lustre/osc/osc_lock.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) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * Implementation of cl_lock for OSC layer.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_OSC
  42
  43 #ifdef __KERNEL__
  44 # include <libcfs/libcfs.h>
  45 #else
  46 # include <liblustre.h>
  47 #endif
  48 /* fid_build_reg_res_name() */
  49 #include <lustre_fid.h>
  50
  51 #include "osc_cl_internal.h"
  52
  53 /** \addtogroup osc
  54  *  @{
  55  */
  56
  57 #define _PAGEREF_MAGIC  (-10000000)
  58
  59 /*****************************************************************************
  60  *
  61  * Type conversions.
  62  *
  63  */
  64
  65 static const struct cl_lock_operations osc_lock_ops;
  66 static const struct cl_lock_operations osc_lock_lockless_ops;
  67 static void osc_lock_to_lockless(const struct lu_env *env,
  68                                  struct osc_lock *ols, int force);
  69 static int osc_lock_has_pages(struct osc_lock *olck);
  70
  71 int osc_lock_is_lockless(const struct osc_lock *olck)
  72 {
  73         return (olck->ols_cl.cls_ops == &osc_lock_lockless_ops);
  74 }
  75
  76 /**
  77  * Returns a weak pointer to the ldlm lock identified by a handle. Returned
  78  * pointer cannot be dereferenced, as lock is not protected from concurrent
  79  * reclaim. This function is a helper for osc_lock_invariant().
  80  */
  81 static struct ldlm_lock *osc_handle_ptr(struct lustre_handle *handle)
  82 {
  83         struct ldlm_lock *lock;
  84
  85         lock = ldlm_handle2lock(handle);
  86         if (lock != NULL)
  87                 LDLM_LOCK_PUT(lock);
  88         return lock;
  89 }
  90
  91 /**
  92  * Invariant that has to be true all of the time.
  93  */
  94 static int osc_lock_invariant(struct osc_lock *ols)
  95 {
  96         struct ldlm_lock *lock        = osc_handle_ptr(&ols->ols_handle);
  97         struct ldlm_lock *olock       = ols->ols_lock;
  98         int               handle_used = lustre_handle_is_used(&ols->ols_handle);
  99
 100         return
 101                 ergo(osc_lock_is_lockless(ols),
 102                      ols->ols_locklessable && ols->ols_lock == NULL)  ||
 103                 (ergo(olock != NULL, handle_used) &&
 104                  ergo(olock != NULL,
 105                       olock->l_handle.h_cookie == ols->ols_handle.cookie) &&
 106                  /*
 107                   * Check that ->ols_handle and ->ols_lock are consistent, but
 108                   * take into account that they are set at the different time.
 109                   */
 110                  ergo(handle_used,
 111                       ergo(lock != NULL && olock != NULL, lock == olock) &&
 112                       ergo(lock == NULL, olock == NULL)) &&
 113                  ergo(ols->ols_state == OLS_CANCELLED,
 114                       olock == NULL && !handle_used) &&
 115                  /*
 116                   * DLM lock is destroyed only after we have seen cancellation
 117                   * ast.
 118                   */
 119                  ergo(olock != NULL && ols->ols_state < OLS_CANCELLED,
 120                       !olock->l_destroyed) &&
 121                  ergo(ols->ols_state == OLS_GRANTED,
 122                       olock != NULL &&
 123                       olock->l_req_mode == olock->l_granted_mode &&
 124                       ols->ols_hold));
 125 }
 126
 127 /*****************************************************************************
 128  *
 129  * Lock operations.
 130  *
 131  */
 132
 133 /**
 134  * Breaks a link between osc_lock and dlm_lock.
 135  */
 136 static void osc_lock_detach(const struct lu_env *env, struct osc_lock *olck)
 137 {
 138         struct ldlm_lock *dlmlock;
 139
 140         spin_lock(&osc_ast_guard);
 141         dlmlock = olck->ols_lock;
 142         if (dlmlock == NULL) {
 143                 spin_unlock(&osc_ast_guard);
 144                 return;
 145         }
 146
 147         olck->ols_lock = NULL;
 148         /* wb(); --- for all who checks (ols->ols_lock != NULL) before
 149          * call to osc_lock_detach() */
 150         dlmlock->l_ast_data = NULL;
 151         olck->ols_handle.cookie = 0ULL;
 152         spin_unlock(&osc_ast_guard);
 153
 154         lock_res_and_lock(dlmlock);
 155         if (dlmlock->l_granted_mode == dlmlock->l_req_mode) {
 156                 struct cl_object *obj = olck->ols_cl.cls_obj;
 157                 struct cl_attr *attr  = &osc_env_info(env)->oti_attr;
 158                 __u64 old_kms;
 159
 160                 cl_object_attr_lock(obj);
 161                 /* Must get the value under the lock to avoid possible races. */
 162                 old_kms = cl2osc(obj)->oo_oinfo->loi_kms;
 163                 /* Update the kms. Need to loop all granted locks.
 164                  * Not a problem for the client */
 165                 attr->cat_kms = ldlm_extent_shift_kms(dlmlock, old_kms);
 166
 167                 cl_object_attr_set(env, obj, attr, CAT_KMS);
 168                 cl_object_attr_unlock(obj);
 169         }
 170         unlock_res_and_lock(dlmlock);
 171
 172         /* release a reference taken in osc_lock_upcall0(). */
 173         LASSERT(olck->ols_has_ref);
 174         lu_ref_del(&dlmlock->l_reference, "osc_lock", olck);
 175         LDLM_LOCK_RELEASE(dlmlock);
 176         olck->ols_has_ref = 0;
 177 }
 178
 179 static int osc_lock_unhold(struct osc_lock *ols)
 180 {
 181         int result = 0;
 182
 183         if (ols->ols_hold) {
 184                 ols->ols_hold = 0;
 185                 result = osc_cancel_base(&ols->ols_handle,
 186                                          ols->ols_einfo.ei_mode);
 187         }
 188         return result;
 189 }
 190
 191 static int osc_lock_unuse(const struct lu_env *env,
 192                           const struct cl_lock_slice *slice)
 193 {
 194         struct osc_lock *ols = cl2osc_lock(slice);
 195
 196         LINVRNT(osc_lock_invariant(ols));
 197
 198         switch (ols->ols_state) {
 199         case OLS_NEW:
 200                 LASSERT(!ols->ols_hold);
 201                 LASSERT(ols->ols_agl);
 202                 return 0;
 203         case OLS_UPCALL_RECEIVED:
 204                 osc_lock_unhold(ols);
 205         case OLS_ENQUEUED:
 206                 LASSERT(!ols->ols_hold);
 207                 osc_lock_detach(env, ols);
 208                 ols->ols_state = OLS_NEW;
 209                 return 0;
 210         case OLS_GRANTED:
 211                 LASSERT(!ols->ols_glimpse);
 212                 LASSERT(ols->ols_hold);
 213                 /*
 214                  * Move lock into OLS_RELEASED state before calling
 215                  * osc_cancel_base() so that possible synchronous cancellation
 216                  * (that always happens e.g., for liblustre) sees that lock is
 217                  * released.
 218                  */
 219                 ols->ols_state = OLS_RELEASED;
 220                 return osc_lock_unhold(ols);
 221         default:
 222                 CERROR("Impossible state: %d\n", ols->ols_state);
 223                 LBUG();
 224         }
 225 }
 226
 227 static void osc_lock_fini(const struct lu_env *env,
 228                           struct cl_lock_slice *slice)
 229 {
 230         struct osc_lock  *ols = cl2osc_lock(slice);
 231
 232         LINVRNT(osc_lock_invariant(ols));
 233         /*
 234          * ->ols_hold can still be true at this point if, for example, a
 235          * thread that requested a lock was killed (and released a reference
 236          * to the lock), before reply from a server was received. In this case
 237          * lock is destroyed immediately after upcall.
 238          */
 239         osc_lock_unhold(ols);
 240         LASSERT(ols->ols_lock == NULL);
 241         LASSERT(cfs_atomic_read(&ols->ols_pageref) == 0 ||
 242                 cfs_atomic_read(&ols->ols_pageref) == _PAGEREF_MAGIC);
 243
 244         OBD_SLAB_FREE_PTR(ols, osc_lock_kmem);
 245 }
 246
 247 static void osc_lock_build_policy(const struct lu_env *env,
 248                                   const struct cl_lock *lock,
 249                                   ldlm_policy_data_t *policy)
 250 {
 251         const struct cl_lock_descr *d = &lock->cll_descr;
 252
 253         osc_index2policy(policy, d->cld_obj, d->cld_start, d->cld_end);
 254         policy->l_extent.gid = d->cld_gid;
 255 }
 256
 257 static __u64 osc_enq2ldlm_flags(__u32 enqflags)
 258 {
 259         __u64 result = 0;
 260
 261         LASSERT((enqflags & ~CEF_MASK) == 0);
 262
 263         if (enqflags & CEF_NONBLOCK)
 264                 result |= LDLM_FL_BLOCK_NOWAIT;
 265         if (enqflags & CEF_ASYNC)
 266                 result |= LDLM_FL_HAS_INTENT;
 267         if (enqflags & CEF_DISCARD_DATA)
 268                 result |= LDLM_AST_DISCARD_DATA;
 269         return result;
 270 }
 271
 272 /**
 273  * Global spin-lock protecting consistency of ldlm_lock::l_ast_data
 274  * pointers. Initialized in osc_init().
 275  */
 276 spinlock_t osc_ast_guard;
 277
 278 static struct osc_lock *osc_ast_data_get(struct ldlm_lock *dlm_lock)
 279 {
 280         struct osc_lock *olck;
 281
 282         lock_res_and_lock(dlm_lock);
 283         spin_lock(&osc_ast_guard);
 284         olck = dlm_lock->l_ast_data;
 285         if (olck != NULL) {
 286                 struct cl_lock *lock = olck->ols_cl.cls_lock;
 287                 /*
 288                  * If osc_lock holds a reference on ldlm lock, return it even
 289                  * when cl_lock is in CLS_FREEING state. This way
 290                  *
 291                  *         osc_ast_data_get(dlmlock) == NULL
 292                  *
 293                  * guarantees that all osc references on dlmlock were
 294                  * released. osc_dlm_blocking_ast0() relies on that.
 295                  */
 296                 if (lock->cll_state < CLS_FREEING || olck->ols_has_ref) {
 297                         cl_lock_get_trust(lock);
 298                         lu_ref_add_atomic(&lock->cll_reference,
 299                                           "ast", cfs_current());
 300                 } else
 301                         olck = NULL;
 302         }
 303         spin_unlock(&osc_ast_guard);
 304         unlock_res_and_lock(dlm_lock);
 305         return olck;
 306 }
 307
 308 static void osc_ast_data_put(const struct lu_env *env, struct osc_lock *olck)
 309 {
 310         struct cl_lock *lock;
 311
 312         lock = olck->ols_cl.cls_lock;
 313         lu_ref_del(&lock->cll_reference, "ast", cfs_current());
 314         cl_lock_put(env, lock);
 315 }
 316
 317 /**
 318  * Updates object attributes from a lock value block (lvb) received together
 319  * with the DLM lock reply from the server. Copy of osc_update_enqueue()
 320  * logic.
 321  *
 322  * This can be optimized to not update attributes when lock is a result of a
 323  * local match.
 324  *
 325  * Called under lock and resource spin-locks.
 326  */
 327 static void osc_lock_lvb_update(const struct lu_env *env, struct osc_lock *olck,
 328                                 int rc)
 329 {
 330         struct ost_lvb    *lvb;
 331         struct cl_object  *obj;
 332         struct lov_oinfo  *oinfo;
 333         struct cl_attr    *attr;
 334         unsigned           valid;
 335
 336         ENTRY;
 337
 338         if (!(olck->ols_flags & LDLM_FL_LVB_READY))
 339                 RETURN_EXIT;
 340
 341         lvb   = &olck->ols_lvb;
 342         obj   = olck->ols_cl.cls_obj;
 343         oinfo = cl2osc(obj)->oo_oinfo;
 344         attr  = &osc_env_info(env)->oti_attr;
 345         valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME | CAT_SIZE;
 346         cl_lvb2attr(attr, lvb);
 347
 348         cl_object_attr_lock(obj);
 349         if (rc == 0) {
 350                 struct ldlm_lock  *dlmlock;
 351                 __u64 size;
 352
 353                 dlmlock = olck->ols_lock;
 354                 LASSERT(dlmlock != NULL);
 355
 356                 /* re-grab LVB from a dlm lock under DLM spin-locks. */
 357                 *lvb = *(struct ost_lvb *)dlmlock->l_lvb_data;
 358                 size = lvb->lvb_size;
 359                 /* Extend KMS up to the end of this lock and no further
 360                  * A lock on [x,y] means a KMS of up to y + 1 bytes! */
 361                 if (size > dlmlock->l_policy_data.l_extent.end)
 362                         size = dlmlock->l_policy_data.l_extent.end + 1;
 363                 if (size >= oinfo->loi_kms) {
 364                         LDLM_DEBUG(dlmlock, "lock acquired, setting rss="LPU64
 365                                    ", kms="LPU64, lvb->lvb_size, size);
 366                         valid |= CAT_KMS;
 367                         attr->cat_kms = size;
 368                 } else {
 369                         LDLM_DEBUG(dlmlock, "lock acquired, setting rss="
 370                                    LPU64"; leaving kms="LPU64", end="LPU64,
 371                                    lvb->lvb_size, oinfo->loi_kms,
 372                                    dlmlock->l_policy_data.l_extent.end);
 373                 }
 374                 ldlm_lock_allow_match_locked(dlmlock);
 375         } else if (rc == -ENAVAIL && olck->ols_glimpse) {
 376                 CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
 377                        " kms="LPU64"\n", lvb->lvb_size, oinfo->loi_kms);
 378         } else
 379                 valid = 0;
 380
 381         if (valid != 0)
 382                 cl_object_attr_set(env, obj, attr, valid);
 383
 384         cl_object_attr_unlock(obj);
 385
 386         EXIT;
 387 }
 388
 389 /**
 390  * Called when a lock is granted, from an upcall (when server returned a
 391  * granted lock), or from completion AST, when server returned a blocked lock.
 392  *
 393  * Called under lock and resource spin-locks, that are released temporarily
 394  * here.
 395  */
 396 static void osc_lock_granted(const struct lu_env *env, struct osc_lock *olck,
 397                              struct ldlm_lock *dlmlock, int rc)
 398 {
 399         struct ldlm_extent   *ext;
 400         struct cl_lock       *lock;
 401         struct cl_lock_descr *descr;
 402
 403         LASSERT(dlmlock->l_granted_mode == dlmlock->l_req_mode);
 404
 405         ENTRY;
 406         if (olck->ols_state < OLS_GRANTED) {
 407                 lock  = olck->ols_cl.cls_lock;
 408                 ext   = &dlmlock->l_policy_data.l_extent;
 409                 descr = &osc_env_info(env)->oti_descr;
 410                 descr->cld_obj = lock->cll_descr.cld_obj;
 411
 412                 /* XXX check that ->l_granted_mode is valid. */
 413                 descr->cld_mode  = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
 414                 descr->cld_start = cl_index(descr->cld_obj, ext->start);
 415                 descr->cld_end   = cl_index(descr->cld_obj, ext->end);
 416                 descr->cld_gid   = ext->gid;
 417                 /*
 418                  * tell upper layers the extent of the lock that was actually
 419                  * granted
 420                  */
 421                 olck->ols_state = OLS_GRANTED;
 422                 osc_lock_lvb_update(env, olck, rc);
 423
 424                 /* release DLM spin-locks to allow cl_lock_{modify,signal}()
 425                  * to take a semaphore on a parent lock. This is safe, because
 426                  * spin-locks are needed to protect consistency of
 427                  * dlmlock->l_*_mode and LVB, and we have finished processing
 428                  * them. */
 429                 unlock_res_and_lock(dlmlock);
 430                 cl_lock_modify(env, lock, descr);
 431                 cl_lock_signal(env, lock);
 432                 LINVRNT(osc_lock_invariant(olck));
 433                 lock_res_and_lock(dlmlock);
 434         }
 435         EXIT;
 436 }
 437
 438 static void osc_lock_upcall0(const struct lu_env *env, struct osc_lock *olck)
 439
 440 {
 441         struct ldlm_lock *dlmlock;
 442
 443         ENTRY;
 444
 445         dlmlock = ldlm_handle2lock_long(&olck->ols_handle, 0);
 446         LASSERT(dlmlock != NULL);
 447
 448         lock_res_and_lock(dlmlock);
 449         spin_lock(&osc_ast_guard);
 450         LASSERT(dlmlock->l_ast_data == olck);
 451         LASSERT(olck->ols_lock == NULL);
 452         olck->ols_lock = dlmlock;
 453         spin_unlock(&osc_ast_guard);
 454
 455         /*
 456          * Lock might be not yet granted. In this case, completion ast
 457          * (osc_ldlm_completion_ast()) comes later and finishes lock
 458          * granting.
 459          */
 460         if (dlmlock->l_granted_mode == dlmlock->l_req_mode)
 461                 osc_lock_granted(env, olck, dlmlock, 0);
 462         unlock_res_and_lock(dlmlock);
 463
 464         /*
 465          * osc_enqueue_interpret() decrefs asynchronous locks, counter
 466          * this.
 467          */
 468         ldlm_lock_addref(&olck->ols_handle, olck->ols_einfo.ei_mode);
 469         olck->ols_hold = 1;
 470
 471         /* lock reference taken by ldlm_handle2lock_long() is owned by
 472          * osc_lock and released in osc_lock_detach() */
 473         lu_ref_add(&dlmlock->l_reference, "osc_lock", olck);
 474         olck->ols_has_ref = 1;
 475 }
 476
 477 /**
 478  * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
 479  * received from a server, or after osc_enqueue_base() matched a local DLM
 480  * lock.
 481  */
 482 static int osc_lock_upcall(void *cookie, int errcode)
 483 {
 484         struct osc_lock         *olck  = cookie;
 485         struct cl_lock_slice    *slice = &olck->ols_cl;
 486         struct cl_lock          *lock  = slice->cls_lock;
 487         struct lu_env           *env;
 488         struct cl_env_nest       nest;
 489
 490         ENTRY;
 491         env = cl_env_nested_get(&nest);
 492         if (!IS_ERR(env)) {
 493                 int rc;
 494
 495                 cl_lock_mutex_get(env, lock);
 496
 497                 LASSERT(lock->cll_state >= CLS_QUEUING);
 498                 if (olck->ols_state == OLS_ENQUEUED) {
 499                         olck->ols_state = OLS_UPCALL_RECEIVED;
 500                         rc = ldlm_error2errno(errcode);
 501                 } else if (olck->ols_state == OLS_CANCELLED) {
 502                         rc = -EIO;
 503                 } else {
 504                         CERROR("Impossible state: %d\n", olck->ols_state);
 505                         LBUG();
 506                 }
 507                 if (rc) {
 508                         struct ldlm_lock *dlmlock;
 509
 510                         dlmlock = ldlm_handle2lock(&olck->ols_handle);
 511                         if (dlmlock != NULL) {
 512                                 lock_res_and_lock(dlmlock);
 513                                 spin_lock(&osc_ast_guard);
 514                                 LASSERT(olck->ols_lock == NULL);
 515                                 dlmlock->l_ast_data = NULL;
 516                                 olck->ols_handle.cookie = 0ULL;
 517                                 spin_unlock(&osc_ast_guard);
 518                                 ldlm_lock_fail_match_locked(dlmlock);
 519                                 unlock_res_and_lock(dlmlock);
 520                                 LDLM_LOCK_PUT(dlmlock);
 521                         }
 522                 } else {
 523                         if (olck->ols_glimpse)
 524                                 olck->ols_glimpse = 0;
 525                         osc_lock_upcall0(env, olck);
 526                 }
 527
 528                 /* Error handling, some errors are tolerable. */
 529                 if (olck->ols_locklessable && rc == -EUSERS) {
 530                         /* This is a tolerable error, turn this lock into
 531                          * lockless lock.
 532                          */
 533                         osc_object_set_contended(cl2osc(slice->cls_obj));
 534                         LASSERT(slice->cls_ops == &osc_lock_ops);
 535
 536                         /* Change this lock to ldlmlock-less lock. */
 537                         osc_lock_to_lockless(env, olck, 1);
 538                         olck->ols_state = OLS_GRANTED;
 539                         rc = 0;
 540                 } else if (olck->ols_glimpse && rc == -ENAVAIL) {
 541                         osc_lock_lvb_update(env, olck, rc);
 542                         cl_lock_delete(env, lock);
 543                         /* Hide the error. */
 544                         rc = 0;
 545                 }
 546
 547                 if (rc == 0) {
 548                         /* For AGL case, the RPC sponsor may exits the cl_lock
 549                         *  processing without wait() called before related OSC
 550                         *  lock upcall(). So update the lock status according
 551                         *  to the enqueue result inside AGL upcall(). */
 552                         if (olck->ols_agl) {
 553                                 lock->cll_flags |= CLF_FROM_UPCALL;
 554                                 cl_wait_try(env, lock);
 555                                 lock->cll_flags &= ~CLF_FROM_UPCALL;
 556                                 if (!olck->ols_glimpse)
 557                                         olck->ols_agl = 0;
 558                         }
 559                         cl_lock_signal(env, lock);
 560                         /* del user for lock upcall cookie */
 561                         cl_unuse_try(env, lock);
 562                 } else {
 563                         /* del user for lock upcall cookie */
 564                         cl_lock_user_del(env, lock);
 565                         cl_lock_error(env, lock, rc);
 566                 }
 567
 568                 /* release cookie reference, acquired by osc_lock_enqueue() */
 569                 cl_lock_hold_release(env, lock, "upcall", lock);
 570                 cl_lock_mutex_put(env, lock);
 571
 572                 lu_ref_del(&lock->cll_reference, "upcall", lock);
 573                 /* This maybe the last reference, so must be called after
 574                  * cl_lock_mutex_put(). */
 575                 cl_lock_put(env, lock);
 576
 577                 cl_env_nested_put(&nest, env);
 578         } else {
 579                 /* should never happen, similar to osc_ldlm_blocking_ast(). */
 580                 LBUG();
 581         }
 582         RETURN(errcode);
 583 }
 584
 585 /**
 586  * Core of osc_dlm_blocking_ast() logic.
 587  */
 588 static void osc_lock_blocking(const struct lu_env *env,
 589                               struct ldlm_lock *dlmlock,
 590                               struct osc_lock *olck, int blocking)
 591 {
 592         struct cl_lock *lock = olck->ols_cl.cls_lock;
 593
 594         LASSERT(olck->ols_lock == dlmlock);
 595         CLASSERT(OLS_BLOCKED < OLS_CANCELLED);
 596         LASSERT(!osc_lock_is_lockless(olck));
 597
 598         /*
 599          * Lock might be still addref-ed here, if e.g., blocking ast
 600          * is sent for a failed lock.
 601          */
 602         osc_lock_unhold(olck);
 603
 604         if (blocking && olck->ols_state < OLS_BLOCKED)
 605                 /*
 606                  * Move osc_lock into OLS_BLOCKED before canceling the lock,
 607                  * because it recursively re-enters osc_lock_blocking(), with
 608                  * the state set to OLS_CANCELLED.
 609                  */
 610                 olck->ols_state = OLS_BLOCKED;
 611         /*
 612          * cancel and destroy lock at least once no matter how blocking ast is
 613          * entered (see comment above osc_ldlm_blocking_ast() for use
 614          * cases). cl_lock_cancel() and cl_lock_delete() are idempotent.
 615          */
 616         cl_lock_cancel(env, lock);
 617         cl_lock_delete(env, lock);
 618 }
 619
 620 /**
 621  * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
 622  * and ldlm_lock caches.
 623  */
 624 static int osc_dlm_blocking_ast0(const struct lu_env *env,
 625                                  struct ldlm_lock *dlmlock,
 626                                  void *data, int flag)
 627 {
 628         struct osc_lock *olck;
 629         struct cl_lock  *lock;
 630         int result;
 631         int cancel;
 632
 633         LASSERT(flag == LDLM_CB_BLOCKING || flag == LDLM_CB_CANCELING);
 634
 635         cancel = 0;
 636         olck = osc_ast_data_get(dlmlock);
 637         if (olck != NULL) {
 638                 lock = olck->ols_cl.cls_lock;
 639                 cl_lock_mutex_get(env, lock);
 640                 LINVRNT(osc_lock_invariant(olck));
 641                 if (olck->ols_ast_wait) {
 642                         /* wake up osc_lock_use() */
 643                         cl_lock_signal(env, lock);
 644                         olck->ols_ast_wait = 0;
 645                 }
 646                 /*
 647                  * Lock might have been canceled while this thread was
 648                  * sleeping for lock mutex, but olck is pinned in memory.
 649                  */
 650                 if (olck == dlmlock->l_ast_data) {
 651                         /*
 652                          * NOTE: DLM sends blocking AST's for failed locks
 653                          *       (that are still in pre-OLS_GRANTED state)
 654                          *       too, and they have to be canceled otherwise
 655                          *       DLM lock is never destroyed and stuck in
 656                          *       the memory.
 657                          *
 658                          *       Alternatively, ldlm_cli_cancel() can be
 659                          *       called here directly for osc_locks with
 660                          *       ols_state < OLS_GRANTED to maintain an
 661                          *       invariant that ->clo_cancel() is only called
 662                          *       for locks that were granted.
 663                          */
 664                         LASSERT(data == olck);
 665                         osc_lock_blocking(env, dlmlock,
 666                                           olck, flag == LDLM_CB_BLOCKING);
 667                 } else
 668                         cancel = 1;
 669                 cl_lock_mutex_put(env, lock);
 670                 osc_ast_data_put(env, olck);
 671         } else
 672                 /*
 673                  * DLM lock exists, but there is no cl_lock attached to it.
 674                  * This is a `normal' race. cl_object and its cl_lock's can be
 675                  * removed by memory pressure, together with all pages.
 676                  */
 677                 cancel = (flag == LDLM_CB_BLOCKING);
 678
 679         if (cancel) {
 680                 struct lustre_handle *lockh;
 681
 682                 lockh = &osc_env_info(env)->oti_handle;
 683                 ldlm_lock2handle(dlmlock, lockh);
 684                 result = ldlm_cli_cancel(lockh, LCF_ASYNC);
 685         } else
 686                 result = 0;
 687         return result;
 688 }
 689
 690 /**
 691  * Blocking ast invoked by ldlm when dlm lock is either blocking progress of
 692  * some other lock, or is canceled. This function is installed as a
 693  * ldlm_lock::l_blocking_ast() for client extent locks.
 694  *
 695  * Control flow is tricky, because ldlm uses the same call-back
 696  * (ldlm_lock::l_blocking_ast()) for both blocking and cancellation ast's.
 697  *
 698  * \param dlmlock lock for which ast occurred.
 699  *
 700  * \param new description of a conflicting lock in case of blocking ast.
 701  *
 702  * \param data value of dlmlock->l_ast_data
 703  *
 704  * \param flag LDLM_CB_BLOCKING or LDLM_CB_CANCELING. Used to distinguish
 705  *             cancellation and blocking ast's.
 706  *
 707  * Possible use cases:
 708  *
 709  *     - ldlm calls dlmlock->l_blocking_ast(..., LDLM_CB_CANCELING) to cancel
 710  *       lock due to lock lru pressure, or explicit user request to purge
 711  *       locks.
 712  *
 713  *     - ldlm calls dlmlock->l_blocking_ast(..., LDLM_CB_BLOCKING) to notify
 714  *       us that dlmlock conflicts with another lock that some client is
 715  *       enqueing. Lock is canceled.
 716  *
 717  *           - cl_lock_cancel() is called. osc_lock_cancel() calls
 718  *             ldlm_cli_cancel() that calls
 719  *
 720  *                  dlmlock->l_blocking_ast(..., LDLM_CB_CANCELING)
 721  *
 722  *             recursively entering osc_ldlm_blocking_ast().
 723  *
 724  *     - client cancels lock voluntary (e.g., as a part of early cancellation):
 725  *
 726  *           cl_lock_cancel()->
 727  *             osc_lock_cancel()->
 728  *               ldlm_cli_cancel()->
 729  *                 dlmlock->l_blocking_ast(..., LDLM_CB_CANCELING)
 730  *
 731  */
 732 static int osc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
 733                                  struct ldlm_lock_desc *new, void *data,
 734                                  int flag)
 735 {
 736         struct lu_env     *env;
 737         struct cl_env_nest nest;
 738         int                result;
 739
 740         /*
 741          * This can be called in the context of outer IO, e.g.,
 742          *
 743          *     cl_enqueue()->...
 744          *       ->osc_enqueue_base()->...
 745          *         ->ldlm_prep_elc_req()->...
 746          *           ->ldlm_cancel_callback()->...
 747          *             ->osc_ldlm_blocking_ast()
 748          *
 749          * new environment has to be created to not corrupt outer context.
 750          */
 751         env = cl_env_nested_get(&nest);
 752         if (!IS_ERR(env)) {
 753                 result = osc_dlm_blocking_ast0(env, dlmlock, data, flag);
 754                 cl_env_nested_put(&nest, env);
 755         } else {
 756                 result = PTR_ERR(env);
 757                 /*
 758                  * XXX This should never happen, as cl_lock is
 759                  * stuck. Pre-allocated environment a la vvp_inode_fini_env
 760                  * should be used.
 761                  */
 762                 LBUG();
 763         }
 764         if (result != 0) {
 765                 if (result == -ENODATA)
 766                         result = 0;
 767                 else
 768                         CERROR("BAST failed: %d\n", result);
 769         }
 770         return result;
 771 }
 772
 773 static int osc_ldlm_completion_ast(struct ldlm_lock *dlmlock,
 774                                    __u64 flags, void *data)
 775 {
 776         struct cl_env_nest nest;
 777         struct lu_env     *env;
 778         struct osc_lock   *olck;
 779         struct cl_lock    *lock;
 780         int result;
 781         int dlmrc;
 782
 783         /* first, do dlm part of the work */
 784         dlmrc = ldlm_completion_ast_async(dlmlock, flags, data);
 785         /* then, notify cl_lock */
 786         env = cl_env_nested_get(&nest);
 787         if (!IS_ERR(env)) {
 788                 olck = osc_ast_data_get(dlmlock);
 789                 if (olck != NULL) {
 790                         lock = olck->ols_cl.cls_lock;
 791                         cl_lock_mutex_get(env, lock);
 792                         /*
 793                          * ldlm_handle_cp_callback() copied LVB from request
 794                          * to lock->l_lvb_data, store it in osc_lock.
 795                          */
 796                         LASSERT(dlmlock->l_lvb_data != NULL);
 797                         lock_res_and_lock(dlmlock);
 798                         olck->ols_lvb = *(struct ost_lvb *)dlmlock->l_lvb_data;
 799                         if (olck->ols_lock == NULL) {
 800                                 /*
 801                                  * upcall (osc_lock_upcall()) hasn't yet been
 802                                  * called. Do nothing now, upcall will bind
 803                                  * olck to dlmlock and signal the waiters.
 804                                  *
 805                                  * This maintains an invariant that osc_lock
 806                                  * and ldlm_lock are always bound when
 807                                  * osc_lock is in OLS_GRANTED state.
 808                                  */
 809                         } else if (dlmlock->l_granted_mode ==
 810                                    dlmlock->l_req_mode) {
 811                                 osc_lock_granted(env, olck, dlmlock, dlmrc);
 812                         }
 813                         unlock_res_and_lock(dlmlock);
 814
 815                         if (dlmrc != 0) {
 816                                 CL_LOCK_DEBUG(D_ERROR, env, lock,
 817                                               "dlmlock returned %d\n", dlmrc);
 818                                 cl_lock_error(env, lock, dlmrc);
 819                         }
 820                         cl_lock_mutex_put(env, lock);
 821                         osc_ast_data_put(env, olck);
 822                         result = 0;
 823                 } else
 824                         result = -ELDLM_NO_LOCK_DATA;
 825                 cl_env_nested_put(&nest, env);
 826         } else
 827                 result = PTR_ERR(env);
 828         return dlmrc ?: result;
 829 }
 830
 831 static int osc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data)
 832 {
 833         struct ptlrpc_request  *req  = data;
 834         struct osc_lock        *olck;
 835         struct cl_lock         *lock;
 836         struct cl_object       *obj;
 837         struct cl_env_nest      nest;
 838         struct lu_env          *env;
 839         struct ost_lvb         *lvb;
 840         struct req_capsule     *cap;
 841         int                     result;
 842
 843         LASSERT(lustre_msg_get_opc(req->rq_reqmsg) == LDLM_GL_CALLBACK);
 844
 845         env = cl_env_nested_get(&nest);
 846         if (!IS_ERR(env)) {
 847                 /* osc_ast_data_get() has to go after environment is
 848                  * allocated, because osc_ast_data() acquires a
 849                  * reference to a lock, and it can only be released in
 850                  * environment.
 851                  */
 852                 olck = osc_ast_data_get(dlmlock);
 853                 if (olck != NULL) {
 854                         lock = olck->ols_cl.cls_lock;
 855                         /* Do not grab the mutex of cl_lock for glimpse.
 856                          * See LU-1274 for details.
 857                          * BTW, it's okay for cl_lock to be cancelled during
 858                          * this period because server can handle this race.
 859                          * See ldlm_server_glimpse_ast() for details.
 860                          * cl_lock_mutex_get(env, lock); */
 861                         cap = &req->rq_pill;
 862                         req_capsule_extend(cap, &RQF_LDLM_GL_CALLBACK);
 863                         req_capsule_set_size(cap, &RMF_DLM_LVB, RCL_SERVER,
 864                                              sizeof *lvb);
 865                         result = req_capsule_server_pack(cap);
 866                         if (result == 0) {
 867                                 lvb = req_capsule_server_get(cap, &RMF_DLM_LVB);
 868                                 obj = lock->cll_descr.cld_obj;
 869                                 result = cl_object_glimpse(env, obj, lvb);
 870                         }
 871                         if (!exp_connect_lvb_type(req->rq_export))
 872                                 req_capsule_shrink(&req->rq_pill,
 873                                                    &RMF_DLM_LVB,
 874                                                    sizeof(struct ost_lvb_v1),
 875                                                    RCL_SERVER);
 876                         osc_ast_data_put(env, olck);
 877                 } else {
 878                         /*
 879                          * These errors are normal races, so we don't want to
 880                          * fill the console with messages by calling
 881                          * ptlrpc_error()
 882                          */
 883                         lustre_pack_reply(req, 1, NULL, NULL);
 884                         result = -ELDLM_NO_LOCK_DATA;
 885                 }
 886                 cl_env_nested_put(&nest, env);
 887         } else
 888                 result = PTR_ERR(env);
 889         req->rq_status = result;
 890         return result;
 891 }
 892
 893 static unsigned long osc_lock_weigh(const struct lu_env *env,
 894                                     const struct cl_lock_slice *slice)
 895 {
 896         /*
 897          * don't need to grab coh_page_guard since we don't care the exact #
 898          * of pages..
 899          */
 900         return cl_object_header(slice->cls_obj)->coh_pages;
 901 }
 902
 903 /**
 904  * Get the weight of dlm lock for early cancellation.
 905  *
 906  * XXX: it should return the pages covered by this \a dlmlock.
 907  */
 908 static unsigned long osc_ldlm_weigh_ast(struct ldlm_lock *dlmlock)
 909 {
 910         struct cl_env_nest       nest;
 911         struct lu_env           *env;
 912         struct osc_lock         *lock;
 913         struct cl_lock          *cll;
 914         unsigned long            weight;
 915         ENTRY;
 916
 917         cfs_might_sleep();
 918         /*
 919          * osc_ldlm_weigh_ast has a complex context since it might be called
 920          * because of lock canceling, or from user's input. We have to make
 921          * a new environment for it. Probably it is implementation safe to use
 922          * the upper context because cl_lock_put don't modify environment
 923          * variables. But in case of ..
 924          */
 925         env = cl_env_nested_get(&nest);
 926         if (IS_ERR(env))
 927                 /* Mostly because lack of memory, tend to eliminate this lock*/
 928                 RETURN(0);
 929
 930         LASSERT(dlmlock->l_resource->lr_type == LDLM_EXTENT);
 931         lock = osc_ast_data_get(dlmlock);
 932         if (lock == NULL) {
 933                 /* cl_lock was destroyed because of memory pressure.
 934                  * It is much reasonable to assign this type of lock
 935                  * a lower cost.
 936                  */
 937                 GOTO(out, weight = 0);
 938         }
 939
 940         cll = lock->ols_cl.cls_lock;
 941         cl_lock_mutex_get(env, cll);
 942         weight = cl_lock_weigh(env, cll);
 943         cl_lock_mutex_put(env, cll);
 944         osc_ast_data_put(env, lock);
 945         EXIT;
 946
 947 out:
 948         cl_env_nested_put(&nest, env);
 949         return weight;
 950 }
 951
 952 static void osc_lock_build_einfo(const struct lu_env *env,
 953                                  const struct cl_lock *clock,
 954                                  struct osc_lock *lock,
 955                                  struct ldlm_enqueue_info *einfo)
 956 {
 957         enum cl_lock_mode mode;
 958
 959         mode = clock->cll_descr.cld_mode;
 960         if (mode == CLM_PHANTOM)
 961                 /*
 962                  * For now, enqueue all glimpse locks in read mode. In the
 963                  * future, client might choose to enqueue LCK_PW lock for
 964                  * glimpse on a file opened for write.
 965                  */
 966                 mode = CLM_READ;
 967
 968         einfo->ei_type   = LDLM_EXTENT;
 969         einfo->ei_mode   = osc_cl_lock2ldlm(mode);
 970         einfo->ei_cb_bl  = osc_ldlm_blocking_ast;
 971         einfo->ei_cb_cp  = osc_ldlm_completion_ast;
 972         einfo->ei_cb_gl  = osc_ldlm_glimpse_ast;
 973         einfo->ei_cb_wg  = osc_ldlm_weigh_ast;
 974         einfo->ei_cbdata = lock; /* value to be put into ->l_ast_data */
 975 }
 976
 977 /**
 978  * Determine if the lock should be converted into a lockless lock.
 979  *
 980  * Steps to check:
 981  * - if the lock has an explicite requirment for a non-lockless lock;
 982  * - if the io lock request type ci_lockreq;
 983  * - send the enqueue rpc to ost to make the further decision;
 984  * - special treat to truncate lockless lock
 985  *
 986  *  Additional policy can be implemented here, e.g., never do lockless-io
 987  *  for large extents.
 988  */
 989 static void osc_lock_to_lockless(const struct lu_env *env,
 990                                  struct osc_lock *ols, int force)
 991 {
 992         struct cl_lock_slice *slice = &ols->ols_cl;
 993
 994         LASSERT(ols->ols_state == OLS_NEW ||
 995                 ols->ols_state == OLS_UPCALL_RECEIVED);
 996
 997         if (force) {
 998                 ols->ols_locklessable = 1;
 999                 slice->cls_ops = &osc_lock_lockless_ops;
1000         } else {
1001                 struct osc_io *oio     = osc_env_io(env);
1002                 struct cl_io  *io      = oio->oi_cl.cis_io;
1003                 struct cl_object *obj  = slice->cls_obj;
1004                 struct osc_object *oob = cl2osc(obj);
1005                 const struct osc_device *osd = lu2osc_dev(obj->co_lu.lo_dev);
1006                 struct obd_connect_data *ocd;
1007
1008                 LASSERT(io->ci_lockreq == CILR_MANDATORY ||
1009                         io->ci_lockreq == CILR_MAYBE ||
1010                         io->ci_lockreq == CILR_NEVER);
1011
1012                 ocd = &class_exp2cliimp(osc_export(oob))->imp_connect_data;
1013                 ols->ols_locklessable = (io->ci_type != CIT_SETATTR) &&
1014                                 (io->ci_lockreq == CILR_MAYBE) &&
1015                                 (ocd->ocd_connect_flags & OBD_CONNECT_SRVLOCK);
1016                 if (io->ci_lockreq == CILR_NEVER ||
1017                         /* lockless IO */
1018                     (ols->ols_locklessable && osc_object_is_contended(oob)) ||
1019                         /* lockless truncate */
1020                     (cl_io_is_trunc(io) &&
1021                      (ocd->ocd_connect_flags & OBD_CONNECT_TRUNCLOCK) &&
1022                       osd->od_lockless_truncate)) {
1023                         ols->ols_locklessable = 1;
1024                         slice->cls_ops = &osc_lock_lockless_ops;
1025                 }
1026         }
1027         LASSERT(ergo(ols->ols_glimpse, !osc_lock_is_lockless(ols)));
1028 }
1029
1030 static int osc_lock_compatible(const struct osc_lock *qing,
1031                                const struct osc_lock *qed)
1032 {
1033         enum cl_lock_mode qing_mode;
1034         enum cl_lock_mode qed_mode;
1035
1036         qing_mode = qing->ols_cl.cls_lock->cll_descr.cld_mode;
1037         if (qed->ols_glimpse &&
1038             (qed->ols_state >= OLS_UPCALL_RECEIVED || qing_mode == CLM_READ))
1039                 return 1;
1040
1041         qed_mode = qed->ols_cl.cls_lock->cll_descr.cld_mode;
1042         return ((qing_mode == CLM_READ) && (qed_mode == CLM_READ));
1043 }
1044
1045 /**
1046  * Cancel all conflicting locks and wait for them to be destroyed.
1047  *
1048  * This function is used for two purposes:
1049  *
1050  *     - early cancel all conflicting locks before starting IO, and
1051  *
1052  *     - guarantee that pages added to the page cache by lockless IO are never
1053  *       covered by locks other than lockless IO lock, and, hence, are not
1054  *       visible to other threads.
1055  */
1056 static int osc_lock_enqueue_wait(const struct lu_env *env,
1057                                  const struct osc_lock *olck)
1058 {
1059         struct cl_lock          *lock    = olck->ols_cl.cls_lock;
1060         struct cl_lock_descr    *descr   = &lock->cll_descr;
1061         struct cl_object_header *hdr     = cl_object_header(descr->cld_obj);
1062         struct cl_lock          *scan;
1063         struct cl_lock          *conflict= NULL;
1064         int lockless                     = osc_lock_is_lockless(olck);
1065         int rc                           = 0;
1066         ENTRY;
1067
1068         LASSERT(cl_lock_is_mutexed(lock));
1069
1070         /* make it enqueue anyway for glimpse lock, because we actually
1071          * don't need to cancel any conflicting locks. */
1072         if (olck->ols_glimpse)
1073                 return 0;
1074
1075         spin_lock(&hdr->coh_lock_guard);
1076         cfs_list_for_each_entry(scan, &hdr->coh_locks, cll_linkage) {
1077                 struct cl_lock_descr *cld = &scan->cll_descr;
1078                 const struct osc_lock *scan_ols;
1079
1080                 if (scan == lock)
1081                         break;
1082
1083                 if (scan->cll_state < CLS_QUEUING ||
1084                     scan->cll_state == CLS_FREEING ||
1085                     cld->cld_start > descr->cld_end ||
1086                     cld->cld_end < descr->cld_start)
1087                         continue;
1088
1089                 /* overlapped and living locks. */
1090
1091                 /* We're not supposed to give up group lock. */
1092                 if (scan->cll_descr.cld_mode == CLM_GROUP) {
1093                         LASSERT(descr->cld_mode != CLM_GROUP ||
1094                                 descr->cld_gid != scan->cll_descr.cld_gid);
1095                         continue;
1096                 }
1097
1098                 scan_ols = osc_lock_at(scan);
1099
1100                 /* We need to cancel the compatible locks if we're enqueuing
1101                  * a lockless lock, for example:
1102                  * imagine that client has PR lock on [0, 1000], and thread T0
1103                  * is doing lockless IO in [500, 1500] region. Concurrent
1104                  * thread T1 can see lockless data in [500, 1000], which is
1105                  * wrong, because these data are possibly stale. */
1106                 if (!lockless && osc_lock_compatible(olck, scan_ols))
1107                         continue;
1108
1109                 cl_lock_get_trust(scan);
1110                 conflict = scan;
1111                 break;
1112         }
1113         spin_unlock(&hdr->coh_lock_guard);
1114
1115         if (conflict) {
1116                 if (lock->cll_descr.cld_mode == CLM_GROUP) {
1117                         /* we want a group lock but a previous lock request
1118                          * conflicts, we do not wait but return 0 so the
1119                          * request is send to the server
1120                          */
1121                         CDEBUG(D_DLMTRACE, "group lock %p is conflicted "
1122                                            "with %p, no wait, send to server\n",
1123                                lock, conflict);
1124                         cl_lock_put(env, conflict);
1125                         rc = 0;
1126                 } else {
1127                         CDEBUG(D_DLMTRACE, "lock %p is conflicted with %p, "
1128                                            "will wait\n",
1129                                lock, conflict);
1130                         LASSERT(lock->cll_conflict == NULL);
1131                         lu_ref_add(&conflict->cll_reference, "cancel-wait",
1132                                    lock);
1133                         lock->cll_conflict = conflict;
1134                         rc = CLO_WAIT;
1135                 }
1136         }
1137         RETURN(rc);
1138 }
1139
1140 /**
1141  * Implementation of cl_lock_operations::clo_enqueue() method for osc
1142  * layer. This initiates ldlm enqueue:
1143  *
1144  *     - cancels conflicting locks early (osc_lock_enqueue_wait());
1145  *
1146  *     - calls osc_enqueue_base() to do actual enqueue.
1147  *
1148  * osc_enqueue_base() is supplied with an upcall function that is executed
1149  * when lock is received either after a local cached ldlm lock is matched, or
1150  * when a reply from the server is received.
1151  *
1152  * This function does not wait for the network communication to complete.
1153  */
1154 static int osc_lock_enqueue(const struct lu_env *env,
1155                             const struct cl_lock_slice *slice,
1156                             struct cl_io *unused, __u32 enqflags)
1157 {
1158         struct osc_lock          *ols     = cl2osc_lock(slice);
1159         struct cl_lock           *lock    = ols->ols_cl.cls_lock;
1160         int result;
1161         ENTRY;
1162
1163         LASSERT(cl_lock_is_mutexed(lock));
1164         LASSERTF(ols->ols_state == OLS_NEW,
1165                  "Impossible state: %d\n", ols->ols_state);
1166
1167         LASSERTF(ergo(ols->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
1168                 "lock = %p, ols = %p\n", lock, ols);
1169
1170         result = osc_lock_enqueue_wait(env, ols);
1171         if (result == 0) {
1172                 if (!osc_lock_is_lockless(ols)) {
1173                         struct osc_object        *obj = cl2osc(slice->cls_obj);
1174                         struct osc_thread_info   *info = osc_env_info(env);
1175                         struct ldlm_res_id       *resname = &info->oti_resname;
1176                         ldlm_policy_data_t       *policy = &info->oti_policy;
1177                         struct ldlm_enqueue_info *einfo = &ols->ols_einfo;
1178
1179                         /* lock will be passed as upcall cookie,
1180                          * hold ref to prevent to be released. */
1181                         cl_lock_hold_add(env, lock, "upcall", lock);
1182                         /* a user for lock also */
1183                         cl_lock_user_add(env, lock);
1184                         ols->ols_state = OLS_ENQUEUED;
1185
1186                         /*
1187                          * XXX: this is possible blocking point as
1188                          * ldlm_lock_match(LDLM_FL_LVB_READY) waits for
1189                          * LDLM_CP_CALLBACK.
1190                          */
1191                         ostid_build_res_name(&obj->oo_oinfo->loi_oi, resname);
1192                         osc_lock_build_policy(env, lock, policy);
1193                         result = osc_enqueue_base(osc_export(obj), resname,
1194                                           &ols->ols_flags, policy,
1195                                           &ols->ols_lvb,
1196                                           obj->oo_oinfo->loi_kms_valid,
1197                                           osc_lock_upcall,
1198                                           ols, einfo, &ols->ols_handle,
1199                                           PTLRPCD_SET, 1, ols->ols_agl);
1200                         if (result != 0) {
1201                                 cl_lock_user_del(env, lock);
1202                                 cl_lock_unhold(env, lock, "upcall", lock);
1203                                 if (unlikely(result == -ECANCELED)) {
1204                                         ols->ols_state = OLS_NEW;
1205                                         result = 0;
1206                                 }
1207                         }
1208                 } else {
1209                         ols->ols_state = OLS_GRANTED;
1210                         ols->ols_owner = osc_env_io(env);
1211                 }
1212         }
1213         LASSERT(ergo(ols->ols_glimpse, !osc_lock_is_lockless(ols)));
1214         RETURN(result);
1215 }
1216
1217 static int osc_lock_wait(const struct lu_env *env,
1218                          const struct cl_lock_slice *slice)
1219 {
1220         struct osc_lock *olck = cl2osc_lock(slice);
1221         struct cl_lock  *lock = olck->ols_cl.cls_lock;
1222
1223         LINVRNT(osc_lock_invariant(olck));
1224
1225         if (olck->ols_glimpse && olck->ols_state >= OLS_UPCALL_RECEIVED) {
1226                 if (olck->ols_flags & LDLM_FL_LVB_READY) {
1227                         return 0;
1228                 } else if (olck->ols_agl) {
1229                         if (lock->cll_flags & CLF_FROM_UPCALL)
1230                                 /* It is from enqueue RPC reply upcall for
1231                                  * updating state. Do not re-enqueue. */
1232                                 return -ENAVAIL;
1233                         else
1234                                 olck->ols_state = OLS_NEW;
1235                 } else {
1236                         LASSERT(lock->cll_error);
1237                         return lock->cll_error;
1238                 }
1239         }
1240
1241         if (olck->ols_state == OLS_NEW) {
1242                 int rc;
1243
1244                 LASSERT(olck->ols_agl);
1245                 olck->ols_agl = 0;
1246                 rc = osc_lock_enqueue(env, slice, NULL, CEF_ASYNC | CEF_MUST);
1247                 if (rc != 0)
1248                         return rc;
1249                 else
1250                         return CLO_REENQUEUED;
1251         }
1252
1253         LASSERT(equi(olck->ols_state >= OLS_UPCALL_RECEIVED &&
1254                      lock->cll_error == 0, olck->ols_lock != NULL));
1255
1256         return lock->cll_error ?: olck->ols_state >= OLS_GRANTED ? 0 : CLO_WAIT;
1257 }
1258
1259 /**
1260  * An implementation of cl_lock_operations::clo_use() method that pins cached
1261  * lock.
1262  */
1263 static int osc_lock_use(const struct lu_env *env,
1264                         const struct cl_lock_slice *slice)
1265 {
1266         struct osc_lock *olck = cl2osc_lock(slice);
1267         int rc;
1268
1269         LASSERT(!olck->ols_hold);
1270
1271         /*
1272          * Atomically check for LDLM_FL_CBPENDING and addref a lock if this
1273          * flag is not set. This protects us from a concurrent blocking ast.
1274          */
1275         rc = ldlm_lock_addref_try(&olck->ols_handle, olck->ols_einfo.ei_mode);
1276         if (rc == 0) {
1277                 olck->ols_hold = 1;
1278                 olck->ols_state = OLS_GRANTED;
1279         } else {
1280                 struct cl_lock *lock;
1281
1282                 /*
1283                  * Lock is being cancelled somewhere within
1284                  * ldlm_handle_bl_callback(): LDLM_FL_CBPENDING is already
1285                  * set, but osc_ldlm_blocking_ast() hasn't yet acquired
1286                  * cl_lock mutex.
1287                  */
1288                 lock = slice->cls_lock;
1289                 LASSERT(lock->cll_state == CLS_INTRANSIT);
1290                 LASSERT(lock->cll_users > 0);
1291                 /* set a flag for osc_dlm_blocking_ast0() to signal the
1292                  * lock.*/
1293                 olck->ols_ast_wait = 1;
1294                 rc = CLO_WAIT;
1295         }
1296         return rc;
1297 }
1298
1299 static int osc_lock_flush(struct osc_lock *ols, int discard)
1300 {
1301         struct cl_lock       *lock  = ols->ols_cl.cls_lock;
1302         struct cl_env_nest    nest;
1303         struct lu_env        *env;
1304         int result = 0;
1305         ENTRY;
1306
1307         env = cl_env_nested_get(&nest);
1308         if (!IS_ERR(env)) {
1309                 struct osc_object    *obj   = cl2osc(ols->ols_cl.cls_obj);
1310                 struct cl_lock_descr *descr = &lock->cll_descr;
1311                 int rc = 0;
1312
1313                 if (descr->cld_mode >= CLM_WRITE) {
1314                         result = osc_cache_writeback_range(env, obj,
1315                                         descr->cld_start, descr->cld_end,
1316                                         1, discard);
1317                         LDLM_DEBUG(ols->ols_lock,
1318                                 "lock %p: %d pages were %s.\n", lock, result,
1319                                 discard ? "discarded" : "written");
1320                         if (result > 0)
1321                                 result = 0;
1322                 }
1323
1324                 rc = cl_lock_discard_pages(env, lock);
1325                 if (result == 0 && rc < 0)
1326                         result = rc;
1327
1328                 cl_env_nested_put(&nest, env);
1329         } else
1330                 result = PTR_ERR(env);
1331         if (result == 0) {
1332                 ols->ols_flush = 1;
1333                 LINVRNT(!osc_lock_has_pages(ols));
1334         }
1335         RETURN(result);
1336 }
1337
1338 /**
1339  * Implements cl_lock_operations::clo_cancel() method for osc layer. This is
1340  * called (as part of cl_lock_cancel()) when lock is canceled either voluntary
1341  * (LRU pressure, early cancellation, umount, etc.) or due to the conflict
1342  * with some other lock some where in the cluster. This function does the
1343  * following:
1344  *
1345  *     - invalidates all pages protected by this lock (after sending dirty
1346  *       ones to the server, as necessary);
1347  *
1348  *     - decref's underlying ldlm lock;
1349  *
1350  *     - cancels ldlm lock (ldlm_cli_cancel()).
1351  */
1352 static void osc_lock_cancel(const struct lu_env *env,
1353                             const struct cl_lock_slice *slice)
1354 {
1355         struct cl_lock   *lock    = slice->cls_lock;
1356         struct osc_lock  *olck    = cl2osc_lock(slice);
1357         struct ldlm_lock *dlmlock = olck->ols_lock;
1358         int               result  = 0;
1359         int               discard;
1360
1361         LASSERT(cl_lock_is_mutexed(lock));
1362         LINVRNT(osc_lock_invariant(olck));
1363
1364         if (dlmlock != NULL) {
1365                 int do_cancel;
1366
1367                 discard = !!(dlmlock->l_flags & LDLM_FL_DISCARD_DATA);
1368                 if (olck->ols_state >= OLS_GRANTED)
1369                         result = osc_lock_flush(olck, discard);
1370                 osc_lock_unhold(olck);
1371
1372                 lock_res_and_lock(dlmlock);
1373                 /* Now that we're the only user of dlm read/write reference,
1374                  * mostly the ->l_readers + ->l_writers should be zero.
1375                  * However, there is a corner case.
1376                  * See bug 18829 for details.*/
1377                 do_cancel = (dlmlock->l_readers == 0 &&
1378                              dlmlock->l_writers == 0);
1379                 dlmlock->l_flags |= LDLM_FL_CBPENDING;
1380                 unlock_res_and_lock(dlmlock);
1381                 if (do_cancel)
1382                         result = ldlm_cli_cancel(&olck->ols_handle, LCF_ASYNC);
1383                 if (result < 0)
1384                         CL_LOCK_DEBUG(D_ERROR, env, lock,
1385                                       "lock %p cancel failure with error(%d)\n",
1386                                       lock, result);
1387         }
1388         olck->ols_state = OLS_CANCELLED;
1389         olck->ols_flags &= ~LDLM_FL_LVB_READY;
1390         osc_lock_detach(env, olck);
1391 }
1392
1393 #ifdef INVARIANT_CHECK
1394 static int check_cb(const struct lu_env *env, struct cl_io *io,
1395                     struct cl_page *page, void *cbdata)
1396 {
1397         struct cl_lock *lock = cbdata;
1398
1399         if (lock->cll_descr.cld_mode == CLM_READ) {
1400                 struct cl_lock *tmp;
1401                 tmp = cl_lock_at_page(env, lock->cll_descr.cld_obj,
1402                                      page, lock, 1, 0);
1403                 if (tmp != NULL) {
1404                         cl_lock_put(env, tmp);
1405                         return CLP_GANG_OKAY;
1406                 }
1407         }
1408
1409         CL_LOCK_DEBUG(D_ERROR, env, lock, "still has pages\n");
1410         CL_PAGE_DEBUG(D_ERROR, env, page, "\n");
1411         return CLP_GANG_ABORT;
1412 }
1413
1414 /**
1415  * Returns true iff there are pages under \a olck not protected by other
1416  * locks.
1417  */
1418 static int osc_lock_has_pages(struct osc_lock *olck)
1419 {
1420         struct cl_lock       *lock;
1421         struct cl_lock_descr *descr;
1422         struct cl_object     *obj;
1423         struct osc_object    *oob;
1424         struct cl_env_nest    nest;
1425         struct cl_io         *io;
1426         struct lu_env        *env;
1427         int                   result;
1428
1429         env = cl_env_nested_get(&nest);
1430         if (IS_ERR(env))
1431                 return 0;
1432
1433         obj   = olck->ols_cl.cls_obj;
1434         oob   = cl2osc(obj);
1435         io    = &oob->oo_debug_io;
1436         lock  = olck->ols_cl.cls_lock;
1437         descr = &lock->cll_descr;
1438
1439         mutex_lock(&oob->oo_debug_mutex);
1440
1441         io->ci_obj = cl_object_top(obj);
1442         io->ci_ignore_layout = 1;
1443         cl_io_init(env, io, CIT_MISC, io->ci_obj);
1444         do {
1445                 result = cl_page_gang_lookup(env, obj, io,
1446                                              descr->cld_start, descr->cld_end,
1447                                              check_cb, (void *)lock);
1448                 if (result == CLP_GANG_ABORT)
1449                         break;
1450                 if (result == CLP_GANG_RESCHED)
1451                         cfs_cond_resched();
1452         } while (result != CLP_GANG_OKAY);
1453         cl_io_fini(env, io);
1454         mutex_unlock(&oob->oo_debug_mutex);
1455         cl_env_nested_put(&nest, env);
1456
1457         return (result == CLP_GANG_ABORT);
1458 }
1459 #else
1460 static int osc_lock_has_pages(struct osc_lock *olck)
1461 {
1462         return 0;
1463 }
1464 #endif /* INVARIANT_CHECK */
1465
1466 static void osc_lock_delete(const struct lu_env *env,
1467                             const struct cl_lock_slice *slice)
1468 {
1469         struct osc_lock *olck;
1470
1471         olck = cl2osc_lock(slice);
1472         if (olck->ols_glimpse) {
1473                 LASSERT(!olck->ols_hold);
1474                 LASSERT(!olck->ols_lock);
1475                 return;
1476         }
1477
1478         LINVRNT(osc_lock_invariant(olck));
1479         LINVRNT(!osc_lock_has_pages(olck));
1480
1481         osc_lock_unhold(olck);
1482         osc_lock_detach(env, olck);
1483 }
1484
1485 /**
1486  * Implements cl_lock_operations::clo_state() method for osc layer.
1487  *
1488  * Maintains osc_lock::ols_owner field.
1489  *
1490  * This assumes that lock always enters CLS_HELD (from some other state) in
1491  * the same IO context as one that requested the lock. This should not be a
1492  * problem, because context is by definition shared by all activity pertaining
1493  * to the same high-level IO.
1494  */
1495 static void osc_lock_state(const struct lu_env *env,
1496                            const struct cl_lock_slice *slice,
1497                            enum cl_lock_state state)
1498 {
1499         struct osc_lock *lock = cl2osc_lock(slice);
1500
1501         /*
1502          * XXX multiple io contexts can use the lock at the same time.
1503          */
1504         LINVRNT(osc_lock_invariant(lock));
1505         if (state == CLS_HELD && slice->cls_lock->cll_state != CLS_HELD) {
1506                 struct osc_io *oio = osc_env_io(env);
1507
1508                 LASSERT(lock->ols_owner == NULL);
1509                 lock->ols_owner = oio;
1510         } else if (state != CLS_HELD)
1511                 lock->ols_owner = NULL;
1512 }
1513
1514 static int osc_lock_print(const struct lu_env *env, void *cookie,
1515                           lu_printer_t p, const struct cl_lock_slice *slice)
1516 {
1517         struct osc_lock *lock = cl2osc_lock(slice);
1518
1519         /*
1520          * XXX print ldlm lock and einfo properly.
1521          */
1522         (*p)(env, cookie, "%p %#16llx "LPX64" %d %p ",
1523              lock->ols_lock, lock->ols_flags, lock->ols_handle.cookie,
1524              lock->ols_state, lock->ols_owner);
1525         osc_lvb_print(env, cookie, p, &lock->ols_lvb);
1526         return 0;
1527 }
1528
1529 static int osc_lock_fits_into(const struct lu_env *env,
1530                               const struct cl_lock_slice *slice,
1531                               const struct cl_lock_descr *need,
1532                               const struct cl_io *io)
1533 {
1534         struct osc_lock *ols = cl2osc_lock(slice);
1535
1536         if (need->cld_enq_flags & CEF_NEVER)
1537                 return 0;
1538
1539         if (ols->ols_state >= OLS_CANCELLED)
1540                 return 0;
1541
1542         if (need->cld_mode == CLM_PHANTOM) {
1543                 if (ols->ols_agl)
1544                         return !(ols->ols_state > OLS_RELEASED);
1545
1546                 /*
1547                  * Note: the QUEUED lock can't be matched here, otherwise
1548                  * it might cause the deadlocks.
1549                  * In read_process,
1550                  * P1: enqueued read lock, create sublock1
1551                  * P2: enqueued write lock, create sublock2(conflicted
1552                  *     with sublock1).
1553                  * P1: Grant read lock.
1554                  * P1: enqueued glimpse lock(with holding sublock1_read),
1555                  *     matched with sublock2, waiting sublock2 to be granted.
1556                  *     But sublock2 can not be granted, because P1
1557                  *     will not release sublock1. Bang!
1558                  */
1559                 if (ols->ols_state < OLS_GRANTED ||
1560                     ols->ols_state > OLS_RELEASED)
1561                         return 0;
1562         } else if (need->cld_enq_flags & CEF_MUST) {
1563                 /*
1564                  * If the lock hasn't ever enqueued, it can't be matched
1565                  * because enqueue process brings in many information
1566                  * which can be used to determine things such as lockless,
1567                  * CEF_MUST, etc.
1568                  */
1569                 if (ols->ols_state < OLS_UPCALL_RECEIVED &&
1570                     ols->ols_locklessable)
1571                         return 0;
1572         }
1573         return 1;
1574 }
1575
1576 static const struct cl_lock_operations osc_lock_ops = {
1577         .clo_fini    = osc_lock_fini,
1578         .clo_enqueue = osc_lock_enqueue,
1579         .clo_wait    = osc_lock_wait,
1580         .clo_unuse   = osc_lock_unuse,
1581         .clo_use     = osc_lock_use,
1582         .clo_delete  = osc_lock_delete,
1583         .clo_state   = osc_lock_state,
1584         .clo_cancel  = osc_lock_cancel,
1585         .clo_weigh   = osc_lock_weigh,
1586         .clo_print   = osc_lock_print,
1587         .clo_fits_into = osc_lock_fits_into,
1588 };
1589
1590 static int osc_lock_lockless_unuse(const struct lu_env *env,
1591                                    const struct cl_lock_slice *slice)
1592 {
1593         struct osc_lock *ols = cl2osc_lock(slice);
1594         struct cl_lock *lock = slice->cls_lock;
1595
1596         LASSERT(ols->ols_state == OLS_GRANTED);
1597         LINVRNT(osc_lock_invariant(ols));
1598
1599         cl_lock_cancel(env, lock);
1600         cl_lock_delete(env, lock);
1601         return 0;
1602 }
1603
1604 static void osc_lock_lockless_cancel(const struct lu_env *env,
1605                                      const struct cl_lock_slice *slice)
1606 {
1607         struct osc_lock   *ols  = cl2osc_lock(slice);
1608         int result;
1609
1610         result = osc_lock_flush(ols, 0);
1611         if (result)
1612                 CERROR("Pages for lockless lock %p were not purged(%d)\n",
1613                        ols, result);
1614         ols->ols_state = OLS_CANCELLED;
1615 }
1616
1617 static int osc_lock_lockless_wait(const struct lu_env *env,
1618                                   const struct cl_lock_slice *slice)
1619 {
1620         struct osc_lock *olck = cl2osc_lock(slice);
1621         struct cl_lock  *lock = olck->ols_cl.cls_lock;
1622
1623         LINVRNT(osc_lock_invariant(olck));
1624         LASSERT(olck->ols_state >= OLS_UPCALL_RECEIVED);
1625
1626         return lock->cll_error;
1627 }
1628
1629 static void osc_lock_lockless_state(const struct lu_env *env,
1630                                     const struct cl_lock_slice *slice,
1631                                     enum cl_lock_state state)
1632 {
1633         struct osc_lock *lock = cl2osc_lock(slice);
1634
1635         LINVRNT(osc_lock_invariant(lock));
1636         if (state == CLS_HELD) {
1637                 struct osc_io *oio  = osc_env_io(env);
1638
1639                 LASSERT(ergo(lock->ols_owner, lock->ols_owner == oio));
1640                 lock->ols_owner = oio;
1641
1642                 /* set the io to be lockless if this lock is for io's
1643                  * host object */
1644                 if (cl_object_same(oio->oi_cl.cis_obj, slice->cls_obj))
1645                         oio->oi_lockless = 1;
1646         }
1647 }
1648
1649 static int osc_lock_lockless_fits_into(const struct lu_env *env,
1650                                        const struct cl_lock_slice *slice,
1651                                        const struct cl_lock_descr *need,
1652                                        const struct cl_io *io)
1653 {
1654         struct osc_lock *lock = cl2osc_lock(slice);
1655
1656         if (!(need->cld_enq_flags & CEF_NEVER))
1657                 return 0;
1658
1659         /* lockless lock should only be used by its owning io. b22147 */
1660         return (lock->ols_owner == osc_env_io(env));
1661 }
1662
1663 static const struct cl_lock_operations osc_lock_lockless_ops = {
1664         .clo_fini      = osc_lock_fini,
1665         .clo_enqueue   = osc_lock_enqueue,
1666         .clo_wait      = osc_lock_lockless_wait,
1667         .clo_unuse     = osc_lock_lockless_unuse,
1668         .clo_state     = osc_lock_lockless_state,
1669         .clo_fits_into = osc_lock_lockless_fits_into,
1670         .clo_cancel    = osc_lock_lockless_cancel,
1671         .clo_print     = osc_lock_print
1672 };
1673
1674 int osc_lock_init(const struct lu_env *env,
1675                   struct cl_object *obj, struct cl_lock *lock,
1676                   const struct cl_io *unused)
1677 {
1678         struct osc_lock *clk;
1679         int result;
1680
1681         OBD_SLAB_ALLOC_PTR_GFP(clk, osc_lock_kmem, CFS_ALLOC_IO);
1682         if (clk != NULL) {
1683                 __u32 enqflags = lock->cll_descr.cld_enq_flags;
1684
1685                 osc_lock_build_einfo(env, lock, clk, &clk->ols_einfo);
1686                 cfs_atomic_set(&clk->ols_pageref, 0);
1687                 clk->ols_state = OLS_NEW;
1688
1689                 clk->ols_flags = osc_enq2ldlm_flags(enqflags);
1690                 clk->ols_agl = !!(enqflags & CEF_AGL);
1691                 if (clk->ols_agl)
1692                         clk->ols_flags |= LDLM_FL_BLOCK_NOWAIT;
1693                 if (clk->ols_flags & LDLM_FL_HAS_INTENT)
1694                         clk->ols_glimpse = 1;
1695
1696                 cl_lock_slice_add(lock, &clk->ols_cl, obj, &osc_lock_ops);
1697
1698                 if (!(enqflags & CEF_MUST))
1699                         /* try to convert this lock to a lockless lock */
1700                         osc_lock_to_lockless(env, clk, (enqflags & CEF_NEVER));
1701                 if (clk->ols_locklessable && !(enqflags & CEF_DISCARD_DATA))
1702                         clk->ols_flags |= LDLM_FL_DENY_ON_CONTENTION;
1703
1704                 LDLM_DEBUG_NOLOCK("lock %p, osc lock %p, flags %llx\n",
1705                                 lock, clk, clk->ols_flags);
1706
1707                 result = 0;
1708         } else
1709                 result = -ENOMEM;
1710         return result;
1711 }
1712
1713 int osc_dlm_lock_pageref(struct ldlm_lock *dlm)
1714 {
1715         struct osc_lock *olock;
1716         int              rc = 0;
1717
1718         spin_lock(&osc_ast_guard);
1719         olock = dlm->l_ast_data;
1720         /*
1721          * there's a very rare race with osc_page_addref_lock(), but that
1722          * doesn't matter because in the worst case we don't cancel a lock
1723          * which we actually can, that's no harm.
1724          */
1725         if (olock != NULL &&
1726             cfs_atomic_add_return(_PAGEREF_MAGIC,
1727                                   &olock->ols_pageref) != _PAGEREF_MAGIC) {
1728                 cfs_atomic_sub(_PAGEREF_MAGIC, &olock->ols_pageref);
1729                 rc = 1;
1730         }
1731         spin_unlock(&osc_ast_guard);
1732         return rc;
1733 }
1734
1735 /** @} osc */