4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * Implementation of cl_lock for OSC layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
41 #define DEBUG_SUBSYSTEM S_OSC
44 # include <libcfs/libcfs.h>
46 # include <liblustre.h>
48 /* fid_build_reg_res_name() */
49 #include <lustre_fid.h>
51 #include "osc_cl_internal.h"
57 #define _PAGEREF_MAGIC (-10000000)
59 /*****************************************************************************
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);
71 int osc_lock_is_lockless(const struct osc_lock *olck)
73 return (olck->ols_cl.cls_ops == &osc_lock_lockless_ops);
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().
81 static struct ldlm_lock *osc_handle_ptr(struct lustre_handle *handle)
83 struct ldlm_lock *lock;
85 lock = ldlm_handle2lock(handle);
92 * Invariant that has to be true all of the time.
94 static int osc_lock_invariant(struct osc_lock *ols)
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);
101 ergo(osc_lock_is_lockless(ols),
102 ols->ols_locklessable && ols->ols_lock == NULL) ||
103 (ergo(olock != NULL, handle_used) &&
105 olock->l_handle.h_cookie == ols->ols_handle.cookie) &&
107 * Check that ->ols_handle and ->ols_lock are consistent, but
108 * take into account that they are set at the different time.
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) &&
116 * DLM lock is destroyed only after we have seen cancellation
119 ergo(olock != NULL && ols->ols_state < OLS_CANCELLED,
120 !olock->l_destroyed) &&
121 ergo(ols->ols_state == OLS_GRANTED,
123 olock->l_req_mode == olock->l_granted_mode &&
127 /*****************************************************************************
134 * Breaks a link between osc_lock and dlm_lock.
136 static void osc_lock_detach(const struct lu_env *env, struct osc_lock *olck)
138 struct ldlm_lock *dlmlock;
140 spin_lock(&osc_ast_guard);
141 dlmlock = olck->ols_lock;
142 if (dlmlock == NULL) {
143 spin_unlock(&osc_ast_guard);
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);
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;
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);
167 cl_object_attr_set(env, obj, attr, CAT_KMS);
168 cl_object_attr_unlock(obj);
170 unlock_res_and_lock(dlmlock);
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;
179 static int osc_lock_unhold(struct osc_lock *ols)
185 result = osc_cancel_base(&ols->ols_handle,
186 ols->ols_einfo.ei_mode);
191 static int osc_lock_unuse(const struct lu_env *env,
192 const struct cl_lock_slice *slice)
194 struct osc_lock *ols = cl2osc_lock(slice);
196 LINVRNT(osc_lock_invariant(ols));
198 switch (ols->ols_state) {
200 LASSERT(!ols->ols_hold);
201 LASSERT(ols->ols_agl);
203 case OLS_UPCALL_RECEIVED:
204 osc_lock_unhold(ols);
206 LASSERT(!ols->ols_hold);
207 osc_lock_detach(env, ols);
208 ols->ols_state = OLS_NEW;
211 LASSERT(!ols->ols_glimpse);
212 LASSERT(ols->ols_hold);
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
219 ols->ols_state = OLS_RELEASED;
220 return osc_lock_unhold(ols);
222 CERROR("Impossible state: %d\n", ols->ols_state);
227 static void osc_lock_fini(const struct lu_env *env,
228 struct cl_lock_slice *slice)
230 struct osc_lock *ols = cl2osc_lock(slice);
232 LINVRNT(osc_lock_invariant(ols));
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.
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);
244 OBD_SLAB_FREE_PTR(ols, osc_lock_kmem);
247 static void osc_lock_build_policy(const struct lu_env *env,
248 const struct cl_lock *lock,
249 ldlm_policy_data_t *policy)
251 const struct cl_lock_descr *d = &lock->cll_descr;
253 osc_index2policy(policy, d->cld_obj, d->cld_start, d->cld_end);
254 policy->l_extent.gid = d->cld_gid;
257 static __u64 osc_enq2ldlm_flags(__u32 enqflags)
261 LASSERT((enqflags & ~CEF_MASK) == 0);
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;
273 * Global spin-lock protecting consistency of ldlm_lock::l_ast_data
274 * pointers. Initialized in osc_init().
276 spinlock_t osc_ast_guard;
278 static struct osc_lock *osc_ast_data_get(struct ldlm_lock *dlm_lock)
280 struct osc_lock *olck;
282 lock_res_and_lock(dlm_lock);
283 spin_lock(&osc_ast_guard);
284 olck = dlm_lock->l_ast_data;
286 struct cl_lock *lock = olck->ols_cl.cls_lock;
288 * If osc_lock holds a reference on ldlm lock, return it even
289 * when cl_lock is in CLS_FREEING state. This way
291 * osc_ast_data_get(dlmlock) == NULL
293 * guarantees that all osc references on dlmlock were
294 * released. osc_dlm_blocking_ast0() relies on that.
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());
303 spin_unlock(&osc_ast_guard);
304 unlock_res_and_lock(dlm_lock);
308 static void osc_ast_data_put(const struct lu_env *env, struct osc_lock *olck)
310 struct cl_lock *lock;
312 lock = olck->ols_cl.cls_lock;
313 lu_ref_del(&lock->cll_reference, "ast", cfs_current());
314 cl_lock_put(env, lock);
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()
322 * This can be optimized to not update attributes when lock is a result of a
325 * Called under lock and resource spin-locks.
327 static void osc_lock_lvb_update(const struct lu_env *env, struct osc_lock *olck,
331 struct cl_object *obj;
332 struct lov_oinfo *oinfo;
333 struct cl_attr *attr;
338 if (!(olck->ols_flags & LDLM_FL_LVB_READY))
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);
348 cl_object_attr_lock(obj);
350 struct ldlm_lock *dlmlock;
353 dlmlock = olck->ols_lock;
354 LASSERT(dlmlock != NULL);
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);
367 attr->cat_kms = size;
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);
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);
382 cl_object_attr_set(env, obj, attr, valid);
384 cl_object_attr_unlock(obj);
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.
393 * Called under lock and resource spin-locks, that are released temporarily
396 static void osc_lock_granted(const struct lu_env *env, struct osc_lock *olck,
397 struct ldlm_lock *dlmlock, int rc)
399 struct ldlm_extent *ext;
400 struct cl_lock *lock;
401 struct cl_lock_descr *descr;
403 LASSERT(dlmlock->l_granted_mode == dlmlock->l_req_mode);
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;
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;
418 * tell upper layers the extent of the lock that was actually
421 olck->ols_state = OLS_GRANTED;
422 osc_lock_lvb_update(env, olck, rc);
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
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);
438 static void osc_lock_upcall0(const struct lu_env *env, struct osc_lock *olck)
441 struct ldlm_lock *dlmlock;
445 dlmlock = ldlm_handle2lock_long(&olck->ols_handle, 0);
446 LASSERT(dlmlock != NULL);
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);
456 * Lock might be not yet granted. In this case, completion ast
457 * (osc_ldlm_completion_ast()) comes later and finishes lock
460 if (dlmlock->l_granted_mode == dlmlock->l_req_mode)
461 osc_lock_granted(env, olck, dlmlock, 0);
462 unlock_res_and_lock(dlmlock);
465 * osc_enqueue_interpret() decrefs asynchronous locks, counter
468 ldlm_lock_addref(&olck->ols_handle, olck->ols_einfo.ei_mode);
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;
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
482 static int osc_lock_upcall(void *cookie, int errcode)
484 struct osc_lock *olck = cookie;
485 struct cl_lock_slice *slice = &olck->ols_cl;
486 struct cl_lock *lock = slice->cls_lock;
488 struct cl_env_nest nest;
491 env = cl_env_nested_get(&nest);
495 cl_lock_mutex_get(env, lock);
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) {
504 CERROR("Impossible state: %d\n", olck->ols_state);
508 struct ldlm_lock *dlmlock;
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);
523 if (olck->ols_glimpse)
524 olck->ols_glimpse = 0;
525 osc_lock_upcall0(env, olck);
528 /* Error handling, some errors are tolerable. */
529 if (olck->ols_locklessable && rc == -EUSERS) {
530 /* This is a tolerable error, turn this lock into
533 osc_object_set_contended(cl2osc(slice->cls_obj));
534 LASSERT(slice->cls_ops == &osc_lock_ops);
536 /* Change this lock to ldlmlock-less lock. */
537 osc_lock_to_lockless(env, olck, 1);
538 olck->ols_state = OLS_GRANTED;
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. */
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(). */
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)
559 cl_lock_signal(env, lock);
560 /* del user for lock upcall cookie */
561 cl_unuse_try(env, lock);
563 /* del user for lock upcall cookie */
564 cl_lock_user_del(env, lock);
565 cl_lock_error(env, lock, rc);
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);
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);
577 cl_env_nested_put(&nest, env);
579 /* should never happen, similar to osc_ldlm_blocking_ast(). */
586 * Core of osc_dlm_blocking_ast() logic.
588 static void osc_lock_blocking(const struct lu_env *env,
589 struct ldlm_lock *dlmlock,
590 struct osc_lock *olck, int blocking)
592 struct cl_lock *lock = olck->ols_cl.cls_lock;
594 LASSERT(olck->ols_lock == dlmlock);
595 CLASSERT(OLS_BLOCKED < OLS_CANCELLED);
596 LASSERT(!osc_lock_is_lockless(olck));
599 * Lock might be still addref-ed here, if e.g., blocking ast
600 * is sent for a failed lock.
602 osc_lock_unhold(olck);
604 if (blocking && olck->ols_state < OLS_BLOCKED)
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.
610 olck->ols_state = OLS_BLOCKED;
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.
616 cl_lock_cancel(env, lock);
617 cl_lock_delete(env, lock);
621 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
622 * and ldlm_lock caches.
624 static int osc_dlm_blocking_ast0(const struct lu_env *env,
625 struct ldlm_lock *dlmlock,
626 void *data, int flag)
628 struct osc_lock *olck;
629 struct cl_lock *lock;
633 LASSERT(flag == LDLM_CB_BLOCKING || flag == LDLM_CB_CANCELING);
636 olck = osc_ast_data_get(dlmlock);
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;
647 * Lock might have been canceled while this thread was
648 * sleeping for lock mutex, but olck is pinned in memory.
650 if (olck == dlmlock->l_ast_data) {
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
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.
664 LASSERT(data == olck);
665 osc_lock_blocking(env, dlmlock,
666 olck, flag == LDLM_CB_BLOCKING);
669 cl_lock_mutex_put(env, lock);
670 osc_ast_data_put(env, olck);
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.
677 cancel = (flag == LDLM_CB_BLOCKING);
680 struct lustre_handle *lockh;
682 lockh = &osc_env_info(env)->oti_handle;
683 ldlm_lock2handle(dlmlock, lockh);
684 result = ldlm_cli_cancel(lockh, LCF_ASYNC);
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.
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.
698 * \param dlmlock lock for which ast occurred.
700 * \param new description of a conflicting lock in case of blocking ast.
702 * \param data value of dlmlock->l_ast_data
704 * \param flag LDLM_CB_BLOCKING or LDLM_CB_CANCELING. Used to distinguish
705 * cancellation and blocking ast's.
707 * Possible use cases:
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
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.
717 * - cl_lock_cancel() is called. osc_lock_cancel() calls
718 * ldlm_cli_cancel() that calls
720 * dlmlock->l_blocking_ast(..., LDLM_CB_CANCELING)
722 * recursively entering osc_ldlm_blocking_ast().
724 * - client cancels lock voluntary (e.g., as a part of early cancellation):
727 * osc_lock_cancel()->
728 * ldlm_cli_cancel()->
729 * dlmlock->l_blocking_ast(..., LDLM_CB_CANCELING)
732 static int osc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
733 struct ldlm_lock_desc *new, void *data,
737 struct cl_env_nest nest;
741 * This can be called in the context of outer IO, e.g.,
744 * ->osc_enqueue_base()->...
745 * ->ldlm_prep_elc_req()->...
746 * ->ldlm_cancel_callback()->...
747 * ->osc_ldlm_blocking_ast()
749 * new environment has to be created to not corrupt outer context.
751 env = cl_env_nested_get(&nest);
753 result = osc_dlm_blocking_ast0(env, dlmlock, data, flag);
754 cl_env_nested_put(&nest, env);
756 result = PTR_ERR(env);
758 * XXX This should never happen, as cl_lock is
759 * stuck. Pre-allocated environment a la vvp_inode_fini_env
765 if (result == -ENODATA)
768 CERROR("BAST failed: %d\n", result);
773 static int osc_ldlm_completion_ast(struct ldlm_lock *dlmlock,
774 __u64 flags, void *data)
776 struct cl_env_nest nest;
778 struct osc_lock *olck;
779 struct cl_lock *lock;
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);
788 olck = osc_ast_data_get(dlmlock);
790 lock = olck->ols_cl.cls_lock;
791 cl_lock_mutex_get(env, lock);
793 * ldlm_handle_cp_callback() copied LVB from request
794 * to lock->l_lvb_data, store it in osc_lock.
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) {
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.
805 * This maintains an invariant that osc_lock
806 * and ldlm_lock are always bound when
807 * osc_lock is in OLS_GRANTED state.
809 } else if (dlmlock->l_granted_mode ==
810 dlmlock->l_req_mode) {
811 osc_lock_granted(env, olck, dlmlock, dlmrc);
813 unlock_res_and_lock(dlmlock);
816 CL_LOCK_DEBUG(D_ERROR, env, lock,
817 "dlmlock returned %d\n", dlmrc);
818 cl_lock_error(env, lock, dlmrc);
820 cl_lock_mutex_put(env, lock);
821 osc_ast_data_put(env, olck);
824 result = -ELDLM_NO_LOCK_DATA;
825 cl_env_nested_put(&nest, env);
827 result = PTR_ERR(env);
828 return dlmrc ?: result;
831 static int osc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data)
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;
840 struct req_capsule *cap;
843 LASSERT(lustre_msg_get_opc(req->rq_reqmsg) == LDLM_GL_CALLBACK);
845 env = cl_env_nested_get(&nest);
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
852 olck = osc_ast_data_get(dlmlock);
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); */
862 req_capsule_extend(cap, &RQF_LDLM_GL_CALLBACK);
863 req_capsule_set_size(cap, &RMF_DLM_LVB, RCL_SERVER,
865 result = req_capsule_server_pack(cap);
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);
871 if (!exp_connect_lvb_type(req->rq_export))
872 req_capsule_shrink(&req->rq_pill,
874 sizeof(struct ost_lvb_v1),
876 osc_ast_data_put(env, olck);
879 * These errors are normal races, so we don't want to
880 * fill the console with messages by calling
883 lustre_pack_reply(req, 1, NULL, NULL);
884 result = -ELDLM_NO_LOCK_DATA;
886 cl_env_nested_put(&nest, env);
888 result = PTR_ERR(env);
889 req->rq_status = result;
893 static unsigned long osc_lock_weigh(const struct lu_env *env,
894 const struct cl_lock_slice *slice)
897 * don't need to grab coh_page_guard since we don't care the exact #
900 return cl_object_header(slice->cls_obj)->coh_pages;
904 * Get the weight of dlm lock for early cancellation.
906 * XXX: it should return the pages covered by this \a dlmlock.
908 static unsigned long osc_ldlm_weigh_ast(struct ldlm_lock *dlmlock)
910 struct cl_env_nest nest;
912 struct osc_lock *lock;
914 unsigned long weight;
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 ..
925 env = cl_env_nested_get(&nest);
927 /* Mostly because lack of memory, tend to eliminate this lock*/
930 LASSERT(dlmlock->l_resource->lr_type == LDLM_EXTENT);
931 lock = osc_ast_data_get(dlmlock);
933 /* cl_lock was destroyed because of memory pressure.
934 * It is much reasonable to assign this type of lock
937 GOTO(out, weight = 0);
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);
948 cl_env_nested_put(&nest, env);
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)
957 enum cl_lock_mode mode;
959 mode = clock->cll_descr.cld_mode;
960 if (mode == CLM_PHANTOM)
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.
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 */
978 * Determine if the lock should be converted into a lockless lock.
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
986 * Additional policy can be implemented here, e.g., never do lockless-io
989 static void osc_lock_to_lockless(const struct lu_env *env,
990 struct osc_lock *ols, int force)
992 struct cl_lock_slice *slice = &ols->ols_cl;
994 LASSERT(ols->ols_state == OLS_NEW ||
995 ols->ols_state == OLS_UPCALL_RECEIVED);
998 ols->ols_locklessable = 1;
999 slice->cls_ops = &osc_lock_lockless_ops;
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;
1008 LASSERT(io->ci_lockreq == CILR_MANDATORY ||
1009 io->ci_lockreq == CILR_MAYBE ||
1010 io->ci_lockreq == CILR_NEVER);
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 ||
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;
1027 LASSERT(ergo(ols->ols_glimpse, !osc_lock_is_lockless(ols)));
1030 static int osc_lock_compatible(const struct osc_lock *qing,
1031 const struct osc_lock *qed)
1033 enum cl_lock_mode qing_mode;
1034 enum cl_lock_mode qed_mode;
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))
1041 qed_mode = qed->ols_cl.cls_lock->cll_descr.cld_mode;
1042 return ((qing_mode == CLM_READ) && (qed_mode == CLM_READ));
1046 * Cancel all conflicting locks and wait for them to be destroyed.
1048 * This function is used for two purposes:
1050 * - early cancel all conflicting locks before starting IO, and
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.
1056 static int osc_lock_enqueue_wait(const struct lu_env *env,
1057 const struct osc_lock *olck)
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);
1068 LASSERT(cl_lock_is_mutexed(lock));
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)
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;
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)
1089 /* overlapped and living locks. */
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);
1098 scan_ols = osc_lock_at(scan);
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))
1109 cl_lock_get_trust(scan);
1113 spin_unlock(&hdr->coh_lock_guard);
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
1121 CDEBUG(D_DLMTRACE, "group lock %p is conflicted "
1122 "with %p, no wait, send to server\n",
1124 cl_lock_put(env, conflict);
1127 CDEBUG(D_DLMTRACE, "lock %p is conflicted with %p, "
1130 LASSERT(lock->cll_conflict == NULL);
1131 lu_ref_add(&conflict->cll_reference, "cancel-wait",
1133 lock->cll_conflict = conflict;
1141 * Implementation of cl_lock_operations::clo_enqueue() method for osc
1142 * layer. This initiates ldlm enqueue:
1144 * - cancels conflicting locks early (osc_lock_enqueue_wait());
1146 * - calls osc_enqueue_base() to do actual enqueue.
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.
1152 * This function does not wait for the network communication to complete.
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)
1158 struct osc_lock *ols = cl2osc_lock(slice);
1159 struct cl_lock *lock = ols->ols_cl.cls_lock;
1163 LASSERT(cl_lock_is_mutexed(lock));
1164 LASSERTF(ols->ols_state == OLS_NEW,
1165 "Impossible state: %d\n", ols->ols_state);
1167 LASSERTF(ergo(ols->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
1168 "lock = %p, ols = %p\n", lock, ols);
1170 result = osc_lock_enqueue_wait(env, ols);
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;
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;
1187 * XXX: this is possible blocking point as
1188 * ldlm_lock_match(LDLM_FL_LVB_READY) waits for
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,
1196 obj->oo_oinfo->loi_kms_valid,
1198 ols, einfo, &ols->ols_handle,
1199 PTLRPCD_SET, 1, ols->ols_agl);
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;
1209 ols->ols_state = OLS_GRANTED;
1210 ols->ols_owner = osc_env_io(env);
1213 LASSERT(ergo(ols->ols_glimpse, !osc_lock_is_lockless(ols)));
1217 static int osc_lock_wait(const struct lu_env *env,
1218 const struct cl_lock_slice *slice)
1220 struct osc_lock *olck = cl2osc_lock(slice);
1221 struct cl_lock *lock = olck->ols_cl.cls_lock;
1223 LINVRNT(osc_lock_invariant(olck));
1225 if (olck->ols_glimpse && olck->ols_state >= OLS_UPCALL_RECEIVED) {
1226 if (olck->ols_flags & LDLM_FL_LVB_READY) {
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. */
1234 olck->ols_state = OLS_NEW;
1236 LASSERT(lock->cll_error);
1237 return lock->cll_error;
1241 if (olck->ols_state == OLS_NEW) {
1244 LASSERT(olck->ols_agl);
1246 rc = osc_lock_enqueue(env, slice, NULL, CEF_ASYNC | CEF_MUST);
1250 return CLO_REENQUEUED;
1253 LASSERT(equi(olck->ols_state >= OLS_UPCALL_RECEIVED &&
1254 lock->cll_error == 0, olck->ols_lock != NULL));
1256 return lock->cll_error ?: olck->ols_state >= OLS_GRANTED ? 0 : CLO_WAIT;
1260 * An implementation of cl_lock_operations::clo_use() method that pins cached
1263 static int osc_lock_use(const struct lu_env *env,
1264 const struct cl_lock_slice *slice)
1266 struct osc_lock *olck = cl2osc_lock(slice);
1269 LASSERT(!olck->ols_hold);
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.
1275 rc = ldlm_lock_addref_try(&olck->ols_handle, olck->ols_einfo.ei_mode);
1278 olck->ols_state = OLS_GRANTED;
1280 struct cl_lock *lock;
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
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
1293 olck->ols_ast_wait = 1;
1299 static int osc_lock_flush(struct osc_lock *ols, int discard)
1301 struct cl_lock *lock = ols->ols_cl.cls_lock;
1302 struct cl_env_nest nest;
1307 env = cl_env_nested_get(&nest);
1309 struct osc_object *obj = cl2osc(ols->ols_cl.cls_obj);
1310 struct cl_lock_descr *descr = &lock->cll_descr;
1313 if (descr->cld_mode >= CLM_WRITE) {
1314 result = osc_cache_writeback_range(env, obj,
1315 descr->cld_start, descr->cld_end,
1317 LDLM_DEBUG(ols->ols_lock,
1318 "lock %p: %d pages were %s.\n", lock, result,
1319 discard ? "discarded" : "written");
1324 rc = cl_lock_discard_pages(env, lock);
1325 if (result == 0 && rc < 0)
1328 cl_env_nested_put(&nest, env);
1330 result = PTR_ERR(env);
1333 LINVRNT(!osc_lock_has_pages(ols));
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
1345 * - invalidates all pages protected by this lock (after sending dirty
1346 * ones to the server, as necessary);
1348 * - decref's underlying ldlm lock;
1350 * - cancels ldlm lock (ldlm_cli_cancel()).
1352 static void osc_lock_cancel(const struct lu_env *env,
1353 const struct cl_lock_slice *slice)
1355 struct cl_lock *lock = slice->cls_lock;
1356 struct osc_lock *olck = cl2osc_lock(slice);
1357 struct ldlm_lock *dlmlock = olck->ols_lock;
1361 LASSERT(cl_lock_is_mutexed(lock));
1362 LINVRNT(osc_lock_invariant(olck));
1364 if (dlmlock != NULL) {
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);
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);
1382 result = ldlm_cli_cancel(&olck->ols_handle, LCF_ASYNC);
1384 CL_LOCK_DEBUG(D_ERROR, env, lock,
1385 "lock %p cancel failure with error(%d)\n",
1388 olck->ols_state = OLS_CANCELLED;
1389 olck->ols_flags &= ~LDLM_FL_LVB_READY;
1390 osc_lock_detach(env, olck);
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)
1397 struct cl_lock *lock = cbdata;
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,
1404 cl_lock_put(env, tmp);
1405 return CLP_GANG_OKAY;
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;
1415 * Returns true iff there are pages under \a olck not protected by other
1418 static int osc_lock_has_pages(struct osc_lock *olck)
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;
1429 env = cl_env_nested_get(&nest);
1433 obj = olck->ols_cl.cls_obj;
1435 io = &oob->oo_debug_io;
1436 lock = olck->ols_cl.cls_lock;
1437 descr = &lock->cll_descr;
1439 mutex_lock(&oob->oo_debug_mutex);
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);
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)
1450 if (result == CLP_GANG_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);
1457 return (result == CLP_GANG_ABORT);
1460 static int osc_lock_has_pages(struct osc_lock *olck)
1464 #endif /* INVARIANT_CHECK */
1466 static void osc_lock_delete(const struct lu_env *env,
1467 const struct cl_lock_slice *slice)
1469 struct osc_lock *olck;
1471 olck = cl2osc_lock(slice);
1472 if (olck->ols_glimpse) {
1473 LASSERT(!olck->ols_hold);
1474 LASSERT(!olck->ols_lock);
1478 LINVRNT(osc_lock_invariant(olck));
1479 LINVRNT(!osc_lock_has_pages(olck));
1481 osc_lock_unhold(olck);
1482 osc_lock_detach(env, olck);
1486 * Implements cl_lock_operations::clo_state() method for osc layer.
1488 * Maintains osc_lock::ols_owner field.
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.
1495 static void osc_lock_state(const struct lu_env *env,
1496 const struct cl_lock_slice *slice,
1497 enum cl_lock_state state)
1499 struct osc_lock *lock = cl2osc_lock(slice);
1502 * XXX multiple io contexts can use the lock at the same time.
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);
1508 LASSERT(lock->ols_owner == NULL);
1509 lock->ols_owner = oio;
1510 } else if (state != CLS_HELD)
1511 lock->ols_owner = NULL;
1514 static int osc_lock_print(const struct lu_env *env, void *cookie,
1515 lu_printer_t p, const struct cl_lock_slice *slice)
1517 struct osc_lock *lock = cl2osc_lock(slice);
1520 * XXX print ldlm lock and einfo properly.
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);
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)
1534 struct osc_lock *ols = cl2osc_lock(slice);
1536 if (need->cld_enq_flags & CEF_NEVER)
1539 if (ols->ols_state >= OLS_CANCELLED)
1542 if (need->cld_mode == CLM_PHANTOM) {
1544 return !(ols->ols_state > OLS_RELEASED);
1547 * Note: the QUEUED lock can't be matched here, otherwise
1548 * it might cause the deadlocks.
1550 * P1: enqueued read lock, create sublock1
1551 * P2: enqueued write lock, create sublock2(conflicted
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!
1559 if (ols->ols_state < OLS_GRANTED ||
1560 ols->ols_state > OLS_RELEASED)
1562 } else if (need->cld_enq_flags & CEF_MUST) {
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,
1569 if (ols->ols_state < OLS_UPCALL_RECEIVED &&
1570 ols->ols_locklessable)
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,
1590 static int osc_lock_lockless_unuse(const struct lu_env *env,
1591 const struct cl_lock_slice *slice)
1593 struct osc_lock *ols = cl2osc_lock(slice);
1594 struct cl_lock *lock = slice->cls_lock;
1596 LASSERT(ols->ols_state == OLS_GRANTED);
1597 LINVRNT(osc_lock_invariant(ols));
1599 cl_lock_cancel(env, lock);
1600 cl_lock_delete(env, lock);
1604 static void osc_lock_lockless_cancel(const struct lu_env *env,
1605 const struct cl_lock_slice *slice)
1607 struct osc_lock *ols = cl2osc_lock(slice);
1610 result = osc_lock_flush(ols, 0);
1612 CERROR("Pages for lockless lock %p were not purged(%d)\n",
1614 ols->ols_state = OLS_CANCELLED;
1617 static int osc_lock_lockless_wait(const struct lu_env *env,
1618 const struct cl_lock_slice *slice)
1620 struct osc_lock *olck = cl2osc_lock(slice);
1621 struct cl_lock *lock = olck->ols_cl.cls_lock;
1623 LINVRNT(osc_lock_invariant(olck));
1624 LASSERT(olck->ols_state >= OLS_UPCALL_RECEIVED);
1626 return lock->cll_error;
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)
1633 struct osc_lock *lock = cl2osc_lock(slice);
1635 LINVRNT(osc_lock_invariant(lock));
1636 if (state == CLS_HELD) {
1637 struct osc_io *oio = osc_env_io(env);
1639 LASSERT(ergo(lock->ols_owner, lock->ols_owner == oio));
1640 lock->ols_owner = oio;
1642 /* set the io to be lockless if this lock is for io's
1644 if (cl_object_same(oio->oi_cl.cis_obj, slice->cls_obj))
1645 oio->oi_lockless = 1;
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)
1654 struct osc_lock *lock = cl2osc_lock(slice);
1656 if (!(need->cld_enq_flags & CEF_NEVER))
1659 /* lockless lock should only be used by its owning io. b22147 */
1660 return (lock->ols_owner == osc_env_io(env));
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
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)
1678 struct osc_lock *clk;
1681 OBD_SLAB_ALLOC_PTR_GFP(clk, osc_lock_kmem, CFS_ALLOC_IO);
1683 __u32 enqflags = lock->cll_descr.cld_enq_flags;
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;
1689 clk->ols_flags = osc_enq2ldlm_flags(enqflags);
1690 clk->ols_agl = !!(enqflags & CEF_AGL);
1692 clk->ols_flags |= LDLM_FL_BLOCK_NOWAIT;
1693 if (clk->ols_flags & LDLM_FL_HAS_INTENT)
1694 clk->ols_glimpse = 1;
1696 cl_lock_slice_add(lock, &clk->ols_cl, obj, &osc_lock_ops);
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;
1704 LDLM_DEBUG_NOLOCK("lock %p, osc lock %p, flags %llx\n",
1705 lock, clk, clk->ols_flags);
1713 int osc_dlm_lock_pageref(struct ldlm_lock *dlm)
1715 struct osc_lock *olock;
1718 spin_lock(&osc_ast_guard);
1719 olock = dlm->l_ast_data;
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.
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);
1731 spin_unlock(&osc_ast_guard);