lustre/ldlm/ldlm_flock.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003 Hewlett-Packard Development Company LP.
  24  * Developed under the sponsorship of the US Government under
  25  * Subcontract No. B514193
  26  *
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2010, 2017, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  */
  35
  36 /**
  37  * This file implements POSIX lock type for Lustre.
  38  * Its policy properties are start and end of extent and PID.
  39  *
  40  * These locks are only done through MDS due to POSIX semantics requiring
  41  * e.g. that locks could be only partially released and as such split into
  42  * two parts, and also that two adjacent locks from the same process may be
  43  * merged into a single wider lock.
  44  *
  45  * Lock modes are mapped like this:
  46  * PR and PW for READ and WRITE locks
  47  * NL to request a releasing of a portion of the lock
  48  *
  49  * These flock locks never timeout.
  50  */
  51
  52 #define DEBUG_SUBSYSTEM S_LDLM
  53
  54 #include <linux/list.h>
  55 #ifdef HAVE_LINUX_FILELOCK_HEADER
  56 #include <linux/filelock.h>
  57 #endif
  58 #include <lustre_dlm.h>
  59 #include <obd_support.h>
  60 #include <obd_class.h>
  61 #include <lustre_lib.h>
  62
  63 #include "ldlm_internal.h"
  64
  65 int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
  66                             void *data, int flag);
  67
  68 static inline int
  69 ldlm_same_flock_owner(struct ldlm_lock *lock, struct ldlm_lock *new)
  70 {
  71         return ((new->l_policy_data.l_flock.owner ==
  72                  lock->l_policy_data.l_flock.owner) &&
  73                 (new->l_export == lock->l_export));
  74 }
  75
  76 static inline int
  77 ldlm_flocks_overlap(struct ldlm_lock *lock, struct ldlm_lock *new)
  78 {
  79         return ((new->l_policy_data.l_flock.start <=
  80                  lock->l_policy_data.l_flock.end) &&
  81                 (new->l_policy_data.l_flock.end >=
  82                  lock->l_policy_data.l_flock.start));
  83 }
  84
  85 static inline void ldlm_flock_blocking_link(struct ldlm_lock *req,
  86                                             struct ldlm_lock *lock)
  87 {
  88         /* For server only */
  89         if (req->l_export == NULL)
  90                 return;
  91
  92         LASSERT(hlist_unhashed(&req->l_exp_flock_hash));
  93
  94         req->l_policy_data.l_flock.blocking_owner =
  95                 lock->l_policy_data.l_flock.owner;
  96         req->l_policy_data.l_flock.blocking_export =
  97                 lock->l_export;
  98         atomic_set(&req->l_policy_data.l_flock.blocking_refs, 0);
  99
 100         cfs_hash_add(req->l_export->exp_flock_hash,
 101                      &req->l_policy_data.l_flock.owner,
 102                      &req->l_exp_flock_hash);
 103 }
 104
 105 static inline void ldlm_flock_blocking_unlink(struct ldlm_lock *req)
 106 {
 107         /* For server only */
 108         if (req->l_export == NULL)
 109                 return;
 110
 111         check_res_locked(req->l_resource);
 112         if (req->l_export->exp_flock_hash != NULL &&
 113             !hlist_unhashed(&req->l_exp_flock_hash))
 114                 cfs_hash_del(req->l_export->exp_flock_hash,
 115                              &req->l_policy_data.l_flock.owner,
 116                              &req->l_exp_flock_hash);
 117 }
 118
 119 static inline void
 120 ldlm_flock_destroy(struct ldlm_lock *lock, enum ldlm_mode mode, __u64 flags)
 121 {
 122         ENTRY;
 123
 124         LDLM_DEBUG(lock, "ldlm_flock_destroy(mode: %d, flags: %#llx)",
 125                    mode, flags);
 126
 127         /* Safe to not lock here, since it should be empty anyway */
 128         LASSERT(hlist_unhashed(&lock->l_exp_flock_hash));
 129
 130         list_del_init(&lock->l_res_link);
 131         if (flags == LDLM_FL_WAIT_NOREPROC) {
 132                 /* client side - set a flag to prevent sending a CANCEL */
 133                 lock->l_flags |= LDLM_FL_LOCAL_ONLY | LDLM_FL_CBPENDING;
 134
 135                 /* when reaching here, it is under lock_res_and_lock(). Thus,
 136                  * need call the nolock version of ldlm_lock_decref_internal
 137                  */
 138                 ldlm_lock_decref_internal_nolock(lock, mode);
 139         }
 140
 141         ldlm_lock_destroy_nolock(lock);
 142         EXIT;
 143 }
 144
 145 #ifdef HAVE_SERVER_SUPPORT
 146 /**
 147  * POSIX locks deadlock detection code.
 148  *
 149  * Given a new lock \a req and an existing lock \a bl_lock it conflicts
 150  * with, we need to iterate through all blocked POSIX locks for this
 151  * export and see if there is a deadlock condition arising. (i.e. when
 152  * one client holds a lock on something and want a lock on something
 153  * else and at the same time another client has the opposite situation).
 154  */
 155
 156 struct ldlm_flock_lookup_cb_data {
 157         __u64 *bl_owner;
 158         struct ldlm_lock *lock;
 159         struct obd_export *exp;
 160 };
 161
 162 static int ldlm_flock_lookup_cb(struct obd_export *exp, void *data)
 163 {
 164         struct ldlm_flock_lookup_cb_data *cb_data = data;
 165         struct ldlm_lock *lock;
 166
 167         if (exp->exp_failed)
 168                 return 0;
 169
 170         lock = cfs_hash_lookup(exp->exp_flock_hash, cb_data->bl_owner);
 171         if (lock == NULL)
 172                 return 0;
 173
 174         /* Stop on first found lock. Same process can't sleep twice */
 175         cb_data->lock = lock;
 176         cb_data->exp = class_export_get(exp);
 177
 178         return 1;
 179 }
 180
 181 static int
 182 ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)
 183 {
 184         struct obd_export *req_exp = req->l_export;
 185         struct obd_export *bl_exp = bl_lock->l_export;
 186         __u64 req_owner = req->l_policy_data.l_flock.owner;
 187         __u64 bl_owner = bl_lock->l_policy_data.l_flock.owner;
 188
 189         /* For server only */
 190         if (req_exp == NULL)
 191                 return 0;
 192
 193         class_export_get(bl_exp);
 194         while (1) {
 195                 struct ldlm_flock_lookup_cb_data cb_data = {
 196                         .bl_owner = &bl_owner,
 197                         .lock = NULL,
 198                         .exp = NULL,
 199                 };
 200                 struct ptlrpc_connection *bl_exp_conn;
 201                 struct obd_export *bl_exp_new;
 202                 struct ldlm_lock *lock = NULL;
 203                 struct ldlm_flock *flock;
 204
 205                 bl_exp_conn = bl_exp->exp_connection;
 206                 if (bl_exp->exp_flock_hash != NULL) {
 207                         int found;
 208
 209                         found = obd_nid_export_for_each(bl_exp->exp_obd,
 210                                                         &bl_exp_conn->c_peer.nid,
 211                                                         ldlm_flock_lookup_cb,
 212                                                         &cb_data);
 213                         if (found)
 214                                 lock = cb_data.lock;
 215                 }
 216                 if (lock == NULL)
 217                         break;
 218
 219                 class_export_put(bl_exp);
 220                 bl_exp = cb_data.exp;
 221
 222                 LASSERT(req != lock);
 223                 flock = &lock->l_policy_data.l_flock;
 224                 LASSERT(flock->owner == bl_owner);
 225                 bl_owner = flock->blocking_owner;
 226                 bl_exp_new = class_export_get(flock->blocking_export);
 227                 class_export_put(bl_exp);
 228
 229                 cfs_hash_put(bl_exp->exp_flock_hash, &lock->l_exp_flock_hash);
 230                 bl_exp = bl_exp_new;
 231
 232                 if (bl_exp->exp_failed)
 233                         break;
 234
 235                 if (bl_owner == req_owner &&
 236                     nid_same(&bl_exp_conn->c_peer.nid,
 237                               &req_exp->exp_connection->c_peer.nid)) {
 238                         class_export_put(bl_exp);
 239                         return 1;
 240                 }
 241         }
 242         class_export_put(bl_exp);
 243
 244         return 0;
 245 }
 246
 247 static void ldlm_flock_cancel_on_deadlock(struct ldlm_lock *lock,
 248                                           struct list_head *work_list)
 249 {
 250         CDEBUG(D_INFO, "reprocess deadlock req=%p\n", lock);
 251
 252         if ((exp_connect_flags(lock->l_export) &
 253              OBD_CONNECT_FLOCK_DEAD) == 0) {
 254                 CERROR("deadlock found, but client doesn't support flock canceliation\n");
 255         } else {
 256                 LASSERT(lock->l_completion_ast);
 257                 LASSERT(!ldlm_is_ast_sent(lock));
 258                 lock->l_flags |= (LDLM_FL_AST_SENT | LDLM_FL_CANCEL_ON_BLOCK |
 259                                   LDLM_FL_FLOCK_DEADLOCK);
 260                 ldlm_flock_blocking_unlink(lock);
 261                 ldlm_resource_unlink_lock(lock);
 262                 ldlm_add_ast_work_item(lock, NULL, work_list);
 263         }
 264 }
 265 #endif /* HAVE_SERVER_SUPPORT */
 266
 267 /**
 268  * Process a granting attempt for flock lock.
 269  * Must be called under ns lock held.
 270  *
 271  * This function looks for any conflicts for \a lock in the granted or
 272  * waiting queues. The lock is granted if no conflicts are found in
 273  * either queue.
 274  */
 275 int
 276 ldlm_process_flock_lock(struct ldlm_lock *req, __u64 *flags,
 277                         enum ldlm_process_intention intention,
 278                         enum ldlm_error *err, struct list_head *work_list)
 279 {
 280         struct ldlm_resource *res = req->l_resource;
 281         struct ldlm_namespace *ns = ldlm_res_to_ns(res);
 282         struct ldlm_lock *tmp;
 283         struct ldlm_lock *ownlocks = NULL;
 284         struct ldlm_lock *lock = NULL;
 285         struct ldlm_lock *new = req;
 286         struct ldlm_lock *new2 = NULL;
 287         enum ldlm_mode mode = req->l_req_mode;
 288         int local = ns_is_client(ns);
 289         int added = (mode == LCK_NL);
 290         int splitted = 0;
 291         const struct ldlm_callback_suite null_cbs = { NULL };
 292 #ifdef HAVE_SERVER_SUPPORT
 293         struct list_head *grant_work = (intention == LDLM_PROCESS_ENQUEUE ?
 294                                         NULL : work_list);
 295 #endif
 296
 297         ENTRY;
 298         CDEBUG(D_DLMTRACE, "flags %#llx owner %llu pid %u mode %u start "
 299                "%llu end %llu\n", *flags,
 300                new->l_policy_data.l_flock.owner,
 301                new->l_policy_data.l_flock.pid, mode,
 302                req->l_policy_data.l_flock.start,
 303                req->l_policy_data.l_flock.end);
 304
 305         *err = ELDLM_OK;
 306
 307         if (local) {
 308                 /* No blocking ASTs are sent to the clients for
 309                  * Posix file & record locks
 310                  */
 311                 req->l_blocking_ast = NULL;
 312         } else {
 313                 /* Called on the server for lock cancels. */
 314                 req->l_blocking_ast = ldlm_flock_blocking_ast;
 315         }
 316
 317 reprocess:
 318         if ((*flags == LDLM_FL_WAIT_NOREPROC) || (mode == LCK_NL)) {
 319                 /* This loop determines where this processes locks start
 320                  * in the resource lr_granted list.
 321                  */
 322                 list_for_each_entry(lock, &res->lr_granted, l_res_link) {
 323                         if (ldlm_same_flock_owner(lock, req)) {
 324                                 ownlocks = lock;
 325                                 break;
 326                         }
 327                 }
 328         }
 329 #ifdef HAVE_SERVER_SUPPORT
 330         else {
 331                 int reprocess_failed = 0;
 332                 lockmode_verify(mode);
 333
 334                 /* This loop determines if there are existing locks
 335                  * that conflict with the new lock request.
 336                  */
 337                 list_for_each_entry(lock, &res->lr_granted, l_res_link) {
 338                         if (ldlm_same_flock_owner(lock, req)) {
 339                                 if (!ownlocks)
 340                                         ownlocks = lock;
 341                                 continue;
 342                         }
 343
 344                         if (req->l_req_mode == LCK_PR &&
 345                             lock->l_granted_mode == LCK_PR &&
 346                             lock->l_policy_data.l_flock.start <=
 347                                 req->l_policy_data.l_flock.start &&
 348                             lock->l_policy_data.l_flock.end >=
 349                                 req->l_policy_data.l_flock.end) {
 350                                 /* there can't be granted WR lock */
 351                                 break;
 352                         }
 353                         /* locks are compatible, overlap doesn't matter */
 354                         if (lockmode_compat(lock->l_granted_mode, mode))
 355                                 continue;
 356
 357                         if (!ldlm_flocks_overlap(lock, req))
 358                                 continue;
 359
 360                         if (intention != LDLM_PROCESS_ENQUEUE) {
 361                                 if (ldlm_flock_deadlock(req, lock)) {
 362                                         ldlm_flock_cancel_on_deadlock(
 363                                                 req, grant_work);
 364                                         RETURN(LDLM_ITER_CONTINUE);
 365                                 }
 366                                 reprocess_failed = 1;
 367                                 break;
 368                         }
 369
 370                         if (*flags & LDLM_FL_BLOCK_NOWAIT) {
 371                                 ldlm_flock_destroy(req, mode, *flags);
 372                                 *err = -EAGAIN;
 373                                 RETURN(LDLM_ITER_STOP);
 374                         }
 375
 376                         if (*flags & LDLM_FL_TEST_LOCK) {
 377                                 ldlm_flock_destroy(req, mode, *flags);
 378                                 req->l_req_mode = lock->l_granted_mode;
 379                                 req->l_policy_data.l_flock.pid =
 380                                         lock->l_policy_data.l_flock.pid;
 381                                 req->l_policy_data.l_flock.start =
 382                                         lock->l_policy_data.l_flock.start;
 383                                 req->l_policy_data.l_flock.end =
 384                                         lock->l_policy_data.l_flock.end;
 385                                 *flags |= LDLM_FL_LOCK_CHANGED;
 386                                 RETURN(LDLM_ITER_STOP);
 387                         }
 388
 389                         /* add lock to blocking list before deadlock
 390                          * check to prevent race
 391                          */
 392                         ldlm_flock_blocking_link(req, lock);
 393
 394                         if (ldlm_flock_deadlock(req, lock)) {
 395                                 ldlm_flock_blocking_unlink(req);
 396                                 ldlm_flock_destroy(req, mode, *flags);
 397                                 *err = -EDEADLK;
 398                                 RETURN(LDLM_ITER_STOP);
 399                         }
 400
 401                         ldlm_resource_add_lock(res, &res->lr_waiting, req);
 402                         *flags |= LDLM_FL_BLOCK_GRANTED;
 403                         RETURN(LDLM_ITER_STOP);
 404                 }
 405                 if (reprocess_failed)
 406                         RETURN(LDLM_ITER_CONTINUE);
 407         }
 408
 409         if (*flags & LDLM_FL_TEST_LOCK) {
 410                 ldlm_flock_destroy(req, mode, *flags);
 411                 req->l_req_mode = LCK_NL;
 412                 *flags |= LDLM_FL_LOCK_CHANGED;
 413                 RETURN(LDLM_ITER_STOP);
 414         }
 415
 416         /* In case we had slept on this lock request take it off of the
 417          * deadlock detection hash list.
 418          */
 419         ldlm_flock_blocking_unlink(req);
 420 #endif /* HAVE_SERVER_SUPPORT */
 421
 422         /* Scan the locks owned by this process to find the insertion point
 423          * (as locks are ordered), and to handle overlaps.
 424          * We may have to merge or split existing locks.
 425          */
 426         if (ownlocks)
 427                 lock = ownlocks;
 428         else
 429                 lock = list_entry(&res->lr_granted,
 430                                   struct ldlm_lock, l_res_link);
 431         list_for_each_entry_safe_from(lock, tmp, &res->lr_granted, l_res_link) {
 432                 if (!ldlm_same_flock_owner(lock, new))
 433                         break;
 434
 435                 if (lock->l_granted_mode == mode) {
 436                         /* If the modes are the same then we need to process
 437                          * locks that overlap OR adjoin the new lock. The extra
 438                          * logic condition is necessary to deal with arithmetic
 439                          * overflow and underflow.
 440                          */
 441                         if ((new->l_policy_data.l_flock.start >
 442                              (lock->l_policy_data.l_flock.end + 1))
 443                             && (lock->l_policy_data.l_flock.end !=
 444                                 OBD_OBJECT_EOF))
 445                                 continue;
 446
 447                         if ((new->l_policy_data.l_flock.end <
 448                              (lock->l_policy_data.l_flock.start - 1))
 449                             && (lock->l_policy_data.l_flock.start != 0))
 450                                 break;
 451
 452                         if (new->l_policy_data.l_flock.start <
 453                             lock->l_policy_data.l_flock.start) {
 454                                 lock->l_policy_data.l_flock.start =
 455                                         new->l_policy_data.l_flock.start;
 456                         } else {
 457                                 new->l_policy_data.l_flock.start =
 458                                         lock->l_policy_data.l_flock.start;
 459                         }
 460
 461                         if (new->l_policy_data.l_flock.end >
 462                             lock->l_policy_data.l_flock.end) {
 463                                 lock->l_policy_data.l_flock.end =
 464                                         new->l_policy_data.l_flock.end;
 465                         } else {
 466                                 new->l_policy_data.l_flock.end =
 467                                         lock->l_policy_data.l_flock.end;
 468                         }
 469
 470                         if (added) {
 471                                 ldlm_flock_destroy(lock, mode, *flags);
 472                         } else {
 473                                 new = lock;
 474                                 added = 1;
 475                         }
 476                         continue;
 477                 }
 478
 479                 if (new->l_policy_data.l_flock.start >
 480                     lock->l_policy_data.l_flock.end)
 481                         continue;
 482
 483                 if (new->l_policy_data.l_flock.end <
 484                     lock->l_policy_data.l_flock.start)
 485                         break;
 486
 487                 res->lr_flock_node.lfn_needs_reprocess = true;
 488
 489                 if (new->l_policy_data.l_flock.start <=
 490                     lock->l_policy_data.l_flock.start) {
 491                         if (new->l_policy_data.l_flock.end <
 492                             lock->l_policy_data.l_flock.end) {
 493                                 lock->l_policy_data.l_flock.start =
 494                                         new->l_policy_data.l_flock.end + 1;
 495                                 break;
 496                         }
 497                         ldlm_flock_destroy(lock, lock->l_req_mode, *flags);
 498                         continue;
 499                 }
 500                 if (new->l_policy_data.l_flock.end >=
 501                     lock->l_policy_data.l_flock.end) {
 502                         lock->l_policy_data.l_flock.end =
 503                                 new->l_policy_data.l_flock.start - 1;
 504                         continue;
 505                 }
 506
 507                 /* split the existing lock into two locks */
 508
 509                 /* if this is an F_UNLCK operation then we could avoid
 510                  * allocating a new lock and use the req lock passed in
 511                  * with the request but this would complicate the reply
 512                  * processing since updates to req get reflected in the
 513                  * reply. The client side replays the lock request so
 514                  * it must see the original lock data in the reply.
 515                  */
 516
 517                 /* XXX - if ldlm_lock_new() can sleep we should
 518                  * release the lr_lock, allocate the new lock,
 519                  * and restart processing this lock.
 520                  */
 521                 if (new2 == NULL) {
 522                         unlock_res_and_lock(req);
 523                         new2 = ldlm_lock_create(ns, &res->lr_name, LDLM_FLOCK,
 524                                                 lock->l_granted_mode, &null_cbs,
 525                                                 NULL, 0, LVB_T_NONE);
 526                         lock_res_and_lock(req);
 527                         if (IS_ERR(new2)) {
 528                                 ldlm_flock_destroy(req, lock->l_granted_mode,
 529                                                    *flags);
 530                                 *err = PTR_ERR(new2);
 531                                 RETURN(LDLM_ITER_STOP);
 532                         }
 533                         goto reprocess;
 534                 }
 535
 536                 splitted = 1;
 537
 538                 new2->l_granted_mode = lock->l_granted_mode;
 539                 new2->l_policy_data.l_flock.pid =
 540                         new->l_policy_data.l_flock.pid;
 541                 new2->l_policy_data.l_flock.owner =
 542                         new->l_policy_data.l_flock.owner;
 543                 new2->l_policy_data.l_flock.start =
 544                         lock->l_policy_data.l_flock.start;
 545                 new2->l_policy_data.l_flock.end =
 546                         new->l_policy_data.l_flock.start - 1;
 547                 lock->l_policy_data.l_flock.start =
 548                         new->l_policy_data.l_flock.end + 1;
 549                 new2->l_conn_export = lock->l_conn_export;
 550                 if (lock->l_export != NULL) {
 551                         new2->l_export = class_export_lock_get(lock->l_export,
 552                                                                new2);
 553                         if (new2->l_export->exp_lock_hash &&
 554                             hlist_unhashed(&new2->l_exp_hash))
 555                                 cfs_hash_add(new2->l_export->exp_lock_hash,
 556                                              &new2->l_remote_handle,
 557                                              &new2->l_exp_hash);
 558                 }
 559                 if (*flags == LDLM_FL_WAIT_NOREPROC)
 560                         ldlm_lock_addref_internal_nolock(new2,
 561                                                          lock->l_granted_mode);
 562
 563                 /* insert new2 at lock */
 564                 ldlm_resource_add_lock(res, &lock->l_res_link, new2);
 565                 LDLM_LOCK_RELEASE(new2);
 566                 break;
 567         }
 568
 569         /* if new2 is created but never used, destroy it*/
 570         if (splitted == 0 && new2 != NULL)
 571                 ldlm_lock_destroy_nolock(new2);
 572
 573         /* At this point we're granting the lock request. */
 574         req->l_granted_mode = req->l_req_mode;
 575
 576         /* Add req to the granted queue before calling ldlm_reprocess_all(). */
 577         if (!added) {
 578                 list_del_init(&req->l_res_link);
 579                 /* insert new lock before "lock", which might be the
 580                  * next lock for this owner, or might be the first
 581                  * lock for the next owner, or might not be a lock at
 582                  * all, but instead points at the head of the list
 583                  */
 584                 ldlm_resource_add_lock(res, &lock->l_res_link, req);
 585         }
 586
 587         if (*flags != LDLM_FL_WAIT_NOREPROC) {
 588 #ifdef HAVE_SERVER_SUPPORT
 589                 if (intention == LDLM_PROCESS_ENQUEUE) {
 590                         /* If this is an unlock, reprocess the waitq and
 591                          * send completions ASTs for locks that can now be
 592                          * granted. The only problem with doing this
 593                          * reprocessing here is that the completion ASTs for
 594                          * newly granted locks will be sent before the unlock
 595                          * completion is sent. It shouldn't be an issue. Also
 596                          * note that ldlm_process_flock_lock() will recurse,
 597                          * but only once because 'intention' won't be
 598                          * LDLM_PROCESS_ENQUEUE from ldlm_reprocess_queue.
 599                          */
 600                         struct ldlm_flock_node *fn = &res->lr_flock_node;
 601 restart:
 602                         if (mode == LCK_NL && fn->lfn_needs_reprocess &&
 603                             atomic_read(&fn->lfn_unlock_pending) == 0) {
 604                                 LIST_HEAD(rpc_list);
 605                                 int rc;
 606
 607                                 ldlm_reprocess_queue(res, &res->lr_waiting,
 608                                                      &rpc_list,
 609                                                      LDLM_PROCESS_RESCAN, 0);
 610                                 fn->lfn_needs_reprocess = false;
 611                                 unlock_res_and_lock(req);
 612                                 rc = ldlm_run_ast_work(ns, &rpc_list,
 613                                                        LDLM_WORK_CP_AST);
 614                                 lock_res_and_lock(req);
 615                                 if (rc == -ERESTART) {
 616                                         fn->lfn_needs_reprocess = true;
 617                                         GOTO(restart, rc);
 618                                 }
 619                         }
 620                 } else {
 621                         LASSERT(req->l_completion_ast);
 622                         ldlm_add_ast_work_item(req, NULL, grant_work);
 623                 }
 624 #else /* !HAVE_SERVER_SUPPORT */
 625                 /* The only one possible case for client-side calls flock
 626                  * policy function is ldlm_flock_completion_ast inside which
 627                  * carries LDLM_FL_WAIT_NOREPROC flag.
 628                  */
 629                 CERROR("Illegal parameter for client-side-only module.\n");
 630                 LBUG();
 631 #endif /* HAVE_SERVER_SUPPORT */
 632         }
 633
 634         /* In case we're reprocessing the requested lock we can't destroy
 635          * it until after calling ldlm_add_ast_work_item() above so that laawi()
 636          * can bump the reference count on \a req. Otherwise \a req
 637          * could be freed before the completion AST can be sent.
 638          */
 639         if (added)
 640                 ldlm_flock_destroy(req, mode, *flags);
 641
 642         ldlm_resource_dump(D_INFO, res);
 643         RETURN(LDLM_ITER_CONTINUE);
 644 }
 645
 646 /**
 647  * Flock completion callback function.
 648  *
 649  * \param lock [in,out]: A lock to be handled
 650  * \param flags    [in]: flags
 651  * \param *data    [in]: ldlm_work_cp_ast_lock() will use ldlm_cb_set_arg
 652  *
 653  * \retval 0    : success
 654  * \retval <0   : failure
 655  */
 656 int
 657 ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
 658 {
 659         struct file_lock *getlk = lock->l_ast_data;
 660         struct obd_device *obd;
 661         enum ldlm_error err;
 662         int rc = 0;
 663         ENTRY;
 664
 665         CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT2, 4);
 666         if (CFS_FAIL_PRECHECK(OBD_FAIL_LDLM_CP_CB_WAIT3)) {
 667                 lock_res_and_lock(lock);
 668                 lock->l_flags |= LDLM_FL_FAIL_LOC;
 669                 unlock_res_and_lock(lock);
 670                 CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT3, 4);
 671         }
 672         CDEBUG(D_DLMTRACE, "flags: %#llx data: %p getlk: %p\n",
 673                flags, data, getlk);
 674
 675         LASSERT(flags != LDLM_FL_WAIT_NOREPROC);
 676
 677         if (flags & LDLM_FL_FAILED)
 678                 goto granted;
 679
 680         if (!(flags & LDLM_FL_BLOCKED_MASK)) {
 681                 if (NULL == data)
 682                         /* mds granted the lock in the reply */
 683                         goto granted;
 684                 /* CP AST RPC: lock get granted, wake it up */
 685                 wake_up(&lock->l_waitq);
 686                 RETURN(0);
 687         }
 688
 689         LDLM_DEBUG(lock,
 690                    "client-side enqueue returned a blocked lock, sleeping");
 691         obd = class_exp2obd(lock->l_conn_export);
 692
 693         /* Go to sleep until the lock is granted. */
 694         rc = l_wait_event_abortable(lock->l_waitq,
 695                                     is_granted_or_cancelled(lock));
 696         if (rc < 0) {
 697                 /* take lock off the deadlock detection hash list. */
 698                 lock_res_and_lock(lock);
 699                 ldlm_flock_blocking_unlink(lock);
 700
 701                 /* client side - set flag to prevent lock from being
 702                  * put on LRU list
 703                  */
 704                 ldlm_set_cbpending(lock);
 705                 unlock_res_and_lock(lock);
 706
 707                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
 708                            rc);
 709                 RETURN(rc);
 710         }
 711
 712 granted:
 713         CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT, 10);
 714
 715         if (CFS_FAIL_PRECHECK(OBD_FAIL_LDLM_CP_CB_WAIT4)) {
 716                 lock_res_and_lock(lock);
 717                 /* DEADLOCK is always set with CBPENDING */
 718                 lock->l_flags |= LDLM_FL_FLOCK_DEADLOCK | LDLM_FL_CBPENDING;
 719                 unlock_res_and_lock(lock);
 720                 CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT4, 4);
 721         }
 722         if (CFS_FAIL_PRECHECK(OBD_FAIL_LDLM_CP_CB_WAIT5)) {
 723                 lock_res_and_lock(lock);
 724                 /* DEADLOCK is always set with CBPENDING */
 725                 lock->l_flags |= (LDLM_FL_FAIL_LOC |
 726                                   LDLM_FL_FLOCK_DEADLOCK | LDLM_FL_CBPENDING);
 727                 unlock_res_and_lock(lock);
 728                 CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT5, 4);
 729         }
 730
 731         lock_res_and_lock(lock);
 732
 733
 734         /* Protect against race where lock could have been just destroyed
 735          * due to overlap in ldlm_process_flock_lock().
 736          */
 737         if (ldlm_is_destroyed(lock)) {
 738                 unlock_res_and_lock(lock);
 739                 LDLM_DEBUG(lock, "client-side enqueue waking up: destroyed");
 740
 741                 /* An error is still to be returned, to propagate it up to
 742                  * ldlm_cli_enqueue_fini() caller. */
 743                 RETURN(-EIO);
 744         }
 745
 746         /* ldlm_lock_enqueue() has already placed lock on the granted list. */
 747         ldlm_resource_unlink_lock(lock);
 748
 749         /* Import invalidation. We need to actually release the lock
 750          * references being held, so that it can go away. No point in
 751          * holding the lock even if app still believes it has it, since
 752          * server already dropped it anyway. Only for granted locks too.
 753          */
 754         /* Do the same for DEADLOCK'ed locks. */
 755         if (ldlm_is_failed(lock) || ldlm_is_flock_deadlock(lock)) {
 756                 int mode;
 757
 758                 if (flags & LDLM_FL_TEST_LOCK)
 759                         LASSERT(ldlm_is_test_lock(lock));
 760
 761                 if (ldlm_is_test_lock(lock) || ldlm_is_flock_deadlock(lock))
 762                         mode = getlk->fl_type;
 763                 else
 764                         mode = lock->l_req_mode;
 765
 766                 if (ldlm_is_flock_deadlock(lock)) {
 767                         LDLM_DEBUG(lock, "client-side enqueue deadlock "
 768                                    "received");
 769                         rc = -EDEADLK;
 770                 }
 771                 ldlm_flock_destroy(lock, mode, LDLM_FL_WAIT_NOREPROC);
 772                 unlock_res_and_lock(lock);
 773
 774                 /* Need to wake up the waiter if we were evicted */
 775                 wake_up(&lock->l_waitq);
 776
 777                 /* An error is still to be returned, to propagate it up to
 778                  * ldlm_cli_enqueue_fini() caller.
 779                  */
 780                 RETURN(rc ? : -EIO);
 781         }
 782
 783         LDLM_DEBUG(lock, "client-side enqueue granted");
 784
 785         if (flags & LDLM_FL_TEST_LOCK) {
 786                 /*
 787                  * fcntl(F_GETLK) request
 788                  * The old mode was saved in getlk->fl_type so that if the mode
 789                  * in the lock changes we can decref the appropriate refcount.
 790                  */
 791                 LASSERT(ldlm_is_test_lock(lock));
 792                 ldlm_flock_destroy(lock, getlk->fl_type, LDLM_FL_WAIT_NOREPROC);
 793                 switch (lock->l_granted_mode) {
 794                 case LCK_PR:
 795                         getlk->fl_type = F_RDLCK;
 796                         break;
 797                 case LCK_PW:
 798                         getlk->fl_type = F_WRLCK;
 799                         break;
 800                 default:
 801                         getlk->fl_type = F_UNLCK;
 802                 }
 803                 getlk->fl_pid = (pid_t)lock->l_policy_data.l_flock.pid;
 804                 getlk->fl_start = (loff_t)lock->l_policy_data.l_flock.start;
 805                 getlk->fl_end = (loff_t)lock->l_policy_data.l_flock.end;
 806         } else {
 807                 __u64 noreproc = LDLM_FL_WAIT_NOREPROC;
 808
 809                 /* We need to reprocess the lock to do merges or splits
 810                  * with existing locks owned by this process.
 811                  */
 812                 ldlm_process_flock_lock(lock, &noreproc, 1, &err, NULL);
 813         }
 814         unlock_res_and_lock(lock);
 815         RETURN(rc);
 816 }
 817 EXPORT_SYMBOL(ldlm_flock_completion_ast);
 818
 819 int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 820                             void *data, int flag)
 821 {
 822         ENTRY;
 823
 824         LASSERT(lock);
 825         LASSERT(flag == LDLM_CB_CANCELING);
 826
 827         /* take lock off the deadlock detection hash list. */
 828         lock_res_and_lock(lock);
 829         ldlm_flock_blocking_unlink(lock);
 830         unlock_res_and_lock(lock);
 831         RETURN(0);
 832 }
 833
 834 void ldlm_flock_policy_wire_to_local(const union ldlm_wire_policy_data *wpolicy,
 835                                      union ldlm_policy_data *lpolicy)
 836 {
 837         lpolicy->l_flock.start = wpolicy->l_flock.lfw_start;
 838         lpolicy->l_flock.end = wpolicy->l_flock.lfw_end;
 839         lpolicy->l_flock.pid = wpolicy->l_flock.lfw_pid;
 840         lpolicy->l_flock.owner = wpolicy->l_flock.lfw_owner;
 841 }
 842
 843 void ldlm_flock_policy_local_to_wire(const union ldlm_policy_data *lpolicy,
 844                                      union ldlm_wire_policy_data *wpolicy)
 845 {
 846         memset(wpolicy, 0, sizeof(*wpolicy));
 847         wpolicy->l_flock.lfw_start = lpolicy->l_flock.start;
 848         wpolicy->l_flock.lfw_end = lpolicy->l_flock.end;
 849         wpolicy->l_flock.lfw_pid = lpolicy->l_flock.pid;
 850         wpolicy->l_flock.lfw_owner = lpolicy->l_flock.owner;
 851 }
 852
 853 /*
 854  * Export handle<->flock hash operations.
 855  */
 856 static unsigned int
 857 ldlm_export_flock_hash(struct cfs_hash *hs, const void *key,
 858                        const unsigned int bits)
 859 {
 860         return cfs_hash_64(*(__u64 *)key, bits);
 861 }
 862
 863 static void *
 864 ldlm_export_flock_key(struct hlist_node *hnode)
 865 {
 866         struct ldlm_lock *lock;
 867
 868         lock = hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 869         return &lock->l_policy_data.l_flock.owner;
 870 }
 871
 872 static int
 873 ldlm_export_flock_keycmp(const void *key, struct hlist_node *hnode)
 874 {
 875         return !memcmp(ldlm_export_flock_key(hnode), key, sizeof(__u64));
 876 }
 877
 878 static void *
 879 ldlm_export_flock_object(struct hlist_node *hnode)
 880 {
 881         return hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 882 }
 883
 884 static void
 885 ldlm_export_flock_get(struct cfs_hash *hs, struct hlist_node *hnode)
 886 {
 887         struct ldlm_lock *lock;
 888         struct ldlm_flock *flock;
 889
 890         lock = hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 891         LDLM_LOCK_GET(lock);
 892
 893         flock = &lock->l_policy_data.l_flock;
 894         LASSERT(flock->blocking_export != NULL);
 895         class_export_get(flock->blocking_export);
 896         atomic_inc(&flock->blocking_refs);
 897 }
 898
 899 static void
 900 ldlm_export_flock_put(struct cfs_hash *hs, struct hlist_node *hnode)
 901 {
 902         struct ldlm_lock *lock;
 903         struct ldlm_flock *flock;
 904
 905         lock = hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 906
 907         flock = &lock->l_policy_data.l_flock;
 908         LASSERT(flock->blocking_export != NULL);
 909         class_export_put(flock->blocking_export);
 910         if (atomic_dec_and_test(&flock->blocking_refs)) {
 911                 flock->blocking_owner = 0;
 912                 flock->blocking_export = NULL;
 913         }
 914         LDLM_LOCK_RELEASE(lock);
 915 }
 916
 917 static struct cfs_hash_ops ldlm_export_flock_ops = {
 918         .hs_hash        = ldlm_export_flock_hash,
 919         .hs_key         = ldlm_export_flock_key,
 920         .hs_keycmp      = ldlm_export_flock_keycmp,
 921         .hs_object      = ldlm_export_flock_object,
 922         .hs_get         = ldlm_export_flock_get,
 923         .hs_put         = ldlm_export_flock_put,
 924         .hs_put_locked  = ldlm_export_flock_put,
 925 };
 926
 927 int ldlm_init_flock_export(struct obd_export *exp)
 928 {
 929         if( strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_MDT_NAME) != 0)
 930                 RETURN(0);
 931
 932         exp->exp_flock_hash =
 933                 cfs_hash_create(obd_uuid2str(&exp->exp_client_uuid),
 934                                 HASH_EXP_LOCK_CUR_BITS,
 935                                 HASH_EXP_LOCK_MAX_BITS,
 936                                 HASH_EXP_LOCK_BKT_BITS, 0,
 937                                 CFS_HASH_MIN_THETA, CFS_HASH_MAX_THETA,
 938                                 &ldlm_export_flock_ops,
 939                                 CFS_HASH_DEFAULT | CFS_HASH_NBLK_CHANGE);
 940         if (!exp->exp_flock_hash)
 941                 RETURN(-ENOMEM);
 942
 943         RETURN(0);
 944 }
 945
 946 void ldlm_destroy_flock_export(struct obd_export *exp)
 947 {
 948         ENTRY;
 949         if (exp->exp_flock_hash) {
 950                 cfs_hash_putref(exp->exp_flock_hash);
 951                 exp->exp_flock_hash = NULL;
 952         }
 953         EXIT;
 954 }