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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003 Hewlett-Packard Development Company LP.
  28  * Developed under the sponsorship of the US Government under
  29  * Subcontract No. B514193
  30  *
  31  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  32  * Use is subject to license terms.
  33  *
  34  * Copyright (c) 2010, 2012, Intel Corporation.
  35  */
  36 /*
  37  * This file is part of Lustre, http://www.lustre.org/
  38  * Lustre is a trademark of Sun Microsystems, Inc.
  39  */
  40
  41 /**
  42  * This file implements POSIX lock type for Lustre.
  43  * Its policy properties are start and end of extent and PID.
  44  *
  45  * These locks are only done through MDS due to POSIX semantics requiring
  46  * e.g. that locks could be only partially released and as such split into
  47  * two parts, and also that two adjacent locks from the same process may be
  48  * merged into a single wider lock.
  49  *
  50  * Lock modes are mapped like this:
  51  * PR and PW for READ and WRITE locks
  52  * NL to request a releasing of a portion of the lock
  53  *
  54  * These flock locks never timeout.
  55  */
  56
  57 #define DEBUG_SUBSYSTEM S_LDLM
  58
  59 #ifdef __KERNEL__
  60 #include <lustre_dlm.h>
  61 #include <obd_support.h>
  62 #include <obd_class.h>
  63 #include <lustre_lib.h>
  64 #include <libcfs/list.h>
  65 #else
  66 #include <liblustre.h>
  67 #include <obd_class.h>
  68 #endif
  69
  70 #include "ldlm_internal.h"
  71
  72 int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
  73                             void *data, int flag);
  74
  75 /**
  76  * list_for_remaining_safe - iterate over the remaining entries in a list
  77  *              and safeguard against removal of a list entry.
  78  * \param pos   the &struct list_head to use as a loop counter. pos MUST
  79  *              have been initialized prior to using it in this macro.
  80  * \param n     another &struct list_head to use as temporary storage
  81  * \param head  the head for your list.
  82  */
  83 #define list_for_remaining_safe(pos, n, head) \
  84         for (n = pos->next; pos != (head); pos = n, n = pos->next)
  85
  86 static inline int
  87 ldlm_same_flock_owner(struct ldlm_lock *lock, struct ldlm_lock *new)
  88 {
  89         return((new->l_policy_data.l_flock.owner ==
  90                 lock->l_policy_data.l_flock.owner) &&
  91                (new->l_export == lock->l_export));
  92 }
  93
  94 static inline int
  95 ldlm_flocks_overlap(struct ldlm_lock *lock, struct ldlm_lock *new)
  96 {
  97         return((new->l_policy_data.l_flock.start <=
  98                 lock->l_policy_data.l_flock.end) &&
  99                (new->l_policy_data.l_flock.end >=
 100                 lock->l_policy_data.l_flock.start));
 101 }
 102
 103 static inline void ldlm_flock_blocking_link(struct ldlm_lock *req,
 104                                             struct ldlm_lock *lock)
 105 {
 106         /* For server only */
 107         if (req->l_export == NULL)
 108                 return;
 109
 110         LASSERT(hlist_unhashed(&req->l_exp_flock_hash));
 111
 112         req->l_policy_data.l_flock.blocking_owner =
 113                 lock->l_policy_data.l_flock.owner;
 114         req->l_policy_data.l_flock.blocking_export =
 115                 lock->l_export;
 116         req->l_policy_data.l_flock.blocking_refs = 0;
 117
 118         cfs_hash_add(req->l_export->exp_flock_hash,
 119                      &req->l_policy_data.l_flock.owner,
 120                      &req->l_exp_flock_hash);
 121 }
 122
 123 static inline void ldlm_flock_blocking_unlink(struct ldlm_lock *req)
 124 {
 125         /* For server only */
 126         if (req->l_export == NULL)
 127                 return;
 128
 129         check_res_locked(req->l_resource);
 130         if (req->l_export->exp_flock_hash != NULL &&
 131             !hlist_unhashed(&req->l_exp_flock_hash))
 132                 cfs_hash_del(req->l_export->exp_flock_hash,
 133                              &req->l_policy_data.l_flock.owner,
 134                              &req->l_exp_flock_hash);
 135 }
 136
 137 static inline void
 138 ldlm_flock_destroy(struct ldlm_lock *lock, ldlm_mode_t mode, __u64 flags)
 139 {
 140         ENTRY;
 141
 142         LDLM_DEBUG(lock, "ldlm_flock_destroy(mode: %d, flags: "LPX64")",
 143                    mode, flags);
 144
 145         /* Safe to not lock here, since it should be empty anyway */
 146         LASSERT(hlist_unhashed(&lock->l_exp_flock_hash));
 147
 148         list_del_init(&lock->l_res_link);
 149         if (flags == LDLM_FL_WAIT_NOREPROC) {
 150                 /* client side - set a flag to prevent sending a CANCEL */
 151                 lock->l_flags |= LDLM_FL_LOCAL_ONLY | LDLM_FL_CBPENDING;
 152
 153                 /* when reaching here, it is under lock_res_and_lock(). Thus,
 154                    need call the nolock version of ldlm_lock_decref_internal*/
 155                 ldlm_lock_decref_internal_nolock(lock, mode);
 156         }
 157
 158         ldlm_lock_destroy_nolock(lock);
 159         EXIT;
 160 }
 161
 162 /**
 163  * POSIX locks deadlock detection code.
 164  *
 165  * Given a new lock \a req and an existing lock \a bl_lock it conflicts
 166  * with, we need to iterate through all blocked POSIX locks for this
 167  * export and see if there is a deadlock condition arising. (i.e. when
 168  * one client holds a lock on something and want a lock on something
 169  * else and at the same time another client has the opposite situation).
 170  */
 171
 172 struct ldlm_flock_lookup_cb_data {
 173         __u64 *bl_owner;
 174         struct ldlm_lock *lock;
 175         struct obd_export *exp;
 176 };
 177
 178 static int ldlm_flock_lookup_cb(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 179                                 struct hlist_node *hnode, void *data)
 180 {
 181         struct ldlm_flock_lookup_cb_data *cb_data = data;
 182         struct obd_export *exp = cfs_hash_object(hs, hnode);
 183         struct ldlm_lock *lock;
 184
 185         lock = cfs_hash_lookup(exp->exp_flock_hash, cb_data->bl_owner);
 186         if (lock == NULL)
 187                 return 0;
 188
 189         /* Stop on first found lock. Same process can't sleep twice */
 190         cb_data->lock = lock;
 191         cb_data->exp = class_export_get(exp);
 192
 193         return 1;
 194 }
 195
 196 static int
 197 ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)
 198 {
 199         struct obd_export *req_exp = req->l_export;
 200         struct obd_export *bl_exp = bl_lock->l_export;
 201         __u64 req_owner = req->l_policy_data.l_flock.owner;
 202         __u64 bl_owner = bl_lock->l_policy_data.l_flock.owner;
 203
 204         /* For server only */
 205         if (req_exp == NULL)
 206                 return 0;
 207
 208         class_export_get(bl_exp);
 209         while (1) {
 210                 struct ldlm_flock_lookup_cb_data cb_data = {
 211                                         .bl_owner = &bl_owner,
 212                                         .lock = NULL,
 213                                         .exp = NULL };
 214                 struct obd_export *bl_exp_new;
 215                 struct ldlm_lock *lock = NULL;
 216                 struct ldlm_flock *flock;
 217
 218                 if (bl_exp->exp_flock_hash != NULL) {
 219                         cfs_hash_for_each_key(bl_exp->exp_obd->obd_nid_hash,
 220                                 &bl_exp->exp_connection->c_peer.nid,
 221                                 ldlm_flock_lookup_cb, &cb_data);
 222                         lock = cb_data.lock;
 223                 }
 224                 if (lock == NULL)
 225                         break;
 226
 227                 class_export_put(bl_exp);
 228                 bl_exp = cb_data.exp;
 229
 230                 LASSERT(req != lock);
 231                 flock = &lock->l_policy_data.l_flock;
 232                 LASSERT(flock->owner == bl_owner);
 233                 bl_owner = flock->blocking_owner;
 234                 bl_exp_new = class_export_get(flock->blocking_export);
 235                 class_export_put(bl_exp);
 236
 237                 cfs_hash_put(bl_exp->exp_flock_hash, &lock->l_exp_flock_hash);
 238                 bl_exp = bl_exp_new;
 239
 240                 if (bl_exp->exp_failed)
 241                         break;
 242
 243                 if (bl_owner == req_owner &&
 244                     (bl_exp->exp_connection->c_peer.nid ==
 245                      req_exp->exp_connection->c_peer.nid)) {
 246                         class_export_put(bl_exp);
 247                         return 1;
 248                 }
 249         }
 250         class_export_put(bl_exp);
 251
 252         return 0;
 253 }
 254
 255 static void ldlm_flock_cancel_on_deadlock(struct ldlm_lock *lock,
 256                                           struct list_head *work_list)
 257 {
 258         CDEBUG(D_INFO, "reprocess deadlock req=%p\n", lock);
 259
 260         if ((exp_connect_flags(lock->l_export) &
 261                                 OBD_CONNECT_FLOCK_DEAD) == 0) {
 262                 CERROR("deadlock found, but client doesn't "
 263                                 "support flock canceliation\n");
 264         } else {
 265                 LASSERT(lock->l_completion_ast);
 266                 LASSERT(!ldlm_is_ast_sent(lock));
 267                 lock->l_flags |= LDLM_FL_AST_SENT | LDLM_FL_CANCEL_ON_BLOCK |
 268                         LDLM_FL_FLOCK_DEADLOCK;
 269                 ldlm_flock_blocking_unlink(lock);
 270                 ldlm_resource_unlink_lock(lock);
 271                 ldlm_add_ast_work_item(lock, NULL, work_list);
 272         }
 273 }
 274
 275 /**
 276  * Process a granting attempt for flock lock.
 277  * Must be called under ns lock held.
 278  *
 279  * This function looks for any conflicts for \a lock in the granted or
 280  * waiting queues. The lock is granted if no conflicts are found in
 281  * either queue.
 282  *
 283  * It is also responsible for splitting a lock if a portion of the lock
 284  * is released.
 285  *
 286  * If \a first_enq is 0 (ie, called from ldlm_reprocess_queue):
 287  *   - blocking ASTs have already been sent
 288  *
 289  * If \a first_enq is 1 (ie, called from ldlm_lock_enqueue):
 290  *   - blocking ASTs have not been sent yet, so list of conflicting locks
 291  *     would be collected and ASTs sent.
 292  */
 293 int
 294 ldlm_process_flock_lock(struct ldlm_lock *req, __u64 *flags, int first_enq,
 295                         ldlm_error_t *err, struct list_head *work_list)
 296 {
 297         struct ldlm_resource *res = req->l_resource;
 298         struct ldlm_namespace *ns = ldlm_res_to_ns(res);
 299         struct list_head *tmp;
 300         struct list_head *ownlocks = NULL;
 301         struct ldlm_lock *lock = NULL;
 302         struct ldlm_lock *new = req;
 303         struct ldlm_lock *new2 = NULL;
 304         ldlm_mode_t mode = req->l_req_mode;
 305         int local = ns_is_client(ns);
 306         int added = (mode == LCK_NL);
 307         int overlaps = 0;
 308         int splitted = 0;
 309         const struct ldlm_callback_suite null_cbs = { NULL };
 310         ENTRY;
 311
 312         CDEBUG(D_DLMTRACE, "flags "LPX64" owner "LPU64" pid %u mode %u start "
 313                LPU64" end "LPU64"\n", *flags,
 314                new->l_policy_data.l_flock.owner,
 315                new->l_policy_data.l_flock.pid, mode,
 316                req->l_policy_data.l_flock.start,
 317                req->l_policy_data.l_flock.end);
 318
 319         *err = ELDLM_OK;
 320
 321         if (local) {
 322                 /* No blocking ASTs are sent to the clients for
 323                  * Posix file & record locks */
 324                 req->l_blocking_ast = NULL;
 325         } else {
 326                 /* Called on the server for lock cancels. */
 327                 req->l_blocking_ast = ldlm_flock_blocking_ast;
 328         }
 329
 330 reprocess:
 331         if ((*flags == LDLM_FL_WAIT_NOREPROC) || (mode == LCK_NL)) {
 332                 /* This loop determines where this processes locks start
 333                  * in the resource lr_granted list. */
 334                 list_for_each(tmp, &res->lr_granted) {
 335                         lock = list_entry(tmp, struct ldlm_lock,
 336                                               l_res_link);
 337                         if (ldlm_same_flock_owner(lock, req)) {
 338                                 ownlocks = tmp;
 339                                 break;
 340                         }
 341                 }
 342         } else {
 343                 int reprocess_failed = 0;
 344                 lockmode_verify(mode);
 345
 346                 /* This loop determines if there are existing locks
 347                  * that conflict with the new lock request. */
 348                 list_for_each(tmp, &res->lr_granted) {
 349                         lock = list_entry(tmp, struct ldlm_lock,
 350                                               l_res_link);
 351
 352                         if (ldlm_same_flock_owner(lock, req)) {
 353                                 if (!ownlocks)
 354                                         ownlocks = tmp;
 355                                 continue;
 356                         }
 357
 358                         /* locks are compatible, overlap doesn't matter */
 359                         if (lockmode_compat(lock->l_granted_mode, mode))
 360                                 continue;
 361
 362                         if (!ldlm_flocks_overlap(lock, req))
 363                                 continue;
 364
 365                         if (!first_enq) {
 366                                 reprocess_failed = 1;
 367                                 if (ldlm_flock_deadlock(req, lock)) {
 368                                         ldlm_flock_cancel_on_deadlock(req,
 369                                                         work_list);
 370                                         RETURN(LDLM_ITER_CONTINUE);
 371                                 }
 372                                 continue;
 373                         }
 374
 375                         if (*flags & LDLM_FL_BLOCK_NOWAIT) {
 376                                 ldlm_flock_destroy(req, mode, *flags);
 377                                 *err = -EAGAIN;
 378                                 RETURN(LDLM_ITER_STOP);
 379                         }
 380
 381                         if (*flags & LDLM_FL_TEST_LOCK) {
 382                                 ldlm_flock_destroy(req, mode, *flags);
 383                                 req->l_req_mode = lock->l_granted_mode;
 384                                 req->l_policy_data.l_flock.pid =
 385                                         lock->l_policy_data.l_flock.pid;
 386                                 req->l_policy_data.l_flock.start =
 387                                         lock->l_policy_data.l_flock.start;
 388                                 req->l_policy_data.l_flock.end =
 389                                         lock->l_policy_data.l_flock.end;
 390                                 *flags |= LDLM_FL_LOCK_CHANGED;
 391                                 RETURN(LDLM_ITER_STOP);
 392                         }
 393
 394                         /* add lock to blocking list before deadlock
 395                          * check to prevent race */
 396                         ldlm_flock_blocking_link(req, lock);
 397
 398                         if (ldlm_flock_deadlock(req, lock)) {
 399                                 ldlm_flock_blocking_unlink(req);
 400                                 ldlm_flock_destroy(req, mode, *flags);
 401                                 *err = -EDEADLK;
 402                                 RETURN(LDLM_ITER_STOP);
 403                         }
 404
 405                         ldlm_resource_add_lock(res, &res->lr_waiting, req);
 406                         *flags |= LDLM_FL_BLOCK_GRANTED;
 407                         RETURN(LDLM_ITER_STOP);
 408                 }
 409                 if (reprocess_failed)
 410                         RETURN(LDLM_ITER_CONTINUE);
 411         }
 412
 413         if (*flags & LDLM_FL_TEST_LOCK) {
 414                 ldlm_flock_destroy(req, mode, *flags);
 415                 req->l_req_mode = LCK_NL;
 416                 *flags |= LDLM_FL_LOCK_CHANGED;
 417                 RETURN(LDLM_ITER_STOP);
 418         }
 419
 420         /* In case we had slept on this lock request take it off of the
 421          * deadlock detection hash list. */
 422         ldlm_flock_blocking_unlink(req);
 423
 424         /* Scan the locks owned by this process that overlap this request.
 425          * We may have to merge or split existing locks. */
 426
 427         if (!ownlocks)
 428                 ownlocks = &res->lr_granted;
 429
 430         list_for_remaining_safe(ownlocks, tmp, &res->lr_granted) {
 431                 lock = list_entry(ownlocks, struct ldlm_lock, l_res_link);
 432
 433                 if (!ldlm_same_flock_owner(lock, new))
 434                         break;
 435
 436                 if (lock->l_granted_mode == mode) {
 437                         /* If the modes are the same then we need to process
 438                          * locks that overlap OR adjoin the new lock. The extra
 439                          * logic condition is necessary to deal with arithmetic
 440                          * overflow and underflow. */
 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                 ++overlaps;
 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                 /* XXX - if ldlm_lock_new() can sleep we should
 517                  * release the lr_lock, allocate the new lock,
 518                  * and restart processing this lock. */
 519                 if (new2 == NULL) {
 520                         unlock_res_and_lock(req);
 521                         new2 = ldlm_lock_create(ns, &res->lr_name, LDLM_FLOCK,
 522                                                 lock->l_granted_mode, &null_cbs,
 523                                                 NULL, 0, LVB_T_NONE);
 524                         lock_res_and_lock(req);
 525                         if (IS_ERR(new2)) {
 526                                 ldlm_flock_destroy(req, lock->l_granted_mode,
 527                                                    *flags);
 528                                 *err = PTR_ERR(new2);
 529                                 RETURN(LDLM_ITER_STOP);
 530                         }
 531                         goto reprocess;
 532                 }
 533
 534                 splitted = 1;
 535
 536                 new2->l_granted_mode = lock->l_granted_mode;
 537                 new2->l_policy_data.l_flock.pid =
 538                         new->l_policy_data.l_flock.pid;
 539                 new2->l_policy_data.l_flock.owner =
 540                         new->l_policy_data.l_flock.owner;
 541                 new2->l_policy_data.l_flock.start =
 542                         lock->l_policy_data.l_flock.start;
 543                 new2->l_policy_data.l_flock.end =
 544                         new->l_policy_data.l_flock.start - 1;
 545                 lock->l_policy_data.l_flock.start =
 546                         new->l_policy_data.l_flock.end + 1;
 547                 new2->l_conn_export = lock->l_conn_export;
 548                 if (lock->l_export != NULL) {
 549                         new2->l_export = class_export_lock_get(lock->l_export, new2);
 550                         if (new2->l_export->exp_lock_hash &&
 551                             hlist_unhashed(&new2->l_exp_hash))
 552                                 cfs_hash_add(new2->l_export->exp_lock_hash,
 553                                              &new2->l_remote_handle,
 554                                              &new2->l_exp_hash);
 555                 }
 556                 if (*flags == LDLM_FL_WAIT_NOREPROC)
 557                         ldlm_lock_addref_internal_nolock(new2,
 558                                                          lock->l_granted_mode);
 559
 560                 /* insert new2 at lock */
 561                 ldlm_resource_add_lock(res, ownlocks, new2);
 562                 LDLM_LOCK_RELEASE(new2);
 563                 break;
 564         }
 565
 566         /* if new2 is created but never used, destroy it*/
 567         if (splitted == 0 && new2 != NULL)
 568                 ldlm_lock_destroy_nolock(new2);
 569
 570         /* At this point we're granting the lock request. */
 571         req->l_granted_mode = req->l_req_mode;
 572
 573         /* Add req to the granted queue before calling ldlm_reprocess_all(). */
 574         if (!added) {
 575                 list_del_init(&req->l_res_link);
 576                 /* insert new lock before ownlocks in list. */
 577                 ldlm_resource_add_lock(res, ownlocks, req);
 578         }
 579
 580         if (*flags != LDLM_FL_WAIT_NOREPROC) {
 581 #ifdef HAVE_SERVER_SUPPORT
 582                 if (first_enq) {
 583                         /* If this is an unlock, reprocess the waitq and
 584                          * send completions ASTs for locks that can now be
 585                          * granted. The only problem with doing this
 586                          * reprocessing here is that the completion ASTs for
 587                          * newly granted locks will be sent before the unlock
 588                          * completion is sent. It shouldn't be an issue. Also
 589                          * note that ldlm_process_flock_lock() will recurse,
 590                          * but only once because first_enq will be false from
 591                          * ldlm_reprocess_queue. */
 592                         if ((mode == LCK_NL) && overlaps) {
 593                                 struct list_head rpc_list;
 594                                 int rc;
 595
 596                                 INIT_LIST_HEAD(&rpc_list);
 597 restart:
 598                                 ldlm_reprocess_queue(res, &res->lr_waiting,
 599                                                      &rpc_list);
 600
 601                                 unlock_res_and_lock(req);
 602                                 rc = ldlm_run_ast_work(ns, &rpc_list,
 603                                                        LDLM_WORK_CP_AST);
 604                                 lock_res_and_lock(req);
 605                                 if (rc == -ERESTART)
 606                                         GOTO(restart, rc);
 607                        }
 608                 } else {
 609                         LASSERT(req->l_completion_ast);
 610                         ldlm_add_ast_work_item(req, NULL, work_list);
 611                 }
 612 #else /* !HAVE_SERVER_SUPPORT */
 613                 /* The only one possible case for client-side calls flock
 614                  * policy function is ldlm_flock_completion_ast inside which
 615                  * carries LDLM_FL_WAIT_NOREPROC flag. */
 616                 CERROR("Illegal parameter for client-side-only module.\n");
 617                 LBUG();
 618 #endif /* HAVE_SERVER_SUPPORT */
 619         }
 620
 621         /* In case we're reprocessing the requested lock we can't destroy
 622          * it until after calling ldlm_add_ast_work_item() above so that laawi()
 623          * can bump the reference count on \a req. Otherwise \a req
 624          * could be freed before the completion AST can be sent.  */
 625         if (added)
 626                 ldlm_flock_destroy(req, mode, *flags);
 627
 628         ldlm_resource_dump(D_INFO, res);
 629         RETURN(LDLM_ITER_CONTINUE);
 630 }
 631
 632 struct ldlm_flock_wait_data {
 633         struct ldlm_lock *fwd_lock;
 634         int               fwd_generation;
 635 };
 636
 637 static void
 638 ldlm_flock_interrupted_wait(void *data)
 639 {
 640         struct ldlm_lock *lock;
 641         ENTRY;
 642
 643         lock = ((struct ldlm_flock_wait_data *)data)->fwd_lock;
 644
 645         /* take lock off the deadlock detection hash list. */
 646         lock_res_and_lock(lock);
 647         ldlm_flock_blocking_unlink(lock);
 648
 649         /* client side - set flag to prevent lock from being put on LRU list */
 650         ldlm_set_cbpending(lock);
 651         unlock_res_and_lock(lock);
 652
 653         EXIT;
 654 }
 655
 656 /**
 657  * Flock completion callback function.
 658  *
 659  * \param lock [in,out]: A lock to be handled
 660  * \param flags    [in]: flags
 661  * \param *data    [in]: ldlm_work_cp_ast_lock() will use ldlm_cb_set_arg
 662  *
 663  * \retval 0    : success
 664  * \retval <0   : failure
 665  */
 666 int
 667 ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
 668 {
 669         struct file_lock                *getlk = lock->l_ast_data;
 670         struct obd_device              *obd;
 671         struct obd_import              *imp = NULL;
 672         struct ldlm_flock_wait_data     fwd;
 673         struct l_wait_info              lwi;
 674         ldlm_error_t                    err;
 675         int                             rc = 0;
 676         ENTRY;
 677
 678         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT2, 4);
 679         if (OBD_FAIL_PRECHECK(OBD_FAIL_LDLM_CP_CB_WAIT3)) {
 680                 lock_res_and_lock(lock);
 681                 lock->l_flags |= LDLM_FL_FAIL_LOC;
 682                 unlock_res_and_lock(lock);
 683                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT3, 4);
 684         }
 685         CDEBUG(D_DLMTRACE, "flags: "LPX64" data: %p getlk: %p\n",
 686                flags, data, getlk);
 687
 688         LASSERT(flags != LDLM_FL_WAIT_NOREPROC);
 689
 690         if (flags & LDLM_FL_FAILED)
 691                 goto granted;
 692
 693         if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
 694                        LDLM_FL_BLOCK_CONV))) {
 695                 if (NULL == data)
 696                         /* mds granted the lock in the reply */
 697                         goto granted;
 698                 /* CP AST RPC: lock get granted, wake it up */
 699                 wake_up(&lock->l_waitq);
 700                 RETURN(0);
 701         }
 702
 703         LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
 704                    "sleeping");
 705         fwd.fwd_lock = lock;
 706         obd = class_exp2obd(lock->l_conn_export);
 707
 708         /* if this is a local lock, there is no import */
 709         if (NULL != obd)
 710                 imp = obd->u.cli.cl_import;
 711
 712         if (NULL != imp) {
 713                 spin_lock(&imp->imp_lock);
 714                 fwd.fwd_generation = imp->imp_generation;
 715                 spin_unlock(&imp->imp_lock);
 716         }
 717
 718         lwi = LWI_TIMEOUT_INTR(0, NULL, ldlm_flock_interrupted_wait, &fwd);
 719
 720         /* Go to sleep until the lock is granted. */
 721         rc = l_wait_event(lock->l_waitq, is_granted_or_cancelled(lock), &lwi);
 722
 723         if (rc) {
 724                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
 725                            rc);
 726                 RETURN(rc);
 727         }
 728
 729 granted:
 730         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT, 10);
 731
 732         if (OBD_FAIL_PRECHECK(OBD_FAIL_LDLM_CP_CB_WAIT4)) {
 733                 lock_res_and_lock(lock);
 734                 /* DEADLOCK is always set with CBPENDING */
 735                 lock->l_flags |= LDLM_FL_FLOCK_DEADLOCK | LDLM_FL_CBPENDING;
 736                 unlock_res_and_lock(lock);
 737                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT4, 4);
 738         }
 739         if (OBD_FAIL_PRECHECK(OBD_FAIL_LDLM_CP_CB_WAIT5)) {
 740                 lock_res_and_lock(lock);
 741                 /* DEADLOCK is always set with CBPENDING */
 742                 lock->l_flags |= LDLM_FL_FAIL_LOC |
 743                                  LDLM_FL_FLOCK_DEADLOCK | LDLM_FL_CBPENDING;
 744                 unlock_res_and_lock(lock);
 745                 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT5, 4);
 746         }
 747
 748         lock_res_and_lock(lock);
 749
 750
 751         /* Protect against race where lock could have been just destroyed
 752          * due to overlap in ldlm_process_flock_lock().
 753          */
 754         if (ldlm_is_destroyed(lock)) {
 755                 unlock_res_and_lock(lock);
 756                 LDLM_DEBUG(lock, "client-side enqueue waking up: destroyed");
 757
 758                 /* An error is still to be returned, to propagate it up to
 759                  * ldlm_cli_enqueue_fini() caller. */
 760                 RETURN(-EIO);
 761         }
 762
 763         /* ldlm_lock_enqueue() has already placed lock on the granted list. */
 764         ldlm_resource_unlink_lock(lock);
 765
 766         /* Import invalidation. We need to actually release the lock
 767          * references being held, so that it can go away. No point in
 768          * holding the lock even if app still believes it has it, since
 769          * server already dropped it anyway. Only for granted locks too. */
 770         /* Do the same for DEADLOCK'ed locks. */
 771         if (ldlm_is_failed(lock) || ldlm_is_flock_deadlock(lock)) {
 772                 int mode;
 773
 774                 if (flags & LDLM_FL_TEST_LOCK)
 775                         LASSERT(ldlm_is_test_lock(lock));
 776
 777                 if (ldlm_is_test_lock(lock) || ldlm_is_flock_deadlock(lock))
 778                         mode = flock_type(getlk);
 779                 else
 780                         mode = lock->l_granted_mode;
 781
 782                 if (ldlm_is_flock_deadlock(lock)) {
 783                         LDLM_DEBUG(lock, "client-side enqueue deadlock "
 784                                    "received");
 785                         rc = -EDEADLK;
 786                 }
 787                 ldlm_flock_destroy(lock, mode, LDLM_FL_WAIT_NOREPROC);
 788                 unlock_res_and_lock(lock);
 789
 790                 /* Need to wake up the waiter if we were evicted */
 791                 wake_up(&lock->l_waitq);
 792
 793                 /* An error is still to be returned, to propagate it up to
 794                  * ldlm_cli_enqueue_fini() caller. */
 795                 RETURN(rc ? : -EIO);
 796         }
 797
 798         LDLM_DEBUG(lock, "client-side enqueue granted");
 799
 800         if (flags & LDLM_FL_TEST_LOCK) {
 801                 /* fcntl(F_GETLK) request */
 802                 /* The old mode was saved in getlk->fl_type so that if the mode
 803                  * in the lock changes we can decref the appropriate refcount.*/
 804                 LASSERT(ldlm_is_test_lock(lock));
 805                 ldlm_flock_destroy(lock, flock_type(getlk),
 806                                    LDLM_FL_WAIT_NOREPROC);
 807                 switch (lock->l_granted_mode) {
 808                 case LCK_PR:
 809                         flock_set_type(getlk, F_RDLCK);
 810                         break;
 811                 case LCK_PW:
 812                         flock_set_type(getlk, F_WRLCK);
 813                         break;
 814                 default:
 815                         flock_set_type(getlk, F_UNLCK);
 816                 }
 817                 flock_set_pid(getlk, (pid_t)lock->l_policy_data.l_flock.pid);
 818                 flock_set_start(getlk,
 819                                 (loff_t)lock->l_policy_data.l_flock.start);
 820                 flock_set_end(getlk,
 821                               (loff_t)lock->l_policy_data.l_flock.end);
 822         } else {
 823                 __u64 noreproc = LDLM_FL_WAIT_NOREPROC;
 824
 825                 /* We need to reprocess the lock to do merges or splits
 826                  * with existing locks owned by this process. */
 827                 ldlm_process_flock_lock(lock, &noreproc, 1, &err, NULL);
 828         }
 829         unlock_res_and_lock(lock);
 830         RETURN(rc);
 831 }
 832 EXPORT_SYMBOL(ldlm_flock_completion_ast);
 833
 834 int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 835                             void *data, int flag)
 836 {
 837         ENTRY;
 838
 839         LASSERT(lock);
 840         LASSERT(flag == LDLM_CB_CANCELING);
 841
 842         /* take lock off the deadlock detection hash list. */
 843         lock_res_and_lock(lock);
 844         ldlm_flock_blocking_unlink(lock);
 845         unlock_res_and_lock(lock);
 846         RETURN(0);
 847 }
 848
 849 void ldlm_flock_policy_wire18_to_local(const ldlm_wire_policy_data_t *wpolicy,
 850                                        ldlm_policy_data_t *lpolicy)
 851 {
 852         memset(lpolicy, 0, sizeof(*lpolicy));
 853         lpolicy->l_flock.start = wpolicy->l_flock.lfw_start;
 854         lpolicy->l_flock.end = wpolicy->l_flock.lfw_end;
 855         lpolicy->l_flock.pid = wpolicy->l_flock.lfw_pid;
 856         /* Compat code, old clients had no idea about owner field and
 857          * relied solely on pid for ownership. Introduced in LU-104, 2.1,
 858          * April 2011 */
 859         lpolicy->l_flock.owner = wpolicy->l_flock.lfw_pid;
 860 }
 861
 862
 863 void ldlm_flock_policy_wire21_to_local(const ldlm_wire_policy_data_t *wpolicy,
 864                                        ldlm_policy_data_t *lpolicy)
 865 {
 866         memset(lpolicy, 0, sizeof(*lpolicy));
 867         lpolicy->l_flock.start = wpolicy->l_flock.lfw_start;
 868         lpolicy->l_flock.end = wpolicy->l_flock.lfw_end;
 869         lpolicy->l_flock.pid = wpolicy->l_flock.lfw_pid;
 870         lpolicy->l_flock.owner = wpolicy->l_flock.lfw_owner;
 871 }
 872
 873 void ldlm_flock_policy_local_to_wire(const ldlm_policy_data_t *lpolicy,
 874                                      ldlm_wire_policy_data_t *wpolicy)
 875 {
 876         memset(wpolicy, 0, sizeof(*wpolicy));
 877         wpolicy->l_flock.lfw_start = lpolicy->l_flock.start;
 878         wpolicy->l_flock.lfw_end = lpolicy->l_flock.end;
 879         wpolicy->l_flock.lfw_pid = lpolicy->l_flock.pid;
 880         wpolicy->l_flock.lfw_owner = lpolicy->l_flock.owner;
 881 }
 882
 883 /*
 884  * Export handle<->flock hash operations.
 885  */
 886 static unsigned
 887 ldlm_export_flock_hash(cfs_hash_t *hs, const void *key, unsigned mask)
 888 {
 889         return cfs_hash_u64_hash(*(__u64 *)key, mask);
 890 }
 891
 892 static void *
 893 ldlm_export_flock_key(struct hlist_node *hnode)
 894 {
 895         struct ldlm_lock *lock;
 896
 897         lock = hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 898         return &lock->l_policy_data.l_flock.owner;
 899 }
 900
 901 static int
 902 ldlm_export_flock_keycmp(const void *key, struct hlist_node *hnode)
 903 {
 904         return !memcmp(ldlm_export_flock_key(hnode), key, sizeof(__u64));
 905 }
 906
 907 static void *
 908 ldlm_export_flock_object(struct hlist_node *hnode)
 909 {
 910         return hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 911 }
 912
 913 static void
 914 ldlm_export_flock_get(cfs_hash_t *hs, struct hlist_node *hnode)
 915 {
 916         struct ldlm_lock *lock;
 917         struct ldlm_flock *flock;
 918
 919         lock = hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 920         LDLM_LOCK_GET(lock);
 921
 922         flock = &lock->l_policy_data.l_flock;
 923         LASSERT(flock->blocking_export != NULL);
 924         class_export_get(flock->blocking_export);
 925         flock->blocking_refs++;
 926 }
 927
 928 static void
 929 ldlm_export_flock_put(cfs_hash_t *hs, struct hlist_node *hnode)
 930 {
 931         struct ldlm_lock *lock;
 932         struct ldlm_flock *flock;
 933
 934         lock = hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 935         LDLM_LOCK_RELEASE(lock);
 936
 937         flock = &lock->l_policy_data.l_flock;
 938         LASSERT(flock->blocking_export != NULL);
 939         class_export_put(flock->blocking_export);
 940         if (--flock->blocking_refs == 0) {
 941                 flock->blocking_owner = 0;
 942                 flock->blocking_export = NULL;
 943         }
 944 }
 945
 946 static cfs_hash_ops_t ldlm_export_flock_ops = {
 947         .hs_hash        = ldlm_export_flock_hash,
 948         .hs_key         = ldlm_export_flock_key,
 949         .hs_keycmp      = ldlm_export_flock_keycmp,
 950         .hs_object      = ldlm_export_flock_object,
 951         .hs_get         = ldlm_export_flock_get,
 952         .hs_put         = ldlm_export_flock_put,
 953         .hs_put_locked  = ldlm_export_flock_put,
 954 };
 955
 956 int ldlm_init_flock_export(struct obd_export *exp)
 957 {
 958         if( strcmp(exp->exp_obd->obd_type->typ_name, LUSTRE_MDT_NAME) != 0)
 959                 RETURN(0);
 960
 961         exp->exp_flock_hash =
 962                 cfs_hash_create(obd_uuid2str(&exp->exp_client_uuid),
 963                                 HASH_EXP_LOCK_CUR_BITS,
 964                                 HASH_EXP_LOCK_MAX_BITS,
 965                                 HASH_EXP_LOCK_BKT_BITS, 0,
 966                                 CFS_HASH_MIN_THETA, CFS_HASH_MAX_THETA,
 967                                 &ldlm_export_flock_ops,
 968                                 CFS_HASH_DEFAULT | CFS_HASH_NBLK_CHANGE);
 969         if (!exp->exp_flock_hash)
 970                 RETURN(-ENOMEM);
 971
 972         RETURN(0);
 973 }
 974 EXPORT_SYMBOL(ldlm_init_flock_export);
 975
 976 void ldlm_destroy_flock_export(struct obd_export *exp)
 977 {
 978         ENTRY;
 979         if (exp->exp_flock_hash) {
 980                 cfs_hash_putref(exp->exp_flock_hash);
 981                 exp->exp_flock_hash = NULL;
 982         }
 983         EXIT;
 984 }
 985 EXPORT_SYMBOL(ldlm_destroy_flock_export);