fs/nfs/pnfs_nfs.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Common NFS I/O  operations for the pnfs file based
   4  * layout drivers.
   5  *
   6  * Copyright (c) 2014, Primary Data, Inc. All rights reserved.
   7  *
   8  * Tom Haynes <loghyr@primarydata.com>
   9  */
  10
  11 #include <linux/nfs_fs.h>
  12 #include <linux/nfs_page.h>
  13 #include <linux/sunrpc/addr.h>
  14 #include <linux/module.h>
  15
  16 #include "nfs4session.h"
  17 #include "internal.h"
  18 #include "pnfs.h"
  19
  20 #define NFSDBG_FACILITY         NFSDBG_PNFS
  21
  22 void pnfs_generic_rw_release(void *data)
  23 {
  24         struct nfs_pgio_header *hdr = data;
  25
  26         nfs_put_client(hdr->ds_clp);
  27         hdr->mds_ops->rpc_release(data);
  28 }
  29 EXPORT_SYMBOL_GPL(pnfs_generic_rw_release);
  30
  31 /* Fake up some data that will cause nfs_commit_release to retry the writes. */
  32 void pnfs_generic_prepare_to_resend_writes(struct nfs_commit_data *data)
  33 {
  34         struct nfs_writeverf *verf = data->res.verf;
  35
  36         data->task.tk_status = 0;
  37         memset(&verf->verifier, 0, sizeof(verf->verifier));
  38         verf->committed = NFS_UNSTABLE;
  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 static struct pnfs_layout_segment *
  63 pnfs_free_bucket_lseg(struct pnfs_commit_bucket *bucket)
  64 {
  65         if (list_empty(&bucket->committing) && list_empty(&bucket->written)) {
  66                 struct pnfs_layout_segment *freeme = bucket->lseg;
  67                 bucket->lseg = NULL;
  68                 return freeme;
  69         }
  70         return NULL;
  71 }
  72
  73 /* The generic layer is about to remove the req from the commit list.
  74  * If this will make the bucket empty, it will need to put the lseg reference.
  75  * Note this must be called holding nfsi->commit_mutex
  76  */
  77 void
  78 pnfs_generic_clear_request_commit(struct nfs_page *req,
  79                                   struct nfs_commit_info *cinfo)
  80 {
  81         struct pnfs_commit_bucket *bucket = NULL;
  82
  83         if (!test_and_clear_bit(PG_COMMIT_TO_DS, &req->wb_flags))
  84                 goto out;
  85         cinfo->ds->nwritten--;
  86         if (list_is_singular(&req->wb_list))
  87                 bucket = list_first_entry(&req->wb_list,
  88                                           struct pnfs_commit_bucket, written);
  89 out:
  90         nfs_request_remove_commit_list(req, cinfo);
  91         if (bucket)
  92                 pnfs_put_lseg(pnfs_free_bucket_lseg(bucket));
  93 }
  94 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
  95
  96 struct pnfs_commit_array *
  97 pnfs_alloc_commit_array(size_t n, gfp_t gfp_flags)
  98 {
  99         struct pnfs_commit_array *p;
 100         struct pnfs_commit_bucket *b;
 101
 102         p = kmalloc(struct_size(p, buckets, n), gfp_flags);
 103         if (!p)
 104                 return NULL;
 105         p->nbuckets = n;
 106         INIT_LIST_HEAD(&p->cinfo_list);
 107         INIT_LIST_HEAD(&p->lseg_list);
 108         p->lseg = NULL;
 109         for (b = &p->buckets[0]; n != 0; b++, n--) {
 110                 INIT_LIST_HEAD(&b->written);
 111                 INIT_LIST_HEAD(&b->committing);
 112                 b->lseg = NULL;
 113                 b->direct_verf.committed = NFS_INVALID_STABLE_HOW;
 114         }
 115         return p;
 116 }
 117 EXPORT_SYMBOL_GPL(pnfs_alloc_commit_array);
 118
 119 void
 120 pnfs_free_commit_array(struct pnfs_commit_array *p)
 121 {
 122         kfree_rcu(p, rcu);
 123 }
 124 EXPORT_SYMBOL_GPL(pnfs_free_commit_array);
 125
 126 static struct pnfs_commit_array *
 127 pnfs_find_commit_array_by_lseg(struct pnfs_ds_commit_info *fl_cinfo,
 128                 struct pnfs_layout_segment *lseg)
 129 {
 130         struct pnfs_commit_array *array;
 131
 132         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 133                 if (array->lseg == lseg)
 134                         return array;
 135         }
 136         return NULL;
 137 }
 138
 139 struct pnfs_commit_array *
 140 pnfs_add_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
 141                 struct pnfs_commit_array *new,
 142                 struct pnfs_layout_segment *lseg)
 143 {
 144         struct pnfs_commit_array *array;
 145
 146         array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 147         if (array)
 148                 return array;
 149         new->lseg = lseg;
 150         refcount_set(&new->refcount, 1);
 151         list_add_rcu(&new->cinfo_list, &fl_cinfo->commits);
 152         list_add(&new->lseg_list, &lseg->pls_commits);
 153         return new;
 154 }
 155 EXPORT_SYMBOL_GPL(pnfs_add_commit_array);
 156
 157 static struct pnfs_commit_array *
 158 pnfs_lookup_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
 159                 struct pnfs_layout_segment *lseg)
 160 {
 161         struct pnfs_commit_array *array;
 162
 163         rcu_read_lock();
 164         array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 165         if (!array) {
 166                 rcu_read_unlock();
 167                 fl_cinfo->ops->setup_ds_info(fl_cinfo, lseg);
 168                 rcu_read_lock();
 169                 array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 170         }
 171         rcu_read_unlock();
 172         return array;
 173 }
 174
 175 static void
 176 pnfs_release_commit_array_locked(struct pnfs_commit_array *array)
 177 {
 178         list_del_rcu(&array->cinfo_list);
 179         list_del(&array->lseg_list);
 180         pnfs_free_commit_array(array);
 181 }
 182
 183 static void
 184 pnfs_put_commit_array_locked(struct pnfs_commit_array *array)
 185 {
 186         if (refcount_dec_and_test(&array->refcount))
 187                 pnfs_release_commit_array_locked(array);
 188 }
 189
 190 static void
 191 pnfs_put_commit_array(struct pnfs_commit_array *array, struct inode *inode)
 192 {
 193         if (refcount_dec_and_lock(&array->refcount, &inode->i_lock)) {
 194                 pnfs_release_commit_array_locked(array);
 195                 spin_unlock(&inode->i_lock);
 196         }
 197 }
 198
 199 static struct pnfs_commit_array *
 200 pnfs_get_commit_array(struct pnfs_commit_array *array)
 201 {
 202         if (refcount_inc_not_zero(&array->refcount))
 203                 return array;
 204         return NULL;
 205 }
 206
 207 static void
 208 pnfs_remove_and_free_commit_array(struct pnfs_commit_array *array)
 209 {
 210         array->lseg = NULL;
 211         list_del_init(&array->lseg_list);
 212         pnfs_put_commit_array_locked(array);
 213 }
 214
 215 void
 216 pnfs_generic_ds_cinfo_release_lseg(struct pnfs_ds_commit_info *fl_cinfo,
 217                 struct pnfs_layout_segment *lseg)
 218 {
 219         struct pnfs_commit_array *array, *tmp;
 220
 221         list_for_each_entry_safe(array, tmp, &lseg->pls_commits, lseg_list)
 222                 pnfs_remove_and_free_commit_array(array);
 223 }
 224 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_release_lseg);
 225
 226 void
 227 pnfs_generic_ds_cinfo_destroy(struct pnfs_ds_commit_info *fl_cinfo)
 228 {
 229         struct pnfs_commit_array *array, *tmp;
 230
 231         list_for_each_entry_safe(array, tmp, &fl_cinfo->commits, cinfo_list)
 232                 pnfs_remove_and_free_commit_array(array);
 233 }
 234 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_destroy);
 235
 236 /*
 237  * Locks the nfs_page requests for commit and moves them to
 238  * @bucket->committing.
 239  */
 240 static int
 241 pnfs_bucket_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
 242                                 struct nfs_commit_info *cinfo,
 243                                 int max)
 244 {
 245         struct list_head *src = &bucket->written;
 246         struct list_head *dst = &bucket->committing;
 247         int ret;
 248
 249         lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 250         ret = nfs_scan_commit_list(src, dst, cinfo, max);
 251         if (ret) {
 252                 cinfo->ds->nwritten -= ret;
 253                 cinfo->ds->ncommitting += ret;
 254         }
 255         return ret;
 256 }
 257
 258 static int pnfs_bucket_scan_array(struct nfs_commit_info *cinfo,
 259                                   struct pnfs_commit_bucket *buckets,
 260                                   unsigned int nbuckets,
 261                                   int max)
 262 {
 263         unsigned int i;
 264         int rv = 0, cnt;
 265
 266         for (i = 0; i < nbuckets && max != 0; i++) {
 267                 cnt = pnfs_bucket_scan_ds_commit_list(&buckets[i], cinfo, max);
 268                 rv += cnt;
 269                 max -= cnt;
 270         }
 271         return rv;
 272 }
 273
 274 /* Move reqs from written to committing lists, returning count
 275  * of number moved.
 276  */
 277 int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo, int max)
 278 {
 279         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 280         struct pnfs_commit_array *array;
 281         int rv = 0, cnt;
 282
 283         rcu_read_lock();
 284         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 285                 if (!array->lseg || !pnfs_get_commit_array(array))
 286                         continue;
 287                 rcu_read_unlock();
 288                 cnt = pnfs_bucket_scan_array(cinfo, array->buckets,
 289                                 array->nbuckets, max);
 290                 rcu_read_lock();
 291                 pnfs_put_commit_array(array, cinfo->inode);
 292                 rv += cnt;
 293                 max -= cnt;
 294                 if (!max)
 295                         break;
 296         }
 297         rcu_read_unlock();
 298         return rv;
 299 }
 300 EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
 301
 302 static unsigned int
 303 pnfs_bucket_recover_commit_reqs(struct list_head *dst,
 304                                 struct pnfs_commit_bucket *buckets,
 305                                 unsigned int nbuckets,
 306                                 struct nfs_commit_info *cinfo)
 307 {
 308         struct pnfs_commit_bucket *b;
 309         struct pnfs_layout_segment *freeme;
 310         unsigned int nwritten, ret = 0;
 311         unsigned int i;
 312
 313 restart:
 314         for (i = 0, b = buckets; i < nbuckets; i++, b++) {
 315                 nwritten = nfs_scan_commit_list(&b->written, dst, cinfo, 0);
 316                 if (!nwritten)
 317                         continue;
 318                 ret += nwritten;
 319                 freeme = pnfs_free_bucket_lseg(b);
 320                 if (freeme) {
 321                         pnfs_put_lseg(freeme);
 322                         goto restart;
 323                 }
 324         }
 325         return ret;
 326 }
 327
 328 /* Pull everything off the committing lists and dump into @dst.  */
 329 void pnfs_generic_recover_commit_reqs(struct list_head *dst,
 330                                       struct nfs_commit_info *cinfo)
 331 {
 332         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 333         struct pnfs_commit_array *array;
 334         unsigned int nwritten;
 335
 336         lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 337         rcu_read_lock();
 338         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 339                 if (!array->lseg || !pnfs_get_commit_array(array))
 340                         continue;
 341                 rcu_read_unlock();
 342                 nwritten = pnfs_bucket_recover_commit_reqs(dst,
 343                                                            array->buckets,
 344                                                            array->nbuckets,
 345                                                            cinfo);
 346                 rcu_read_lock();
 347                 pnfs_put_commit_array(array, cinfo->inode);
 348                 fl_cinfo->nwritten -= nwritten;
 349         }
 350         rcu_read_unlock();
 351 }
 352 EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
 353
 354 static struct nfs_page *
 355 pnfs_bucket_search_commit_reqs(struct pnfs_commit_bucket *buckets,
 356                 unsigned int nbuckets, struct page *page)
 357 {
 358         struct nfs_page *req;
 359         struct pnfs_commit_bucket *b;
 360         unsigned int i;
 361
 362         /* Linearly search the commit lists for each bucket until a matching
 363          * request is found */
 364         for (i = 0, b = buckets; i < nbuckets; i++, b++) {
 365                 list_for_each_entry(req, &b->written, wb_list) {
 366                         if (req->wb_page == page)
 367                                 return req->wb_head;
 368                 }
 369                 list_for_each_entry(req, &b->committing, wb_list) {
 370                         if (req->wb_page == page)
 371                                 return req->wb_head;
 372                 }
 373         }
 374         return NULL;
 375 }
 376
 377 /* pnfs_generic_search_commit_reqs - Search lists in @cinfo for the head reqest
 378  *                                 for @page
 379  * @cinfo - commit info for current inode
 380  * @page - page to search for matching head request
 381  *
 382  * Returns a the head request if one is found, otherwise returns NULL.
 383  */
 384 struct nfs_page *
 385 pnfs_generic_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
 386 {
 387         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 388         struct pnfs_commit_array *array;
 389         struct nfs_page *req;
 390
 391         list_for_each_entry(array, &fl_cinfo->commits, cinfo_list) {
 392                 req = pnfs_bucket_search_commit_reqs(array->buckets,
 393                                 array->nbuckets, page);
 394                 if (req)
 395                         return req;
 396         }
 397         return NULL;
 398 }
 399 EXPORT_SYMBOL_GPL(pnfs_generic_search_commit_reqs);
 400
 401 static struct pnfs_layout_segment *
 402 pnfs_bucket_get_committing(struct list_head *head,
 403                            struct pnfs_commit_bucket *bucket,
 404                            struct nfs_commit_info *cinfo)
 405 {
 406         struct pnfs_layout_segment *lseg;
 407         struct list_head *pos;
 408
 409         list_for_each(pos, &bucket->committing)
 410                 cinfo->ds->ncommitting--;
 411         list_splice_init(&bucket->committing, head);
 412         lseg = pnfs_free_bucket_lseg(bucket);
 413         if (!lseg)
 414                 lseg = pnfs_get_lseg(bucket->lseg);
 415         return lseg;
 416 }
 417
 418 static struct nfs_commit_data *
 419 pnfs_bucket_fetch_commitdata(struct pnfs_commit_bucket *bucket,
 420                              struct nfs_commit_info *cinfo)
 421 {
 422         struct nfs_commit_data *data = nfs_commitdata_alloc(false);
 423
 424         if (!data)
 425                 return NULL;
 426         data->lseg = pnfs_bucket_get_committing(&data->pages, bucket, cinfo);
 427         return data;
 428 }
 429
 430 static void pnfs_generic_retry_commit(struct pnfs_commit_bucket *buckets,
 431                                       unsigned int nbuckets,
 432                                       struct nfs_commit_info *cinfo,
 433                                       unsigned int idx)
 434 {
 435         struct pnfs_commit_bucket *bucket;
 436         struct pnfs_layout_segment *freeme;
 437         LIST_HEAD(pages);
 438
 439         for (bucket = buckets; idx < nbuckets; bucket++, idx++) {
 440                 if (list_empty(&bucket->committing))
 441                         continue;
 442                 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 443                 freeme = pnfs_bucket_get_committing(&pages, bucket, cinfo);
 444                 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 445                 nfs_retry_commit(&pages, freeme, cinfo, idx);
 446                 pnfs_put_lseg(freeme);
 447         }
 448 }
 449
 450 static unsigned int
 451 pnfs_bucket_alloc_ds_commits(struct list_head *list,
 452                              struct pnfs_commit_bucket *buckets,
 453                              unsigned int nbuckets,
 454                              struct nfs_commit_info *cinfo)
 455 {
 456         struct pnfs_commit_bucket *bucket;
 457         struct nfs_commit_data *data;
 458         unsigned int i;
 459         unsigned int nreq = 0;
 460
 461         for (i = 0, bucket = buckets; i < nbuckets; i++, bucket++) {
 462                 if (list_empty(&bucket->committing))
 463                         continue;
 464                 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 465                 if (!list_empty(&bucket->committing)) {
 466                         data = pnfs_bucket_fetch_commitdata(bucket, cinfo);
 467                         if (!data)
 468                                 goto out_error;
 469                         data->ds_commit_index = i;
 470                         list_add_tail(&data->list, list);
 471                         nreq++;
 472                 }
 473                 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 474         }
 475         return nreq;
 476 out_error:
 477         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 478         /* Clean up on error */
 479         pnfs_generic_retry_commit(buckets, nbuckets, cinfo, i);
 480         return nreq;
 481 }
 482
 483 static unsigned int
 484 pnfs_alloc_ds_commits_list(struct list_head *list,
 485                            struct pnfs_ds_commit_info *fl_cinfo,
 486                            struct nfs_commit_info *cinfo)
 487 {
 488         struct pnfs_commit_array *array;
 489         unsigned int ret = 0;
 490
 491         rcu_read_lock();
 492         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 493                 if (!array->lseg || !pnfs_get_commit_array(array))
 494                         continue;
 495                 rcu_read_unlock();
 496                 ret += pnfs_bucket_alloc_ds_commits(list, array->buckets,
 497                                 array->nbuckets, cinfo);
 498                 rcu_read_lock();
 499                 pnfs_put_commit_array(array, cinfo->inode);
 500         }
 501         rcu_read_unlock();
 502         return ret;
 503 }
 504
 505 /* This follows nfs_commit_list pretty closely */
 506 int
 507 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
 508                              int how, struct nfs_commit_info *cinfo,
 509                              int (*initiate_commit)(struct nfs_commit_data *data,
 510                                                     int how))
 511 {
 512         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 513         struct nfs_commit_data *data, *tmp;
 514         LIST_HEAD(list);
 515         unsigned int nreq = 0;
 516
 517         if (!list_empty(mds_pages)) {
 518                 data = nfs_commitdata_alloc(true);
 519                 data->ds_commit_index = -1;
 520                 list_splice_init(mds_pages, &data->pages);
 521                 list_add_tail(&data->list, &list);
 522                 nreq++;
 523         }
 524
 525         nreq += pnfs_alloc_ds_commits_list(&list, fl_cinfo, cinfo);
 526         if (nreq == 0)
 527                 goto out;
 528
 529         list_for_each_entry_safe(data, tmp, &list, list) {
 530                 list_del(&data->list);
 531                 if (data->ds_commit_index < 0) {
 532                         nfs_init_commit(data, NULL, NULL, cinfo);
 533                         nfs_initiate_commit(NFS_CLIENT(inode), data,
 534                                             NFS_PROTO(data->inode),
 535                                             data->mds_ops, how,
 536                                             RPC_TASK_CRED_NOREF);
 537                 } else {
 538                         nfs_init_commit(data, NULL, data->lseg, cinfo);
 539                         initiate_commit(data, how);
 540                 }
 541         }
 542 out:
 543         return PNFS_ATTEMPTED;
 544 }
 545 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
 546
 547 /*
 548  * Data server cache
 549  *
 550  * Data servers can be mapped to different device ids.
 551  * nfs4_pnfs_ds reference counting
 552  *   - set to 1 on allocation
 553  *   - incremented when a device id maps a data server already in the cache.
 554  *   - decremented when deviceid is removed from the cache.
 555  */
 556 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
 557 static LIST_HEAD(nfs4_data_server_cache);
 558
 559 /* Debug routines */
 560 static void
 561 print_ds(struct nfs4_pnfs_ds *ds)
 562 {
 563         if (ds == NULL) {
 564                 printk(KERN_WARNING "%s NULL device\n", __func__);
 565                 return;
 566         }
 567         printk(KERN_WARNING "        ds %s\n"
 568                 "        ref count %d\n"
 569                 "        client %p\n"
 570                 "        cl_exchange_flags %x\n",
 571                 ds->ds_remotestr,
 572                 refcount_read(&ds->ds_count), ds->ds_clp,
 573                 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
 574 }
 575
 576 static bool
 577 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
 578 {
 579         struct sockaddr_in *a, *b;
 580         struct sockaddr_in6 *a6, *b6;
 581
 582         if (addr1->sa_family != addr2->sa_family)
 583                 return false;
 584
 585         switch (addr1->sa_family) {
 586         case AF_INET:
 587                 a = (struct sockaddr_in *)addr1;
 588                 b = (struct sockaddr_in *)addr2;
 589
 590                 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
 591                     a->sin_port == b->sin_port)
 592                         return true;
 593                 break;
 594
 595         case AF_INET6:
 596                 a6 = (struct sockaddr_in6 *)addr1;
 597                 b6 = (struct sockaddr_in6 *)addr2;
 598
 599                 /* LINKLOCAL addresses must have matching scope_id */
 600                 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
 601                     IPV6_ADDR_SCOPE_LINKLOCAL &&
 602                     a6->sin6_scope_id != b6->sin6_scope_id)
 603                         return false;
 604
 605                 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
 606                     a6->sin6_port == b6->sin6_port)
 607                         return true;
 608                 break;
 609
 610         default:
 611                 dprintk("%s: unhandled address family: %u\n",
 612                         __func__, addr1->sa_family);
 613                 return false;
 614         }
 615
 616         return false;
 617 }
 618
 619 /*
 620  * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
 621  * declare a match.
 622  */
 623 static bool
 624 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 625                                const struct list_head *dsaddrs2)
 626 {
 627         struct nfs4_pnfs_ds_addr *da1, *da2;
 628         struct sockaddr *sa1, *sa2;
 629         bool match = false;
 630
 631         list_for_each_entry(da1, dsaddrs1, da_node) {
 632                 sa1 = (struct sockaddr *)&da1->da_addr;
 633                 match = false;
 634                 list_for_each_entry(da2, dsaddrs2, da_node) {
 635                         sa2 = (struct sockaddr *)&da2->da_addr;
 636                         match = same_sockaddr(sa1, sa2);
 637                         if (match)
 638                                 break;
 639                 }
 640                 if (!match)
 641                         break;
 642         }
 643         return match;
 644 }
 645
 646 /*
 647  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 648  */
 649 static struct nfs4_pnfs_ds *
 650 _data_server_lookup_locked(const struct list_head *dsaddrs)
 651 {
 652         struct nfs4_pnfs_ds *ds;
 653
 654         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 655                 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 656                         return ds;
 657         return NULL;
 658 }
 659
 660 static struct nfs4_pnfs_ds_addr *nfs4_pnfs_ds_addr_alloc(gfp_t gfp_flags)
 661 {
 662         struct nfs4_pnfs_ds_addr *da = kzalloc(sizeof(*da), gfp_flags);
 663         if (da)
 664                 INIT_LIST_HEAD(&da->da_node);
 665         return da;
 666 }
 667
 668 static void nfs4_pnfs_ds_addr_free(struct nfs4_pnfs_ds_addr *da)
 669 {
 670         kfree(da->da_remotestr);
 671         kfree(da->da_netid);
 672         kfree(da);
 673 }
 674
 675 static void destroy_ds(struct nfs4_pnfs_ds *ds)
 676 {
 677         struct nfs4_pnfs_ds_addr *da;
 678
 679         dprintk("--> %s\n", __func__);
 680         ifdebug(FACILITY)
 681                 print_ds(ds);
 682
 683         nfs_put_client(ds->ds_clp);
 684
 685         while (!list_empty(&ds->ds_addrs)) {
 686                 da = list_first_entry(&ds->ds_addrs,
 687                                       struct nfs4_pnfs_ds_addr,
 688                                       da_node);
 689                 list_del_init(&da->da_node);
 690                 nfs4_pnfs_ds_addr_free(da);
 691         }
 692
 693         kfree(ds->ds_remotestr);
 694         kfree(ds);
 695 }
 696
 697 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
 698 {
 699         if (refcount_dec_and_lock(&ds->ds_count,
 700                                 &nfs4_ds_cache_lock)) {
 701                 list_del_init(&ds->ds_node);
 702                 spin_unlock(&nfs4_ds_cache_lock);
 703                 destroy_ds(ds);
 704         }
 705 }
 706 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
 707
 708 /*
 709  * Create a string with a human readable address and port to avoid
 710  * complicated setup around many dprinks.
 711  */
 712 static char *
 713 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 714 {
 715         struct nfs4_pnfs_ds_addr *da;
 716         char *remotestr;
 717         size_t len;
 718         char *p;
 719
 720         len = 3;        /* '{', '}' and eol */
 721         list_for_each_entry(da, dsaddrs, da_node) {
 722                 len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 723         }
 724
 725         remotestr = kzalloc(len, gfp_flags);
 726         if (!remotestr)
 727                 return NULL;
 728
 729         p = remotestr;
 730         *(p++) = '{';
 731         len--;
 732         list_for_each_entry(da, dsaddrs, da_node) {
 733                 size_t ll = strlen(da->da_remotestr);
 734
 735                 if (ll > len)
 736                         goto out_err;
 737
 738                 memcpy(p, da->da_remotestr, ll);
 739                 p += ll;
 740                 len -= ll;
 741
 742                 if (len < 1)
 743                         goto out_err;
 744                 (*p++) = ',';
 745                 len--;
 746         }
 747         if (len < 2)
 748                 goto out_err;
 749         *(p++) = '}';
 750         *p = '\0';
 751         return remotestr;
 752 out_err:
 753         kfree(remotestr);
 754         return NULL;
 755 }
 756
 757 /*
 758  * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
 759  * uncached and return cached struct nfs4_pnfs_ds.
 760  */
 761 struct nfs4_pnfs_ds *
 762 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
 763 {
 764         struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 765         char *remotestr;
 766
 767         if (list_empty(dsaddrs)) {
 768                 dprintk("%s: no addresses defined\n", __func__);
 769                 goto out;
 770         }
 771
 772         ds = kzalloc(sizeof(*ds), gfp_flags);
 773         if (!ds)
 774                 goto out;
 775
 776         /* this is only used for debugging, so it's ok if its NULL */
 777         remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 778
 779         spin_lock(&nfs4_ds_cache_lock);
 780         tmp_ds = _data_server_lookup_locked(dsaddrs);
 781         if (tmp_ds == NULL) {
 782                 INIT_LIST_HEAD(&ds->ds_addrs);
 783                 list_splice_init(dsaddrs, &ds->ds_addrs);
 784                 ds->ds_remotestr = remotestr;
 785                 refcount_set(&ds->ds_count, 1);
 786                 INIT_LIST_HEAD(&ds->ds_node);
 787                 ds->ds_clp = NULL;
 788                 list_add(&ds->ds_node, &nfs4_data_server_cache);
 789                 dprintk("%s add new data server %s\n", __func__,
 790                         ds->ds_remotestr);
 791         } else {
 792                 kfree(remotestr);
 793                 kfree(ds);
 794                 refcount_inc(&tmp_ds->ds_count);
 795                 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 796                         __func__, tmp_ds->ds_remotestr,
 797                         refcount_read(&tmp_ds->ds_count));
 798                 ds = tmp_ds;
 799         }
 800         spin_unlock(&nfs4_ds_cache_lock);
 801 out:
 802         return ds;
 803 }
 804 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
 805
 806 static int nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
 807 {
 808         might_sleep();
 809         return wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING, TASK_KILLABLE);
 810 }
 811
 812 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
 813 {
 814         smp_mb__before_atomic();
 815         clear_and_wake_up_bit(NFS4DS_CONNECTING, &ds->ds_state);
 816 }
 817
 818 static struct nfs_client *(*get_v3_ds_connect)(
 819                         struct nfs_server *mds_srv,
 820                         const struct sockaddr *ds_addr,
 821                         int ds_addrlen,
 822                         int ds_proto,
 823                         unsigned int ds_timeo,
 824                         unsigned int ds_retrans);
 825
 826 static bool load_v3_ds_connect(void)
 827 {
 828         if (!get_v3_ds_connect) {
 829                 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
 830                 WARN_ON_ONCE(!get_v3_ds_connect);
 831         }
 832
 833         return(get_v3_ds_connect != NULL);
 834 }
 835
 836 void nfs4_pnfs_v3_ds_connect_unload(void)
 837 {
 838         if (get_v3_ds_connect) {
 839                 symbol_put(nfs3_set_ds_client);
 840                 get_v3_ds_connect = NULL;
 841         }
 842 }
 843
 844 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
 845                                  struct nfs4_pnfs_ds *ds,
 846                                  unsigned int timeo,
 847                                  unsigned int retrans)
 848 {
 849         struct nfs_client *clp = ERR_PTR(-EIO);
 850         struct nfs4_pnfs_ds_addr *da;
 851         int status = 0;
 852
 853         dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 854
 855         if (!load_v3_ds_connect())
 856                 return -EPROTONOSUPPORT;
 857
 858         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 859                 dprintk("%s: DS %s: trying address %s\n",
 860                         __func__, ds->ds_remotestr, da->da_remotestr);
 861
 862                 if (!IS_ERR(clp)) {
 863                         struct xprt_create xprt_args = {
 864                                 .ident = da->da_transport,
 865                                 .net = clp->cl_net,
 866                                 .dstaddr = (struct sockaddr *)&da->da_addr,
 867                                 .addrlen = da->da_addrlen,
 868                                 .servername = clp->cl_hostname,
 869                         };
 870
 871                         if (da->da_transport != clp->cl_proto)
 872                                 continue;
 873                         if (da->da_addr.ss_family != clp->cl_addr.ss_family)
 874                                 continue;
 875                         /* Add this address as an alias */
 876                         rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 877                                         rpc_clnt_test_and_add_xprt, NULL);
 878                         continue;
 879                 }
 880                 clp = get_v3_ds_connect(mds_srv,
 881                                 (struct sockaddr *)&da->da_addr,
 882                                 da->da_addrlen, da->da_transport,
 883                                 timeo, retrans);
 884                 if (IS_ERR(clp))
 885                         continue;
 886                 clp->cl_rpcclient->cl_softerr = 0;
 887                 clp->cl_rpcclient->cl_softrtry = 0;
 888         }
 889
 890         if (IS_ERR(clp)) {
 891                 status = PTR_ERR(clp);
 892                 goto out;
 893         }
 894
 895         smp_wmb();
 896         WRITE_ONCE(ds->ds_clp, clp);
 897         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 898 out:
 899         return status;
 900 }
 901
 902 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
 903                                  struct nfs4_pnfs_ds *ds,
 904                                  unsigned int timeo,
 905                                  unsigned int retrans,
 906                                  u32 minor_version)
 907 {
 908         struct nfs_client *clp = ERR_PTR(-EIO);
 909         struct nfs4_pnfs_ds_addr *da;
 910         int status = 0;
 911
 912         dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 913
 914         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 915                 dprintk("%s: DS %s: trying address %s\n",
 916                         __func__, ds->ds_remotestr, da->da_remotestr);
 917
 918                 if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) {
 919                         struct xprt_create xprt_args = {
 920                                 .ident = da->da_transport,
 921                                 .net = clp->cl_net,
 922                                 .dstaddr = (struct sockaddr *)&da->da_addr,
 923                                 .addrlen = da->da_addrlen,
 924                                 .servername = clp->cl_hostname,
 925                         };
 926                         struct nfs4_add_xprt_data xprtdata = {
 927                                 .clp = clp,
 928                         };
 929                         struct rpc_add_xprt_test rpcdata = {
 930                                 .add_xprt_test = clp->cl_mvops->session_trunk,
 931                                 .data = &xprtdata,
 932                         };
 933
 934                         if (da->da_transport != clp->cl_proto)
 935                                 continue;
 936                         if (da->da_addr.ss_family != clp->cl_addr.ss_family)
 937                                 continue;
 938                         /**
 939                         * Test this address for session trunking and
 940                         * add as an alias
 941                         */
 942                         xprtdata.cred = nfs4_get_clid_cred(clp),
 943                         rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 944                                           rpc_clnt_setup_test_and_add_xprt,
 945                                           &rpcdata);
 946                         if (xprtdata.cred)
 947                                 put_cred(xprtdata.cred);
 948                 } else {
 949                         clp = nfs4_set_ds_client(mds_srv,
 950                                                 (struct sockaddr *)&da->da_addr,
 951                                                 da->da_addrlen,
 952                                                 da->da_transport, timeo,
 953                                                 retrans, minor_version);
 954                         if (IS_ERR(clp))
 955                                 continue;
 956
 957                         status = nfs4_init_ds_session(clp,
 958                                         mds_srv->nfs_client->cl_lease_time);
 959                         if (status) {
 960                                 nfs_put_client(clp);
 961                                 clp = ERR_PTR(-EIO);
 962                                 continue;
 963                         }
 964
 965                 }
 966         }
 967
 968         if (IS_ERR(clp)) {
 969                 status = PTR_ERR(clp);
 970                 goto out;
 971         }
 972
 973         smp_wmb();
 974         WRITE_ONCE(ds->ds_clp, clp);
 975         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 976 out:
 977         return status;
 978 }
 979
 980 /*
 981  * Create an rpc connection to the nfs4_pnfs_ds data server.
 982  * Currently only supports IPv4 and IPv6 addresses.
 983  * If connection fails, make devid unavailable and return a -errno.
 984  */
 985 int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
 986                           struct nfs4_deviceid_node *devid, unsigned int timeo,
 987                           unsigned int retrans, u32 version, u32 minor_version)
 988 {
 989         int err;
 990
 991         do {
 992                 err = nfs4_wait_ds_connect(ds);
 993                 if (err || ds->ds_clp)
 994                         goto out;
 995                 if (nfs4_test_deviceid_unavailable(devid))
 996                         return -ENODEV;
 997         } while (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) != 0);
 998
 999         if (ds->ds_clp)
1000                 goto connect_done;
1001
1002         switch (version) {
1003         case 3:
1004                 err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo, retrans);
1005                 break;
1006         case 4:
1007                 err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo, retrans,
1008                                                minor_version);
1009                 break;
1010         default:
1011                 dprintk("%s: unsupported DS version %d\n", __func__, version);
1012                 err = -EPROTONOSUPPORT;
1013         }
1014
1015 connect_done:
1016         nfs4_clear_ds_conn_bit(ds);
1017 out:
1018         /*
1019          * At this point the ds->ds_clp should be ready, but it might have
1020          * hit an error.
1021          */
1022         if (!err) {
1023                 if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) {
1024                         WARN_ON_ONCE(ds->ds_clp ||
1025                                 !nfs4_test_deviceid_unavailable(devid));
1026                         return -EINVAL;
1027                 }
1028                 err = nfs_client_init_status(ds->ds_clp);
1029         }
1030
1031         return err;
1032 }
1033 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
1034
1035 /*
1036  * Currently only supports ipv4, ipv6 and one multi-path address.
1037  */
1038 struct nfs4_pnfs_ds_addr *
1039 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
1040 {
1041         struct nfs4_pnfs_ds_addr *da = NULL;
1042         char *buf, *portstr;
1043         __be16 port;
1044         ssize_t nlen, rlen;
1045         int tmp[2];
1046         char *netid;
1047         size_t len;
1048         char *startsep = "";
1049         char *endsep = "";
1050
1051
1052         /* r_netid */
1053         nlen = xdr_stream_decode_string_dup(xdr, &netid, XDR_MAX_NETOBJ,
1054                                             gfp_flags);
1055         if (unlikely(nlen < 0))
1056                 goto out_err;
1057
1058         /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
1059         /* port is ".ABC.DEF", 8 chars max */
1060         rlen = xdr_stream_decode_string_dup(xdr, &buf, INET6_ADDRSTRLEN +
1061                                             IPV6_SCOPE_ID_LEN + 8, gfp_flags);
1062         if (unlikely(rlen < 0))
1063                 goto out_free_netid;
1064
1065         /* replace port '.' with '-' */
1066         portstr = strrchr(buf, '.');
1067         if (!portstr) {
1068                 dprintk("%s: Failed finding expected dot in port\n",
1069                         __func__);
1070                 goto out_free_buf;
1071         }
1072         *portstr = '-';
1073
1074         /* find '.' between address and port */
1075         portstr = strrchr(buf, '.');
1076         if (!portstr) {
1077                 dprintk("%s: Failed finding expected dot between address and "
1078                         "port\n", __func__);
1079                 goto out_free_buf;
1080         }
1081         *portstr = '\0';
1082
1083         da = nfs4_pnfs_ds_addr_alloc(gfp_flags);
1084         if (unlikely(!da))
1085                 goto out_free_buf;
1086
1087         if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
1088                       sizeof(da->da_addr))) {
1089                 dprintk("%s: error parsing address %s\n", __func__, buf);
1090                 goto out_free_da;
1091         }
1092
1093         portstr++;
1094         sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
1095         port = htons((tmp[0] << 8) | (tmp[1]));
1096
1097         switch (da->da_addr.ss_family) {
1098         case AF_INET:
1099                 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
1100                 da->da_addrlen = sizeof(struct sockaddr_in);
1101                 break;
1102
1103         case AF_INET6:
1104                 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
1105                 da->da_addrlen = sizeof(struct sockaddr_in6);
1106                 startsep = "[";
1107                 endsep = "]";
1108                 break;
1109
1110         default:
1111                 dprintk("%s: unsupported address family: %u\n",
1112                         __func__, da->da_addr.ss_family);
1113                 goto out_free_da;
1114         }
1115
1116         da->da_transport = xprt_find_transport_ident(netid);
1117         if (da->da_transport < 0) {
1118                 dprintk("%s: ERROR: unknown r_netid \"%s\"\n",
1119                         __func__, netid);
1120                 goto out_free_da;
1121         }
1122
1123         da->da_netid = netid;
1124
1125         /* save human readable address */
1126         len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
1127         da->da_remotestr = kzalloc(len, gfp_flags);
1128
1129         /* NULL is ok, only used for dprintk */
1130         if (da->da_remotestr)
1131                 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
1132                          buf, endsep, ntohs(port));
1133
1134         dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
1135         kfree(buf);
1136         return da;
1137
1138 out_free_da:
1139         kfree(da);
1140 out_free_buf:
1141         dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
1142         kfree(buf);
1143 out_free_netid:
1144         kfree(netid);
1145 out_err:
1146         return NULL;
1147 }
1148 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
1149
1150 void
1151 pnfs_layout_mark_request_commit(struct nfs_page *req,
1152                                 struct pnfs_layout_segment *lseg,
1153                                 struct nfs_commit_info *cinfo,
1154                                 u32 ds_commit_idx)
1155 {
1156         struct list_head *list;
1157         struct pnfs_commit_array *array;
1158         struct pnfs_commit_bucket *bucket;
1159
1160         mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
1161         array = pnfs_lookup_commit_array(cinfo->ds, lseg);
1162         if (!array || !pnfs_is_valid_lseg(lseg))
1163                 goto out_resched;
1164         bucket = &array->buckets[ds_commit_idx];
1165         list = &bucket->written;
1166         /* Non-empty buckets hold a reference on the lseg.  That ref
1167          * is normally transferred to the COMMIT call and released
1168          * there.  It could also be released if the last req is pulled
1169          * off due to a rewrite, in which case it will be done in
1170          * pnfs_common_clear_request_commit
1171          */
1172         if (!bucket->lseg)
1173                 bucket->lseg = pnfs_get_lseg(lseg);
1174         set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1175         cinfo->ds->nwritten++;
1176
1177         nfs_request_add_commit_list_locked(req, list, cinfo);
1178         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1179         nfs_mark_page_unstable(req->wb_page, cinfo);
1180         return;
1181 out_resched:
1182         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1183         cinfo->completion_ops->resched_write(cinfo, req);
1184 }
1185 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
1186
1187 int
1188 pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
1189 {
1190         int ret;
1191
1192         if (!pnfs_layoutcommit_outstanding(inode))
1193                 return 0;
1194         ret = nfs_commit_inode(inode, FLUSH_SYNC);
1195         if (ret < 0)
1196                 return ret;
1197         if (datasync)
1198                 return 0;
1199         return pnfs_layoutcommit_inode(inode, true);
1200 }
1201 EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
1202