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 const struct cl_lock *lock,
43 union ldlm_policy_data *policy)
45 memset(policy, 0, sizeof *policy);
46 policy->l_inodebits.bits = MDS_INODELOCK_DOM;
48 policy->l_inodebits.li_gid = lock->cll_descr.cld_gid;
52 int mdc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data)
54 return osc_ldlm_glimpse_ast(dlmlock, data);
57 static void mdc_lock_build_einfo(const struct lu_env *env,
58 const struct cl_lock *lock,
59 struct osc_object *osc,
60 struct ldlm_enqueue_info *einfo)
62 einfo->ei_type = LDLM_IBITS;
63 einfo->ei_mode = osc_cl_lock2ldlm(lock->cll_descr.cld_mode);
64 einfo->ei_cb_bl = mdc_ldlm_blocking_ast;
65 einfo->ei_cb_cp = ldlm_completion_ast;
66 einfo->ei_cb_gl = mdc_ldlm_glimpse_ast;
67 einfo->ei_cbdata = osc; /* value to be put into ->l_ast_data */
70 static void mdc_lock_lvb_update(const struct lu_env *env,
71 struct osc_object *osc,
72 struct ldlm_lock *dlmlock,
75 static int mdc_set_dom_lock_data(struct ldlm_lock *lock, void *data)
79 LASSERT(lock != NULL);
80 LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
82 lock_res_and_lock(lock);
84 if (lock->l_ast_data == NULL)
85 lock->l_ast_data = data;
86 if (lock->l_ast_data == data)
89 unlock_res_and_lock(lock);
94 int mdc_dom_lock_match(const struct lu_env *env, struct obd_export *exp,
95 struct ldlm_res_id *res_id, enum ldlm_type type,
96 union ldlm_policy_data *policy, enum ldlm_mode mode,
97 __u64 *flags, struct osc_object *obj,
98 struct lustre_handle *lockh,
99 enum ldlm_match_flags match_flags)
101 struct obd_device *obd = exp->exp_obd;
102 __u64 lflags = *flags;
107 rc = ldlm_lock_match_with_skip(obd->obd_namespace, lflags, 0,
108 res_id, type, policy, mode, lockh, match_flags);
109 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
113 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
115 LASSERT(lock != NULL);
116 if (mdc_set_dom_lock_data(lock, obj)) {
117 lock_res_and_lock(lock);
118 if (!ldlm_is_lvb_cached(lock)) {
119 LASSERT(lock->l_ast_data == obj);
120 mdc_lock_lvb_update(env, obj, lock, NULL);
121 ldlm_set_lvb_cached(lock);
123 unlock_res_and_lock(lock);
125 ldlm_lock_decref(lockh, rc);
134 * Finds an existing lock covering a page with given index.
135 * Copy of osc_obj_dlmlock_at_pgoff() but for DoM IBITS lock.
137 struct ldlm_lock *mdc_dlmlock_at_pgoff(const struct lu_env *env,
138 struct osc_object *obj, pgoff_t index,
139 enum osc_dap_flags dap_flags)
141 struct osc_thread_info *info = osc_env_info(env);
142 struct ldlm_res_id *resname = &info->oti_resname;
143 union ldlm_policy_data *policy = &info->oti_policy;
144 struct lustre_handle lockh;
145 struct ldlm_lock *lock = NULL;
148 enum ldlm_match_flags match_flags = 0;
152 fid_build_reg_res_name(lu_object_fid(osc2lu(obj)), resname);
153 mdc_lock_build_policy(env, NULL, policy);
154 policy->l_inodebits.li_gid = LDLM_GID_ANY;
156 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
157 if (dap_flags & OSC_DAP_FL_TEST_LOCK)
158 flags |= LDLM_FL_TEST_LOCK;
160 if (dap_flags & OSC_DAP_FL_AST)
161 match_flags |= LDLM_MATCH_AST;
163 if (dap_flags & OSC_DAP_FL_CANCELING)
164 match_flags |= LDLM_MATCH_UNREF;
167 /* Next, search for already existing extent locks that will cover us */
168 /* If we're trying to read, we also search for an existing PW lock. The
169 * VFS and page cache already protect us locally, so lots of readers/
170 * writers can share a single PW lock. */
171 mode = mdc_dom_lock_match(env, osc_export(obj), resname, LDLM_IBITS,
172 policy, LCK_PR | LCK_PW | LCK_GROUP, &flags,
173 obj, &lockh, match_flags);
175 lock = ldlm_handle2lock(&lockh);
176 /* RACE: the lock is cancelled so let's try again */
177 if (unlikely(lock == NULL))
185 * Check if page @page is covered by an extra lock or discard it.
187 static bool mdc_check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
188 void **pvec, int count, void *cbdata)
190 struct osc_thread_info *info = osc_env_info(env);
191 struct osc_object *osc = cbdata;
195 for (i = 0; i < count; i++) {
196 struct osc_page *ops = pvec[i];
198 index = osc_index(ops);
199 if (index >= info->oti_fn_index) {
200 struct ldlm_lock *tmp;
201 struct cl_page *page = ops->ops_cl.cpl_page;
203 /* refresh non-overlapped index */
204 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
205 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_AST);
207 info->oti_fn_index = CL_PAGE_EOF;
209 } else if (cl_page_own(env, io, page) == 0) {
210 /* discard the page */
211 cl_page_discard(env, io, page);
212 cl_page_disown(env, io, page);
214 LASSERT(page->cp_state == CPS_FREEING);
218 info->oti_next_index = index + 1;
224 * Discard pages protected by the given lock. This function traverses radix
225 * tree to find all covering pages and discard them. If a page is being covered
226 * by other locks, it should remain in cache.
228 * If error happens on any step, the process continues anyway (the reasoning
229 * behind this being that lock cancellation cannot be delayed indefinitely).
231 static int mdc_lock_discard_pages(const struct lu_env *env,
232 struct osc_object *osc,
233 pgoff_t start, pgoff_t end,
236 struct osc_thread_info *info = osc_env_info(env);
237 struct cl_io *io = &info->oti_io;
238 osc_page_gang_cbt cb;
243 io->ci_obj = cl_object_top(osc2cl(osc));
244 io->ci_ignore_layout = 1;
245 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
249 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
250 info->oti_fn_index = info->oti_next_index = start;
252 osc_page_gang_lookup(env, io, osc, info->oti_next_index,
253 end, cb, (void *)osc);
259 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
260 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
268 if (mode == CLM_WRITE) {
269 result = osc_cache_writeback_range(env, obj, start, end, 1,
271 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
272 obj, start, end, result,
273 discard ? "discarded" : "written back");
278 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
279 rc = mdc_lock_discard_pages(env, obj, start, end,
280 mode == CLM_WRITE || discard);
281 if (result == 0 && rc < 0)
287 void mdc_lock_lockless_cancel(const struct lu_env *env,
288 const struct cl_lock_slice *slice)
290 struct osc_lock *ols = cl2osc_lock(slice);
291 struct osc_object *osc = cl2osc(slice->cls_obj);
292 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
295 LASSERT(ols->ols_dlmlock == NULL);
296 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
299 CERROR("Pages for lockless lock %p were not purged(%d)\n",
302 osc_lock_wake_waiters(env, osc, ols);
306 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
307 * and ldlm_lock caches.
309 static int mdc_dlm_canceling(const struct lu_env *env,
310 struct ldlm_lock *dlmlock)
312 struct cl_object *obj = NULL;
315 enum cl_lock_mode mode = CLM_READ;
319 lock_res_and_lock(dlmlock);
320 if (!ldlm_is_granted(dlmlock)) {
321 dlmlock->l_ast_data = NULL;
322 unlock_res_and_lock(dlmlock);
326 discard = ldlm_is_discard_data(dlmlock);
327 if (dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))
330 if (dlmlock->l_ast_data != NULL) {
331 obj = osc2cl(dlmlock->l_ast_data);
334 unlock_res_and_lock(dlmlock);
336 /* if l_ast_data is NULL, the dlmlock was enqueued by AGL or
337 * the object has been destroyed. */
339 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
341 /* Destroy pages covered by the extent of the DLM lock */
342 result = mdc_lock_flush(env, cl2osc(obj), cl_index(obj, 0),
343 CL_PAGE_EOF, mode, discard);
344 /* Losing a lock, set KMS to 0.
345 * NB: assumed that DOM lock covers whole data on MDT.
347 /* losing a lock, update kms */
348 lock_res_and_lock(dlmlock);
349 dlmlock->l_ast_data = NULL;
350 cl_object_attr_lock(obj);
352 cl_object_attr_update(env, obj, attr, CAT_KMS);
353 cl_object_attr_unlock(obj);
354 unlock_res_and_lock(dlmlock);
355 cl_object_put(env, obj);
360 int mdc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
361 struct ldlm_lock_desc *new, void *data, int reason)
368 case LDLM_CB_BLOCKING: {
369 struct lustre_handle lockh;
371 ldlm_lock2handle(dlmlock, &lockh);
372 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
377 case LDLM_CB_CANCELING: {
382 * This can be called in the context of outer IO, e.g.,
384 * osc_enqueue_base()->...
385 * ->ldlm_prep_elc_req()->...
386 * ->ldlm_cancel_callback()->...
387 * ->osc_ldlm_blocking_ast()
389 * new environment has to be created to not corrupt outer
392 env = cl_env_get(&refcheck);
398 rc = mdc_dlm_canceling(env, dlmlock);
399 cl_env_put(env, &refcheck);
409 * Updates object attributes from a lock value block (lvb) received together
410 * with the DLM lock reply from the server.
411 * This can be optimized to not update attributes when lock is a result of a
414 * Called under lock and resource spin-locks.
416 void mdc_lock_lvb_update(const struct lu_env *env, struct osc_object *osc,
417 struct ldlm_lock *dlmlock, struct ost_lvb *lvb)
419 struct cl_object *obj = osc2cl(osc);
420 struct lov_oinfo *oinfo = osc->oo_oinfo;
421 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
422 unsigned valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME |
424 unsigned int setkms = 0;
429 LASSERT(dlmlock != NULL);
430 lvb = &dlmlock->l_ost_lvb;
432 cl_lvb2attr(attr, lvb);
434 cl_object_attr_lock(obj);
435 if (dlmlock != NULL) {
438 check_res_locked(dlmlock->l_resource);
439 size = lvb->lvb_size;
441 if (size >= oinfo->loi_kms) {
443 attr->cat_kms = size;
446 ldlm_lock_allow_match_locked(dlmlock);
449 /* The size should not be less than the kms */
450 if (attr->cat_size < oinfo->loi_kms)
451 attr->cat_size = oinfo->loi_kms;
453 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
454 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
455 setkms ? "" : " leaving",
456 setkms ? attr->cat_kms : oinfo->loi_kms,
457 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
459 cl_object_attr_update(env, obj, attr, valid);
460 cl_object_attr_unlock(obj);
464 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
465 struct lustre_handle *lockh)
467 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
468 struct ldlm_lock *dlmlock;
472 dlmlock = ldlm_handle2lock_long(lockh, 0);
473 LASSERT(dlmlock != NULL);
475 /* lock reference taken by ldlm_handle2lock_long() is
476 * owned by osc_lock and released in osc_lock_detach()
478 lu_ref_add(&dlmlock->l_reference, "osc_lock", oscl);
479 oscl->ols_has_ref = 1;
481 LASSERT(oscl->ols_dlmlock == NULL);
482 oscl->ols_dlmlock = dlmlock;
484 /* This may be a matched lock for glimpse request, do not hold
485 * lock reference in that case. */
486 if (!oscl->ols_glimpse) {
487 /* hold a refc for non glimpse lock which will
488 * be released in osc_lock_cancel() */
489 lustre_handle_copy(&oscl->ols_handle, lockh);
490 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
494 /* Lock must have been granted. */
495 lock_res_and_lock(dlmlock);
496 if (ldlm_is_granted(dlmlock)) {
497 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
499 /* extend the lock extent, otherwise it will have problem when
500 * we decide whether to grant a lockless lock. */
501 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
502 descr->cld_start = cl_index(descr->cld_obj, 0);
503 descr->cld_end = CL_PAGE_EOF;
505 /* no lvb update for matched lock */
506 if (!ldlm_is_lvb_cached(dlmlock)) {
507 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
508 LASSERT(osc == dlmlock->l_ast_data);
509 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
510 ldlm_set_lvb_cached(dlmlock);
513 unlock_res_and_lock(dlmlock);
515 LASSERT(oscl->ols_state != OLS_GRANTED);
516 oscl->ols_state = OLS_GRANTED;
521 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
522 * received from a server, or after mdc_enqueue_send() matched a local DLM
525 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
528 struct osc_lock *oscl = cookie;
529 struct cl_lock_slice *slice = &oscl->ols_cl;
535 env = cl_env_percpu_get();
536 /* should never happen, similar to osc_ldlm_blocking_ast(). */
537 LASSERT(!IS_ERR(env));
539 rc = ldlm_error2errno(errcode);
540 if (oscl->ols_state == OLS_ENQUEUED) {
541 oscl->ols_state = OLS_UPCALL_RECEIVED;
542 } else if (oscl->ols_state == OLS_CANCELLED) {
545 CERROR("Impossible state: %d\n", oscl->ols_state);
549 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
551 mdc_lock_granted(env, oscl, lockh);
553 /* Error handling, some errors are tolerable. */
554 if (oscl->ols_glimpse && rc == -ENAVAIL) {
555 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
556 mdc_lock_lvb_update(env, cl2osc(slice->cls_obj),
557 NULL, &oscl->ols_lvb);
558 /* Hide the error. */
562 if (oscl->ols_owner != NULL)
563 cl_sync_io_note(env, oscl->ols_owner, rc);
564 cl_env_percpu_put(env);
569 /* This is needed only for old servers (before 2.14) support */
570 int mdc_fill_lvb(struct req_capsule *pill, struct ost_lvb *lvb)
572 struct mdt_body *body;
574 /* get LVB data from mdt_body otherwise */
575 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
579 if (!(body->mbo_valid & OBD_MD_DOM_SIZE))
582 mdc_body2lvb(body, lvb);
586 int mdc_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
587 osc_enqueue_upcall_f upcall, void *cookie,
588 struct lustre_handle *lockh, enum ldlm_mode mode,
589 __u64 *flags, int errcode)
591 struct osc_lock *ols = cookie;
592 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
597 /* needed only for glimpse from an old server (< 2.14) */
598 if (glimpse && !exp_connect_dom_lvb(exp))
599 rc = mdc_fill_lvb(&req->rq_pill, &ols->ols_lvb);
601 if (glimpse && errcode == ELDLM_LOCK_ABORTED) {
602 struct ldlm_reply *rep;
604 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
606 rep->lock_policy_res2 =
607 ptlrpc_status_ntoh(rep->lock_policy_res2);
608 if (rep->lock_policy_res2)
609 errcode = rep->lock_policy_res2;
613 *flags |= LDLM_FL_LVB_READY;
614 } else if (errcode == ELDLM_OK) {
615 struct ldlm_lock *lock;
617 /* Callers have references, should be valid always */
618 lock = ldlm_handle2lock(lockh);
620 /* At this point ols_lvb must be filled with correct LVB either
621 * by mdc_fill_lvb() above or by ldlm_cli_enqueue_fini().
622 * DoM uses l_ost_lvb to store LVB data, so copy it here from
623 * just updated ols_lvb.
625 lock_res_and_lock(lock);
626 memcpy(&lock->l_ost_lvb, &ols->ols_lvb,
627 sizeof(lock->l_ost_lvb));
628 unlock_res_and_lock(lock);
630 *flags |= LDLM_FL_LVB_READY;
633 /* Call the update callback. */
634 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
636 /* release the reference taken in ldlm_cli_enqueue() */
637 if (errcode == ELDLM_LOCK_MATCHED)
639 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
640 ldlm_lock_decref(lockh, mode);
645 int mdc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
648 struct osc_enqueue_args *aa = args;
649 struct ldlm_lock *lock;
650 struct lustre_handle *lockh = &aa->oa_lockh;
651 enum ldlm_mode mode = aa->oa_mode;
652 struct ldlm_enqueue_info einfo = {
653 .ei_type = aa->oa_type,
659 LASSERT(!aa->oa_speculative);
661 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
663 lock = ldlm_handle2lock(lockh);
664 LASSERTF(lock != NULL,
665 "lockh %#llx, req %p, aa %p - client evicted?\n",
666 lockh->cookie, req, aa);
668 /* Take an additional reference so that a blocking AST that
669 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
670 * to arrive after an upcall has been executed by
671 * mdc_enqueue_fini().
673 ldlm_lock_addref(lockh, mode);
675 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
676 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
678 /* Let CP AST to grant the lock first. */
679 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
681 /* Complete obtaining the lock procedure. */
682 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, &einfo, 1, aa->oa_flags,
683 aa->oa_lvb, aa->oa_lvb ?
684 sizeof(*aa->oa_lvb) : 0, lockh, rc);
685 /* Complete mdc stuff. */
686 rc = mdc_enqueue_fini(aa->oa_exp, req, aa->oa_upcall, aa->oa_cookie,
687 lockh, mode, aa->oa_flags, rc);
689 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
691 ldlm_lock_decref(lockh, mode);
696 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
697 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
698 * other synchronous requests, however keeping some locks and trying to obtain
699 * others may take a considerable amount of time in a case of ost failure; and
700 * when other sync requests do not get released lock from a client, the client
701 * is excluded from the cluster -- such scenarious make the life difficult, so
702 * release locks just after they are obtained. */
703 int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
704 struct ldlm_res_id *res_id, __u64 *flags,
705 union ldlm_policy_data *policy, struct ost_lvb *lvb,
706 osc_enqueue_upcall_f upcall, void *cookie,
707 struct ldlm_enqueue_info *einfo, int async)
709 struct obd_device *obd = exp->exp_obd;
710 struct lustre_handle lockh = { 0 };
711 struct ptlrpc_request *req = NULL;
712 struct ldlm_intent *lit;
714 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
715 __u64 match_flags = *flags;
719 bool compat_glimpse = glimpse && !exp_connect_dom_lvb(exp);
723 mode = einfo->ei_mode;
724 if (einfo->ei_mode == LCK_PR)
727 match_flags |= LDLM_FL_LVB_READY;
729 match_flags |= LDLM_FL_BLOCK_GRANTED;
730 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
731 einfo->ei_type, policy, mode, &lockh);
733 struct ldlm_lock *matched;
735 if (*flags & LDLM_FL_TEST_LOCK)
738 matched = ldlm_handle2lock(&lockh);
740 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
741 ldlm_set_kms_ignore(matched);
743 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
744 *flags |= LDLM_FL_LVB_READY;
746 /* We already have a lock, and it's referenced. */
747 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
749 ldlm_lock_decref(&lockh, mode);
750 LDLM_LOCK_PUT(matched);
753 ldlm_lock_decref(&lockh, mode);
754 LDLM_LOCK_PUT(matched);
757 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
760 /* Glimpse is intent on old server */
761 req = ptlrpc_request_alloc(class_exp2cliimp(exp), compat_glimpse ?
762 &RQF_LDLM_INTENT : &RQF_LDLM_ENQUEUE);
766 /* For WRITE lock cancel other locks on resource early if any */
767 if (einfo->ei_mode & LCK_PW)
768 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
774 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
776 ptlrpc_request_free(req);
780 if (compat_glimpse) {
781 /* pack the glimpse intent */
782 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
783 lit->opc = IT_GLIMPSE;
786 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
787 *flags &= ~LDLM_FL_BLOCK_GRANTED;
789 if (compat_glimpse) {
790 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
791 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
794 lvb_size = sizeof(*lvb);
795 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
798 ptlrpc_request_set_replen(req);
800 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
801 lvb_size, LVB_T_OST, &lockh, async);
804 struct osc_enqueue_args *aa;
806 aa = ptlrpc_req_async_args(aa, req);
808 aa->oa_mode = einfo->ei_mode;
809 aa->oa_type = einfo->ei_type;
810 lustre_handle_copy(&aa->oa_lockh, &lockh);
811 aa->oa_upcall = upcall;
812 aa->oa_cookie = cookie;
813 aa->oa_speculative = false;
814 aa->oa_flags = flags;
815 aa->oa_lvb = compat_glimpse ? NULL : lvb;
817 req->rq_interpret_reply = mdc_enqueue_interpret;
818 ptlrpcd_add_req(req);
820 ptlrpc_req_finished(req);
825 rc = mdc_enqueue_fini(exp, req, upcall, cookie, &lockh, einfo->ei_mode,
827 ptlrpc_req_finished(req);
832 * Implementation of cl_lock_operations::clo_enqueue() method for osc
833 * layer. This initiates ldlm enqueue:
835 * - cancels conflicting locks early (osc_lock_enqueue_wait());
837 * - calls osc_enqueue_base() to do actual enqueue.
839 * osc_enqueue_base() is supplied with an upcall function that is executed
840 * when lock is received either after a local cached ldlm lock is matched, or
841 * when a reply from the server is received.
843 * This function does not wait for the network communication to complete.
845 static int mdc_lock_enqueue(const struct lu_env *env,
846 const struct cl_lock_slice *slice,
847 struct cl_io *unused, struct cl_sync_io *anchor)
849 struct osc_thread_info *info = osc_env_info(env);
850 struct osc_io *oio = osc_env_io(env);
851 struct osc_object *osc = cl2osc(slice->cls_obj);
852 struct osc_lock *oscl = cl2osc_lock(slice);
853 struct cl_lock *lock = slice->cls_lock;
854 struct ldlm_res_id *resname = &info->oti_resname;
855 union ldlm_policy_data *policy = &info->oti_policy;
856 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
857 void *cookie = (void *)oscl;
863 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
864 "lock = %p, ols = %p\n", lock, oscl);
866 if (oscl->ols_state == OLS_GRANTED)
869 /* Lockahead is not supported on MDT yet */
870 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
871 result = -EOPNOTSUPP;
875 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
876 GOTO(enqueue_base, 0);
878 if (oscl->ols_glimpse) {
879 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
881 GOTO(enqueue_base, 0);
884 result = osc_lock_enqueue_wait(env, osc, oscl);
888 /* we can grant lockless lock right after all conflicting locks
890 if (osc_lock_is_lockless(oscl)) {
891 oscl->ols_state = OLS_GRANTED;
892 oio->oi_lockless = 1;
897 oscl->ols_state = OLS_ENQUEUED;
898 if (anchor != NULL) {
899 atomic_inc(&anchor->csi_sync_nr);
900 oscl->ols_owner = anchor;
904 * DLM lock's ast data must be osc_object;
905 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
908 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
909 mdc_lock_build_policy(env, lock, policy);
910 LASSERT(!oscl->ols_speculative);
911 result = mdc_enqueue_send(env, osc_export(osc), resname,
912 &oscl->ols_flags, policy, &oscl->ols_lvb,
913 upcall, cookie, &oscl->ols_einfo, async);
915 if (osc_lock_is_lockless(oscl)) {
916 oio->oi_lockless = 1;
918 LASSERT(oscl->ols_state == OLS_GRANTED);
919 LASSERT(oscl->ols_hold);
920 LASSERT(oscl->ols_dlmlock != NULL);
925 oscl->ols_state = OLS_CANCELLED;
926 osc_lock_wake_waiters(env, osc, oscl);
929 cl_sync_io_note(env, anchor, result);
934 static const struct cl_lock_operations mdc_lock_lockless_ops = {
935 .clo_fini = osc_lock_fini,
936 .clo_enqueue = mdc_lock_enqueue,
937 .clo_cancel = mdc_lock_lockless_cancel,
938 .clo_print = osc_lock_print
941 static const struct cl_lock_operations mdc_lock_ops = {
942 .clo_fini = osc_lock_fini,
943 .clo_enqueue = mdc_lock_enqueue,
944 .clo_cancel = osc_lock_cancel,
945 .clo_print = osc_lock_print,
948 int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
949 struct cl_lock *lock, const struct cl_io *io)
951 struct osc_lock *ols;
952 __u32 enqflags = lock->cll_descr.cld_enq_flags;
953 __u64 flags = osc_enq2ldlm_flags(enqflags);
957 /* Ignore AGL for Data-on-MDT, stat returns size data */
958 if ((enqflags & CEF_SPECULATIVE) != 0)
961 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
962 if (unlikely(ols == NULL))
965 ols->ols_state = OLS_NEW;
966 spin_lock_init(&ols->ols_lock);
967 INIT_LIST_HEAD(&ols->ols_waiting_list);
968 INIT_LIST_HEAD(&ols->ols_wait_entry);
969 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
970 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
972 ols->ols_flags = flags;
973 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
974 if (lock->cll_descr.cld_mode == CLM_GROUP)
975 ols->ols_flags |= LDLM_FL_ATOMIC_CB;
977 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
978 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
979 ols->ols_glimpse = 1;
981 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
983 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
985 if (!(enqflags & CEF_MUST))
986 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
988 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
989 osc_lock_set_writer(env, io, obj, ols);
991 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx\n",
992 lock, ols, ols->ols_flags);
999 * An implementation of cl_io_operations specific methods for MDC layer.
1002 static int mdc_async_upcall(void *a, int rc)
1004 struct osc_async_cbargs *args = a;
1007 complete(&args->opc_sync);
1011 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
1012 pgoff_t index, struct lustre_handle *lh)
1014 struct ldlm_lock *lock;
1016 /* find DOM lock protecting object */
1017 lock = mdc_dlmlock_at_pgoff(env, osc, index,
1018 OSC_DAP_FL_TEST_LOCK |
1019 OSC_DAP_FL_CANCELING);
1021 struct ldlm_resource *res;
1022 struct ldlm_res_id *resname;
1024 resname = &osc_env_info(env)->oti_resname;
1025 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1026 res = ldlm_resource_get(osc_export(osc)->exp_obd->obd_namespace,
1027 NULL, resname, LDLM_IBITS, 0);
1028 ldlm_resource_dump(D_ERROR, res);
1029 libcfs_debug_dumpstack(NULL);
1032 *lh = lock->l_remote_handle;
1033 LDLM_LOCK_PUT(lock);
1038 static int mdc_io_setattr_start(const struct lu_env *env,
1039 const struct cl_io_slice *slice)
1041 struct cl_io *io = slice->cis_io;
1042 struct osc_io *oio = cl2osc_io(env, slice);
1043 struct cl_object *obj = slice->cis_obj;
1044 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1045 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1046 struct obdo *oa = &oio->oi_oa;
1047 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1048 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1049 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1050 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1053 /* silently ignore non-truncate setattr for Data-on-MDT object */
1054 if (cl_io_is_trunc(io)) {
1055 /* truncate cache dirty pages first */
1056 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1062 if (cl_io_is_fallocate(io))
1065 if (oio->oi_lockless == 0) {
1066 cl_object_attr_lock(obj);
1067 rc = cl_object_attr_get(env, obj, attr);
1069 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1070 unsigned int cl_valid = 0;
1072 if (ia_avalid & ATTR_SIZE) {
1073 attr->cat_size = size;
1074 attr->cat_kms = size;
1075 cl_valid = (CAT_SIZE | CAT_KMS);
1077 if (ia_avalid & ATTR_MTIME_SET) {
1078 attr->cat_mtime = lvb->lvb_mtime;
1079 cl_valid |= CAT_MTIME;
1081 if (ia_avalid & ATTR_ATIME_SET) {
1082 attr->cat_atime = lvb->lvb_atime;
1083 cl_valid |= CAT_ATIME;
1085 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1086 attr->cat_ctime = lvb->lvb_ctime;
1087 cl_valid |= CAT_CTIME;
1089 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1091 cl_object_attr_unlock(obj);
1096 if (!(ia_avalid & ATTR_SIZE))
1099 memset(oa, 0, sizeof(*oa));
1100 oa->o_oi = loi->loi_oi;
1101 oa->o_mtime = attr->cat_mtime;
1102 oa->o_atime = attr->cat_atime;
1103 oa->o_ctime = attr->cat_ctime;
1106 oa->o_blocks = OBD_OBJECT_EOF;
1107 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1108 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1110 if (oio->oi_lockless) {
1111 oa->o_flags = OBD_FL_SRVLOCK;
1112 oa->o_valid |= OBD_MD_FLFLAGS;
1114 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1117 oa->o_valid |= OBD_MD_FLHANDLE;
1120 init_completion(&cbargs->opc_sync);
1122 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1123 mdc_async_upcall, cbargs);
1124 cbargs->opc_rpc_sent = rc == 0;
1128 static int mdc_io_read_ahead(const struct lu_env *env,
1129 const struct cl_io_slice *ios,
1130 pgoff_t start, struct cl_read_ahead *ra)
1132 struct osc_object *osc = cl2osc(ios->cis_obj);
1133 struct osc_io *oio = cl2osc_io(env, ios);
1134 struct ldlm_lock *dlmlock;
1138 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1139 if (dlmlock == NULL)
1142 oio->oi_is_readahead = 1;
1143 if (dlmlock->l_req_mode != LCK_PR) {
1144 struct lustre_handle lockh;
1146 ldlm_lock2handle(dlmlock, &lockh);
1147 ldlm_lock_addref(&lockh, LCK_PR);
1148 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1151 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1152 ra->cra_end_idx = CL_PAGE_EOF;
1153 ra->cra_release = osc_read_ahead_release;
1154 ra->cra_dlmlock = dlmlock;
1160 int mdc_io_fsync_start(const struct lu_env *env,
1161 const struct cl_io_slice *slice)
1163 struct cl_io *io = slice->cis_io;
1164 struct cl_fsync_io *fio = &io->u.ci_fsync;
1165 struct cl_object *obj = slice->cis_obj;
1166 struct osc_object *osc = cl2osc(obj);
1171 /* a MDC lock always covers whole object, do sync for whole
1172 * possible range despite of supplied start/end values.
1174 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1175 fio->fi_mode == CL_FSYNC_DISCARD);
1177 fio->fi_nr_written += result;
1180 if (fio->fi_mode == CL_FSYNC_ALL) {
1183 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1186 /* Use OSC sync code because it is asynchronous.
1187 * It is to be added into MDC and avoid the using of
1188 * OST_SYNC at both MDC and MDT.
1190 rc = osc_fsync_ost(env, osc, fio);
1198 struct mdc_data_version_args {
1199 struct osc_io *dva_oio;
1203 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1206 struct mdc_data_version_args *dva = args;
1207 struct osc_io *oio = dva->dva_oio;
1208 const struct mdt_body *body;
1214 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1216 GOTO(out, rc = -EPROTO);
1218 /* Prepare OBDO from mdt_body for CLIO */
1219 oio->oi_oa.o_valid = body->mbo_valid;
1220 oio->oi_oa.o_flags = body->mbo_flags;
1221 oio->oi_oa.o_data_version = body->mbo_version;
1222 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1225 oio->oi_cbarg.opc_rc = rc;
1226 complete(&oio->oi_cbarg.opc_sync);
1230 static int mdc_io_data_version_start(const struct lu_env *env,
1231 const struct cl_io_slice *slice)
1233 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1234 struct osc_io *oio = cl2osc_io(env, slice);
1235 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1236 struct osc_object *obj = cl2osc(slice->cis_obj);
1237 struct obd_export *exp = osc_export(obj);
1238 struct ptlrpc_request *req;
1239 struct mdt_body *body;
1240 struct mdc_data_version_args *dva;
1245 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1246 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1247 oio->oi_oa.o_valid = OBD_MD_FLID;
1249 init_completion(&cbargs->opc_sync);
1251 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1255 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1257 ptlrpc_request_free(req);
1261 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1262 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1263 body->mbo_valid = OBD_MD_FLID;
1264 /* Indicate that data version is needed */
1265 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1266 body->mbo_flags = 0;
1268 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1269 body->mbo_valid |= OBD_MD_FLFLAGS;
1270 body->mbo_flags |= OBD_FL_SRVLOCK;
1271 if (dv->dv_flags & LL_DV_WR_FLUSH)
1272 body->mbo_flags |= OBD_FL_FLUSH;
1275 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1276 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1277 ptlrpc_request_set_replen(req);
1279 req->rq_interpret_reply = mdc_data_version_interpret;
1280 dva = ptlrpc_req_async_args(dva, req);
1283 ptlrpcd_add_req(req);
1288 static void mdc_io_data_version_end(const struct lu_env *env,
1289 const struct cl_io_slice *slice)
1291 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1292 struct osc_io *oio = cl2osc_io(env, slice);
1293 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1296 wait_for_completion(&cbargs->opc_sync);
1298 if (cbargs->opc_rc != 0) {
1299 slice->cis_io->ci_result = cbargs->opc_rc;
1301 slice->cis_io->ci_result = 0;
1302 if (!(oio->oi_oa.o_valid &
1303 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1304 slice->cis_io->ci_result = -ENOTSUPP;
1306 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1307 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1308 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1309 dv->dv_data_version = oio->oi_oa.o_data_version;
1315 static const struct cl_io_operations mdc_io_ops = {
1318 .cio_iter_init = osc_io_iter_init,
1319 .cio_iter_fini = osc_io_rw_iter_fini,
1320 .cio_start = osc_io_read_start,
1323 .cio_iter_init = osc_io_iter_init,
1324 .cio_iter_fini = osc_io_rw_iter_fini,
1325 .cio_start = osc_io_write_start,
1326 .cio_end = osc_io_end,
1329 .cio_iter_init = osc_io_iter_init,
1330 .cio_iter_fini = osc_io_iter_fini,
1331 .cio_start = mdc_io_setattr_start,
1332 .cio_end = osc_io_setattr_end,
1334 [CIT_DATA_VERSION] = {
1335 .cio_start = mdc_io_data_version_start,
1336 .cio_end = mdc_io_data_version_end,
1339 .cio_iter_init = osc_io_iter_init,
1340 .cio_iter_fini = osc_io_iter_fini,
1341 .cio_start = osc_io_fault_start,
1342 .cio_end = osc_io_end,
1345 .cio_start = mdc_io_fsync_start,
1346 .cio_end = osc_io_fsync_end,
1349 .cio_start = osc_io_lseek_start,
1350 .cio_end = osc_io_lseek_end,
1353 .cio_read_ahead = mdc_io_read_ahead,
1354 .cio_lru_reserve = osc_io_lru_reserve,
1355 .cio_submit = osc_io_submit,
1356 .cio_commit_async = osc_io_commit_async,
1357 .cio_extent_release = osc_io_extent_release,
1360 int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1363 struct osc_io *oio = osc_env_io(env);
1365 CL_IO_SLICE_CLEAN(oio, oi_cl);
1366 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1370 static void mdc_build_res_name(struct osc_object *osc,
1371 struct ldlm_res_id *resname)
1373 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1377 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1378 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1381 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1382 struct cl_req_attr *attr)
1384 u64 flags = attr->cra_flags;
1386 /* Copy object FID to cl_attr */
1387 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1389 if (flags & OBD_MD_FLGROUP)
1390 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1392 if (flags & OBD_MD_FLID)
1393 attr->cra_oa->o_valid |= OBD_MD_FLID;
1395 if (flags & OBD_MD_FLHANDLE) {
1396 struct osc_page *opg;
1398 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1399 if (!opg->ops_srvlock) {
1402 rc = mdc_get_lock_handle(env, cl2osc(obj),
1404 &attr->cra_oa->o_handle);
1406 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1407 "uncovered page!\n");
1410 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1416 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1417 struct cl_attr *attr)
1419 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1421 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1422 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1424 return osc_attr_get(env, obj, attr);
1427 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1429 struct osc_object *osc = (struct osc_object *)data;
1430 struct ost_lvb *lvb = &lock->l_ost_lvb;
1431 struct lov_oinfo *oinfo;
1434 if (lock->l_ast_data == data) {
1435 lock->l_ast_data = NULL;
1437 LASSERT(osc != NULL);
1438 LASSERT(osc->oo_oinfo != NULL);
1439 LASSERT(lvb != NULL);
1441 /* Updates lvb in lock by the cached oinfo */
1442 oinfo = osc->oo_oinfo;
1444 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1445 "%llu %llu %llu by oinfo size %llu blocks %llu "
1446 "[cma]time %llu %llu %llu", lvb->lvb_size,
1447 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1448 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1449 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1450 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1451 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1453 cl_object_attr_lock(&osc->oo_cl);
1454 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1455 cl_object_attr_unlock(&osc->oo_cl);
1456 ldlm_clear_lvb_cached(lock);
1458 RETURN(LDLM_ITER_CONTINUE);
1461 int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1463 struct osc_object *osc = cl2osc(obj);
1464 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1466 /* DLM locks don't hold a reference of osc_object so we have to
1467 * clear it before the object is being destroyed. */
1468 osc_build_res_name(osc, resname);
1469 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1470 mdc_object_ast_clear, osc);
1474 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1475 struct ldlm_lock *lock)
1477 /* if lock cancel is initiated from llite then it is combined
1478 * lock with DOM bit and it may have no l_ast_data initialized yet,
1479 * so init it here with given osc_object.
1481 mdc_set_dom_lock_data(lock, cl2osc(obj));
1482 RETURN(mdc_dlm_canceling(env, lock));
1485 static const struct cl_object_operations mdc_ops = {
1486 .coo_page_init = osc_page_init,
1487 .coo_lock_init = mdc_lock_init,
1488 .coo_io_init = mdc_io_init,
1489 .coo_attr_get = mdc_attr_get,
1490 .coo_attr_update = osc_attr_update,
1491 .coo_glimpse = osc_object_glimpse,
1492 .coo_req_attr_set = mdc_req_attr_set,
1493 .coo_prune = mdc_object_prune,
1494 .coo_object_flush = mdc_object_flush
1497 static const struct osc_object_operations mdc_object_ops = {
1498 .oto_build_res_name = mdc_build_res_name,
1499 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1502 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1503 const struct lu_object_conf *conf)
1505 struct osc_object *osc = lu2osc(obj);
1507 if (osc->oo_initialized)
1510 osc->oo_initialized = true;
1512 return osc_object_init(env, obj, conf);
1515 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1517 osc_object_free(env, obj);
1520 static const struct lu_object_operations mdc_lu_obj_ops = {
1521 .loo_object_init = mdc_object_init,
1522 .loo_object_delete = NULL,
1523 .loo_object_release = NULL,
1524 .loo_object_free = mdc_object_free,
1525 .loo_object_print = osc_object_print,
1526 .loo_object_invariant = NULL
1529 struct lu_object *mdc_object_alloc(const struct lu_env *env,
1530 const struct lu_object_header *unused,
1531 struct lu_device *dev)
1533 struct osc_object *osc;
1534 struct lu_object *obj;
1536 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1539 lu_object_init(obj, NULL, dev);
1540 osc->oo_cl.co_ops = &mdc_ops;
1541 obj->lo_ops = &mdc_lu_obj_ops;
1542 osc->oo_obj_ops = &mdc_object_ops;
1543 osc->oo_initialized = false;
1550 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1551 struct lustre_cfg *cfg)
1553 size_t count = class_modify_config(cfg, PARAM_MDC,
1554 &d->ld_obd->obd_kset.kobj);
1555 return count > 0 ? 0 : count;
1558 const struct lu_device_operations mdc_lu_ops = {
1559 .ldo_object_alloc = mdc_object_alloc,
1560 .ldo_process_config = mdc_process_config,
1561 .ldo_recovery_complete = NULL,
1564 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1565 struct lu_device_type *t,
1566 struct lustre_cfg *cfg)
1568 struct lu_device *d;
1569 struct osc_device *od;
1570 struct obd_device *obd;
1575 RETURN(ERR_PTR(-ENOMEM));
1577 cl_device_init(&od->od_cl, t);
1579 d->ld_ops = &mdc_lu_ops;
1582 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1584 RETURN(ERR_PTR(-ENODEV));
1586 rc = mdc_setup(obd, cfg);
1588 osc_device_free(env, d);
1589 RETURN(ERR_PTR(rc));
1591 od->od_exp = obd->obd_self_export;
1595 static const struct lu_device_type_operations mdc_device_type_ops = {
1596 .ldto_device_alloc = mdc_device_alloc,
1597 .ldto_device_free = osc_device_free,
1598 .ldto_device_init = osc_device_init,
1599 .ldto_device_fini = osc_device_fini
1602 struct lu_device_type mdc_device_type = {
1603 .ldt_tags = LU_DEVICE_CL,
1604 .ldt_name = LUSTRE_MDC_NAME,
1605 .ldt_ops = &mdc_device_type_ops,
1606 .ldt_ctx_tags = LCT_CL_THREAD