_initiate_commit(NFS_CLIENT(inode), data,
 281                                             data->mds_ops, how, 0);
 282                 } else {
 283                         struct pnfs_commit_bucket *buckets;
 284
 285                         buckets = cinfo->ds->buckets;
 286                         nfs_init_commit(data,
 287                                         &buckets[data->ds_commit_index].committing,
 288                                         data->lseg,
 289                                         cinfo);
 290                         initiate_commit(data, how);
 291                 }
 292         }
 293 out:
 294         cinfo->ds->ncommitting = 0;
 295         return PNFS_ATTEMPTED;
 296 }
 297 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
 298
 299 /*
 300  * Data server cache
 301  *
 302  * Data servers can be mapped to different device ids.
 303  * nfs4_pnfs_ds reference counting
 304  *   - set to 1 on allocation
 305  *   - incremented when a device id maps a data server already in the cache.
 306  *   - decremented when deviceid is removed from the cache.
 307  */
 308 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
 309 static LIST_HEAD(nfs4_data_server_cache);
 310
 311 /* Debug routines */
 312 static void
 313 print_ds(struct nfs4_pnfs_ds *ds)
 314 {
 315         if (ds == NULL) {
 316                 printk(KERN_WARNING "%s NULL device\n", __func__);
 317                 return;
 318         }
 319         printk(KERN_WARNING "        ds %s\n"
 320                 "        ref count %d\n"
 321                 "        client %p\n"
 322                 "        cl_exchange_flags %x\n",
 323                 ds->ds_remotestr,
 324                 atomic_read(&ds->ds_count), ds->ds_clp,
 325                 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
 326 }
 327
 328 static bool
 329 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
 330 {
 331         struct sockaddr_in *a, *b;
 332         struct sockaddr_in6 *a6, *b6;
 333
 334         if (addr1->sa_family != addr2->sa_family)
 335                 return false;
 336
 337         switch (addr1->sa_family) {
 338         case AF_INET:
 339                 a = (struct sockaddr_in *)addr1;
 340                 b = (struct sockaddr_in *)addr2;
 341
 342                 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
 343                     a->sin_port == b->sin_port)
 344                         return true;
 345                 break;
 346
 347         case AF_INET6:
 348                 a6 = (struct sockaddr_in6 *)addr1;
 349                 b6 = (struct sockaddr_in6 *)addr2;
 350
 351                 /* LINKLOCAL addresses must have matching scope_id */
 352                 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
 353                     IPV6_ADDR_SCOPE_LINKLOCAL &&
 354                     a6->sin6_scope_id != b6->sin6_scope_id)
 355                         return false;
 356
 357                 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
 358                     a6->sin6_port == b6->sin6_port)
 359                         return true;
 360                 break;
 361
 362         default:
 363                 dprintk("%s: unhandled address family: %u\n",
 364                         __func__, addr1->sa_family);
 365                 return false;
 366         }
 367
 368         return false;
 369 }
 370
 371 static bool
 372 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 373                                const struct list_head *dsaddrs2)
 374 {
 375         struct nfs4_pnfs_ds_addr *da1, *da2;
 376
 377         /* step through both lists, comparing as we go */
 378         for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
 379              da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
 380              da1 != NULL && da2 != NULL;
 381              da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
 382              da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
 383                 if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
 384                                    (struct sockaddr *)&da2->da_addr))
 385                         return false;
 386         }
 387         if (da1 == NULL && da2 == NULL)
 388                 return true;
 389
 390         return false;
 391 }
 392
 393 /*
 394  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 395  */
 396 static struct nfs4_pnfs_ds *
 397 _data_server_lookup_locked(const struct list_head *dsaddrs)
 398 {
 399         struct nfs4_pnfs_ds *ds;
 400
 401         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 402                 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 403                         return ds;
 404         return NULL;
 405 }
 406
 407 static void destroy_ds(struct nfs4_pnfs_ds *ds)
 408 {
 409         struct nfs4_pnfs_ds_addr *da;
 410
 411         dprintk("--> %s\n", __func__);
 412         ifdebug(FACILITY)
 413                 print_ds(ds);
 414
 415         nfs_put_client(ds->ds_clp);
 416
 417         while (!list_empty(&ds->ds_addrs)) {
 418                 da = list_first_entry(&ds->ds_addrs,
 419                                       struct nfs4_pnfs_ds_addr,
 420                                       da_node);
 421                 list_del_init(&da->da_node);
 422                 kfree(da->da_remotestr);
 423                 kfree(da);
 424         }
 425
 426         kfree(ds->ds_remotestr);
 427         kfree(ds);
 428 }
 429
 430 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
 431 {
 432         if (atomic_dec_and_lock(&ds->ds_count,
 433                                 &nfs4_ds_cache_lock)) {
 434                 list_del_init(&ds->ds_node);
 435                 spin_unlock(&nfs4_ds_cache_lock);
 436                 destroy_ds(ds);
 437         }
 438 }
 439 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
 440
 441 /*
 442  * Create a string with a human readable address and port to avoid
 443  * complicated setup around many dprinks.
 444  */
 445 static char *
 446 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 447 {
 448         struct nfs4_pnfs_ds_addr *da;
 449         char *remotestr;
 450         size_t len;
 451         char *p;
 452
 453         len = 3;        /* '{', '}' and eol */
 454         list_for_each_entry(da, dsaddrs, da_node) {
 455                 len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 456         }
 457
 458         remotestr = kzalloc(len, gfp_flags);
 459         if (!remotestr)
 460                 return NULL;
 461
 462         p = remotestr;
 463         *(p++) = '{';
 464         len--;
 465         list_for_each_entry(da, dsaddrs, da_node) {
 466                 size_t ll = strlen(da->da_remotestr);
 467
 468                 if (ll > len)
 469                         goto out_err;
 470
 471                 memcpy(p, da->da_remotestr, ll);
 472                 p += ll;
 473                 len -= ll;
 474
 475                 if (len < 1)
 476                         goto out_err;
 477                 (*p++) = ',';
 478                 len--;
 479         }
 480         if (len < 2)
 481                 goto out_err;
 482         *(p++) = '}';
 483         *p = '\0';
 484         return remotestr;
 485 out_err:
 486         kfree(remotestr);
 487         return NULL;
 488 }
 489
 490 /*
 491  * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
 492  * uncached and return cached struct nfs4_pnfs_ds.
 493  */
 494 struct nfs4_pnfs_ds *
 495 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
 496 {
 497         struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 498         char *remotestr;
 499
 500         if (list_empty(dsaddrs)) {
 501                 dprintk("%s: no addresses defined\n", __func__);
 502                 goto out;
 503         }
 504
 505         ds = kzalloc(sizeof(*ds), gfp_flags);
 506         if (!ds)
 507                 goto out;
 508
 509         /* this is only used for debugging, so it's ok if its NULL */
 510         remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 511
 512         spin_lock(&nfs4_ds_cache_lock);
 513         tmp_ds = _data_server_lookup_locked(dsaddrs);
 514         if (tmp_ds == NULL) {
 515                 INIT_LIST_HEAD(&ds->ds_addrs);
 516                 list_splice_init(dsaddrs, &ds->ds_addrs);
 517                 ds->ds_remotestr = remotestr;
 518                 atomic_set(&ds->ds_count, 1);
 519                 INIT_LIST_HEAD(&ds->ds_node);
 520                 ds->ds_clp = NULL;
 521                 list_add(&ds->ds_node, &nfs4_data_server_cache);
 522                 dprintk("%s add new data server %s\n", __func__,
 523                         ds->ds_remotestr);
 524         } else {
 525                 kfree(remotestr);
 526                 kfree(ds);
 527                 atomic_inc(&tmp_ds->ds_count);
 528                 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 529                         __func__, tmp_ds->ds_remotestr,
 530                         atomic_read(&tmp_ds->ds_count));
 531                 ds = tmp_ds;
 532         }
 533         spin_unlock(&nfs4_ds_cache_lock);
 534 out:
 535         return ds;
 536 }
 537 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
 538
 539 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
 540 {
 541         might_sleep();
 542         wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
 543                         TASK_KILLABLE);
 544 }
 545
 546 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
 547 {
 548         smp_mb__before_atomic();
 549         clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
 550         smp_mb__after_atomic();
 551         wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
 552 }
 553
 554 static struct nfs_client *(*get_v3_ds_connect)(
 555                         struct nfs_client *mds_clp,
 556                         const struct sockaddr *ds_addr,
 557                         int ds_addrlen,
 558                         int ds_proto,
 559                         unsigned int ds_timeo,
 560                         unsigned int ds_retrans,
 561                         rpc_authflavor_t au_flavor);
 562
 563 static bool load_v3_ds_connect(void)
 564 {
 565         if (!get_v3_ds_connect) {
 566                 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
 567                 WARN_ON_ONCE(!get_v3_ds_connect);
 568         }
 569
 570         return(get_v3_ds_connect != NULL);
 571 }
 572
 573 void __exit nfs4_pnfs_v3_ds_connect_unload(void)
 574 {
 575         if (get_v3_ds_connect) {
 576                 symbol_put(nfs3_set_ds_client);
 577                 get_v3_ds_connect = NULL;
 578         }
 579 }
 580 EXPORT_SYMBOL_GPL(nfs4_pnfs_v3_ds_connect_unload);
 581
 582 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
 583                                  struct nfs4_pnfs_ds *ds,
 584                                  unsigned int timeo,
 585                                  unsigned int retrans,
 586                                  rpc_authflavor_t au_flavor)
 587 {
 588         struct nfs_client *clp = ERR_PTR(-EIO);
 589         struct nfs4_pnfs_ds_addr *da;
 590         int status = 0;
 591
 592         dprintk("--> %s DS %s au_flavor %d\n", __func__,
 593                 ds->ds_remotestr, au_flavor);
 594
 595         if (!load_v3_ds_connect())
 596                 goto out;
 597
 598         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 599                 dprintk("%s: DS %s: trying address %s\n",
 600                         __func__, ds->ds_remotestr, da->da_remotestr);
 601
 602                 clp = get_v3_ds_connect(mds_srv->nfs_client,
 603                                         (struct sockaddr *)&da->da_addr,
 604                                         da->da_addrlen, IPPROTO_TCP,
 605                                         timeo, retrans, au_flavor);
 606                 if (!IS_ERR(clp))
 607                         break;
 608         }
 609
 610         if (IS_ERR(clp)) {
 611                 status = PTR_ERR(clp);
 612                 goto out;
 613         }
 614
 615         smp_wmb();
 616         ds->ds_clp = clp;
 617         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 618 out:
 619         return status;
 620 }
 621
 622 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
 623                                  struct nfs4_pnfs_ds *ds,
 624                                  unsigned int timeo,
 625                                  unsigned int retrans,
 626                                  u32 minor_version,
 627                                  rpc_authflavor_t au_flavor)
 628 {
 629         struct nfs_client *clp = ERR_PTR(-EIO);
 630         struct nfs4_pnfs_ds_addr *da;
 631         int status = 0;
 632
 633         dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
 634                 au_flavor);
 635
 636         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 637                 dprintk("%s: DS %s: trying address %s\n",
 638                         __func__, ds->ds_remotestr, da->da_remotestr);
 639
 640                 clp = nfs4_set_ds_client(mds_srv->nfs_client,
 641                                         (struct sockaddr *)&da->da_addr,
 642                                         da->da_addrlen, IPPROTO_TCP,
 643                                         timeo, retrans, minor_version,
 644                                         au_flavor);
 645                 if (!IS_ERR(clp))
 646                         break;
 647         }
 648
 649         if (IS_ERR(clp)) {
 650                 status = PTR_ERR(clp);
 651                 goto out;
 652         }
 653
 654         status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time);
 655         if (status)
 656                 goto out_put;
 657
 658         smp_wmb();
 659         ds->ds_clp = clp;
 660         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 661 out:
 662         return status;
 663 out_put:
 664         nfs_put_client(clp);
 665         goto out;
 666 }
 667
 668 /*
 669  * Create an rpc connection to the nfs4_pnfs_ds data server.
 670  * Currently only supports IPv4 and IPv6 addresses.
 671  * If connection fails, make devid unavailable.
 672  */
 673 void nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
 674                           struct nfs4_deviceid_node *devid, unsigned int timeo,
 675                           unsigned int retrans, u32 version,
 676                           u32 minor_version, rpc_authflavor_t au_flavor)
 677 {
 678         if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
 679                 int err = 0;
 680
 681                 if (version == 3) {
 682                         err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
 683                                                        retrans, au_flavor);
 684                 } else if (version == 4) {
 685                         err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
 686                                                        retrans, minor_version,
 687                                                        au_flavor);
 688                 } else {
 689                         dprintk("%s: unsupported DS version %d\n", __func__,
 690                                 version);
 691                         err = -EPROTONOSUPPORT;
 692                 }
 693
 694                 if (err)
 695                         nfs4_mark_deviceid_unavailable(devid);
 696                 nfs4_clear_ds_conn_bit(ds);
 697         } else {
 698                 nfs4_wait_ds_connect(ds);
 699         }
 700 }
 701 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
 702
 703 /*
 704  * Currently only supports ipv4, ipv6 and one multi-path address.
 705  */
 706 struct nfs4_pnfs_ds_addr *
 707 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
 708 {
 709         struct nfs4_pnfs_ds_addr *da = NULL;
 710         char *buf, *portstr;
 711         __be16 port;
 712         int nlen, rlen;
 713         int tmp[2];
 714         __be32 *p;
 715         char *netid, *match_netid;
 716         size_t len, match_netid_len;
 717         char *startsep = "";
 718         char *endsep = "";
 719
 720
 721         /* r_netid */
 722         p = xdr_inline_decode(xdr, 4);
 723         if (unlikely(!p))
 724                 goto out_err;
 725         nlen = be32_to_cpup(p++);
 726
 727         p = xdr_inline_decode(xdr, nlen);
 728         if (unlikely(!p))
 729                 goto out_err;
 730
 731         netid = kmalloc(nlen+1, gfp_flags);
 732         if (unlikely(!netid))
 733                 goto out_err;
 734
 735         netid[nlen] = '\0';
 736         memcpy(netid, p, nlen);
 737
 738         /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
 739         p = xdr_inline_decode(xdr, 4);
 740         if (unlikely(!p))
 741                 goto out_free_netid;
 742         rlen = be32_to_cpup(p);
 743
 744         p = xdr_inline_decode(xdr, rlen);
 745         if (unlikely(!p))
 746                 goto out_free_netid;
 747
 748         /* port is ".ABC.DEF", 8 chars max */
 749         if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
 750                 dprintk("%s: Invalid address, length %d\n", __func__,
 751                         rlen);
 752                 goto out_free_netid;
 753         }
 754         buf = kmalloc(rlen + 1, gfp_flags);
 755         if (!buf) {
 756                 dprintk("%s: Not enough memory\n", __func__);
 757                 goto out_free_netid;
 758         }
 759         buf[rlen] = '\0';
 760         memcpy(buf, p, rlen);
 761
 762         /* replace port '.' with '-' */
 763         portstr = strrchr(buf, '.');
 764         if (!portstr) {
 765                 dprintk("%s: Failed finding expected dot in port\n",
 766                         __func__);
 767                 goto out_free_buf;
 768         }
 769         *portstr = '-';
 770
 771         /* find '.' between address and port */
 772         portstr = strrchr(buf, '.');
 773         if (!portstr) {
 774                 dprintk("%s: Failed finding expected dot between address and "
 775                         "port\n", __func__);
 776                 goto out_free_buf;
 777         }
 778         *portstr = '\0';
 779
 780         da = kzalloc(sizeof(*da), gfp_flags);
 781         if (unlikely(!da))
 782                 goto out_free_buf;
 783
 784         INIT_LIST_HEAD(&da->da_node);
 785
 786         if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
 787                       sizeof(da->da_addr))) {
 788                 dprintk("%s: error parsing address %s\n", __func__, buf);
 789                 goto out_free_da;
 790         }
 791
 792         portstr++;
 793         sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
 794         port = htons((tmp[0] << 8) | (tmp[1]));
 795
 796         switch (da->da_addr.ss_family) {
 797         case AF_INET:
 798                 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
 799                 da->da_addrlen = sizeof(struct sockaddr_in);
 800                 match_netid = "tcp";
 801                 match_netid_len = 3;
 802                 break;
 803
 804         case AF_INET6:
 805                 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
 806                 da->da_addrlen = sizeof(struct sockaddr_in6);
 807                 match_netid = "tcp6";
 808                 match_netid_len = 4;
 809                 startsep = "[";
 810                 endsep = "]";
 811                 break;
 812
 813         default:
 814                 dprintk("%s: unsupported address family: %u\n",
 815                         __func__, da->da_addr.ss_family);
 816                 goto out_free_da;
 817         }
 818
 819         if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
 820                 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
 821                         __func__, netid, match_netid);
 822                 goto out_free_da;
 823         }
 824
 825         /* save human readable address */
 826         len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
 827         da->da_remotestr = kzalloc(len, gfp_flags);
 828
 829         /* NULL is ok, only used for dprintk */
 830         if (da->da_remotestr)
 831                 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
 832                          buf, endsep, ntohs(port));
 833
 834         dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
 835         kfree(buf);
 836         kfree(netid);
 837         return da;
 838
 839 out_free_da:
 840         kfree(da);
 841 out_free_buf:
 842         dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
 843         kfree(buf);
 844 out_free_netid:
 845         kfree(netid);
 846 out_err:
 847         return NULL;
 848 }
 849 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);