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 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, enum ldlm_mode mode,
99 __u64 *flags, struct osc_object *obj,
100 struct lustre_handle *lockh,
101 enum ldlm_match_flags match_flags)
103 struct obd_device *obd = exp->exp_obd;
104 __u64 lflags = *flags;
109 rc = ldlm_lock_match_with_skip(obd->obd_namespace, lflags, 0,
110 res_id, type, policy, mode, lockh, match_flags);
111 if (rc == 0 || lflags & LDLM_FL_TEST_LOCK)
115 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
117 LASSERT(lock != NULL);
118 if (mdc_set_dom_lock_data(lock, obj)) {
119 lock_res_and_lock(lock);
120 if (!ldlm_is_lvb_cached(lock)) {
121 LASSERT(lock->l_ast_data == obj);
122 mdc_lock_lvb_update(env, obj, lock, NULL);
123 ldlm_set_lvb_cached(lock);
125 unlock_res_and_lock(lock);
127 ldlm_lock_decref(lockh, rc);
136 * Finds an existing lock covering a page with given index.
137 * Copy of osc_obj_dlmlock_at_pgoff() but for DoM IBITS lock.
139 struct ldlm_lock *mdc_dlmlock_at_pgoff(const struct lu_env *env,
140 struct osc_object *obj, pgoff_t index,
141 enum osc_dap_flags dap_flags)
143 struct osc_thread_info *info = osc_env_info(env);
144 struct ldlm_res_id *resname = &info->oti_resname;
145 union ldlm_policy_data *policy = &info->oti_policy;
146 struct lustre_handle lockh;
147 struct ldlm_lock *lock = NULL;
150 enum ldlm_match_flags match_flags = 0;
154 fid_build_reg_res_name(lu_object_fid(osc2lu(obj)), resname);
155 mdc_lock_build_policy(env, NULL, policy);
156 policy->l_inodebits.li_gid = LDLM_GID_ANY;
158 flags = LDLM_FL_BLOCK_GRANTED | LDLM_FL_CBPENDING;
159 if (dap_flags & OSC_DAP_FL_TEST_LOCK)
160 flags |= LDLM_FL_TEST_LOCK;
162 if (dap_flags & OSC_DAP_FL_AST)
163 match_flags |= LDLM_MATCH_AST;
165 if (dap_flags & OSC_DAP_FL_CANCELING)
166 match_flags |= LDLM_MATCH_UNREF;
169 /* Next, search for already existing extent locks that will cover us */
170 /* If we're trying to read, we also search for an existing PW lock. The
171 * VFS and page cache already protect us locally, so lots of readers/
172 * writers can share a single PW lock. */
173 mode = mdc_dom_lock_match(env, osc_export(obj), resname, LDLM_IBITS,
174 policy, LCK_PR | LCK_PW | LCK_GROUP, &flags,
175 obj, &lockh, match_flags);
177 lock = ldlm_handle2lock(&lockh);
178 /* RACE: the lock is cancelled so let's try again */
179 if (unlikely(lock == NULL))
187 * Check if page @page is covered by an extra lock or discard it.
189 static bool mdc_check_and_discard_cb(const struct lu_env *env, struct cl_io *io,
190 void **pvec, int count, void *cbdata)
192 struct osc_thread_info *info = osc_env_info(env);
193 struct osc_object *osc = cbdata;
197 for (i = 0; i < count; i++) {
198 struct osc_page *ops = pvec[i];
200 index = osc_index(ops);
201 if (index >= info->oti_fn_index) {
202 struct ldlm_lock *tmp;
203 struct cl_page *page = ops->ops_cl.cpl_page;
205 /* refresh non-overlapped index */
206 tmp = mdc_dlmlock_at_pgoff(env, osc, index,
207 OSC_DAP_FL_TEST_LOCK | OSC_DAP_FL_AST);
209 info->oti_fn_index = CL_PAGE_EOF;
211 } else if (cl_page_own(env, io, page) == 0) {
212 /* discard the page */
213 cl_page_discard(env, io, page);
214 cl_page_disown(env, io, page);
216 LASSERT(page->cp_state == CPS_FREEING);
220 info->oti_next_index = index + 1;
226 * Discard pages protected by the given lock. This function traverses radix
227 * tree to find all covering pages and discard them. If a page is being covered
228 * by other locks, it should remain in cache.
230 * If error happens on any step, the process continues anyway (the reasoning
231 * behind this being that lock cancellation cannot be delayed indefinitely).
233 static int mdc_lock_discard_pages(const struct lu_env *env,
234 struct osc_object *osc,
235 pgoff_t start, pgoff_t end,
238 struct osc_thread_info *info = osc_env_info(env);
239 struct cl_io *io = &info->oti_io;
240 osc_page_gang_cbt cb;
245 io->ci_obj = cl_object_top(osc2cl(osc));
246 io->ci_ignore_layout = 1;
247 result = cl_io_init(env, io, CIT_MISC, io->ci_obj);
251 cb = discard ? osc_discard_cb : mdc_check_and_discard_cb;
252 info->oti_fn_index = info->oti_next_index = start;
254 osc_page_gang_lookup(env, io, osc, info->oti_next_index,
255 end, cb, (void *)osc);
261 static int mdc_lock_flush(const struct lu_env *env, struct osc_object *obj,
262 pgoff_t start, pgoff_t end, enum cl_lock_mode mode,
270 if (mode == CLM_WRITE) {
271 result = osc_cache_writeback_range(env, obj, start, end, 1,
273 CDEBUG(D_CACHE, "object %p: [%lu -> %lu] %d pages were %s.\n",
274 obj, start, end, result,
275 discard ? "discarded" : "written back");
280 /* Avoid lock matching with CLM_WRITE, there can be no other locks */
281 rc = mdc_lock_discard_pages(env, obj, start, end,
282 mode == CLM_WRITE || discard);
283 if (result == 0 && rc < 0)
289 void mdc_lock_lockless_cancel(const struct lu_env *env,
290 const struct cl_lock_slice *slice)
292 struct osc_lock *ols = cl2osc_lock(slice);
293 struct osc_object *osc = cl2osc(slice->cls_obj);
294 struct cl_lock_descr *descr = &slice->cls_lock->cll_descr;
297 LASSERT(ols->ols_dlmlock == NULL);
298 rc = mdc_lock_flush(env, osc, descr->cld_start, descr->cld_end,
301 CERROR("Pages for lockless lock %p were not purged(%d)\n",
304 osc_lock_wake_waiters(env, osc, ols);
308 * Helper for osc_dlm_blocking_ast() handling discrepancies between cl_lock
309 * and ldlm_lock caches.
311 static int mdc_dlm_canceling(const struct lu_env *env,
312 struct ldlm_lock *dlmlock)
314 struct cl_object *obj = NULL;
317 enum cl_lock_mode mode = CLM_READ;
321 lock_res_and_lock(dlmlock);
322 if (!ldlm_is_granted(dlmlock)) {
323 dlmlock->l_ast_data = NULL;
324 unlock_res_and_lock(dlmlock);
328 discard = ldlm_is_discard_data(dlmlock);
329 if (dlmlock->l_granted_mode & (LCK_PW | LCK_GROUP))
332 if (dlmlock->l_ast_data != NULL) {
333 obj = osc2cl(dlmlock->l_ast_data);
336 unlock_res_and_lock(dlmlock);
338 /* if l_ast_data is NULL, the dlmlock was enqueued by AGL or
339 * the object has been destroyed. */
341 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
344 /* Destroy pages covered by the extent of the DLM lock */
345 result = mdc_lock_flush(env, cl2osc(obj), cl_index(obj, 0),
346 CL_PAGE_EOF, mode, discard);
347 /* Losing a lock, set KMS to 0.
348 * NB: assumed that DOM lock covers whole data on MDT.
350 /* losing a lock, update kms */
351 lock_res_and_lock(dlmlock);
352 data = dlmlock->l_ast_data;
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);
360 /* Skip dec in case mdc_object_ast_clear() did it */
361 if (data && dlmlock->l_req_mode == LCK_GROUP)
362 osc_grouplock_dec(cl2osc(obj), dlmlock);
363 cl_object_put(env, obj);
368 int mdc_ldlm_blocking_ast(struct ldlm_lock *dlmlock,
369 struct ldlm_lock_desc *new, void *data, int reason)
376 case LDLM_CB_BLOCKING: {
377 struct lustre_handle lockh;
379 ldlm_lock2handle(dlmlock, &lockh);
380 rc = ldlm_cli_cancel(&lockh, LCF_ASYNC);
385 case LDLM_CB_CANCELING: {
390 * This can be called in the context of outer IO, e.g.,
392 * osc_enqueue_base()->...
393 * ->ldlm_prep_elc_req()->...
394 * ->ldlm_cancel_callback()->...
395 * ->osc_ldlm_blocking_ast()
397 * new environment has to be created to not corrupt outer
400 env = cl_env_get(&refcheck);
406 rc = mdc_dlm_canceling(env, dlmlock);
407 cl_env_put(env, &refcheck);
417 * Updates object attributes from a lock value block (lvb) received together
418 * with the DLM lock reply from the server.
419 * This can be optimized to not update attributes when lock is a result of a
422 * Called under lock and resource spin-locks.
424 void mdc_lock_lvb_update(const struct lu_env *env, struct osc_object *osc,
425 struct ldlm_lock *dlmlock, struct ost_lvb *lvb)
427 struct cl_object *obj = osc2cl(osc);
428 struct lov_oinfo *oinfo = osc->oo_oinfo;
429 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
430 unsigned valid = CAT_BLOCKS | CAT_ATIME | CAT_CTIME | CAT_MTIME |
432 unsigned int setkms = 0;
437 LASSERT(dlmlock != NULL);
438 lvb = &dlmlock->l_ost_lvb;
440 cl_lvb2attr(attr, lvb);
442 cl_object_attr_lock(obj);
443 if (dlmlock != NULL) {
446 check_res_locked(dlmlock->l_resource);
447 size = lvb->lvb_size;
449 if (size >= oinfo->loi_kms) {
451 attr->cat_kms = size;
454 ldlm_lock_allow_match_locked(dlmlock);
457 /* The size should not be less than the kms */
458 if (attr->cat_size < oinfo->loi_kms)
459 attr->cat_size = oinfo->loi_kms;
461 LDLM_DEBUG(dlmlock, "acquired size %llu, setting rss=%llu;%s "
462 "kms=%llu, end=%llu", lvb->lvb_size, attr->cat_size,
463 setkms ? "" : " leaving",
464 setkms ? attr->cat_kms : oinfo->loi_kms,
465 dlmlock ? dlmlock->l_policy_data.l_extent.end : -1ull);
467 cl_object_attr_update(env, obj, attr, valid);
468 cl_object_attr_unlock(obj);
472 static void mdc_lock_granted(const struct lu_env *env, struct osc_lock *oscl,
473 struct lustre_handle *lockh, int errcode)
475 struct osc_object *osc = cl2osc(oscl->ols_cl.cls_obj);
476 struct ldlm_lock *dlmlock;
480 dlmlock = ldlm_handle2lock_long(lockh, 0);
481 LASSERT(dlmlock != NULL);
483 /* lock reference taken by ldlm_handle2lock_long() is
484 * owned by osc_lock and released in osc_lock_detach()
486 lu_ref_add_atomic(&dlmlock->l_reference, "osc_lock", oscl);
487 oscl->ols_has_ref = 1;
489 LASSERT(oscl->ols_dlmlock == NULL);
490 oscl->ols_dlmlock = dlmlock;
492 /* This may be a matched lock for glimpse request, do not hold
493 * lock reference in that case. */
494 if (!oscl->ols_glimpse) {
495 /* hold a refc for non glimpse lock which will
496 * be released in osc_lock_cancel() */
497 lustre_handle_copy(&oscl->ols_handle, lockh);
498 ldlm_lock_addref(lockh, oscl->ols_einfo.ei_mode);
502 /* Lock must have been granted. */
503 lock_res_and_lock(dlmlock);
504 if (ldlm_is_granted(dlmlock)) {
505 struct cl_lock_descr *descr = &oscl->ols_cl.cls_lock->cll_descr;
507 /* extend the lock extent, otherwise it will have problem when
508 * we decide whether to grant a lockless lock. */
509 descr->cld_mode = osc_ldlm2cl_lock(dlmlock->l_granted_mode);
510 descr->cld_start = cl_index(descr->cld_obj, 0);
511 descr->cld_end = CL_PAGE_EOF;
513 /* no lvb update for matched lock */
514 if (!ldlm_is_lvb_cached(dlmlock)) {
515 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
516 LASSERT(osc == dlmlock->l_ast_data);
517 mdc_lock_lvb_update(env, osc, dlmlock, NULL);
518 ldlm_set_lvb_cached(dlmlock);
521 unlock_res_and_lock(dlmlock);
523 LASSERT(oscl->ols_state != OLS_GRANTED);
524 oscl->ols_state = OLS_GRANTED;
526 if (errcode != ELDLM_LOCK_MATCHED && dlmlock->l_req_mode == LCK_GROUP)
527 osc_grouplock_inc_locked(osc, dlmlock);
532 * Lock upcall function that is executed either when a reply to ENQUEUE rpc is
533 * received from a server, or after mdc_enqueue_send() matched a local DLM
536 static int mdc_lock_upcall(void *cookie, struct lustre_handle *lockh,
539 struct osc_lock *oscl = cookie;
540 struct cl_lock_slice *slice = &oscl->ols_cl;
546 env = cl_env_percpu_get();
547 /* should never happen, similar to osc_ldlm_blocking_ast(). */
548 LASSERT(!IS_ERR(env));
550 rc = ldlm_error2errno(errcode);
551 if (oscl->ols_state == OLS_ENQUEUED) {
552 oscl->ols_state = OLS_UPCALL_RECEIVED;
553 } else if (oscl->ols_state == OLS_CANCELLED) {
556 CERROR("Impossible state: %d\n", oscl->ols_state);
560 CDEBUG(D_INODE, "rc %d, err %d\n", rc, errcode);
562 mdc_lock_granted(env, oscl, lockh, errcode);
564 /* Error handling, some errors are tolerable. */
565 if (oscl->ols_glimpse && rc == -ENAVAIL) {
566 LASSERT(oscl->ols_flags & LDLM_FL_LVB_READY);
567 mdc_lock_lvb_update(env, cl2osc(slice->cls_obj),
568 NULL, &oscl->ols_lvb);
569 /* Hide the error. */
573 if (oscl->ols_owner != NULL)
574 cl_sync_io_note(env, oscl->ols_owner, rc);
575 cl_env_percpu_put(env);
580 /* This is needed only for old servers (before 2.14) support */
581 int mdc_fill_lvb(struct req_capsule *pill, struct ost_lvb *lvb)
583 struct mdt_body *body;
585 /* get LVB data from mdt_body otherwise */
586 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
590 if (!(body->mbo_valid & OBD_MD_DOM_SIZE))
593 mdc_body2lvb(body, lvb);
597 int mdc_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
598 osc_enqueue_upcall_f upcall, void *cookie,
599 struct lustre_handle *lockh, enum ldlm_mode mode,
600 __u64 *flags, int errcode)
602 struct osc_lock *ols = cookie;
603 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
608 /* needed only for glimpse from an old server (< 2.14) */
609 if (glimpse && !exp_connect_dom_lvb(exp))
610 rc = mdc_fill_lvb(&req->rq_pill, &ols->ols_lvb);
612 if (glimpse && errcode == ELDLM_LOCK_ABORTED) {
613 struct ldlm_reply *rep;
615 rep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
617 rep->lock_policy_res2 =
618 ptlrpc_status_ntoh(rep->lock_policy_res2);
619 if (rep->lock_policy_res2)
620 errcode = rep->lock_policy_res2;
624 *flags |= LDLM_FL_LVB_READY;
625 } else if (errcode == ELDLM_OK) {
626 struct ldlm_lock *lock;
628 /* Callers have references, should be valid always */
629 lock = ldlm_handle2lock(lockh);
631 /* At this point ols_lvb must be filled with correct LVB either
632 * by mdc_fill_lvb() above or by ldlm_cli_enqueue_fini().
633 * DoM uses l_ost_lvb to store LVB data, so copy it here from
634 * just updated ols_lvb.
636 lock_res_and_lock(lock);
637 memcpy(&lock->l_ost_lvb, &ols->ols_lvb,
638 sizeof(lock->l_ost_lvb));
639 unlock_res_and_lock(lock);
641 *flags |= LDLM_FL_LVB_READY;
644 /* Call the update callback. */
645 rc = (*upcall)(cookie, lockh, rc < 0 ? rc : errcode);
647 /* release the reference taken in ldlm_cli_enqueue() */
648 if (errcode == ELDLM_LOCK_MATCHED)
650 if (errcode == ELDLM_OK && lustre_handle_is_used(lockh))
651 ldlm_lock_decref(lockh, mode);
656 int mdc_enqueue_interpret(const struct lu_env *env, struct ptlrpc_request *req,
659 struct osc_enqueue_args *aa = args;
660 struct ldlm_lock *lock;
661 struct lustre_handle *lockh = &aa->oa_lockh;
662 enum ldlm_mode mode = aa->oa_mode;
663 struct ldlm_enqueue_info einfo = {
664 .ei_type = aa->oa_type,
670 LASSERT(!aa->oa_speculative);
672 /* ldlm_cli_enqueue is holding a reference on the lock, so it must
674 lock = ldlm_handle2lock(lockh);
675 LASSERTF(lock != NULL,
676 "lockh %#llx, req %p, aa %p - client evicted?\n",
677 lockh->cookie, req, aa);
679 /* Take an additional reference so that a blocking AST that
680 * ldlm_cli_enqueue_fini() might post for a failed lock, is guaranteed
681 * to arrive after an upcall has been executed by
682 * mdc_enqueue_fini().
684 ldlm_lock_addref(lockh, mode);
686 /* Let cl_lock_state_wait fail with -ERESTARTSYS to unuse sublocks. */
687 OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_ENQUEUE_HANG, 2);
689 /* Let CP AST to grant the lock first. */
690 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
692 /* Complete obtaining the lock procedure. */
693 rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, &einfo, 1, aa->oa_flags,
694 aa->oa_lvb, aa->oa_lvb ?
695 sizeof(*aa->oa_lvb) : 0, lockh, rc, true);
696 /* Complete mdc stuff. */
697 rc = mdc_enqueue_fini(aa->oa_exp, req, aa->oa_upcall, aa->oa_cookie,
698 lockh, mode, aa->oa_flags, rc);
700 OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_CANCEL_RACE, 10);
702 ldlm_lock_decref(lockh, mode);
707 /* When enqueuing asynchronously, locks are not ordered, we can obtain a lock
708 * from the 2nd OSC before a lock from the 1st one. This does not deadlock with
709 * other synchronous requests, however keeping some locks and trying to obtain
710 * others may take a considerable amount of time in a case of ost failure; and
711 * when other sync requests do not get released lock from a client, the client
712 * is excluded from the cluster -- such scenarious make the life difficult, so
713 * release locks just after they are obtained. */
714 int mdc_enqueue_send(const struct lu_env *env, struct obd_export *exp,
715 struct ldlm_res_id *res_id, __u64 *flags,
716 union ldlm_policy_data *policy, struct ost_lvb *lvb,
717 osc_enqueue_upcall_f upcall, void *cookie,
718 struct ldlm_enqueue_info *einfo, int async)
720 struct obd_device *obd = exp->exp_obd;
721 struct lustre_handle lockh = { 0 };
722 struct ptlrpc_request *req = NULL;
723 struct ldlm_intent *lit;
725 bool glimpse = *flags & LDLM_FL_HAS_INTENT;
726 __u64 match_flags = *flags;
730 bool compat_glimpse = glimpse && !exp_connect_dom_lvb(exp);
734 mode = einfo->ei_mode;
735 if (einfo->ei_mode == LCK_PR)
738 match_flags |= LDLM_FL_LVB_READY;
740 match_flags |= LDLM_FL_BLOCK_GRANTED;
741 mode = ldlm_lock_match(obd->obd_namespace, match_flags, res_id,
742 einfo->ei_type, policy, mode, &lockh);
744 struct ldlm_lock *matched;
746 if (*flags & LDLM_FL_TEST_LOCK)
749 matched = ldlm_handle2lock(&lockh);
751 if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GLIMPSE_DDOS))
752 ldlm_set_kms_ignore(matched);
754 if (mdc_set_dom_lock_data(matched, einfo->ei_cbdata)) {
755 *flags |= LDLM_FL_LVB_READY;
757 /* We already have a lock, and it's referenced. */
758 (*upcall)(cookie, &lockh, ELDLM_LOCK_MATCHED);
760 ldlm_lock_decref(&lockh, mode);
761 LDLM_LOCK_PUT(matched);
764 ldlm_lock_decref(&lockh, mode);
765 LDLM_LOCK_PUT(matched);
768 if (*flags & (LDLM_FL_TEST_LOCK | LDLM_FL_MATCH_LOCK))
771 /* Glimpse is intent on old server */
772 req = ptlrpc_request_alloc(class_exp2cliimp(exp), compat_glimpse ?
773 &RQF_LDLM_INTENT : &RQF_LDLM_ENQUEUE);
777 /* For WRITE lock cancel other locks on resource early if any */
778 if (einfo->ei_mode & LCK_PW)
779 count = mdc_resource_get_unused_res(exp, res_id, &cancels,
785 rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
787 ptlrpc_request_free(req);
791 if (compat_glimpse) {
792 /* pack the glimpse intent */
793 lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
794 lit->opc = IT_GLIMPSE;
797 /* users of mdc_enqueue() can pass this flag for ldlm_lock_match() */
798 *flags &= ~LDLM_FL_BLOCK_GRANTED;
800 if (compat_glimpse) {
801 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
802 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
805 lvb_size = sizeof(*lvb);
806 req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
809 ptlrpc_request_set_replen(req);
811 rc = ldlm_cli_enqueue(exp, &req, einfo, res_id, policy, flags, lvb,
812 lvb_size, LVB_T_OST, &lockh, async);
815 struct osc_enqueue_args *aa;
817 aa = ptlrpc_req_async_args(aa, req);
819 aa->oa_mode = einfo->ei_mode;
820 aa->oa_type = einfo->ei_type;
821 lustre_handle_copy(&aa->oa_lockh, &lockh);
822 aa->oa_upcall = upcall;
823 aa->oa_cookie = cookie;
824 aa->oa_speculative = false;
825 aa->oa_flags = flags;
826 aa->oa_lvb = compat_glimpse ? NULL : lvb;
828 req->rq_interpret_reply = mdc_enqueue_interpret;
829 ptlrpcd_add_req(req);
831 ptlrpc_req_finished(req);
836 rc = mdc_enqueue_fini(exp, req, upcall, cookie, &lockh, einfo->ei_mode,
838 ptlrpc_req_finished(req);
843 * Implementation of cl_lock_operations::clo_enqueue() method for osc
844 * layer. This initiates ldlm enqueue:
846 * - cancels conflicting locks early (osc_lock_enqueue_wait());
848 * - calls osc_enqueue_base() to do actual enqueue.
850 * osc_enqueue_base() is supplied with an upcall function that is executed
851 * when lock is received either after a local cached ldlm lock is matched, or
852 * when a reply from the server is received.
854 * This function does not wait for the network communication to complete.
856 static int __mdc_lock_enqueue(const struct lu_env *env,
857 const struct cl_lock_slice *slice,
858 struct cl_io *unused, struct cl_sync_io *anchor)
860 struct osc_thread_info *info = osc_env_info(env);
861 struct osc_io *oio = osc_env_io(env);
862 struct osc_object *osc = cl2osc(slice->cls_obj);
863 struct osc_lock *oscl = cl2osc_lock(slice);
864 struct cl_lock *lock = slice->cls_lock;
865 struct ldlm_res_id *resname = &info->oti_resname;
866 union ldlm_policy_data *policy = &info->oti_policy;
867 osc_enqueue_upcall_f upcall = mdc_lock_upcall;
868 void *cookie = (void *)oscl;
874 LASSERTF(ergo(oscl->ols_glimpse, lock->cll_descr.cld_mode <= CLM_READ),
875 "lock = %p, ols = %p\n", lock, oscl);
877 if (oscl->ols_state == OLS_GRANTED)
880 /* Lockahead is not supported on MDT yet */
881 if (oscl->ols_flags & LDLM_FL_NO_EXPANSION) {
882 result = -EOPNOTSUPP;
886 if (oscl->ols_flags & LDLM_FL_TEST_LOCK)
887 GOTO(enqueue_base, 0);
889 if (oscl->ols_glimpse) {
890 LASSERT(equi(oscl->ols_speculative, anchor == NULL));
892 GOTO(enqueue_base, 0);
895 result = osc_lock_enqueue_wait(env, osc, oscl);
899 /* we can grant lockless lock right after all conflicting locks
901 if (osc_lock_is_lockless(oscl)) {
902 oscl->ols_state = OLS_GRANTED;
903 oio->oi_lockless = 1;
908 oscl->ols_state = OLS_ENQUEUED;
909 if (anchor != NULL) {
910 atomic_inc(&anchor->csi_sync_nr);
911 oscl->ols_owner = anchor;
915 * DLM lock's ast data must be osc_object;
916 * DLM's enqueue callback set to osc_lock_upcall() with cookie as
919 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
920 mdc_lock_build_policy(env, lock, policy);
921 LASSERT(!oscl->ols_speculative);
922 result = mdc_enqueue_send(env, osc_export(osc), resname,
923 &oscl->ols_flags, policy, &oscl->ols_lvb,
924 upcall, cookie, &oscl->ols_einfo, async);
926 if (osc_lock_is_lockless(oscl)) {
927 oio->oi_lockless = 1;
929 LASSERT(oscl->ols_state == OLS_GRANTED);
930 LASSERT(oscl->ols_hold);
931 LASSERT(oscl->ols_dlmlock != NULL);
936 oscl->ols_state = OLS_CANCELLED;
937 osc_lock_wake_waiters(env, osc, oscl);
940 cl_sync_io_note(env, anchor, result);
945 static int mdc_lock_enqueue(const struct lu_env *env,
946 const struct cl_lock_slice *slice,
947 struct cl_io *unused, struct cl_sync_io *anchor)
949 struct osc_object *obj = cl2osc(slice->cls_obj);
950 struct osc_lock *oscl = cl2osc_lock(slice);
951 struct lustre_handle lh = { 0 };
954 if (oscl->ols_cl.cls_lock->cll_descr.cld_mode == CLM_GROUP) {
955 rc = osc_grouplock_enqueue_init(env, obj, oscl, &lh);
960 rc = __mdc_lock_enqueue(env, slice, unused, anchor);
962 if (oscl->ols_cl.cls_lock->cll_descr.cld_mode == CLM_GROUP)
963 osc_grouplock_enqueue_fini(env, obj, oscl, &lh);
967 static const struct cl_lock_operations mdc_lock_lockless_ops = {
968 .clo_fini = osc_lock_fini,
969 .clo_enqueue = mdc_lock_enqueue,
970 .clo_cancel = mdc_lock_lockless_cancel,
971 .clo_print = osc_lock_print
974 static const struct cl_lock_operations mdc_lock_ops = {
975 .clo_fini = osc_lock_fini,
976 .clo_enqueue = mdc_lock_enqueue,
977 .clo_cancel = osc_lock_cancel,
978 .clo_print = osc_lock_print,
981 int mdc_lock_init(const struct lu_env *env, struct cl_object *obj,
982 struct cl_lock *lock, const struct cl_io *io)
984 struct osc_lock *ols;
985 __u32 enqflags = lock->cll_descr.cld_enq_flags;
986 __u64 flags = osc_enq2ldlm_flags(enqflags);
990 /* Ignore AGL for Data-on-MDT, stat returns size data */
991 if ((enqflags & CEF_SPECULATIVE) != 0)
994 OBD_SLAB_ALLOC_PTR_GFP(ols, osc_lock_kmem, GFP_NOFS);
995 if (unlikely(ols == NULL))
998 ols->ols_state = OLS_NEW;
999 spin_lock_init(&ols->ols_lock);
1000 INIT_LIST_HEAD(&ols->ols_waiting_list);
1001 INIT_LIST_HEAD(&ols->ols_wait_entry);
1002 INIT_LIST_HEAD(&ols->ols_nextlock_oscobj);
1003 ols->ols_lockless_ops = &mdc_lock_lockless_ops;
1005 ols->ols_flags = flags;
1006 ols->ols_speculative = !!(enqflags & CEF_SPECULATIVE);
1008 if (ols->ols_flags & LDLM_FL_HAS_INTENT) {
1009 ols->ols_flags |= LDLM_FL_BLOCK_GRANTED;
1010 ols->ols_glimpse = 1;
1012 mdc_lock_build_einfo(env, lock, cl2osc(obj), &ols->ols_einfo);
1014 cl_lock_slice_add(lock, &ols->ols_cl, obj, &mdc_lock_ops);
1016 if (!(enqflags & CEF_MUST))
1017 osc_lock_to_lockless(env, ols, (enqflags & CEF_NEVER));
1019 if (io->ci_type == CIT_WRITE || cl_io_is_mkwrite(io))
1020 osc_lock_set_writer(env, io, obj, ols);
1021 else if (io->ci_type == CIT_READ ||
1022 (io->ci_type == CIT_FAULT && !io->u.ci_fault.ft_mkwrite))
1023 osc_lock_set_reader(env, io, obj, ols);
1025 LDLM_DEBUG_NOLOCK("lock %p, mdc lock %p, flags %llx\n",
1026 lock, ols, ols->ols_flags);
1033 * An implementation of cl_io_operations specific methods for MDC layer.
1036 static int mdc_async_upcall(void *a, int rc)
1038 struct osc_async_cbargs *args = a;
1041 complete(&args->opc_sync);
1045 static int mdc_get_lock_handle(const struct lu_env *env, struct osc_object *osc,
1046 pgoff_t index, struct lustre_handle *lh)
1048 struct ldlm_lock *lock;
1050 /* find DOM lock protecting object */
1051 lock = mdc_dlmlock_at_pgoff(env, osc, index,
1052 OSC_DAP_FL_TEST_LOCK |
1053 OSC_DAP_FL_CANCELING);
1055 struct ldlm_resource *res;
1056 struct ldlm_res_id *resname;
1058 resname = &osc_env_info(env)->oti_resname;
1059 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1060 res = ldlm_resource_get(osc_export(osc)->
1061 exp_obd->obd_namespace,
1062 resname, LDLM_IBITS, 0);
1064 CERROR("No lock resource for "DFID"\n",
1065 PFID(lu_object_fid(osc2lu(osc))));
1067 ldlm_resource_dump(D_ERROR, res);
1068 libcfs_debug_dumpstack(NULL);
1071 *lh = lock->l_remote_handle;
1072 LDLM_LOCK_PUT(lock);
1077 static int mdc_io_setattr_start(const struct lu_env *env,
1078 const struct cl_io_slice *slice)
1080 struct cl_io *io = slice->cis_io;
1081 struct osc_io *oio = cl2osc_io(env, slice);
1082 struct cl_object *obj = slice->cis_obj;
1083 struct lov_oinfo *loi = cl2osc(obj)->oo_oinfo;
1084 struct cl_attr *attr = &osc_env_info(env)->oti_attr;
1085 struct obdo *oa = &oio->oi_oa;
1086 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1087 __u64 size = io->u.ci_setattr.sa_attr.lvb_size;
1088 unsigned int ia_avalid = io->u.ci_setattr.sa_avalid;
1089 enum op_xvalid ia_xvalid = io->u.ci_setattr.sa_xvalid;
1092 /* silently ignore non-truncate setattr for Data-on-MDT object */
1093 if (cl_io_is_trunc(io)) {
1094 /* truncate cache dirty pages first */
1095 rc = osc_cache_truncate_start(env, cl2osc(obj), size,
1099 } else if (cl_io_is_fallocate(io) &&
1100 io->u.ci_setattr.sa_falloc_mode & FALLOC_FL_PUNCH_HOLE) {
1101 rc = osc_punch_start(env, io, obj);
1106 if (oio->oi_lockless == 0) {
1107 cl_object_attr_lock(obj);
1108 rc = cl_object_attr_get(env, obj, attr);
1110 struct ost_lvb *lvb = &io->u.ci_setattr.sa_attr;
1111 unsigned int cl_valid = 0;
1113 if (ia_avalid & ATTR_SIZE) {
1114 attr->cat_size = size;
1115 attr->cat_kms = size;
1116 cl_valid = (CAT_SIZE | CAT_KMS);
1118 if (ia_avalid & ATTR_MTIME_SET) {
1119 attr->cat_mtime = lvb->lvb_mtime;
1120 cl_valid |= CAT_MTIME;
1122 if (ia_avalid & ATTR_ATIME_SET) {
1123 attr->cat_atime = lvb->lvb_atime;
1124 cl_valid |= CAT_ATIME;
1126 if (ia_xvalid & OP_XVALID_CTIME_SET) {
1127 attr->cat_ctime = lvb->lvb_ctime;
1128 cl_valid |= CAT_CTIME;
1130 rc = cl_object_attr_update(env, obj, attr, cl_valid);
1132 cl_object_attr_unlock(obj);
1137 if (!(ia_avalid & ATTR_SIZE) && !cl_io_is_fallocate(io))
1140 memset(oa, 0, sizeof(*oa));
1141 oa->o_oi = loi->loi_oi;
1142 oa->o_mtime = attr->cat_mtime;
1143 oa->o_atime = attr->cat_atime;
1144 oa->o_ctime = attr->cat_ctime;
1145 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLATIME |
1146 OBD_MD_FLCTIME | OBD_MD_FLMTIME | OBD_MD_FLSIZE |
1149 if (oio->oi_lockless) {
1150 oa->o_flags = OBD_FL_SRVLOCK;
1151 oa->o_valid |= OBD_MD_FLFLAGS;
1153 rc = mdc_get_lock_handle(env, cl2osc(obj), CL_PAGE_EOF,
1156 oa->o_valid |= OBD_MD_FLHANDLE;
1159 init_completion(&cbargs->opc_sync);
1160 if (cl_io_is_fallocate(io)) {
1161 int falloc_mode = io->u.ci_setattr.sa_falloc_mode;
1163 oa->o_size = io->u.ci_setattr.sa_falloc_offset;
1164 oa->o_blocks = io->u.ci_setattr.sa_falloc_end;
1165 rc = osc_fallocate_base(osc_export(cl2osc(obj)), oa,
1166 mdc_async_upcall, cbargs, falloc_mode);
1169 oa->o_blocks = OBD_OBJECT_EOF;
1170 rc = osc_punch_send(osc_export(cl2osc(obj)), oa,
1171 mdc_async_upcall, cbargs);
1173 cbargs->opc_rpc_sent = rc == 0;
1177 static int mdc_io_read_ahead(const struct lu_env *env,
1178 const struct cl_io_slice *ios,
1179 pgoff_t start, struct cl_read_ahead *ra)
1181 struct osc_object *osc = cl2osc(ios->cis_obj);
1182 struct osc_io *oio = cl2osc_io(env, ios);
1183 struct ldlm_lock *dlmlock;
1187 dlmlock = mdc_dlmlock_at_pgoff(env, osc, start, 0);
1188 if (dlmlock == NULL)
1191 oio->oi_is_readahead = 1;
1192 if (dlmlock->l_req_mode != LCK_PR) {
1193 struct lustre_handle lockh;
1195 ldlm_lock2handle(dlmlock, &lockh);
1196 ldlm_lock_addref(&lockh, LCK_PR);
1197 ldlm_lock_decref(&lockh, dlmlock->l_req_mode);
1200 ra->cra_rpc_pages = osc_cli(osc)->cl_max_pages_per_rpc;
1201 ra->cra_end_idx = CL_PAGE_EOF;
1202 ra->cra_release = osc_read_ahead_release;
1203 ra->cra_dlmlock = dlmlock;
1209 int mdc_io_fsync_start(const struct lu_env *env,
1210 const struct cl_io_slice *slice)
1212 struct cl_io *io = slice->cis_io;
1213 struct cl_fsync_io *fio = &io->u.ci_fsync;
1214 struct cl_object *obj = slice->cis_obj;
1215 struct osc_object *osc = cl2osc(obj);
1220 /* a MDC lock always covers whole object, do sync for whole
1221 * possible range despite of supplied start/end values.
1223 result = osc_cache_writeback_range(env, osc, 0, CL_PAGE_EOF, 0,
1224 fio->fi_mode == CL_FSYNC_DISCARD);
1226 fio->fi_nr_written += result;
1229 if (fio->fi_mode == CL_FSYNC_ALL) {
1232 rc = osc_cache_wait_range(env, osc, 0, CL_PAGE_EOF);
1235 /* Use OSC sync code because it is asynchronous.
1236 * It is to be added into MDC and avoid the using of
1237 * OST_SYNC at both MDC and MDT.
1239 rc = osc_fsync_ost(env, osc, fio);
1247 struct mdc_data_version_args {
1248 struct osc_io *dva_oio;
1252 mdc_data_version_interpret(const struct lu_env *env, struct ptlrpc_request *req,
1255 struct mdc_data_version_args *dva = args;
1256 struct osc_io *oio = dva->dva_oio;
1257 const struct mdt_body *body;
1263 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1265 GOTO(out, rc = -EPROTO);
1267 /* Prepare OBDO from mdt_body for CLIO */
1268 oio->oi_oa.o_valid = body->mbo_valid;
1269 oio->oi_oa.o_flags = body->mbo_flags;
1270 oio->oi_oa.o_data_version = body->mbo_version;
1271 oio->oi_oa.o_layout_version = body->mbo_layout_gen;
1274 oio->oi_cbarg.opc_rc = rc;
1275 complete(&oio->oi_cbarg.opc_sync);
1279 static int mdc_io_data_version_start(const struct lu_env *env,
1280 const struct cl_io_slice *slice)
1282 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1283 struct osc_io *oio = cl2osc_io(env, slice);
1284 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1285 struct osc_object *obj = cl2osc(slice->cis_obj);
1286 struct obd_export *exp = osc_export(obj);
1287 struct ptlrpc_request *req;
1288 struct mdt_body *body;
1289 struct mdc_data_version_args *dva;
1294 memset(&oio->oi_oa, 0, sizeof(oio->oi_oa));
1295 oio->oi_oa.o_oi.oi_fid = *lu_object_fid(osc2lu(obj));
1296 oio->oi_oa.o_valid = OBD_MD_FLID;
1298 init_completion(&cbargs->opc_sync);
1300 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
1304 rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
1306 ptlrpc_request_free(req);
1310 body = req_capsule_client_get(&req->rq_pill, &RMF_MDT_BODY);
1311 body->mbo_fid1 = *lu_object_fid(osc2lu(obj));
1312 body->mbo_valid = OBD_MD_FLID;
1313 /* Indicate that data version is needed */
1314 body->mbo_valid |= OBD_MD_FLDATAVERSION;
1315 body->mbo_flags = 0;
1317 if (dv->dv_flags & (LL_DV_RD_FLUSH | LL_DV_WR_FLUSH)) {
1318 body->mbo_valid |= OBD_MD_FLFLAGS;
1319 body->mbo_flags |= OBD_FL_SRVLOCK;
1320 if (dv->dv_flags & LL_DV_WR_FLUSH)
1321 body->mbo_flags |= OBD_FL_FLUSH;
1324 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
1325 req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, 0);
1326 req_capsule_set_size(&req->rq_pill, &RMF_FILE_ENCCTX, RCL_SERVER, 0);
1327 ptlrpc_request_set_replen(req);
1329 req->rq_interpret_reply = mdc_data_version_interpret;
1330 dva = ptlrpc_req_async_args(dva, req);
1333 ptlrpcd_add_req(req);
1338 static void mdc_io_data_version_end(const struct lu_env *env,
1339 const struct cl_io_slice *slice)
1341 struct cl_data_version_io *dv = &slice->cis_io->u.ci_data_version;
1342 struct osc_io *oio = cl2osc_io(env, slice);
1343 struct osc_async_cbargs *cbargs = &oio->oi_cbarg;
1346 wait_for_completion(&cbargs->opc_sync);
1348 if (cbargs->opc_rc != 0) {
1349 slice->cis_io->ci_result = cbargs->opc_rc;
1351 slice->cis_io->ci_result = 0;
1352 if (!(oio->oi_oa.o_valid &
1353 (OBD_MD_LAYOUT_VERSION | OBD_MD_FLDATAVERSION)))
1354 slice->cis_io->ci_result = -ENOTSUPP;
1356 if (oio->oi_oa.o_valid & OBD_MD_LAYOUT_VERSION)
1357 dv->dv_layout_version = oio->oi_oa.o_layout_version;
1358 if (oio->oi_oa.o_valid & OBD_MD_FLDATAVERSION)
1359 dv->dv_data_version = oio->oi_oa.o_data_version;
1365 static const struct cl_io_operations mdc_io_ops = {
1368 .cio_iter_init = osc_io_iter_init,
1369 .cio_iter_fini = osc_io_rw_iter_fini,
1370 .cio_start = osc_io_read_start,
1373 .cio_iter_init = osc_io_iter_init,
1374 .cio_iter_fini = osc_io_rw_iter_fini,
1375 .cio_start = osc_io_write_start,
1376 .cio_end = osc_io_end,
1379 .cio_iter_init = osc_io_iter_init,
1380 .cio_iter_fini = osc_io_iter_fini,
1381 .cio_start = mdc_io_setattr_start,
1382 .cio_end = osc_io_setattr_end,
1384 [CIT_DATA_VERSION] = {
1385 .cio_start = mdc_io_data_version_start,
1386 .cio_end = mdc_io_data_version_end,
1389 .cio_iter_init = osc_io_iter_init,
1390 .cio_iter_fini = osc_io_iter_fini,
1391 .cio_start = osc_io_fault_start,
1392 .cio_end = osc_io_end,
1395 .cio_start = mdc_io_fsync_start,
1396 .cio_end = osc_io_fsync_end,
1399 .cio_start = osc_io_lseek_start,
1400 .cio_end = osc_io_lseek_end,
1403 .cio_read_ahead = mdc_io_read_ahead,
1404 .cio_lru_reserve = osc_io_lru_reserve,
1405 .cio_submit = osc_io_submit,
1406 .cio_commit_async = osc_io_commit_async,
1407 .cio_extent_release = osc_io_extent_release,
1410 int mdc_io_init(const struct lu_env *env, struct cl_object *obj,
1413 struct osc_io *oio = osc_env_io(env);
1415 CL_IO_SLICE_CLEAN(oio, oi_cl);
1416 cl_io_slice_add(io, &oio->oi_cl, obj, &mdc_io_ops);
1420 static void mdc_build_res_name(struct osc_object *osc,
1421 struct ldlm_res_id *resname)
1423 fid_build_reg_res_name(lu_object_fid(osc2lu(osc)), resname);
1427 * Implementation of struct cl_req_operations::cro_attr_set() for MDC
1428 * layer. MDC is responsible for struct obdo::o_id and struct obdo::o_seq
1431 static void mdc_req_attr_set(const struct lu_env *env, struct cl_object *obj,
1432 struct cl_req_attr *attr)
1434 u64 flags = attr->cra_flags;
1436 /* Copy object FID to cl_attr */
1437 attr->cra_oa->o_oi.oi_fid = *lu_object_fid(&obj->co_lu);
1439 if (flags & OBD_MD_FLGROUP)
1440 attr->cra_oa->o_valid |= OBD_MD_FLGROUP;
1442 if (flags & OBD_MD_FLID)
1443 attr->cra_oa->o_valid |= OBD_MD_FLID;
1445 if (flags & OBD_MD_FLHANDLE) {
1446 struct osc_page *opg;
1448 opg = osc_cl_page_osc(attr->cra_page, cl2osc(obj));
1449 if (!opg->ops_srvlock) {
1452 rc = mdc_get_lock_handle(env, cl2osc(obj),
1454 &attr->cra_oa->o_handle);
1456 CL_PAGE_DEBUG(D_ERROR, env, attr->cra_page,
1457 "uncovered page!\n");
1460 attr->cra_oa->o_valid |= OBD_MD_FLHANDLE;
1466 static int mdc_attr_get(const struct lu_env *env, struct cl_object *obj,
1467 struct cl_attr *attr)
1469 struct lov_oinfo *oinfo = cl2osc(obj)->oo_oinfo;
1471 if (OST_LVB_IS_ERR(oinfo->loi_lvb.lvb_blocks))
1472 return OST_LVB_GET_ERR(oinfo->loi_lvb.lvb_blocks);
1474 return osc_attr_get(env, obj, attr);
1477 static int mdc_object_ast_clear(struct ldlm_lock *lock, void *data)
1479 struct osc_object *osc = (struct osc_object *)data;
1480 struct ost_lvb *lvb = &lock->l_ost_lvb;
1481 struct lov_oinfo *oinfo;
1484 if (lock->l_ast_data == data) {
1485 lock->l_ast_data = NULL;
1487 LASSERT(osc != NULL);
1488 LASSERT(osc->oo_oinfo != NULL);
1489 LASSERT(lvb != NULL);
1491 /* Updates lvb in lock by the cached oinfo */
1492 oinfo = osc->oo_oinfo;
1494 LDLM_DEBUG(lock, "update lock size %llu blocks %llu [cma]time: "
1495 "%llu %llu %llu by oinfo size %llu blocks %llu "
1496 "[cma]time %llu %llu %llu", lvb->lvb_size,
1497 lvb->lvb_blocks, lvb->lvb_ctime, lvb->lvb_mtime,
1498 lvb->lvb_atime, oinfo->loi_lvb.lvb_size,
1499 oinfo->loi_lvb.lvb_blocks, oinfo->loi_lvb.lvb_ctime,
1500 oinfo->loi_lvb.lvb_mtime, oinfo->loi_lvb.lvb_atime);
1501 LASSERT(oinfo->loi_lvb.lvb_size >= oinfo->loi_kms);
1503 cl_object_attr_lock(&osc->oo_cl);
1504 memcpy(lvb, &oinfo->loi_lvb, sizeof(oinfo->loi_lvb));
1505 cl_object_attr_unlock(&osc->oo_cl);
1506 ldlm_clear_lvb_cached(lock);
1508 if (lock->l_req_mode == LCK_GROUP)
1509 osc_grouplock_dec(osc, lock);
1511 RETURN(LDLM_ITER_CONTINUE);
1514 int mdc_object_prune(const struct lu_env *env, struct cl_object *obj)
1516 struct osc_object *osc = cl2osc(obj);
1517 struct ldlm_res_id *resname = &osc_env_info(env)->oti_resname;
1519 /* DLM locks don't hold a reference of osc_object so we have to
1520 * clear it before the object is being destroyed. */
1521 osc_build_res_name(osc, resname);
1522 ldlm_resource_iterate(osc_export(osc)->exp_obd->obd_namespace, resname,
1523 mdc_object_ast_clear, osc);
1527 static int mdc_object_flush(const struct lu_env *env, struct cl_object *obj,
1528 struct ldlm_lock *lock)
1530 /* if lock cancel is initiated from llite then it is combined
1531 * lock with DOM bit and it may have no l_ast_data initialized yet,
1532 * so init it here with given osc_object.
1534 mdc_set_dom_lock_data(lock, cl2osc(obj));
1535 RETURN(mdc_dlm_canceling(env, lock));
1538 static const struct cl_object_operations mdc_ops = {
1539 .coo_page_init = osc_page_init,
1540 .coo_lock_init = mdc_lock_init,
1541 .coo_io_init = mdc_io_init,
1542 .coo_attr_get = mdc_attr_get,
1543 .coo_attr_update = osc_attr_update,
1544 .coo_glimpse = osc_object_glimpse,
1545 .coo_req_attr_set = mdc_req_attr_set,
1546 .coo_prune = mdc_object_prune,
1547 .coo_object_flush = mdc_object_flush
1550 static const struct osc_object_operations mdc_object_ops = {
1551 .oto_build_res_name = mdc_build_res_name,
1552 .oto_dlmlock_at_pgoff = mdc_dlmlock_at_pgoff,
1555 static int mdc_object_init(const struct lu_env *env, struct lu_object *obj,
1556 const struct lu_object_conf *conf)
1558 struct osc_object *osc = lu2osc(obj);
1560 if (osc->oo_initialized)
1563 osc->oo_initialized = true;
1565 return osc_object_init(env, obj, conf);
1568 static void mdc_object_free(const struct lu_env *env, struct lu_object *obj)
1570 osc_object_free(env, obj);
1573 static const struct lu_object_operations mdc_lu_obj_ops = {
1574 .loo_object_init = mdc_object_init,
1575 .loo_object_delete = NULL,
1576 .loo_object_release = NULL,
1577 .loo_object_free = mdc_object_free,
1578 .loo_object_print = osc_object_print,
1579 .loo_object_invariant = NULL
1582 struct lu_object *mdc_object_alloc(const struct lu_env *env,
1583 const struct lu_object_header *unused,
1584 struct lu_device *dev)
1586 struct osc_object *osc;
1587 struct lu_object *obj;
1589 OBD_SLAB_ALLOC_PTR_GFP(osc, osc_object_kmem, GFP_NOFS);
1592 lu_object_init(obj, NULL, dev);
1593 osc->oo_cl.co_ops = &mdc_ops;
1594 obj->lo_ops = &mdc_lu_obj_ops;
1595 osc->oo_obj_ops = &mdc_object_ops;
1596 osc->oo_initialized = false;
1603 static int mdc_process_config(const struct lu_env *env, struct lu_device *d,
1604 struct lustre_cfg *cfg)
1606 size_t count = class_modify_config(cfg, PARAM_MDC,
1607 &d->ld_obd->obd_kset.kobj);
1608 return count > 0 ? 0 : count;
1611 const struct lu_device_operations mdc_lu_ops = {
1612 .ldo_object_alloc = mdc_object_alloc,
1613 .ldo_process_config = mdc_process_config,
1614 .ldo_recovery_complete = NULL,
1617 static struct lu_device *mdc_device_alloc(const struct lu_env *env,
1618 struct lu_device_type *t,
1619 struct lustre_cfg *cfg)
1621 struct lu_device *d;
1622 struct osc_device *od;
1623 struct obd_device *obd;
1628 RETURN(ERR_PTR(-ENOMEM));
1630 cl_device_init(&od->od_cl, t);
1632 d->ld_ops = &mdc_lu_ops;
1635 obd = class_name2obd(lustre_cfg_string(cfg, 0));
1637 RETURN(ERR_PTR(-ENODEV));
1639 rc = mdc_setup(obd, cfg);
1641 osc_device_free(env, d);
1642 RETURN(ERR_PTR(rc));
1644 od->od_exp = obd->obd_self_export;
1648 static const struct lu_device_type_operations mdc_device_type_ops = {
1649 .ldto_device_alloc = mdc_device_alloc,
1650 .ldto_device_free = osc_device_free,
1651 .ldto_device_init = osc_device_init,
1652 .ldto_device_fini = osc_device_fini
1655 struct lu_device_type mdc_device_type = {
1656 .ldt_tags = LU_DEVICE_CL,
1657 .ldt_name = LUSTRE_MDC_NAME,
1658 .ldt_ops = &mdc_device_type_ops,
1659 .ldt_ctx_tags = LCT_CL_THREAD