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_NOPNFS)
 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         if (!stp)
 194                 return NULL;
 195         stp->sc_free = nfsd4_free_layout_stateid;
 196         get_nfs4_file(fp);
 197         stp->sc_file = fp;
 198
 199         ls = layoutstateid(stp);
 200         INIT_LIST_HEAD(&ls->ls_perclnt);
 201         INIT_LIST_HEAD(&ls->ls_perfile);
 202         spin_lock_init(&ls->ls_lock);
 203         INIT_LIST_HEAD(&ls->ls_layouts);
 204         ls->ls_layout_type = layout_type;
 205         nfsd4_init_cb(&ls->ls_recall, clp, &nfsd4_cb_layout_ops,
 206                         NFSPROC4_CLNT_CB_LAYOUT);
 207
 208         if (parent->sc_type == NFS4_DELEG_STID)
 209                 ls->ls_file = get_file(fp->fi_deleg_file);
 210         else
 211                 ls->ls_file = find_any_file(fp);
 212         BUG_ON(!ls->ls_file);
 213
 214         if (nfsd4_layout_setlease(ls)) {
 215                 put_nfs4_file(fp);
 216                 kmem_cache_free(nfs4_layout_stateid_cache, ls);
 217                 return NULL;
 218         }
 219
 220         spin_lock(&clp->cl_lock);
 221         stp->sc_type = NFS4_LAYOUT_STID;
 222         list_add(&ls->ls_perclnt, &clp->cl_lo_states);
 223         spin_unlock(&clp->cl_lock);
 224
 225         spin_lock(&fp->fi_lock);
 226         list_add(&ls->ls_perfile, &fp->fi_lo_states);
 227         spin_unlock(&fp->fi_lock);
 228
 229         trace_layoutstate_alloc(&ls->ls_stid.sc_stateid);
 230         return ls;
 231 }
 232
 233 __be32
 234 nfsd4_preprocess_layout_stateid(struct svc_rqst *rqstp,
 235                 struct nfsd4_compound_state *cstate, stateid_t *stateid,
 236                 bool create, u32 layout_type, struct nfs4_layout_stateid **lsp)
 237 {
 238         struct nfs4_layout_stateid *ls;
 239         struct nfs4_stid *stid;
 240         unsigned char typemask = NFS4_LAYOUT_STID;
 241         __be32 status;
 242
 243         if (create)
 244                 typemask |= (NFS4_OPEN_STID | NFS4_LOCK_STID | NFS4_DELEG_STID);
 245
 246         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &stid,
 247                         net_generic(SVC_NET(rqstp), nfsd_net_id));
 248         if (status)
 249                 goto out;
 250
 251         if (!fh_match(&cstate->current_fh.fh_handle,
 252                       &stid->sc_file->fi_fhandle)) {
 253                 status = nfserr_bad_stateid;
 254                 goto out_put_stid;
 255         }
 256
 257         if (stid->sc_type != NFS4_LAYOUT_STID) {
 258                 ls = nfsd4_alloc_layout_stateid(cstate, stid, layout_type);
 259                 nfs4_put_stid(stid);
 260
 261                 status = nfserr_jukebox;
 262                 if (!ls)
 263                         goto out;
 264         } else {
 265                 ls = container_of(stid, struct nfs4_layout_stateid, ls_stid);
 266
 267                 status = nfserr_bad_stateid;
 268                 if (stateid->si_generation > stid->sc_stateid.si_generation)
 269                         goto out_put_stid;
 270                 if (layout_type != ls->ls_layout_type)
 271                         goto out_put_stid;
 272         }
 273
 274         *lsp = ls;
 275         return 0;
 276
 277 out_put_stid:
 278         nfs4_put_stid(stid);
 279 out:
 280         return status;
 281 }
 282
 283 static void
 284 nfsd4_recall_file_layout(struct nfs4_layout_stateid *ls)
 285 {
 286         spin_lock(&ls->ls_lock);
 287         if (ls->ls_recalled)
 288                 goto out_unlock;
 289
 290         ls->ls_recalled = true;
 291         atomic_inc(&ls->ls_stid.sc_file->fi_lo_recalls);
 292         if (list_empty(&ls->ls_layouts))
 293                 goto out_unlock;
 294
 295         trace_layout_recall(&ls->ls_stid.sc_stateid);
 296
 297         atomic_inc(&ls->ls_stid.sc_count);
 298         update_stateid(&ls->ls_stid.sc_stateid);
 299         memcpy(&ls->ls_recall_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t));
 300         nfsd4_run_cb(&ls->ls_recall);
 301
 302 out_unlock:
 303         spin_unlock(&ls->ls_lock);
 304 }
 305
 306 static inline u64
 307 layout_end(struct nfsd4_layout_seg *seg)
 308 {
 309         u64 end = seg->offset + seg->length;
 310         return end >= seg->offset ? end : NFS4_MAX_UINT64;
 311 }
 312
 313 static void
 314 layout_update_len(struct nfsd4_layout_seg *lo, u64 end)
 315 {
 316         if (end == NFS4_MAX_UINT64)
 317                 lo->length = NFS4_MAX_UINT64;
 318         else
 319                 lo->length = end - lo->offset;
 320 }
 321
 322 static bool
 323 layouts_overlapping(struct nfs4_layout *lo, struct nfsd4_layout_seg *s)
 324 {
 325         if (s->iomode != IOMODE_ANY && s->iomode != lo->lo_seg.iomode)
 326                 return false;
 327         if (layout_end(&lo->lo_seg) <= s->offset)
 328                 return false;
 329         if (layout_end(s) <= lo->lo_seg.offset)
 330                 return false;
 331         return true;
 332 }
 333
 334 static bool
 335 layouts_try_merge(struct nfsd4_layout_seg *lo, struct nfsd4_layout_seg *new)
 336 {
 337         if (lo->iomode != new->iomode)
 338                 return false;
 339         if (layout_end(new) < lo->offset)
 340                 return false;
 341         if (layout_end(lo) < new->offset)
 342                 return false;
 343
 344         lo->offset = min(lo->offset, new->offset);
 345         layout_update_len(lo, max(layout_end(lo), layout_end(new)));
 346         return true;
 347 }
 348
 349 static __be32
 350 nfsd4_recall_conflict(struct nfs4_layout_stateid *ls)
 351 {
 352         struct nfs4_file *fp = ls->ls_stid.sc_file;
 353         struct nfs4_layout_stateid *l, *n;
 354         __be32 nfserr = nfs_ok;
 355
 356         assert_spin_locked(&fp->fi_lock);
 357
 358         list_for_each_entry_safe(l, n, &fp->fi_lo_states, ls_perfile) {
 359                 if (l != ls) {
 360                         nfsd4_recall_file_layout(l);
 361                         nfserr = nfserr_recallconflict;
 362                 }
 363         }
 364
 365         return nfserr;
 366 }
 367
 368 __be32
 369 nfsd4_insert_layout(struct nfsd4_layoutget *lgp, struct nfs4_layout_stateid *ls)
 370 {
 371         struct nfsd4_layout_seg *seg = &lgp->lg_seg;
 372         struct nfs4_file *fp = ls->ls_stid.sc_file;
 373         struct nfs4_layout *lp, *new = NULL;
 374         __be32 nfserr;
 375
 376         spin_lock(&fp->fi_lock);
 377         nfserr = nfsd4_recall_conflict(ls);
 378         if (nfserr)
 379                 goto out;
 380         spin_lock(&ls->ls_lock);
 381         list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
 382                 if (layouts_try_merge(&lp->lo_seg, seg))
 383                         goto done;
 384         }
 385         spin_unlock(&ls->ls_lock);
 386         spin_unlock(&fp->fi_lock);
 387
 388         new = kmem_cache_alloc(nfs4_layout_cache, GFP_KERNEL);
 389         if (!new)
 390                 return nfserr_jukebox;
 391         memcpy(&new->lo_seg, seg, sizeof(lp->lo_seg));
 392         new->lo_state = ls;
 393
 394         spin_lock(&fp->fi_lock);
 395         nfserr = nfsd4_recall_conflict(ls);
 396         if (nfserr)
 397                 goto out;
 398         spin_lock(&ls->ls_lock);
 399         list_for_each_entry(lp, &ls->ls_layouts, lo_perstate) {
 400                 if (layouts_try_merge(&lp->lo_seg, seg))
 401                         goto done;
 402         }
 403
 404         atomic_inc(&ls->ls_stid.sc_count);
 405         list_add_tail(&new->lo_perstate, &ls->ls_layouts);
 406         new = NULL;
 407 done:
 408         update_stateid(&ls->ls_stid.sc_stateid);
 409         memcpy(&lgp->lg_sid, &ls->ls_stid.sc_stateid, sizeof(stateid_t));
 410         spin_unlock(&ls->ls_lock);
 411 out:
 412         spin_unlock(&fp->fi_lock);
 413         if (new)
 414                 kmem_cache_free(nfs4_layout_cache, new);
 415         return nfserr;
 416 }
 417
 418 static void
 419 nfsd4_free_layouts(struct list_head *reaplist)
 420 {
 421         while (!list_empty(reaplist)) {
 422                 struct nfs4_layout *lp = list_first_entry(reaplist,
 423                                 struct nfs4_layout, lo_perstate);
 424
 425                 list_del(&lp->lo_perstate);
 426                 nfs4_put_stid(&lp->lo_state->ls_stid);
 427                 kmem_cache_free(nfs4_layout_cache, lp);
 428         }
 429 }
 430
 431 static void
 432 nfsd4_return_file_layout(struct nfs4_layout *lp, struct nfsd4_layout_seg *seg,
 433                 struct list_head *reaplist)
 434 {
 435         struct nfsd4_layout_seg *lo = &lp->lo_seg;
 436         u64 end = layout_end(lo);
 437
 438         if (seg->offset <= lo->offset) {
 439                 if (layout_end(seg) >= end) {
 440                         list_move_tail(&lp->lo_perstate, reaplist);
 441                         return;
 442                 }
 443                 end = seg->offset;
 444         } else {
 445                 /* retain the whole layout segment on a split. */
 446                 if (layout_end(seg) < end) {
 447                         dprintk("%s: split not supported\n", __func__);
 448                         return;
 449                 }
 450
 451                 lo->offset = layout_end(seg);
 452         }
 453
 454         layout_update_len(lo, end);
 455 }
 456
 457 __be32
 458 nfsd4_return_file_layouts(struct svc_rqst *rqstp,
 459                 struct nfsd4_compound_state *cstate,
 460                 struct nfsd4_layoutreturn *lrp)
 461 {
 462         struct nfs4_layout_stateid *ls;
 463         struct nfs4_layout *lp, *n;
 464         LIST_HEAD(reaplist);
 465         __be32 nfserr;
 466         int found = 0;
 467
 468         nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lrp->lr_sid,
 469                                                 false, lrp->lr_layout_type,
 470                                                 &ls);
 471         if (nfserr) {
 472                 trace_layout_return_lookup_fail(&lrp->lr_sid);
 473                 return nfserr;
 474         }
 475
 476         spin_lock(&ls->ls_lock);
 477         list_for_each_entry_safe(lp, n, &ls->ls_layouts, lo_perstate) {
 478                 if (layouts_overlapping(lp, &lrp->lr_seg)) {
 479                         nfsd4_return_file_layout(lp, &lrp->lr_seg, &reaplist);
 480                         found++;
 481                 }
 482         }
 483         if (!list_empty(&ls->ls_layouts)) {
 484                 if (found) {
 485                         update_stateid(&ls->ls_stid.sc_stateid);
 486                         memcpy(&lrp->lr_sid, &ls->ls_stid.sc_stateid,
 487                                 sizeof(stateid_t));
 488                 }
 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         nfs4_put_stid(&ls->ls_stid);
 498         nfsd4_free_layouts(&reaplist);
 499         return nfs_ok;
 500 }
 501
 502 __be32
 503 nfsd4_return_client_layouts(struct svc_rqst *rqstp,
 504                 struct nfsd4_compound_state *cstate,
 505                 struct nfsd4_layoutreturn *lrp)
 506 {
 507         struct nfs4_layout_stateid *ls, *n;
 508         struct nfs4_client *clp = cstate->clp;
 509         struct nfs4_layout *lp, *t;
 510         LIST_HEAD(reaplist);
 511
 512         lrp->lrs_present = 0;
 513
 514         spin_lock(&clp->cl_lock);
 515         list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt) {
 516                 if (lrp->lr_return_type == RETURN_FSID &&
 517                     !fh_fsid_match(&ls->ls_stid.sc_file->fi_fhandle,
 518                                    &cstate->current_fh.fh_handle))
 519                         continue;
 520
 521                 spin_lock(&ls->ls_lock);
 522                 list_for_each_entry_safe(lp, t, &ls->ls_layouts, lo_perstate) {
 523                         if (lrp->lr_seg.iomode == IOMODE_ANY ||
 524                             lrp->lr_seg.iomode == lp->lo_seg.iomode)
 525                                 list_move_tail(&lp->lo_perstate, &reaplist);
 526                 }
 527                 spin_unlock(&ls->ls_lock);
 528         }
 529         spin_unlock(&clp->cl_lock);
 530
 531         nfsd4_free_layouts(&reaplist);
 532         return 0;
 533 }
 534
 535 static void
 536 nfsd4_return_all_layouts(struct nfs4_layout_stateid *ls,
 537                 struct list_head *reaplist)
 538 {
 539         spin_lock(&ls->ls_lock);
 540         list_splice_init(&ls->ls_layouts, reaplist);
 541         spin_unlock(&ls->ls_lock);
 542 }
 543
 544 void
 545 nfsd4_return_all_client_layouts(struct nfs4_client *clp)
 546 {
 547         struct nfs4_layout_stateid *ls, *n;
 548         LIST_HEAD(reaplist);
 549
 550         spin_lock(&clp->cl_lock);
 551         list_for_each_entry_safe(ls, n, &clp->cl_lo_states, ls_perclnt)
 552                 nfsd4_return_all_layouts(ls, &reaplist);
 553         spin_unlock(&clp->cl_lock);
 554
 555         nfsd4_free_layouts(&reaplist);
 556 }
 557
 558 void
 559 nfsd4_return_all_file_layouts(struct nfs4_client *clp, struct nfs4_file *fp)
 560 {
 561         struct nfs4_layout_stateid *ls, *n;
 562         LIST_HEAD(reaplist);
 563
 564         spin_lock(&fp->fi_lock);
 565         list_for_each_entry_safe(ls, n, &fp->fi_lo_states, ls_perfile) {
 566                 if (ls->ls_stid.sc_client == clp)
 567                         nfsd4_return_all_layouts(ls, &reaplist);
 568         }
 569         spin_unlock(&fp->fi_lock);
 570
 571         nfsd4_free_layouts(&reaplist);
 572 }
 573
 574 static void
 575 nfsd4_cb_layout_fail(struct nfs4_layout_stateid *ls)
 576 {
 577         struct nfs4_client *clp = ls->ls_stid.sc_client;
 578         char addr_str[INET6_ADDRSTRLEN];
 579         static char *envp[] = {
 580                 "HOME=/",
 581                 "TERM=linux",
 582                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
 583                 NULL
 584         };
 585         char *argv[8];
 586         int error;
 587
 588         rpc_ntop((struct sockaddr *)&clp->cl_addr, addr_str, sizeof(addr_str));
 589
 590         nfsd4_cb_layout_fail(ls);
 591
 592         printk(KERN_WARNING
 593                 "nfsd: client %s failed to respond to layout recall. "
 594                 "  Fencing..\n", addr_str);
 595
 596         argv[0] = "/sbin/nfsd-recall-failed";
 597         argv[1] = addr_str;
 598         argv[2] = ls->ls_file->f_path.mnt->mnt_sb->s_id;
 599         argv[3] = NULL;
 600
 601         error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
 602         if (error) {
 603                 printk(KERN_ERR "nfsd: fence failed for client %s: %d!\n",
 604                         addr_str, error);
 605         }
 606 }
 607
 608 static int
 609 nfsd4_cb_layout_done(struct nfsd4_callback *cb, struct rpc_task *task)
 610 {
 611         struct nfs4_layout_stateid *ls =
 612                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 613         LIST_HEAD(reaplist);
 614
 615         switch (task->tk_status) {
 616         case 0:
 617                 return 1;
 618         case -NFS4ERR_NOMATCHING_LAYOUT:
 619                 trace_layout_recall_done(&ls->ls_stid.sc_stateid);
 620                 task->tk_status = 0;
 621                 return 1;
 622         case -NFS4ERR_DELAY:
 623                 /* Poll the client until it's done with the layout */
 624                 /* FIXME: cap number of retries.
 625                  * The pnfs standard states that we need to only expire
 626                  * the client after at-least "lease time" .eg lease-time * 2
 627                  * when failing to communicate a recall
 628                  */
 629                 rpc_delay(task, HZ/100); /* 10 mili-seconds */
 630                 return 0;
 631         default:
 632                 /*
 633                  * Unknown error or non-responding client, we'll need to fence.
 634                  */
 635                 nfsd4_cb_layout_fail(ls);
 636                 return -1;
 637         }
 638 }
 639
 640 static void
 641 nfsd4_cb_layout_release(struct nfsd4_callback *cb)
 642 {
 643         struct nfs4_layout_stateid *ls =
 644                 container_of(cb, struct nfs4_layout_stateid, ls_recall);
 645         LIST_HEAD(reaplist);
 646
 647         trace_layout_recall_release(&ls->ls_stid.sc_stateid);
 648
 649         nfsd4_return_all_layouts(ls, &reaplist);
 650         nfsd4_free_layouts(&reaplist);
 651         nfs4_put_stid(&ls->ls_stid);
 652 }
 653
 654 static struct nfsd4_callback_ops nfsd4_cb_layout_ops = {
 655         .done           = nfsd4_cb_layout_done,
 656         .release        = nfsd4_cb_layout_release,
 657 };
 658
 659 static bool
 660 nfsd4_layout_lm_break(struct file_lock *fl)
 661 {
 662         /*
 663          * We don't want the locks code to timeout the lease for us;
 664          * we'll remove it ourself if a layout isn't returned
 665          * in time:
 666          */
 667         fl->fl_break_time = 0;
 668         nfsd4_recall_file_layout(fl->fl_owner);
 669         return false;
 670 }
 671
 672 static int
 673 nfsd4_layout_lm_change(struct file_lock *onlist, int arg,
 674                 struct list_head *dispose)
 675 {
 676         BUG_ON(!(arg & F_UNLCK));
 677         return lease_modify(onlist, arg, dispose);
 678 }
 679
 680 static const struct lock_manager_operations nfsd4_layouts_lm_ops = {
 681         .lm_break       = nfsd4_layout_lm_break,
 682         .lm_change      = nfsd4_layout_lm_change,
 683 };
 684
 685 int
 686 nfsd4_init_pnfs(void)
 687 {
 688         int i;
 689
 690         for (i = 0; i < DEVID_HASH_SIZE; i++)
 691                 INIT_LIST_HEAD(&nfsd_devid_hash[i]);
 692
 693         nfs4_layout_cache = kmem_cache_create("nfs4_layout",
 694                         sizeof(struct nfs4_layout), 0, 0, NULL);
 695         if (!nfs4_layout_cache)
 696                 return -ENOMEM;
 697
 698         nfs4_layout_stateid_cache = kmem_cache_create("nfs4_layout_stateid",
 699                         sizeof(struct nfs4_layout_stateid), 0, 0, NULL);
 700         if (!nfs4_layout_stateid_cache) {
 701                 kmem_cache_destroy(nfs4_layout_cache);
 702                 return -ENOMEM;
 703         }
 704         return 0;
 705 }
 706
 707 void
 708 nfsd4_exit_pnfs(void)
 709 {
 710         int i;
 711
 712         kmem_cache_destroy(nfs4_layout_cache);
 713         kmem_cache_destroy(nfs4_layout_stateid_cache);
 714
 715         for (i = 0; i < DEVID_HASH_SIZE; i++) {
 716                 struct nfsd4_deviceid_map *map, *n;
 717
 718                 list_for_each_entry_safe(map, n, &nfsd_devid_hash[i], hash)
 719                         kfree(map);
 720         }
 721 }