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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2017, Intel Corporation.
26 * This file is part of Lustre, http://www.lustre.org/
28 * Implementation of cl_device, cl_req for MDC layer.
30 * Author: Mikhail Pershin <mike.pershin@intel.com>
33 #define DEBUG_SUBSYSTEM S_MDC
35 #include <obd_class.h>
36 #include <lustre_osc.h>
37 #include <uapi/linux/lustre/lustre_param.h>
39 #include "mdc_internal.h"
41 static void mdc_lock_build_policy(const struct lu_env *env,
42 union ldlm_policy_data *policy)
44 memset(policy, 0, sizeof *policy);
45 policy->l_inodebits.bits = MDS_INODELOCK_DOM;
48 int mdc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data)
50 return osc_ldlm_glimpse_ast(dlmlock, data);
53 static void mdc_lock_build_einfo(const struct lu_env *env,
54 const struct cl_lock *lock,
55 struct osc_object *osc,
56 struct ldlm_enqueue_info *einfo)
58 einfo->ei_type = LDLM_IBITS;
59 einfo->ei_mode = osc_cl_lock2ldlm(lock->cll_descr.cld_mode);
60 einfo->ei_cb_bl = mdc_ldlm_blocking_ast;
61 einfo->ei_cb_cp = ldlm_completion_ast;
62 einfo->ei_cb_gl = mdc_ldlm_glimpse_ast;
63 einfo->ei_cbdata = osc; /* value to be put into ->l_ast_data */
66 static void mdc_lock_lvb_update(const struct lu_env *env,
67 struct osc_object *osc,
68 struct ldlm_lock *dlmlock,
71 static int mdc_set_dom_lock_data(struct ldlm_lock *lock, void *data)
75 LASSERT(lock != NULL);
76 LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
78 lock_res_and_lock(lock);
80 if (lock->l_ast_data == NULL)
81 lock->l_ast_data = data;
82 if (lock->l_ast_data == data)
85 unlock_res_and_lock(lock);
90 int mdc_dom_lock_match(const struct lu_env *env, struct obd_export *exp,
91 struct ldlm_res_id *res_id, enum ldlm_type type,
92 union ldlm_policy_data *policy, enum ldlm_mode mode,
93 __u64 *flags, struct osc_object *obj,
94 struct lustre_handle *lockh, int unref)
96 struct obd_device *obd = exp->exp_obd;
97 __u64 lflags = *flags;
102 rc = ldlm_lock_match(obd->obd_namespace, lflags,
103 res_id, type, policy, mode, lockh, unref);
104 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
108 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
110 LASSERT(lock != NULL);
111 if (mdc_set_dom_lock_data(lock, obj)) {
112 lock_res_and_lock(lock);
113 if (!ldlm_is_lvb_cached(lock)) {
114 LASSERT(lock->l_ast_data == obj);
115 mdc_lock_lvb_update(env, obj, lock, NULL);
116 ldlm_set_lvb_cached(lock);
118 unlock_res_and_lock(lock);
120 ldlm_lock_decref(lockh, rc);
129 * Finds an existing lock covering a page with given index.
130 * Copy of osc_obj_dlmlock_at_pgoff() but for DoM IBITS lock.
132 struct ldlm_lock *mdc_dlmlock_at_pgoff(const struct lu_env *env,
133 struct osc_object *obj, pgoff_t index,
134 enum osc_dap_flags dap_flags)
136 struct osc_thread_info *info = osc_env_info(env);
137 struct ldlm_res_id *resname = &info->oti_resname;
138 union ldlm_policy_data *policy = &info->oti_policy;
139 struct lustre_handle lockh;
140 struct ldlm_lock *lock = NULL;
146 fid_build_reg_res_name(lu_object_fid(osc2lu(obj)), resname);
147 mdc_lock_build_policy(env, policy);
149 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
150 if (dap_flags & OSC_DAP_FL_TEST_LOCK)
151 flags |= LDLM_FL_TEST_LOCK;
154 /* Next, search for already existing extent locks that will cover us */
155 /* If we're trying to read, we also search for an existing PW lock. The
156 * VFS and page cache already protect us locally, so lots of readers/
157 * writers can share a single PW lock. */
158 mode = mdc_dom_lock_match(env, osc_export(obj), resname, LDLM_IBITS,
159 policy, LCK_PR | LCK_PW | LCK_GROUP, &flags,
161 dap_flags & OSC_DAP_FL_CANCELING);
163 lock = ldlm_handle2lock(&lockh);
164 /* RACE: the lock is cancelled so let's try again */
165 if (unlikely(lock == NULL))
173 * Check if page @page is covered by an extra lock or discard it.
175 static int mdc_check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
176 struct osc_page *ops, void *cbdata)
178 struct osc_thread_info *info = osc_env_info(env);
179 struct osc_object *osc = cbdata;
182 index = osc_index(ops);
183 if (index >= info->oti_fn_index) {
184 struct ldlm_lock *tmp;
185 struct cl_page *page = ops->ops_cl.cpl_page;
187 /* refresh non-overlapped index */
188 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
189 OSC_DAP_FL_TEST_LOCK);
191 info->oti_fn_index = CL_PAGE_EOF;
193 } else if (cl_page_own(env, io, page) == 0) {
194 /* discard the page */
195 cl_page_discard(env, io, page);
196 cl_page_disown(env, io, page);
198 LASSERT(page->cp_state == CPS_FREEING);
202 info->oti_next_index = index + 1;
203 return CLP_GANG_OKAY;
207 * Discard pages protected by the given lock. This function traverses radix
208 * tree to find all covering pages and discard them. If a page is being covered
209 * by other locks, it should remain in cache.
211 * If error happens on any step, the process continues anyway (the reasoning
212 * behind this being that lock cancellation cannot be delayed indefinitely).
214 static int mdc_lock_discard_pages(const struct lu_env *env,
215 struct osc_object *osc,
216 pgoff_t start, pgoff_t end,
219 struct osc_thread_info *info = osc_env_info(env);
220 struct cl_io *io = &info->oti_io;
221 osc_page_gang_cbt cb;
227 io->ci_obj = cl_object_top(osc2cl(osc));
228 io->ci_ignore_layout = 1;
229 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
233 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
234 info->oti_fn_index = info->oti_next_index = start;
236 res = osc_page_gang_lookup(env, io, osc, info->oti_next_index,
237 end, cb, (void *)osc);
238 if (info->oti_next_index > end)
241 if (res == CLP_GANG_RESCHED)
243 } while (res != CLP_GANG_OKAY);
249 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
250 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
258 if (mode == CLM_WRITE) {
259 result = osc_cache_writeback_range(env, obj, start, end, 1,
261 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
262 obj, start, end, result,
263 discard ? "discarded" : "written back");
268 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
269 rc = mdc_lock_discard_pages(env, obj, start, end,
270 mode == CLM_WRITE || discard);
271 if (result == 0 && rc < 0)
277 void mdc_lock_lockless_cancel(const struct lu_env *env,
278 const struct cl_lock_slice *slice)
280 struct osc_lock *ols = cl2osc_lock(slice);
281 struct osc_object *osc = cl2osc(slice->cls_obj);
282 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
285 LASSERT(ols->ols_dlmlock == NULL);
286 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
289 CERROR("Pages for lockless lock %p were not purged(%d)\n",
292 osc_lock_wake_waiters(env, osc, ols);
296 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
297 * and ldlm_lock caches.
299 static int mdc_dlm_blocking_ast0(const struct lu_env *env,
300 struct ldlm_lock *dlmlock,
303 struct cl_object *obj = NULL;
306 enum cl_lock_mode mode = CLM_READ;
310 LASSERT(flag == LDLM_CB_CANCELING);
311 LASSERT(dlmlock != NULL);
313 lock_res_and_lock(dlmlock);
314 if (dlmlock->l_granted_mode != dlmlock->l_req_mode) {
315 dlmlock->l_ast_data = NULL;
316 unlock_res_and_lock(dlmlock);
320 discard = ldlm_is_discard_data(dlmlock);
321 if (dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))
324 if (dlmlock->l_ast_data != NULL) {
325 obj = osc2cl(dlmlock->l_ast_data);
328 unlock_res_and_lock(dlmlock);
330 /* if l_ast_data is NULL, the dlmlock was enqueued by AGL or
331 * the object has been destroyed. */
333 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
335 /* Destroy pages covered by the extent of the DLM lock */
336 result = mdc_lock_flush(env, cl2osc(obj), cl_index(obj, 0),
337 CL_PAGE_EOF, mode, discard);
338 /* Losing a lock, set KMS to 0.
339 * NB: assumed that DOM lock covers whole data on MDT.
341 /* losing a lock, update kms */
342 lock_res_and_lock(dlmlock);
343 dlmlock->l_ast_data = NULL;
344 cl_object_attr_lock(obj);
346 cl_object_attr_update(env, obj, attr, CAT_KMS);
347 cl_object_attr_unlock(obj);
348 unlock_res_and_lock(dlmlock);
349 cl_object_put(env, obj);
354 int mdc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
355 struct ldlm_lock_desc *new, void *data, int flag)
362 case LDLM_CB_BLOCKING: {
363 struct lustre_handle lockh;
365 ldlm_lock2handle(dlmlock, &lockh);
366 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
371 case LDLM_CB_CANCELING: {
376 * This can be called in the context of outer IO, e.g.,
378 * osc_enqueue_base()->...
379 * ->ldlm_prep_elc_req()->...
380 * ->ldlm_cancel_callback()->...
381 * ->osc_ldlm_blocking_ast()
383 * new environment has to be created to not corrupt outer
386 env = cl_env_get(&refcheck);
392 rc = mdc_dlm_blocking_ast0(env, dlmlock, flag);
393 cl_env_put(env, &refcheck);
403 * Updates object attributes from a lock value block (lvb) received together
404 * with the DLM lock reply from the server.
405 * This can be optimized to not update attributes when lock is a result of a
408 * Called under lock and resource spin-locks.
410 void mdc_lock_lvb_update(const struct lu_env *env, struct osc_object *osc,
411 struct ldlm_lock *dlmlock, struct ost_lvb *lvb)
413 struct cl_object *obj = osc2cl(osc);
414 struct lov_oinfo *oinfo = osc->oo_oinfo;
415 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
416 unsigned valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME |
418 unsigned int setkms = 0;
423 LASSERT(dlmlock != NULL);
424 lvb = &dlmlock->l_ost_lvb;
426 cl_lvb2attr(attr, lvb);
428 cl_object_attr_lock(obj);
429 if (dlmlock != NULL) {
432 check_res_locked(dlmlock->l_resource);
433 size = lvb->lvb_size;
435 if (size >= oinfo->loi_kms) {
437 attr->cat_kms = size;
442 /* The size should not be less than the kms */
443 if (attr->cat_size < oinfo->loi_kms)
444 attr->cat_size = oinfo->loi_kms;
446 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
447 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
448 setkms ? "" : " leaving",
449 setkms ? attr->cat_kms : oinfo->loi_kms,
450 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
452 cl_object_attr_update(env, obj, attr, valid);
453 cl_object_attr_unlock(obj);
457 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
458 struct lustre_handle *lockh)
460 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
461 struct ldlm_lock *dlmlock;
465 dlmlock = ldlm_handle2lock_long(lockh, 0);
466 LASSERT(dlmlock != NULL);
468 /* lock reference taken by ldlm_handle2lock_long() is
469 * owned by osc_lock and released in osc_lock_detach()
471 lu_ref_add(&dlmlock->l_reference, "osc_lock", oscl);
472 oscl->ols_has_ref = 1;
474 LASSERT(oscl->ols_dlmlock == NULL);
475 oscl->ols_dlmlock = dlmlock;
477 /* This may be a matched lock for glimpse request, do not hold
478 * lock reference in that case. */
479 if (!oscl->ols_glimpse) {
480 /* hold a refc for non glimpse lock which will
481 * be released in osc_lock_cancel() */
482 lustre_handle_copy(&oscl->ols_handle, lockh);
483 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
487 /* Lock must have been granted. */
488 lock_res_and_lock(dlmlock);
489 if (dlmlock->l_granted_mode == dlmlock->l_req_mode) {
490 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
492 /* extend the lock extent, otherwise it will have problem when
493 * we decide whether to grant a lockless lock. */
494 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
495 descr->cld_start = cl_index(descr->cld_obj, 0);
496 descr->cld_end = CL_PAGE_EOF;
498 /* no lvb update for matched lock */
499 if (!ldlm_is_lvb_cached(dlmlock)) {
500 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
501 LASSERT(osc == dlmlock->l_ast_data);
502 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
503 ldlm_set_lvb_cached(dlmlock);
506 unlock_res_and_lock(dlmlock);
508 LASSERT(oscl->ols_state != OLS_GRANTED);
509 oscl->ols_state = OLS_GRANTED;
514 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
515 * received from a server, or after osc_enqueue_base() matched a local DLM
518 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
521 struct osc_lock *oscl = cookie;
522 struct cl_lock_slice *slice = &oscl->ols_cl;
528 env = cl_env_percpu_get();
529 /* should never happen, similar to osc_ldlm_blocking_ast(). */
530 LASSERT(!IS_ERR(env));
532 rc = ldlm_error2errno(errcode);
533 if (oscl->ols_state == OLS_ENQUEUED) {
534 oscl->ols_state = OLS_UPCALL_RECEIVED;
535 } else if (oscl->ols_state == OLS_CANCELLED) {
538 CERROR("Impossible state: %d\n", oscl->ols_state);
542 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
544 mdc_lock_granted(env, oscl, lockh);
546 /* Error handling, some errors are tolerable. */
547 if (oscl->ols_locklessable && rc == -EUSERS) {
548 /* This is a tolerable error, turn this lock into
551 osc_object_set_contended(cl2osc(slice->cls_obj));
552 LASSERT(slice->cls_ops != oscl->ols_lockless_ops);
554 /* Change this lock to ldlmlock-less lock. */
555 osc_lock_to_lockless(env, oscl, 1);
556 oscl->ols_state = OLS_GRANTED;
558 } else if (oscl->ols_glimpse && rc == -ENAVAIL) {
559 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
560 mdc_lock_lvb_update(env, cl2osc(slice->cls_obj),
561 NULL, &oscl->ols_lvb);
562 /* Hide the error. */
566 if (oscl->ols_owner != NULL)
567 cl_sync_io_note(env, oscl->ols_owner, rc);
568 cl_env_percpu_put(env);
573 int mdc_fill_lvb(struct ptlrpc_request *req, struct ost_lvb *lvb)
575 struct mdt_body *body;
577 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
581 lvb->lvb_mtime = body->mbo_mtime;
582 lvb->lvb_atime = body->mbo_atime;
583 lvb->lvb_ctime = body->mbo_ctime;
584 lvb->lvb_blocks = body->mbo_dom_blocks;
585 lvb->lvb_size = body->mbo_dom_size;
590 int mdc_enqueue_fini(struct ptlrpc_request *req, osc_enqueue_upcall_f upcall,
591 void *cookie, struct lustre_handle *lockh,
592 enum ldlm_mode mode, __u64 *flags, int errcode)
594 struct osc_lock *ols = cookie;
595 struct ldlm_lock *lock;
600 /* The request was created before ldlm_cli_enqueue call. */
601 if (errcode == ELDLM_LOCK_ABORTED) {
602 struct ldlm_reply *rep;
604 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
605 LASSERT(rep != NULL);
607 rep->lock_policy_res2 =
608 ptlrpc_status_ntoh(rep->lock_policy_res2);
609 if (rep->lock_policy_res2)
610 errcode = rep->lock_policy_res2;
612 rc = mdc_fill_lvb(req, &ols->ols_lvb);
613 *flags |= LDLM_FL_LVB_READY;
614 } else if (errcode == ELDLM_OK) {
615 /* Callers have references, should be valid always */
616 lock = ldlm_handle2lock(lockh);
619 rc = mdc_fill_lvb(req, &lock->l_ost_lvb);
621 *flags |= LDLM_FL_LVB_READY;
624 /* Call the update callback. */
625 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
627 /* release the reference taken in ldlm_cli_enqueue() */
628 if (errcode == ELDLM_LOCK_MATCHED)
630 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
631 ldlm_lock_decref(lockh, mode);
636 int mdc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
639 struct osc_enqueue_args *aa = args;
640 struct ldlm_lock *lock;
641 struct lustre_handle *lockh = &aa->oa_lockh;
642 enum ldlm_mode mode = aa->oa_mode;
646 LASSERT(!aa->oa_speculative);
648 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
650 lock = ldlm_handle2lock(lockh);
651 LASSERTF(lock != NULL,
652 "lockh %#llx, req %p, aa %p - client evicted?\n",
653 lockh->cookie, req, aa);
655 /* Take an additional reference so that a blocking AST that
656 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
657 * to arrive after an upcall has been executed by
658 * osc_enqueue_fini(). */
659 ldlm_lock_addref(lockh, mode);
661 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
662 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
664 /* Let CP AST to grant the lock first. */
665 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
667 /* Complete obtaining the lock procedure. */
668 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
669 aa->oa_mode, aa->oa_flags, NULL, 0,
671 /* Complete mdc stuff. */
672 rc = mdc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
675 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
677 ldlm_lock_decref(lockh, mode);
682 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
683 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
684 * other synchronous requests, however keeping some locks and trying to obtain
685 * others may take a considerable amount of time in a case of ost failure; and
686 * when other sync requests do not get released lock from a client, the client
687 * is excluded from the cluster -- such scenarious make the life difficult, so
688 * release locks just after they are obtained. */
689 int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
690 struct ldlm_res_id *res_id, __u64 *flags,
691 union ldlm_policy_data *policy,
692 struct ost_lvb *lvb, int kms_valid,
693 osc_enqueue_upcall_f upcall, void *cookie,
694 struct ldlm_enqueue_info *einfo, int async)
696 struct obd_device *obd = exp->exp_obd;
697 struct lustre_handle lockh = { 0 };
698 struct ptlrpc_request *req = NULL;
699 struct ldlm_intent *lit;
701 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
702 __u64 match_flags = *flags;
708 mode = einfo->ei_mode;
709 if (einfo->ei_mode == LCK_PR)
713 match_flags |= LDLM_FL_BLOCK_GRANTED;
714 /* DOM locking uses LDLM_FL_KMS_IGNORE to mark locks wich have no valid
715 * LVB information, e.g. canceled locks or locks of just pruned object,
716 * such locks should be skipped.
718 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
719 einfo->ei_type, policy, mode, &lockh, 0);
721 struct ldlm_lock *matched;
723 if (*flags & LDLM_FL_TEST_LOCK)
726 matched = ldlm_handle2lock(&lockh);
728 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
729 ldlm_set_kms_ignore(matched);
731 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
732 *flags |= LDLM_FL_LVB_READY;
734 /* We already have a lock, and it's referenced. */
735 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
737 ldlm_lock_decref(&lockh, mode);
738 LDLM_LOCK_PUT(matched);
741 ldlm_lock_decref(&lockh, mode);
742 LDLM_LOCK_PUT(matched);
745 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
748 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_INTENT);
752 /* For WRITE lock cancel other locks on resource early if any */
753 if (einfo->ei_mode & LCK_PW)
754 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
760 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
762 ptlrpc_request_free(req);
766 /* pack the intent */
767 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
768 lit->opc = glimpse ? IT_GLIMPSE : IT_BRW;
770 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
771 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
772 ptlrpc_request_set_replen(req);
774 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
775 *flags &= ~LDLM_FL_BLOCK_GRANTED;
776 /* All MDC IO locks are intents */
777 *flags |= LDLM_FL_HAS_INTENT;
778 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, NULL,
779 0, LVB_T_NONE, &lockh, async);
782 struct osc_enqueue_args *aa;
784 aa = ptlrpc_req_async_args(aa, req);
786 aa->oa_mode = einfo->ei_mode;
787 aa->oa_type = einfo->ei_type;
788 lustre_handle_copy(&aa->oa_lockh, &lockh);
789 aa->oa_upcall = upcall;
790 aa->oa_cookie = cookie;
791 aa->oa_speculative = false;
792 aa->oa_flags = flags;
795 req->rq_interpret_reply = mdc_enqueue_interpret;
796 ptlrpcd_add_req(req);
798 ptlrpc_req_finished(req);
803 rc = mdc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
805 ptlrpc_req_finished(req);
810 * Implementation of cl_lock_operations::clo_enqueue() method for osc
811 * layer. This initiates ldlm enqueue:
813 * - cancels conflicting locks early (osc_lock_enqueue_wait());
815 * - calls osc_enqueue_base() to do actual enqueue.
817 * osc_enqueue_base() is supplied with an upcall function that is executed
818 * when lock is received either after a local cached ldlm lock is matched, or
819 * when a reply from the server is received.
821 * This function does not wait for the network communication to complete.
823 static int mdc_lock_enqueue(const struct lu_env *env,
824 const struct cl_lock_slice *slice,
825 struct cl_io *unused, struct cl_sync_io *anchor)
827 struct osc_thread_info *info = osc_env_info(env);
828 struct osc_io *oio = osc_env_io(env);
829 struct osc_object *osc = cl2osc(slice->cls_obj);
830 struct osc_lock *oscl = cl2osc_lock(slice);
831 struct cl_lock *lock = slice->cls_lock;
832 struct ldlm_res_id *resname = &info->oti_resname;
833 union ldlm_policy_data *policy = &info->oti_policy;
834 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
835 void *cookie = (void *)oscl;
841 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
842 "lock = %p, ols = %p\n", lock, oscl);
844 if (oscl->ols_state == OLS_GRANTED)
847 /* Lockahead is not supported on MDT yet */
848 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
849 result = -EOPNOTSUPP;
853 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
854 GOTO(enqueue_base, 0);
856 if (oscl->ols_glimpse) {
857 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
859 GOTO(enqueue_base, 0);
862 result = osc_lock_enqueue_wait(env, osc, oscl);
866 /* we can grant lockless lock right after all conflicting locks
868 if (osc_lock_is_lockless(oscl)) {
869 oscl->ols_state = OLS_GRANTED;
870 oio->oi_lockless = 1;
875 oscl->ols_state = OLS_ENQUEUED;
876 if (anchor != NULL) {
877 atomic_inc(&anchor->csi_sync_nr);
878 oscl->ols_owner = anchor;
882 * DLM lock's ast data must be osc_object;
883 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
886 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
887 mdc_lock_build_policy(env, policy);
888 LASSERT(!oscl->ols_speculative);
889 result = mdc_enqueue_send(env, osc_export(osc), resname,
890 &oscl->ols_flags, policy,
891 &oscl->ols_lvb, osc->oo_oinfo->loi_kms_valid,
892 upcall, cookie, &oscl->ols_einfo, async);
894 if (osc_lock_is_lockless(oscl)) {
895 oio->oi_lockless = 1;
897 LASSERT(oscl->ols_state == OLS_GRANTED);
898 LASSERT(oscl->ols_hold);
899 LASSERT(oscl->ols_dlmlock != NULL);
904 oscl->ols_state = OLS_CANCELLED;
905 osc_lock_wake_waiters(env, osc, oscl);
908 cl_sync_io_note(env, anchor, result);
913 static const struct cl_lock_operations mdc_lock_lockless_ops = {
914 .clo_fini = osc_lock_fini,
915 .clo_enqueue = mdc_lock_enqueue,
916 .clo_cancel = mdc_lock_lockless_cancel,
917 .clo_print = osc_lock_print
920 static const struct cl_lock_operations mdc_lock_ops = {
921 .clo_fini = osc_lock_fini,
922 .clo_enqueue = mdc_lock_enqueue,
923 .clo_cancel = osc_lock_cancel,
924 .clo_print = osc_lock_print,
927 int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
928 struct cl_lock *lock, const struct cl_io *io)
930 struct osc_lock *ols;
931 __u32 enqflags = lock->cll_descr.cld_enq_flags;
932 __u64 flags = osc_enq2ldlm_flags(enqflags);
936 /* Ignore AGL for Data-on-MDT, stat returns size data */
937 if ((enqflags & CEF_SPECULATIVE) != 0)
940 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
941 if (unlikely(ols == NULL))
944 ols->ols_state = OLS_NEW;
945 spin_lock_init(&ols->ols_lock);
946 INIT_LIST_HEAD(&ols->ols_waiting_list);
947 INIT_LIST_HEAD(&ols->ols_wait_entry);
948 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
949 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
951 ols->ols_flags = flags;
952 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
954 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
955 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
956 ols->ols_glimpse = 1;
958 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
960 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
962 if (!(enqflags & CEF_MUST))
963 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
964 if (ols->ols_locklessable && !(enqflags & CEF_DISCARD_DATA))
965 ols->ols_flags |= LDLM_FL_DENY_ON_CONTENTION;
967 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
968 osc_lock_set_writer(env, io, obj, ols);
970 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx\n",
971 lock, ols, ols->ols_flags);
978 * An implementation of cl_io_operations specific methods for MDC layer.
981 static int mdc_async_upcall(void *a, int rc)
983 struct osc_async_cbargs *args = a;
986 complete(&args->opc_sync);
990 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
991 pgoff_t index, struct lustre_handle *lh)
993 struct ldlm_lock *lock;
995 /* find DOM lock protecting object */
996 lock = mdc_dlmlock_at_pgoff(env, osc, index,
997 OSC_DAP_FL_TEST_LOCK |
998 OSC_DAP_FL_CANCELING);
1000 struct ldlm_resource *res;
1001 struct ldlm_res_id *resname;
1003 resname = &osc_env_info(env)->oti_resname;
1004 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1005 res = ldlm_resource_get(osc_export(osc)->exp_obd->obd_namespace,
1006 NULL, resname, LDLM_IBITS, 0);
1007 ldlm_resource_dump(D_ERROR, res);
1008 libcfs_debug_dumpstack(NULL);
1011 *lh = lock->l_remote_handle;
1012 LDLM_LOCK_PUT(lock);
1017 static int mdc_io_setattr_start(const struct lu_env *env,
1018 const struct cl_io_slice *slice)
1020 struct cl_io *io = slice->cis_io;
1021 struct osc_io *oio = cl2osc_io(env, slice);
1022 struct cl_object *obj = slice->cis_obj;
1023 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1024 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1025 struct obdo *oa = &oio->oi_oa;
1026 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1027 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1028 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1029 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1032 /* silently ignore non-truncate setattr for Data-on-MDT object */
1033 if (cl_io_is_trunc(io)) {
1034 /* truncate cache dirty pages first */
1035 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1041 if (oio->oi_lockless == 0) {
1042 cl_object_attr_lock(obj);
1043 rc = cl_object_attr_get(env, obj, attr);
1045 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1046 unsigned int cl_valid = 0;
1048 if (ia_avalid & ATTR_SIZE) {
1049 attr->cat_size = size;
1050 attr->cat_kms = size;
1051 cl_valid = (CAT_SIZE | CAT_KMS);
1053 if (ia_avalid & ATTR_MTIME_SET) {
1054 attr->cat_mtime = lvb->lvb_mtime;
1055 cl_valid |= CAT_MTIME;
1057 if (ia_avalid & ATTR_ATIME_SET) {
1058 attr->cat_atime = lvb->lvb_atime;
1059 cl_valid |= CAT_ATIME;
1061 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1062 attr->cat_ctime = lvb->lvb_ctime;
1063 cl_valid |= CAT_CTIME;
1065 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1067 cl_object_attr_unlock(obj);
1072 if (!(ia_avalid & ATTR_SIZE))
1075 memset(oa, 0, sizeof(*oa));
1076 oa->o_oi = loi->loi_oi;
1077 oa->o_mtime = attr->cat_mtime;
1078 oa->o_atime = attr->cat_atime;
1079 oa->o_ctime = attr->cat_ctime;
1082 oa->o_blocks = OBD_OBJECT_EOF;
1083 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1084 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1086 if (oio->oi_lockless) {
1087 oa->o_flags = OBD_FL_SRVLOCK;
1088 oa->o_valid |= OBD_MD_FLFLAGS;
1090 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1093 oa->o_valid |= OBD_MD_FLHANDLE;
1096 init_completion(&cbargs->opc_sync);
1098 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1099 mdc_async_upcall, cbargs);
1100 cbargs->opc_rpc_sent = rc == 0;
1104 static int mdc_io_read_ahead(const struct lu_env *env,
1105 const struct cl_io_slice *ios,
1106 pgoff_t start, struct cl_read_ahead *ra)
1108 struct osc_object *osc = cl2osc(ios->cis_obj);
1109 struct ldlm_lock *dlmlock;
1113 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1114 if (dlmlock == NULL)
1117 if (dlmlock->l_req_mode != LCK_PR) {
1118 struct lustre_handle lockh;
1120 ldlm_lock2handle(dlmlock, &lockh);
1121 ldlm_lock_addref(&lockh, LCK_PR);
1122 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1125 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1126 ra->cra_end_idx = CL_PAGE_EOF;
1127 ra->cra_release = osc_read_ahead_release;
1128 ra->cra_cbdata = dlmlock;
1133 int mdc_io_fsync_start(const struct lu_env *env,
1134 const struct cl_io_slice *slice)
1136 struct cl_io *io = slice->cis_io;
1137 struct cl_fsync_io *fio = &io->u.ci_fsync;
1138 struct cl_object *obj = slice->cis_obj;
1139 struct osc_object *osc = cl2osc(obj);
1144 /* a MDC lock always covers whole object, do sync for whole
1145 * possible range despite of supplied start/end values.
1147 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1148 fio->fi_mode == CL_FSYNC_DISCARD);
1150 fio->fi_nr_written += result;
1153 if (fio->fi_mode == CL_FSYNC_ALL) {
1156 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1159 /* Use OSC sync code because it is asynchronous.
1160 * It is to be added into MDC and avoid the using of
1161 * OST_SYNC at both MDC and MDT.
1163 rc = osc_fsync_ost(env, osc, fio);
1171 struct mdc_data_version_args {
1172 struct osc_io *dva_oio;
1176 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1179 struct mdc_data_version_args *dva = args;
1180 struct osc_io *oio = dva->dva_oio;
1181 const struct mdt_body *body;
1187 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1189 GOTO(out, rc = -EPROTO);
1191 /* Prepare OBDO from mdt_body for CLIO */
1192 oio->oi_oa.o_valid = body->mbo_valid;
1193 oio->oi_oa.o_flags = body->mbo_flags;
1194 oio->oi_oa.o_data_version = body->mbo_version;
1195 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1198 oio->oi_cbarg.opc_rc = rc;
1199 complete(&oio->oi_cbarg.opc_sync);
1203 static int mdc_io_data_version_start(const struct lu_env *env,
1204 const struct cl_io_slice *slice)
1206 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1207 struct osc_io *oio = cl2osc_io(env, slice);
1208 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1209 struct osc_object *obj = cl2osc(slice->cis_obj);
1210 struct obd_export *exp = osc_export(obj);
1211 struct ptlrpc_request *req;
1212 struct mdt_body *body;
1213 struct mdc_data_version_args *dva;
1218 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1219 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1220 oio->oi_oa.o_valid = OBD_MD_FLID;
1222 init_completion(&cbargs->opc_sync);
1224 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1228 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1230 ptlrpc_request_free(req);
1234 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1235 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1236 body->mbo_valid = OBD_MD_FLID;
1237 /* Indicate that data version is needed */
1238 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1239 body->mbo_flags = 0;
1241 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1242 body->mbo_valid |= OBD_MD_FLFLAGS;
1243 body->mbo_flags |= OBD_FL_SRVLOCK;
1244 if (dv->dv_flags & LL_DV_WR_FLUSH)
1245 body->mbo_flags |= OBD_FL_FLUSH;
1248 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1249 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1250 ptlrpc_request_set_replen(req);
1252 req->rq_interpret_reply = mdc_data_version_interpret;
1253 dva = ptlrpc_req_async_args(dva, req);
1256 ptlrpcd_add_req(req);
1261 static void mdc_io_data_version_end(const struct lu_env *env,
1262 const struct cl_io_slice *slice)
1264 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1265 struct osc_io *oio = cl2osc_io(env, slice);
1266 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1269 wait_for_completion(&cbargs->opc_sync);
1271 if (cbargs->opc_rc != 0) {
1272 slice->cis_io->ci_result = cbargs->opc_rc;
1274 slice->cis_io->ci_result = 0;
1275 if (!(oio->oi_oa.o_valid &
1276 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1277 slice->cis_io->ci_result = -ENOTSUPP;
1279 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1280 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1281 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1282 dv->dv_data_version = oio->oi_oa.o_data_version;
1288 static struct cl_io_operations mdc_io_ops = {
1291 .cio_iter_init = osc_io_rw_iter_init,
1292 .cio_iter_fini = osc_io_rw_iter_fini,
1293 .cio_start = osc_io_read_start,
1296 .cio_iter_init = osc_io_rw_iter_init,
1297 .cio_iter_fini = osc_io_rw_iter_fini,
1298 .cio_start = osc_io_write_start,
1299 .cio_end = osc_io_end,
1302 .cio_iter_init = osc_io_iter_init,
1303 .cio_iter_fini = osc_io_iter_fini,
1304 .cio_start = mdc_io_setattr_start,
1305 .cio_end = osc_io_setattr_end,
1307 [CIT_DATA_VERSION] = {
1308 .cio_start = mdc_io_data_version_start,
1309 .cio_end = mdc_io_data_version_end,
1312 .cio_iter_init = osc_io_iter_init,
1313 .cio_iter_fini = osc_io_iter_fini,
1314 .cio_start = osc_io_fault_start,
1315 .cio_end = osc_io_end,
1318 .cio_start = mdc_io_fsync_start,
1319 .cio_end = osc_io_fsync_end,
1322 .cio_read_ahead = mdc_io_read_ahead,
1323 .cio_submit = osc_io_submit,
1324 .cio_commit_async = osc_io_commit_async,
1327 int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1330 struct osc_io *oio = osc_env_io(env);
1332 CL_IO_SLICE_CLEAN(oio, oi_cl);
1333 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1337 static void mdc_build_res_name(struct osc_object *osc,
1338 struct ldlm_res_id *resname)
1340 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1344 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1345 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1348 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1349 struct cl_req_attr *attr)
1351 u64 flags = attr->cra_flags;
1353 /* Copy object FID to cl_attr */
1354 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1356 if (flags & OBD_MD_FLGROUP)
1357 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1359 if (flags & OBD_MD_FLID)
1360 attr->cra_oa->o_valid |= OBD_MD_FLID;
1362 if (flags & OBD_MD_FLHANDLE) {
1363 struct osc_page *opg;
1365 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1366 if (!opg->ops_srvlock) {
1369 rc = mdc_get_lock_handle(env, cl2osc(obj),
1371 &attr->cra_oa->o_handle);
1373 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1374 "uncovered page!\n");
1377 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1383 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1384 struct cl_attr *attr)
1386 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1388 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1389 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1391 return osc_attr_get(env, obj, attr);
1394 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1396 struct osc_object *osc = (struct osc_object *)data;
1397 struct ost_lvb *lvb = &lock->l_ost_lvb;
1398 struct lov_oinfo *oinfo;
1401 if (lock->l_ast_data == data) {
1402 lock->l_ast_data = NULL;
1404 LASSERT(osc != NULL);
1405 LASSERT(osc->oo_oinfo != NULL);
1406 LASSERT(lvb != NULL);
1408 /* Updates lvb in lock by the cached oinfo */
1409 oinfo = osc->oo_oinfo;
1411 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1412 "%llu %llu %llu by oinfo size %llu blocks %llu "
1413 "[cma]time %llu %llu %llu", lvb->lvb_size,
1414 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1415 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1416 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1417 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1418 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1420 cl_object_attr_lock(&osc->oo_cl);
1421 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1422 cl_object_attr_unlock(&osc->oo_cl);
1423 ldlm_clear_lvb_cached(lock);
1425 RETURN(LDLM_ITER_CONTINUE);
1428 int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1430 struct osc_object *osc = cl2osc(obj);
1431 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1433 /* DLM locks don't hold a reference of osc_object so we have to
1434 * clear it before the object is being destroyed. */
1435 osc_build_res_name(osc, resname);
1436 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1437 mdc_object_ast_clear, osc);
1441 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1442 struct ldlm_lock *lock)
1444 RETURN(mdc_dlm_blocking_ast0(env, lock, LDLM_CB_CANCELING));
1447 static const struct cl_object_operations mdc_ops = {
1448 .coo_page_init = osc_page_init,
1449 .coo_lock_init = mdc_lock_init,
1450 .coo_io_init = mdc_io_init,
1451 .coo_attr_get = mdc_attr_get,
1452 .coo_attr_update = osc_attr_update,
1453 .coo_glimpse = osc_object_glimpse,
1454 .coo_req_attr_set = mdc_req_attr_set,
1455 .coo_prune = mdc_object_prune,
1456 .coo_object_flush = mdc_object_flush
1459 static const struct osc_object_operations mdc_object_ops = {
1460 .oto_build_res_name = mdc_build_res_name,
1461 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1464 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1465 const struct lu_object_conf *conf)
1467 struct osc_object *osc = lu2osc(obj);
1469 if (osc->oo_initialized)
1472 osc->oo_initialized = true;
1474 return osc_object_init(env, obj, conf);
1477 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1479 osc_object_free(env, obj);
1482 static const struct lu_object_operations mdc_lu_obj_ops = {
1483 .loo_object_init = mdc_object_init,
1484 .loo_object_delete = NULL,
1485 .loo_object_release = NULL,
1486 .loo_object_free = mdc_object_free,
1487 .loo_object_print = osc_object_print,
1488 .loo_object_invariant = NULL
1491 struct lu_object *mdc_object_alloc(const struct lu_env *env,
1492 const struct lu_object_header *unused,
1493 struct lu_device *dev)
1495 struct osc_object *osc;
1496 struct lu_object *obj;
1498 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1501 lu_object_init(obj, NULL, dev);
1502 osc->oo_cl.co_ops = &mdc_ops;
1503 obj->lo_ops = &mdc_lu_obj_ops;
1504 osc->oo_obj_ops = &mdc_object_ops;
1505 osc->oo_initialized = false;
1512 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1513 struct lustre_cfg *cfg)
1515 size_t count = class_modify_config(cfg, PARAM_MDC,
1516 &d->ld_obd->obd_kset.kobj);
1517 return count > 0 ? 0 : count;
1520 const struct lu_device_operations mdc_lu_ops = {
1521 .ldo_object_alloc = mdc_object_alloc,
1522 .ldo_process_config = mdc_process_config,
1523 .ldo_recovery_complete = NULL,
1526 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1527 struct lu_device_type *t,
1528 struct lustre_cfg *cfg)
1530 struct lu_device *d;
1531 struct osc_device *od;
1532 struct obd_device *obd;
1537 RETURN(ERR_PTR(-ENOMEM));
1539 cl_device_init(&od->od_cl, t);
1541 d->ld_ops = &mdc_lu_ops;
1544 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1546 RETURN(ERR_PTR(-ENODEV));
1548 rc = mdc_setup(obd, cfg);
1550 osc_device_free(env, d);
1551 RETURN(ERR_PTR(rc));
1553 od->od_exp = obd->obd_self_export;
1557 static const struct lu_device_type_operations mdc_device_type_ops = {
1558 .ldto_device_alloc = mdc_device_alloc,
1559 .ldto_device_free = osc_device_free,
1560 .ldto_device_init = osc_device_init,
1561 .ldto_device_fini = osc_device_fini
1564 struct lu_device_type mdc_device_type = {
1565 .ldt_tags = LU_DEVICE_CL,
1566 .ldt_name = LUSTRE_MDC_NAME,
1567 .ldt_ops = &mdc_device_type_ops,
1568 .ldt_ctx_tags = LCT_CL_THREAD