kernel/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 void pnfs_generic_rw_release(void *data)
  22 {
  23         struct nfs_pgio_header *hdr = data;
  24
  25         nfs_put_client(hdr->ds_clp);
  26         hdr->mds_ops->rpc_release(data);
  27 }
  28 EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
  29
  30 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
  31 void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data)
  32 {
  33         struct nfs_page *first = nfs_list_entry(data->pages.next);
  34
  35         data->task.tk_status = 0;
  36         memcpy(&data->verf.verifier, &first->wb_verf,
  37                sizeof(data->verf.verifier));
  38         data->verf.verifier.data[0]++; /* ensure verifier mismatch */
  39 }
  40 EXPORT_SYMBOL_GPL(pnfs_generic_prepare_to_resend_writes);
  41
  42 void pnfs_generic_write_commit_done(struct rpc_task *task, void *data)
  43 {
  44         struct nfs_commit_data *wdata = data;
  45
  46         /* Note this may cause RPC to be resent */
  47         wdata->mds_ops->rpc_call_done(task, data);
  48 }
  49 EXPORT_SYMBOL_GPL(pnfs_generic_write_commit_done);
  50
  51 void pnfs_generic_commit_release(void *calldata)
  52 {
  53         struct nfs_commit_data *data = calldata;
  54
  55         data->completion_ops->completion(data);
  56         pnfs_put_lseg(data->lseg);
  57         nfs_put_client(data->ds_clp);
  58         nfs_commitdata_release(data);
  59 }
  60 EXPORT_SYMBOL_GPL(pnfs_generic_commit_release);
  61
  62 /* The generic layer is about to remove the req from the commit list.
  63  * If this will make the bucket empty, it will need to put the lseg reference.
  64  * Note this must be called holding the inode (/cinfo) lock
  65  */
  66 void
  67 pnfs_generic_clear_request_commit(struct nfs_page *req,
  68                                   struct nfs_commit_info *cinfo)
  69 {
  70         struct pnfs_layout_segment *freeme = NULL;
  71
  72         if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
  73                 goto out;
  74         cinfo->ds->nwritten--;
  75         if (list_is_singular(&req->wb_list)) {
  76                 struct pnfs_commit_bucket *bucket;
  77
  78                 bucket = list_first_entry(&req->wb_list,
  79                                           struct pnfs_commit_bucket,
  80                                           written);
  81                 freeme = bucket->wlseg;
  82                 bucket->wlseg = NULL;
  83         }
  84 out:
  85         nfs_request_remove_commit_list(req, cinfo);
  86         pnfs_put_lseg_locked(freeme);
  87 }
  88 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
  89
  90 static int
  91 pnfs_generic_transfer_commit_list(struct list_head *src, struct list_head *dst,
  92                                   struct nfs_commit_info *cinfo, int max)
  93 {
  94         struct nfs_page *req, *tmp;
  95         int ret = 0;
  96
  97         list_for_each_entry_safe(req, tmp, src, wb_list) {
  98                 if (!nfs_lock_request(req))
  99                         continue;
 100                 kref_get(&req->wb_kref);
 101                 if (cond_resched_lock(cinfo->lock))
 102                         list_safe_reset_next(req, tmp, wb_list);
 103                 nfs_request_remove_commit_list(req, cinfo);
 104                 clear_bit(PG_COMMIT_TO_DS, &req->wb_flags);
 105                 nfs_list_add_request(req, dst);
 106                 ret++;
 107                 if ((ret == max) && !cinfo->dreq)
 108                         break;
 109         }
 110         return ret;
 111 }
 112
 113 static int
 114 pnfs_generic_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
 115                                  struct nfs_commit_info *cinfo,
 116                                  int max)
 117 {
 118         struct list_head *src = &bucket->written;
 119         struct list_head *dst = &bucket->committing;
 120         int ret;
 121
 122         lockdep_assert_held(cinfo->lock);
 123         ret = pnfs_generic_transfer_commit_list(src, dst, cinfo, max);
 124         if (ret) {
 125                 cinfo->ds->nwritten -= ret;
 126                 cinfo->ds->ncommitting += ret;
 127                 bucket->clseg = bucket->wlseg;
 128                 if (list_empty(src))
 129                         bucket->wlseg = NULL;
 130                 else
 131                         pnfs_get_lseg(bucket->clseg);
 132         }
 133         return ret;
 134 }
 135
 136 /* Move reqs from written to committing lists, returning count
 137  * of number moved.
 138  */
 139 int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo,
 140                                    int max)
 141 {
 142         int i, rv = 0, cnt;
 143
 144         lockdep_assert_held(cinfo->lock);
 145         for (i = 0; i < cinfo->ds->nbuckets && max != 0; i++) {
 146                 cnt = pnfs_generic_scan_ds_commit_list(&cinfo->ds->buckets[i],
 147                                                        cinfo, max);
 148                 max -= cnt;
 149                 rv += cnt;
 150         }
 151         return rv;
 152 }
 153 EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
 154
 155 /* Pull everything off the committing lists and dump into @dst.  */
 156 void pnfs_generic_recover_commit_reqs(struct list_head *dst,
 157                                       struct nfs_commit_info *cinfo)
 158 {
 159         struct pnfs_commit_bucket *b;
 160         struct pnfs_layout_segment *freeme;
 161         int i;
 162
 163         lockdep_assert_held(cinfo->lock);
 164 restart:
 165         for (i = 0, b = cinfo->ds->buckets; i < cinfo->ds->nbuckets; i++, b++) {
 166                 if (pnfs_generic_transfer_commit_list(&b->written, dst,
 167                                                       cinfo, 0)) {
 168                         freeme = b->wlseg;
 169                         b->wlseg = NULL;
 170                         spin_unlock(cinfo->lock);
 171                         pnfs_put_lseg(freeme);
 172                         spin_lock(cinfo->lock);
 173                         goto restart;
 174                 }
 175         }
 176         cinfo->ds->nwritten = 0;
 177 }
 178 EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
 179
 180 static void pnfs_generic_retry_commit(struct nfs_commit_info *cinfo, int idx)
 181 {
 182         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 183         struct pnfs_commit_bucket *bucket;
 184         struct pnfs_layout_segment *freeme;
 185         int i;
 186
 187         for (i = idx; i < fl_cinfo->nbuckets; i++) {
 188                 bucket = &fl_cinfo->buckets[i];
 189                 if (list_empty(&bucket->committing))
 190                         continue;
 191                 nfs_retry_commit(&bucket->committing, bucket->clseg, cinfo, i);
 192                 spin_lock(cinfo->lock);
 193                 freeme = bucket->clseg;
 194                 bucket->clseg = NULL;
 195                 spin_unlock(cinfo->lock);
 196                 pnfs_put_lseg(freeme);
 197         }
 198 }
 199
 200 static unsigned int
 201 pnfs_generic_alloc_ds_commits(struct nfs_commit_info *cinfo,
 202                               struct list_head *list)
 203 {
 204         struct pnfs_ds_commit_info *fl_cinfo;
 205         struct pnfs_commit_bucket *bucket;
 206         struct nfs_commit_data *data;
 207         int i;
 208         unsigned int nreq = 0;
 209
 210         fl_cinfo = cinfo->ds;
 211         bucket = fl_cinfo->buckets;
 212         for (i = 0; i < fl_cinfo->nbuckets; i++, bucket++) {
 213                 if (list_empty(&bucket->committing))
 214                         continue;
 215                 data = nfs_commitdata_alloc();
 216                 if (!data)
 217                         break;
 218                 data->ds_commit_index = i;
 219                 spin_lock(cinfo->lock);
 220                 data->lseg = bucket->clseg;
 221                 bucket->clseg = NULL;
 222                 spin_unlock(cinfo->lock);
 223                 list_add(&data->pages, list);
 224                 nreq++;
 225         }
 226
 227         /* Clean up on error */
 228         pnfs_generic_retry_commit(cinfo, i);
 229         return nreq;
 230 }
 231
 232 /* This follows nfs_commit_list pretty closely */
 233 int
 234 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
 235                              int how, struct nfs_commit_info *cinfo,
 236                              int (*initiate_commit)(struct nfs_commit_data *data,
 237                                                     int how))
 238 {
 239         struct nfs_commit_data *data, *tmp;
 240         LIST_HEAD(list);
 241         unsigned int nreq = 0;
 242
 243         if (!list_empty(mds_pages)) {
 244                 data = nfs_commitdata_alloc();
 245                 if (data != NULL) {
 246                         data->lseg = NULL;
 247                         list_add(&data->pages, &list);
 248                         nreq++;
 249                 } else {
 250                         nfs_retry_commit(mds_pages, NULL, cinfo, 0);
 251                         pnfs_generic_retry_commit(cinfo, 0);
 252                         cinfo->completion_ops->error_cleanup(NFS_I(inode));
 253                         return -ENOMEM;
 254                 }
 255         }
 256
 257         nreq += pnfs_generic_alloc_ds_commits(cinfo, &list);
 258
 259         if (nreq == 0) {
 260                 cinfo->completion_ops->error_cleanup(NFS_I(inode));
 261                 goto out;
 262         }
 263
 264         atomic_add(nreq, &cinfo->mds->rpcs_out);
 265
 266         list_for_each_entry_safe(data, tmp, &list, pages) {
 267                 list_del_init(&data->pages);
 268                 if (!data->lseg) {
 269                         nfs_init_commit(data, mds_pages, NULL, cinfo);
 270                         nfs_initiate_commit(NFS_CLIENT(inode), data,
 271                                             NFS_PROTO(data->inode),
 272                                             data->mds_ops, how, 0);
 273                 } else {
 274                         struct pnfs_commit_bucket *buckets;
 275
 276                         buckets = cinfo->ds->buckets;
 277                         nfs_init_commit(data,
 278                                         &buckets[data->ds_commit_index].committing,
 279                                         data->lseg,
 280                                         cinfo);
 281                         initiate_commit(data, how);
 282                 }
 283         }
 284 out:
 285         cinfo->ds->ncommitting = 0;
 286         return PNFS_ATTEMPTED;
 287 }
 288 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
 289
 290 /*
 291  * Data server cache
 292  *
 293  * Data servers can be mapped to different device ids.
 294  * nfs4_pnfs_ds reference counting
 295  *   - set to 1 on allocation
 296  *   - incremented when a device id maps a data server already in the cache.
 297  *   - decremented when deviceid is removed from the cache.
 298  */
 299 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
 300 static LIST_HEAD(nfs4_data_server_cache);
 301
 302 /* Debug routines */
 303 static void
 304 print_ds(struct nfs4_pnfs_ds *ds)
 305 {
 306         if (ds == NULL) {
 307                 printk(KERN_WARNING "%s NULL device\n", __func__);
 308                 return;
 309         }
 310         printk(KERN_WARNING "        ds %s\n"
 311                 "        ref count %d\n"
 312                 "        client %p\n"
 313                 "        cl_exchange_flags %x\n",
 314                 ds->ds_remotestr,
 315                 atomic_read(&ds->ds_count), ds->ds_clp,
 316                 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
 317 }
 318
 319 static bool
 320 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
 321 {
 322         struct sockaddr_in *a, *b;
 323         struct sockaddr_in6 *a6, *b6;
 324
 325         if (addr1->sa_family != addr2->sa_family)
 326                 return false;
 327
 328         switch (addr1->sa_family) {
 329         case AF_INET:
 330                 a = (struct sockaddr_in *)addr1;
 331                 b = (struct sockaddr_in *)addr2;
 332
 333                 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
 334                     a->sin_port == b->sin_port)
 335                         return true;
 336                 break;
 337
 338         case AF_INET6:
 339                 a6 = (struct sockaddr_in6 *)addr1;
 340                 b6 = (struct sockaddr_in6 *)addr2;
 341
 342                 /* LINKLOCAL addresses must have matching scope_id */
 343                 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
 344                     IPV6_ADDR_SCOPE_LINKLOCAL &&
 345                     a6->sin6_scope_id != b6->sin6_scope_id)
 346                         return false;
 347
 348                 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
 349                     a6->sin6_port == b6->sin6_port)
 350                         return true;
 351                 break;
 352
 353         default:
 354                 dprintk("%s: unhandled address family: %u\n",
 355                         __func__, addr1->sa_family);
 356                 return false;
 357         }
 358
 359         return false;
 360 }
 361
 362 static bool
 363 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 364                                const struct list_head *dsaddrs2)
 365 {
 366         struct nfs4_pnfs_ds_addr *da1, *da2;
 367
 368         /* step through both lists, comparing as we go */
 369         for (da1 = list_first_entry(dsaddrs1, typeof(*da1), da_node),
 370              da2 = list_first_entry(dsaddrs2, typeof(*da2), da_node);
 371              da1 != NULL && da2 != NULL;
 372              da1 = list_entry(da1->da_node.next, typeof(*da1), da_node),
 373              da2 = list_entry(da2->da_node.next, typeof(*da2), da_node)) {
 374                 if (!same_sockaddr((struct sockaddr *)&da1->da_addr,
 375                                    (struct sockaddr *)&da2->da_addr))
 376                         return false;
 377         }
 378         if (da1 == NULL && da2 == NULL)
 379                 return true;
 380
 381         return false;
 382 }
 383
 384 /*
 385  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 386  */
 387 static struct nfs4_pnfs_ds *
 388 _data_server_lookup_locked(const struct list_head *dsaddrs)
 389 {
 390         struct nfs4_pnfs_ds *ds;
 391
 392         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 393                 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 394                         return ds;
 395         return NULL;
 396 }
 397
 398 static void destroy_ds(struct nfs4_pnfs_ds *ds)
 399 {
 400         struct nfs4_pnfs_ds_addr *da;
 401
 402         dprintk("--> %s\n", __func__);
 403         ifdebug(FACILITY)
 404                 print_ds(ds);
 405
 406         nfs_put_client(ds->ds_clp);
 407
 408         while (!list_empty(&ds->ds_addrs)) {
 409                 da = list_first_entry(&ds->ds_addrs,
 410                                       struct nfs4_pnfs_ds_addr,
 411                                       da_node);
 412                 list_del_init(&da->da_node);
 413                 kfree(da->da_remotestr);
 414                 kfree(da);
 415         }
 416
 417         kfree(ds->ds_remotestr);
 418         kfree(ds);
 419 }
 420
 421 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
 422 {
 423         if (atomic_dec_and_lock(&ds->ds_count,
 424                                 &nfs4_ds_cache_lock)) {
 425                 list_del_init(&ds->ds_node);
 426                 spin_unlock(&nfs4_ds_cache_lock);
 427                 destroy_ds(ds);
 428         }
 429 }
 430 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
 431
 432 /*
 433  * Create a string with a human readable address and port to avoid
 434  * complicated setup around many dprinks.
 435  */
 436 static char *
 437 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 438 {
 439         struct nfs4_pnfs_ds_addr *da;
 440         char *remotestr;
 441         size_t len;
 442         char *p;
 443
 444         len = 3;        /* '{', '}' and eol */
 445         list_for_each_entry(da, dsaddrs, da_node) {
 446                 len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 447         }
 448
 449         remotestr = kzalloc(len, gfp_flags);
 450         if (!remotestr)
 451                 return NULL;
 452
 453         p = remotestr;
 454         *(p++) = '{';
 455         len--;
 456         list_for_each_entry(da, dsaddrs, da_node) {
 457                 size_t ll = strlen(da->da_remotestr);
 458
 459                 if (ll > len)
 460                         goto out_err;
 461
 462                 memcpy(p, da->da_remotestr, ll);
 463                 p += ll;
 464                 len -= ll;
 465
 466                 if (len < 1)
 467                         goto out_err;
 468                 (*p++) = ',';
 469                 len--;
 470         }
 471         if (len < 2)
 472                 goto out_err;
 473         *(p++) = '}';
 474         *p = '\0';
 475         return remotestr;
 476 out_err:
 477         kfree(remotestr);
 478         return NULL;
 479 }
 480
 481 /*
 482  * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
 483  * uncached and return cached struct nfs4_pnfs_ds.
 484  */
 485 struct nfs4_pnfs_ds *
 486 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
 487 {
 488         struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 489         char *remotestr;
 490
 491         if (list_empty(dsaddrs)) {
 492                 dprintk("%s: no addresses defined\n", __func__);
 493                 goto out;
 494         }
 495
 496         ds = kzalloc(sizeof(*ds), gfp_flags);
 497         if (!ds)
 498                 goto out;
 499
 500         /* this is only used for debugging, so it's ok if its NULL */
 501         remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 502
 503         spin_lock(&nfs4_ds_cache_lock);
 504         tmp_ds = _data_server_lookup_locked(dsaddrs);
 505         if (tmp_ds == NULL) {
 506                 INIT_LIST_HEAD(&ds->ds_addrs);
 507                 list_splice_init(dsaddrs, &ds->ds_addrs);
 508                 ds->ds_remotestr = remotestr;
 509                 atomic_set(&ds->ds_count, 1);
 510                 INIT_LIST_HEAD(&ds->ds_node);
 511                 ds->ds_clp = NULL;
 512                 list_add(&ds->ds_node, &nfs4_data_server_cache);
 513                 dprintk("%s add new data server %s\n", __func__,
 514                         ds->ds_remotestr);
 515         } else {
 516                 kfree(remotestr);
 517                 kfree(ds);
 518                 atomic_inc(&tmp_ds->ds_count);
 519                 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 520                         __func__, tmp_ds->ds_remotestr,
 521                         atomic_read(&tmp_ds->ds_count));
 522                 ds = tmp_ds;
 523         }
 524         spin_unlock(&nfs4_ds_cache_lock);
 525 out:
 526         return ds;
 527 }
 528 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
 529
 530 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
 531 {
 532         might_sleep();
 533         wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
 534                         TASK_KILLABLE);
 535 }
 536
 537 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
 538 {
 539         smp_mb__before_atomic();
 540         clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
 541         smp_mb__after_atomic();
 542         wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
 543 }
 544
 545 static struct nfs_client *(*get_v3_ds_connect)(
 546                         struct nfs_client *mds_clp,
 547                         const struct sockaddr *ds_addr,
 548                         int ds_addrlen,
 549                         int ds_proto,
 550                         unsigned int ds_timeo,
 551                         unsigned int ds_retrans,
 552                         rpc_authflavor_t au_flavor);
 553
 554 static bool load_v3_ds_connect(void)
 555 {
 556         if (!get_v3_ds_connect) {
 557                 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
 558                 WARN_ON_ONCE(!get_v3_ds_connect);
 559         }
 560
 561         return(get_v3_ds_connect != NULL);
 562 }
 563
 564 void nfs4_pnfs_v3_ds_connect_unload(void)
 565 {
 566         if (get_v3_ds_connect) {
 567                 symbol_put(nfs3_set_ds_client);
 568                 get_v3_ds_connect = NULL;
 569         }
 570 }
 571 EXPORT_SYMBOL_GPL(nfs4_pnfs_v3_ds_connect_unload);
 572
 573 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
 574                                  struct nfs4_pnfs_ds *ds,
 575                                  unsigned int timeo,
 576                                  unsigned int retrans,
 577                                  rpc_authflavor_t au_flavor)
 578 {
 579         struct nfs_client *clp = ERR_PTR(-EIO);
 580         struct nfs4_pnfs_ds_addr *da;
 581         int status = 0;
 582
 583         dprintk("--> %s DS %s au_flavor %d\n", __func__,
 584                 ds->ds_remotestr, au_flavor);
 585
 586         if (!load_v3_ds_connect())
 587                 goto out;
 588
 589         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 590                 dprintk("%s: DS %s: trying address %s\n",
 591                         __func__, ds->ds_remotestr, da->da_remotestr);
 592
 593                 clp = get_v3_ds_connect(mds_srv->nfs_client,
 594                                         (struct sockaddr *)&da->da_addr,
 595                                         da->da_addrlen, IPPROTO_TCP,
 596                                         timeo, retrans, au_flavor);
 597                 if (!IS_ERR(clp))
 598                         break;
 599         }
 600
 601         if (IS_ERR(clp)) {
 602                 status = PTR_ERR(clp);
 603                 goto out;
 604         }
 605
 606         smp_wmb();
 607         ds->ds_clp = clp;
 608         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 609 out:
 610         return status;
 611 }
 612
 613 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
 614                                  struct nfs4_pnfs_ds *ds,
 615                                  unsigned int timeo,
 616                                  unsigned int retrans,
 617                                  u32 minor_version,
 618                                  rpc_authflavor_t au_flavor)
 619 {
 620         struct nfs_client *clp = ERR_PTR(-EIO);
 621         struct nfs4_pnfs_ds_addr *da;
 622         int status = 0;
 623
 624         dprintk("--> %s DS %s au_flavor %d\n", __func__, ds->ds_remotestr,
 625                 au_flavor);
 626
 627         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 628                 dprintk("%s: DS %s: trying address %s\n",
 629                         __func__, ds->ds_remotestr, da->da_remotestr);
 630
 631                 clp = nfs4_set_ds_client(mds_srv->nfs_client,
 632                                         (struct sockaddr *)&da->da_addr,
 633                                         da->da_addrlen, IPPROTO_TCP,
 634                                         timeo, retrans, minor_version,
 635                                         au_flavor);
 636                 if (!IS_ERR(clp))
 637                         break;
 638         }
 639
 640         if (IS_ERR(clp)) {
 641                 status = PTR_ERR(clp);
 642                 goto out;
 643         }
 644
 645         status = nfs4_init_ds_session(clp, mds_srv->nfs_client->cl_lease_time);
 646         if (status)
 647                 goto out_put;
 648
 649         smp_wmb();
 650         ds->ds_clp = clp;
 651         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 652 out:
 653         return status;
 654 out_put:
 655         nfs_put_client(clp);
 656         goto out;
 657 }
 658
 659 /*
 660  * Create an rpc connection to the nfs4_pnfs_ds data server.
 661  * Currently only supports IPv4 and IPv6 addresses.
 662  * If connection fails, make devid unavailable.
 663  */
 664 void nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
 665                           struct nfs4_deviceid_node *devid, unsigned int timeo,
 666                           unsigned int retrans, u32 version,
 667                           u32 minor_version, rpc_authflavor_t au_flavor)
 668 {
 669         if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
 670                 int err = 0;
 671
 672                 if (version == 3) {
 673                         err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
 674                                                        retrans, au_flavor);
 675                 } else if (version == 4) {
 676                         err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
 677                                                        retrans, minor_version,
 678                                                        au_flavor);
 679                 } else {
 680                         dprintk("%s: unsupported DS version %d\n", __func__,
 681                                 version);
 682                         err = -EPROTONOSUPPORT;
 683                 }
 684
 685                 if (err)
 686                         nfs4_mark_deviceid_unavailable(devid);
 687                 nfs4_clear_ds_conn_bit(ds);
 688         } else {
 689                 nfs4_wait_ds_connect(ds);
 690         }
 691 }
 692 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
 693
 694 /*
 695  * Currently only supports ipv4, ipv6 and one multi-path address.
 696  */
 697 struct nfs4_pnfs_ds_addr *
 698 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
 699 {
 700         struct nfs4_pnfs_ds_addr *da = NULL;
 701         char *buf, *portstr;
 702         __be16 port;
 703         int nlen, rlen;
 704         int tmp[2];
 705         __be32 *p;
 706         char *netid, *match_netid;
 707         size_t len, match_netid_len;
 708         char *startsep = "";
 709         char *endsep = "";
 710
 711
 712         /* r_netid */
 713         p = xdr_inline_decode(xdr, 4);
 714         if (unlikely(!p))
 715                 goto out_err;
 716         nlen = be32_to_cpup(p++);
 717
 718         p = xdr_inline_decode(xdr, nlen);
 719         if (unlikely(!p))
 720                 goto out_err;
 721
 722         netid = kmalloc(nlen+1, gfp_flags);
 723         if (unlikely(!netid))
 724                 goto out_err;
 725
 726         netid[nlen] = '\0';
 727         memcpy(netid, p, nlen);
 728
 729         /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
 730         p = xdr_inline_decode(xdr, 4);
 731         if (unlikely(!p))
 732                 goto out_free_netid;
 733         rlen = be32_to_cpup(p);
 734
 735         p = xdr_inline_decode(xdr, rlen);
 736         if (unlikely(!p))
 737                 goto out_free_netid;
 738
 739         /* port is ".ABC.DEF", 8 chars max */
 740         if (rlen > INET6_ADDRSTRLEN + IPV6_SCOPE_ID_LEN + 8) {
 741                 dprintk("%s: Invalid address, length %d\n", __func__,
 742                         rlen);
 743                 goto out_free_netid;
 744         }
 745         buf = kmalloc(rlen + 1, gfp_flags);
 746         if (!buf) {
 747                 dprintk("%s: Not enough memory\n", __func__);
 748                 goto out_free_netid;
 749         }
 750         buf[rlen] = '\0';
 751         memcpy(buf, p, rlen);
 752
 753         /* replace port '.' with '-' */
 754         portstr = strrchr(buf, '.');
 755         if (!portstr) {
 756                 dprintk("%s: Failed finding expected dot in port\n",
 757                         __func__);
 758                 goto out_free_buf;
 759         }
 760         *portstr = '-';
 761
 762         /* find '.' between address and port */
 763         portstr = strrchr(buf, '.');
 764         if (!portstr) {
 765                 dprintk("%s: Failed finding expected dot between address and "
 766                         "port\n", __func__);
 767                 goto out_free_buf;
 768         }
 769         *portstr = '\0';
 770
 771         da = kzalloc(sizeof(*da), gfp_flags);
 772         if (unlikely(!da))
 773                 goto out_free_buf;
 774
 775         INIT_LIST_HEAD(&da->da_node);
 776
 777         if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
 778                       sizeof(da->da_addr))) {
 779                 dprintk("%s: error parsing address %s\n", __func__, buf);
 780                 goto out_free_da;
 781         }
 782
 783         portstr++;
 784         sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
 785         port = htons((tmp[0] << 8) | (tmp[1]));
 786
 787         switch (da->da_addr.ss_family) {
 788         case AF_INET:
 789                 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
 790                 da->da_addrlen = sizeof(struct sockaddr_in);
 791                 match_netid = "tcp";
 792                 match_netid_len = 3;
 793                 break;
 794
 795         case AF_INET6:
 796                 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
 797                 da->da_addrlen = sizeof(struct sockaddr_in6);
 798                 match_netid = "tcp6";
 799                 match_netid_len = 4;
 800                 startsep = "[";
 801                 endsep = "]";
 802                 break;
 803
 804         default:
 805                 dprintk("%s: unsupported address family: %u\n",
 806                         __func__, da->da_addr.ss_family);
 807                 goto out_free_da;
 808         }
 809
 810         if (nlen != match_netid_len || strncmp(netid, match_netid, nlen)) {
 811                 dprintk("%s: ERROR: r_netid \"%s\" != \"%s\"\n",
 812                         __func__, netid, match_netid);
 813                 goto out_free_da;
 814         }
 815
 816         /* save human readable address */
 817         len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
 818         da->da_remotestr = kzalloc(len, gfp_flags);
 819
 820         /* NULL is ok, only used for dprintk */
 821         if (da->da_remotestr)
 822                 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
 823                          buf, endsep, ntohs(port));
 824
 825         dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
 826         kfree(buf);
 827         kfree(netid);
 828         return da;
 829
 830 out_free_da:
 831         kfree(da);
 832 out_free_buf:
 833         dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
 834         kfree(buf);
 835 out_free_netid:
 836         kfree(netid);
 837 out_err:
 838         return NULL;
 839 }
 840 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
 841
 842 void
 843 pnfs_layout_mark_request_commit(struct nfs_page *req,
 844                                 struct pnfs_layout_segment *lseg,
 845                                 struct nfs_commit_info *cinfo,
 846                                 u32 ds_commit_idx)
 847 {
 848         struct list_head *list;
 849         struct pnfs_commit_bucket *buckets;
 850
 851         spin_lock(cinfo->lock);
 852         buckets = cinfo->ds->buckets;
 853         list = &buckets[ds_commit_idx].written;
 854         if (list_empty(list)) {
 855                 /* Non-empty buckets hold a reference on the lseg.  That ref
 856                  * is normally transferred to the COMMIT call and released
 857                  * there.  It could also be released if the last req is pulled
 858                  * off due to a rewrite, in which case it will be done in
 859                  * pnfs_common_clear_request_commit
 860                  */
 861                 WARN_ON_ONCE(buckets[ds_commit_idx].wlseg != NULL);
 862                 buckets[ds_commit_idx].wlseg = pnfs_get_lseg(lseg);
 863         }
 864         set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
 865         cinfo->ds->nwritten++;
 866         spin_unlock(cinfo->lock);
 867
 868         nfs_request_add_commit_list(req, list, cinfo);
 869 }
 870 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
 871
 872 int
 873 pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
 874 {
 875         if (datasync)
 876                 return 0;
 877         return pnfs_layoutcommit_inode(inode, true);
 878 }
 879 EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
 880