fs/nfsd/nfs4layouts.c

   1 /*
   2  * Copyright (c) 2014 Christoph Hellwig.
   3  */
   4 #include <linux/kmod.h>
   5 #include <linux/file.h>
   6 #include <linux/jhash.h>
   7 #include <linux/sched.h>
   8 #include <linux/sunrpc/addr.h>
   9
  10 #include "pnfs.h"
  11 #include "netns.h"
  12 #include "trace.h"
  13
  14 #define NFSDDBG_FACILITY                NFSDDBG_PNFS
  15
  16 struct nfs4_layout {
  17         struct list_head                lo_perstate;
  18         struct nfs4_layout_stateid      *lo_state;
  19         struct nfsd4_layout_seg         lo_seg;
  20 };
  21
  22 static struct kmem_cache *nfs4_layout_cache;
  23 static struct kmem_cache *nfs4_layout_stateid_cache;
  24
  25 static struct nfsd4_callback_ops nfsd4_cb_layout_ops;
  26 static const struct lock_manager_operations nfsd4_layouts_lm_ops;
  27
  28 const struct nfsd4_layout_ops *nfsd4_layout_ops[LAYOUT_TYPE_MAX] =  {
  29         [LAYOUT_BLOCK_VOLUME]   = &bl_layout_ops,
  30 };
  31
  32 /* pNFS device ID to export fsid mapping */
  33 #define DEVID_HASH_BITS 8
  34 #define DEVID_HASH_SIZE (1 << DEVID_HASH_BITS)
  35 #define DEVID_HASH_MASK (DEVID_HASH_SIZE - 1)
  36 static u64 nfsd_devid_seq = 1;
  37 static struct list_head nfsd_devid_hash[DEVID_HASH_SIZE];
  38 static DEFINE_SPINLOCK(nfsd_devid_lock);
  39
  40 static inline u32 devid_hashfn(u64 idx)
  41 {
  42         return jhash_2words(idx, idx >> 32, 0) & DEVID_HASH_MASK;
  43 }
  44
  45 static void
  46 nfsd4_alloc_devid_map(const struct svc_fh *fhp)
  47 {
  48         const struct knfsd_fh *fh = &fhp->fh_handle;
  49         size_t fsid_len = key_len(fh->fh_fsid_type);
  50         struct nfsd4_deviceid_map *map, *old;
  51         int i;
  52
  53         map = kzalloc(sizeof(*map) + fsid_len, GFP_KERNEL);
  54         if (!map)
  55                 return;
  56
  57         map->fsid_type = fh->fh_fsid_type;
  58         memcpy(&map->fsid, fh->fh_fsid, fsid_len);
  59
  60         spin_lock(&nfsd_devid_lock);
  61         if (fhp->fh_export->ex_devid_map)
  62                 goto out_unlock;
  63
  64         for (i = 0; i < DEVID_HASH_SIZE; i++) {
  65                 list_for_each_entry(old, &nfsd_devid_hash[i], hash) {
  66                         if (old->fsid_type != fh->fh_fsid_type)
  67                                 continue;
  68                         if (memcmp(old->fsid, fh->fh_fsid,
  69                                         key_len(old->fsid_type)))
  70                                 continue;
  71
  72                         fhp->fh_export->ex_devid_map = old;
  73                         goto out_unlock;
  74                 }
  75         }
  76
  77         map->idx = nfsd_devid_seq++;
  78         list_add_tail_rcu(&map->hash, &nfsd_devid_hash[devid_hashfn(map->idx)]);
  79         fhp->fh_export->ex_devid_map = map;
  80         map = NULL;
  81
  82 out_unlock:
  83         spin_unlock(&nfsd_devid_lock);
  84         kfree(map);
  85 }
  86
  87 struct nfsd4_deviceid_map *
  88 nfsd4_find_devid_map(int idx)
  89 {
  90         struct nfsd4_deviceid_map *map, *ret = NULL;
  91
  92         rcu_read_lock();
  93         list_for_each_entry_rcu(map, &nfsd_devid_hash[devid_hashfn(idx)], hash)
  94                 if (map->idx == idx)
  95                         ret = map;
  96         rcu_read_unlock();
  97
  98         return ret;
  99 }
 100
 101 int
 102 nfsd4_set_deviceid(struct nfsd4_deviceid *id, const struct svc_fh *fhp,
 103                 u32 device_generation)
 104 {
 105         if (!fhp->fh_export->ex_devid_map) {
 106                 nfsd4_alloc_devid_map(fhp);
 107                 if (!fhp->fh_export->ex_devid_map)
 108                         return -ENOMEM;
 109         }
 110
 111         id->fsid_idx = fhp->fh_export->ex_devid_map->idx;
 112         id->generation = device_generation;
 113         id->pad = 0;
 114         return 0;
 115 }
 116
 117 void nfsd4_setup_layout_type(struct svc_export *exp)
 118 {
 119         struct super_block *sb = exp->ex_path.mnt->mnt_sb;
 120
 121         if (!(exp->ex_flags & NFSEXP_PNFS))
 122                 return;
 123
 124         if (sb->s_export_op->get_uuid &&
 125             sb->s_export_op->map_blocks &&
 126             sb->s_export_op->commit_blocks)
 127                 exp->ex_layout_type = LAYOUT_BLOCK_VOLUME;
 128 }
 129
 130 static void
 131 nfsd4_free_layout_stateid(struct nfs4_stid *stid)
 132 {
 133         struct nfs4_layout_stateid *ls = layoutstateid(stid);
 134         struct nfs4_client *clp = ls->ls_stid.sc_client;
 135         struct nfs4_file *fp = ls->ls_stid.sc_file;
 136
 137         trace_layoutstate_free(&ls->ls_stid.sc_stateid);
 138
 139         spin_lock(&clp->cl_lock);
 140         list_del_init(&ls->ls_perclnt);
 141         spin_unlock(&clp->cl_lock);
 142
 143         spin_lock(&fp->fi_lock);
 144         list_del_init(&ls->ls_perfile);
 145         spin_unlock(&fp->fi_lock);
 146
 147         vfs_setlease(ls->ls_file, F_UNLCK, NULL, (void **)&ls);
 148         fput(ls->ls_file);
 149
 150         if (ls->ls_recalled)
 151                 atomic_dec(&ls->ls_stid.sc_file->fi_lo_recalls);
 152
 153         kmem_cache_free(nfs4_layout_stateid_cache, ls);
 154 }
 155
 156 static int
 157 nfsd4_layout_setlease(struct nfs4_layout_stateid *ls)
 158 {
 159         struct file_lock *fl;
 160         int status;
 161
 162         fl = locks_alloc_lock();
 163         if (!fl)
 164                 return -ENOMEM;
 165         locks_init_lock(fl);
 166         fl->fl_lmops = &nfsd4_layouts_lm_ops;
 167         fl->fl_flags = FL_LAYOUT;
 168         fl->fl_type = F_RDLCK;
 169         fl->fl_end = OFFSET_MAX;
 170         fl->fl_owner = ls;
 171         fl->fl_pid = current->tgid;
 172         fl->fl_file = ls->ls_file;
 173
 174         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl, NULL);
 175         if (status) {
 176                 locks_free_lock(fl);
 177                 return status;
 178         }
 179         BUG_ON(fl != NULL);
 180         return 0;
 181 }
 182
 183 static struct nfs4_layout_stateid *
 184 nfsd4_alloc_layout_stateid(struct nfsd4_compound_state *cstate,
 185                 struct nfs4_stid *parent, u32 layout_type)
 186 {
 187         struct nfs4_client *clp = cstate->clp;
 188         struct nfs4_file *fp = parent->sc_file;
 189         struct nfs4_layout_stateid *ls;
 190         struct nfs4_stid *stp;
 191
 192         stp = nfs4_alloc_stid(cstate->clp, nfs4_layout_stateid_cache,
 193                                         nfsd4_free_layout_stateid);
 194         if (!stp)
 195                 return NULL;
 196
 197         get_nfs4_file(fp);
 198         stp->sc_file = fp;
 199
 200         ls = layoutstateid(stp);
 201         INIT_LIST_HEAD(&ls->ls_perclnt);
 202         INIT_LIST_HEAD(&ls->ls_perfile);
 203         spin_lock_init(&ls->ls_lock);
 204         INIT_LIST_HEAD(&ls->ls_layouts);
 205         mutex_init(&ls->ls_mutex);
 206         ls->ls_layout_type = layout_type;
 207         nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
 208                         NFSPROC4_CLNT_CB_LAYOUT);
 209
 210         if (parent->sc_type == NFS4_DELEG_STID)
 211                 ls->ls_file = get_file(fp->fi_deleg_file);
 212         else
 213                 ls->ls_file = find_any_file(fp);
 214         BUG_ON(!ls->ls_file);
 215
 216         if (nfsd4_layout_setlease(ls)) {
 217                 fput(ls->ls_file);
 218                 put_nfs4_file(fp);
 219                 kmem_cache_free(nfs4_layout_stateid_cache, ls);
 220                 return NULL;
 221         }
 222
 223         spin_lock(&clp->cl_lock);
 224         stp->sc_type = NFS4_LAYOUT_STID;
 225         list_add(&ls->ls_perclnt, &clp->cl_lo_states);
 226         spin_unlock(&clp->cl_lock);
 227
 228         spin_lock(&fp->fi_lock);
 229         list_add(&ls->ls_perfile, &fp->fi_lo_states);
 230         spin_unlock(&fp->fi_lock);
 231
 232         trace_layoutstate_alloc(&ls->ls_stid.sc_stateid);
 233         return ls;
 234 }
 235
 236 __be32
 237 nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
 238                 struct nfsd4_compound_state *cstate, stateid_t *stateid,
 239                 bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
 240 {
 241         struct nfs4_layout_stateid *ls;
 242         struct nfs4_stid *stid;
 243         unsigned char typemask = NFS4_LAYOUT_STID;
 244         __be32 status;
 245
 246         if (create)
 247                 typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
 248
 249         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
 250                         net_generic(SVC_NET(rqstp), nfsd_net_id));
 251         if (status)
 252                 goto out;
 253
 254         if (!fh_match(&cstate->current_fh.fh_handle,
 255                       &stid->sc_file->fi_fhandle)) {
 256                 status = nfserr_bad_stateid;
 257                 goto out_put_stid;
 258         }
 259
 260         if (stid->sc_type != NFS4_LAYOUT_STID) {
 261                 ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
 262                 nfs4_put_stid(stid);
 263
 264                 status = nfserr_jukebox;
 265                 if (!ls)
 266                         goto out;
 267                 mutex_lock(&ls->ls_mutex);
 268         } else {
 269                 ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
 270
 271                 status = nfserr_bad_stateid;
 272                 mutex_lock(&ls->ls_mutex);
 273                 if (stateid->si_generation > stid->sc_stateid.si_generation)
 274                         goto out_unlock_stid;
 275                 if (layout_type != ls->ls_layout_type)
 276                         goto out_unlock_stid;
 277         }
 278
 279         *lsp = ls;
 280         return 0;
 281
 282 out_unlock_stid:
 283         mutex_unlock(&ls->ls_mutex);
 284 out_put_stid:
 285         nfs4_put_stid(stid);
 286 out:
 287         return status;
 288 }
 289
 290 static void
 291 nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
 292 {
 293         spin_lock(&ls->ls_lock);
 294         if (ls->ls_recalled)
 295                 goto out_unlock;
 296
 297         ls->ls_recalled = true;
 298         atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
 299         if (list_empty(&ls->ls_layouts))
 300                 goto out_unlock;
 301
 302         trace_layout_recall(&ls->ls_stid.sc_stateid);
 303
 304         atomic_inc(&ls->ls_stid.sc_count);
 305         nfsd4_run_cb(&ls->ls_recall);
 306
 307 out_unlock:
 308         spin_unlock(&ls->ls_lock);
 309 }
 310
 311 static inline u64
 312 layout_end(struct nfsd4_layout_seg *seg)
 313 {
 314         u64 end = seg->offset + seg->length;
 315         return end >= seg->offset ? end : NFS4_MAX_UINT64;
 316 }
 317
 318 static void
 319 layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
 320 {
 321         if (end == NFS4_MAX_UINT64)
 322                 lo->length = NFS4_MAX_UINT64;
 323         else
 324                 lo->length = end - lo->offset;
 325 }
 326
 327 static bool
 328 layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
 329 {
 330         if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
 331                 return false;
 332         if (layout_end(&lo->lo_seg) <= s->offset)
 333                 return false;
 334         if (layout_end(s) <= lo->lo_seg.offset)
 335                 return false;
 336         return true;
 337 }
 338
 339 static bool
 340 layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
 341 {
 342         if (lo->iomode != new->iomode)
 343                 return false;
 344         if (layout_end(new) < lo->offset)
 345                 return false;
 346         if (layout_end(lo) < new->offset)
 347                 return false;
 348
 349         lo->offset = min(lo->offset, new->offset);
 350         layout_update_len(lo, max(layout_end(lo), layout_end(new)));
 351         return true;
 352 }
 353
 354 static __be32
 355 nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
 356 {
 357         struct nfs4_file *fp = ls->ls_stid.sc_file;
 358         struct nfs4_layout_stateid *l, *n;
 359         __be32 nfserr = nfs_ok;
 360
 361         assert_spin_locked(&fp->fi_lock);
 362
 363         list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
 364                 if (l != ls) {
 365                         nfsd4_recall_file_layout(l);
 366                         nfserr = nfserr_recallconflict;
 367                 }
 368         }
 369
 370         return nfserr;
 371 }
 372
 373 __be32
 374 nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
 375 {
 376         struct nfsd4_layout_seg *seg = &lgp->lg_seg;
 377         struct nfs4_file *fp = ls->ls_stid.sc_file;
 378         struct nfs4_layout *lp, *new = NULL;
 379         __be32 nfserr;
 380
 381         spin_lock(&fp->fi_lock);
 382         nfserr = nfsd4_recall_conflict(ls);
 383         if (nfserr)
 384                 goto out;
 385         spin_lock(&ls->ls_lock);
 386         list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
 387                 if (layouts_try_merge(&lp->lo_seg, seg))
 388                         goto done;
 389         }
 390         spin_unlock(&ls->ls_lock);
 391         spin_unlock(&fp->fi_lock);
 392
 393         new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
 394         if (!new)
 395                 return nfserr_jukebox;
 396         memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg));
 397         new->lo_state = ls;
 398
 399         spin_lock(&fp->fi_lock);
 400         nfserr = nfsd4_recall_conflict(ls);
 401         if (nfserr)
 402                 goto out;
 403         spin_lock(&ls->ls_lock);
 404         list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
 405                 if (layouts_try_merge(&lp->lo_seg, seg))
 406                         goto done;
 407         }
 408
 409         atomic_inc(&ls->ls_stid.sc_count);
 410         list_add_tail(&new->lo_perstate, &ls->ls_layouts);
 411         new = NULL;
 412 done:
 413         nfs4_inc_and_copy_stateid(&lgp->lg_sid, &ls->ls_stid);
 414         spin_unlock(&ls->ls_lock);
 415 out:
 416         spin_unlock(&fp->fi_lock);
 417         if (new)
 418                 kmem_cache_free(nfs4_layout_cache, new);
 419         return nfserr;
 420 }
 421
 422 static void
 423 nfsd4_free_layouts(struct list_head *reaplist)
 424 {
 425         while (!list_empty(reaplist)) {
 426                 struct nfs4_layout *lp = list_first_entry(reaplist,
 427                                 struct nfs4_layout, lo_perstate);
 428
 429                 list_del(&lp->lo_perstate);
 430                 nfs4_put_stid(&lp->lo_state->ls_stid);
 431                 kmem_cache_free(nfs4_layout_cache, lp);
 432         }
 433 }
 434
 435 static void
 436 nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
 437                 struct list_head *reaplist)
 438 {
 439         struct nfsd4_layout_seg *lo = &lp->lo_seg;
 440         u64 end = layout_end(lo);
 441
 442         if (seg->offset <= lo->offset) {
 443                 if (layout_end(seg) >= end) {
 444                         list_move_tail(&lp->lo_perstate, reaplist);
 445                         return;
 446                 }
 447                 lo->offset = layout_end(seg);
 448         } else {
 449                 /* retain the whole layout segment on a split. */
 450                 if (layout_end(seg) < end) {
 451                         dprintk("%s: split not supported\n", __func__);
 452                         return;
 453                 }
 454                 end = seg->offset;
 455         }
 456
 457         layout_update_len(lo, end);
 458 }
 459
 460 __be32
 461 nfsd4_return_file_layouts(struct svc_rqst *rqstp,
 462                 struct nfsd4_compound_state *cstate,
 463                 struct nfsd4_layoutreturn *lrp)
 464 {
 465         struct nfs4_layout_stateid *ls;
 466         struct nfs4_layout *lp, *n;
 467         LIST_HEAD(reaplist);
 468         __be32 nfserr;
 469         int found = 0;
 470
 471         nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
 472                                                 false, lrp->lr_layout_type,
 473                                                 &ls);
 474         if (nfserr) {
 475                 trace_layout_return_lookup_fail(&lrp->lr_sid);
 476                 return nfserr;
 477         }
 478
 479         spin_lock(&ls->ls_lock);
 480         list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
 481                 if (layouts_overlapping(lp, &lrp->lr_seg)) {
 482                         nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
 483                         found++;
 484                 }
 485         }
 486         if (!list_empty(&ls->ls_layouts)) {
 487                 if (found)
 488                         nfs4_inc_and_copy_stateid(&lrp->lr_sid, &ls->ls_stid);
 489                 lrp->lrs_present = 1;
 490         } else {
 491                 trace_layoutstate_unhash(&ls->ls_stid.sc_stateid);
 492                 nfs4_unhash_stid(&ls->ls_stid);
 493                 lrp->lrs_present = 0;
 494         }
 495         spin_unlock(&ls->ls_lock);
 496
 497         mutex_unlock(&ls->ls_mutex);
 498         nfs4_put_stid(&ls->ls_stid);
 499         nfsd4_free_layouts(&reaplist);
 500         return nfs_ok;
 501 }
 502
 503 __be32
 504 nfsd4_return_client_layouts(struct svc_rqst *rqstp,
 505                 struct nfsd4_compound_state *cstate,
 506                 struct nfsd4_layoutreturn *lrp)
 507 {
 508         struct nfs4_layout_stateid *ls, *n;
 509         struct nfs4_client *clp = cstate->clp;
 510         struct nfs4_layout *lp, *t;
 511         LIST_HEAD(reaplist);
 512
 513         lrp->lrs_present = 0;
 514
 515         spin_lock(&clp->cl_lock);
 516         list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
 517                 if (ls->ls_layout_type != lrp->lr_layout_type)
 518                         continue;
 519
 520                 if (lrp->lr_return_type == RETURN_FSID &&
 521                     !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
 522                                    &cstate->current_fh.fh_handle))
 523                         continue;
 524
 525                 spin_lock(&ls->ls_lock);
 526                 list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
 527                         if (lrp->lr_seg.iomode == IOMODE_ANY ||
 528                             lrp->lr_seg.iomode == lp->lo_seg.iomode)
 529                                 list_move_tail(&lp->lo_perstate, &reaplist);
 530                 }
 531                 spin_unlock(&ls->ls_lock);
 532         }
 533         spin_unlock(&clp->cl_lock);
 534
 535         nfsd4_free_layouts(&reaplist);
 536         return 0;
 537 }
 538
 539 static void
 540 nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
 541                 struct list_head *reaplist)
 542 {
 543         spin_lock(&ls->ls_lock);
 544         list_splice_init(&ls->ls_layouts, reaplist);
 545         spin_unlock(&ls->ls_lock);
 546 }
 547
 548 void
 549 nfsd4_return_all_client_layouts(struct nfs4_client *clp)
 550 {
 551         struct nfs4_layout_stateid *ls, *n;
 552         LIST_HEAD(reaplist);
 553
 554         spin_lock(&clp->cl_lock);
 555         list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
 556                 nfsd4_return_all_layouts(ls, &reaplist);
 557         spin_unlock(&clp->cl_lock);
 558
 559         nfsd4_free_layouts(&reaplist);
 560 }
 561
 562 void
 563 nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
 564 {
 565         struct nfs4_layout_stateid *ls, *n;
 566         LIST_HEAD(reaplist);
 567
 568         spin_lock(&fp->fi_lock);
 569         list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
 570                 if (ls->ls_stid.sc_client == clp)
 571                         nfsd4_return_all_layouts(ls, &reaplist);
 572         }
 573         spin_unlock(&fp->fi_lock);
 574
 575         nfsd4_free_layouts(&reaplist);
 576 }
 577
 578 static void
 579 nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
 580 {
 581         struct nfs4_client *clp = ls->ls_stid.sc_client;
 582         char addr_str[INET6_ADDRSTRLEN];
 583         static char *envp[] = {
 584                 "HOME=/",
 585                 "TERM=linux",
 586                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 587                 NULL
 588         };
 589         char *argv[8];
 590         int error;
 591
 592         rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
 593
 594         trace_layout_recall_fail(&ls->ls_stid.sc_stateid);
 595
 596         printk(KERN_WARNING
 597                 "nfsd: client %s failed to respond to layout recall. "
 598                 "  Fencing..\n", addr_str);
 599
 600         argv[0] = "/sbin/nfsd-recall-failed";
 601         argv[1] = addr_str;
 602         argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id;
 603         argv[3] = NULL;
 604
 605         error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
 606         if (error) {
 607                 printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
 608                         addr_str, error);
 609         }
 610 }
 611
 612 static void
 613 nfsd4_cb_layout_prepare(struct nfsd4_callback *cb)
 614 {
 615         struct nfs4_layout_stateid *ls =
 616                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 617
 618         mutex_lock(&ls->ls_mutex);
 619         nfs4_inc_and_copy_stateid(&ls->ls_recall_sid, &ls->ls_stid);
 620         mutex_unlock(&ls->ls_mutex);
 621 }
 622
 623 static int
 624 nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
 625 {
 626         struct nfs4_layout_stateid *ls =
 627                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 628         LIST_HEAD(reaplist);
 629
 630         switch (task->tk_status) {
 631         case 0:
 632                 return 1;
 633         case -NFS4ERR_NOMATCHING_LAYOUT:
 634                 trace_layout_recall_done(&ls->ls_stid.sc_stateid);
 635                 task->tk_status = 0;
 636                 return 1;
 637         case -NFS4ERR_DELAY:
 638                 /* Poll the client until it's done with the layout */
 639                 /* FIXME: cap number of retries.
 640                  * The pnfs standard states that we need to only expire
 641                  * the client after at-least "lease time" .eg lease-time * 2
 642                  * when failing to communicate a recall
 643                  */
 644                 rpc_delay(task, HZ/100); /* 10 mili-seconds */
 645                 return 0;
 646         default:
 647                 /*
 648                  * Unknown error or non-responding client, we'll need to fence.
 649                  */
 650                 nfsd4_cb_layout_fail(ls);
 651                 return -1;
 652         }
 653 }
 654
 655 static void
 656 nfsd4_cb_layout_release(struct nfsd4_callback *cb)
 657 {
 658         struct nfs4_layout_stateid *ls =
 659                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 660         LIST_HEAD(reaplist);
 661
 662         trace_layout_recall_release(&ls->ls_stid.sc_stateid);
 663
 664         nfsd4_return_all_layouts(ls, &reaplist);
 665         nfsd4_free_layouts(&reaplist);
 666         nfs4_put_stid(&ls->ls_stid);
 667 }
 668
 669 static struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
 670         .prepare        = nfsd4_cb_layout_prepare,
 671         .done           = nfsd4_cb_layout_done,
 672         .release        = nfsd4_cb_layout_release,
 673 };
 674
 675 static bool
 676 nfsd4_layout_lm_break(struct file_lock *fl)
 677 {
 678         /*
 679          * We don't want the locks code to timeout the lease for us;
 680          * we'll remove it ourself if a layout isn't returned
 681          * in time:
 682          */
 683         fl->fl_break_time = 0;
 684         nfsd4_recall_file_layout(fl->fl_owner);
 685         return false;
 686 }
 687
 688 static int
 689 nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
 690                 struct list_head *dispose)
 691 {
 692         BUG_ON(!(arg & F_UNLCK));
 693         return lease_modify(onlist, arg, dispose);
 694 }
 695
 696 static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
 697         .lm_break       = nfsd4_layout_lm_break,
 698         .lm_change      = nfsd4_layout_lm_change,
 699 };
 700
 701 int
 702 nfsd4_init_pnfs(void)
 703 {
 704         int i;
 705
 706         for (i = 0; i < DEVID_HASH_SIZE; i++)
 707                 INIT_LIST_HEAD(&nfsd_devid_hash[i]);
 708
 709         nfs4_layout_cache = kmem_cache_create("nfs4_layout",
 710                         sizeof(struct nfs4_layout), 0, 0, NULL);
 711         if (!nfs4_layout_cache)
 712                 return -ENOMEM;
 713
 714         nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
 715                         sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
 716         if (!nfs4_layout_stateid_cache) {
 717                 kmem_cache_destroy(nfs4_layout_cache);
 718                 return -ENOMEM;
 719         }
 720         return 0;
 721 }
 722
 723 void
 724 nfsd4_exit_pnfs(void)
 725 {
 726         int i;
 727
 728         kmem_cache_destroy(nfs4_layout_cache);
 729         kmem_cache_destroy(nfs4_layout_stateid_cache);
 730
 731         for (i = 0; i < DEVID_HASH_SIZE; i++) {
 732                 struct nfsd4_deviceid_map *map, *n;
 733
 734                 list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
 735                         kfree(map);
 736         }
 737 }