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 int ldlm_flock_blocking_link(struct ldlm_lock *req,
 104                                            struct ldlm_lock *lock)
 105 {
 106         int rc = 0;
 107
 108         /* For server only */
 109         if (req->l_export == NULL)
 110                 return 0;
 111
 112         if (unlikely(req->l_export->exp_flock_hash == NULL)) {
 113                 rc = ldlm_init_flock_export(req->l_export);
 114                 if (rc)
 115                         goto error;
 116         }
 117
 118         LASSERT(cfs_hlist_unhashed(&req->l_exp_flock_hash));
 119
 120         req->l_policy_data.l_flock.blocking_owner =
 121                 lock->l_policy_data.l_flock.owner;
 122         req->l_policy_data.l_flock.blocking_export =
 123                 lock->l_export;
 124         req->l_policy_data.l_flock.blocking_refs = 0;
 125
 126         cfs_hash_add(req->l_export->exp_flock_hash,
 127                      &req->l_policy_data.l_flock.owner,
 128                      &req->l_exp_flock_hash);
 129 error:
 130         return rc;
 131 }
 132
 133 static inline void ldlm_flock_blocking_unlink(struct ldlm_lock *req)
 134 {
 135         /* For server only */
 136         if (req->l_export == NULL)
 137                 return;
 138
 139         check_res_locked(req->l_resource);
 140         if (req->l_export->exp_flock_hash != NULL &&
 141             !cfs_hlist_unhashed(&req->l_exp_flock_hash))
 142                 cfs_hash_del(req->l_export->exp_flock_hash,
 143                              &req->l_policy_data.l_flock.owner,
 144                              &req->l_exp_flock_hash);
 145 }
 146
 147 static inline void
 148 ldlm_flock_destroy(struct ldlm_lock *lock, ldlm_mode_t mode, __u64 flags)
 149 {
 150         ENTRY;
 151
 152         LDLM_DEBUG(lock, "ldlm_flock_destroy(mode: %d, flags: 0x%llx)",
 153                    mode, flags);
 154
 155         /* Safe to not lock here, since it should be empty anyway */
 156         LASSERT(cfs_hlist_unhashed(&lock->l_exp_flock_hash));
 157
 158         cfs_list_del_init(&lock->l_res_link);
 159         if (flags == LDLM_FL_WAIT_NOREPROC &&
 160             !(lock->l_flags & LDLM_FL_FAILED)) {
 161                 /* client side - set a flag to prevent sending a CANCEL */
 162                 lock->l_flags |= LDLM_FL_LOCAL_ONLY | LDLM_FL_CBPENDING;
 163
 164                 /* when reaching here, it is under lock_res_and_lock(). Thus,
 165                    need call the nolock version of ldlm_lock_decref_internal*/
 166                 ldlm_lock_decref_internal_nolock(lock, mode);
 167         }
 168
 169         ldlm_lock_destroy_nolock(lock);
 170         EXIT;
 171 }
 172
 173 /**
 174  * POSIX locks deadlock detection code.
 175  *
 176  * Given a new lock \a req and an existing lock \a bl_lock it conflicts
 177  * with, we need to iterate through all blocked POSIX locks for this
 178  * export and see if there is a deadlock condition arising. (i.e. when
 179  * one client holds a lock on something and want a lock on something
 180  * else and at the same time another client has the opposite situation).
 181  */
 182 static int
 183 ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)
 184 {
 185         struct obd_export *req_exp = req->l_export;
 186         struct obd_export *bl_exp = bl_lock->l_export;
 187         __u64 req_owner = req->l_policy_data.l_flock.owner;
 188         __u64 bl_owner = bl_lock->l_policy_data.l_flock.owner;
 189
 190         /* For server only */
 191         if (req_exp == NULL)
 192                 return 0;
 193
 194         class_export_get(bl_exp);
 195         while (1) {
 196                 struct obd_export *bl_exp_new;
 197                 struct ldlm_lock *lock = NULL;
 198                 struct ldlm_flock *flock;
 199
 200                 if (bl_exp->exp_flock_hash != NULL)
 201                         lock = cfs_hash_lookup(bl_exp->exp_flock_hash,
 202                                                &bl_owner);
 203                 if (lock == NULL)
 204                         break;
 205
 206                 LASSERT(req != lock);
 207                 flock = &lock->l_policy_data.l_flock;
 208                 LASSERT(flock->owner == bl_owner);
 209                 bl_owner = flock->blocking_owner;
 210                 bl_exp_new = class_export_get(flock->blocking_export);
 211                 class_export_put(bl_exp);
 212
 213                 cfs_hash_put(bl_exp->exp_flock_hash, &lock->l_exp_flock_hash);
 214                 bl_exp = bl_exp_new;
 215
 216                 if (bl_owner == req_owner && bl_exp == req_exp) {
 217                         class_export_put(bl_exp);
 218                         return 1;
 219                 }
 220         }
 221         class_export_put(bl_exp);
 222
 223         return 0;
 224 }
 225
 226 /**
 227  * Process a granting attempt for flock lock.
 228  * Must be called under ns lock held.
 229  *
 230  * This function looks for any conflicts for \a lock in the granted or
 231  * waiting queues. The lock is granted if no conflicts are found in
 232  * either queue.
 233  *
 234  * It is also responsible for splitting a lock if a portion of the lock
 235  * is released.
 236  *
 237  * If \a first_enq is 0 (ie, called from ldlm_reprocess_queue):
 238  *   - blocking ASTs have already been sent
 239  *
 240  * If \a first_enq is 1 (ie, called from ldlm_lock_enqueue):
 241  *   - blocking ASTs have not been sent yet, so list of conflicting locks
 242  *     would be collected and ASTs sent.
 243  */
 244 int
 245 ldlm_process_flock_lock(struct ldlm_lock *req, __u64 *flags, int first_enq,
 246                         ldlm_error_t *err, cfs_list_t *work_list)
 247 {
 248         struct ldlm_resource *res = req->l_resource;
 249         struct ldlm_namespace *ns = ldlm_res_to_ns(res);
 250         cfs_list_t *tmp;
 251         cfs_list_t *ownlocks = NULL;
 252         struct ldlm_lock *lock = NULL;
 253         struct ldlm_lock *new = req;
 254         struct ldlm_lock *new2 = NULL;
 255         ldlm_mode_t mode = req->l_req_mode;
 256         int local = ns_is_client(ns);
 257         int added = (mode == LCK_NL);
 258         int overlaps = 0;
 259         int splitted = 0;
 260         const struct ldlm_callback_suite null_cbs = { NULL };
 261         int rc;
 262         ENTRY;
 263
 264         CDEBUG(D_DLMTRACE, "flags %#llx owner "LPU64" pid %u mode %u start "
 265                LPU64" end "LPU64"\n", *flags,
 266                new->l_policy_data.l_flock.owner,
 267                new->l_policy_data.l_flock.pid, mode,
 268                req->l_policy_data.l_flock.start,
 269                req->l_policy_data.l_flock.end);
 270
 271         *err = ELDLM_OK;
 272
 273         if (local) {
 274                 /* No blocking ASTs are sent to the clients for
 275                  * Posix file & record locks */
 276                 req->l_blocking_ast = NULL;
 277         } else {
 278                 /* Called on the server for lock cancels. */
 279                 req->l_blocking_ast = ldlm_flock_blocking_ast;
 280         }
 281
 282 reprocess:
 283         if ((*flags == LDLM_FL_WAIT_NOREPROC) || (mode == LCK_NL)) {
 284                 /* This loop determines where this processes locks start
 285                  * in the resource lr_granted list. */
 286                 cfs_list_for_each(tmp, &res->lr_granted) {
 287                         lock = cfs_list_entry(tmp, struct ldlm_lock,
 288                                               l_res_link);
 289                         if (ldlm_same_flock_owner(lock, req)) {
 290                                 ownlocks = tmp;
 291                                 break;
 292                         }
 293                 }
 294         } else {
 295                 lockmode_verify(mode);
 296
 297                 /* This loop determines if there are existing locks
 298                  * that conflict with the new lock request. */
 299                 cfs_list_for_each(tmp, &res->lr_granted) {
 300                         lock = cfs_list_entry(tmp, struct ldlm_lock,
 301                                               l_res_link);
 302
 303                         if (ldlm_same_flock_owner(lock, req)) {
 304                                 if (!ownlocks)
 305                                         ownlocks = tmp;
 306                                 continue;
 307                         }
 308
 309                         /* locks are compatible, overlap doesn't matter */
 310                         if (lockmode_compat(lock->l_granted_mode, mode))
 311                                 continue;
 312
 313                         if (!ldlm_flocks_overlap(lock, req))
 314                                 continue;
 315
 316                         if (!first_enq)
 317                                 RETURN(LDLM_ITER_CONTINUE);
 318
 319                         if (*flags & LDLM_FL_BLOCK_NOWAIT) {
 320                                 ldlm_flock_destroy(req, mode, *flags);
 321                                 *err = -EAGAIN;
 322                                 RETURN(LDLM_ITER_STOP);
 323                         }
 324
 325                         if (*flags & LDLM_FL_TEST_LOCK) {
 326                                 ldlm_flock_destroy(req, mode, *flags);
 327                                 req->l_req_mode = lock->l_granted_mode;
 328                                 req->l_policy_data.l_flock.pid =
 329                                         lock->l_policy_data.l_flock.pid;
 330                                 req->l_policy_data.l_flock.start =
 331                                         lock->l_policy_data.l_flock.start;
 332                                 req->l_policy_data.l_flock.end =
 333                                         lock->l_policy_data.l_flock.end;
 334                                 *flags |= LDLM_FL_LOCK_CHANGED;
 335                                 RETURN(LDLM_ITER_STOP);
 336                         }
 337
 338                         /* add lock to blocking list before deadlock
 339                          * check to prevent race */
 340                         rc = ldlm_flock_blocking_link(req, lock);
 341                         if (rc) {
 342                                 ldlm_flock_destroy(req, mode, *flags);
 343                                 *err = rc;
 344                                 RETURN(LDLM_ITER_STOP);
 345                         }
 346                         if (ldlm_flock_deadlock(req, lock)) {
 347                                 ldlm_flock_blocking_unlink(req);
 348                                 ldlm_flock_destroy(req, mode, *flags);
 349                                 *err = -EDEADLK;
 350                                 RETURN(LDLM_ITER_STOP);
 351                         }
 352
 353                         ldlm_resource_add_lock(res, &res->lr_waiting, req);
 354                         *flags |= LDLM_FL_BLOCK_GRANTED;
 355                         RETURN(LDLM_ITER_STOP);
 356                 }
 357         }
 358
 359         if (*flags & LDLM_FL_TEST_LOCK) {
 360                 ldlm_flock_destroy(req, mode, *flags);
 361                 req->l_req_mode = LCK_NL;
 362                 *flags |= LDLM_FL_LOCK_CHANGED;
 363                 RETURN(LDLM_ITER_STOP);
 364         }
 365
 366         /* In case we had slept on this lock request take it off of the
 367          * deadlock detection hash list. */
 368         ldlm_flock_blocking_unlink(req);
 369
 370         /* Scan the locks owned by this process that overlap this request.
 371          * We may have to merge or split existing locks. */
 372
 373         if (!ownlocks)
 374                 ownlocks = &res->lr_granted;
 375
 376         list_for_remaining_safe(ownlocks, tmp, &res->lr_granted) {
 377                 lock = cfs_list_entry(ownlocks, struct ldlm_lock, l_res_link);
 378
 379                 if (!ldlm_same_flock_owner(lock, new))
 380                         break;
 381
 382                 if (lock->l_granted_mode == mode) {
 383                         /* If the modes are the same then we need to process
 384                          * locks that overlap OR adjoin the new lock. The extra
 385                          * logic condition is necessary to deal with arithmetic
 386                          * overflow and underflow. */
 387                         if ((new->l_policy_data.l_flock.start >
 388                              (lock->l_policy_data.l_flock.end + 1))
 389                             && (lock->l_policy_data.l_flock.end !=
 390                                 OBD_OBJECT_EOF))
 391                                 continue;
 392
 393                         if ((new->l_policy_data.l_flock.end <
 394                              (lock->l_policy_data.l_flock.start - 1))
 395                             && (lock->l_policy_data.l_flock.start != 0))
 396                                 break;
 397
 398                         if (new->l_policy_data.l_flock.start <
 399                             lock->l_policy_data.l_flock.start) {
 400                                 lock->l_policy_data.l_flock.start =
 401                                         new->l_policy_data.l_flock.start;
 402                         } else {
 403                                 new->l_policy_data.l_flock.start =
 404                                         lock->l_policy_data.l_flock.start;
 405                         }
 406
 407                         if (new->l_policy_data.l_flock.end >
 408                             lock->l_policy_data.l_flock.end) {
 409                                 lock->l_policy_data.l_flock.end =
 410                                         new->l_policy_data.l_flock.end;
 411                         } else {
 412                                 new->l_policy_data.l_flock.end =
 413                                         lock->l_policy_data.l_flock.end;
 414                         }
 415
 416                         if (added) {
 417                                 ldlm_flock_destroy(lock, mode, *flags);
 418                         } else {
 419                                 new = lock;
 420                                 added = 1;
 421                         }
 422                         continue;
 423                 }
 424
 425                 if (new->l_policy_data.l_flock.start >
 426                     lock->l_policy_data.l_flock.end)
 427                         continue;
 428
 429                 if (new->l_policy_data.l_flock.end <
 430                     lock->l_policy_data.l_flock.start)
 431                         break;
 432
 433                 ++overlaps;
 434
 435                 if (new->l_policy_data.l_flock.start <=
 436                     lock->l_policy_data.l_flock.start) {
 437                         if (new->l_policy_data.l_flock.end <
 438                             lock->l_policy_data.l_flock.end) {
 439                                 lock->l_policy_data.l_flock.start =
 440                                         new->l_policy_data.l_flock.end + 1;
 441                                 break;
 442                         }
 443                         ldlm_flock_destroy(lock, lock->l_req_mode, *flags);
 444                         continue;
 445                 }
 446                 if (new->l_policy_data.l_flock.end >=
 447                     lock->l_policy_data.l_flock.end) {
 448                         lock->l_policy_data.l_flock.end =
 449                                 new->l_policy_data.l_flock.start - 1;
 450                         continue;
 451                 }
 452
 453                 /* split the existing lock into two locks */
 454
 455                 /* if this is an F_UNLCK operation then we could avoid
 456                  * allocating a new lock and use the req lock passed in
 457                  * with the request but this would complicate the reply
 458                  * processing since updates to req get reflected in the
 459                  * reply. The client side replays the lock request so
 460                  * it must see the original lock data in the reply. */
 461
 462                 /* XXX - if ldlm_lock_new() can sleep we should
 463                  * release the lr_lock, allocate the new lock,
 464                  * and restart processing this lock. */
 465                 if (!new2) {
 466                         unlock_res_and_lock(req);
 467                         new2 = ldlm_lock_create(ns, &res->lr_name, LDLM_FLOCK,
 468                                                 lock->l_granted_mode, &null_cbs,
 469                                                 NULL, 0, LVB_T_NONE);
 470                         lock_res_and_lock(req);
 471                         if (!new2) {
 472                                 ldlm_flock_destroy(req, lock->l_granted_mode,
 473                                                    *flags);
 474                                 *err = -ENOLCK;
 475                                 RETURN(LDLM_ITER_STOP);
 476                         }
 477                         goto reprocess;
 478                 }
 479
 480                 splitted = 1;
 481
 482                 new2->l_granted_mode = lock->l_granted_mode;
 483                 new2->l_policy_data.l_flock.pid =
 484                         new->l_policy_data.l_flock.pid;
 485                 new2->l_policy_data.l_flock.owner =
 486                         new->l_policy_data.l_flock.owner;
 487                 new2->l_policy_data.l_flock.start =
 488                         lock->l_policy_data.l_flock.start;
 489                 new2->l_policy_data.l_flock.end =
 490                         new->l_policy_data.l_flock.start - 1;
 491                 lock->l_policy_data.l_flock.start =
 492                         new->l_policy_data.l_flock.end + 1;
 493                 new2->l_conn_export = lock->l_conn_export;
 494                 if (lock->l_export != NULL) {
 495                         new2->l_export = class_export_lock_get(lock->l_export, new2);
 496                         if (new2->l_export->exp_lock_hash &&
 497                             cfs_hlist_unhashed(&new2->l_exp_hash))
 498                                 cfs_hash_add(new2->l_export->exp_lock_hash,
 499                                              &new2->l_remote_handle,
 500                                              &new2->l_exp_hash);
 501                 }
 502                 if (*flags == LDLM_FL_WAIT_NOREPROC)
 503                         ldlm_lock_addref_internal_nolock(new2,
 504                                                          lock->l_granted_mode);
 505
 506                 /* insert new2 at lock */
 507                 ldlm_resource_add_lock(res, ownlocks, new2);
 508                 LDLM_LOCK_RELEASE(new2);
 509                 break;
 510         }
 511
 512         /* if new2 is created but never used, destroy it*/
 513         if (splitted == 0 && new2 != NULL)
 514                 ldlm_lock_destroy_nolock(new2);
 515
 516         /* At this point we're granting the lock request. */
 517         req->l_granted_mode = req->l_req_mode;
 518
 519         /* Add req to the granted queue before calling ldlm_reprocess_all(). */
 520         if (!added) {
 521                 cfs_list_del_init(&req->l_res_link);
 522                 /* insert new lock before ownlocks in list. */
 523                 ldlm_resource_add_lock(res, ownlocks, req);
 524         }
 525
 526         if (*flags != LDLM_FL_WAIT_NOREPROC) {
 527 #ifdef HAVE_SERVER_SUPPORT
 528                 if (first_enq) {
 529                         /* If this is an unlock, reprocess the waitq and
 530                          * send completions ASTs for locks that can now be
 531                          * granted. The only problem with doing this
 532                          * reprocessing here is that the completion ASTs for
 533                          * newly granted locks will be sent before the unlock
 534                          * completion is sent. It shouldn't be an issue. Also
 535                          * note that ldlm_process_flock_lock() will recurse,
 536                          * but only once because first_enq will be false from
 537                          * ldlm_reprocess_queue. */
 538                         if ((mode == LCK_NL) && overlaps) {
 539                                 CFS_LIST_HEAD(rpc_list);
 540                                 int rc;
 541 restart:
 542                                 ldlm_reprocess_queue(res, &res->lr_waiting,
 543                                                      &rpc_list);
 544
 545                                 unlock_res_and_lock(req);
 546                                 rc = ldlm_run_ast_work(ns, &rpc_list,
 547                                                        LDLM_WORK_CP_AST);
 548                                 lock_res_and_lock(req);
 549                                 if (rc == -ERESTART)
 550                                         GOTO(restart, -ERESTART);
 551                        }
 552                 } else {
 553                         LASSERT(req->l_completion_ast);
 554                         ldlm_add_ast_work_item(req, NULL, work_list);
 555                 }
 556 #else /* !HAVE_SERVER_SUPPORT */
 557                 /* The only one possible case for client-side calls flock
 558                  * policy function is ldlm_flock_completion_ast inside which
 559                  * carries LDLM_FL_WAIT_NOREPROC flag. */
 560                 CERROR("Illegal parameter for client-side-only module.\n");
 561                 LBUG();
 562 #endif /* HAVE_SERVER_SUPPORT */
 563         }
 564
 565         /* In case we're reprocessing the requested lock we can't destroy
 566          * it until after calling ldlm_add_ast_work_item() above so that laawi()
 567          * can bump the reference count on \a req. Otherwise \a req
 568          * could be freed before the completion AST can be sent.  */
 569         if (added)
 570                 ldlm_flock_destroy(req, mode, *flags);
 571
 572         ldlm_resource_dump(D_INFO, res);
 573         RETURN(LDLM_ITER_CONTINUE);
 574 }
 575
 576 struct ldlm_flock_wait_data {
 577         struct ldlm_lock *fwd_lock;
 578         int               fwd_generation;
 579 };
 580
 581 static void
 582 ldlm_flock_interrupted_wait(void *data)
 583 {
 584         struct ldlm_lock *lock;
 585         ENTRY;
 586
 587         lock = ((struct ldlm_flock_wait_data *)data)->fwd_lock;
 588
 589         /* take lock off the deadlock detection hash list. */
 590         lock_res_and_lock(lock);
 591         ldlm_flock_blocking_unlink(lock);
 592
 593         /* client side - set flag to prevent lock from being put on LRU list */
 594         lock->l_flags |= LDLM_FL_CBPENDING;
 595         unlock_res_and_lock(lock);
 596
 597         EXIT;
 598 }
 599
 600 /**
 601  * Flock completion callback function.
 602  *
 603  * \param lock [in,out]: A lock to be handled
 604  * \param flags    [in]: flags
 605  * \param *data    [in]: ldlm_work_cp_ast_lock() will use ldlm_cb_set_arg
 606  *
 607  * \retval 0    : success
 608  * \retval <0   : failure
 609  */
 610 int
 611 ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data)
 612 {
 613         struct file_lock                *getlk = lock->l_ast_data;
 614         struct obd_device              *obd;
 615         struct obd_import              *imp = NULL;
 616         struct ldlm_flock_wait_data     fwd;
 617         struct l_wait_info              lwi;
 618         ldlm_error_t                    err;
 619         int                             rc = 0;
 620         ENTRY;
 621
 622         CDEBUG(D_DLMTRACE, "flags: 0x%llx data: %p getlk: %p\n",
 623                flags, data, getlk);
 624
 625         /* Import invalidation. We need to actually release the lock
 626          * references being held, so that it can go away. No point in
 627          * holding the lock even if app still believes it has it, since
 628          * server already dropped it anyway. Only for granted locks too. */
 629         if ((lock->l_flags & (LDLM_FL_FAILED|LDLM_FL_LOCAL_ONLY)) ==
 630             (LDLM_FL_FAILED|LDLM_FL_LOCAL_ONLY)) {
 631                 if (lock->l_req_mode == lock->l_granted_mode &&
 632                     lock->l_granted_mode != LCK_NL &&
 633                     NULL == data)
 634                         ldlm_lock_decref_internal(lock, lock->l_req_mode);
 635
 636                 /* Need to wake up the waiter if we were evicted */
 637                 cfs_waitq_signal(&lock->l_waitq);
 638                 RETURN(0);
 639         }
 640
 641         LASSERT(flags != LDLM_FL_WAIT_NOREPROC);
 642
 643         if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
 644                        LDLM_FL_BLOCK_CONV))) {
 645                 if (NULL == data)
 646                         /* mds granted the lock in the reply */
 647                         goto granted;
 648                 /* CP AST RPC: lock get granted, wake it up */
 649                 cfs_waitq_signal(&lock->l_waitq);
 650                 RETURN(0);
 651         }
 652
 653         LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
 654                    "sleeping");
 655         fwd.fwd_lock = lock;
 656         obd = class_exp2obd(lock->l_conn_export);
 657
 658         /* if this is a local lock, there is no import */
 659         if (NULL != obd)
 660                 imp = obd->u.cli.cl_import;
 661
 662         if (NULL != imp) {
 663                 spin_lock(&imp->imp_lock);
 664                 fwd.fwd_generation = imp->imp_generation;
 665                 spin_unlock(&imp->imp_lock);
 666         }
 667
 668         lwi = LWI_TIMEOUT_INTR(0, NULL, ldlm_flock_interrupted_wait, &fwd);
 669
 670         /* Go to sleep until the lock is granted. */
 671         rc = l_wait_event(lock->l_waitq, is_granted_or_cancelled(lock), &lwi);
 672
 673         if (rc) {
 674                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
 675                            rc);
 676                 RETURN(rc);
 677         }
 678
 679 granted:
 680         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT, 10);
 681
 682         if (lock->l_flags & LDLM_FL_DESTROYED) {
 683                 LDLM_DEBUG(lock, "client-side enqueue waking up: destroyed");
 684                 RETURN(0);
 685         }
 686
 687         if (lock->l_flags & LDLM_FL_FAILED) {
 688                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed");
 689                 RETURN(-EIO);
 690         }
 691
 692         if (rc) {
 693                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
 694                            rc);
 695                 RETURN(rc);
 696         }
 697
 698         LDLM_DEBUG(lock, "client-side enqueue granted");
 699
 700         lock_res_and_lock(lock);
 701
 702         /* take lock off the deadlock detection hash list. */
 703         ldlm_flock_blocking_unlink(lock);
 704
 705         /* ldlm_lock_enqueue() has already placed lock on the granted list. */
 706         cfs_list_del_init(&lock->l_res_link);
 707
 708         if (flags & LDLM_FL_TEST_LOCK) {
 709                 /* fcntl(F_GETLK) request */
 710                 /* The old mode was saved in getlk->fl_type so that if the mode
 711                  * in the lock changes we can decref the appropriate refcount.*/
 712                 ldlm_flock_destroy(lock, flock_type(getlk),
 713                                    LDLM_FL_WAIT_NOREPROC);
 714                 switch (lock->l_granted_mode) {
 715                 case LCK_PR:
 716                         flock_set_type(getlk, F_RDLCK);
 717                         break;
 718                 case LCK_PW:
 719                         flock_set_type(getlk, F_WRLCK);
 720                         break;
 721                 default:
 722                         flock_set_type(getlk, F_UNLCK);
 723                 }
 724                 flock_set_pid(getlk, (pid_t)lock->l_policy_data.l_flock.pid);
 725                 flock_set_start(getlk,
 726                                 (loff_t)lock->l_policy_data.l_flock.start);
 727                 flock_set_end(getlk,
 728                               (loff_t)lock->l_policy_data.l_flock.end);
 729         } else {
 730                 __u64 noreproc = LDLM_FL_WAIT_NOREPROC;
 731
 732                 /* We need to reprocess the lock to do merges or splits
 733                  * with existing locks owned by this process. */
 734                 ldlm_process_flock_lock(lock, &noreproc, 1, &err, NULL);
 735         }
 736         unlock_res_and_lock(lock);
 737         RETURN(0);
 738 }
 739 EXPORT_SYMBOL(ldlm_flock_completion_ast);
 740
 741 int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 742                             void *data, int flag)
 743 {
 744         ENTRY;
 745
 746         LASSERT(lock);
 747         LASSERT(flag == LDLM_CB_CANCELING);
 748
 749         /* take lock off the deadlock detection hash list. */
 750         lock_res_and_lock(lock);
 751         ldlm_flock_blocking_unlink(lock);
 752         unlock_res_and_lock(lock);
 753         RETURN(0);
 754 }
 755
 756 void ldlm_flock_policy_wire18_to_local(const ldlm_wire_policy_data_t *wpolicy,
 757                                        ldlm_policy_data_t *lpolicy)
 758 {
 759         memset(lpolicy, 0, sizeof(*lpolicy));
 760         lpolicy->l_flock.start = wpolicy->l_flock.lfw_start;
 761         lpolicy->l_flock.end = wpolicy->l_flock.lfw_end;
 762         lpolicy->l_flock.pid = wpolicy->l_flock.lfw_pid;
 763         /* Compat code, old clients had no idea about owner field and
 764          * relied solely on pid for ownership. Introduced in LU-104, 2.1,
 765          * April 2011 */
 766         lpolicy->l_flock.owner = wpolicy->l_flock.lfw_pid;
 767 }
 768
 769
 770 void ldlm_flock_policy_wire21_to_local(const ldlm_wire_policy_data_t *wpolicy,
 771                                        ldlm_policy_data_t *lpolicy)
 772 {
 773         memset(lpolicy, 0, sizeof(*lpolicy));
 774         lpolicy->l_flock.start = wpolicy->l_flock.lfw_start;
 775         lpolicy->l_flock.end = wpolicy->l_flock.lfw_end;
 776         lpolicy->l_flock.pid = wpolicy->l_flock.lfw_pid;
 777         lpolicy->l_flock.owner = wpolicy->l_flock.lfw_owner;
 778 }
 779
 780 void ldlm_flock_policy_local_to_wire(const ldlm_policy_data_t *lpolicy,
 781                                      ldlm_wire_policy_data_t *wpolicy)
 782 {
 783         memset(wpolicy, 0, sizeof(*wpolicy));
 784         wpolicy->l_flock.lfw_start = lpolicy->l_flock.start;
 785         wpolicy->l_flock.lfw_end = lpolicy->l_flock.end;
 786         wpolicy->l_flock.lfw_pid = lpolicy->l_flock.pid;
 787         wpolicy->l_flock.lfw_owner = lpolicy->l_flock.owner;
 788 }
 789
 790 /*
 791  * Export handle<->flock hash operations.
 792  */
 793 static unsigned
 794 ldlm_export_flock_hash(cfs_hash_t *hs, const void *key, unsigned mask)
 795 {
 796         return cfs_hash_u64_hash(*(__u64 *)key, mask);
 797 }
 798
 799 static void *
 800 ldlm_export_flock_key(cfs_hlist_node_t *hnode)
 801 {
 802         struct ldlm_lock *lock;
 803
 804         lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 805         return &lock->l_policy_data.l_flock.owner;
 806 }
 807
 808 static int
 809 ldlm_export_flock_keycmp(const void *key, cfs_hlist_node_t *hnode)
 810 {
 811         return !memcmp(ldlm_export_flock_key(hnode), key, sizeof(__u64));
 812 }
 813
 814 static void *
 815 ldlm_export_flock_object(cfs_hlist_node_t *hnode)
 816 {
 817         return cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 818 }
 819
 820 static void
 821 ldlm_export_flock_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 822 {
 823         struct ldlm_lock *lock;
 824         struct ldlm_flock *flock;
 825
 826         lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 827         LDLM_LOCK_GET(lock);
 828
 829         flock = &lock->l_policy_data.l_flock;
 830         LASSERT(flock->blocking_export != NULL);
 831         class_export_get(flock->blocking_export);
 832         flock->blocking_refs++;
 833 }
 834
 835 static void
 836 ldlm_export_flock_put(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 837 {
 838         struct ldlm_lock *lock;
 839         struct ldlm_flock *flock;
 840
 841         lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 842         LDLM_LOCK_RELEASE(lock);
 843
 844         flock = &lock->l_policy_data.l_flock;
 845         LASSERT(flock->blocking_export != NULL);
 846         class_export_put(flock->blocking_export);
 847         if (--flock->blocking_refs == 0) {
 848                 flock->blocking_owner = 0;
 849                 flock->blocking_export = NULL;
 850         }
 851 }
 852
 853 static cfs_hash_ops_t ldlm_export_flock_ops = {
 854         .hs_hash        = ldlm_export_flock_hash,
 855         .hs_key         = ldlm_export_flock_key,
 856         .hs_keycmp      = ldlm_export_flock_keycmp,
 857         .hs_object      = ldlm_export_flock_object,
 858         .hs_get         = ldlm_export_flock_get,
 859         .hs_put         = ldlm_export_flock_put,
 860         .hs_put_locked  = ldlm_export_flock_put,
 861 };
 862
 863 int ldlm_init_flock_export(struct obd_export *exp)
 864 {
 865         exp->exp_flock_hash =
 866                 cfs_hash_create(obd_uuid2str(&exp->exp_client_uuid),
 867                                 HASH_EXP_LOCK_CUR_BITS,
 868                                 HASH_EXP_LOCK_MAX_BITS,
 869                                 HASH_EXP_LOCK_BKT_BITS, 0,
 870                                 CFS_HASH_MIN_THETA, CFS_HASH_MAX_THETA,
 871                                 &ldlm_export_flock_ops,
 872                                 CFS_HASH_DEFAULT | CFS_HASH_NBLK_CHANGE);
 873         if (!exp->exp_flock_hash)
 874                 RETURN(-ENOMEM);
 875
 876         RETURN(0);
 877 }
 878 EXPORT_SYMBOL(ldlm_init_flock_export);
 879
 880 void ldlm_destroy_flock_export(struct obd_export *exp)
 881 {
 882         ENTRY;
 883         if (exp->exp_flock_hash) {
 884                 cfs_hash_putref(exp->exp_flock_hash);
 885                 exp->exp_flock_hash = NULL;
 886         }
 887         EXIT;
 888 }
 889 EXPORT_SYMBOL(ldlm_destroy_flock_export);