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 struct osc_page *ops, void *cbdata)
190 struct osc_thread_info *info = osc_env_info(env);
191 struct osc_object *osc = cbdata;
194 index = osc_index(ops);
195 if (index >= info->oti_fn_index) {
196 struct ldlm_lock *tmp;
197 struct cl_page *page = ops->ops_cl.cpl_page;
199 /* refresh non-overlapped index */
200 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
201 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_AST);
203 info->oti_fn_index = CL_PAGE_EOF;
205 } else if (cl_page_own(env, io, page) == 0) {
206 /* discard the page */
207 cl_page_discard(env, io, page);
208 cl_page_disown(env, io, page);
210 LASSERT(page->cp_state == CPS_FREEING);
214 info->oti_next_index = index + 1;
219 * Discard pages protected by the given lock. This function traverses radix
220 * tree to find all covering pages and discard them. If a page is being covered
221 * by other locks, it should remain in cache.
223 * If error happens on any step, the process continues anyway (the reasoning
224 * behind this being that lock cancellation cannot be delayed indefinitely).
226 static int mdc_lock_discard_pages(const struct lu_env *env,
227 struct osc_object *osc,
228 pgoff_t start, pgoff_t end,
231 struct osc_thread_info *info = osc_env_info(env);
232 struct cl_io *io = &info->oti_io;
233 osc_page_gang_cbt cb;
238 io->ci_obj = cl_object_top(osc2cl(osc));
239 io->ci_ignore_layout = 1;
240 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
244 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
245 info->oti_fn_index = info->oti_next_index = start;
247 osc_page_gang_lookup(env, io, osc, info->oti_next_index,
248 end, cb, (void *)osc);
254 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
255 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
263 if (mode == CLM_WRITE) {
264 result = osc_cache_writeback_range(env, obj, start, end, 1,
266 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
267 obj, start, end, result,
268 discard ? "discarded" : "written back");
273 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
274 rc = mdc_lock_discard_pages(env, obj, start, end,
275 mode == CLM_WRITE || discard);
276 if (result == 0 && rc < 0)
282 void mdc_lock_lockless_cancel(const struct lu_env *env,
283 const struct cl_lock_slice *slice)
285 struct osc_lock *ols = cl2osc_lock(slice);
286 struct osc_object *osc = cl2osc(slice->cls_obj);
287 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
290 LASSERT(ols->ols_dlmlock == NULL);
291 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
294 CERROR("Pages for lockless lock %p were not purged(%d)\n",
297 osc_lock_wake_waiters(env, osc, ols);
301 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
302 * and ldlm_lock caches.
304 static int mdc_dlm_canceling(const struct lu_env *env,
305 struct ldlm_lock *dlmlock)
307 struct cl_object *obj = NULL;
310 enum cl_lock_mode mode = CLM_READ;
314 lock_res_and_lock(dlmlock);
315 if (!ldlm_is_granted(dlmlock)) {
316 dlmlock->l_ast_data = NULL;
317 unlock_res_and_lock(dlmlock);
321 discard = ldlm_is_discard_data(dlmlock);
322 if (dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))
325 if (dlmlock->l_ast_data != NULL) {
326 obj = osc2cl(dlmlock->l_ast_data);
329 unlock_res_and_lock(dlmlock);
331 /* if l_ast_data is NULL, the dlmlock was enqueued by AGL or
332 * the object has been destroyed. */
334 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
336 /* Destroy pages covered by the extent of the DLM lock */
337 result = mdc_lock_flush(env, cl2osc(obj), cl_index(obj, 0),
338 CL_PAGE_EOF, mode, discard);
339 /* Losing a lock, set KMS to 0.
340 * NB: assumed that DOM lock covers whole data on MDT.
342 /* losing a lock, update kms */
343 lock_res_and_lock(dlmlock);
344 dlmlock->l_ast_data = NULL;
345 cl_object_attr_lock(obj);
347 cl_object_attr_update(env, obj, attr, CAT_KMS);
348 cl_object_attr_unlock(obj);
349 unlock_res_and_lock(dlmlock);
350 cl_object_put(env, obj);
355 int mdc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
356 struct ldlm_lock_desc *new, void *data, int reason)
363 case LDLM_CB_BLOCKING: {
364 struct lustre_handle lockh;
366 ldlm_lock2handle(dlmlock, &lockh);
367 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
372 case LDLM_CB_CANCELING: {
377 * This can be called in the context of outer IO, e.g.,
379 * osc_enqueue_base()->...
380 * ->ldlm_prep_elc_req()->...
381 * ->ldlm_cancel_callback()->...
382 * ->osc_ldlm_blocking_ast()
384 * new environment has to be created to not corrupt outer
387 env = cl_env_get(&refcheck);
393 rc = mdc_dlm_canceling(env, dlmlock);
394 cl_env_put(env, &refcheck);
404 * Updates object attributes from a lock value block (lvb) received together
405 * with the DLM lock reply from the server.
406 * This can be optimized to not update attributes when lock is a result of a
409 * Called under lock and resource spin-locks.
411 void mdc_lock_lvb_update(const struct lu_env *env, struct osc_object *osc,
412 struct ldlm_lock *dlmlock, struct ost_lvb *lvb)
414 struct cl_object *obj = osc2cl(osc);
415 struct lov_oinfo *oinfo = osc->oo_oinfo;
416 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
417 unsigned valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME |
419 unsigned int setkms = 0;
424 LASSERT(dlmlock != NULL);
425 lvb = &dlmlock->l_ost_lvb;
427 cl_lvb2attr(attr, lvb);
429 cl_object_attr_lock(obj);
430 if (dlmlock != NULL) {
433 check_res_locked(dlmlock->l_resource);
434 size = lvb->lvb_size;
436 if (size >= oinfo->loi_kms) {
438 attr->cat_kms = size;
441 ldlm_lock_allow_match_locked(dlmlock);
444 /* The size should not be less than the kms */
445 if (attr->cat_size < oinfo->loi_kms)
446 attr->cat_size = oinfo->loi_kms;
448 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
449 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
450 setkms ? "" : " leaving",
451 setkms ? attr->cat_kms : oinfo->loi_kms,
452 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
454 cl_object_attr_update(env, obj, attr, valid);
455 cl_object_attr_unlock(obj);
459 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
460 struct lustre_handle *lockh)
462 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
463 struct ldlm_lock *dlmlock;
467 dlmlock = ldlm_handle2lock_long(lockh, 0);
468 LASSERT(dlmlock != NULL);
470 /* lock reference taken by ldlm_handle2lock_long() is
471 * owned by osc_lock and released in osc_lock_detach()
473 lu_ref_add(&dlmlock->l_reference, "osc_lock", oscl);
474 oscl->ols_has_ref = 1;
476 LASSERT(oscl->ols_dlmlock == NULL);
477 oscl->ols_dlmlock = dlmlock;
479 /* This may be a matched lock for glimpse request, do not hold
480 * lock reference in that case. */
481 if (!oscl->ols_glimpse) {
482 /* hold a refc for non glimpse lock which will
483 * be released in osc_lock_cancel() */
484 lustre_handle_copy(&oscl->ols_handle, lockh);
485 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
489 /* Lock must have been granted. */
490 lock_res_and_lock(dlmlock);
491 if (ldlm_is_granted(dlmlock)) {
492 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
494 /* extend the lock extent, otherwise it will have problem when
495 * we decide whether to grant a lockless lock. */
496 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
497 descr->cld_start = cl_index(descr->cld_obj, 0);
498 descr->cld_end = CL_PAGE_EOF;
500 /* no lvb update for matched lock */
501 if (!ldlm_is_lvb_cached(dlmlock)) {
502 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
503 LASSERT(osc == dlmlock->l_ast_data);
504 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
505 ldlm_set_lvb_cached(dlmlock);
508 unlock_res_and_lock(dlmlock);
510 LASSERT(oscl->ols_state != OLS_GRANTED);
511 oscl->ols_state = OLS_GRANTED;
516 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
517 * received from a server, or after mdc_enqueue_send() matched a local DLM
520 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
523 struct osc_lock *oscl = cookie;
524 struct cl_lock_slice *slice = &oscl->ols_cl;
530 env = cl_env_percpu_get();
531 /* should never happen, similar to osc_ldlm_blocking_ast(). */
532 LASSERT(!IS_ERR(env));
534 rc = ldlm_error2errno(errcode);
535 if (oscl->ols_state == OLS_ENQUEUED) {
536 oscl->ols_state = OLS_UPCALL_RECEIVED;
537 } else if (oscl->ols_state == OLS_CANCELLED) {
540 CERROR("Impossible state: %d\n", oscl->ols_state);
544 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
546 mdc_lock_granted(env, oscl, lockh);
548 /* Error handling, some errors are tolerable. */
549 if (oscl->ols_locklessable && rc == -EUSERS) {
550 /* This is a tolerable error, turn this lock into
553 osc_object_set_contended(cl2osc(slice->cls_obj));
554 LASSERT(slice->cls_ops != oscl->ols_lockless_ops);
556 /* Change this lock to ldlmlock-less lock. */
557 osc_lock_to_lockless(env, oscl, 1);
558 oscl->ols_state = OLS_GRANTED;
560 } else if (oscl->ols_glimpse && rc == -ENAVAIL) {
561 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
562 mdc_lock_lvb_update(env, cl2osc(slice->cls_obj),
563 NULL, &oscl->ols_lvb);
564 /* Hide the error. */
568 if (oscl->ols_owner != NULL)
569 cl_sync_io_note(env, oscl->ols_owner, rc);
570 cl_env_percpu_put(env);
575 /* This is needed only for old servers (before 2.14) support */
576 int mdc_fill_lvb(struct ptlrpc_request *req, struct ost_lvb *lvb)
578 struct mdt_body *body;
580 /* get LVB data from mdt_body otherwise */
581 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
585 if (!(body->mbo_valid & OBD_MD_DOM_SIZE))
588 mdc_body2lvb(body, lvb);
592 int mdc_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
593 osc_enqueue_upcall_f upcall, void *cookie,
594 struct lustre_handle *lockh, enum ldlm_mode mode,
595 __u64 *flags, int errcode)
597 struct osc_lock *ols = cookie;
598 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
603 /* needed only for glimpse from an old server (< 2.14) */
604 if (glimpse && !exp_connect_dom_lvb(exp))
605 rc = mdc_fill_lvb(req, &ols->ols_lvb);
607 if (glimpse && errcode == ELDLM_LOCK_ABORTED) {
608 struct ldlm_reply *rep;
610 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
612 rep->lock_policy_res2 =
613 ptlrpc_status_ntoh(rep->lock_policy_res2);
614 if (rep->lock_policy_res2)
615 errcode = rep->lock_policy_res2;
619 *flags |= LDLM_FL_LVB_READY;
620 } else if (errcode == ELDLM_OK) {
621 struct ldlm_lock *lock;
623 /* Callers have references, should be valid always */
624 lock = ldlm_handle2lock(lockh);
626 /* At this point ols_lvb must be filled with correct LVB either
627 * by mdc_fill_lvb() above or by ldlm_cli_enqueue_fini().
628 * DoM uses l_ost_lvb to store LVB data, so copy it here from
629 * just updated ols_lvb.
631 lock_res_and_lock(lock);
632 memcpy(&lock->l_ost_lvb, &ols->ols_lvb,
633 sizeof(lock->l_ost_lvb));
634 unlock_res_and_lock(lock);
636 *flags |= LDLM_FL_LVB_READY;
639 /* Call the update callback. */
640 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
642 /* release the reference taken in ldlm_cli_enqueue() */
643 if (errcode == ELDLM_LOCK_MATCHED)
645 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
646 ldlm_lock_decref(lockh, mode);
651 int mdc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
654 struct osc_enqueue_args *aa = args;
655 struct ldlm_lock *lock;
656 struct lustre_handle *lockh = &aa->oa_lockh;
657 enum ldlm_mode mode = aa->oa_mode;
658 struct ldlm_enqueue_info einfo = {
659 .ei_type = aa->oa_type,
665 LASSERT(!aa->oa_speculative);
667 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
669 lock = ldlm_handle2lock(lockh);
670 LASSERTF(lock != NULL,
671 "lockh %#llx, req %p, aa %p - client evicted?\n",
672 lockh->cookie, req, aa);
674 /* Take an additional reference so that a blocking AST that
675 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
676 * to arrive after an upcall has been executed by
677 * mdc_enqueue_fini().
679 ldlm_lock_addref(lockh, mode);
681 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
682 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
684 /* Let CP AST to grant the lock first. */
685 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
687 /* Complete obtaining the lock procedure. */
688 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, &einfo, 1, aa->oa_flags,
689 aa->oa_lvb, aa->oa_lvb ?
690 sizeof(*aa->oa_lvb) : 0, lockh, rc);
691 /* Complete mdc stuff. */
692 rc = mdc_enqueue_fini(aa->oa_exp, req, aa->oa_upcall, aa->oa_cookie,
693 lockh, mode, aa->oa_flags, rc);
695 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
697 ldlm_lock_decref(lockh, mode);
702 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
703 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
704 * other synchronous requests, however keeping some locks and trying to obtain
705 * others may take a considerable amount of time in a case of ost failure; and
706 * when other sync requests do not get released lock from a client, the client
707 * is excluded from the cluster -- such scenarious make the life difficult, so
708 * release locks just after they are obtained. */
709 int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
710 struct ldlm_res_id *res_id, __u64 *flags,
711 union ldlm_policy_data *policy, struct ost_lvb *lvb,
712 osc_enqueue_upcall_f upcall, void *cookie,
713 struct ldlm_enqueue_info *einfo, int async)
715 struct obd_device *obd = exp->exp_obd;
716 struct lustre_handle lockh = { 0 };
717 struct ptlrpc_request *req = NULL;
718 struct ldlm_intent *lit;
720 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
721 __u64 match_flags = *flags;
725 bool compat_glimpse = glimpse && !exp_connect_dom_lvb(exp);
729 mode = einfo->ei_mode;
730 if (einfo->ei_mode == LCK_PR)
733 match_flags |= LDLM_FL_LVB_READY;
735 match_flags |= LDLM_FL_BLOCK_GRANTED;
736 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
737 einfo->ei_type, policy, mode, &lockh);
739 struct ldlm_lock *matched;
741 if (*flags & LDLM_FL_TEST_LOCK)
744 matched = ldlm_handle2lock(&lockh);
746 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
747 ldlm_set_kms_ignore(matched);
749 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
750 *flags |= LDLM_FL_LVB_READY;
752 /* We already have a lock, and it's referenced. */
753 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
755 ldlm_lock_decref(&lockh, mode);
756 LDLM_LOCK_PUT(matched);
759 ldlm_lock_decref(&lockh, mode);
760 LDLM_LOCK_PUT(matched);
763 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
766 /* Glimpse is intent on old server */
767 req = ptlrpc_request_alloc(class_exp2cliimp(exp), compat_glimpse ?
768 &RQF_LDLM_INTENT : &RQF_LDLM_ENQUEUE);
772 /* For WRITE lock cancel other locks on resource early if any */
773 if (einfo->ei_mode & LCK_PW)
774 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
780 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
782 ptlrpc_request_free(req);
786 if (compat_glimpse) {
787 /* pack the glimpse intent */
788 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
789 lit->opc = IT_GLIMPSE;
792 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
793 *flags &= ~LDLM_FL_BLOCK_GRANTED;
795 if (compat_glimpse) {
796 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
797 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
800 lvb_size = sizeof(*lvb);
801 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
804 ptlrpc_request_set_replen(req);
806 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
807 lvb_size, LVB_T_OST, &lockh, async);
810 struct osc_enqueue_args *aa;
812 aa = ptlrpc_req_async_args(aa, req);
814 aa->oa_mode = einfo->ei_mode;
815 aa->oa_type = einfo->ei_type;
816 lustre_handle_copy(&aa->oa_lockh, &lockh);
817 aa->oa_upcall = upcall;
818 aa->oa_cookie = cookie;
819 aa->oa_speculative = false;
820 aa->oa_flags = flags;
821 aa->oa_lvb = compat_glimpse ? NULL : lvb;
823 req->rq_interpret_reply = mdc_enqueue_interpret;
824 ptlrpcd_add_req(req);
826 ptlrpc_req_finished(req);
831 rc = mdc_enqueue_fini(exp, req, upcall, cookie, &lockh, einfo->ei_mode,
833 ptlrpc_req_finished(req);
838 * Implementation of cl_lock_operations::clo_enqueue() method for osc
839 * layer. This initiates ldlm enqueue:
841 * - cancels conflicting locks early (osc_lock_enqueue_wait());
843 * - calls osc_enqueue_base() to do actual enqueue.
845 * osc_enqueue_base() is supplied with an upcall function that is executed
846 * when lock is received either after a local cached ldlm lock is matched, or
847 * when a reply from the server is received.
849 * This function does not wait for the network communication to complete.
851 static int mdc_lock_enqueue(const struct lu_env *env,
852 const struct cl_lock_slice *slice,
853 struct cl_io *unused, struct cl_sync_io *anchor)
855 struct osc_thread_info *info = osc_env_info(env);
856 struct osc_io *oio = osc_env_io(env);
857 struct osc_object *osc = cl2osc(slice->cls_obj);
858 struct osc_lock *oscl = cl2osc_lock(slice);
859 struct cl_lock *lock = slice->cls_lock;
860 struct ldlm_res_id *resname = &info->oti_resname;
861 union ldlm_policy_data *policy = &info->oti_policy;
862 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
863 void *cookie = (void *)oscl;
869 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
870 "lock = %p, ols = %p\n", lock, oscl);
872 if (oscl->ols_state == OLS_GRANTED)
875 /* Lockahead is not supported on MDT yet */
876 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
877 result = -EOPNOTSUPP;
881 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
882 GOTO(enqueue_base, 0);
884 if (oscl->ols_glimpse) {
885 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
887 GOTO(enqueue_base, 0);
890 result = osc_lock_enqueue_wait(env, osc, oscl);
894 /* we can grant lockless lock right after all conflicting locks
896 if (osc_lock_is_lockless(oscl)) {
897 oscl->ols_state = OLS_GRANTED;
898 oio->oi_lockless = 1;
903 oscl->ols_state = OLS_ENQUEUED;
904 if (anchor != NULL) {
905 atomic_inc(&anchor->csi_sync_nr);
906 oscl->ols_owner = anchor;
910 * DLM lock's ast data must be osc_object;
911 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
914 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
915 mdc_lock_build_policy(env, lock, policy);
916 LASSERT(!oscl->ols_speculative);
917 result = mdc_enqueue_send(env, osc_export(osc), resname,
918 &oscl->ols_flags, policy, &oscl->ols_lvb,
919 upcall, cookie, &oscl->ols_einfo, async);
921 if (osc_lock_is_lockless(oscl)) {
922 oio->oi_lockless = 1;
924 LASSERT(oscl->ols_state == OLS_GRANTED);
925 LASSERT(oscl->ols_hold);
926 LASSERT(oscl->ols_dlmlock != NULL);
931 oscl->ols_state = OLS_CANCELLED;
932 osc_lock_wake_waiters(env, osc, oscl);
935 cl_sync_io_note(env, anchor, result);
940 static const struct cl_lock_operations mdc_lock_lockless_ops = {
941 .clo_fini = osc_lock_fini,
942 .clo_enqueue = mdc_lock_enqueue,
943 .clo_cancel = mdc_lock_lockless_cancel,
944 .clo_print = osc_lock_print
947 static const struct cl_lock_operations mdc_lock_ops = {
948 .clo_fini = osc_lock_fini,
949 .clo_enqueue = mdc_lock_enqueue,
950 .clo_cancel = osc_lock_cancel,
951 .clo_print = osc_lock_print,
954 int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
955 struct cl_lock *lock, const struct cl_io *io)
957 struct osc_lock *ols;
958 __u32 enqflags = lock->cll_descr.cld_enq_flags;
959 __u64 flags = osc_enq2ldlm_flags(enqflags);
963 /* Ignore AGL for Data-on-MDT, stat returns size data */
964 if ((enqflags & CEF_SPECULATIVE) != 0)
967 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
968 if (unlikely(ols == NULL))
971 ols->ols_state = OLS_NEW;
972 spin_lock_init(&ols->ols_lock);
973 INIT_LIST_HEAD(&ols->ols_waiting_list);
974 INIT_LIST_HEAD(&ols->ols_wait_entry);
975 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
976 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
978 ols->ols_flags = flags;
979 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
980 if (lock->cll_descr.cld_mode == CLM_GROUP)
981 ols->ols_flags |= LDLM_FL_ATOMIC_CB;
983 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
984 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
985 ols->ols_glimpse = 1;
987 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
989 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
991 if (!(enqflags & CEF_MUST))
992 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
993 if (ols->ols_locklessable && !(enqflags & CEF_DISCARD_DATA))
994 ols->ols_flags |= LDLM_FL_DENY_ON_CONTENTION;
996 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
997 osc_lock_set_writer(env, io, obj, ols);
999 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx\n",
1000 lock, ols, ols->ols_flags);
1007 * An implementation of cl_io_operations specific methods for MDC layer.
1010 static int mdc_async_upcall(void *a, int rc)
1012 struct osc_async_cbargs *args = a;
1015 complete(&args->opc_sync);
1019 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
1020 pgoff_t index, struct lustre_handle *lh)
1022 struct ldlm_lock *lock;
1024 /* find DOM lock protecting object */
1025 lock = mdc_dlmlock_at_pgoff(env, osc, index,
1026 OSC_DAP_FL_TEST_LOCK |
1027 OSC_DAP_FL_CANCELING);
1029 struct ldlm_resource *res;
1030 struct ldlm_res_id *resname;
1032 resname = &osc_env_info(env)->oti_resname;
1033 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1034 res = ldlm_resource_get(osc_export(osc)->exp_obd->obd_namespace,
1035 NULL, resname, LDLM_IBITS, 0);
1036 ldlm_resource_dump(D_ERROR, res);
1037 libcfs_debug_dumpstack(NULL);
1040 *lh = lock->l_remote_handle;
1041 LDLM_LOCK_PUT(lock);
1046 static int mdc_io_setattr_start(const struct lu_env *env,
1047 const struct cl_io_slice *slice)
1049 struct cl_io *io = slice->cis_io;
1050 struct osc_io *oio = cl2osc_io(env, slice);
1051 struct cl_object *obj = slice->cis_obj;
1052 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1053 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1054 struct obdo *oa = &oio->oi_oa;
1055 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1056 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1057 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1058 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1061 /* silently ignore non-truncate setattr for Data-on-MDT object */
1062 if (cl_io_is_trunc(io)) {
1063 /* truncate cache dirty pages first */
1064 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1070 if (cl_io_is_fallocate(io))
1073 if (oio->oi_lockless == 0) {
1074 cl_object_attr_lock(obj);
1075 rc = cl_object_attr_get(env, obj, attr);
1077 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1078 unsigned int cl_valid = 0;
1080 if (ia_avalid & ATTR_SIZE) {
1081 attr->cat_size = size;
1082 attr->cat_kms = size;
1083 cl_valid = (CAT_SIZE | CAT_KMS);
1085 if (ia_avalid & ATTR_MTIME_SET) {
1086 attr->cat_mtime = lvb->lvb_mtime;
1087 cl_valid |= CAT_MTIME;
1089 if (ia_avalid & ATTR_ATIME_SET) {
1090 attr->cat_atime = lvb->lvb_atime;
1091 cl_valid |= CAT_ATIME;
1093 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1094 attr->cat_ctime = lvb->lvb_ctime;
1095 cl_valid |= CAT_CTIME;
1097 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1099 cl_object_attr_unlock(obj);
1104 if (!(ia_avalid & ATTR_SIZE))
1107 memset(oa, 0, sizeof(*oa));
1108 oa->o_oi = loi->loi_oi;
1109 oa->o_mtime = attr->cat_mtime;
1110 oa->o_atime = attr->cat_atime;
1111 oa->o_ctime = attr->cat_ctime;
1114 oa->o_blocks = OBD_OBJECT_EOF;
1115 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1116 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1118 if (oio->oi_lockless) {
1119 oa->o_flags = OBD_FL_SRVLOCK;
1120 oa->o_valid |= OBD_MD_FLFLAGS;
1122 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1125 oa->o_valid |= OBD_MD_FLHANDLE;
1128 init_completion(&cbargs->opc_sync);
1130 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1131 mdc_async_upcall, cbargs);
1132 cbargs->opc_rpc_sent = rc == 0;
1136 static int mdc_io_read_ahead(const struct lu_env *env,
1137 const struct cl_io_slice *ios,
1138 pgoff_t start, struct cl_read_ahead *ra)
1140 struct osc_object *osc = cl2osc(ios->cis_obj);
1141 struct osc_io *oio = cl2osc_io(env, ios);
1142 struct ldlm_lock *dlmlock;
1146 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1147 if (dlmlock == NULL)
1150 oio->oi_is_readahead = 1;
1151 if (dlmlock->l_req_mode != LCK_PR) {
1152 struct lustre_handle lockh;
1154 ldlm_lock2handle(dlmlock, &lockh);
1155 ldlm_lock_addref(&lockh, LCK_PR);
1156 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1159 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1160 ra->cra_end_idx = CL_PAGE_EOF;
1161 ra->cra_release = osc_read_ahead_release;
1162 ra->cra_dlmlock = dlmlock;
1168 int mdc_io_fsync_start(const struct lu_env *env,
1169 const struct cl_io_slice *slice)
1171 struct cl_io *io = slice->cis_io;
1172 struct cl_fsync_io *fio = &io->u.ci_fsync;
1173 struct cl_object *obj = slice->cis_obj;
1174 struct osc_object *osc = cl2osc(obj);
1179 /* a MDC lock always covers whole object, do sync for whole
1180 * possible range despite of supplied start/end values.
1182 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1183 fio->fi_mode == CL_FSYNC_DISCARD);
1185 fio->fi_nr_written += result;
1188 if (fio->fi_mode == CL_FSYNC_ALL) {
1191 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1194 /* Use OSC sync code because it is asynchronous.
1195 * It is to be added into MDC and avoid the using of
1196 * OST_SYNC at both MDC and MDT.
1198 rc = osc_fsync_ost(env, osc, fio);
1206 struct mdc_data_version_args {
1207 struct osc_io *dva_oio;
1211 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1214 struct mdc_data_version_args *dva = args;
1215 struct osc_io *oio = dva->dva_oio;
1216 const struct mdt_body *body;
1222 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1224 GOTO(out, rc = -EPROTO);
1226 /* Prepare OBDO from mdt_body for CLIO */
1227 oio->oi_oa.o_valid = body->mbo_valid;
1228 oio->oi_oa.o_flags = body->mbo_flags;
1229 oio->oi_oa.o_data_version = body->mbo_version;
1230 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1233 oio->oi_cbarg.opc_rc = rc;
1234 complete(&oio->oi_cbarg.opc_sync);
1238 static int mdc_io_data_version_start(const struct lu_env *env,
1239 const struct cl_io_slice *slice)
1241 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1242 struct osc_io *oio = cl2osc_io(env, slice);
1243 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1244 struct osc_object *obj = cl2osc(slice->cis_obj);
1245 struct obd_export *exp = osc_export(obj);
1246 struct ptlrpc_request *req;
1247 struct mdt_body *body;
1248 struct mdc_data_version_args *dva;
1253 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1254 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1255 oio->oi_oa.o_valid = OBD_MD_FLID;
1257 init_completion(&cbargs->opc_sync);
1259 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1263 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1265 ptlrpc_request_free(req);
1269 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1270 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1271 body->mbo_valid = OBD_MD_FLID;
1272 /* Indicate that data version is needed */
1273 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1274 body->mbo_flags = 0;
1276 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1277 body->mbo_valid |= OBD_MD_FLFLAGS;
1278 body->mbo_flags |= OBD_FL_SRVLOCK;
1279 if (dv->dv_flags & LL_DV_WR_FLUSH)
1280 body->mbo_flags |= OBD_FL_FLUSH;
1283 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1284 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1285 ptlrpc_request_set_replen(req);
1287 req->rq_interpret_reply = mdc_data_version_interpret;
1288 dva = ptlrpc_req_async_args(dva, req);
1291 ptlrpcd_add_req(req);
1296 static void mdc_io_data_version_end(const struct lu_env *env,
1297 const struct cl_io_slice *slice)
1299 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1300 struct osc_io *oio = cl2osc_io(env, slice);
1301 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1304 wait_for_completion(&cbargs->opc_sync);
1306 if (cbargs->opc_rc != 0) {
1307 slice->cis_io->ci_result = cbargs->opc_rc;
1309 slice->cis_io->ci_result = 0;
1310 if (!(oio->oi_oa.o_valid &
1311 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1312 slice->cis_io->ci_result = -ENOTSUPP;
1314 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1315 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1316 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1317 dv->dv_data_version = oio->oi_oa.o_data_version;
1323 static const struct cl_io_operations mdc_io_ops = {
1326 .cio_iter_init = osc_io_iter_init,
1327 .cio_iter_fini = osc_io_rw_iter_fini,
1328 .cio_start = osc_io_read_start,
1331 .cio_iter_init = osc_io_iter_init,
1332 .cio_iter_fini = osc_io_rw_iter_fini,
1333 .cio_start = osc_io_write_start,
1334 .cio_end = osc_io_end,
1337 .cio_iter_init = osc_io_iter_init,
1338 .cio_iter_fini = osc_io_iter_fini,
1339 .cio_start = mdc_io_setattr_start,
1340 .cio_end = osc_io_setattr_end,
1342 [CIT_DATA_VERSION] = {
1343 .cio_start = mdc_io_data_version_start,
1344 .cio_end = mdc_io_data_version_end,
1347 .cio_iter_init = osc_io_iter_init,
1348 .cio_iter_fini = osc_io_iter_fini,
1349 .cio_start = osc_io_fault_start,
1350 .cio_end = osc_io_end,
1353 .cio_start = mdc_io_fsync_start,
1354 .cio_end = osc_io_fsync_end,
1357 .cio_start = osc_io_lseek_start,
1358 .cio_end = osc_io_lseek_end,
1361 .cio_read_ahead = mdc_io_read_ahead,
1362 .cio_lru_reserve = osc_io_lru_reserve,
1363 .cio_submit = osc_io_submit,
1364 .cio_commit_async = osc_io_commit_async,
1365 .cio_extent_release = osc_io_extent_release,
1368 int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1371 struct osc_io *oio = osc_env_io(env);
1373 CL_IO_SLICE_CLEAN(oio, oi_cl);
1374 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1378 static void mdc_build_res_name(struct osc_object *osc,
1379 struct ldlm_res_id *resname)
1381 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1385 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1386 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1389 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1390 struct cl_req_attr *attr)
1392 u64 flags = attr->cra_flags;
1394 /* Copy object FID to cl_attr */
1395 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1397 if (flags & OBD_MD_FLGROUP)
1398 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1400 if (flags & OBD_MD_FLID)
1401 attr->cra_oa->o_valid |= OBD_MD_FLID;
1403 if (flags & OBD_MD_FLHANDLE) {
1404 struct osc_page *opg;
1406 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1407 if (!opg->ops_srvlock) {
1410 rc = mdc_get_lock_handle(env, cl2osc(obj),
1412 &attr->cra_oa->o_handle);
1414 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1415 "uncovered page!\n");
1418 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1424 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1425 struct cl_attr *attr)
1427 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1429 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1430 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1432 return osc_attr_get(env, obj, attr);
1435 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1437 struct osc_object *osc = (struct osc_object *)data;
1438 struct ost_lvb *lvb = &lock->l_ost_lvb;
1439 struct lov_oinfo *oinfo;
1442 if (lock->l_ast_data == data) {
1443 lock->l_ast_data = NULL;
1445 LASSERT(osc != NULL);
1446 LASSERT(osc->oo_oinfo != NULL);
1447 LASSERT(lvb != NULL);
1449 /* Updates lvb in lock by the cached oinfo */
1450 oinfo = osc->oo_oinfo;
1452 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1453 "%llu %llu %llu by oinfo size %llu blocks %llu "
1454 "[cma]time %llu %llu %llu", lvb->lvb_size,
1455 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1456 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1457 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1458 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1459 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1461 cl_object_attr_lock(&osc->oo_cl);
1462 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1463 cl_object_attr_unlock(&osc->oo_cl);
1464 ldlm_clear_lvb_cached(lock);
1466 RETURN(LDLM_ITER_CONTINUE);
1469 int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1471 struct osc_object *osc = cl2osc(obj);
1472 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1474 /* DLM locks don't hold a reference of osc_object so we have to
1475 * clear it before the object is being destroyed. */
1476 osc_build_res_name(osc, resname);
1477 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1478 mdc_object_ast_clear, osc);
1482 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1483 struct ldlm_lock *lock)
1485 /* if lock cancel is initiated from llite then it is combined
1486 * lock with DOM bit and it may have no l_ast_data initialized yet,
1487 * so init it here with given osc_object.
1489 mdc_set_dom_lock_data(lock, cl2osc(obj));
1490 RETURN(mdc_dlm_canceling(env, lock));
1493 static const struct cl_object_operations mdc_ops = {
1494 .coo_page_init = osc_page_init,
1495 .coo_lock_init = mdc_lock_init,
1496 .coo_io_init = mdc_io_init,
1497 .coo_attr_get = mdc_attr_get,
1498 .coo_attr_update = osc_attr_update,
1499 .coo_glimpse = osc_object_glimpse,
1500 .coo_req_attr_set = mdc_req_attr_set,
1501 .coo_prune = mdc_object_prune,
1502 .coo_object_flush = mdc_object_flush
1505 static const struct osc_object_operations mdc_object_ops = {
1506 .oto_build_res_name = mdc_build_res_name,
1507 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1510 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1511 const struct lu_object_conf *conf)
1513 struct osc_object *osc = lu2osc(obj);
1515 if (osc->oo_initialized)
1518 osc->oo_initialized = true;
1520 return osc_object_init(env, obj, conf);
1523 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1525 osc_object_free(env, obj);
1528 static const struct lu_object_operations mdc_lu_obj_ops = {
1529 .loo_object_init = mdc_object_init,
1530 .loo_object_delete = NULL,
1531 .loo_object_release = NULL,
1532 .loo_object_free = mdc_object_free,
1533 .loo_object_print = osc_object_print,
1534 .loo_object_invariant = NULL
1537 struct lu_object *mdc_object_alloc(const struct lu_env *env,
1538 const struct lu_object_header *unused,
1539 struct lu_device *dev)
1541 struct osc_object *osc;
1542 struct lu_object *obj;
1544 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1547 lu_object_init(obj, NULL, dev);
1548 osc->oo_cl.co_ops = &mdc_ops;
1549 obj->lo_ops = &mdc_lu_obj_ops;
1550 osc->oo_obj_ops = &mdc_object_ops;
1551 osc->oo_initialized = false;
1558 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1559 struct lustre_cfg *cfg)
1561 size_t count = class_modify_config(cfg, PARAM_MDC,
1562 &d->ld_obd->obd_kset.kobj);
1563 return count > 0 ? 0 : count;
1566 const struct lu_device_operations mdc_lu_ops = {
1567 .ldo_object_alloc = mdc_object_alloc,
1568 .ldo_process_config = mdc_process_config,
1569 .ldo_recovery_complete = NULL,
1572 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1573 struct lu_device_type *t,
1574 struct lustre_cfg *cfg)
1576 struct lu_device *d;
1577 struct osc_device *od;
1578 struct obd_device *obd;
1583 RETURN(ERR_PTR(-ENOMEM));
1585 cl_device_init(&od->od_cl, t);
1587 d->ld_ops = &mdc_lu_ops;
1590 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1592 RETURN(ERR_PTR(-ENODEV));
1594 rc = mdc_setup(obd, cfg);
1596 osc_device_free(env, d);
1597 RETURN(ERR_PTR(rc));
1599 od->od_exp = obd->obd_self_export;
1603 static const struct lu_device_type_operations mdc_device_type_ops = {
1604 .ldto_device_alloc = mdc_device_alloc,
1605 .ldto_device_free = osc_device_free,
1606 .ldto_device_init = osc_device_init,
1607 .ldto_device_fini = osc_device_fini
1610 struct lu_device_type mdc_device_type = {
1611 .ldt_tags = LU_DEVICE_CL,
1612 .ldt_name = LUSTRE_MDC_NAME,
1613 .ldt_ops = &mdc_device_type_ops,
1614 .ldt_ctx_tags = LCT_CL_THREAD