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 <[email protected]>
   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 pnfs_layout_segment *
 355 pnfs_bucket_get_committing(struct list_head *head,
 356                            struct pnfs_commit_bucket *bucket,
 357                            struct nfs_commit_info *cinfo)
 358 {
 359         struct pnfs_layout_segment *lseg;
 360         struct list_head *pos;
 361
 362         list_for_each(pos, &bucket->committing)
 363                 cinfo->ds->ncommitting--;
 364         list_splice_init(&bucket->committing, head);
 365         lseg = pnfs_free_bucket_lseg(bucket);
 366         if (!lseg)
 367                 lseg = pnfs_get_lseg(bucket->lseg);
 368         return lseg;
 369 }
 370
 371 static struct nfs_commit_data *
 372 pnfs_bucket_fetch_commitdata(struct pnfs_commit_bucket *bucket,
 373                              struct nfs_commit_info *cinfo)
 374 {
 375         struct nfs_commit_data *data = nfs_commitdata_alloc();
 376
 377         if (!data)
 378                 return NULL;
 379         data->lseg = pnfs_bucket_get_committing(&data->pages, bucket, cinfo);
 380         return data;
 381 }
 382
 383 static void pnfs_generic_retry_commit(struct pnfs_commit_bucket *buckets,
 384                                       unsigned int nbuckets,
 385                                       struct nfs_commit_info *cinfo,
 386                                       unsigned int idx)
 387 {
 388         struct pnfs_commit_bucket *bucket;
 389         struct pnfs_layout_segment *freeme;
 390         LIST_HEAD(pages);
 391
 392         for (bucket = buckets; idx < nbuckets; bucket++, idx++) {
 393                 if (list_empty(&bucket->committing))
 394                         continue;
 395                 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 396                 freeme = pnfs_bucket_get_committing(&pages, bucket, cinfo);
 397                 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 398                 nfs_retry_commit(&pages, freeme, cinfo, idx);
 399                 pnfs_put_lseg(freeme);
 400         }
 401 }
 402
 403 static unsigned int
 404 pnfs_bucket_alloc_ds_commits(struct list_head *list,
 405                              struct pnfs_commit_bucket *buckets,
 406                              unsigned int nbuckets,
 407                              struct nfs_commit_info *cinfo)
 408 {
 409         struct pnfs_commit_bucket *bucket;
 410         struct nfs_commit_data *data;
 411         unsigned int i;
 412         unsigned int nreq = 0;
 413
 414         for (i = 0, bucket = buckets; i < nbuckets; i++, bucket++) {
 415                 if (list_empty(&bucket->committing))
 416                         continue;
 417                 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 418                 if (!list_empty(&bucket->committing)) {
 419                         data = pnfs_bucket_fetch_commitdata(bucket, cinfo);
 420                         if (!data)
 421                                 goto out_error;
 422                         data->ds_commit_index = i;
 423                         list_add_tail(&data->list, list);
 424                         nreq++;
 425                 }
 426                 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 427         }
 428         return nreq;
 429 out_error:
 430         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 431         /* Clean up on error */
 432         pnfs_generic_retry_commit(buckets, nbuckets, cinfo, i);
 433         return nreq;
 434 }
 435
 436 static unsigned int
 437 pnfs_alloc_ds_commits_list(struct list_head *list,
 438                            struct pnfs_ds_commit_info *fl_cinfo,
 439                            struct nfs_commit_info *cinfo)
 440 {
 441         struct pnfs_commit_array *array;
 442         unsigned int ret = 0;
 443
 444         rcu_read_lock();
 445         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 446                 if (!array->lseg || !pnfs_get_commit_array(array))
 447                         continue;
 448                 rcu_read_unlock();
 449                 ret += pnfs_bucket_alloc_ds_commits(list, array->buckets,
 450                                 array->nbuckets, cinfo);
 451                 rcu_read_lock();
 452                 pnfs_put_commit_array(array, cinfo->inode);
 453         }
 454         rcu_read_unlock();
 455         return ret;
 456 }
 457
 458 /* This follows nfs_commit_list pretty closely */
 459 int
 460 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
 461                              int how, struct nfs_commit_info *cinfo,
 462                              int (*initiate_commit)(struct nfs_commit_data *data,
 463                                                     int how))
 464 {
 465         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 466         struct nfs_commit_data *data, *tmp;
 467         LIST_HEAD(list);
 468         unsigned int nreq = 0;
 469
 470         if (!list_empty(mds_pages)) {
 471                 data = nfs_commitdata_alloc();
 472                 if (!data) {
 473                         nfs_retry_commit(mds_pages, NULL, cinfo, -1);
 474                         return -ENOMEM;
 475                 }
 476                 data->ds_commit_index = -1;
 477                 list_splice_init(mds_pages, &data->pages);
 478                 list_add_tail(&data->list, &list);
 479                 nreq++;
 480         }
 481
 482         nreq += pnfs_alloc_ds_commits_list(&list, fl_cinfo, cinfo);
 483         if (nreq == 0)
 484                 goto out;
 485
 486         list_for_each_entry_safe(data, tmp, &list, list) {
 487                 list_del(&data->list);
 488                 if (data->ds_commit_index < 0) {
 489                         nfs_init_commit(data, NULL, NULL, cinfo);
 490                         nfs_initiate_commit(NFS_CLIENT(inode), data,
 491                                             NFS_PROTO(data->inode),
 492                                             data->mds_ops, how,
 493                                             RPC_TASK_CRED_NOREF);
 494                 } else {
 495                         nfs_init_commit(data, NULL, data->lseg, cinfo);
 496                         initiate_commit(data, how);
 497                 }
 498         }
 499 out:
 500         return PNFS_ATTEMPTED;
 501 }
 502 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
 503
 504 /*
 505  * Data server cache
 506  *
 507  * Data servers can be mapped to different device ids.
 508  * nfs4_pnfs_ds reference counting
 509  *   - set to 1 on allocation
 510  *   - incremented when a device id maps a data server already in the cache.
 511  *   - decremented when deviceid is removed from the cache.
 512  */
 513 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
 514 static LIST_HEAD(nfs4_data_server_cache);
 515
 516 /* Debug routines */
 517 static void
 518 print_ds(struct nfs4_pnfs_ds *ds)
 519 {
 520         if (ds == NULL) {
 521                 printk(KERN_WARNING "%s NULL device\n", __func__);
 522                 return;
 523         }
 524         printk(KERN_WARNING "        ds %s\n"
 525                 "        ref count %d\n"
 526                 "        client %p\n"
 527                 "        cl_exchange_flags %x\n",
 528                 ds->ds_remotestr,
 529                 refcount_read(&ds->ds_count), ds->ds_clp,
 530                 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
 531 }
 532
 533 static bool
 534 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
 535 {
 536         struct sockaddr_in *a, *b;
 537         struct sockaddr_in6 *a6, *b6;
 538
 539         if (addr1->sa_family != addr2->sa_family)
 540                 return false;
 541
 542         switch (addr1->sa_family) {
 543         case AF_INET:
 544                 a = (struct sockaddr_in *)addr1;
 545                 b = (struct sockaddr_in *)addr2;
 546
 547                 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
 548                     a->sin_port == b->sin_port)
 549                         return true;
 550                 break;
 551
 552         case AF_INET6:
 553                 a6 = (struct sockaddr_in6 *)addr1;
 554                 b6 = (struct sockaddr_in6 *)addr2;
 555
 556                 /* LINKLOCAL addresses must have matching scope_id */
 557                 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
 558                     IPV6_ADDR_SCOPE_LINKLOCAL &&
 559                     a6->sin6_scope_id != b6->sin6_scope_id)
 560                         return false;
 561
 562                 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
 563                     a6->sin6_port == b6->sin6_port)
 564                         return true;
 565                 break;
 566
 567         default:
 568                 dprintk("%s: unhandled address family: %u\n",
 569                         __func__, addr1->sa_family);
 570                 return false;
 571         }
 572
 573         return false;
 574 }
 575
 576 /*
 577  * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
 578  * declare a match.
 579  */
 580 static bool
 581 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 582                                const struct list_head *dsaddrs2)
 583 {
 584         struct nfs4_pnfs_ds_addr *da1, *da2;
 585         struct sockaddr *sa1, *sa2;
 586         bool match = false;
 587
 588         list_for_each_entry(da1, dsaddrs1, da_node) {
 589                 sa1 = (struct sockaddr *)&da1->da_addr;
 590                 match = false;
 591                 list_for_each_entry(da2, dsaddrs2, da_node) {
 592                         sa2 = (struct sockaddr *)&da2->da_addr;
 593                         match = same_sockaddr(sa1, sa2);
 594                         if (match)
 595                                 break;
 596                 }
 597                 if (!match)
 598                         break;
 599         }
 600         return match;
 601 }
 602
 603 /*
 604  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 605  */
 606 static struct nfs4_pnfs_ds *
 607 _data_server_lookup_locked(const struct list_head *dsaddrs)
 608 {
 609         struct nfs4_pnfs_ds *ds;
 610
 611         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 612                 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 613                         return ds;
 614         return NULL;
 615 }
 616
 617 static struct nfs4_pnfs_ds_addr *nfs4_pnfs_ds_addr_alloc(gfp_t gfp_flags)
 618 {
 619         struct nfs4_pnfs_ds_addr *da = kzalloc(sizeof(*da), gfp_flags);
 620         if (da)
 621                 INIT_LIST_HEAD(&da->da_node);
 622         return da;
 623 }
 624
 625 static void nfs4_pnfs_ds_addr_free(struct nfs4_pnfs_ds_addr *da)
 626 {
 627         kfree(da->da_remotestr);
 628         kfree(da->da_netid);
 629         kfree(da);
 630 }
 631
 632 static void destroy_ds(struct nfs4_pnfs_ds *ds)
 633 {
 634         struct nfs4_pnfs_ds_addr *da;
 635
 636         dprintk("--> %s\n", __func__);
 637         ifdebug(FACILITY)
 638                 print_ds(ds);
 639
 640         nfs_put_client(ds->ds_clp);
 641
 642         while (!list_empty(&ds->ds_addrs)) {
 643                 da = list_first_entry(&ds->ds_addrs,
 644                                       struct nfs4_pnfs_ds_addr,
 645                                       da_node);
 646                 list_del_init(&da->da_node);
 647                 nfs4_pnfs_ds_addr_free(da);
 648         }
 649
 650         kfree(ds->ds_remotestr);
 651         kfree(ds);
 652 }
 653
 654 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
 655 {
 656         if (refcount_dec_and_lock(&ds->ds_count,
 657                                 &nfs4_ds_cache_lock)) {
 658                 list_del_init(&ds->ds_node);
 659                 spin_unlock(&nfs4_ds_cache_lock);
 660                 destroy_ds(ds);
 661         }
 662 }
 663 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
 664
 665 /*
 666  * Create a string with a human readable address and port to avoid
 667  * complicated setup around many dprinks.
 668  */
 669 static char *
 670 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 671 {
 672         struct nfs4_pnfs_ds_addr *da;
 673         char *remotestr;
 674         size_t len;
 675         char *p;
 676
 677         len = 3;        /* '{', '}' and eol */
 678         list_for_each_entry(da, dsaddrs, da_node) {
 679                 len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 680         }
 681
 682         remotestr = kzalloc(len, gfp_flags);
 683         if (!remotestr)
 684                 return NULL;
 685
 686         p = remotestr;
 687         *(p++) = '{';
 688         len--;
 689         list_for_each_entry(da, dsaddrs, da_node) {
 690                 size_t ll = strlen(da->da_remotestr);
 691
 692                 if (ll > len)
 693                         goto out_err;
 694
 695                 memcpy(p, da->da_remotestr, ll);
 696                 p += ll;
 697                 len -= ll;
 698
 699                 if (len < 1)
 700                         goto out_err;
 701                 (*p++) = ',';
 702                 len--;
 703         }
 704         if (len < 2)
 705                 goto out_err;
 706         *(p++) = '}';
 707         *p = '\0';
 708         return remotestr;
 709 out_err:
 710         kfree(remotestr);
 711         return NULL;
 712 }
 713
 714 /*
 715  * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
 716  * uncached and return cached struct nfs4_pnfs_ds.
 717  */
 718 struct nfs4_pnfs_ds *
 719 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
 720 {
 721         struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 722         char *remotestr;
 723
 724         if (list_empty(dsaddrs)) {
 725                 dprintk("%s: no addresses defined\n", __func__);
 726                 goto out;
 727         }
 728
 729         ds = kzalloc(sizeof(*ds), gfp_flags);
 730         if (!ds)
 731                 goto out;
 732
 733         /* this is only used for debugging, so it's ok if its NULL */
 734         remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 735
 736         spin_lock(&nfs4_ds_cache_lock);
 737         tmp_ds = _data_server_lookup_locked(dsaddrs);
 738         if (tmp_ds == NULL) {
 739                 INIT_LIST_HEAD(&ds->ds_addrs);
 740                 list_splice_init(dsaddrs, &ds->ds_addrs);
 741                 ds->ds_remotestr = remotestr;
 742                 refcount_set(&ds->ds_count, 1);
 743                 INIT_LIST_HEAD(&ds->ds_node);
 744                 ds->ds_clp = NULL;
 745                 list_add(&ds->ds_node, &nfs4_data_server_cache);
 746                 dprintk("%s add new data server %s\n", __func__,
 747                         ds->ds_remotestr);
 748         } else {
 749                 kfree(remotestr);
 750                 kfree(ds);
 751                 refcount_inc(&tmp_ds->ds_count);
 752                 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 753                         __func__, tmp_ds->ds_remotestr,
 754                         refcount_read(&tmp_ds->ds_count));
 755                 ds = tmp_ds;
 756         }
 757         spin_unlock(&nfs4_ds_cache_lock);
 758 out:
 759         return ds;
 760 }
 761 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
 762
 763 static int nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
 764 {
 765         might_sleep();
 766         return wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING, TASK_KILLABLE);
 767 }
 768
 769 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
 770 {
 771         smp_mb__before_atomic();
 772         clear_and_wake_up_bit(NFS4DS_CONNECTING, &ds->ds_state);
 773 }
 774
 775 static struct nfs_client *(*get_v3_ds_connect)(
 776                         struct nfs_server *mds_srv,
 777                         const struct sockaddr_storage *ds_addr,
 778                         int ds_addrlen,
 779                         int ds_proto,
 780                         unsigned int ds_timeo,
 781                         unsigned int ds_retrans);
 782
 783 static bool load_v3_ds_connect(void)
 784 {
 785         if (!get_v3_ds_connect) {
 786                 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
 787                 WARN_ON_ONCE(!get_v3_ds_connect);
 788         }
 789
 790         return(get_v3_ds_connect != NULL);
 791 }
 792
 793 void nfs4_pnfs_v3_ds_connect_unload(void)
 794 {
 795         if (get_v3_ds_connect) {
 796                 symbol_put(nfs3_set_ds_client);
 797                 get_v3_ds_connect = NULL;
 798         }
 799 }
 800
 801 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
 802                                  struct nfs4_pnfs_ds *ds,
 803                                  unsigned int timeo,
 804                                  unsigned int retrans)
 805 {
 806         struct nfs_client *clp = ERR_PTR(-EIO);
 807         struct nfs4_pnfs_ds_addr *da;
 808         unsigned long connect_timeout = timeo * (retrans + 1) * HZ / 10;
 809         int status = 0;
 810
 811         dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 812
 813         if (!load_v3_ds_connect())
 814                 return -EPROTONOSUPPORT;
 815
 816         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 817                 dprintk("%s: DS %s: trying address %s\n",
 818                         __func__, ds->ds_remotestr, da->da_remotestr);
 819
 820                 if (!IS_ERR(clp)) {
 821                         struct xprt_create xprt_args = {
 822                                 .ident = da->da_transport,
 823                                 .net = clp->cl_net,
 824                                 .dstaddr = (struct sockaddr *)&da->da_addr,
 825                                 .addrlen = da->da_addrlen,
 826                                 .servername = clp->cl_hostname,
 827                                 .connect_timeout = connect_timeout,
 828                                 .reconnect_timeout = connect_timeout,
 829                         };
 830
 831                         if (da->da_transport != clp->cl_proto)
 832                                 continue;
 833                         if (da->da_addr.ss_family != clp->cl_addr.ss_family)
 834                                 continue;
 835                         /* Add this address as an alias */
 836                         rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 837                                         rpc_clnt_test_and_add_xprt, NULL);
 838                         continue;
 839                 }
 840                 clp = get_v3_ds_connect(mds_srv,
 841                                 &da->da_addr,
 842                                 da->da_addrlen, da->da_transport,
 843                                 timeo, retrans);
 844                 if (IS_ERR(clp))
 845                         continue;
 846                 clp->cl_rpcclient->cl_softerr = 0;
 847                 clp->cl_rpcclient->cl_softrtry = 0;
 848         }
 849
 850         if (IS_ERR(clp)) {
 851                 status = PTR_ERR(clp);
 852                 goto out;
 853         }
 854
 855         smp_wmb();
 856         WRITE_ONCE(ds->ds_clp, clp);
 857         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 858 out:
 859         return status;
 860 }
 861
 862 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
 863                                  struct nfs4_pnfs_ds *ds,
 864                                  unsigned int timeo,
 865                                  unsigned int retrans,
 866                                  u32 minor_version)
 867 {
 868         struct nfs_client *clp = ERR_PTR(-EIO);
 869         struct nfs4_pnfs_ds_addr *da;
 870         int status = 0;
 871
 872         dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 873
 874         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 875                 char servername[48];
 876
 877                 dprintk("%s: DS %s: trying address %s\n",
 878                         __func__, ds->ds_remotestr, da->da_remotestr);
 879
 880                 if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) {
 881                         struct xprt_create xprt_args = {
 882                                 .ident = da->da_transport,
 883                                 .net = clp->cl_net,
 884                                 .dstaddr = (struct sockaddr *)&da->da_addr,
 885                                 .addrlen = da->da_addrlen,
 886                                 .servername = clp->cl_hostname,
 887                                 .xprtsec = clp->cl_xprtsec,
 888                         };
 889                         struct nfs4_add_xprt_data xprtdata = {
 890                                 .clp = clp,
 891                         };
 892                         struct rpc_add_xprt_test rpcdata = {
 893                                 .add_xprt_test = clp->cl_mvops->session_trunk,
 894                                 .data = &xprtdata,
 895                         };
 896
 897                         if (da->da_transport != clp->cl_proto &&
 898                                         clp->cl_proto != XPRT_TRANSPORT_TCP_TLS)
 899                                 continue;
 900                         if (da->da_transport == XPRT_TRANSPORT_TCP &&
 901                                 mds_srv->nfs_client->cl_proto ==
 902                                         XPRT_TRANSPORT_TCP_TLS) {
 903                                 struct sockaddr *addr =
 904                                         (struct sockaddr *)&da->da_addr;
 905                                 struct sockaddr_in *sin =
 906                                         (struct sockaddr_in *)&da->da_addr;
 907                                 struct sockaddr_in6 *sin6 =
 908                                         (struct sockaddr_in6 *)&da->da_addr;
 909
 910                                 /* for NFS with TLS we need to supply a correct
 911                                  * servername of the trunked transport, not the
 912                                  * servername of the main transport stored in
 913                                  * clp->cl_hostname. And set the protocol to
 914                                  * indicate to use TLS
 915                                  */
 916                                 servername[0] = '\0';
 917                                 switch(addr->sa_family) {
 918                                 case AF_INET:
 919                                         snprintf(servername, sizeof(servername),
 920                                                 "%pI4", &sin->sin_addr.s_addr);
 921                                         break;
 922                                 case AF_INET6:
 923                                         snprintf(servername, sizeof(servername),
 924                                                 "%pI6", &sin6->sin6_addr);
 925                                         break;
 926                                 default:
 927                                         /* do not consider this address */
 928                                         continue;
 929                                 }
 930                                 xprt_args.ident = XPRT_TRANSPORT_TCP_TLS;
 931                                 xprt_args.servername = servername;
 932                         }
 933                         if (da->da_addr.ss_family != clp->cl_addr.ss_family)
 934                                 continue;
 935
 936                         /**
 937                         * Test this address for session trunking and
 938                         * add as an alias
 939                         */
 940                         xprtdata.cred = nfs4_get_clid_cred(clp);
 941                         rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 942                                           rpc_clnt_setup_test_and_add_xprt,
 943                                           &rpcdata);
 944                         if (xprtdata.cred)
 945                                 put_cred(xprtdata.cred);
 946                 } else {
 947                         if (da->da_transport == XPRT_TRANSPORT_TCP &&
 948                                 mds_srv->nfs_client->cl_proto ==
 949                                         XPRT_TRANSPORT_TCP_TLS)
 950                                 da->da_transport = XPRT_TRANSPORT_TCP_TLS;
 951                         clp = nfs4_set_ds_client(mds_srv,
 952                                                 &da->da_addr,
 953                                                 da->da_addrlen,
 954                                                 da->da_transport, timeo,
 955                                                 retrans, minor_version);
 956                         if (IS_ERR(clp))
 957                                 continue;
 958
 959                         status = nfs4_init_ds_session(clp,
 960                                         mds_srv->nfs_client->cl_lease_time);
 961                         if (status) {
 962                                 nfs_put_client(clp);
 963                                 clp = ERR_PTR(-EIO);
 964                                 continue;
 965                         }
 966
 967                 }
 968         }
 969
 970         if (IS_ERR(clp)) {
 971                 status = PTR_ERR(clp);
 972                 goto out;
 973         }
 974
 975         smp_wmb();
 976         WRITE_ONCE(ds->ds_clp, clp);
 977         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 978 out:
 979         return status;
 980 }
 981
 982 /*
 983  * Create an rpc connection to the nfs4_pnfs_ds data server.
 984  * Currently only supports IPv4 and IPv6 addresses.
 985  * If connection fails, make devid unavailable and return a -errno.
 986  */
 987 int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
 988                           struct nfs4_deviceid_node *devid, unsigned int timeo,
 989                           unsigned int retrans, u32 version, u32 minor_version)
 990 {
 991         int err;
 992
 993         do {
 994                 err = nfs4_wait_ds_connect(ds);
 995                 if (err || ds->ds_clp)
 996                         goto out;
 997                 if (nfs4_test_deviceid_unavailable(devid))
 998                         return -ENODEV;
 999         } while (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) != 0);
1000
1001         if (ds->ds_clp)
1002                 goto connect_done;
1003
1004         switch (version) {
1005         case 3:
1006                 err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo, retrans);
1007                 break;
1008         case 4:
1009                 err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo, retrans,
1010                                                minor_version);
1011                 break;
1012         default:
1013                 dprintk("%s: unsupported DS version %d\n", __func__, version);
1014                 err = -EPROTONOSUPPORT;
1015         }
1016
1017 connect_done:
1018         nfs4_clear_ds_conn_bit(ds);
1019 out:
1020         /*
1021          * At this point the ds->ds_clp should be ready, but it might have
1022          * hit an error.
1023          */
1024         if (!err) {
1025                 if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) {
1026                         WARN_ON_ONCE(ds->ds_clp ||
1027                                 !nfs4_test_deviceid_unavailable(devid));
1028                         return -EINVAL;
1029                 }
1030                 err = nfs_client_init_status(ds->ds_clp);
1031         }
1032
1033         return err;
1034 }
1035 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
1036
1037 /*
1038  * Currently only supports ipv4, ipv6 and one multi-path address.
1039  */
1040 struct nfs4_pnfs_ds_addr *
1041 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
1042 {
1043         struct nfs4_pnfs_ds_addr *da = NULL;
1044         char *buf, *portstr;
1045         __be16 port;
1046         ssize_t nlen, rlen;
1047         int tmp[2];
1048         char *netid;
1049         size_t len;
1050         char *startsep = "";
1051         char *endsep = "";
1052
1053
1054         /* r_netid */
1055         nlen = xdr_stream_decode_string_dup(xdr, &netid, XDR_MAX_NETOBJ,
1056                                             gfp_flags);
1057         if (unlikely(nlen < 0))
1058                 goto out_err;
1059
1060         /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
1061         /* port is ".ABC.DEF", 8 chars max */
1062         rlen = xdr_stream_decode_string_dup(xdr, &buf, INET6_ADDRSTRLEN +
1063                                             IPV6_SCOPE_ID_LEN + 8, gfp_flags);
1064         if (unlikely(rlen < 0))
1065                 goto out_free_netid;
1066
1067         /* replace port '.' with '-' */
1068         portstr = strrchr(buf, '.');
1069         if (!portstr) {
1070                 dprintk("%s: Failed finding expected dot in port\n",
1071                         __func__);
1072                 goto out_free_buf;
1073         }
1074         *portstr = '-';
1075
1076         /* find '.' between address and port */
1077         portstr = strrchr(buf, '.');
1078         if (!portstr) {
1079                 dprintk("%s: Failed finding expected dot between address and "
1080                         "port\n", __func__);
1081                 goto out_free_buf;
1082         }
1083         *portstr = '\0';
1084
1085         da = nfs4_pnfs_ds_addr_alloc(gfp_flags);
1086         if (unlikely(!da))
1087                 goto out_free_buf;
1088
1089         if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
1090                       sizeof(da->da_addr))) {
1091                 dprintk("%s: error parsing address %s\n", __func__, buf);
1092                 goto out_free_da;
1093         }
1094
1095         portstr++;
1096         sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
1097         port = htons((tmp[0] << 8) | (tmp[1]));
1098
1099         switch (da->da_addr.ss_family) {
1100         case AF_INET:
1101                 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
1102                 da->da_addrlen = sizeof(struct sockaddr_in);
1103                 break;
1104
1105         case AF_INET6:
1106                 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
1107                 da->da_addrlen = sizeof(struct sockaddr_in6);
1108                 startsep = "[";
1109                 endsep = "]";
1110                 break;
1111
1112         default:
1113                 dprintk("%s: unsupported address family: %u\n",
1114                         __func__, da->da_addr.ss_family);
1115                 goto out_free_da;
1116         }
1117
1118         da->da_transport = xprt_find_transport_ident(netid);
1119         if (da->da_transport < 0) {
1120                 dprintk("%s: ERROR: unknown r_netid \"%s\"\n",
1121                         __func__, netid);
1122                 goto out_free_da;
1123         }
1124
1125         da->da_netid = netid;
1126
1127         /* save human readable address */
1128         len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
1129         da->da_remotestr = kzalloc(len, gfp_flags);
1130
1131         /* NULL is ok, only used for dprintk */
1132         if (da->da_remotestr)
1133                 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
1134                          buf, endsep, ntohs(port));
1135
1136         dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
1137         kfree(buf);
1138         return da;
1139
1140 out_free_da:
1141         kfree(da);
1142 out_free_buf:
1143         dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
1144         kfree(buf);
1145 out_free_netid:
1146         kfree(netid);
1147 out_err:
1148         return NULL;
1149 }
1150 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
1151
1152 void
1153 pnfs_layout_mark_request_commit(struct nfs_page *req,
1154                                 struct pnfs_layout_segment *lseg,
1155                                 struct nfs_commit_info *cinfo,
1156                                 u32 ds_commit_idx)
1157 {
1158         struct list_head *list;
1159         struct pnfs_commit_array *array;
1160         struct pnfs_commit_bucket *bucket;
1161
1162         mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
1163         array = pnfs_lookup_commit_array(cinfo->ds, lseg);
1164         if (!array || !pnfs_is_valid_lseg(lseg))
1165                 goto out_resched;
1166         bucket = &array->buckets[ds_commit_idx];
1167         list = &bucket->written;
1168         /* Non-empty buckets hold a reference on the lseg.  That ref
1169          * is normally transferred to the COMMIT call and released
1170          * there.  It could also be released if the last req is pulled
1171          * off due to a rewrite, in which case it will be done in
1172          * pnfs_common_clear_request_commit
1173          */
1174         if (!bucket->lseg)
1175                 bucket->lseg = pnfs_get_lseg(lseg);
1176         set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1177         cinfo->ds->nwritten++;
1178
1179         nfs_request_add_commit_list_locked(req, list, cinfo);
1180         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1181         nfs_folio_mark_unstable(nfs_page_to_folio(req), cinfo);
1182         return;
1183 out_resched:
1184         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1185         cinfo->completion_ops->resched_write(cinfo, req);
1186 }
1187 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
1188
1189 int
1190 pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
1191 {
1192         int ret;
1193
1194         if (!pnfs_layoutcommit_outstanding(inode))
1195                 return 0;
1196         ret = nfs_commit_inode(inode, FLUSH_SYNC);
1197         if (ret < 0)
1198                 return ret;
1199         if (datasync)
1200                 return 0;
1201         return pnfs_layoutcommit_inode(inode, true);
1202 }
1203 EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
1204