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 */
71 static void mdc_lock_lvb_update(const struct lu_env *env,
72 struct osc_object *osc,
73 struct ldlm_lock *dlmlock,
76 static int mdc_set_dom_lock_data(struct ldlm_lock *lock, void *data)
80 LASSERT(lock != NULL);
81 LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
83 lock_res_and_lock(lock);
85 if (lock->l_ast_data == NULL)
86 lock->l_ast_data = data;
87 if (lock->l_ast_data == data)
90 unlock_res_and_lock(lock);
95 int mdc_dom_lock_match(const struct lu_env *env, struct obd_export *exp,
96 struct ldlm_res_id *res_id, enum ldlm_type type,
97 union ldlm_policy_data *policy, enum ldlm_mode mode,
98 __u64 *flags, struct osc_object *obj,
99 struct lustre_handle *lockh,
100 enum ldlm_match_flags match_flags)
102 struct obd_device *obd = exp->exp_obd;
103 __u64 lflags = *flags;
108 rc = ldlm_lock_match_with_skip(obd->obd_namespace, lflags, 0,
109 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 void **pvec, int count, void *cbdata)
191 struct osc_thread_info *info = osc_env_info(env);
192 struct osc_object *osc = cbdata;
196 for (i = 0; i < count; i++) {
197 struct osc_page *ops = pvec[i];
199 index = osc_index(ops);
200 if (index >= info->oti_fn_index) {
201 struct ldlm_lock *tmp;
202 struct cl_page *page = ops->ops_cl.cpl_page;
204 /* refresh non-overlapped index */
205 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
206 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_AST);
208 info->oti_fn_index = CL_PAGE_EOF;
210 } else if (cl_page_own(env, io, page) == 0) {
211 /* discard the page */
212 cl_page_discard(env, io, page);
213 cl_page_disown(env, io, page);
215 LASSERT(page->cp_state == CPS_FREEING);
219 info->oti_next_index = index + 1;
225 * Discard pages protected by the given lock. This function traverses radix
226 * tree to find all covering pages and discard them. If a page is being covered
227 * by other locks, it should remain in cache.
229 * If error happens on any step, the process continues anyway (the reasoning
230 * behind this being that lock cancellation cannot be delayed indefinitely).
232 static int mdc_lock_discard_pages(const struct lu_env *env,
233 struct osc_object *osc,
234 pgoff_t start, pgoff_t end,
237 struct osc_thread_info *info = osc_env_info(env);
238 struct cl_io *io = &info->oti_io;
239 osc_page_gang_cbt cb;
244 io->ci_obj = cl_object_top(osc2cl(osc));
245 io->ci_ignore_layout = 1;
246 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
250 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
251 info->oti_fn_index = info->oti_next_index = start;
253 osc_page_gang_lookup(env, io, osc, info->oti_next_index,
254 end, cb, (void *)osc);
260 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
261 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
269 if (mode == CLM_WRITE) {
270 result = osc_cache_writeback_range(env, obj, start, end, 1,
272 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
273 obj, start, end, result,
274 discard ? "discarded" : "written back");
279 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
280 rc = mdc_lock_discard_pages(env, obj, start, end,
281 mode == CLM_WRITE || discard);
282 if (result == 0 && rc < 0)
288 void mdc_lock_lockless_cancel(const struct lu_env *env,
289 const struct cl_lock_slice *slice)
291 struct osc_lock *ols = cl2osc_lock(slice);
292 struct osc_object *osc = cl2osc(slice->cls_obj);
293 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
296 LASSERT(ols->ols_dlmlock == NULL);
297 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
300 CERROR("Pages for lockless lock %p were not purged(%d)\n",
303 osc_lock_wake_waiters(env, osc, ols);
307 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
308 * and ldlm_lock caches.
310 static int mdc_dlm_canceling(const struct lu_env *env,
311 struct ldlm_lock *dlmlock)
313 struct cl_object *obj = NULL;
316 enum cl_lock_mode mode = CLM_READ;
320 lock_res_and_lock(dlmlock);
321 if (!ldlm_is_granted(dlmlock)) {
322 dlmlock->l_ast_data = NULL;
323 unlock_res_and_lock(dlmlock);
327 discard = ldlm_is_discard_data(dlmlock);
328 if (dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))
331 if (dlmlock->l_ast_data != NULL) {
332 obj = osc2cl(dlmlock->l_ast_data);
335 unlock_res_and_lock(dlmlock);
337 /* if l_ast_data is NULL, the dlmlock was enqueued by AGL or
338 * the object has been destroyed. */
340 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
342 /* Destroy pages covered by the extent of the DLM lock */
343 result = mdc_lock_flush(env, cl2osc(obj), cl_index(obj, 0),
344 CL_PAGE_EOF, mode, discard);
345 /* Losing a lock, set KMS to 0.
346 * NB: assumed that DOM lock covers whole data on MDT.
348 /* losing a lock, update kms */
349 lock_res_and_lock(dlmlock);
350 dlmlock->l_ast_data = NULL;
351 cl_object_attr_lock(obj);
353 cl_object_attr_update(env, obj, attr, CAT_KMS);
354 cl_object_attr_unlock(obj);
355 unlock_res_and_lock(dlmlock);
356 cl_object_put(env, obj);
361 int mdc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
362 struct ldlm_lock_desc *new, void *data, int reason)
369 case LDLM_CB_BLOCKING: {
370 struct lustre_handle lockh;
372 ldlm_lock2handle(dlmlock, &lockh);
373 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
378 case LDLM_CB_CANCELING: {
383 * This can be called in the context of outer IO, e.g.,
385 * osc_enqueue_base()->...
386 * ->ldlm_prep_elc_req()->...
387 * ->ldlm_cancel_callback()->...
388 * ->osc_ldlm_blocking_ast()
390 * new environment has to be created to not corrupt outer
393 env = cl_env_get(&refcheck);
399 rc = mdc_dlm_canceling(env, dlmlock);
400 cl_env_put(env, &refcheck);
410 * Updates object attributes from a lock value block (lvb) received together
411 * with the DLM lock reply from the server.
412 * This can be optimized to not update attributes when lock is a result of a
415 * Called under lock and resource spin-locks.
417 void mdc_lock_lvb_update(const struct lu_env *env, struct osc_object *osc,
418 struct ldlm_lock *dlmlock, struct ost_lvb *lvb)
420 struct cl_object *obj = osc2cl(osc);
421 struct lov_oinfo *oinfo = osc->oo_oinfo;
422 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
423 unsigned valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME |
425 unsigned int setkms = 0;
430 LASSERT(dlmlock != NULL);
431 lvb = &dlmlock->l_ost_lvb;
433 cl_lvb2attr(attr, lvb);
435 cl_object_attr_lock(obj);
436 if (dlmlock != NULL) {
439 check_res_locked(dlmlock->l_resource);
440 size = lvb->lvb_size;
442 if (size >= oinfo->loi_kms) {
444 attr->cat_kms = size;
447 ldlm_lock_allow_match_locked(dlmlock);
450 /* The size should not be less than the kms */
451 if (attr->cat_size < oinfo->loi_kms)
452 attr->cat_size = oinfo->loi_kms;
454 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
455 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
456 setkms ? "" : " leaving",
457 setkms ? attr->cat_kms : oinfo->loi_kms,
458 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
460 cl_object_attr_update(env, obj, attr, valid);
461 cl_object_attr_unlock(obj);
465 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
466 struct lustre_handle *lockh)
468 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
469 struct ldlm_lock *dlmlock;
473 dlmlock = ldlm_handle2lock_long(lockh, 0);
474 LASSERT(dlmlock != NULL);
476 /* lock reference taken by ldlm_handle2lock_long() is
477 * owned by osc_lock and released in osc_lock_detach()
479 lu_ref_add(&dlmlock->l_reference, "osc_lock", oscl);
480 oscl->ols_has_ref = 1;
482 LASSERT(oscl->ols_dlmlock == NULL);
483 oscl->ols_dlmlock = dlmlock;
485 /* This may be a matched lock for glimpse request, do not hold
486 * lock reference in that case. */
487 if (!oscl->ols_glimpse) {
488 /* hold a refc for non glimpse lock which will
489 * be released in osc_lock_cancel() */
490 lustre_handle_copy(&oscl->ols_handle, lockh);
491 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
495 /* Lock must have been granted. */
496 lock_res_and_lock(dlmlock);
497 if (ldlm_is_granted(dlmlock)) {
498 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
500 /* extend the lock extent, otherwise it will have problem when
501 * we decide whether to grant a lockless lock. */
502 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
503 descr->cld_start = cl_index(descr->cld_obj, 0);
504 descr->cld_end = CL_PAGE_EOF;
506 /* no lvb update for matched lock */
507 if (!ldlm_is_lvb_cached(dlmlock)) {
508 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
509 LASSERT(osc == dlmlock->l_ast_data);
510 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
511 ldlm_set_lvb_cached(dlmlock);
514 unlock_res_and_lock(dlmlock);
516 LASSERT(oscl->ols_state != OLS_GRANTED);
517 oscl->ols_state = OLS_GRANTED;
522 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
523 * received from a server, or after mdc_enqueue_send() matched a local DLM
526 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
529 struct osc_lock *oscl = cookie;
530 struct cl_lock_slice *slice = &oscl->ols_cl;
536 env = cl_env_percpu_get();
537 /* should never happen, similar to osc_ldlm_blocking_ast(). */
538 LASSERT(!IS_ERR(env));
540 rc = ldlm_error2errno(errcode);
541 if (oscl->ols_state == OLS_ENQUEUED) {
542 oscl->ols_state = OLS_UPCALL_RECEIVED;
543 } else if (oscl->ols_state == OLS_CANCELLED) {
546 CERROR("Impossible state: %d\n", oscl->ols_state);
550 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
552 mdc_lock_granted(env, oscl, lockh);
554 /* Error handling, some errors are tolerable. */
555 if (oscl->ols_glimpse && rc == -ENAVAIL) {
556 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
557 mdc_lock_lvb_update(env, cl2osc(slice->cls_obj),
558 NULL, &oscl->ols_lvb);
559 /* Hide the error. */
563 if (oscl->ols_owner != NULL)
564 cl_sync_io_note(env, oscl->ols_owner, rc);
565 cl_env_percpu_put(env);
570 /* This is needed only for old servers (before 2.14) support */
571 int mdc_fill_lvb(struct req_capsule *pill, struct ost_lvb *lvb)
573 struct mdt_body *body;
575 /* get LVB data from mdt_body otherwise */
576 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
580 if (!(body->mbo_valid & OBD_MD_DOM_SIZE))
583 mdc_body2lvb(body, lvb);
587 int mdc_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
588 osc_enqueue_upcall_f upcall, void *cookie,
589 struct lustre_handle *lockh, enum ldlm_mode mode,
590 __u64 *flags, int errcode)
592 struct osc_lock *ols = cookie;
593 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
598 /* needed only for glimpse from an old server (< 2.14) */
599 if (glimpse && !exp_connect_dom_lvb(exp))
600 rc = mdc_fill_lvb(&req->rq_pill, &ols->ols_lvb);
602 if (glimpse && errcode == ELDLM_LOCK_ABORTED) {
603 struct ldlm_reply *rep;
605 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
607 rep->lock_policy_res2 =
608 ptlrpc_status_ntoh(rep->lock_policy_res2);
609 if (rep->lock_policy_res2)
610 errcode = rep->lock_policy_res2;
614 *flags |= LDLM_FL_LVB_READY;
615 } else if (errcode == ELDLM_OK) {
616 struct ldlm_lock *lock;
618 /* Callers have references, should be valid always */
619 lock = ldlm_handle2lock(lockh);
621 /* At this point ols_lvb must be filled with correct LVB either
622 * by mdc_fill_lvb() above or by ldlm_cli_enqueue_fini().
623 * DoM uses l_ost_lvb to store LVB data, so copy it here from
624 * just updated ols_lvb.
626 lock_res_and_lock(lock);
627 memcpy(&lock->l_ost_lvb, &ols->ols_lvb,
628 sizeof(lock->l_ost_lvb));
629 unlock_res_and_lock(lock);
631 *flags |= LDLM_FL_LVB_READY;
634 /* Call the update callback. */
635 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
637 /* release the reference taken in ldlm_cli_enqueue() */
638 if (errcode == ELDLM_LOCK_MATCHED)
640 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
641 ldlm_lock_decref(lockh, mode);
646 int mdc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
649 struct osc_enqueue_args *aa = args;
650 struct ldlm_lock *lock;
651 struct lustre_handle *lockh = &aa->oa_lockh;
652 enum ldlm_mode mode = aa->oa_mode;
653 struct ldlm_enqueue_info einfo = {
654 .ei_type = aa->oa_type,
660 LASSERT(!aa->oa_speculative);
662 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
664 lock = ldlm_handle2lock(lockh);
665 LASSERTF(lock != NULL,
666 "lockh %#llx, req %p, aa %p - client evicted?\n",
667 lockh->cookie, req, aa);
669 /* Take an additional reference so that a blocking AST that
670 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
671 * to arrive after an upcall has been executed by
672 * mdc_enqueue_fini().
674 ldlm_lock_addref(lockh, mode);
676 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
677 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
679 /* Let CP AST to grant the lock first. */
680 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
682 /* Complete obtaining the lock procedure. */
683 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, &einfo, 1, aa->oa_flags,
684 aa->oa_lvb, aa->oa_lvb ?
685 sizeof(*aa->oa_lvb) : 0, lockh, rc);
686 /* Complete mdc stuff. */
687 rc = mdc_enqueue_fini(aa->oa_exp, req, aa->oa_upcall, aa->oa_cookie,
688 lockh, mode, aa->oa_flags, rc);
690 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
692 ldlm_lock_decref(lockh, mode);
697 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
698 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
699 * other synchronous requests, however keeping some locks and trying to obtain
700 * others may take a considerable amount of time in a case of ost failure; and
701 * when other sync requests do not get released lock from a client, the client
702 * is excluded from the cluster -- such scenarious make the life difficult, so
703 * release locks just after they are obtained. */
704 int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
705 struct ldlm_res_id *res_id, __u64 *flags,
706 union ldlm_policy_data *policy, struct ost_lvb *lvb,
707 osc_enqueue_upcall_f upcall, void *cookie,
708 struct ldlm_enqueue_info *einfo, int async)
710 struct obd_device *obd = exp->exp_obd;
711 struct lustre_handle lockh = { 0 };
712 struct ptlrpc_request *req = NULL;
713 struct ldlm_intent *lit;
715 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
716 __u64 match_flags = *flags;
720 bool compat_glimpse = glimpse && !exp_connect_dom_lvb(exp);
724 mode = einfo->ei_mode;
725 if (einfo->ei_mode == LCK_PR)
728 match_flags |= LDLM_FL_LVB_READY;
730 match_flags |= LDLM_FL_BLOCK_GRANTED;
731 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
732 einfo->ei_type, policy, mode, &lockh);
734 struct ldlm_lock *matched;
736 if (*flags & LDLM_FL_TEST_LOCK)
739 matched = ldlm_handle2lock(&lockh);
741 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
742 ldlm_set_kms_ignore(matched);
744 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
745 *flags |= LDLM_FL_LVB_READY;
747 /* We already have a lock, and it's referenced. */
748 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
750 ldlm_lock_decref(&lockh, mode);
751 LDLM_LOCK_PUT(matched);
754 ldlm_lock_decref(&lockh, mode);
755 LDLM_LOCK_PUT(matched);
758 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
761 /* Glimpse is intent on old server */
762 req = ptlrpc_request_alloc(class_exp2cliimp(exp), compat_glimpse ?
763 &RQF_LDLM_INTENT : &RQF_LDLM_ENQUEUE);
767 /* For WRITE lock cancel other locks on resource early if any */
768 if (einfo->ei_mode & LCK_PW)
769 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
775 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
777 ptlrpc_request_free(req);
781 if (compat_glimpse) {
782 /* pack the glimpse intent */
783 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
784 lit->opc = IT_GLIMPSE;
787 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
788 *flags &= ~LDLM_FL_BLOCK_GRANTED;
790 if (compat_glimpse) {
791 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
792 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
795 lvb_size = sizeof(*lvb);
796 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
799 ptlrpc_request_set_replen(req);
801 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
802 lvb_size, LVB_T_OST, &lockh, async);
805 struct osc_enqueue_args *aa;
807 aa = ptlrpc_req_async_args(aa, req);
809 aa->oa_mode = einfo->ei_mode;
810 aa->oa_type = einfo->ei_type;
811 lustre_handle_copy(&aa->oa_lockh, &lockh);
812 aa->oa_upcall = upcall;
813 aa->oa_cookie = cookie;
814 aa->oa_speculative = false;
815 aa->oa_flags = flags;
816 aa->oa_lvb = compat_glimpse ? NULL : lvb;
818 req->rq_interpret_reply = mdc_enqueue_interpret;
819 ptlrpcd_add_req(req);
821 ptlrpc_req_finished(req);
826 rc = mdc_enqueue_fini(exp, req, upcall, cookie, &lockh, einfo->ei_mode,
828 ptlrpc_req_finished(req);
833 * Implementation of cl_lock_operations::clo_enqueue() method for osc
834 * layer. This initiates ldlm enqueue:
836 * - cancels conflicting locks early (osc_lock_enqueue_wait());
838 * - calls osc_enqueue_base() to do actual enqueue.
840 * osc_enqueue_base() is supplied with an upcall function that is executed
841 * when lock is received either after a local cached ldlm lock is matched, or
842 * when a reply from the server is received.
844 * This function does not wait for the network communication to complete.
846 static int mdc_lock_enqueue(const struct lu_env *env,
847 const struct cl_lock_slice *slice,
848 struct cl_io *unused, struct cl_sync_io *anchor)
850 struct osc_thread_info *info = osc_env_info(env);
851 struct osc_io *oio = osc_env_io(env);
852 struct osc_object *osc = cl2osc(slice->cls_obj);
853 struct osc_lock *oscl = cl2osc_lock(slice);
854 struct cl_lock *lock = slice->cls_lock;
855 struct ldlm_res_id *resname = &info->oti_resname;
856 union ldlm_policy_data *policy = &info->oti_policy;
857 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
858 void *cookie = (void *)oscl;
864 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
865 "lock = %p, ols = %p\n", lock, oscl);
867 if (oscl->ols_state == OLS_GRANTED)
870 /* Lockahead is not supported on MDT yet */
871 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
872 result = -EOPNOTSUPP;
876 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
877 GOTO(enqueue_base, 0);
879 if (oscl->ols_glimpse) {
880 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
882 GOTO(enqueue_base, 0);
885 result = osc_lock_enqueue_wait(env, osc, oscl);
889 /* we can grant lockless lock right after all conflicting locks
891 if (osc_lock_is_lockless(oscl)) {
892 oscl->ols_state = OLS_GRANTED;
893 oio->oi_lockless = 1;
898 oscl->ols_state = OLS_ENQUEUED;
899 if (anchor != NULL) {
900 atomic_inc(&anchor->csi_sync_nr);
901 oscl->ols_owner = anchor;
905 * DLM lock's ast data must be osc_object;
906 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
909 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
910 mdc_lock_build_policy(env, lock, policy);
911 LASSERT(!oscl->ols_speculative);
912 result = mdc_enqueue_send(env, osc_export(osc), resname,
913 &oscl->ols_flags, policy, &oscl->ols_lvb,
914 upcall, cookie, &oscl->ols_einfo, async);
916 if (osc_lock_is_lockless(oscl)) {
917 oio->oi_lockless = 1;
919 LASSERT(oscl->ols_state == OLS_GRANTED);
920 LASSERT(oscl->ols_hold);
921 LASSERT(oscl->ols_dlmlock != NULL);
926 oscl->ols_state = OLS_CANCELLED;
927 osc_lock_wake_waiters(env, osc, oscl);
930 cl_sync_io_note(env, anchor, result);
935 static const struct cl_lock_operations mdc_lock_lockless_ops = {
936 .clo_fini = osc_lock_fini,
937 .clo_enqueue = mdc_lock_enqueue,
938 .clo_cancel = mdc_lock_lockless_cancel,
939 .clo_print = osc_lock_print
942 static const struct cl_lock_operations mdc_lock_ops = {
943 .clo_fini = osc_lock_fini,
944 .clo_enqueue = mdc_lock_enqueue,
945 .clo_cancel = osc_lock_cancel,
946 .clo_print = osc_lock_print,
949 int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
950 struct cl_lock *lock, const struct cl_io *io)
952 struct osc_lock *ols;
953 __u32 enqflags = lock->cll_descr.cld_enq_flags;
954 __u64 flags = osc_enq2ldlm_flags(enqflags);
958 /* Ignore AGL for Data-on-MDT, stat returns size data */
959 if ((enqflags & CEF_SPECULATIVE) != 0)
962 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
963 if (unlikely(ols == NULL))
966 ols->ols_state = OLS_NEW;
967 spin_lock_init(&ols->ols_lock);
968 INIT_LIST_HEAD(&ols->ols_waiting_list);
969 INIT_LIST_HEAD(&ols->ols_wait_entry);
970 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
971 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
973 ols->ols_flags = flags;
974 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
975 if (lock->cll_descr.cld_mode == CLM_GROUP)
976 ols->ols_flags |= LDLM_FL_ATOMIC_CB;
978 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
979 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
980 ols->ols_glimpse = 1;
982 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
984 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
986 if (!(enqflags & CEF_MUST))
987 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
989 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
990 osc_lock_set_writer(env, io, obj, ols);
992 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx\n",
993 lock, ols, ols->ols_flags);
1000 * An implementation of cl_io_operations specific methods for MDC layer.
1003 static int mdc_async_upcall(void *a, int rc)
1005 struct osc_async_cbargs *args = a;
1008 complete(&args->opc_sync);
1012 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
1013 pgoff_t index, struct lustre_handle *lh)
1015 struct ldlm_lock *lock;
1017 /* find DOM lock protecting object */
1018 lock = mdc_dlmlock_at_pgoff(env, osc, index,
1019 OSC_DAP_FL_TEST_LOCK |
1020 OSC_DAP_FL_CANCELING);
1022 struct ldlm_resource *res;
1023 struct ldlm_res_id *resname;
1025 resname = &osc_env_info(env)->oti_resname;
1026 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1027 res = ldlm_resource_get(osc_export(osc)->exp_obd->obd_namespace,
1028 NULL, resname, LDLM_IBITS, 0);
1029 ldlm_resource_dump(D_ERROR, res);
1030 libcfs_debug_dumpstack(NULL);
1033 *lh = lock->l_remote_handle;
1034 LDLM_LOCK_PUT(lock);
1039 static int mdc_io_setattr_start(const struct lu_env *env,
1040 const struct cl_io_slice *slice)
1042 struct cl_io *io = slice->cis_io;
1043 struct osc_io *oio = cl2osc_io(env, slice);
1044 struct cl_object *obj = slice->cis_obj;
1045 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1046 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1047 struct obdo *oa = &oio->oi_oa;
1048 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1049 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1050 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1051 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1054 /* silently ignore non-truncate setattr for Data-on-MDT object */
1055 if (cl_io_is_trunc(io)) {
1056 /* truncate cache dirty pages first */
1057 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1061 } else if (cl_io_is_fallocate(io) &&
1062 io->u.ci_setattr.sa_falloc_mode & FALLOC_FL_PUNCH_HOLE) {
1063 rc = osc_punch_start(env, io, obj);
1068 if (oio->oi_lockless == 0) {
1069 cl_object_attr_lock(obj);
1070 rc = cl_object_attr_get(env, obj, attr);
1072 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1073 unsigned int cl_valid = 0;
1075 if (ia_avalid & ATTR_SIZE) {
1076 attr->cat_size = size;
1077 attr->cat_kms = size;
1078 cl_valid = (CAT_SIZE | CAT_KMS);
1080 if (ia_avalid & ATTR_MTIME_SET) {
1081 attr->cat_mtime = lvb->lvb_mtime;
1082 cl_valid |= CAT_MTIME;
1084 if (ia_avalid & ATTR_ATIME_SET) {
1085 attr->cat_atime = lvb->lvb_atime;
1086 cl_valid |= CAT_ATIME;
1088 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1089 attr->cat_ctime = lvb->lvb_ctime;
1090 cl_valid |= CAT_CTIME;
1092 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1094 cl_object_attr_unlock(obj);
1099 if (!(ia_avalid & ATTR_SIZE) && !cl_io_is_fallocate(io))
1102 memset(oa, 0, sizeof(*oa));
1103 oa->o_oi = loi->loi_oi;
1104 oa->o_mtime = attr->cat_mtime;
1105 oa->o_atime = attr->cat_atime;
1106 oa->o_ctime = attr->cat_ctime;
1107 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1108 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1111 if (oio->oi_lockless) {
1112 oa->o_flags = OBD_FL_SRVLOCK;
1113 oa->o_valid |= OBD_MD_FLFLAGS;
1115 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1118 oa->o_valid |= OBD_MD_FLHANDLE;
1121 init_completion(&cbargs->opc_sync);
1122 if (cl_io_is_fallocate(io)) {
1123 int falloc_mode = io->u.ci_setattr.sa_falloc_mode;
1125 oa->o_size = io->u.ci_setattr.sa_falloc_offset;
1126 oa->o_blocks = io->u.ci_setattr.sa_falloc_end;
1127 rc = osc_fallocate_base(osc_export(cl2osc(obj)), oa,
1128 mdc_async_upcall, cbargs, falloc_mode);
1131 oa->o_blocks = OBD_OBJECT_EOF;
1132 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1133 mdc_async_upcall, cbargs);
1135 cbargs->opc_rpc_sent = rc == 0;
1139 static int mdc_io_read_ahead(const struct lu_env *env,
1140 const struct cl_io_slice *ios,
1141 pgoff_t start, struct cl_read_ahead *ra)
1143 struct osc_object *osc = cl2osc(ios->cis_obj);
1144 struct osc_io *oio = cl2osc_io(env, ios);
1145 struct ldlm_lock *dlmlock;
1149 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1150 if (dlmlock == NULL)
1153 oio->oi_is_readahead = 1;
1154 if (dlmlock->l_req_mode != LCK_PR) {
1155 struct lustre_handle lockh;
1157 ldlm_lock2handle(dlmlock, &lockh);
1158 ldlm_lock_addref(&lockh, LCK_PR);
1159 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1162 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1163 ra->cra_end_idx = CL_PAGE_EOF;
1164 ra->cra_release = osc_read_ahead_release;
1165 ra->cra_dlmlock = dlmlock;
1171 int mdc_io_fsync_start(const struct lu_env *env,
1172 const struct cl_io_slice *slice)
1174 struct cl_io *io = slice->cis_io;
1175 struct cl_fsync_io *fio = &io->u.ci_fsync;
1176 struct cl_object *obj = slice->cis_obj;
1177 struct osc_object *osc = cl2osc(obj);
1182 /* a MDC lock always covers whole object, do sync for whole
1183 * possible range despite of supplied start/end values.
1185 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1186 fio->fi_mode == CL_FSYNC_DISCARD);
1188 fio->fi_nr_written += result;
1191 if (fio->fi_mode == CL_FSYNC_ALL) {
1194 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1197 /* Use OSC sync code because it is asynchronous.
1198 * It is to be added into MDC and avoid the using of
1199 * OST_SYNC at both MDC and MDT.
1201 rc = osc_fsync_ost(env, osc, fio);
1209 struct mdc_data_version_args {
1210 struct osc_io *dva_oio;
1214 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1217 struct mdc_data_version_args *dva = args;
1218 struct osc_io *oio = dva->dva_oio;
1219 const struct mdt_body *body;
1225 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1227 GOTO(out, rc = -EPROTO);
1229 /* Prepare OBDO from mdt_body for CLIO */
1230 oio->oi_oa.o_valid = body->mbo_valid;
1231 oio->oi_oa.o_flags = body->mbo_flags;
1232 oio->oi_oa.o_data_version = body->mbo_version;
1233 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1236 oio->oi_cbarg.opc_rc = rc;
1237 complete(&oio->oi_cbarg.opc_sync);
1241 static int mdc_io_data_version_start(const struct lu_env *env,
1242 const struct cl_io_slice *slice)
1244 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1245 struct osc_io *oio = cl2osc_io(env, slice);
1246 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1247 struct osc_object *obj = cl2osc(slice->cis_obj);
1248 struct obd_export *exp = osc_export(obj);
1249 struct ptlrpc_request *req;
1250 struct mdt_body *body;
1251 struct mdc_data_version_args *dva;
1256 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1257 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1258 oio->oi_oa.o_valid = OBD_MD_FLID;
1260 init_completion(&cbargs->opc_sync);
1262 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1266 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1268 ptlrpc_request_free(req);
1272 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1273 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1274 body->mbo_valid = OBD_MD_FLID;
1275 /* Indicate that data version is needed */
1276 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1277 body->mbo_flags = 0;
1279 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1280 body->mbo_valid |= OBD_MD_FLFLAGS;
1281 body->mbo_flags |= OBD_FL_SRVLOCK;
1282 if (dv->dv_flags & LL_DV_WR_FLUSH)
1283 body->mbo_flags |= OBD_FL_FLUSH;
1286 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1287 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1288 ptlrpc_request_set_replen(req);
1290 req->rq_interpret_reply = mdc_data_version_interpret;
1291 dva = ptlrpc_req_async_args(dva, req);
1294 ptlrpcd_add_req(req);
1299 static void mdc_io_data_version_end(const struct lu_env *env,
1300 const struct cl_io_slice *slice)
1302 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1303 struct osc_io *oio = cl2osc_io(env, slice);
1304 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1307 wait_for_completion(&cbargs->opc_sync);
1309 if (cbargs->opc_rc != 0) {
1310 slice->cis_io->ci_result = cbargs->opc_rc;
1312 slice->cis_io->ci_result = 0;
1313 if (!(oio->oi_oa.o_valid &
1314 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1315 slice->cis_io->ci_result = -ENOTSUPP;
1317 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1318 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1319 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1320 dv->dv_data_version = oio->oi_oa.o_data_version;
1326 static const struct cl_io_operations mdc_io_ops = {
1329 .cio_iter_init = osc_io_iter_init,
1330 .cio_iter_fini = osc_io_rw_iter_fini,
1331 .cio_start = osc_io_read_start,
1334 .cio_iter_init = osc_io_iter_init,
1335 .cio_iter_fini = osc_io_rw_iter_fini,
1336 .cio_start = osc_io_write_start,
1337 .cio_end = osc_io_end,
1340 .cio_iter_init = osc_io_iter_init,
1341 .cio_iter_fini = osc_io_iter_fini,
1342 .cio_start = mdc_io_setattr_start,
1343 .cio_end = osc_io_setattr_end,
1345 [CIT_DATA_VERSION] = {
1346 .cio_start = mdc_io_data_version_start,
1347 .cio_end = mdc_io_data_version_end,
1350 .cio_iter_init = osc_io_iter_init,
1351 .cio_iter_fini = osc_io_iter_fini,
1352 .cio_start = osc_io_fault_start,
1353 .cio_end = osc_io_end,
1356 .cio_start = mdc_io_fsync_start,
1357 .cio_end = osc_io_fsync_end,
1360 .cio_start = osc_io_lseek_start,
1361 .cio_end = osc_io_lseek_end,
1364 .cio_read_ahead = mdc_io_read_ahead,
1365 .cio_lru_reserve = osc_io_lru_reserve,
1366 .cio_submit = osc_io_submit,
1367 .cio_commit_async = osc_io_commit_async,
1368 .cio_extent_release = osc_io_extent_release,
1371 int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1374 struct osc_io *oio = osc_env_io(env);
1376 CL_IO_SLICE_CLEAN(oio, oi_cl);
1377 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1381 static void mdc_build_res_name(struct osc_object *osc,
1382 struct ldlm_res_id *resname)
1384 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1388 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1389 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1392 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1393 struct cl_req_attr *attr)
1395 u64 flags = attr->cra_flags;
1397 /* Copy object FID to cl_attr */
1398 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1400 if (flags & OBD_MD_FLGROUP)
1401 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1403 if (flags & OBD_MD_FLID)
1404 attr->cra_oa->o_valid |= OBD_MD_FLID;
1406 if (flags & OBD_MD_FLHANDLE) {
1407 struct osc_page *opg;
1409 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1410 if (!opg->ops_srvlock) {
1413 rc = mdc_get_lock_handle(env, cl2osc(obj),
1415 &attr->cra_oa->o_handle);
1417 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1418 "uncovered page!\n");
1421 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1427 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1428 struct cl_attr *attr)
1430 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1432 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1433 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1435 return osc_attr_get(env, obj, attr);
1438 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1440 struct osc_object *osc = (struct osc_object *)data;
1441 struct ost_lvb *lvb = &lock->l_ost_lvb;
1442 struct lov_oinfo *oinfo;
1445 if (lock->l_ast_data == data) {
1446 lock->l_ast_data = NULL;
1448 LASSERT(osc != NULL);
1449 LASSERT(osc->oo_oinfo != NULL);
1450 LASSERT(lvb != NULL);
1452 /* Updates lvb in lock by the cached oinfo */
1453 oinfo = osc->oo_oinfo;
1455 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1456 "%llu %llu %llu by oinfo size %llu blocks %llu "
1457 "[cma]time %llu %llu %llu", lvb->lvb_size,
1458 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1459 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1460 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1461 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1462 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1464 cl_object_attr_lock(&osc->oo_cl);
1465 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1466 cl_object_attr_unlock(&osc->oo_cl);
1467 ldlm_clear_lvb_cached(lock);
1469 RETURN(LDLM_ITER_CONTINUE);
1472 int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1474 struct osc_object *osc = cl2osc(obj);
1475 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1477 /* DLM locks don't hold a reference of osc_object so we have to
1478 * clear it before the object is being destroyed. */
1479 osc_build_res_name(osc, resname);
1480 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1481 mdc_object_ast_clear, osc);
1485 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1486 struct ldlm_lock *lock)
1488 /* if lock cancel is initiated from llite then it is combined
1489 * lock with DOM bit and it may have no l_ast_data initialized yet,
1490 * so init it here with given osc_object.
1492 mdc_set_dom_lock_data(lock, cl2osc(obj));
1493 RETURN(mdc_dlm_canceling(env, lock));
1496 static const struct cl_object_operations mdc_ops = {
1497 .coo_page_init = osc_page_init,
1498 .coo_lock_init = mdc_lock_init,
1499 .coo_io_init = mdc_io_init,
1500 .coo_attr_get = mdc_attr_get,
1501 .coo_attr_update = osc_attr_update,
1502 .coo_glimpse = osc_object_glimpse,
1503 .coo_req_attr_set = mdc_req_attr_set,
1504 .coo_prune = mdc_object_prune,
1505 .coo_object_flush = mdc_object_flush
1508 static const struct osc_object_operations mdc_object_ops = {
1509 .oto_build_res_name = mdc_build_res_name,
1510 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1513 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1514 const struct lu_object_conf *conf)
1516 struct osc_object *osc = lu2osc(obj);
1518 if (osc->oo_initialized)
1521 osc->oo_initialized = true;
1523 return osc_object_init(env, obj, conf);
1526 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1528 osc_object_free(env, obj);
1531 static const struct lu_object_operations mdc_lu_obj_ops = {
1532 .loo_object_init = mdc_object_init,
1533 .loo_object_delete = NULL,
1534 .loo_object_release = NULL,
1535 .loo_object_free = mdc_object_free,
1536 .loo_object_print = osc_object_print,
1537 .loo_object_invariant = NULL
1540 struct lu_object *mdc_object_alloc(const struct lu_env *env,
1541 const struct lu_object_header *unused,
1542 struct lu_device *dev)
1544 struct osc_object *osc;
1545 struct lu_object *obj;
1547 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1550 lu_object_init(obj, NULL, dev);
1551 osc->oo_cl.co_ops = &mdc_ops;
1552 obj->lo_ops = &mdc_lu_obj_ops;
1553 osc->oo_obj_ops = &mdc_object_ops;
1554 osc->oo_initialized = false;
1561 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1562 struct lustre_cfg *cfg)
1564 size_t count = class_modify_config(cfg, PARAM_MDC,
1565 &d->ld_obd->obd_kset.kobj);
1566 return count > 0 ? 0 : count;
1569 const struct lu_device_operations mdc_lu_ops = {
1570 .ldo_object_alloc = mdc_object_alloc,
1571 .ldo_process_config = mdc_process_config,
1572 .ldo_recovery_complete = NULL,
1575 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1576 struct lu_device_type *t,
1577 struct lustre_cfg *cfg)
1579 struct lu_device *d;
1580 struct osc_device *od;
1581 struct obd_device *obd;
1586 RETURN(ERR_PTR(-ENOMEM));
1588 cl_device_init(&od->od_cl, t);
1590 d->ld_ops = &mdc_lu_ops;
1593 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1595 RETURN(ERR_PTR(-ENODEV));
1597 rc = mdc_setup(obd, cfg);
1599 osc_device_free(env, d);
1600 RETURN(ERR_PTR(rc));
1602 od->od_exp = obd->obd_self_export;
1606 static const struct lu_device_type_operations mdc_device_type_ops = {
1607 .ldto_device_alloc = mdc_device_alloc,
1608 .ldto_device_free = osc_device_free,
1609 .ldto_device_init = osc_device_init,
1610 .ldto_device_fini = osc_device_fini
1613 struct lu_device_type mdc_device_type = {
1614 .ldt_tags = LU_DEVICE_CL,
1615 .ldt_name = LUSTRE_MDC_NAME,
1616 .ldt_ops = &mdc_device_type_ops,
1617 .ldt_ctx_tags = LCT_CL_THREAD