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 <linux/falloc.h>
38 #include <uapi/linux/lustre/lustre_param.h>
40 #include "mdc_internal.h"
42 static void mdc_lock_build_policy(const struct lu_env *env,
43 const struct cl_lock *lock,
44 union ldlm_policy_data *policy)
46 memset(policy, 0, sizeof *policy);
47 policy->l_inodebits.bits = MDS_INODELOCK_DOM;
49 policy->l_inodebits.li_gid = lock->cll_descr.cld_gid;
53 int mdc_ldlm_glimpse_ast(struct ldlm_lock *dlmlock, void *data)
55 return osc_ldlm_glimpse_ast(dlmlock, data);
58 static void mdc_lock_build_einfo(const struct lu_env *env,
59 const struct cl_lock *lock,
60 struct osc_object *osc,
61 struct ldlm_enqueue_info *einfo)
63 einfo->ei_type = LDLM_IBITS;
64 einfo->ei_mode = osc_cl_lock2ldlm(lock->cll_descr.cld_mode);
65 einfo->ei_cb_bl = mdc_ldlm_blocking_ast;
66 einfo->ei_cb_cp = ldlm_completion_ast;
67 einfo->ei_cb_gl = mdc_ldlm_glimpse_ast;
68 einfo->ei_cbdata = osc; /* value to be put into ->l_ast_data */
69 einfo->ei_req_slot = 1;
72 static void mdc_lock_lvb_update(const struct lu_env *env,
73 struct osc_object *osc,
74 struct ldlm_lock *dlmlock,
77 static int mdc_set_dom_lock_data(struct ldlm_lock *lock, void *data)
81 LASSERT(lock != NULL);
82 LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
84 lock_res_and_lock(lock);
86 if (lock->l_ast_data == NULL)
87 lock->l_ast_data = data;
88 if (lock->l_ast_data == data)
91 unlock_res_and_lock(lock);
96 static int mdc_dom_lock_match(const struct lu_env *env, struct obd_export *exp,
97 struct ldlm_res_id *res_id, enum ldlm_type type,
98 union ldlm_policy_data *policy,
99 enum ldlm_mode mode, __u64 *flags,
100 struct osc_object *obj,
101 struct lustre_handle *lockh,
102 enum ldlm_match_flags match_flags)
104 struct obd_device *obd = exp->exp_obd;
105 __u64 lflags = *flags;
110 rc = ldlm_lock_match_with_skip(obd->obd_namespace, lflags, 0,
111 res_id, type, policy, mode, lockh, match_flags);
112 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
116 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
118 LASSERT(lock != NULL);
119 if (mdc_set_dom_lock_data(lock, obj)) {
120 lock_res_and_lock(lock);
121 if (!ldlm_is_lvb_cached(lock)) {
122 LASSERT(lock->l_ast_data == obj);
123 mdc_lock_lvb_update(env, obj, lock, NULL);
124 ldlm_set_lvb_cached(lock);
126 unlock_res_and_lock(lock);
128 ldlm_lock_decref(lockh, rc);
137 * Finds an existing lock covering a page with given index.
138 * Copy of osc_obj_dlmlock_at_pgoff() but for DoM IBITS lock.
140 static struct ldlm_lock *mdc_dlmlock_at_pgoff(const struct lu_env *env,
141 struct osc_object *obj,
143 enum osc_dap_flags dap_flags)
145 struct osc_thread_info *info = osc_env_info(env);
146 struct ldlm_res_id *resname = &info->oti_resname;
147 union ldlm_policy_data *policy = &info->oti_policy;
148 struct lustre_handle lockh;
149 struct ldlm_lock *lock = NULL;
152 enum ldlm_match_flags match_flags = 0;
156 fid_build_reg_res_name(lu_object_fid(osc2lu(obj)), resname);
157 mdc_lock_build_policy(env, NULL, policy);
158 policy->l_inodebits.li_gid = LDLM_GID_ANY;
160 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
161 if (dap_flags & OSC_DAP_FL_TEST_LOCK)
162 flags |= LDLM_FL_TEST_LOCK;
164 if (dap_flags & OSC_DAP_FL_AST)
165 match_flags |= LDLM_MATCH_AST;
167 if (dap_flags & OSC_DAP_FL_CANCELING)
168 match_flags |= LDLM_MATCH_UNREF;
171 /* Next, search for already existing extent locks that will cover us */
172 /* If we're trying to read, we also search for an existing PW lock. The
173 * VFS and page cache already protect us locally, so lots of readers/
174 * writers can share a single PW lock. */
175 mode = mdc_dom_lock_match(env, osc_export(obj), resname, LDLM_IBITS,
176 policy, LCK_PR | LCK_PW | LCK_GROUP, &flags,
177 obj, &lockh, match_flags);
179 lock = ldlm_handle2lock(&lockh);
180 /* RACE: the lock is cancelled so let's try again */
181 if (unlikely(lock == NULL))
189 * Check if page @page is covered by an extra lock or discard it.
191 static bool mdc_check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
192 void **pvec, int count, void *cbdata)
194 struct osc_thread_info *info = osc_env_info(env);
195 struct osc_object *osc = cbdata;
199 for (i = 0; i < count; i++) {
200 struct osc_page *ops = pvec[i];
202 index = osc_index(ops);
203 if (index >= info->oti_fn_index) {
204 struct ldlm_lock *tmp;
205 struct cl_page *page = ops->ops_cl.cpl_page;
207 /* refresh non-overlapped index */
208 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
209 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_AST);
211 info->oti_fn_index = CL_PAGE_EOF;
213 } else if (cl_page_own(env, io, page) == 0) {
214 /* discard the page */
215 cl_page_discard(env, io, page);
216 cl_page_disown(env, io, page);
218 LASSERT(page->cp_state == CPS_FREEING);
222 info->oti_next_index = index + 1;
228 * Discard pages protected by the given lock. This function traverses radix
229 * tree to find all covering pages and discard them. If a page is being covered
230 * by other locks, it should remain in cache.
232 * If error happens on any step, the process continues anyway (the reasoning
233 * behind this being that lock cancellation cannot be delayed indefinitely).
235 static int mdc_lock_discard_pages(const struct lu_env *env,
236 struct osc_object *osc,
237 pgoff_t start, pgoff_t end,
240 struct osc_thread_info *info = osc_env_info(env);
241 struct cl_io *io = &info->oti_io;
242 osc_page_gang_cbt cb;
247 io->ci_obj = cl_object_top(osc2cl(osc));
248 io->ci_ignore_layout = 1;
249 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
253 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
254 info->oti_fn_index = info->oti_next_index = start;
256 osc_page_gang_lookup(env, io, osc, info->oti_next_index,
257 end, cb, (void *)osc);
263 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
264 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
272 if (mode == CLM_WRITE) {
273 result = osc_cache_writeback_range(env, obj, start, end, 1,
275 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
276 obj, start, end, result,
277 discard ? "discarded" : "written back");
282 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
283 rc = mdc_lock_discard_pages(env, obj, start, end,
284 mode == CLM_WRITE || discard);
285 if (result == 0 && rc < 0)
291 static void mdc_lock_lockless_cancel(const struct lu_env *env,
292 const struct cl_lock_slice *slice)
294 struct osc_lock *ols = cl2osc_lock(slice);
295 struct osc_object *osc = cl2osc(slice->cls_obj);
296 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
299 LASSERT(ols->ols_dlmlock == NULL);
300 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
303 CERROR("Pages for lockless lock %p were not purged(%d)\n",
306 osc_lock_wake_waiters(env, osc, ols);
310 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
311 * and ldlm_lock caches.
313 static int mdc_dlm_canceling(const struct lu_env *env,
314 struct ldlm_lock *dlmlock)
316 struct cl_object *obj = NULL;
319 enum cl_lock_mode mode = CLM_READ;
323 lock_res_and_lock(dlmlock);
324 if (!ldlm_is_granted(dlmlock)) {
325 dlmlock->l_ast_data = NULL;
326 unlock_res_and_lock(dlmlock);
330 discard = ldlm_is_discard_data(dlmlock);
331 if (dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))
334 if (dlmlock->l_ast_data != NULL) {
335 obj = osc2cl(dlmlock->l_ast_data);
338 unlock_res_and_lock(dlmlock);
340 /* if l_ast_data is NULL, the dlmlock was enqueued by AGL or
341 * the object has been destroyed. */
343 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
345 /* Destroy pages covered by the extent of the DLM lock */
346 result = mdc_lock_flush(env, cl2osc(obj), 0,
347 CL_PAGE_EOF, mode, discard);
348 /* Losing a lock, set KMS to 0.
349 * NB: assumed that DOM lock covers whole data on MDT.
351 /* losing a lock, update kms */
352 lock_res_and_lock(dlmlock);
353 dlmlock->l_ast_data = NULL;
354 cl_object_attr_lock(obj);
356 cl_object_attr_update(env, obj, attr, CAT_KMS);
357 cl_object_attr_unlock(obj);
358 unlock_res_and_lock(dlmlock);
359 cl_object_put(env, obj);
364 int mdc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
365 struct ldlm_lock_desc *new, void *data, int reason)
372 case LDLM_CB_BLOCKING: {
373 struct lustre_handle lockh;
375 ldlm_lock2handle(dlmlock, &lockh);
376 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
381 case LDLM_CB_CANCELING: {
386 * This can be called in the context of outer IO, e.g.,
388 * osc_enqueue_base()->...
389 * ->ldlm_prep_elc_req()->...
390 * ->ldlm_cancel_callback()->...
391 * ->osc_ldlm_blocking_ast()
393 * new environment has to be created to not corrupt outer
396 env = cl_env_get(&refcheck);
402 rc = mdc_dlm_canceling(env, dlmlock);
403 cl_env_put(env, &refcheck);
413 * Updates object attributes from a lock value block (lvb) received together
414 * with the DLM lock reply from the server.
415 * This can be optimized to not update attributes when lock is a result of a
418 * Called under lock and resource spin-locks.
420 void mdc_lock_lvb_update(const struct lu_env *env, struct osc_object *osc,
421 struct ldlm_lock *dlmlock, struct ost_lvb *lvb)
423 struct cl_object *obj = osc2cl(osc);
424 struct lov_oinfo *oinfo = osc->oo_oinfo;
425 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
426 unsigned valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME |
428 unsigned int setkms = 0;
433 LASSERT(dlmlock != NULL);
434 lvb = &dlmlock->l_ost_lvb;
436 cl_lvb2attr(attr, lvb);
438 cl_object_attr_lock(obj);
439 if (dlmlock != NULL) {
442 check_res_locked(dlmlock->l_resource);
443 size = lvb->lvb_size;
445 if (size >= oinfo->loi_kms) {
447 attr->cat_kms = size;
450 ldlm_lock_allow_match_locked(dlmlock);
453 /* The size should not be less than the kms */
454 if (attr->cat_size < oinfo->loi_kms)
455 attr->cat_size = oinfo->loi_kms;
457 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
458 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
459 setkms ? "" : " leaving",
460 setkms ? attr->cat_kms : oinfo->loi_kms,
461 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
463 cl_object_attr_update(env, obj, attr, valid);
464 cl_object_attr_unlock(obj);
468 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
469 struct lustre_handle *lockh)
471 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
472 struct ldlm_lock *dlmlock;
476 dlmlock = ldlm_handle2lock_long(lockh, 0);
477 LASSERT(dlmlock != NULL);
479 /* lock reference taken by ldlm_handle2lock_long() is
480 * owned by osc_lock and released in osc_lock_detach()
482 lu_ref_add_atomic(&dlmlock->l_reference, "osc_lock", oscl);
483 oscl->ols_has_ref = 1;
485 LASSERT(oscl->ols_dlmlock == NULL);
486 oscl->ols_dlmlock = dlmlock;
488 /* This may be a matched lock for glimpse request, do not hold
489 * lock reference in that case. */
490 if (!oscl->ols_glimpse) {
491 /* hold a refc for non glimpse lock which will
492 * be released in osc_lock_cancel() */
493 lustre_handle_copy(&oscl->ols_handle, lockh);
494 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
498 /* Lock must have been granted. */
499 lock_res_and_lock(dlmlock);
500 if (ldlm_is_granted(dlmlock)) {
501 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
503 /* extend the lock extent, otherwise it will have problem when
504 * we decide whether to grant a lockless lock. */
505 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
506 descr->cld_start = 0;
507 descr->cld_end = CL_PAGE_EOF;
509 /* no lvb update for matched lock */
510 if (!ldlm_is_lvb_cached(dlmlock)) {
511 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
512 LASSERT(osc == dlmlock->l_ast_data);
513 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
514 ldlm_set_lvb_cached(dlmlock);
517 unlock_res_and_lock(dlmlock);
519 LASSERT(oscl->ols_state != OLS_GRANTED);
520 oscl->ols_state = OLS_GRANTED;
525 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
526 * received from a server, or after mdc_enqueue_send() matched a local DLM
529 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
532 struct osc_lock *oscl = cookie;
533 struct cl_lock_slice *slice = &oscl->ols_cl;
539 env = cl_env_percpu_get();
540 /* should never happen, similar to osc_ldlm_blocking_ast(). */
541 LASSERT(!IS_ERR(env));
543 rc = ldlm_error2errno(errcode);
544 if (oscl->ols_state == OLS_ENQUEUED) {
545 oscl->ols_state = OLS_UPCALL_RECEIVED;
546 } else if (oscl->ols_state == OLS_CANCELLED) {
549 CERROR("Impossible state: %d\n", oscl->ols_state);
553 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
555 mdc_lock_granted(env, oscl, lockh);
557 /* Error handling, some errors are tolerable. */
558 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 /* This is needed only for old servers (before 2.14) support */
574 int mdc_fill_lvb(struct req_capsule *pill, struct ost_lvb *lvb)
576 struct mdt_body *body;
578 /* get LVB data from mdt_body otherwise */
579 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
583 if (!(body->mbo_valid & OBD_MD_DOM_SIZE))
586 mdc_body2lvb(body, lvb);
590 static int mdc_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
591 osc_enqueue_upcall_f upcall, void *cookie,
592 struct lustre_handle *lockh, enum ldlm_mode mode,
593 __u64 *flags, int errcode)
595 struct osc_lock *ols = cookie;
596 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
601 /* needed only for glimpse from an old server (< 2.14) */
602 if (glimpse && !exp_connect_dom_lvb(exp) && errcode >= 0)
603 rc = mdc_fill_lvb(&req->rq_pill, &ols->ols_lvb);
605 if (glimpse && errcode == ELDLM_LOCK_ABORTED) {
606 struct ldlm_reply *rep;
608 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
610 rep->lock_policy_res2 =
611 ptlrpc_status_ntoh(rep->lock_policy_res2);
612 if (rep->lock_policy_res2)
613 errcode = rep->lock_policy_res2;
617 *flags |= LDLM_FL_LVB_READY;
618 } else if (errcode == ELDLM_OK) {
619 struct ldlm_lock *lock;
621 /* Callers have references, should be valid always */
622 lock = ldlm_handle2lock(lockh);
624 /* At this point ols_lvb must be filled with correct LVB either
625 * by mdc_fill_lvb() above or by ldlm_cli_enqueue_fini().
626 * DoM uses l_ost_lvb to store LVB data, so copy it here from
627 * just updated ols_lvb.
629 lock_res_and_lock(lock);
630 memcpy(&lock->l_ost_lvb, &ols->ols_lvb,
631 sizeof(lock->l_ost_lvb));
632 unlock_res_and_lock(lock);
634 *flags |= LDLM_FL_LVB_READY;
637 /* Call the update callback. */
638 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
640 /* release the reference taken in ldlm_cli_enqueue() */
641 if (errcode == ELDLM_LOCK_MATCHED)
643 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
644 ldlm_lock_decref(lockh, mode);
649 static int mdc_enqueue_interpret(const struct lu_env *env,
650 struct ptlrpc_request *req,
653 struct osc_enqueue_args *aa = args;
654 struct ldlm_lock *lock;
655 struct lustre_handle *lockh = &aa->oa_lockh;
656 enum ldlm_mode mode = aa->oa_mode;
657 struct ldlm_enqueue_info einfo = {
658 .ei_type = aa->oa_type,
664 LASSERT(!aa->oa_speculative);
666 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
668 lock = ldlm_handle2lock(lockh);
669 LASSERTF(lock != NULL,
670 "lockh %#llx, req %px, aa %px - client evicted?\n",
671 lockh->cookie, req, aa);
673 /* Take an additional reference so that a blocking AST that
674 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
675 * to arrive after an upcall has been executed by
676 * mdc_enqueue_fini().
678 ldlm_lock_addref(lockh, mode);
680 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
681 CFS_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
683 /* Let CP AST to grant the lock first. */
684 CFS_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
686 /* Complete obtaining the lock procedure. */
687 rc = ldlm_cli_enqueue_fini(aa->oa_exp, &req->rq_pill, &einfo, 1,
688 aa->oa_flags, aa->oa_lvb, aa->oa_lvb ?
689 sizeof(*aa->oa_lvb) : 0, lockh, rc, true);
690 /* Complete mdc stuff. */
691 rc = mdc_enqueue_fini(aa->oa_exp, req, aa->oa_upcall, aa->oa_cookie,
692 lockh, mode, aa->oa_flags, rc);
694 CFS_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
696 ldlm_lock_decref(lockh, mode);
701 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
702 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
703 * other synchronous requests, however keeping some locks and trying to obtain
704 * others may take a considerable amount of time in a case of ost failure; and
705 * when other sync requests do not get released lock from a client, the client
706 * is excluded from the cluster -- such scenarious make the life difficult, so
707 * release locks just after they are obtained. */
708 static int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
709 struct ldlm_res_id *res_id, __u64 *flags,
710 union ldlm_policy_data *policy, struct ost_lvb *lvb,
711 osc_enqueue_upcall_f upcall, void *cookie,
712 struct ldlm_enqueue_info *einfo, int async)
714 struct obd_device *obd = exp->exp_obd;
715 struct lustre_handle lockh = { 0 };
716 struct ptlrpc_request *req = NULL;
717 struct ldlm_intent *lit;
719 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
720 __u64 search_flags = *flags;
721 __u64 match_flags = 0;
725 bool compat_glimpse = glimpse && !exp_connect_dom_lvb(exp);
729 mode = einfo->ei_mode;
730 if (einfo->ei_mode == LCK_PR)
733 search_flags |= LDLM_FL_LVB_READY;
735 search_flags |= LDLM_FL_BLOCK_GRANTED;
736 if (mode == LCK_GROUP)
737 match_flags = LDLM_MATCH_GROUP;
738 mode = ldlm_lock_match_with_skip(obd->obd_namespace, search_flags, 0,
739 res_id, einfo->ei_type, policy, mode,
740 &lockh, match_flags);
742 struct ldlm_lock *matched;
744 if (*flags & LDLM_FL_TEST_LOCK)
747 matched = ldlm_handle2lock(&lockh);
749 if (CFS_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
750 ldlm_set_kms_ignore(matched);
752 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
753 *flags |= LDLM_FL_LVB_READY;
755 /* We already have a lock, and it's referenced. */
756 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
758 ldlm_lock_decref(&lockh, mode);
759 LDLM_LOCK_PUT(matched);
762 ldlm_lock_decref(&lockh, mode);
763 LDLM_LOCK_PUT(matched);
766 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
769 /* Glimpse is intent on old server */
770 req = ptlrpc_request_alloc(class_exp2cliimp(exp), compat_glimpse ?
771 &RQF_LDLM_INTENT : &RQF_LDLM_ENQUEUE);
775 /* For WRITE lock cancel other locks on resource early if any */
776 if (einfo->ei_mode & LCK_PW)
777 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
783 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
785 ptlrpc_request_free(req);
789 if (compat_glimpse) {
790 /* pack the glimpse intent */
791 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
792 lit->opc = IT_GLIMPSE;
795 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
796 *flags &= ~LDLM_FL_BLOCK_GRANTED;
798 if (compat_glimpse) {
799 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
800 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
803 lvb_size = sizeof(*lvb);
804 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
807 ptlrpc_request_set_replen(req);
809 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
810 lvb_size, LVB_T_OST, &lockh, async);
813 struct osc_enqueue_args *aa;
815 aa = ptlrpc_req_async_args(aa, req);
817 aa->oa_mode = einfo->ei_mode;
818 aa->oa_type = einfo->ei_type;
819 lustre_handle_copy(&aa->oa_lockh, &lockh);
820 aa->oa_upcall = upcall;
821 aa->oa_cookie = cookie;
822 aa->oa_speculative = false;
823 aa->oa_flags = flags;
824 aa->oa_lvb = compat_glimpse ? NULL : lvb;
826 req->rq_interpret_reply = mdc_enqueue_interpret;
827 ptlrpcd_add_req(req);
829 ptlrpc_req_finished(req);
834 rc = mdc_enqueue_fini(exp, req, upcall, cookie, &lockh, einfo->ei_mode,
836 ptlrpc_req_finished(req);
841 * Implementation of cl_lock_operations::clo_enqueue() method for osc
842 * layer. This initiates ldlm enqueue:
844 * - cancels conflicting locks early (osc_lock_enqueue_wait());
846 * - calls osc_enqueue_base() to do actual enqueue.
848 * osc_enqueue_base() is supplied with an upcall function that is executed
849 * when lock is received either after a local cached ldlm lock is matched, or
850 * when a reply from the server is received.
852 * This function does not wait for the network communication to complete.
854 static int mdc_lock_enqueue(const struct lu_env *env,
855 const struct cl_lock_slice *slice,
856 struct cl_io *unused, struct cl_sync_io *anchor)
858 struct osc_thread_info *info = osc_env_info(env);
859 struct osc_io *oio = osc_env_io(env);
860 struct osc_object *osc = cl2osc(slice->cls_obj);
861 struct osc_lock *oscl = cl2osc_lock(slice);
862 struct cl_lock *lock = slice->cls_lock;
863 struct ldlm_res_id *resname = &info->oti_resname;
864 union ldlm_policy_data *policy = &info->oti_policy;
865 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
866 void *cookie = (void *)oscl;
872 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
873 "lock = %px, ols = %px\n", lock, oscl);
875 if (oscl->ols_state == OLS_GRANTED)
878 /* Lockahead is not supported on MDT yet */
879 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
880 result = -EOPNOTSUPP;
884 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
885 GOTO(enqueue_base, 0);
887 if (oscl->ols_glimpse) {
888 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
890 GOTO(enqueue_base, 0);
893 result = osc_lock_enqueue_wait(env, osc, oscl);
897 /* we can grant lockless lock right after all conflicting locks
899 if (osc_lock_is_lockless(oscl)) {
900 oscl->ols_state = OLS_GRANTED;
901 oio->oi_lockless = 1;
906 oscl->ols_state = OLS_ENQUEUED;
907 if (anchor != NULL) {
908 atomic_inc(&anchor->csi_sync_nr);
909 oscl->ols_owner = anchor;
913 * DLM lock's ast data must be osc_object;
914 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
917 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
918 mdc_lock_build_policy(env, lock, policy);
919 LASSERT(!oscl->ols_speculative);
920 result = mdc_enqueue_send(env, osc_export(osc), resname,
921 &oscl->ols_flags, policy, &oscl->ols_lvb,
922 upcall, cookie, &oscl->ols_einfo, async);
924 if (osc_lock_is_lockless(oscl)) {
925 oio->oi_lockless = 1;
927 LASSERT(oscl->ols_state == OLS_GRANTED);
928 LASSERT(oscl->ols_hold);
929 LASSERT(oscl->ols_dlmlock != NULL);
934 oscl->ols_state = OLS_CANCELLED;
935 osc_lock_wake_waiters(env, osc, oscl);
938 cl_sync_io_note(env, anchor, result);
943 static const struct cl_lock_operations mdc_lock_lockless_ops = {
944 .clo_fini = osc_lock_fini,
945 .clo_enqueue = mdc_lock_enqueue,
946 .clo_cancel = mdc_lock_lockless_cancel,
947 .clo_print = osc_lock_print
950 static const struct cl_lock_operations mdc_lock_ops = {
951 .clo_fini = osc_lock_fini,
952 .clo_enqueue = mdc_lock_enqueue,
953 .clo_cancel = osc_lock_cancel,
954 .clo_print = osc_lock_print,
957 static int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
958 struct cl_lock *lock, const struct cl_io *io)
960 struct osc_lock *ols;
961 __u32 enqflags = lock->cll_descr.cld_enq_flags;
962 __u64 flags = osc_enq2ldlm_flags(enqflags);
966 /* Ignore AGL for Data-on-MDT, stat returns size data */
967 if ((enqflags & CEF_SPECULATIVE) != 0)
970 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
971 if (unlikely(ols == NULL))
974 ols->ols_state = OLS_NEW;
975 spin_lock_init(&ols->ols_lock);
976 INIT_LIST_HEAD(&ols->ols_waiting_list);
977 INIT_LIST_HEAD(&ols->ols_wait_entry);
978 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
979 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
981 ols->ols_flags = flags;
982 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
984 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
985 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
986 ols->ols_glimpse = 1;
988 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
990 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
992 if (!(enqflags & CEF_MUST))
993 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
995 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
996 osc_lock_set_writer(env, io, obj, ols);
997 else if (io->ci_type == CIT_READ ||
998 (io->ci_type == CIT_FAULT && !io->u.ci_fault.ft_mkwrite))
999 osc_lock_set_reader(env, io, obj, ols);
1001 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx",
1002 lock, ols, ols->ols_flags);
1009 * An implementation of cl_io_operations specific methods for MDC layer.
1012 static int mdc_async_upcall(void *a, int rc)
1014 struct osc_async_cbargs *args = a;
1017 complete(&args->opc_sync);
1021 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
1022 pgoff_t index, struct lustre_handle *lh)
1024 struct ldlm_lock *lock;
1026 /* find DOM lock protecting object */
1027 lock = mdc_dlmlock_at_pgoff(env, osc, index,
1028 OSC_DAP_FL_TEST_LOCK |
1029 OSC_DAP_FL_CANCELING);
1031 struct ldlm_resource *res;
1032 struct ldlm_res_id *resname;
1034 resname = &osc_env_info(env)->oti_resname;
1035 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1036 res = ldlm_resource_get(osc_export(osc)->
1037 exp_obd->obd_namespace,
1038 resname, LDLM_IBITS, 0);
1040 CERROR("No lock resource for "DFID"\n",
1041 PFID(lu_object_fid(osc2lu(osc))));
1043 ldlm_resource_dump(D_ERROR, res);
1047 *lh = lock->l_remote_handle;
1048 LDLM_LOCK_PUT(lock);
1053 static int mdc_io_setattr_start(const struct lu_env *env,
1054 const struct cl_io_slice *slice)
1056 struct cl_io *io = slice->cis_io;
1057 struct osc_io *oio = cl2osc_io(env, slice);
1058 struct cl_object *obj = slice->cis_obj;
1059 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1060 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1061 struct obdo *oa = &oio->oi_oa;
1062 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1063 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1064 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1065 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1068 /* silently ignore non-truncate setattr for Data-on-MDT object */
1069 if (cl_io_is_trunc(io)) {
1070 /* truncate cache dirty pages first */
1071 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1075 } else if (cl_io_is_fallocate(io) &&
1076 io->u.ci_setattr.sa_falloc_mode & FALLOC_FL_PUNCH_HOLE) {
1077 rc = osc_punch_start(env, io, obj);
1082 if (oio->oi_lockless == 0) {
1083 cl_object_attr_lock(obj);
1084 rc = cl_object_attr_get(env, obj, attr);
1086 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1087 unsigned int cl_valid = 0;
1089 if (ia_avalid & ATTR_SIZE) {
1090 attr->cat_size = size;
1091 attr->cat_kms = size;
1092 cl_valid = (CAT_SIZE | CAT_KMS);
1094 if (ia_avalid & ATTR_MTIME_SET) {
1095 attr->cat_mtime = lvb->lvb_mtime;
1096 cl_valid |= CAT_MTIME;
1098 if (ia_avalid & ATTR_ATIME_SET) {
1099 attr->cat_atime = lvb->lvb_atime;
1100 cl_valid |= CAT_ATIME;
1102 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1103 attr->cat_ctime = lvb->lvb_ctime;
1104 cl_valid |= CAT_CTIME;
1106 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1108 cl_object_attr_unlock(obj);
1113 if (!(ia_avalid & ATTR_SIZE) && !cl_io_is_fallocate(io))
1116 memset(oa, 0, sizeof(*oa));
1117 oa->o_oi = loi->loi_oi;
1118 oa->o_mtime = attr->cat_mtime;
1119 oa->o_atime = attr->cat_atime;
1120 oa->o_ctime = attr->cat_ctime;
1121 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1122 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1125 if (oio->oi_lockless) {
1126 oa->o_flags = OBD_FL_SRVLOCK;
1127 oa->o_valid |= OBD_MD_FLFLAGS;
1129 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1132 oa->o_valid |= OBD_MD_FLHANDLE;
1135 init_completion(&cbargs->opc_sync);
1136 if (cl_io_is_fallocate(io)) {
1137 int falloc_mode = io->u.ci_setattr.sa_falloc_mode;
1139 oa->o_size = io->u.ci_setattr.sa_falloc_offset;
1140 oa->o_blocks = io->u.ci_setattr.sa_falloc_end;
1141 rc = osc_fallocate_base(osc_export(cl2osc(obj)), oa,
1142 mdc_async_upcall, cbargs, falloc_mode);
1145 oa->o_blocks = OBD_OBJECT_EOF;
1146 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1147 mdc_async_upcall, cbargs);
1149 cbargs->opc_rpc_sent = rc == 0;
1153 static int mdc_io_read_ahead(const struct lu_env *env,
1154 const struct cl_io_slice *ios,
1155 pgoff_t start, struct cl_read_ahead *ra)
1157 struct osc_object *osc = cl2osc(ios->cis_obj);
1158 struct osc_io *oio = cl2osc_io(env, ios);
1159 struct ldlm_lock *dlmlock;
1163 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1164 if (dlmlock == NULL)
1167 oio->oi_is_readahead = 1;
1168 if (dlmlock->l_req_mode != LCK_PR) {
1169 struct lustre_handle lockh;
1171 ldlm_lock2handle(dlmlock, &lockh);
1172 ldlm_lock_addref(&lockh, LCK_PR);
1173 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1176 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1177 ra->cra_end_idx = CL_PAGE_EOF;
1178 ra->cra_release = osc_read_ahead_release;
1179 ra->cra_dlmlock = dlmlock;
1185 static int mdc_io_fsync_start(const struct lu_env *env,
1186 const struct cl_io_slice *slice)
1188 struct cl_io *io = slice->cis_io;
1189 struct cl_fsync_io *fio = &io->u.ci_fsync;
1190 struct cl_object *obj = slice->cis_obj;
1191 struct osc_object *osc = cl2osc(obj);
1196 if (fio->fi_mode == CL_FSYNC_RECLAIM) {
1197 struct client_obd *cli = osc_cli(osc);
1199 if (!atomic_long_read(&cli->cl_unstable_count)) {
1200 /* Stop flush when there are no unstable pages? */
1201 CDEBUG(D_CACHE, "unstable count is zero\n");
1206 /* a MDC lock always covers whole object, do sync for whole
1207 * possible range despite of supplied start/end values.
1209 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1210 fio->fi_mode == CL_FSYNC_DISCARD);
1212 fio->fi_nr_written += result;
1215 if (fio->fi_mode == CL_FSYNC_ALL || fio->fi_mode == CL_FSYNC_RECLAIM) {
1216 struct osc_io *oio = cl2osc_io(env, slice);
1217 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1220 if (fio->fi_mode == CL_FSYNC_ALL) {
1221 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1225 /* Use OSC sync code because it is asynchronous.
1226 * It is to be added into MDC and avoid the using of
1227 * OST_SYNC at both MDC and MDT.
1229 rc = osc_fsync_ost(env, osc, fio);
1231 cbargs->opc_rpc_sent = 1;
1239 struct mdc_data_version_args {
1240 struct osc_io *dva_oio;
1244 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1247 struct mdc_data_version_args *dva = args;
1248 struct osc_io *oio = dva->dva_oio;
1249 const struct mdt_body *body;
1255 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1257 GOTO(out, rc = -EPROTO);
1259 /* Prepare OBDO from mdt_body for CLIO */
1260 oio->oi_oa.o_valid = body->mbo_valid;
1261 oio->oi_oa.o_flags = body->mbo_flags;
1262 oio->oi_oa.o_data_version = body->mbo_version;
1263 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1266 oio->oi_cbarg.opc_rc = rc;
1267 complete(&oio->oi_cbarg.opc_sync);
1271 static int mdc_io_data_version_start(const struct lu_env *env,
1272 const struct cl_io_slice *slice)
1274 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1275 struct osc_io *oio = cl2osc_io(env, slice);
1276 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1277 struct osc_object *obj = cl2osc(slice->cis_obj);
1278 struct obd_export *exp = osc_export(obj);
1279 struct ptlrpc_request *req;
1280 struct mdt_body *body;
1281 struct mdc_data_version_args *dva;
1286 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1287 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1288 oio->oi_oa.o_valid = OBD_MD_FLID;
1290 init_completion(&cbargs->opc_sync);
1292 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1296 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1298 ptlrpc_request_free(req);
1302 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1303 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1304 body->mbo_valid = OBD_MD_FLID;
1305 /* Indicate that data version is needed */
1306 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1307 body->mbo_flags = 0;
1309 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1310 body->mbo_valid |= OBD_MD_FLFLAGS;
1311 body->mbo_flags |= OBD_FL_SRVLOCK;
1312 if (dv->dv_flags & LL_DV_WR_FLUSH)
1313 body->mbo_flags |= OBD_FL_FLUSH;
1316 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1317 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1318 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX, RCL_SERVER, 0);
1319 ptlrpc_request_set_replen(req);
1321 req->rq_interpret_reply = mdc_data_version_interpret;
1322 dva = ptlrpc_req_async_args(dva, req);
1325 ptlrpcd_add_req(req);
1330 static void mdc_io_data_version_end(const struct lu_env *env,
1331 const struct cl_io_slice *slice)
1333 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1334 struct osc_io *oio = cl2osc_io(env, slice);
1335 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1338 wait_for_completion(&cbargs->opc_sync);
1340 if (cbargs->opc_rc != 0) {
1341 slice->cis_io->ci_result = cbargs->opc_rc;
1343 slice->cis_io->ci_result = 0;
1344 if (!(oio->oi_oa.o_valid &
1345 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1346 slice->cis_io->ci_result = -EOPNOTSUPP;
1348 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1349 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1350 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1351 dv->dv_data_version = oio->oi_oa.o_data_version;
1357 static const struct cl_io_operations mdc_io_ops = {
1360 .cio_iter_init = osc_io_iter_init,
1361 .cio_iter_fini = osc_io_rw_iter_fini,
1362 .cio_start = osc_io_read_start,
1365 .cio_iter_init = osc_io_iter_init,
1366 .cio_iter_fini = osc_io_rw_iter_fini,
1367 .cio_start = osc_io_write_start,
1368 .cio_end = osc_io_end,
1371 .cio_iter_init = osc_io_iter_init,
1372 .cio_iter_fini = osc_io_iter_fini,
1373 .cio_start = mdc_io_setattr_start,
1374 .cio_end = osc_io_setattr_end,
1376 [CIT_DATA_VERSION] = {
1377 .cio_start = mdc_io_data_version_start,
1378 .cio_end = mdc_io_data_version_end,
1381 .cio_iter_init = osc_io_iter_init,
1382 .cio_iter_fini = osc_io_iter_fini,
1383 .cio_start = osc_io_fault_start,
1384 .cio_end = osc_io_end,
1387 .cio_start = mdc_io_fsync_start,
1388 .cio_end = osc_io_fsync_end,
1391 .cio_start = osc_io_lseek_start,
1392 .cio_end = osc_io_lseek_end,
1395 .cio_read_ahead = mdc_io_read_ahead,
1396 .cio_lru_reserve = osc_io_lru_reserve,
1397 .cio_submit = osc_io_submit,
1398 .cio_commit_async = osc_io_commit_async,
1399 .cio_extent_release = osc_io_extent_release,
1402 static int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1405 struct osc_io *oio = osc_env_io(env);
1407 CL_IO_SLICE_CLEAN(oio, oi_cl);
1408 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1412 static void mdc_build_res_name(struct osc_object *osc,
1413 struct ldlm_res_id *resname)
1415 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1419 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1420 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1423 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1424 struct cl_req_attr *attr)
1426 u64 flags = attr->cra_flags;
1428 /* Copy object FID to cl_attr */
1429 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1431 if (flags & OBD_MD_FLGROUP)
1432 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1434 if (flags & OBD_MD_FLID)
1435 attr->cra_oa->o_valid |= OBD_MD_FLID;
1437 if (flags & OBD_MD_FLHANDLE) {
1438 struct osc_page *opg;
1440 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1441 if (!opg->ops_srvlock) {
1444 rc = mdc_get_lock_handle(env, cl2osc(obj),
1446 &attr->cra_oa->o_handle);
1448 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1449 "uncovered page!\n");
1452 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1458 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1459 struct cl_attr *attr)
1461 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1463 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1464 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1466 return osc_attr_get(env, obj, attr);
1469 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1471 struct osc_object *osc = (struct osc_object *)data;
1472 struct ost_lvb *lvb = &lock->l_ost_lvb;
1473 struct lov_oinfo *oinfo;
1476 if (lock->l_ast_data == data) {
1477 lock->l_ast_data = NULL;
1479 LASSERT(osc != NULL);
1480 LASSERT(osc->oo_oinfo != NULL);
1481 LASSERT(lvb != NULL);
1483 /* Updates lvb in lock by the cached oinfo */
1484 oinfo = osc->oo_oinfo;
1486 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1487 "%llu %llu %llu by oinfo size %llu blocks %llu "
1488 "[cma]time %llu %llu %llu", lvb->lvb_size,
1489 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1490 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1491 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1492 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1493 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1495 cl_object_attr_lock(&osc->oo_cl);
1496 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1497 cl_object_attr_unlock(&osc->oo_cl);
1498 ldlm_clear_lvb_cached(lock);
1500 RETURN(LDLM_ITER_CONTINUE);
1503 static int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1505 struct osc_object *osc = cl2osc(obj);
1506 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1508 /* DLM locks don't hold a reference of osc_object so we have to
1509 * clear it before the object is being destroyed. */
1510 osc_build_res_name(osc, resname);
1511 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1512 mdc_object_ast_clear, osc);
1516 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1517 struct ldlm_lock *lock)
1519 /* if lock cancel is initiated from llite then it is combined
1520 * lock with DOM bit and it may have no l_ast_data initialized yet,
1521 * so init it here with given osc_object.
1523 mdc_set_dom_lock_data(lock, cl2osc(obj));
1524 RETURN(mdc_dlm_canceling(env, lock));
1527 static const struct cl_object_operations mdc_ops = {
1528 .coo_page_init = osc_page_init,
1529 .coo_lock_init = mdc_lock_init,
1530 .coo_io_init = mdc_io_init,
1531 .coo_attr_get = mdc_attr_get,
1532 .coo_attr_update = osc_attr_update,
1533 .coo_glimpse = osc_object_glimpse,
1534 .coo_req_attr_set = mdc_req_attr_set,
1535 .coo_prune = mdc_object_prune,
1536 .coo_object_flush = mdc_object_flush
1539 static const struct osc_object_operations mdc_object_ops = {
1540 .oto_build_res_name = mdc_build_res_name,
1541 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1544 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1545 const struct lu_object_conf *conf)
1547 struct osc_object *osc = lu2osc(obj);
1549 if (osc->oo_initialized)
1552 osc->oo_initialized = true;
1554 return osc_object_init(env, obj, conf);
1557 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1559 osc_object_free(env, obj);
1562 static const struct lu_object_operations mdc_lu_obj_ops = {
1563 .loo_object_init = mdc_object_init,
1564 .loo_object_delete = NULL,
1565 .loo_object_release = NULL,
1566 .loo_object_free = mdc_object_free,
1567 .loo_object_print = osc_object_print,
1568 .loo_object_invariant = NULL
1571 static struct lu_object *mdc_object_alloc(const struct lu_env *env,
1572 const struct lu_object_header *unused,
1573 struct lu_device *dev)
1575 struct osc_object *osc;
1576 struct lu_object *obj;
1578 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1581 lu_object_init(obj, NULL, dev);
1582 osc->oo_cl.co_ops = &mdc_ops;
1583 obj->lo_ops = &mdc_lu_obj_ops;
1584 osc->oo_obj_ops = &mdc_object_ops;
1585 osc->oo_initialized = false;
1592 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1593 struct lustre_cfg *cfg)
1595 size_t count = class_modify_config(cfg, PARAM_MDC,
1596 &d->ld_obd->obd_kset.kobj);
1597 return count > 0 ? 0 : count;
1600 const struct lu_device_operations mdc_lu_ops = {
1601 .ldo_object_alloc = mdc_object_alloc,
1602 .ldo_process_config = mdc_process_config,
1603 .ldo_recovery_complete = NULL,
1606 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1607 struct lu_device_type *t,
1608 struct lustre_cfg *cfg)
1610 struct lu_device *d;
1611 struct osc_device *oc;
1612 struct obd_device *obd;
1617 RETURN(ERR_PTR(-ENOMEM));
1619 cl_device_init(&oc->osc_cl, t);
1621 d->ld_ops = &mdc_lu_ops;
1624 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1626 RETURN(ERR_PTR(-ENODEV));
1628 rc = mdc_setup(obd, cfg);
1630 osc_device_free(env, d);
1631 RETURN(ERR_PTR(rc));
1633 oc->osc_exp = obd->obd_self_export;
1634 oc->osc_stats.os_init = ktime_get_real();
1638 static const struct lu_device_type_operations mdc_device_type_ops = {
1639 .ldto_device_alloc = mdc_device_alloc,
1640 .ldto_device_free = osc_device_free,
1641 .ldto_device_init = osc_device_init,
1642 .ldto_device_fini = osc_device_fini
1645 struct lu_device_type mdc_device_type = {
1646 .ldt_tags = LU_DEVICE_CL,
1647 .ldt_name = LUSTRE_MDC_NAME,
1648 .ldt_ops = &mdc_device_type_ops,
1649 .ldt_ctx_tags = LCT_CL_THREAD