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) 2012, 2017 Intel Corporation.
28 * Author: Mikhail Pershin <mike.pershin@intel.com>
31 #define DEBUG_SUBSYSTEM S_FILTER
33 #include <dt_object.h>
34 #include "mdt_internal.h"
36 /* functions below are stubs for now, they will be implemented with
37 * grant support on MDT */
38 static inline void mdt_io_counter_incr(struct obd_export *exp, int opcode,
39 char *jobid, long amount)
44 static inline void mdt_dom_read_lock(struct mdt_object *mo)
46 down_read(&mo->mot_dom_sem);
49 static inline void mdt_dom_read_unlock(struct mdt_object *mo)
51 up_read(&mo->mot_dom_sem);
54 static inline void mdt_dom_write_lock(struct mdt_object *mo)
56 down_write(&mo->mot_dom_sem);
59 static inline void mdt_dom_write_unlock(struct mdt_object *mo)
61 up_write(&mo->mot_dom_sem);
64 static int mdt_preprw_read(const struct lu_env *env, struct obd_export *exp,
65 struct mdt_device *mdt, struct mdt_object *mo,
66 struct lu_attr *la, int niocount,
67 struct niobuf_remote *rnb, int *nr_local,
68 struct niobuf_local *lnb, char *jobid)
70 struct dt_object *dob;
71 int i, j, rc, tot_bytes = 0;
75 mdt_dom_read_lock(mo);
76 if (!mdt_object_exists(mo))
77 GOTO(unlock, rc = -ENOENT);
80 /* parse remote buffers to local buffers and prepare the latter */
82 for (i = 0, j = 0; i < niocount; i++) {
83 rc = dt_bufs_get(env, dob, rnb + i, lnb + j, 0);
86 /* correct index for local buffers to continue with */
89 tot_bytes += rnb[i].rnb_len;
92 rc = dt_attr_get(env, dob, la);
96 rc = dt_read_prep(env, dob, lnb, *nr_local);
100 mdt_io_counter_incr(exp, LPROC_MDT_IO_READ, jobid, tot_bytes);
103 dt_bufs_put(env, dob, lnb, *nr_local);
105 mdt_dom_read_unlock(mo);
109 static int mdt_preprw_write(const struct lu_env *env, struct obd_export *exp,
110 struct mdt_device *mdt, struct mdt_object *mo,
111 struct lu_attr *la, struct obdo *oa,
112 int objcount, struct obd_ioobj *obj,
113 struct niobuf_remote *rnb, int *nr_local,
114 struct niobuf_local *lnb, char *jobid)
116 struct dt_object *dob;
117 int i, j, k, rc = 0, tot_bytes = 0;
121 /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
122 * space back if possible */
123 tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
125 mdt_dom_read_lock(mo);
126 if (!mdt_object_exists(mo)) {
127 CDEBUG(D_ERROR, "%s: BRW to missing obj "DFID"\n",
128 exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)));
129 GOTO(unlock, rc = -ENOENT);
132 dob = mdt_obj2dt(mo);
133 /* parse remote buffers to local buffers and prepare the latter */
135 for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
136 rc = dt_bufs_get(env, dob, rnb + i, lnb + j, 1);
137 if (unlikely(rc < 0))
139 /* correct index for local buffers to continue with */
140 for (k = 0; k < rc; k++) {
141 lnb[j + k].lnb_flags = rnb[i].rnb_flags;
142 if (!(rnb[i].rnb_flags & OBD_BRW_GRANTED))
143 lnb[j + k].lnb_rc = -ENOSPC;
147 tot_bytes += rnb[i].rnb_len;
150 rc = dt_write_prep(env, dob, lnb, *nr_local);
154 mdt_io_counter_incr(exp, LPROC_MDT_IO_WRITE, jobid, tot_bytes);
157 dt_bufs_put(env, dob, lnb, *nr_local);
159 mdt_dom_read_unlock(mo);
160 /* tgt_grant_prepare_write() was called, so we must commit */
161 tgt_grant_commit(exp, oa->o_grant_used, rc);
162 /* let's still process incoming grant information packed in the oa,
163 * but without enforcing grant since we won't proceed with the write.
164 * Just like a read request actually. */
165 tgt_grant_prepare_read(env, exp, oa);
169 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
170 struct obdo *oa, int objcount, struct obd_ioobj *obj,
171 struct niobuf_remote *rnb, int *nr_local,
172 struct niobuf_local *lnb)
174 struct tgt_session_info *tsi = tgt_ses_info(env);
175 struct mdt_thread_info *info = tsi2mdt_info(tsi);
176 struct lu_attr *la = &info->mti_attr.ma_attr;
177 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
178 struct mdt_object *mo;
182 /* The default value PTLRPC_MAX_BRW_PAGES is set in tgt_brw_write()
183 * but for MDT it is different, correct it here. */
184 if (*nr_local > MD_MAX_BRW_PAGES)
185 *nr_local = MD_MAX_BRW_PAGES;
187 jobid = tsi->tsi_jobid;
189 if (!oa || objcount != 1 || obj->ioo_bufcnt == 0) {
190 CERROR("%s: bad parameters %p/%i/%i\n",
191 exp->exp_obd->obd_name, oa, objcount, obj->ioo_bufcnt);
195 mo = mdt_object_find(env, mdt, &tsi->tsi_fid);
197 GOTO(out, rc = PTR_ERR(mo));
199 LASSERT(info->mti_object == NULL);
200 info->mti_object = mo;
202 if (cmd == OBD_BRW_WRITE) {
203 la_from_obdo(la, oa, OBD_MD_FLGETATTR);
204 rc = mdt_preprw_write(env, exp, mdt, mo, la, oa,
205 objcount, obj, rnb, nr_local, lnb,
207 } else if (cmd == OBD_BRW_READ) {
208 tgt_grant_prepare_read(env, exp, oa);
209 rc = mdt_preprw_read(env, exp, mdt, mo, la,
210 obj->ioo_bufcnt, rnb, nr_local, lnb,
212 obdo_from_la(oa, la, LA_ATIME);
214 CERROR("%s: wrong cmd %d received!\n",
215 exp->exp_obd->obd_name, cmd);
219 lu_object_put(env, &mo->mot_obj);
220 info->mti_object = NULL;
226 static int mdt_commitrw_read(const struct lu_env *env, struct mdt_device *mdt,
227 struct mdt_object *mo, int objcount, int niocount,
228 struct niobuf_local *lnb)
230 struct dt_object *dob;
235 LASSERT(niocount > 0);
237 dob = mdt_obj2dt(mo);
239 dt_bufs_put(env, dob, lnb, niocount);
241 mdt_dom_read_unlock(mo);
245 static int mdt_commitrw_write(const struct lu_env *env, struct obd_export *exp,
246 struct mdt_device *mdt, struct mdt_object *mo,
247 struct lu_attr *la, int objcount, int niocount,
248 struct niobuf_local *lnb, unsigned long granted,
251 struct dt_device *dt = mdt->mdt_bottom;
252 struct dt_object *dob;
260 dob = mdt_obj2dt(mo);
263 GOTO(out, rc = old_rc);
265 la->la_valid &= LA_ATIME | LA_MTIME | LA_CTIME;
267 if (!dt_object_exists(dob))
268 GOTO(out, rc = -ENOENT);
270 th = dt_trans_create(env, dt);
272 GOTO(out, rc = PTR_ERR(th));
274 for (i = 0; i < niocount; i++) {
275 if (!(lnb[i].lnb_flags & OBD_BRW_ASYNC)) {
281 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DQACQ_NET))
282 GOTO(out_stop, rc = -EINPROGRESS);
284 rc = dt_declare_write_commit(env, dob, lnb, niocount, th);
289 /* update [mac]time if needed */
290 rc = dt_declare_attr_set(env, dob, la, th);
295 rc = dt_trans_start(env, dt, th);
299 dt_write_lock(env, dob, 0);
300 rc = dt_write_commit(env, dob, lnb, niocount, th);
305 rc = dt_attr_set(env, dob, la, th);
309 /* get attr to return */
310 rc = dt_attr_get(env, dob, la);
312 dt_write_unlock(env, dob);
315 /* Force commit to make the just-deleted blocks
316 * reusable. LU-456 */
321 if (rc == 0 && granted > 0) {
322 if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
327 dt_trans_stop(env, dt, th);
328 if (rc == -ENOSPC && retries++ < 3) {
329 CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
335 dt_bufs_put(env, dob, lnb, niocount);
336 mdt_dom_read_unlock(mo);
338 tgt_grant_commit(exp, granted, old_rc);
342 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
345 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
346 struct ldlm_res_id resid;
347 struct ldlm_resource *res;
349 fid_build_reg_res_name(mdt_object_fid(mo), &resid);
350 res = ldlm_resource_get(mdt->mdt_namespace, NULL, &resid,
355 /* Update lvbo data if exists. */
356 if (mdt_dom_lvb_is_valid(res))
357 mdt_dom_disk_lvbo_update(env, mo, res, increase_only);
358 ldlm_resource_putref(res);
361 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
362 struct obdo *oa, int objcount, struct obd_ioobj *obj,
363 struct niobuf_remote *rnb, int npages,
364 struct niobuf_local *lnb, int old_rc)
366 struct mdt_thread_info *info = mdt_th_info(env);
367 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
368 struct mdt_object *mo = info->mti_object;
369 struct lu_attr *la = &info->mti_attr.ma_attr;
374 CERROR("%s: no pages to commit\n",
375 exp->exp_obd->obd_name);
381 if (cmd == OBD_BRW_WRITE) {
382 /* Don't update timestamps if this write is older than a
383 * setattr which modifies the timestamps. b=10150 */
385 /* XXX when we start having persistent reservations this needs
386 * to be changed to ofd_fmd_get() to create the fmd if it
387 * doesn't already exist so we can store the reservation handle
389 valid = OBD_MD_FLUID | OBD_MD_FLGID;
390 valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
392 la_from_obdo(la, oa, valid);
394 rc = mdt_commitrw_write(env, exp, mdt, mo, la, objcount,
395 npages, lnb, oa->o_grant_used, old_rc);
397 obdo_from_la(oa, la, VALID_FLAGS | LA_GID | LA_UID);
399 obdo_from_la(oa, la, LA_GID | LA_UID);
401 mdt_dom_obj_lvb_update(env, mo, false);
402 /* don't report overquota flag if we failed before reaching
404 if (old_rc == 0 && (rc == 0 || rc == -EDQUOT)) {
405 /* return the overquota flags to client */
406 if (lnb[0].lnb_flags & OBD_BRW_OVER_USRQUOTA) {
407 if (oa->o_valid & OBD_MD_FLFLAGS)
408 oa->o_flags |= OBD_FL_NO_USRQUOTA;
410 oa->o_flags = OBD_FL_NO_USRQUOTA;
413 if (lnb[0].lnb_flags & OBD_BRW_OVER_GRPQUOTA) {
414 if (oa->o_valid & OBD_MD_FLFLAGS)
415 oa->o_flags |= OBD_FL_NO_GRPQUOTA;
417 oa->o_flags = OBD_FL_NO_GRPQUOTA;
420 oa->o_valid |= OBD_MD_FLFLAGS | OBD_MD_FLUSRQUOTA |
423 } else if (cmd == OBD_BRW_READ) {
424 /* If oa != NULL then mdt_preprw_read updated the inode
425 * atime and we should update the lvb so that other glimpses
426 * will also get the updated value. bug 5972 */
428 mdt_dom_obj_lvb_update(env, mo, true);
429 rc = mdt_commitrw_read(env, mdt, mo, objcount, npages, lnb);
435 /* this put is pair to object_get in ofd_preprw_write */
436 mdt_thread_info_fini(info);
440 int mdt_object_punch(const struct lu_env *env, struct dt_device *dt,
441 struct dt_object *dob, __u64 start, __u64 end,
449 /* we support truncate, not punch yet */
450 LASSERT(end == OBD_OBJECT_EOF);
452 if (!dt_object_exists(dob))
455 th = dt_trans_create(env, dt);
459 rc = dt_declare_attr_set(env, dob, la, th);
463 rc = dt_declare_punch(env, dob, start, OBD_OBJECT_EOF, th);
467 tgt_vbr_obj_set(env, dob);
468 rc = dt_trans_start(env, dt, th);
472 dt_write_lock(env, dob, 0);
473 rc = dt_punch(env, dob, start, OBD_OBJECT_EOF, th);
476 rc = dt_attr_set(env, dob, la, th);
480 dt_write_unlock(env, dob);
483 dt_trans_stop(env, dt, th);
487 int mdt_punch_hdl(struct tgt_session_info *tsi)
489 const struct obdo *oa = &tsi->tsi_ost_body->oa;
490 struct ost_body *repbody;
491 struct mdt_thread_info *info;
493 struct ldlm_namespace *ns = tsi->tsi_tgt->lut_obd->obd_namespace;
494 struct obd_export *exp = tsi->tsi_exp;
495 struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
496 struct mdt_object *mo;
497 struct dt_object *dob;
499 struct lustre_handle lh = { 0, };
506 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
507 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
509 if ((oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
510 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
511 RETURN(err_serious(-EPROTO));
513 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
515 RETURN(err_serious(-ENOMEM));
517 /* punch start,end are passed in o_size,o_blocks throught wire */
521 if (end != OBD_OBJECT_EOF) /* Only truncate is supported */
524 info = tsi2mdt_info(tsi);
525 la = &info->mti_attr.ma_attr;
526 /* standard truncate optimization: if file body is completely
527 * destroyed, don't send data back to the server. */
529 flags |= LDLM_FL_AST_DISCARD_DATA;
531 repbody->oa.o_oi = oa->o_oi;
532 repbody->oa.o_valid = OBD_MD_FLID;
534 srvlock = (exp_connect_flags(exp) & OBD_CONNECT_SRVLOCK) &&
535 oa->o_valid & OBD_MD_FLFLAGS &&
536 oa->o_flags & OBD_FL_SRVLOCK;
539 rc = tgt_mdt_data_lock(ns, &tsi->tsi_resid, &lh, LCK_PW,
545 CDEBUG(D_INODE, "calling punch for object "DFID", valid = %#llx"
546 ", start = %lld, end = %lld\n", PFID(&tsi->tsi_fid),
547 oa->o_valid, start, end);
549 mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
551 GOTO(out_unlock, rc = PTR_ERR(mo));
553 mdt_dom_write_lock(mo);
554 if (!mdt_object_exists(mo))
555 GOTO(out_put, rc = -ENOENT);
556 dob = mdt_obj2dt(mo);
558 la_from_obdo(la, oa, OBD_MD_FLMTIME | OBD_MD_FLATIME | OBD_MD_FLCTIME);
560 la->la_valid |= LA_SIZE;
562 rc = mdt_object_punch(tsi->tsi_env, mdt->mdt_bottom, dob,
564 mdt_dom_write_unlock(mo);
568 mdt_dom_obj_lvb_update(tsi->tsi_env, mo, false);
569 mdt_io_counter_incr(tsi->tsi_exp, LPROC_MDT_IO_PUNCH,
573 lu_object_put(tsi->tsi_env, &mo->mot_obj);
576 mdt_save_lock(info, &lh, LCK_PW, rc);
578 mdt_thread_info_fini(info);
583 * MDT glimpse for Data-on-MDT
585 * If there is write lock on client then function issues glimpse_ast to get
586 * an actual size from that client.
589 int mdt_do_glimpse(const struct lu_env *env, struct ldlm_namespace *ns,
590 struct ldlm_resource *res)
592 union ldlm_policy_data policy;
593 struct lustre_handle lockh;
595 struct ldlm_lock *lock;
596 struct ldlm_glimpse_work *gl_work;
597 struct list_head gl_list;
602 /* There can be only one write lock covering data, try to match it. */
603 policy.l_inodebits.bits = MDS_INODELOCK_DOM;
604 mode = ldlm_lock_match(ns, LDLM_FL_BLOCK_GRANTED | LDLM_FL_TEST_LOCK,
605 &res->lr_name, LDLM_IBITS, &policy,
608 /* There is no PW lock on this object; finished. */
612 lock = ldlm_handle2lock(&lockh);
617 * This check is for lock taken in mdt_reint_unlink() that does
618 * not have l_glimpse_ast set. So the logic is: if there is a lock
619 * with no l_glimpse_ast set, this object is being destroyed already.
620 * Hence, if you are grabbing DLM locks on the server, always set
621 * non-NULL glimpse_ast (e.g., ldlm_request.c::ldlm_glimpse_ast()).
623 if (lock->l_glimpse_ast == NULL) {
624 LDLM_DEBUG(lock, "no l_glimpse_ast");
625 GOTO(out, rc = -ENOENT);
628 OBD_SLAB_ALLOC_PTR_GFP(gl_work, ldlm_glimpse_work_kmem, GFP_ATOMIC);
630 GOTO(out, rc = -ENOMEM);
632 /* Populate the gl_work structure.
633 * Grab additional reference on the lock which will be released in
634 * ldlm_work_gl_ast_lock() */
635 gl_work->gl_lock = LDLM_LOCK_GET(lock);
636 /* The glimpse callback is sent to one single IO lock. As a result,
637 * the gl_work list is just composed of one element */
638 INIT_LIST_HEAD(&gl_list);
639 list_add_tail(&gl_work->gl_list, &gl_list);
640 /* There is actually no need for a glimpse descriptor when glimpsing
642 gl_work->gl_desc = NULL;
643 /* the ldlm_glimpse_work structure is allocated on the stack */
644 gl_work->gl_flags = LDLM_GL_WORK_SLAB_ALLOCATED;
646 ldlm_glimpse_locks(res, &gl_list); /* this will update the LVB */
648 /* If the list is not empty, we failed to glimpse a lock and
649 * must clean it up. Usually due to a race with unlink.*/
650 if (!list_empty(&gl_list)) {
651 LDLM_LOCK_RELEASE(lock);
652 OBD_SLAB_FREE_PTR(gl_work, ldlm_glimpse_work_kmem);
661 static void mdt_lvb2body(struct ldlm_resource *res, struct mdt_body *mb)
663 struct ost_lvb *res_lvb;
666 res_lvb = res->lr_lvb_data;
667 mb->mbo_dom_size = res_lvb->lvb_size;
668 mb->mbo_dom_blocks = res_lvb->lvb_blocks;
669 mb->mbo_mtime = res_lvb->lvb_mtime;
670 mb->mbo_ctime = res_lvb->lvb_ctime;
671 mb->mbo_atime = res_lvb->lvb_atime;
673 CDEBUG(D_DLMTRACE, "size %llu\n", res_lvb->lvb_size);
675 mb->mbo_valid |= OBD_MD_FLATIME | OBD_MD_FLCTIME | OBD_MD_FLMTIME |
681 * MDT glimpse for Data-on-MDT
683 * This function is called when MDT get attributes for the DoM object.
684 * If there is write lock on client then function issues glimpse_ast to get
685 * an actual size from that client.
687 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
688 const struct lu_fid *fid, struct mdt_body *mb,
691 struct ldlm_res_id resid;
692 struct ldlm_resource *res;
697 fid_build_reg_res_name(fid, &resid);
698 res = ldlm_resource_get(mdt->mdt_namespace, NULL, &resid,
703 /* Update lvbo data if DoM lock returned or if LVB is not yet valid. */
704 if (dom_lock || !mdt_dom_lvb_is_valid(res))
705 mdt_dom_lvbo_update(res, NULL, NULL, false);
707 mdt_lvb2body(res, mb);
708 ldlm_resource_putref(res);
713 * MDT DoM lock intent policy (glimpse)
715 * Intent policy is called when lock has an intent, for DoM file that
716 * means glimpse lock and policy fills Lock Value Block (LVB).
718 * If already granted lock is found it will be placed in \a lockp and
719 * returned back to caller function.
721 * \param[in] tsi session info
722 * \param[in,out] lockp pointer to the lock
723 * \param[in] flags LDLM flags
725 * \retval ELDLM_LOCK_REPLACED if already granted lock was found
726 * and placed in \a lockp
727 * \retval ELDLM_LOCK_ABORTED in other cases except error
728 * \retval negative value on error
730 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
731 struct ldlm_lock **lockp, __u64 flags)
733 struct ldlm_lock *lock = *lockp;
734 struct ldlm_resource *res = lock->l_resource;
735 ldlm_processing_policy policy;
736 struct ldlm_reply *rep;
737 struct mdt_body *mbo;
742 policy = ldlm_get_processing_policy(res);
743 LASSERT(policy != NULL);
745 req_capsule_set_size(mti->mti_pill, &RMF_MDT_MD, RCL_SERVER, 0);
746 req_capsule_set_size(mti->mti_pill, &RMF_ACL, RCL_SERVER, 0);
747 rc = req_capsule_server_pack(mti->mti_pill);
749 RETURN(err_serious(rc));
751 rep = req_capsule_server_get(mti->mti_pill, &RMF_DLM_REP);
755 mbo = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
760 /* Check if this is a resend case (MSG_RESENT is set on RPC) and a
761 * lock was found by ldlm_handle_enqueue(); if so no need to grant
763 if (flags & LDLM_FL_RESENT) {
764 rc = LDLM_ITER_CONTINUE;
769 rc = policy(lock, &tmpflags, LDLM_PROCESS_RESCAN, &err, NULL);
770 check_res_locked(res);
774 /* The lock met with no resistance; we're finished. */
775 if (rc == LDLM_ITER_CONTINUE) {
776 GOTO(fill_mbo, rc = ELDLM_LOCK_REPLACED);
777 } else if (flags & LDLM_FL_BLOCK_NOWAIT) {
778 /* LDLM_FL_BLOCK_NOWAIT means it is for AGL. Do not send glimpse
779 * callback for glimpse size. The real size user will trigger
780 * the glimpse callback when necessary. */
781 GOTO(fill_mbo, rc = ELDLM_LOCK_ABORTED);
784 rc = mdt_do_glimpse(mti->mti_env, ns, res);
786 /* We are racing with unlink(); just return -ENOENT */
787 rep->lock_policy_res2 = ptlrpc_status_hton(-ENOENT);
789 } else if (rc == -EINVAL) {
790 /* this is possible is client lock has been cancelled but
791 * still exists on server. If that lock was found on server
792 * as only conflicting lock then the client has already
793 * size authority and glimpse is not needed. */
794 CDEBUG(D_DLMTRACE, "Glimpse from the client owning lock\n");
799 rc = ELDLM_LOCK_ABORTED;
801 /* LVB can be without valid data in case of DOM */
802 if (!mdt_dom_lvb_is_valid(res))
803 mdt_dom_lvbo_update(res, lock, NULL, false);
804 mdt_lvb2body(res, mbo);
808 int mdt_brw_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
809 struct ldlm_lock **lockp, __u64 flags)
811 struct tgt_session_info *tsi = tgt_ses_info(mti->mti_env);
812 struct lu_fid *fid = &tsi->tsi_fid;
813 struct ldlm_lock *lock = *lockp;
814 struct ldlm_resource *res = lock->l_resource;
815 struct ldlm_reply *rep;
816 struct mdt_body *mbo;
817 struct mdt_lock_handle *lhc = &mti->mti_lh[MDT_LH_RMT];
818 struct mdt_object *mo;
823 /* Get lock from request for possible resent case. */
824 mdt_intent_fixup_resent(mti, *lockp, lhc, flags);
825 req_capsule_set_size(mti->mti_pill, &RMF_MDT_MD, RCL_SERVER, 0);
826 req_capsule_set_size(mti->mti_pill, &RMF_ACL, RCL_SERVER, 0);
827 rc = req_capsule_server_pack(mti->mti_pill);
829 RETURN(err_serious(rc));
831 rep = req_capsule_server_get(mti->mti_pill, &RMF_DLM_REP);
835 mbo = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY);
839 fid_extract_from_res_name(fid, &res->lr_name);
840 mo = mdt_object_find(mti->mti_env, mti->mti_mdt, fid);
841 if (unlikely(IS_ERR(mo)))
844 if (!mdt_object_exists(mo))
845 GOTO(out, rc = -ENOENT);
847 if (mdt_object_remote(mo))
848 GOTO(out, rc = -EPROTO);
851 if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
852 mdt_lock_handle_init(lhc);
853 mdt_lock_reg_init(lhc, (*lockp)->l_req_mode);
854 /* This will block MDT thread but it should be fine until
855 * client caches small amount of data for DoM, which should be
856 * smaller than one BRW RPC and should be able to be
857 * piggybacked by lock cancel RPC.
858 * If the client could hold the lock too long, this code can be
859 * revised to call mdt_object_lock_try(). And if fails, it will
860 * return ELDLM_OK here and fall back into normal lock enqueue
863 rc = mdt_object_lock(mti, mo, lhc, MDS_INODELOCK_DOM);
868 if (!mdt_dom_lvb_is_valid(res)) {
869 rc = mdt_dom_lvb_alloc(res);
872 mdt_dom_disk_lvbo_update(mti->mti_env, mo, res, false);
874 mdt_lvb2body(res, mbo);
876 rep->lock_policy_res2 = clear_serious(rc);
877 if (rep->lock_policy_res2) {
878 lhc->mlh_reg_lh.cookie = 0ull;
879 GOTO(out, rc = ELDLM_LOCK_ABORTED);
882 rc = mdt_intent_lock_replace(mti, lockp, lhc, flags, rc);
884 mdt_object_put(mti->mti_env, mo);
888 void mdt_dom_discard_data(struct mdt_thread_info *info,
889 const struct lu_fid *fid)
891 struct mdt_device *mdt = info->mti_mdt;
892 union ldlm_policy_data *policy = &info->mti_policy;
893 struct ldlm_res_id *res_id = &info->mti_res_id;
894 struct lustre_handle dom_lh;
895 __u64 flags = LDLM_FL_AST_DISCARD_DATA;
898 policy->l_inodebits.bits = MDS_INODELOCK_DOM;
899 policy->l_inodebits.try_bits = 0;
900 fid_build_reg_res_name(fid, res_id);
902 /* Tell the clients that the object is gone now and that they should
903 * throw away any cached pages. */
904 rc = ldlm_cli_enqueue_local(mdt->mdt_namespace, res_id, LDLM_IBITS,
905 policy, LCK_PW, &flags, ldlm_blocking_ast,
906 ldlm_completion_ast, NULL, NULL, 0,
907 LVB_T_NONE, NULL, &dom_lh);
909 /* We only care about the side-effects, just drop the lock. */
911 ldlm_lock_decref(&dom_lh, LCK_PW);
914 /* check if client has already DoM lock for given resource */
915 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
916 const struct lu_fid *fid)
918 struct mdt_device *mdt = info->mti_mdt;
919 union ldlm_policy_data *policy = &info->mti_policy;
920 struct ldlm_res_id *res_id = &info->mti_res_id;
921 struct lustre_handle lockh;
923 struct ldlm_lock *lock;
926 policy->l_inodebits.bits = MDS_INODELOCK_DOM;
927 fid_build_reg_res_name(fid, res_id);
929 mode = ldlm_lock_match(mdt->mdt_namespace, LDLM_FL_BLOCK_GRANTED |
930 LDLM_FL_TEST_LOCK, res_id, LDLM_IBITS, policy,
933 /* There is no other PW lock on this object; finished. */
937 lock = ldlm_handle2lock(&lockh);
941 /* check if lock from the same client */
942 rc = (lock->l_export->exp_handle.h_cookie ==
943 info->mti_exp->exp_handle.h_cookie);