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);
110 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
114 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
116 LASSERT(lock != NULL);
117 if (mdc_set_dom_lock_data(lock, obj)) {
118 lock_res_and_lock(lock);
119 if (!ldlm_is_lvb_cached(lock)) {
120 LASSERT(lock->l_ast_data == obj);
121 mdc_lock_lvb_update(env, obj, lock, NULL);
122 ldlm_set_lvb_cached(lock);
124 unlock_res_and_lock(lock);
126 ldlm_lock_decref(lockh, rc);
135 * Finds an existing lock covering a page with given index.
136 * Copy of osc_obj_dlmlock_at_pgoff() but for DoM IBITS lock.
138 struct ldlm_lock *mdc_dlmlock_at_pgoff(const struct lu_env *env,
139 struct osc_object *obj, pgoff_t index,
140 enum osc_dap_flags dap_flags)
142 struct osc_thread_info *info = osc_env_info(env);
143 struct ldlm_res_id *resname = &info->oti_resname;
144 union ldlm_policy_data *policy = &info->oti_policy;
145 struct lustre_handle lockh;
146 struct ldlm_lock *lock = NULL;
149 enum ldlm_match_flags match_flags = 0;
153 fid_build_reg_res_name(lu_object_fid(osc2lu(obj)), resname);
154 mdc_lock_build_policy(env, NULL, policy);
155 policy->l_inodebits.li_gid = LDLM_GID_ANY;
157 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
158 if (dap_flags & OSC_DAP_FL_TEST_LOCK)
159 flags |= LDLM_FL_TEST_LOCK;
161 if (dap_flags & OSC_DAP_FL_AST)
162 match_flags |= LDLM_MATCH_AST;
164 if (dap_flags & OSC_DAP_FL_CANCELING)
165 match_flags |= LDLM_MATCH_UNREF;
168 /* Next, search for already existing extent locks that will cover us */
169 /* If we're trying to read, we also search for an existing PW lock. The
170 * VFS and page cache already protect us locally, so lots of readers/
171 * writers can share a single PW lock. */
172 mode = mdc_dom_lock_match(env, osc_export(obj), resname, LDLM_IBITS,
173 policy, LCK_PR | LCK_PW | LCK_GROUP, &flags,
174 obj, &lockh, match_flags);
176 lock = ldlm_handle2lock(&lockh);
177 /* RACE: the lock is cancelled so let's try again */
178 if (unlikely(lock == NULL))
186 * Check if page @page is covered by an extra lock or discard it.
188 static bool mdc_check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
189 struct osc_page *ops, void *cbdata)
191 struct osc_thread_info *info = osc_env_info(env);
192 struct osc_object *osc = cbdata;
195 index = osc_index(ops);
196 if (index >= info->oti_fn_index) {
197 struct ldlm_lock *tmp;
198 struct cl_page *page = ops->ops_cl.cpl_page;
200 /* refresh non-overlapped index */
201 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
202 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_AST);
204 info->oti_fn_index = CL_PAGE_EOF;
206 } else if (cl_page_own(env, io, page) == 0) {
207 /* discard the page */
208 cl_page_discard(env, io, page);
209 cl_page_disown(env, io, page);
211 LASSERT(page->cp_state == CPS_FREEING);
215 info->oti_next_index = index + 1;
220 * Discard pages protected by the given lock. This function traverses radix
221 * tree to find all covering pages and discard them. If a page is being covered
222 * by other locks, it should remain in cache.
224 * If error happens on any step, the process continues anyway (the reasoning
225 * behind this being that lock cancellation cannot be delayed indefinitely).
227 static int mdc_lock_discard_pages(const struct lu_env *env,
228 struct osc_object *osc,
229 pgoff_t start, pgoff_t end,
232 struct osc_thread_info *info = osc_env_info(env);
233 struct cl_io *io = &info->oti_io;
234 osc_page_gang_cbt cb;
239 io->ci_obj = cl_object_top(osc2cl(osc));
240 io->ci_ignore_layout = 1;
241 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
245 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
246 info->oti_fn_index = info->oti_next_index = start;
248 osc_page_gang_lookup(env, io, osc, info->oti_next_index,
249 end, cb, (void *)osc);
255 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
256 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
264 if (mode == CLM_WRITE) {
265 result = osc_cache_writeback_range(env, obj, start, end, 1,
267 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
268 obj, start, end, result,
269 discard ? "discarded" : "written back");
274 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
275 rc = mdc_lock_discard_pages(env, obj, start, end,
276 mode == CLM_WRITE || discard);
277 if (result == 0 && rc < 0)
283 void mdc_lock_lockless_cancel(const struct lu_env *env,
284 const struct cl_lock_slice *slice)
286 struct osc_lock *ols = cl2osc_lock(slice);
287 struct osc_object *osc = cl2osc(slice->cls_obj);
288 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
291 LASSERT(ols->ols_dlmlock == NULL);
292 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
295 CERROR("Pages for lockless lock %p were not purged(%d)\n",
298 osc_lock_wake_waiters(env, osc, ols);
302 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
303 * and ldlm_lock caches.
305 static int mdc_dlm_blocking_ast0(const struct lu_env *env,
306 struct ldlm_lock *dlmlock,
309 struct cl_object *obj = NULL;
312 enum cl_lock_mode mode = CLM_READ;
316 LASSERT(flag == LDLM_CB_CANCELING);
317 LASSERT(dlmlock != NULL);
319 lock_res_and_lock(dlmlock);
320 if (dlmlock->l_granted_mode != dlmlock->l_req_mode) {
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 flag)
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_blocking_ast0(env, dlmlock, flag);
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;
448 /* The size should not be less than the kms */
449 if (attr->cat_size < oinfo->loi_kms)
450 attr->cat_size = oinfo->loi_kms;
452 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
453 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
454 setkms ? "" : " leaving",
455 setkms ? attr->cat_kms : oinfo->loi_kms,
456 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
458 cl_object_attr_update(env, obj, attr, valid);
459 cl_object_attr_unlock(obj);
463 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
464 struct lustre_handle *lockh)
466 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
467 struct ldlm_lock *dlmlock;
471 dlmlock = ldlm_handle2lock_long(lockh, 0);
472 LASSERT(dlmlock != NULL);
474 /* lock reference taken by ldlm_handle2lock_long() is
475 * owned by osc_lock and released in osc_lock_detach()
477 lu_ref_add(&dlmlock->l_reference, "osc_lock", oscl);
478 oscl->ols_has_ref = 1;
480 LASSERT(oscl->ols_dlmlock == NULL);
481 oscl->ols_dlmlock = dlmlock;
483 /* This may be a matched lock for glimpse request, do not hold
484 * lock reference in that case. */
485 if (!oscl->ols_glimpse) {
486 /* hold a refc for non glimpse lock which will
487 * be released in osc_lock_cancel() */
488 lustre_handle_copy(&oscl->ols_handle, lockh);
489 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
493 /* Lock must have been granted. */
494 lock_res_and_lock(dlmlock);
495 if (dlmlock->l_granted_mode == dlmlock->l_req_mode) {
496 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
498 /* extend the lock extent, otherwise it will have problem when
499 * we decide whether to grant a lockless lock. */
500 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
501 descr->cld_start = cl_index(descr->cld_obj, 0);
502 descr->cld_end = CL_PAGE_EOF;
504 /* no lvb update for matched lock */
505 if (!ldlm_is_lvb_cached(dlmlock)) {
506 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
507 LASSERT(osc == dlmlock->l_ast_data);
508 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
509 ldlm_set_lvb_cached(dlmlock);
512 unlock_res_and_lock(dlmlock);
514 LASSERT(oscl->ols_state != OLS_GRANTED);
515 oscl->ols_state = OLS_GRANTED;
520 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
521 * received from a server, or after osc_enqueue_base() matched a local DLM
524 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
527 struct osc_lock *oscl = cookie;
528 struct cl_lock_slice *slice = &oscl->ols_cl;
534 env = cl_env_percpu_get();
535 /* should never happen, similar to osc_ldlm_blocking_ast(). */
536 LASSERT(!IS_ERR(env));
538 rc = ldlm_error2errno(errcode);
539 if (oscl->ols_state == OLS_ENQUEUED) {
540 oscl->ols_state = OLS_UPCALL_RECEIVED;
541 } else if (oscl->ols_state == OLS_CANCELLED) {
544 CERROR("Impossible state: %d\n", oscl->ols_state);
548 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
550 mdc_lock_granted(env, oscl, lockh);
552 /* Error handling, some errors are tolerable. */
553 if (oscl->ols_locklessable && rc == -EUSERS) {
554 /* This is a tolerable error, turn this lock into
557 osc_object_set_contended(cl2osc(slice->cls_obj));
558 LASSERT(slice->cls_ops != oscl->ols_lockless_ops);
560 /* Change this lock to ldlmlock-less lock. */
561 osc_lock_to_lockless(env, oscl, 1);
562 oscl->ols_state = OLS_GRANTED;
564 } else if (oscl->ols_glimpse && rc == -ENAVAIL) {
565 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
566 mdc_lock_lvb_update(env, cl2osc(slice->cls_obj),
567 NULL, &oscl->ols_lvb);
568 /* Hide the error. */
572 if (oscl->ols_owner != NULL)
573 cl_sync_io_note(env, oscl->ols_owner, rc);
574 cl_env_percpu_put(env);
579 int mdc_fill_lvb(struct ptlrpc_request *req, struct ost_lvb *lvb)
581 struct mdt_body *body;
583 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
587 lvb->lvb_mtime = body->mbo_mtime;
588 lvb->lvb_atime = body->mbo_atime;
589 lvb->lvb_ctime = body->mbo_ctime;
590 lvb->lvb_blocks = body->mbo_dom_blocks;
591 lvb->lvb_size = body->mbo_dom_size;
596 int mdc_enqueue_fini(struct ptlrpc_request *req, osc_enqueue_upcall_f upcall,
597 void *cookie, struct lustre_handle *lockh,
598 enum ldlm_mode mode, __u64 *flags, int errcode)
600 struct osc_lock *ols = cookie;
601 struct ldlm_lock *lock;
606 /* The request was created before ldlm_cli_enqueue call. */
607 if (errcode == ELDLM_LOCK_ABORTED) {
608 struct ldlm_reply *rep;
610 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
611 LASSERT(rep != NULL);
613 rep->lock_policy_res2 =
614 ptlrpc_status_ntoh(rep->lock_policy_res2);
615 if (rep->lock_policy_res2)
616 errcode = rep->lock_policy_res2;
618 rc = mdc_fill_lvb(req, &ols->ols_lvb);
619 *flags |= LDLM_FL_LVB_READY;
620 } else if (errcode == ELDLM_OK) {
621 /* Callers have references, should be valid always */
622 lock = ldlm_handle2lock(lockh);
625 rc = mdc_fill_lvb(req, &lock->l_ost_lvb);
627 *flags |= LDLM_FL_LVB_READY;
630 /* Call the update callback. */
631 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
633 /* release the reference taken in ldlm_cli_enqueue() */
634 if (errcode == ELDLM_LOCK_MATCHED)
636 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
637 ldlm_lock_decref(lockh, mode);
642 int mdc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
645 struct osc_enqueue_args *aa = args;
646 struct ldlm_lock *lock;
647 struct lustre_handle *lockh = &aa->oa_lockh;
648 enum ldlm_mode mode = aa->oa_mode;
652 LASSERT(!aa->oa_speculative);
654 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
656 lock = ldlm_handle2lock(lockh);
657 LASSERTF(lock != NULL,
658 "lockh %#llx, req %p, aa %p - client evicted?\n",
659 lockh->cookie, req, aa);
661 /* Take an additional reference so that a blocking AST that
662 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
663 * to arrive after an upcall has been executed by
664 * osc_enqueue_fini(). */
665 ldlm_lock_addref(lockh, mode);
667 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
668 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
670 /* Let CP AST to grant the lock first. */
671 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
673 /* Complete obtaining the lock procedure. */
674 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_type, 1,
675 aa->oa_mode, aa->oa_flags, NULL, 0,
677 /* Complete mdc stuff. */
678 rc = mdc_enqueue_fini(req, aa->oa_upcall, aa->oa_cookie, lockh, mode,
681 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
683 ldlm_lock_decref(lockh, mode);
688 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
689 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
690 * other synchronous requests, however keeping some locks and trying to obtain
691 * others may take a considerable amount of time in a case of ost failure; and
692 * when other sync requests do not get released lock from a client, the client
693 * is excluded from the cluster -- such scenarious make the life difficult, so
694 * release locks just after they are obtained. */
695 int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
696 struct ldlm_res_id *res_id, __u64 *flags,
697 union ldlm_policy_data *policy,
698 struct ost_lvb *lvb, int kms_valid,
699 osc_enqueue_upcall_f upcall, void *cookie,
700 struct ldlm_enqueue_info *einfo, int async)
702 struct obd_device *obd = exp->exp_obd;
703 struct lustre_handle lockh = { 0 };
704 struct ptlrpc_request *req = NULL;
705 struct ldlm_intent *lit;
707 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
708 __u64 match_flags = *flags;
714 mode = einfo->ei_mode;
715 if (einfo->ei_mode == LCK_PR)
719 match_flags |= LDLM_FL_BLOCK_GRANTED;
720 /* DOM locking uses LDLM_FL_KMS_IGNORE to mark locks wich have no valid
721 * LVB information, e.g. canceled locks or locks of just pruned object,
722 * such locks should be skipped.
724 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
725 einfo->ei_type, policy, mode, &lockh);
727 struct ldlm_lock *matched;
729 if (*flags & LDLM_FL_TEST_LOCK)
732 matched = ldlm_handle2lock(&lockh);
734 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
735 ldlm_set_kms_ignore(matched);
737 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
738 *flags |= LDLM_FL_LVB_READY;
740 /* We already have a lock, and it's referenced. */
741 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
743 ldlm_lock_decref(&lockh, mode);
744 LDLM_LOCK_PUT(matched);
747 ldlm_lock_decref(&lockh, mode);
748 LDLM_LOCK_PUT(matched);
751 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
754 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_INTENT);
758 /* For WRITE lock cancel other locks on resource early if any */
759 if (einfo->ei_mode & LCK_PW)
760 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
766 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
768 ptlrpc_request_free(req);
772 /* pack the intent */
773 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
774 lit->opc = glimpse ? IT_GLIMPSE : IT_BRW;
776 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
777 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
778 ptlrpc_request_set_replen(req);
780 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
781 *flags &= ~LDLM_FL_BLOCK_GRANTED;
782 /* All MDC IO locks are intents */
783 *flags |= LDLM_FL_HAS_INTENT;
784 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, NULL,
785 0, LVB_T_NONE, &lockh, async);
788 struct osc_enqueue_args *aa;
790 aa = ptlrpc_req_async_args(aa, req);
792 aa->oa_mode = einfo->ei_mode;
793 aa->oa_type = einfo->ei_type;
794 lustre_handle_copy(&aa->oa_lockh, &lockh);
795 aa->oa_upcall = upcall;
796 aa->oa_cookie = cookie;
797 aa->oa_speculative = false;
798 aa->oa_flags = flags;
801 req->rq_interpret_reply = mdc_enqueue_interpret;
802 ptlrpcd_add_req(req);
804 ptlrpc_req_finished(req);
809 rc = mdc_enqueue_fini(req, upcall, cookie, &lockh, einfo->ei_mode,
811 ptlrpc_req_finished(req);
816 * Implementation of cl_lock_operations::clo_enqueue() method for osc
817 * layer. This initiates ldlm enqueue:
819 * - cancels conflicting locks early (osc_lock_enqueue_wait());
821 * - calls osc_enqueue_base() to do actual enqueue.
823 * osc_enqueue_base() is supplied with an upcall function that is executed
824 * when lock is received either after a local cached ldlm lock is matched, or
825 * when a reply from the server is received.
827 * This function does not wait for the network communication to complete.
829 static int mdc_lock_enqueue(const struct lu_env *env,
830 const struct cl_lock_slice *slice,
831 struct cl_io *unused, struct cl_sync_io *anchor)
833 struct osc_thread_info *info = osc_env_info(env);
834 struct osc_io *oio = osc_env_io(env);
835 struct osc_object *osc = cl2osc(slice->cls_obj);
836 struct osc_lock *oscl = cl2osc_lock(slice);
837 struct cl_lock *lock = slice->cls_lock;
838 struct ldlm_res_id *resname = &info->oti_resname;
839 union ldlm_policy_data *policy = &info->oti_policy;
840 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
841 void *cookie = (void *)oscl;
847 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
848 "lock = %p, ols = %p\n", lock, oscl);
850 if (oscl->ols_state == OLS_GRANTED)
853 /* Lockahead is not supported on MDT yet */
854 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
855 result = -EOPNOTSUPP;
859 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
860 GOTO(enqueue_base, 0);
862 if (oscl->ols_glimpse) {
863 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
865 GOTO(enqueue_base, 0);
868 result = osc_lock_enqueue_wait(env, osc, oscl);
872 /* we can grant lockless lock right after all conflicting locks
874 if (osc_lock_is_lockless(oscl)) {
875 oscl->ols_state = OLS_GRANTED;
876 oio->oi_lockless = 1;
881 oscl->ols_state = OLS_ENQUEUED;
882 if (anchor != NULL) {
883 atomic_inc(&anchor->csi_sync_nr);
884 oscl->ols_owner = anchor;
888 * DLM lock's ast data must be osc_object;
889 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
892 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
893 mdc_lock_build_policy(env, lock, policy);
894 LASSERT(!oscl->ols_speculative);
895 result = mdc_enqueue_send(env, osc_export(osc), resname,
896 &oscl->ols_flags, policy,
897 &oscl->ols_lvb, osc->oo_oinfo->loi_kms_valid,
898 upcall, cookie, &oscl->ols_einfo, async);
900 if (osc_lock_is_lockless(oscl)) {
901 oio->oi_lockless = 1;
903 LASSERT(oscl->ols_state == OLS_GRANTED);
904 LASSERT(oscl->ols_hold);
905 LASSERT(oscl->ols_dlmlock != NULL);
910 oscl->ols_state = OLS_CANCELLED;
911 osc_lock_wake_waiters(env, osc, oscl);
914 cl_sync_io_note(env, anchor, result);
919 static const struct cl_lock_operations mdc_lock_lockless_ops = {
920 .clo_fini = osc_lock_fini,
921 .clo_enqueue = mdc_lock_enqueue,
922 .clo_cancel = mdc_lock_lockless_cancel,
923 .clo_print = osc_lock_print
926 static const struct cl_lock_operations mdc_lock_ops = {
927 .clo_fini = osc_lock_fini,
928 .clo_enqueue = mdc_lock_enqueue,
929 .clo_cancel = osc_lock_cancel,
930 .clo_print = osc_lock_print,
933 int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
934 struct cl_lock *lock, const struct cl_io *io)
936 struct osc_lock *ols;
937 __u32 enqflags = lock->cll_descr.cld_enq_flags;
938 __u64 flags = osc_enq2ldlm_flags(enqflags);
942 /* Ignore AGL for Data-on-MDT, stat returns size data */
943 if ((enqflags & CEF_SPECULATIVE) != 0)
946 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
947 if (unlikely(ols == NULL))
950 ols->ols_state = OLS_NEW;
951 spin_lock_init(&ols->ols_lock);
952 INIT_LIST_HEAD(&ols->ols_waiting_list);
953 INIT_LIST_HEAD(&ols->ols_wait_entry);
954 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
955 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
957 ols->ols_flags = flags;
958 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
959 if (lock->cll_descr.cld_mode == CLM_GROUP)
960 ols->ols_flags |= LDLM_FL_ATOMIC_CB;
962 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
963 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
964 ols->ols_glimpse = 1;
966 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
968 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
970 if (!(enqflags & CEF_MUST))
971 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
972 if (ols->ols_locklessable && !(enqflags & CEF_DISCARD_DATA))
973 ols->ols_flags |= LDLM_FL_DENY_ON_CONTENTION;
975 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
976 osc_lock_set_writer(env, io, obj, ols);
978 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx\n",
979 lock, ols, ols->ols_flags);
986 * An implementation of cl_io_operations specific methods for MDC layer.
989 static int mdc_async_upcall(void *a, int rc)
991 struct osc_async_cbargs *args = a;
994 complete(&args->opc_sync);
998 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
999 pgoff_t index, struct lustre_handle *lh)
1001 struct ldlm_lock *lock;
1003 /* find DOM lock protecting object */
1004 lock = mdc_dlmlock_at_pgoff(env, osc, index,
1005 OSC_DAP_FL_TEST_LOCK |
1006 OSC_DAP_FL_CANCELING);
1008 struct ldlm_resource *res;
1009 struct ldlm_res_id *resname;
1011 resname = &osc_env_info(env)->oti_resname;
1012 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1013 res = ldlm_resource_get(osc_export(osc)->exp_obd->obd_namespace,
1014 NULL, resname, LDLM_IBITS, 0);
1015 ldlm_resource_dump(D_ERROR, res);
1016 libcfs_debug_dumpstack(NULL);
1019 *lh = lock->l_remote_handle;
1020 LDLM_LOCK_PUT(lock);
1025 static int mdc_io_setattr_start(const struct lu_env *env,
1026 const struct cl_io_slice *slice)
1028 struct cl_io *io = slice->cis_io;
1029 struct osc_io *oio = cl2osc_io(env, slice);
1030 struct cl_object *obj = slice->cis_obj;
1031 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1032 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1033 struct obdo *oa = &oio->oi_oa;
1034 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1035 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1036 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1037 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1040 /* silently ignore non-truncate setattr for Data-on-MDT object */
1041 if (cl_io_is_trunc(io)) {
1042 /* truncate cache dirty pages first */
1043 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1049 if (oio->oi_lockless == 0) {
1050 cl_object_attr_lock(obj);
1051 rc = cl_object_attr_get(env, obj, attr);
1053 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1054 unsigned int cl_valid = 0;
1056 if (ia_avalid & ATTR_SIZE) {
1057 attr->cat_size = size;
1058 attr->cat_kms = size;
1059 cl_valid = (CAT_SIZE | CAT_KMS);
1061 if (ia_avalid & ATTR_MTIME_SET) {
1062 attr->cat_mtime = lvb->lvb_mtime;
1063 cl_valid |= CAT_MTIME;
1065 if (ia_avalid & ATTR_ATIME_SET) {
1066 attr->cat_atime = lvb->lvb_atime;
1067 cl_valid |= CAT_ATIME;
1069 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1070 attr->cat_ctime = lvb->lvb_ctime;
1071 cl_valid |= CAT_CTIME;
1073 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1075 cl_object_attr_unlock(obj);
1080 if (!(ia_avalid & ATTR_SIZE))
1083 memset(oa, 0, sizeof(*oa));
1084 oa->o_oi = loi->loi_oi;
1085 oa->o_mtime = attr->cat_mtime;
1086 oa->o_atime = attr->cat_atime;
1087 oa->o_ctime = attr->cat_ctime;
1090 oa->o_blocks = OBD_OBJECT_EOF;
1091 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1092 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1094 if (oio->oi_lockless) {
1095 oa->o_flags = OBD_FL_SRVLOCK;
1096 oa->o_valid |= OBD_MD_FLFLAGS;
1098 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1101 oa->o_valid |= OBD_MD_FLHANDLE;
1104 init_completion(&cbargs->opc_sync);
1106 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1107 mdc_async_upcall, cbargs);
1108 cbargs->opc_rpc_sent = rc == 0;
1112 static int mdc_io_read_ahead(const struct lu_env *env,
1113 const struct cl_io_slice *ios,
1114 pgoff_t start, struct cl_read_ahead *ra)
1116 struct osc_object *osc = cl2osc(ios->cis_obj);
1117 struct ldlm_lock *dlmlock;
1121 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1122 if (dlmlock == NULL)
1125 if (dlmlock->l_req_mode != LCK_PR) {
1126 struct lustre_handle lockh;
1128 ldlm_lock2handle(dlmlock, &lockh);
1129 ldlm_lock_addref(&lockh, LCK_PR);
1130 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1133 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1134 ra->cra_end_idx = CL_PAGE_EOF;
1135 ra->cra_release = osc_read_ahead_release;
1136 ra->cra_cbdata = dlmlock;
1141 int mdc_io_fsync_start(const struct lu_env *env,
1142 const struct cl_io_slice *slice)
1144 struct cl_io *io = slice->cis_io;
1145 struct cl_fsync_io *fio = &io->u.ci_fsync;
1146 struct cl_object *obj = slice->cis_obj;
1147 struct osc_object *osc = cl2osc(obj);
1152 /* a MDC lock always covers whole object, do sync for whole
1153 * possible range despite of supplied start/end values.
1155 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1156 fio->fi_mode == CL_FSYNC_DISCARD);
1158 fio->fi_nr_written += result;
1161 if (fio->fi_mode == CL_FSYNC_ALL) {
1164 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1167 /* Use OSC sync code because it is asynchronous.
1168 * It is to be added into MDC and avoid the using of
1169 * OST_SYNC at both MDC and MDT.
1171 rc = osc_fsync_ost(env, osc, fio);
1179 struct mdc_data_version_args {
1180 struct osc_io *dva_oio;
1184 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1187 struct mdc_data_version_args *dva = args;
1188 struct osc_io *oio = dva->dva_oio;
1189 const struct mdt_body *body;
1195 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1197 GOTO(out, rc = -EPROTO);
1199 /* Prepare OBDO from mdt_body for CLIO */
1200 oio->oi_oa.o_valid = body->mbo_valid;
1201 oio->oi_oa.o_flags = body->mbo_flags;
1202 oio->oi_oa.o_data_version = body->mbo_version;
1203 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1206 oio->oi_cbarg.opc_rc = rc;
1207 complete(&oio->oi_cbarg.opc_sync);
1211 static int mdc_io_data_version_start(const struct lu_env *env,
1212 const struct cl_io_slice *slice)
1214 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1215 struct osc_io *oio = cl2osc_io(env, slice);
1216 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1217 struct osc_object *obj = cl2osc(slice->cis_obj);
1218 struct obd_export *exp = osc_export(obj);
1219 struct ptlrpc_request *req;
1220 struct mdt_body *body;
1221 struct mdc_data_version_args *dva;
1226 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1227 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1228 oio->oi_oa.o_valid = OBD_MD_FLID;
1230 init_completion(&cbargs->opc_sync);
1232 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1236 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1238 ptlrpc_request_free(req);
1242 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1243 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1244 body->mbo_valid = OBD_MD_FLID;
1245 /* Indicate that data version is needed */
1246 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1247 body->mbo_flags = 0;
1249 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1250 body->mbo_valid |= OBD_MD_FLFLAGS;
1251 body->mbo_flags |= OBD_FL_SRVLOCK;
1252 if (dv->dv_flags & LL_DV_WR_FLUSH)
1253 body->mbo_flags |= OBD_FL_FLUSH;
1256 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1257 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1258 ptlrpc_request_set_replen(req);
1260 req->rq_interpret_reply = mdc_data_version_interpret;
1261 dva = ptlrpc_req_async_args(dva, req);
1264 ptlrpcd_add_req(req);
1269 static void mdc_io_data_version_end(const struct lu_env *env,
1270 const struct cl_io_slice *slice)
1272 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1273 struct osc_io *oio = cl2osc_io(env, slice);
1274 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1277 wait_for_completion(&cbargs->opc_sync);
1279 if (cbargs->opc_rc != 0) {
1280 slice->cis_io->ci_result = cbargs->opc_rc;
1282 slice->cis_io->ci_result = 0;
1283 if (!(oio->oi_oa.o_valid &
1284 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1285 slice->cis_io->ci_result = -ENOTSUPP;
1287 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1288 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1289 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1290 dv->dv_data_version = oio->oi_oa.o_data_version;
1296 static struct cl_io_operations mdc_io_ops = {
1299 .cio_iter_init = osc_io_rw_iter_init,
1300 .cio_iter_fini = osc_io_rw_iter_fini,
1301 .cio_start = osc_io_read_start,
1304 .cio_iter_init = osc_io_rw_iter_init,
1305 .cio_iter_fini = osc_io_rw_iter_fini,
1306 .cio_start = osc_io_write_start,
1307 .cio_end = osc_io_end,
1310 .cio_iter_init = osc_io_iter_init,
1311 .cio_iter_fini = osc_io_iter_fini,
1312 .cio_start = mdc_io_setattr_start,
1313 .cio_end = osc_io_setattr_end,
1315 [CIT_DATA_VERSION] = {
1316 .cio_start = mdc_io_data_version_start,
1317 .cio_end = mdc_io_data_version_end,
1320 .cio_iter_init = osc_io_iter_init,
1321 .cio_iter_fini = osc_io_iter_fini,
1322 .cio_start = osc_io_fault_start,
1323 .cio_end = osc_io_end,
1326 .cio_start = mdc_io_fsync_start,
1327 .cio_end = osc_io_fsync_end,
1330 .cio_read_ahead = mdc_io_read_ahead,
1331 .cio_submit = osc_io_submit,
1332 .cio_commit_async = osc_io_commit_async,
1335 int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1338 struct osc_io *oio = osc_env_io(env);
1340 CL_IO_SLICE_CLEAN(oio, oi_cl);
1341 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1345 static void mdc_build_res_name(struct osc_object *osc,
1346 struct ldlm_res_id *resname)
1348 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1352 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1353 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1356 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1357 struct cl_req_attr *attr)
1359 u64 flags = attr->cra_flags;
1361 /* Copy object FID to cl_attr */
1362 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1364 if (flags & OBD_MD_FLGROUP)
1365 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1367 if (flags & OBD_MD_FLID)
1368 attr->cra_oa->o_valid |= OBD_MD_FLID;
1370 if (flags & OBD_MD_FLHANDLE) {
1371 struct osc_page *opg;
1373 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1374 if (!opg->ops_srvlock) {
1377 rc = mdc_get_lock_handle(env, cl2osc(obj),
1379 &attr->cra_oa->o_handle);
1381 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1382 "uncovered page!\n");
1385 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1391 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1392 struct cl_attr *attr)
1394 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1396 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1397 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1399 return osc_attr_get(env, obj, attr);
1402 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1404 struct osc_object *osc = (struct osc_object *)data;
1405 struct ost_lvb *lvb = &lock->l_ost_lvb;
1406 struct lov_oinfo *oinfo;
1409 if (lock->l_ast_data == data) {
1410 lock->l_ast_data = NULL;
1412 LASSERT(osc != NULL);
1413 LASSERT(osc->oo_oinfo != NULL);
1414 LASSERT(lvb != NULL);
1416 /* Updates lvb in lock by the cached oinfo */
1417 oinfo = osc->oo_oinfo;
1419 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1420 "%llu %llu %llu by oinfo size %llu blocks %llu "
1421 "[cma]time %llu %llu %llu", lvb->lvb_size,
1422 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1423 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1424 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1425 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1426 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1428 cl_object_attr_lock(&osc->oo_cl);
1429 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1430 cl_object_attr_unlock(&osc->oo_cl);
1431 ldlm_clear_lvb_cached(lock);
1433 RETURN(LDLM_ITER_CONTINUE);
1436 int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1438 struct osc_object *osc = cl2osc(obj);
1439 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1441 /* DLM locks don't hold a reference of osc_object so we have to
1442 * clear it before the object is being destroyed. */
1443 osc_build_res_name(osc, resname);
1444 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1445 mdc_object_ast_clear, osc);
1449 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1450 struct ldlm_lock *lock)
1452 /* if lock cancel is initiated from llite then it is combined
1453 * lock with DOM bit and it may have no l_ast_data initialized yet,
1454 * so init it here with given osc_object.
1456 mdc_set_dom_lock_data(lock, cl2osc(obj));
1457 RETURN(mdc_dlm_blocking_ast0(env, lock, LDLM_CB_CANCELING));
1460 static const struct cl_object_operations mdc_ops = {
1461 .coo_page_init = osc_page_init,
1462 .coo_lock_init = mdc_lock_init,
1463 .coo_io_init = mdc_io_init,
1464 .coo_attr_get = mdc_attr_get,
1465 .coo_attr_update = osc_attr_update,
1466 .coo_glimpse = osc_object_glimpse,
1467 .coo_req_attr_set = mdc_req_attr_set,
1468 .coo_prune = mdc_object_prune,
1469 .coo_object_flush = mdc_object_flush
1472 static const struct osc_object_operations mdc_object_ops = {
1473 .oto_build_res_name = mdc_build_res_name,
1474 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1477 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1478 const struct lu_object_conf *conf)
1480 struct osc_object *osc = lu2osc(obj);
1482 if (osc->oo_initialized)
1485 osc->oo_initialized = true;
1487 return osc_object_init(env, obj, conf);
1490 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1492 osc_object_free(env, obj);
1495 static const struct lu_object_operations mdc_lu_obj_ops = {
1496 .loo_object_init = mdc_object_init,
1497 .loo_object_delete = NULL,
1498 .loo_object_release = NULL,
1499 .loo_object_free = mdc_object_free,
1500 .loo_object_print = osc_object_print,
1501 .loo_object_invariant = NULL
1504 struct lu_object *mdc_object_alloc(const struct lu_env *env,
1505 const struct lu_object_header *unused,
1506 struct lu_device *dev)
1508 struct osc_object *osc;
1509 struct lu_object *obj;
1511 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1514 lu_object_init(obj, NULL, dev);
1515 osc->oo_cl.co_ops = &mdc_ops;
1516 obj->lo_ops = &mdc_lu_obj_ops;
1517 osc->oo_obj_ops = &mdc_object_ops;
1518 osc->oo_initialized = false;
1525 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1526 struct lustre_cfg *cfg)
1528 size_t count = class_modify_config(cfg, PARAM_MDC,
1529 &d->ld_obd->obd_kset.kobj);
1530 return count > 0 ? 0 : count;
1533 const struct lu_device_operations mdc_lu_ops = {
1534 .ldo_object_alloc = mdc_object_alloc,
1535 .ldo_process_config = mdc_process_config,
1536 .ldo_recovery_complete = NULL,
1539 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1540 struct lu_device_type *t,
1541 struct lustre_cfg *cfg)
1543 struct lu_device *d;
1544 struct osc_device *od;
1545 struct obd_device *obd;
1550 RETURN(ERR_PTR(-ENOMEM));
1552 cl_device_init(&od->od_cl, t);
1554 d->ld_ops = &mdc_lu_ops;
1557 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1559 RETURN(ERR_PTR(-ENODEV));
1561 rc = mdc_setup(obd, cfg);
1563 osc_device_free(env, d);
1564 RETURN(ERR_PTR(rc));
1566 od->od_exp = obd->obd_self_export;
1570 static const struct lu_device_type_operations mdc_device_type_ops = {
1571 .ldto_device_alloc = mdc_device_alloc,
1572 .ldto_device_free = osc_device_free,
1573 .ldto_device_init = osc_device_init,
1574 .ldto_device_fini = osc_device_fini
1577 struct lu_device_type mdc_device_type = {
1578 .ldt_tags = LU_DEVICE_CL,
1579 .ldt_name = LUSTRE_MDC_NAME,
1580 .ldt_ops = &mdc_device_type_ops,
1581 .ldt_ctx_tags = LCT_CL_THREAD