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, 2011, Whamcloud, Inc.
  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 #define DEBUG_SUBSYSTEM S_LDLM
  42
  43 #ifdef __KERNEL__
  44 #include <lustre_dlm.h>
  45 #include <obd_support.h>
  46 #include <obd_class.h>
  47 #include <lustre_lib.h>
  48 #include <libcfs/list.h>
  49 #else
  50 #include <liblustre.h>
  51 #include <obd_class.h>
  52 #endif
  53
  54 #include "ldlm_internal.h"
  55
  56 int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
  57                             void *data, int flag);
  58
  59 /**
  60  * list_for_remaining_safe - iterate over the remaining entries in a list
  61  *              and safeguard against removal of a list entry.
  62  * \param pos   the &struct list_head to use as a loop counter. pos MUST
  63  *              have been initialized prior to using it in this macro.
  64  * \param n     another &struct list_head to use as temporary storage
  65  * \param head  the head for your list.
  66  */
  67 #define list_for_remaining_safe(pos, n, head) \
  68         for (n = pos->next; pos != (head); pos = n, n = pos->next)
  69
  70 static inline int
  71 ldlm_same_flock_owner(struct ldlm_lock *lock, struct ldlm_lock *new)
  72 {
  73         return((new->l_policy_data.l_flock.owner ==
  74                 lock->l_policy_data.l_flock.owner) &&
  75                (new->l_export == lock->l_export));
  76 }
  77
  78 static inline int
  79 ldlm_flocks_overlap(struct ldlm_lock *lock, struct ldlm_lock *new)
  80 {
  81         return((new->l_policy_data.l_flock.start <=
  82                 lock->l_policy_data.l_flock.end) &&
  83                (new->l_policy_data.l_flock.end >=
  84                 lock->l_policy_data.l_flock.start));
  85 }
  86
  87 static inline int ldlm_flock_blocking_link(struct ldlm_lock *req,
  88                                            struct ldlm_lock *lock)
  89 {
  90         int rc = 0;
  91
  92         /* For server only */
  93         if (req->l_export == NULL)
  94                 return 0;
  95
  96         if (unlikely(req->l_export->exp_flock_hash == NULL)) {
  97                 rc = ldlm_init_flock_export(req->l_export);
  98                 if (rc)
  99                         goto error;
 100         }
 101
 102         LASSERT(cfs_hlist_unhashed(&req->l_exp_flock_hash));
 103
 104         req->l_policy_data.l_flock.blocking_owner =
 105                 lock->l_policy_data.l_flock.owner;
 106         req->l_policy_data.l_flock.blocking_export =
 107                 lock->l_export;
 108         req->l_policy_data.l_flock.blocking_refs = 0;
 109
 110         cfs_hash_add(req->l_export->exp_flock_hash,
 111                      &req->l_policy_data.l_flock.owner,
 112                      &req->l_exp_flock_hash);
 113 error:
 114         return rc;
 115 }
 116
 117 static inline void ldlm_flock_blocking_unlink(struct ldlm_lock *req)
 118 {
 119         /* For server only */
 120         if (req->l_export == NULL)
 121                 return;
 122
 123         check_res_locked(req->l_resource);
 124         if (req->l_export->exp_flock_hash != NULL &&
 125             !cfs_hlist_unhashed(&req->l_exp_flock_hash))
 126                 cfs_hash_del(req->l_export->exp_flock_hash,
 127                              &req->l_policy_data.l_flock.owner,
 128                              &req->l_exp_flock_hash);
 129 }
 130
 131 static inline void
 132 ldlm_flock_destroy(struct ldlm_lock *lock, ldlm_mode_t mode, int flags)
 133 {
 134         ENTRY;
 135
 136         LDLM_DEBUG(lock, "ldlm_flock_destroy(mode: %d, flags: 0x%x)",
 137                    mode, flags);
 138
 139         /* Safe to not lock here, since it should be empty anyway */
 140         LASSERT(cfs_hlist_unhashed(&lock->l_exp_flock_hash));
 141
 142         cfs_list_del_init(&lock->l_res_link);
 143         if (flags == LDLM_FL_WAIT_NOREPROC &&
 144             !(lock->l_flags & LDLM_FL_FAILED)) {
 145                 /* client side - set a flag to prevent sending a CANCEL */
 146                 lock->l_flags |= LDLM_FL_LOCAL_ONLY | LDLM_FL_CBPENDING;
 147
 148                 /* when reaching here, it is under lock_res_and_lock(). Thus,
 149                    need call the nolock version of ldlm_lock_decref_internal*/
 150                 ldlm_lock_decref_internal_nolock(lock, mode);
 151         }
 152
 153         ldlm_lock_destroy_nolock(lock);
 154         EXIT;
 155 }
 156
 157 static int
 158 ldlm_flock_deadlock(struct ldlm_lock *req, struct ldlm_lock *bl_lock)
 159 {
 160         struct obd_export *req_exp = req->l_export;
 161         struct obd_export *bl_exp = bl_lock->l_export;
 162         __u64 req_owner = req->l_policy_data.l_flock.owner;
 163         __u64 bl_owner = bl_lock->l_policy_data.l_flock.owner;
 164
 165         /* For server only */
 166         if (req_exp == NULL)
 167                 return 0;
 168
 169         class_export_get(bl_exp);
 170         while (1) {
 171                 struct obd_export *bl_exp_new;
 172                 struct ldlm_lock *lock = NULL;
 173                 struct ldlm_flock *flock;
 174
 175                 if (bl_exp->exp_flock_hash != NULL)
 176                         lock = cfs_hash_lookup(bl_exp->exp_flock_hash,
 177                                                &bl_owner);
 178                 if (lock == NULL)
 179                         break;
 180
 181                 flock = &lock->l_policy_data.l_flock;
 182                 LASSERT(flock->owner == bl_owner);
 183                 bl_owner = flock->blocking_owner;
 184                 bl_exp_new = class_export_get(flock->blocking_export);
 185                 class_export_put(bl_exp);
 186
 187                 cfs_hash_put(bl_exp->exp_flock_hash, &lock->l_exp_flock_hash);
 188                 bl_exp = bl_exp_new;
 189
 190                 if (bl_owner == req_owner && bl_exp == req_exp) {
 191                         class_export_put(bl_exp);
 192                         return 1;
 193                 }
 194         }
 195         class_export_put(bl_exp);
 196
 197         return 0;
 198 }
 199
 200 int
 201 ldlm_process_flock_lock(struct ldlm_lock *req, int *flags, int first_enq,
 202                         ldlm_error_t *err, cfs_list_t *work_list)
 203 {
 204         struct ldlm_resource *res = req->l_resource;
 205         struct ldlm_namespace *ns = ldlm_res_to_ns(res);
 206         cfs_list_t *tmp;
 207         cfs_list_t *ownlocks = NULL;
 208         struct ldlm_lock *lock = NULL;
 209         struct ldlm_lock *new = req;
 210         struct ldlm_lock *new2 = NULL;
 211         ldlm_mode_t mode = req->l_req_mode;
 212         int local = ns_is_client(ns);
 213         int added = (mode == LCK_NL);
 214         int overlaps = 0;
 215         int splitted = 0;
 216         const struct ldlm_callback_suite null_cbs = { NULL };
 217         int rc;
 218         ENTRY;
 219
 220         CDEBUG(D_DLMTRACE, "flags %#x owner "LPU64" pid %u mode %u start "LPU64
 221                " end "LPU64"\n", *flags, new->l_policy_data.l_flock.owner,
 222                new->l_policy_data.l_flock.pid, mode,
 223                req->l_policy_data.l_flock.start,
 224                req->l_policy_data.l_flock.end);
 225
 226         *err = ELDLM_OK;
 227
 228         if (local) {
 229                 /* No blocking ASTs are sent to the clients for
 230                  * Posix file & record locks */
 231                 req->l_blocking_ast = NULL;
 232         } else {
 233                 /* Called on the server for lock cancels. */
 234                 req->l_blocking_ast = ldlm_flock_blocking_ast;
 235         }
 236
 237 reprocess:
 238         if ((*flags == LDLM_FL_WAIT_NOREPROC) || (mode == LCK_NL)) {
 239                 /* This loop determines where this processes locks start
 240                  * in the resource lr_granted list. */
 241                 cfs_list_for_each(tmp, &res->lr_granted) {
 242                         lock = cfs_list_entry(tmp, struct ldlm_lock,
 243                                               l_res_link);
 244                         if (ldlm_same_flock_owner(lock, req)) {
 245                                 ownlocks = tmp;
 246                                 break;
 247                         }
 248                 }
 249         } else {
 250                 lockmode_verify(mode);
 251
 252                 /* This loop determines if there are existing locks
 253                  * that conflict with the new lock request. */
 254                 cfs_list_for_each(tmp, &res->lr_granted) {
 255                         lock = cfs_list_entry(tmp, struct ldlm_lock,
 256                                               l_res_link);
 257
 258                         if (ldlm_same_flock_owner(lock, req)) {
 259                                 if (!ownlocks)
 260                                         ownlocks = tmp;
 261                                 continue;
 262                         }
 263
 264                         /* locks are compatible, overlap doesn't matter */
 265                         if (lockmode_compat(lock->l_granted_mode, mode))
 266                                 continue;
 267
 268                         if (!ldlm_flocks_overlap(lock, req))
 269                                 continue;
 270
 271                         if (!first_enq)
 272                                 RETURN(LDLM_ITER_CONTINUE);
 273
 274                         if (*flags & LDLM_FL_BLOCK_NOWAIT) {
 275                                 ldlm_flock_destroy(req, mode, *flags);
 276                                 *err = -EAGAIN;
 277                                 RETURN(LDLM_ITER_STOP);
 278                         }
 279
 280                         if (*flags & LDLM_FL_TEST_LOCK) {
 281                                 ldlm_flock_destroy(req, mode, *flags);
 282                                 req->l_req_mode = lock->l_granted_mode;
 283                                 req->l_policy_data.l_flock.pid =
 284                                         lock->l_policy_data.l_flock.pid;
 285                                 req->l_policy_data.l_flock.start =
 286                                         lock->l_policy_data.l_flock.start;
 287                                 req->l_policy_data.l_flock.end =
 288                                         lock->l_policy_data.l_flock.end;
 289                                 *flags |= LDLM_FL_LOCK_CHANGED;
 290                                 RETURN(LDLM_ITER_STOP);
 291                         }
 292
 293                         if (ldlm_flock_deadlock(req, lock)) {
 294                                 ldlm_flock_destroy(req, mode, *flags);
 295                                 *err = -EDEADLK;
 296                                 RETURN(LDLM_ITER_STOP);
 297                         }
 298
 299                         rc = ldlm_flock_blocking_link(req, lock);
 300                         if (rc) {
 301                                 ldlm_flock_destroy(req, mode, *flags);
 302                                 *err = rc;
 303                                 RETURN(LDLM_ITER_STOP);
 304                         }
 305                         ldlm_resource_add_lock(res, &res->lr_waiting, req);
 306                         *flags |= LDLM_FL_BLOCK_GRANTED;
 307                         RETURN(LDLM_ITER_STOP);
 308                 }
 309         }
 310
 311         if (*flags & LDLM_FL_TEST_LOCK) {
 312                 ldlm_flock_destroy(req, mode, *flags);
 313                 req->l_req_mode = LCK_NL;
 314                 *flags |= LDLM_FL_LOCK_CHANGED;
 315                 RETURN(LDLM_ITER_STOP);
 316         }
 317
 318         /* In case we had slept on this lock request take it off of the
 319          * deadlock detection hash list. */
 320         ldlm_flock_blocking_unlink(req);
 321
 322         /* Scan the locks owned by this process that overlap this request.
 323          * We may have to merge or split existing locks. */
 324
 325         if (!ownlocks)
 326                 ownlocks = &res->lr_granted;
 327
 328         list_for_remaining_safe(ownlocks, tmp, &res->lr_granted) {
 329                 lock = cfs_list_entry(ownlocks, struct ldlm_lock, l_res_link);
 330
 331                 if (!ldlm_same_flock_owner(lock, new))
 332                         break;
 333
 334                 if (lock->l_granted_mode == mode) {
 335                         /* If the modes are the same then we need to process
 336                          * locks that overlap OR adjoin the new lock. The extra
 337                          * logic condition is necessary to deal with arithmetic
 338                          * overflow and underflow. */
 339                         if ((new->l_policy_data.l_flock.start >
 340                              (lock->l_policy_data.l_flock.end + 1))
 341                             && (lock->l_policy_data.l_flock.end !=
 342                                 OBD_OBJECT_EOF))
 343                                 continue;
 344
 345                         if ((new->l_policy_data.l_flock.end <
 346                              (lock->l_policy_data.l_flock.start - 1))
 347                             && (lock->l_policy_data.l_flock.start != 0))
 348                                 break;
 349
 350                         if (new->l_policy_data.l_flock.start <
 351                             lock->l_policy_data.l_flock.start) {
 352                                 lock->l_policy_data.l_flock.start =
 353                                         new->l_policy_data.l_flock.start;
 354                         } else {
 355                                 new->l_policy_data.l_flock.start =
 356                                         lock->l_policy_data.l_flock.start;
 357                         }
 358
 359                         if (new->l_policy_data.l_flock.end >
 360                             lock->l_policy_data.l_flock.end) {
 361                                 lock->l_policy_data.l_flock.end =
 362                                         new->l_policy_data.l_flock.end;
 363                         } else {
 364                                 new->l_policy_data.l_flock.end =
 365                                         lock->l_policy_data.l_flock.end;
 366                         }
 367
 368                         if (added) {
 369                                 ldlm_flock_destroy(lock, mode, *flags);
 370                         } else {
 371                                 new = lock;
 372                                 added = 1;
 373                         }
 374                         continue;
 375                 }
 376
 377                 if (new->l_policy_data.l_flock.start >
 378                     lock->l_policy_data.l_flock.end)
 379                         continue;
 380
 381                 if (new->l_policy_data.l_flock.end <
 382                     lock->l_policy_data.l_flock.start)
 383                         break;
 384
 385                 ++overlaps;
 386
 387                 if (new->l_policy_data.l_flock.start <=
 388                     lock->l_policy_data.l_flock.start) {
 389                         if (new->l_policy_data.l_flock.end <
 390                             lock->l_policy_data.l_flock.end) {
 391                                 lock->l_policy_data.l_flock.start =
 392                                         new->l_policy_data.l_flock.end + 1;
 393                                 break;
 394                         }
 395                         ldlm_flock_destroy(lock, lock->l_req_mode, *flags);
 396                         continue;
 397                 }
 398                 if (new->l_policy_data.l_flock.end >=
 399                     lock->l_policy_data.l_flock.end) {
 400                         lock->l_policy_data.l_flock.end =
 401                                 new->l_policy_data.l_flock.start - 1;
 402                         continue;
 403                 }
 404
 405                 /* split the existing lock into two locks */
 406
 407                 /* if this is an F_UNLCK operation then we could avoid
 408                  * allocating a new lock and use the req lock passed in
 409                  * with the request but this would complicate the reply
 410                  * processing since updates to req get reflected in the
 411                  * reply. The client side replays the lock request so
 412                  * it must see the original lock data in the reply. */
 413
 414                 /* XXX - if ldlm_lock_new() can sleep we should
 415                  * release the lr_lock, allocate the new lock,
 416                  * and restart processing this lock. */
 417                 if (!new2) {
 418                         unlock_res_and_lock(req);
 419                          new2 = ldlm_lock_create(ns, &res->lr_name, LDLM_FLOCK,
 420                                         lock->l_granted_mode, &null_cbs,
 421                                         NULL, 0);
 422                         lock_res_and_lock(req);
 423                         if (!new2) {
 424                                 ldlm_flock_destroy(req, lock->l_granted_mode,
 425                                                    *flags);
 426                                 *err = -ENOLCK;
 427                                 RETURN(LDLM_ITER_STOP);
 428                         }
 429                         goto reprocess;
 430                 }
 431
 432                 splitted = 1;
 433
 434                 new2->l_granted_mode = lock->l_granted_mode;
 435                 new2->l_policy_data.l_flock.pid =
 436                         new->l_policy_data.l_flock.pid;
 437                 new2->l_policy_data.l_flock.owner =
 438                         new->l_policy_data.l_flock.owner;
 439                 new2->l_policy_data.l_flock.start =
 440                         lock->l_policy_data.l_flock.start;
 441                 new2->l_policy_data.l_flock.end =
 442                         new->l_policy_data.l_flock.start - 1;
 443                 lock->l_policy_data.l_flock.start =
 444                         new->l_policy_data.l_flock.end + 1;
 445                 new2->l_conn_export = lock->l_conn_export;
 446                 if (lock->l_export != NULL) {
 447                         new2->l_export = class_export_lock_get(lock->l_export, new2);
 448                         if (new2->l_export->exp_lock_hash &&
 449                             cfs_hlist_unhashed(&new2->l_exp_hash))
 450                                 cfs_hash_add(new2->l_export->exp_lock_hash,
 451                                              &new2->l_remote_handle,
 452                                              &new2->l_exp_hash);
 453                 }
 454                 if (*flags == LDLM_FL_WAIT_NOREPROC)
 455                         ldlm_lock_addref_internal_nolock(new2,
 456                                                          lock->l_granted_mode);
 457
 458                 /* insert new2 at lock */
 459                 ldlm_resource_add_lock(res, ownlocks, new2);
 460                 LDLM_LOCK_RELEASE(new2);
 461                 break;
 462         }
 463
 464         /* if new2 is created but never used, destroy it*/
 465         if (splitted == 0 && new2 != NULL)
 466                 ldlm_lock_destroy_nolock(new2);
 467
 468         /* At this point we're granting the lock request. */
 469         req->l_granted_mode = req->l_req_mode;
 470
 471         /* Add req to the granted queue before calling ldlm_reprocess_all(). */
 472         if (!added) {
 473                 cfs_list_del_init(&req->l_res_link);
 474                 /* insert new lock before ownlocks in list. */
 475                 ldlm_resource_add_lock(res, ownlocks, req);
 476         }
 477
 478         if (*flags != LDLM_FL_WAIT_NOREPROC) {
 479 #ifdef HAVE_SERVER_SUPPORT
 480                 if (first_enq) {
 481                         /* If this is an unlock, reprocess the waitq and
 482                          * send completions ASTs for locks that can now be
 483                          * granted. The only problem with doing this
 484                          * reprocessing here is that the completion ASTs for
 485                          * newly granted locks will be sent before the unlock
 486                          * completion is sent. It shouldn't be an issue. Also
 487                          * note that ldlm_process_flock_lock() will recurse,
 488                          * but only once because first_enq will be false from
 489                          * ldlm_reprocess_queue. */
 490                         if ((mode == LCK_NL) && overlaps) {
 491                                 CFS_LIST_HEAD(rpc_list);
 492                                 int rc;
 493 restart:
 494                                 ldlm_reprocess_queue(res, &res->lr_waiting,
 495                                                      &rpc_list);
 496
 497                                 unlock_res_and_lock(req);
 498                                 rc = ldlm_run_ast_work(ns, &rpc_list,
 499                                                        LDLM_WORK_CP_AST);
 500                                 lock_res_and_lock(req);
 501                                 if (rc == -ERESTART)
 502                                         GOTO(restart, -ERESTART);
 503                        }
 504                 } else {
 505                         LASSERT(req->l_completion_ast);
 506                         ldlm_add_ast_work_item(req, NULL, work_list);
 507                 }
 508 #else /* !HAVE_SERVER_SUPPORT */
 509                 /* The only one possible case for client-side calls flock
 510                  * policy function is ldlm_flock_completion_ast inside which
 511                  * carries LDLM_FL_WAIT_NOREPROC flag. */
 512                 CERROR("Illegal parameter for client-side-only module.\n");
 513                 LBUG();
 514 #endif /* HAVE_SERVER_SUPPORT */
 515         }
 516
 517         /* In case we're reprocessing the requested lock we can't destroy
 518          * it until after calling ldlm_ast_work_item() above so that lawi()
 519          * can bump the reference count on req. Otherwise req could be freed
 520          * before the completion AST can be sent.  */
 521         if (added)
 522                 ldlm_flock_destroy(req, mode, *flags);
 523
 524         ldlm_resource_dump(D_INFO, res);
 525         RETURN(LDLM_ITER_CONTINUE);
 526 }
 527
 528 struct ldlm_flock_wait_data {
 529         struct ldlm_lock *fwd_lock;
 530         int               fwd_generation;
 531 };
 532
 533 static void
 534 ldlm_flock_interrupted_wait(void *data)
 535 {
 536         struct ldlm_lock *lock;
 537         ENTRY;
 538
 539         lock = ((struct ldlm_flock_wait_data *)data)->fwd_lock;
 540
 541         /* take lock off the deadlock detection hash list. */
 542         lock_res_and_lock(lock);
 543         ldlm_flock_blocking_unlink(lock);
 544
 545         /* client side - set flag to prevent lock from being put on lru list */
 546         lock->l_flags |= LDLM_FL_CBPENDING;
 547         unlock_res_and_lock(lock);
 548
 549         EXIT;
 550 }
 551
 552 /**
 553  * Flock completion calback function.
 554  *
 555  * \param lock [in,out]: A lock to be handled
 556  * \param flags    [in]: flags
 557  * \param *data    [in]: ldlm_work_cp_ast_lock() will use ldlm_cb_set_arg
 558  *
 559  * \retval 0    : success
 560  * \retval <0   : failure
 561  */
 562 int
 563 ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data)
 564 {
 565         cfs_flock_t                    *getlk = lock->l_ast_data;
 566         struct obd_device              *obd;
 567         struct obd_import              *imp = NULL;
 568         struct ldlm_flock_wait_data     fwd;
 569         struct l_wait_info              lwi;
 570         ldlm_error_t                    err;
 571         int                             rc = 0;
 572         ENTRY;
 573
 574         CDEBUG(D_DLMTRACE, "flags: 0x%x data: %p getlk: %p\n",
 575                flags, data, getlk);
 576
 577         /* Import invalidation. We need to actually release the lock
 578          * references being held, so that it can go away. No point in
 579          * holding the lock even if app still believes it has it, since
 580          * server already dropped it anyway. Only for granted locks too. */
 581         if ((lock->l_flags & (LDLM_FL_FAILED|LDLM_FL_LOCAL_ONLY)) ==
 582             (LDLM_FL_FAILED|LDLM_FL_LOCAL_ONLY)) {
 583                 if (lock->l_req_mode == lock->l_granted_mode &&
 584                     lock->l_granted_mode != LCK_NL &&
 585                     NULL == data)
 586                         ldlm_lock_decref_internal(lock, lock->l_req_mode);
 587
 588                 /* Need to wake up the waiter if we were evicted */
 589                 cfs_waitq_signal(&lock->l_waitq);
 590                 RETURN(0);
 591         }
 592
 593         LASSERT(flags != LDLM_FL_WAIT_NOREPROC);
 594
 595         if (!(flags & (LDLM_FL_BLOCK_WAIT | LDLM_FL_BLOCK_GRANTED |
 596                        LDLM_FL_BLOCK_CONV))) {
 597                 if (NULL == data)
 598                         /* mds granted the lock in the reply */
 599                         goto granted;
 600                 /* CP AST RPC: lock get granted, wake it up */
 601                 cfs_waitq_signal(&lock->l_waitq);
 602                 RETURN(0);
 603         }
 604
 605         LDLM_DEBUG(lock, "client-side enqueue returned a blocked lock, "
 606                    "sleeping");
 607         fwd.fwd_lock = lock;
 608         obd = class_exp2obd(lock->l_conn_export);
 609
 610         /* if this is a local lock, there is no import */
 611         if (NULL != obd)
 612                 imp = obd->u.cli.cl_import;
 613
 614         if (NULL != imp) {
 615                 cfs_spin_lock(&imp->imp_lock);
 616                 fwd.fwd_generation = imp->imp_generation;
 617                 cfs_spin_unlock(&imp->imp_lock);
 618         }
 619
 620         lwi = LWI_TIMEOUT_INTR(0, NULL, ldlm_flock_interrupted_wait, &fwd);
 621
 622         /* Go to sleep until the lock is granted. */
 623         rc = l_wait_event(lock->l_waitq, is_granted_or_cancelled(lock), &lwi);
 624
 625         if (rc) {
 626                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
 627                            rc);
 628                 RETURN(rc);
 629         }
 630
 631 granted:
 632         OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT, 10);
 633
 634         if (lock->l_destroyed) {
 635                 LDLM_DEBUG(lock, "client-side enqueue waking up: destroyed");
 636                 RETURN(0);
 637         }
 638
 639         if (lock->l_flags & LDLM_FL_FAILED) {
 640                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed");
 641                 RETURN(-EIO);
 642         }
 643
 644         if (rc) {
 645                 LDLM_DEBUG(lock, "client-side enqueue waking up: failed (%d)",
 646                            rc);
 647                 RETURN(rc);
 648         }
 649
 650         LDLM_DEBUG(lock, "client-side enqueue granted");
 651
 652         lock_res_and_lock(lock);
 653
 654         /* take lock off the deadlock detection hash list. */
 655         ldlm_flock_blocking_unlink(lock);
 656
 657         /* ldlm_lock_enqueue() has already placed lock on the granted list. */
 658         cfs_list_del_init(&lock->l_res_link);
 659
 660         if (flags & LDLM_FL_TEST_LOCK) {
 661                 /* fcntl(F_GETLK) request */
 662                 /* The old mode was saved in getlk->fl_type so that if the mode
 663                  * in the lock changes we can decref the appropriate refcount.*/
 664                 ldlm_flock_destroy(lock, cfs_flock_type(getlk),
 665                                    LDLM_FL_WAIT_NOREPROC);
 666                 switch (lock->l_granted_mode) {
 667                 case LCK_PR:
 668                         cfs_flock_set_type(getlk, F_RDLCK);
 669                         break;
 670                 case LCK_PW:
 671                         cfs_flock_set_type(getlk, F_WRLCK);
 672                         break;
 673                 default:
 674                         cfs_flock_set_type(getlk, F_UNLCK);
 675                 }
 676                 cfs_flock_set_pid(getlk,
 677                                   (pid_t)lock->l_policy_data.l_flock.pid);
 678                 cfs_flock_set_start(getlk,
 679                                     (loff_t)lock->l_policy_data.l_flock.start);
 680                 cfs_flock_set_end(getlk,
 681                                   (loff_t)lock->l_policy_data.l_flock.end);
 682         } else {
 683                 int noreproc = LDLM_FL_WAIT_NOREPROC;
 684
 685                 /* We need to reprocess the lock to do merges or splits
 686                  * with existing locks owned by this process. */
 687                 ldlm_process_flock_lock(lock, &noreproc, 1, &err, NULL);
 688         }
 689         unlock_res_and_lock(lock);
 690         RETURN(0);
 691 }
 692 EXPORT_SYMBOL(ldlm_flock_completion_ast);
 693
 694 int ldlm_flock_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 695                             void *data, int flag)
 696 {
 697         ENTRY;
 698
 699         LASSERT(lock);
 700         LASSERT(flag == LDLM_CB_CANCELING);
 701
 702         /* take lock off the deadlock detection hash list. */
 703         lock_res_and_lock(lock);
 704         ldlm_flock_blocking_unlink(lock);
 705         unlock_res_and_lock(lock);
 706         RETURN(0);
 707 }
 708
 709 void ldlm_flock_policy_wire18_to_local(const ldlm_wire_policy_data_t *wpolicy,
 710                                        ldlm_policy_data_t *lpolicy)
 711 {
 712         memset(lpolicy, 0, sizeof(*lpolicy));
 713         lpolicy->l_flock.start = wpolicy->l_flock.lfw_start;
 714         lpolicy->l_flock.end = wpolicy->l_flock.lfw_end;
 715         lpolicy->l_flock.pid = wpolicy->l_flock.lfw_pid;
 716         /* Compat code, old clients had no idea about owner field and
 717          * relied solely on pid for ownership. Introduced in LU-104, 2.1,
 718          * April 2011 */
 719         lpolicy->l_flock.owner = wpolicy->l_flock.lfw_pid;
 720 }
 721
 722
 723 void ldlm_flock_policy_wire21_to_local(const ldlm_wire_policy_data_t *wpolicy,
 724                                        ldlm_policy_data_t *lpolicy)
 725 {
 726         memset(lpolicy, 0, sizeof(*lpolicy));
 727         lpolicy->l_flock.start = wpolicy->l_flock.lfw_start;
 728         lpolicy->l_flock.end = wpolicy->l_flock.lfw_end;
 729         lpolicy->l_flock.pid = wpolicy->l_flock.lfw_pid;
 730         lpolicy->l_flock.owner = wpolicy->l_flock.lfw_owner;
 731 }
 732
 733 void ldlm_flock_policy_local_to_wire(const ldlm_policy_data_t *lpolicy,
 734                                      ldlm_wire_policy_data_t *wpolicy)
 735 {
 736         memset(wpolicy, 0, sizeof(*wpolicy));
 737         wpolicy->l_flock.lfw_start = lpolicy->l_flock.start;
 738         wpolicy->l_flock.lfw_end = lpolicy->l_flock.end;
 739         wpolicy->l_flock.lfw_pid = lpolicy->l_flock.pid;
 740         wpolicy->l_flock.lfw_owner = lpolicy->l_flock.owner;
 741 }
 742
 743 /*
 744  * Export handle<->flock hash operations.
 745  */
 746 static unsigned
 747 ldlm_export_flock_hash(cfs_hash_t *hs, const void *key, unsigned mask)
 748 {
 749         return cfs_hash_u64_hash(*(__u64 *)key, mask);
 750 }
 751
 752 static void *
 753 ldlm_export_flock_key(cfs_hlist_node_t *hnode)
 754 {
 755         struct ldlm_lock *lock;
 756
 757         lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 758         return &lock->l_policy_data.l_flock.owner;
 759 }
 760
 761 static int
 762 ldlm_export_flock_keycmp(const void *key, cfs_hlist_node_t *hnode)
 763 {
 764         return !memcmp(ldlm_export_flock_key(hnode), key, sizeof(__u64));
 765 }
 766
 767 static void *
 768 ldlm_export_flock_object(cfs_hlist_node_t *hnode)
 769 {
 770         return cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 771 }
 772
 773 static void
 774 ldlm_export_flock_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 775 {
 776         struct ldlm_lock *lock;
 777         struct ldlm_flock *flock;
 778
 779         lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 780         LDLM_LOCK_GET(lock);
 781
 782         flock = &lock->l_policy_data.l_flock;
 783         LASSERT(flock->blocking_export != NULL);
 784         class_export_get(flock->blocking_export);
 785         flock->blocking_refs++;
 786 }
 787
 788 static void
 789 ldlm_export_flock_put(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 790 {
 791         struct ldlm_lock *lock;
 792         struct ldlm_flock *flock;
 793
 794         lock = cfs_hlist_entry(hnode, struct ldlm_lock, l_exp_flock_hash);
 795         LDLM_LOCK_RELEASE(lock);
 796
 797         flock = &lock->l_policy_data.l_flock;
 798         LASSERT(flock->blocking_export != NULL);
 799         class_export_put(flock->blocking_export);
 800         if (--flock->blocking_refs == 0) {
 801                 flock->blocking_owner = 0;
 802                 flock->blocking_export = NULL;
 803         }
 804 }
 805
 806 static cfs_hash_ops_t ldlm_export_flock_ops = {
 807         .hs_hash        = ldlm_export_flock_hash,
 808         .hs_key         = ldlm_export_flock_key,
 809         .hs_keycmp      = ldlm_export_flock_keycmp,
 810         .hs_object      = ldlm_export_flock_object,
 811         .hs_get         = ldlm_export_flock_get,
 812         .hs_put         = ldlm_export_flock_put,
 813         .hs_put_locked  = ldlm_export_flock_put,
 814 };
 815
 816 int ldlm_init_flock_export(struct obd_export *exp)
 817 {
 818         exp->exp_flock_hash =
 819                 cfs_hash_create(obd_uuid2str(&exp->exp_client_uuid),
 820                                 HASH_EXP_LOCK_CUR_BITS,
 821                                 HASH_EXP_LOCK_MAX_BITS,
 822                                 HASH_EXP_LOCK_BKT_BITS, 0,
 823                                 CFS_HASH_MIN_THETA, CFS_HASH_MAX_THETA,
 824                                 &ldlm_export_flock_ops,
 825                                 CFS_HASH_DEFAULT | CFS_HASH_NBLK_CHANGE);
 826         if (!exp->exp_flock_hash)
 827                 RETURN(-ENOMEM);
 828
 829         RETURN(0);
 830 }
 831 EXPORT_SYMBOL(ldlm_init_flock_export);
 832
 833 void ldlm_destroy_flock_export(struct obd_export *exp)
 834 {
 835         ENTRY;
 836         if (exp->exp_flock_hash) {
 837                 cfs_hash_putref(exp->exp_flock_hash);
 838                 exp->exp_flock_hash = NULL;
 839         }
 840         EXIT;
 841 }
 842 EXPORT_SYMBOL(ldlm_destroy_flock_export);