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_layout_segment *freeme = 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                 struct pnfs_commit_bucket *bucket;
  88
  89                 bucket = list_first_entry(&req->wb_list,
  90                                           struct pnfs_commit_bucket,
  91                                           written);
  92                 freeme = pnfs_free_bucket_lseg(bucket);
  93         }
  94 out:
  95         nfs_request_remove_commit_list(req, cinfo);
  96         pnfs_put_lseg(freeme);
  97 }
  98 EXPORT_SYMBOL_GPL(pnfs_generic_clear_request_commit);
  99
 100 struct pnfs_commit_array *
 101 pnfs_alloc_commit_array(size_t n, gfp_t gfp_flags)
 102 {
 103         struct pnfs_commit_array *p;
 104         struct pnfs_commit_bucket *b;
 105
 106         p = kmalloc(struct_size(p, buckets, n), gfp_flags);
 107         if (!p)
 108                 return NULL;
 109         p->nbuckets = n;
 110         INIT_LIST_HEAD(&p->cinfo_list);
 111         INIT_LIST_HEAD(&p->lseg_list);
 112         p->lseg = NULL;
 113         for (b = &p->buckets[0]; n != 0; b++, n--) {
 114                 INIT_LIST_HEAD(&b->written);
 115                 INIT_LIST_HEAD(&b->committing);
 116                 b->lseg = NULL;
 117                 b->direct_verf.committed = NFS_INVALID_STABLE_HOW;
 118         }
 119         return p;
 120 }
 121 EXPORT_SYMBOL_GPL(pnfs_alloc_commit_array);
 122
 123 void
 124 pnfs_free_commit_array(struct pnfs_commit_array *p)
 125 {
 126         kfree_rcu(p, rcu);
 127 }
 128 EXPORT_SYMBOL_GPL(pnfs_free_commit_array);
 129
 130 static struct pnfs_commit_array *
 131 pnfs_find_commit_array_by_lseg(struct pnfs_ds_commit_info *fl_cinfo,
 132                 struct pnfs_layout_segment *lseg)
 133 {
 134         struct pnfs_commit_array *array;
 135
 136         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 137                 if (array->lseg == lseg)
 138                         return array;
 139         }
 140         return NULL;
 141 }
 142
 143 struct pnfs_commit_array *
 144 pnfs_add_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
 145                 struct pnfs_commit_array *new,
 146                 struct pnfs_layout_segment *lseg)
 147 {
 148         struct pnfs_commit_array *array;
 149
 150         array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 151         if (array)
 152                 return array;
 153         new->lseg = lseg;
 154         refcount_set(&new->refcount, 1);
 155         list_add_rcu(&new->cinfo_list, &fl_cinfo->commits);
 156         list_add(&new->lseg_list, &lseg->pls_commits);
 157         return new;
 158 }
 159 EXPORT_SYMBOL_GPL(pnfs_add_commit_array);
 160
 161 static struct pnfs_commit_array *
 162 pnfs_lookup_commit_array(struct pnfs_ds_commit_info *fl_cinfo,
 163                 struct pnfs_layout_segment *lseg)
 164 {
 165         struct pnfs_commit_array *array;
 166
 167         rcu_read_lock();
 168         array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 169         if (!array) {
 170                 rcu_read_unlock();
 171                 fl_cinfo->ops->setup_ds_info(fl_cinfo, lseg);
 172                 rcu_read_lock();
 173                 array = pnfs_find_commit_array_by_lseg(fl_cinfo, lseg);
 174         }
 175         rcu_read_unlock();
 176         return array;
 177 }
 178
 179 static void
 180 pnfs_release_commit_array_locked(struct pnfs_commit_array *array)
 181 {
 182         list_del_rcu(&array->cinfo_list);
 183         list_del(&array->lseg_list);
 184         pnfs_free_commit_array(array);
 185 }
 186
 187 static void
 188 pnfs_put_commit_array_locked(struct pnfs_commit_array *array)
 189 {
 190         if (refcount_dec_and_test(&array->refcount))
 191                 pnfs_release_commit_array_locked(array);
 192 }
 193
 194 static void
 195 pnfs_put_commit_array(struct pnfs_commit_array *array, struct inode *inode)
 196 {
 197         if (refcount_dec_and_lock(&array->refcount, &inode->i_lock)) {
 198                 pnfs_release_commit_array_locked(array);
 199                 spin_unlock(&inode->i_lock);
 200         }
 201 }
 202
 203 static struct pnfs_commit_array *
 204 pnfs_get_commit_array(struct pnfs_commit_array *array)
 205 {
 206         if (refcount_inc_not_zero(&array->refcount))
 207                 return array;
 208         return NULL;
 209 }
 210
 211 static void
 212 pnfs_remove_and_free_commit_array(struct pnfs_commit_array *array)
 213 {
 214         array->lseg = NULL;
 215         list_del_init(&array->lseg_list);
 216         pnfs_put_commit_array_locked(array);
 217 }
 218
 219 void
 220 pnfs_generic_ds_cinfo_release_lseg(struct pnfs_ds_commit_info *fl_cinfo,
 221                 struct pnfs_layout_segment *lseg)
 222 {
 223         struct pnfs_commit_array *array, *tmp;
 224
 225         list_for_each_entry_safe(array, tmp, &lseg->pls_commits, lseg_list)
 226                 pnfs_remove_and_free_commit_array(array);
 227 }
 228 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_release_lseg);
 229
 230 void
 231 pnfs_generic_ds_cinfo_destroy(struct pnfs_ds_commit_info *fl_cinfo)
 232 {
 233         struct pnfs_commit_array *array, *tmp;
 234
 235         list_for_each_entry_safe(array, tmp, &fl_cinfo->commits, cinfo_list)
 236                 pnfs_remove_and_free_commit_array(array);
 237 }
 238 EXPORT_SYMBOL_GPL(pnfs_generic_ds_cinfo_destroy);
 239
 240 /*
 241  * Locks the nfs_page requests for commit and moves them to
 242  * @bucket->committing.
 243  */
 244 static int
 245 pnfs_bucket_scan_ds_commit_list(struct pnfs_commit_bucket *bucket,
 246                                 struct nfs_commit_info *cinfo,
 247                                 int max)
 248 {
 249         struct list_head *src = &bucket->written;
 250         struct list_head *dst = &bucket->committing;
 251         int ret;
 252
 253         lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 254         ret = nfs_scan_commit_list(src, dst, cinfo, max);
 255         if (ret) {
 256                 cinfo->ds->nwritten -= ret;
 257                 cinfo->ds->ncommitting += ret;
 258         }
 259         return ret;
 260 }
 261
 262 static int pnfs_bucket_scan_array(struct nfs_commit_info *cinfo,
 263                                   struct pnfs_commit_bucket *buckets,
 264                                   unsigned int nbuckets,
 265                                   int max)
 266 {
 267         unsigned int i;
 268         int rv = 0, cnt;
 269
 270         for (i = 0; i < nbuckets && max != 0; i++) {
 271                 cnt = pnfs_bucket_scan_ds_commit_list(&buckets[i], cinfo, max);
 272                 rv += cnt;
 273                 max -= cnt;
 274         }
 275         return rv;
 276 }
 277
 278 /* Move reqs from written to committing lists, returning count
 279  * of number moved.
 280  */
 281 int pnfs_generic_scan_commit_lists(struct nfs_commit_info *cinfo, int max)
 282 {
 283         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 284         struct pnfs_commit_array *array;
 285         int rv = 0, cnt;
 286
 287         rcu_read_lock();
 288         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 289                 if (!array->lseg || !pnfs_get_commit_array(array))
 290                         continue;
 291                 rcu_read_unlock();
 292                 cnt = pnfs_bucket_scan_array(cinfo, array->buckets,
 293                                 array->nbuckets, max);
 294                 rcu_read_lock();
 295                 pnfs_put_commit_array(array, cinfo->inode);
 296                 rv += cnt;
 297                 max -= cnt;
 298                 if (!max)
 299                         break;
 300         }
 301         rcu_read_unlock();
 302         return rv;
 303 }
 304 EXPORT_SYMBOL_GPL(pnfs_generic_scan_commit_lists);
 305
 306 static unsigned int
 307 pnfs_bucket_recover_commit_reqs(struct list_head *dst,
 308                                 struct pnfs_commit_bucket *buckets,
 309                                 unsigned int nbuckets,
 310                                 struct nfs_commit_info *cinfo)
 311 {
 312         struct pnfs_commit_bucket *b;
 313         struct pnfs_layout_segment *freeme;
 314         unsigned int nwritten, ret = 0;
 315         unsigned int i;
 316
 317 restart:
 318         for (i = 0, b = buckets; i < nbuckets; i++, b++) {
 319                 nwritten = nfs_scan_commit_list(&b->written, dst, cinfo, 0);
 320                 if (!nwritten)
 321                         continue;
 322                 ret += nwritten;
 323                 freeme = pnfs_free_bucket_lseg(b);
 324                 if (freeme) {
 325                         pnfs_put_lseg(freeme);
 326                         goto restart;
 327                 }
 328         }
 329         return ret;
 330 }
 331
 332 /* Pull everything off the committing lists and dump into @dst.  */
 333 void pnfs_generic_recover_commit_reqs(struct list_head *dst,
 334                                       struct nfs_commit_info *cinfo)
 335 {
 336         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 337         struct pnfs_commit_array *array;
 338         unsigned int nwritten;
 339
 340         lockdep_assert_held(&NFS_I(cinfo->inode)->commit_mutex);
 341         rcu_read_lock();
 342         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 343                 if (!array->lseg || !pnfs_get_commit_array(array))
 344                         continue;
 345                 rcu_read_unlock();
 346                 nwritten = pnfs_bucket_recover_commit_reqs(dst,
 347                                                            array->buckets,
 348                                                            array->nbuckets,
 349                                                            cinfo);
 350                 rcu_read_lock();
 351                 pnfs_put_commit_array(array, cinfo->inode);
 352                 fl_cinfo->nwritten -= nwritten;
 353         }
 354         rcu_read_unlock();
 355 }
 356 EXPORT_SYMBOL_GPL(pnfs_generic_recover_commit_reqs);
 357
 358 static struct nfs_page *
 359 pnfs_bucket_search_commit_reqs(struct pnfs_commit_bucket *buckets,
 360                 unsigned int nbuckets, struct page *page)
 361 {
 362         struct nfs_page *req;
 363         struct pnfs_commit_bucket *b;
 364         unsigned int i;
 365
 366         /* Linearly search the commit lists for each bucket until a matching
 367          * request is found */
 368         for (i = 0, b = buckets; i < nbuckets; i++, b++) {
 369                 list_for_each_entry(req, &b->written, wb_list) {
 370                         if (req->wb_page == page)
 371                                 return req->wb_head;
 372                 }
 373                 list_for_each_entry(req, &b->committing, wb_list) {
 374                         if (req->wb_page == page)
 375                                 return req->wb_head;
 376                 }
 377         }
 378         return NULL;
 379 }
 380
 381 /* pnfs_generic_search_commit_reqs - Search lists in @cinfo for the head reqest
 382  *                                 for @page
 383  * @cinfo - commit info for current inode
 384  * @page - page to search for matching head request
 385  *
 386  * Returns a the head request if one is found, otherwise returns NULL.
 387  */
 388 struct nfs_page *
 389 pnfs_generic_search_commit_reqs(struct nfs_commit_info *cinfo, struct page *page)
 390 {
 391         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 392         struct pnfs_commit_array *array;
 393         struct nfs_page *req;
 394
 395         list_for_each_entry(array, &fl_cinfo->commits, cinfo_list) {
 396                 req = pnfs_bucket_search_commit_reqs(array->buckets,
 397                                 array->nbuckets, page);
 398                 if (req)
 399                         return req;
 400         }
 401         return NULL;
 402 }
 403 EXPORT_SYMBOL_GPL(pnfs_generic_search_commit_reqs);
 404
 405 static struct pnfs_layout_segment *
 406 pnfs_bucket_get_committing(struct list_head *head,
 407                            struct pnfs_commit_bucket *bucket,
 408                            struct nfs_commit_info *cinfo)
 409 {
 410         struct list_head *pos;
 411
 412         list_for_each(pos, &bucket->committing)
 413                 cinfo->ds->ncommitting--;
 414         list_splice_init(&bucket->committing, head);
 415         return pnfs_free_bucket_lseg(bucket);
 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         if (!data->lseg)
 428                 data->lseg = pnfs_get_lseg(bucket->lseg);
 429         return data;
 430 }
 431
 432 static void pnfs_generic_retry_commit(struct pnfs_commit_bucket *buckets,
 433                                       unsigned int nbuckets,
 434                                       struct nfs_commit_info *cinfo,
 435                                       unsigned int idx)
 436 {
 437         struct pnfs_commit_bucket *bucket;
 438         struct pnfs_layout_segment *freeme;
 439         LIST_HEAD(pages);
 440
 441         for (bucket = buckets; idx < nbuckets; bucket++, idx++) {
 442                 if (list_empty(&bucket->committing))
 443                         continue;
 444                 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 445                 freeme = pnfs_bucket_get_committing(&pages, bucket, cinfo);
 446                 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 447                 nfs_retry_commit(&pages, freeme, cinfo, idx);
 448                 pnfs_put_lseg(freeme);
 449         }
 450 }
 451
 452 static unsigned int
 453 pnfs_bucket_alloc_ds_commits(struct list_head *list,
 454                              struct pnfs_commit_bucket *buckets,
 455                              unsigned int nbuckets,
 456                              struct nfs_commit_info *cinfo)
 457 {
 458         struct pnfs_commit_bucket *bucket;
 459         struct nfs_commit_data *data;
 460         unsigned int i;
 461         unsigned int nreq = 0;
 462
 463         for (i = 0, bucket = buckets; i < nbuckets; i++, bucket++) {
 464                 if (list_empty(&bucket->committing))
 465                         continue;
 466                 mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
 467                 if (!list_empty(&bucket->committing)) {
 468                         data = pnfs_bucket_fetch_commitdata(bucket, cinfo);
 469                         if (!data)
 470                                 goto out_error;
 471                         data->ds_commit_index = i;
 472                         list_add_tail(&data->list, list);
 473                         atomic_inc(&cinfo->mds->rpcs_out);
 474                         nreq++;
 475                 }
 476                 mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 477         }
 478         return nreq;
 479 out_error:
 480         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
 481         /* Clean up on error */
 482         pnfs_generic_retry_commit(buckets, nbuckets, cinfo, i);
 483         return nreq;
 484 }
 485
 486 static unsigned int
 487 pnfs_alloc_ds_commits_list(struct list_head *list,
 488                            struct pnfs_ds_commit_info *fl_cinfo,
 489                            struct nfs_commit_info *cinfo)
 490 {
 491         struct pnfs_commit_array *array;
 492         unsigned int ret = 0;
 493
 494         rcu_read_lock();
 495         list_for_each_entry_rcu(array, &fl_cinfo->commits, cinfo_list) {
 496                 if (!array->lseg || !pnfs_get_commit_array(array))
 497                         continue;
 498                 rcu_read_unlock();
 499                 ret += pnfs_bucket_alloc_ds_commits(list, array->buckets,
 500                                 array->nbuckets, cinfo);
 501                 rcu_read_lock();
 502                 pnfs_put_commit_array(array, cinfo->inode);
 503         }
 504         rcu_read_unlock();
 505         return ret;
 506 }
 507
 508 /* This follows nfs_commit_list pretty closely */
 509 int
 510 pnfs_generic_commit_pagelist(struct inode *inode, struct list_head *mds_pages,
 511                              int how, struct nfs_commit_info *cinfo,
 512                              int (*initiate_commit)(struct nfs_commit_data *data,
 513                                                     int how))
 514 {
 515         struct pnfs_ds_commit_info *fl_cinfo = cinfo->ds;
 516         struct nfs_commit_data *data, *tmp;
 517         LIST_HEAD(list);
 518         unsigned int nreq = 0;
 519
 520         if (!list_empty(mds_pages)) {
 521                 data = nfs_commitdata_alloc(true);
 522                 data->ds_commit_index = -1;
 523                 list_splice_init(mds_pages, &data->pages);
 524                 list_add_tail(&data->list, &list);
 525                 atomic_inc(&cinfo->mds->rpcs_out);
 526                 nreq++;
 527         }
 528
 529         nreq += pnfs_alloc_ds_commits_list(&list, fl_cinfo, cinfo);
 530         if (nreq == 0)
 531                 goto out;
 532
 533         list_for_each_entry_safe(data, tmp, &list, list) {
 534                 list_del(&data->list);
 535                 if (data->ds_commit_index < 0) {
 536                         nfs_init_commit(data, NULL, NULL, cinfo);
 537                         nfs_initiate_commit(NFS_CLIENT(inode), data,
 538                                             NFS_PROTO(data->inode),
 539                                             data->mds_ops, how,
 540                                             RPC_TASK_CRED_NOREF);
 541                 } else {
 542                         nfs_init_commit(data, NULL, data->lseg, cinfo);
 543                         initiate_commit(data, how);
 544                 }
 545         }
 546 out:
 547         return PNFS_ATTEMPTED;
 548 }
 549 EXPORT_SYMBOL_GPL(pnfs_generic_commit_pagelist);
 550
 551 /*
 552  * Data server cache
 553  *
 554  * Data servers can be mapped to different device ids.
 555  * nfs4_pnfs_ds reference counting
 556  *   - set to 1 on allocation
 557  *   - incremented when a device id maps a data server already in the cache.
 558  *   - decremented when deviceid is removed from the cache.
 559  */
 560 static DEFINE_SPINLOCK(nfs4_ds_cache_lock);
 561 static LIST_HEAD(nfs4_data_server_cache);
 562
 563 /* Debug routines */
 564 static void
 565 print_ds(struct nfs4_pnfs_ds *ds)
 566 {
 567         if (ds == NULL) {
 568                 printk(KERN_WARNING "%s NULL device\n", __func__);
 569                 return;
 570         }
 571         printk(KERN_WARNING "        ds %s\n"
 572                 "        ref count %d\n"
 573                 "        client %p\n"
 574                 "        cl_exchange_flags %x\n",
 575                 ds->ds_remotestr,
 576                 refcount_read(&ds->ds_count), ds->ds_clp,
 577                 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
 578 }
 579
 580 static bool
 581 same_sockaddr(struct sockaddr *addr1, struct sockaddr *addr2)
 582 {
 583         struct sockaddr_in *a, *b;
 584         struct sockaddr_in6 *a6, *b6;
 585
 586         if (addr1->sa_family != addr2->sa_family)
 587                 return false;
 588
 589         switch (addr1->sa_family) {
 590         case AF_INET:
 591                 a = (struct sockaddr_in *)addr1;
 592                 b = (struct sockaddr_in *)addr2;
 593
 594                 if (a->sin_addr.s_addr == b->sin_addr.s_addr &&
 595                     a->sin_port == b->sin_port)
 596                         return true;
 597                 break;
 598
 599         case AF_INET6:
 600                 a6 = (struct sockaddr_in6 *)addr1;
 601                 b6 = (struct sockaddr_in6 *)addr2;
 602
 603                 /* LINKLOCAL addresses must have matching scope_id */
 604                 if (ipv6_addr_src_scope(&a6->sin6_addr) ==
 605                     IPV6_ADDR_SCOPE_LINKLOCAL &&
 606                     a6->sin6_scope_id != b6->sin6_scope_id)
 607                         return false;
 608
 609                 if (ipv6_addr_equal(&a6->sin6_addr, &b6->sin6_addr) &&
 610                     a6->sin6_port == b6->sin6_port)
 611                         return true;
 612                 break;
 613
 614         default:
 615                 dprintk("%s: unhandled address family: %u\n",
 616                         __func__, addr1->sa_family);
 617                 return false;
 618         }
 619
 620         return false;
 621 }
 622
 623 /*
 624  * Checks if 'dsaddrs1' contains a subset of 'dsaddrs2'. If it does,
 625  * declare a match.
 626  */
 627 static bool
 628 _same_data_server_addrs_locked(const struct list_head *dsaddrs1,
 629                                const struct list_head *dsaddrs2)
 630 {
 631         struct nfs4_pnfs_ds_addr *da1, *da2;
 632         struct sockaddr *sa1, *sa2;
 633         bool match = false;
 634
 635         list_for_each_entry(da1, dsaddrs1, da_node) {
 636                 sa1 = (struct sockaddr *)&da1->da_addr;
 637                 match = false;
 638                 list_for_each_entry(da2, dsaddrs2, da_node) {
 639                         sa2 = (struct sockaddr *)&da2->da_addr;
 640                         match = same_sockaddr(sa1, sa2);
 641                         if (match)
 642                                 break;
 643                 }
 644                 if (!match)
 645                         break;
 646         }
 647         return match;
 648 }
 649
 650 /*
 651  * Lookup DS by addresses.  nfs4_ds_cache_lock is held
 652  */
 653 static struct nfs4_pnfs_ds *
 654 _data_server_lookup_locked(const struct list_head *dsaddrs)
 655 {
 656         struct nfs4_pnfs_ds *ds;
 657
 658         list_for_each_entry(ds, &nfs4_data_server_cache, ds_node)
 659                 if (_same_data_server_addrs_locked(&ds->ds_addrs, dsaddrs))
 660                         return ds;
 661         return NULL;
 662 }
 663
 664 static struct nfs4_pnfs_ds_addr *nfs4_pnfs_ds_addr_alloc(gfp_t gfp_flags)
 665 {
 666         struct nfs4_pnfs_ds_addr *da = kzalloc(sizeof(*da), gfp_flags);
 667         if (da)
 668                 INIT_LIST_HEAD(&da->da_node);
 669         return da;
 670 }
 671
 672 static void nfs4_pnfs_ds_addr_free(struct nfs4_pnfs_ds_addr *da)
 673 {
 674         kfree(da->da_remotestr);
 675         kfree(da->da_netid);
 676         kfree(da);
 677 }
 678
 679 static void destroy_ds(struct nfs4_pnfs_ds *ds)
 680 {
 681         struct nfs4_pnfs_ds_addr *da;
 682
 683         dprintk("--> %s\n", __func__);
 684         ifdebug(FACILITY)
 685                 print_ds(ds);
 686
 687         nfs_put_client(ds->ds_clp);
 688
 689         while (!list_empty(&ds->ds_addrs)) {
 690                 da = list_first_entry(&ds->ds_addrs,
 691                                       struct nfs4_pnfs_ds_addr,
 692                                       da_node);
 693                 list_del_init(&da->da_node);
 694                 nfs4_pnfs_ds_addr_free(da);
 695         }
 696
 697         kfree(ds->ds_remotestr);
 698         kfree(ds);
 699 }
 700
 701 void nfs4_pnfs_ds_put(struct nfs4_pnfs_ds *ds)
 702 {
 703         if (refcount_dec_and_lock(&ds->ds_count,
 704                                 &nfs4_ds_cache_lock)) {
 705                 list_del_init(&ds->ds_node);
 706                 spin_unlock(&nfs4_ds_cache_lock);
 707                 destroy_ds(ds);
 708         }
 709 }
 710 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_put);
 711
 712 /*
 713  * Create a string with a human readable address and port to avoid
 714  * complicated setup around many dprinks.
 715  */
 716 static char *
 717 nfs4_pnfs_remotestr(struct list_head *dsaddrs, gfp_t gfp_flags)
 718 {
 719         struct nfs4_pnfs_ds_addr *da;
 720         char *remotestr;
 721         size_t len;
 722         char *p;
 723
 724         len = 3;        /* '{', '}' and eol */
 725         list_for_each_entry(da, dsaddrs, da_node) {
 726                 len += strlen(da->da_remotestr) + 1;    /* string plus comma */
 727         }
 728
 729         remotestr = kzalloc(len, gfp_flags);
 730         if (!remotestr)
 731                 return NULL;
 732
 733         p = remotestr;
 734         *(p++) = '{';
 735         len--;
 736         list_for_each_entry(da, dsaddrs, da_node) {
 737                 size_t ll = strlen(da->da_remotestr);
 738
 739                 if (ll > len)
 740                         goto out_err;
 741
 742                 memcpy(p, da->da_remotestr, ll);
 743                 p += ll;
 744                 len -= ll;
 745
 746                 if (len < 1)
 747                         goto out_err;
 748                 (*p++) = ',';
 749                 len--;
 750         }
 751         if (len < 2)
 752                 goto out_err;
 753         *(p++) = '}';
 754         *p = '\0';
 755         return remotestr;
 756 out_err:
 757         kfree(remotestr);
 758         return NULL;
 759 }
 760
 761 /*
 762  * Given a list of multipath struct nfs4_pnfs_ds_addr, add it to ds cache if
 763  * uncached and return cached struct nfs4_pnfs_ds.
 764  */
 765 struct nfs4_pnfs_ds *
 766 nfs4_pnfs_ds_add(struct list_head *dsaddrs, gfp_t gfp_flags)
 767 {
 768         struct nfs4_pnfs_ds *tmp_ds, *ds = NULL;
 769         char *remotestr;
 770
 771         if (list_empty(dsaddrs)) {
 772                 dprintk("%s: no addresses defined\n", __func__);
 773                 goto out;
 774         }
 775
 776         ds = kzalloc(sizeof(*ds), gfp_flags);
 777         if (!ds)
 778                 goto out;
 779
 780         /* this is only used for debugging, so it's ok if its NULL */
 781         remotestr = nfs4_pnfs_remotestr(dsaddrs, gfp_flags);
 782
 783         spin_lock(&nfs4_ds_cache_lock);
 784         tmp_ds = _data_server_lookup_locked(dsaddrs);
 785         if (tmp_ds == NULL) {
 786                 INIT_LIST_HEAD(&ds->ds_addrs);
 787                 list_splice_init(dsaddrs, &ds->ds_addrs);
 788                 ds->ds_remotestr = remotestr;
 789                 refcount_set(&ds->ds_count, 1);
 790                 INIT_LIST_HEAD(&ds->ds_node);
 791                 ds->ds_clp = NULL;
 792                 list_add(&ds->ds_node, &nfs4_data_server_cache);
 793                 dprintk("%s add new data server %s\n", __func__,
 794                         ds->ds_remotestr);
 795         } else {
 796                 kfree(remotestr);
 797                 kfree(ds);
 798                 refcount_inc(&tmp_ds->ds_count);
 799                 dprintk("%s data server %s found, inc'ed ds_count to %d\n",
 800                         __func__, tmp_ds->ds_remotestr,
 801                         refcount_read(&tmp_ds->ds_count));
 802                 ds = tmp_ds;
 803         }
 804         spin_unlock(&nfs4_ds_cache_lock);
 805 out:
 806         return ds;
 807 }
 808 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_add);
 809
 810 static void nfs4_wait_ds_connect(struct nfs4_pnfs_ds *ds)
 811 {
 812         might_sleep();
 813         wait_on_bit(&ds->ds_state, NFS4DS_CONNECTING,
 814                         TASK_KILLABLE);
 815 }
 816
 817 static void nfs4_clear_ds_conn_bit(struct nfs4_pnfs_ds *ds)
 818 {
 819         smp_mb__before_atomic();
 820         clear_bit(NFS4DS_CONNECTING, &ds->ds_state);
 821         smp_mb__after_atomic();
 822         wake_up_bit(&ds->ds_state, NFS4DS_CONNECTING);
 823 }
 824
 825 static struct nfs_client *(*get_v3_ds_connect)(
 826                         struct nfs_server *mds_srv,
 827                         const struct sockaddr *ds_addr,
 828                         int ds_addrlen,
 829                         int ds_proto,
 830                         unsigned int ds_timeo,
 831                         unsigned int ds_retrans);
 832
 833 static bool load_v3_ds_connect(void)
 834 {
 835         if (!get_v3_ds_connect) {
 836                 get_v3_ds_connect = symbol_request(nfs3_set_ds_client);
 837                 WARN_ON_ONCE(!get_v3_ds_connect);
 838         }
 839
 840         return(get_v3_ds_connect != NULL);
 841 }
 842
 843 void nfs4_pnfs_v3_ds_connect_unload(void)
 844 {
 845         if (get_v3_ds_connect) {
 846                 symbol_put(nfs3_set_ds_client);
 847                 get_v3_ds_connect = NULL;
 848         }
 849 }
 850
 851 static int _nfs4_pnfs_v3_ds_connect(struct nfs_server *mds_srv,
 852                                  struct nfs4_pnfs_ds *ds,
 853                                  unsigned int timeo,
 854                                  unsigned int retrans)
 855 {
 856         struct nfs_client *clp = ERR_PTR(-EIO);
 857         struct nfs4_pnfs_ds_addr *da;
 858         int status = 0;
 859
 860         dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 861
 862         if (!load_v3_ds_connect())
 863                 goto out;
 864
 865         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 866                 dprintk("%s: DS %s: trying address %s\n",
 867                         __func__, ds->ds_remotestr, da->da_remotestr);
 868
 869                 if (!IS_ERR(clp)) {
 870                         struct xprt_create xprt_args = {
 871                                 .ident = da->da_transport,
 872                                 .net = clp->cl_net,
 873                                 .dstaddr = (struct sockaddr *)&da->da_addr,
 874                                 .addrlen = da->da_addrlen,
 875                                 .servername = clp->cl_hostname,
 876                         };
 877
 878                         if (da->da_transport != clp->cl_proto)
 879                                 continue;
 880                         if (da->da_addr.ss_family != clp->cl_addr.ss_family)
 881                                 continue;
 882                         /* Add this address as an alias */
 883                         rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 884                                         rpc_clnt_test_and_add_xprt, NULL);
 885                         continue;
 886                 }
 887                 clp = get_v3_ds_connect(mds_srv,
 888                                 (struct sockaddr *)&da->da_addr,
 889                                 da->da_addrlen, da->da_transport,
 890                                 timeo, retrans);
 891                 if (IS_ERR(clp))
 892                         continue;
 893                 clp->cl_rpcclient->cl_softerr = 0;
 894                 clp->cl_rpcclient->cl_softrtry = 0;
 895         }
 896
 897         if (IS_ERR(clp)) {
 898                 status = PTR_ERR(clp);
 899                 goto out;
 900         }
 901
 902         smp_wmb();
 903         ds->ds_clp = clp;
 904         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 905 out:
 906         return status;
 907 }
 908
 909 static int _nfs4_pnfs_v4_ds_connect(struct nfs_server *mds_srv,
 910                                  struct nfs4_pnfs_ds *ds,
 911                                  unsigned int timeo,
 912                                  unsigned int retrans,
 913                                  u32 minor_version)
 914 {
 915         struct nfs_client *clp = ERR_PTR(-EIO);
 916         struct nfs4_pnfs_ds_addr *da;
 917         int status = 0;
 918
 919         dprintk("--> %s DS %s\n", __func__, ds->ds_remotestr);
 920
 921         list_for_each_entry(da, &ds->ds_addrs, da_node) {
 922                 dprintk("%s: DS %s: trying address %s\n",
 923                         __func__, ds->ds_remotestr, da->da_remotestr);
 924
 925                 if (!IS_ERR(clp) && clp->cl_mvops->session_trunk) {
 926                         struct xprt_create xprt_args = {
 927                                 .ident = da->da_transport,
 928                                 .net = clp->cl_net,
 929                                 .dstaddr = (struct sockaddr *)&da->da_addr,
 930                                 .addrlen = da->da_addrlen,
 931                                 .servername = clp->cl_hostname,
 932                         };
 933                         struct nfs4_add_xprt_data xprtdata = {
 934                                 .clp = clp,
 935                         };
 936                         struct rpc_add_xprt_test rpcdata = {
 937                                 .add_xprt_test = clp->cl_mvops->session_trunk,
 938                                 .data = &xprtdata,
 939                         };
 940
 941                         if (da->da_transport != clp->cl_proto)
 942                                 continue;
 943                         if (da->da_addr.ss_family != clp->cl_addr.ss_family)
 944                                 continue;
 945                         /**
 946                         * Test this address for session trunking and
 947                         * add as an alias
 948                         */
 949                         xprtdata.cred = nfs4_get_clid_cred(clp),
 950                         rpc_clnt_add_xprt(clp->cl_rpcclient, &xprt_args,
 951                                           rpc_clnt_setup_test_and_add_xprt,
 952                                           &rpcdata);
 953                         if (xprtdata.cred)
 954                                 put_cred(xprtdata.cred);
 955                 } else {
 956                         clp = nfs4_set_ds_client(mds_srv,
 957                                                 (struct sockaddr *)&da->da_addr,
 958                                                 da->da_addrlen,
 959                                                 da->da_transport, timeo,
 960                                                 retrans, minor_version);
 961                         if (IS_ERR(clp))
 962                                 continue;
 963
 964                         status = nfs4_init_ds_session(clp,
 965                                         mds_srv->nfs_client->cl_lease_time);
 966                         if (status) {
 967                                 nfs_put_client(clp);
 968                                 clp = ERR_PTR(-EIO);
 969                                 continue;
 970                         }
 971
 972                 }
 973         }
 974
 975         if (IS_ERR(clp)) {
 976                 status = PTR_ERR(clp);
 977                 goto out;
 978         }
 979
 980         smp_wmb();
 981         ds->ds_clp = clp;
 982         dprintk("%s [new] addr: %s\n", __func__, ds->ds_remotestr);
 983 out:
 984         return status;
 985 }
 986
 987 /*
 988  * Create an rpc connection to the nfs4_pnfs_ds data server.
 989  * Currently only supports IPv4 and IPv6 addresses.
 990  * If connection fails, make devid unavailable and return a -errno.
 991  */
 992 int nfs4_pnfs_ds_connect(struct nfs_server *mds_srv, struct nfs4_pnfs_ds *ds,
 993                           struct nfs4_deviceid_node *devid, unsigned int timeo,
 994                           unsigned int retrans, u32 version, u32 minor_version)
 995 {
 996         int err;
 997
 998 again:
 999         err = 0;
1000         if (test_and_set_bit(NFS4DS_CONNECTING, &ds->ds_state) == 0) {
1001                 if (version == 3) {
1002                         err = _nfs4_pnfs_v3_ds_connect(mds_srv, ds, timeo,
1003                                                        retrans);
1004                 } else if (version == 4) {
1005                         err = _nfs4_pnfs_v4_ds_connect(mds_srv, ds, timeo,
1006                                                        retrans, minor_version);
1007                 } else {
1008                         dprintk("%s: unsupported DS version %d\n", __func__,
1009                                 version);
1010                         err = -EPROTONOSUPPORT;
1011                 }
1012
1013                 nfs4_clear_ds_conn_bit(ds);
1014         } else {
1015                 nfs4_wait_ds_connect(ds);
1016
1017                 /* what was waited on didn't connect AND didn't mark unavail */
1018                 if (!ds->ds_clp && !nfs4_test_deviceid_unavailable(devid))
1019                         goto again;
1020         }
1021
1022         /*
1023          * At this point the ds->ds_clp should be ready, but it might have
1024          * hit an error.
1025          */
1026         if (!err) {
1027                 if (!ds->ds_clp || !nfs_client_init_is_complete(ds->ds_clp)) {
1028                         WARN_ON_ONCE(ds->ds_clp ||
1029                                 !nfs4_test_deviceid_unavailable(devid));
1030                         return -EINVAL;
1031                 }
1032                 err = nfs_client_init_status(ds->ds_clp);
1033         }
1034
1035         return err;
1036 }
1037 EXPORT_SYMBOL_GPL(nfs4_pnfs_ds_connect);
1038
1039 /*
1040  * Currently only supports ipv4, ipv6 and one multi-path address.
1041  */
1042 struct nfs4_pnfs_ds_addr *
1043 nfs4_decode_mp_ds_addr(struct net *net, struct xdr_stream *xdr, gfp_t gfp_flags)
1044 {
1045         struct nfs4_pnfs_ds_addr *da = NULL;
1046         char *buf, *portstr;
1047         __be16 port;
1048         ssize_t nlen, rlen;
1049         int tmp[2];
1050         char *netid;
1051         size_t len;
1052         char *startsep = "";
1053         char *endsep = "";
1054
1055
1056         /* r_netid */
1057         nlen = xdr_stream_decode_string_dup(xdr, &netid, XDR_MAX_NETOBJ,
1058                                             gfp_flags);
1059         if (unlikely(nlen < 0))
1060                 goto out_err;
1061
1062         /* r_addr: ip/ip6addr with port in dec octets - see RFC 5665 */
1063         /* port is ".ABC.DEF", 8 chars max */
1064         rlen = xdr_stream_decode_string_dup(xdr, &buf, INET6_ADDRSTRLEN +
1065                                             IPV6_SCOPE_ID_LEN + 8, gfp_flags);
1066         if (unlikely(rlen < 0))
1067                 goto out_free_netid;
1068
1069         /* replace port '.' with '-' */
1070         portstr = strrchr(buf, '.');
1071         if (!portstr) {
1072                 dprintk("%s: Failed finding expected dot in port\n",
1073                         __func__);
1074                 goto out_free_buf;
1075         }
1076         *portstr = '-';
1077
1078         /* find '.' between address and port */
1079         portstr = strrchr(buf, '.');
1080         if (!portstr) {
1081                 dprintk("%s: Failed finding expected dot between address and "
1082                         "port\n", __func__);
1083                 goto out_free_buf;
1084         }
1085         *portstr = '\0';
1086
1087         da = nfs4_pnfs_ds_addr_alloc(gfp_flags);
1088         if (unlikely(!da))
1089                 goto out_free_buf;
1090
1091         if (!rpc_pton(net, buf, portstr-buf, (struct sockaddr *)&da->da_addr,
1092                       sizeof(da->da_addr))) {
1093                 dprintk("%s: error parsing address %s\n", __func__, buf);
1094                 goto out_free_da;
1095         }
1096
1097         portstr++;
1098         sscanf(portstr, "%d-%d", &tmp[0], &tmp[1]);
1099         port = htons((tmp[0] << 8) | (tmp[1]));
1100
1101         switch (da->da_addr.ss_family) {
1102         case AF_INET:
1103                 ((struct sockaddr_in *)&da->da_addr)->sin_port = port;
1104                 da->da_addrlen = sizeof(struct sockaddr_in);
1105                 break;
1106
1107         case AF_INET6:
1108                 ((struct sockaddr_in6 *)&da->da_addr)->sin6_port = port;
1109                 da->da_addrlen = sizeof(struct sockaddr_in6);
1110                 startsep = "[";
1111                 endsep = "]";
1112                 break;
1113
1114         default:
1115                 dprintk("%s: unsupported address family: %u\n",
1116                         __func__, da->da_addr.ss_family);
1117                 goto out_free_da;
1118         }
1119
1120         da->da_transport = xprt_find_transport_ident(netid);
1121         if (da->da_transport < 0) {
1122                 dprintk("%s: ERROR: unknown r_netid \"%s\"\n",
1123                         __func__, netid);
1124                 goto out_free_da;
1125         }
1126
1127         da->da_netid = netid;
1128
1129         /* save human readable address */
1130         len = strlen(startsep) + strlen(buf) + strlen(endsep) + 7;
1131         da->da_remotestr = kzalloc(len, gfp_flags);
1132
1133         /* NULL is ok, only used for dprintk */
1134         if (da->da_remotestr)
1135                 snprintf(da->da_remotestr, len, "%s%s%s:%u", startsep,
1136                          buf, endsep, ntohs(port));
1137
1138         dprintk("%s: Parsed DS addr %s\n", __func__, da->da_remotestr);
1139         kfree(buf);
1140         return da;
1141
1142 out_free_da:
1143         kfree(da);
1144 out_free_buf:
1145         dprintk("%s: Error parsing DS addr: %s\n", __func__, buf);
1146         kfree(buf);
1147 out_free_netid:
1148         kfree(netid);
1149 out_err:
1150         return NULL;
1151 }
1152 EXPORT_SYMBOL_GPL(nfs4_decode_mp_ds_addr);
1153
1154 void
1155 pnfs_layout_mark_request_commit(struct nfs_page *req,
1156                                 struct pnfs_layout_segment *lseg,
1157                                 struct nfs_commit_info *cinfo,
1158                                 u32 ds_commit_idx)
1159 {
1160         struct list_head *list;
1161         struct pnfs_commit_array *array;
1162         struct pnfs_commit_bucket *bucket;
1163
1164         mutex_lock(&NFS_I(cinfo->inode)->commit_mutex);
1165         array = pnfs_lookup_commit_array(cinfo->ds, lseg);
1166         if (!array || !pnfs_is_valid_lseg(lseg))
1167                 goto out_resched;
1168         bucket = &array->buckets[ds_commit_idx];
1169         list = &bucket->written;
1170         /* Non-empty buckets hold a reference on the lseg.  That ref
1171          * is normally transferred to the COMMIT call and released
1172          * there.  It could also be released if the last req is pulled
1173          * off due to a rewrite, in which case it will be done in
1174          * pnfs_common_clear_request_commit
1175          */
1176         if (!bucket->lseg)
1177                 bucket->lseg = pnfs_get_lseg(lseg);
1178         set_bit(PG_COMMIT_TO_DS, &req->wb_flags);
1179         cinfo->ds->nwritten++;
1180
1181         nfs_request_add_commit_list_locked(req, list, cinfo);
1182         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1183         nfs_mark_page_unstable(req->wb_page, cinfo);
1184         return;
1185 out_resched:
1186         mutex_unlock(&NFS_I(cinfo->inode)->commit_mutex);
1187         cinfo->completion_ops->resched_write(cinfo, req);
1188 }
1189 EXPORT_SYMBOL_GPL(pnfs_layout_mark_request_commit);
1190
1191 int
1192 pnfs_nfs_generic_sync(struct inode *inode, bool datasync)
1193 {
1194         int ret;
1195
1196         if (!pnfs_layoutcommit_outstanding(inode))
1197                 return 0;
1198         ret = nfs_commit_inode(inode, FLUSH_SYNC);
1199         if (ret < 0)
1200                 return ret;
1201         if (datasync)
1202                 return 0;
1203         return pnfs_layoutcommit_inode(inode, true);
1204 }
1205 EXPORT_SYMBOL_GPL(pnfs_nfs_generic_sync);
1206