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) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
32 #define DEBUG_SUBSYSTEM S_LMV
34 #include <linux/file.h>
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/user_namespace.h>
38 #include <linux/uidgid.h>
39 #include <linux/slab.h>
40 #include <linux/pagemap.h>
42 #include <linux/math64.h>
43 #include <linux/seq_file.h>
44 #include <linux/namei.h>
46 #include <obd_support.h>
47 #include <lustre_lib.h>
48 #include <lustre_net.h>
49 #include <obd_class.h>
50 #include <lustre_lmv.h>
51 #include <lprocfs_status.h>
52 #include <cl_object.h>
53 #include <lustre_fid.h>
54 #include <uapi/linux/lustre/lustre_ioctl.h>
55 #include <lustre_kernelcomm.h>
56 #include "lmv_internal.h"
58 static int lmv_check_connect(struct obd_device *obd);
60 void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt,
63 if (tgt->ltd_active == activate)
66 tgt->ltd_active = activate;
67 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count +=
70 tgt->ltd_exp->exp_obd->obd_inactive = !activate;
76 * -EINVAL : UUID can't be found in the LMV's target list
77 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
78 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
80 static int lmv_set_mdc_active(struct lmv_obd *lmv,
81 const struct obd_uuid *uuid,
84 struct lu_tgt_desc *tgt = NULL;
85 struct obd_device *obd;
90 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
91 lmv, uuid->uuid, activate);
93 spin_lock(&lmv->lmv_lock);
94 lmv_foreach_connected_tgt(lmv, tgt) {
95 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n",
96 tgt->ltd_index, tgt->ltd_uuid.uuid,
97 tgt->ltd_exp->exp_handle.h_cookie);
99 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
104 GOTO(out_lmv_lock, rc = -EINVAL);
106 obd = class_exp2obd(tgt->ltd_exp);
108 GOTO(out_lmv_lock, rc = -ENOTCONN);
110 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
111 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
112 obd->obd_type->typ_name, tgt->ltd_index);
113 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
115 if (tgt->ltd_active == activate) {
116 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
117 activate ? "" : "in");
118 GOTO(out_lmv_lock, rc);
121 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
122 activate ? "" : "in");
123 lmv_activate_target(lmv, tgt, activate);
127 spin_unlock(&lmv->lmv_lock);
131 static struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
133 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
134 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
136 return tgt ? obd_get_uuid(tgt->ltd_exp) : NULL;
139 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
140 enum obd_notify_event ev)
142 struct obd_connect_data *conn_data;
143 struct lmv_obd *lmv = &obd->u.lmv;
144 struct obd_uuid *uuid;
148 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
149 CERROR("unexpected notification of %s %s!\n",
150 watched->obd_type->typ_name,
155 uuid = &watched->u.cli.cl_target_uuid;
156 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
158 * Set MDC as active before notifying the observer, so the
159 * observer can use the MDC normally.
161 rc = lmv_set_mdc_active(lmv, uuid,
162 ev == OBD_NOTIFY_ACTIVE);
164 CERROR("%sactivation of %s failed: %d\n",
165 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
169 } else if (ev == OBD_NOTIFY_OCD) {
170 conn_data = &watched->u.cli.cl_import->imp_connect_data;
172 * XXX: Make sure that ocd_connect_flags from all targets are
173 * the same. Otherwise one of MDTs runs wrong version or
174 * something like this. --umka
176 obd->obd_self_export->exp_connect_data = *conn_data;
180 * Pass the notification up the chain.
182 if (obd->obd_observer)
183 rc = obd_notify(obd->obd_observer, watched, ev);
188 static int lmv_connect(const struct lu_env *env,
189 struct obd_export **pexp, struct obd_device *obd,
190 struct obd_uuid *cluuid, struct obd_connect_data *data,
193 struct lmv_obd *lmv = &obd->u.lmv;
194 struct lustre_handle conn = { 0 };
195 struct obd_export *exp;
199 rc = class_connect(&conn, obd, cluuid);
201 CERROR("class_connection() returned %d\n", rc);
205 exp = class_conn2export(&conn);
208 lmv->conn_data = *data;
209 lmv->lmv_cache = localdata;
211 lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
212 &obd->obd_kset.kobj);
213 if (!lmv->lmv_tgts_kobj) {
214 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
215 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
218 rc = lmv_check_connect(obd);
227 if (lmv->lmv_tgts_kobj)
228 kobject_put(lmv->lmv_tgts_kobj);
230 class_disconnect(exp);
235 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
238 struct obd_device *obd = exp->exp_obd;
239 struct lmv_obd *lmv = &obd->u.lmv;
240 struct lmv_tgt_desc *tgt;
246 if (lmv->max_easize < easize) {
247 lmv->max_easize = easize;
250 if (lmv->max_def_easize < def_easize) {
251 lmv->max_def_easize = def_easize;
258 if (lmv->connected == 0)
261 lmv_foreach_connected_tgt(lmv, tgt) {
262 if (!tgt->ltd_active)
265 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
267 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
268 " rc = %d\n", obd->obd_name, tgt->ltd_index, rc);
275 #define MAX_STRING_SIZE 128
277 static int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
279 struct lmv_obd *lmv = &obd->u.lmv;
280 struct obd_device *mdc_obd;
281 struct obd_export *mdc_exp;
282 struct lu_fld_target target;
286 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
289 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
293 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
294 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
295 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
297 if (!mdc_obd->obd_set_up) {
298 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
302 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
303 &lmv->conn_data, lmv->lmv_cache);
305 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
310 * Init fid sequence client for this mdc and add new fld target.
312 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
316 target.ft_srv = NULL;
317 target.ft_exp = mdc_exp;
318 target.ft_idx = tgt->ltd_index;
320 fld_client_add_target(&lmv->lmv_fld, &target);
322 rc = obd_register_observer(mdc_obd, obd);
324 obd_disconnect(mdc_exp);
325 CERROR("target %s register_observer error %d\n",
326 tgt->ltd_uuid.uuid, rc);
330 if (obd->obd_observer) {
332 * Tell the observer about the new target.
334 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
337 obd_disconnect(mdc_exp);
343 tgt->ltd_exp = mdc_exp;
344 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count++;
346 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
348 rc = lu_qos_add_tgt(&lmv->lmv_qos, tgt);
350 obd_disconnect(mdc_exp);
354 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
355 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
356 atomic_read(&obd->obd_refcount));
358 lmv_statfs_check_update(obd, tgt);
360 if (lmv->lmv_tgts_kobj)
361 /* Even if we failed to create the link, that's fine */
362 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
363 &mdc_obd->obd_kset.kobj,
368 static void lmv_del_target(struct lmv_obd *lmv, struct lu_tgt_desc *tgt)
371 ltd_del_tgt(&lmv->lmv_mdt_descs, tgt);
375 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
376 __u32 index, int gen)
378 struct obd_device *mdc_obd;
379 struct lmv_obd *lmv = &obd->u.lmv;
380 struct lmv_tgt_desc *tgt;
381 struct lu_tgt_descs *ltd = &lmv->lmv_mdt_descs;
386 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
387 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
390 CERROR("%s: Target %s not attached: rc = %d\n",
391 obd->obd_name, uuidp->uuid, -EINVAL);
399 mutex_init(&tgt->ltd_fid_mutex);
400 tgt->ltd_index = index;
401 tgt->ltd_uuid = *uuidp;
404 mutex_lock(<d->ltd_mutex);
405 rc = ltd_add_tgt(ltd, tgt);
406 mutex_unlock(<d->ltd_mutex);
412 /* lmv_check_connect() will connect this target. */
415 rc = lmv_connect_mdc(obd, tgt);
417 int easize = sizeof(struct lmv_stripe_md) +
418 lmv->lmv_mdt_count * sizeof(struct lu_fid);
420 lmv_init_ea_size(obd->obd_self_export, easize, 0);
430 static int lmv_check_connect(struct obd_device *obd)
432 struct lmv_obd *lmv = &obd->u.lmv;
433 struct lmv_tgt_desc *tgt;
442 mutex_lock(&lmv->lmv_mdt_descs.ltd_mutex);
444 GOTO(unlock, rc = 0);
446 if (!lmv->lmv_mdt_count) {
447 CERROR("%s: no targets configured: rc = -EINVAL\n",
449 GOTO(unlock, rc = -EINVAL);
452 if (!lmv_mdt0_inited(lmv)) {
453 CERROR("%s: no target configured for index 0: rc = -EINVAL.\n",
455 GOTO(unlock, rc = -EINVAL);
458 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
459 obd->obd_uuid.uuid, obd->obd_name);
461 lmv_foreach_tgt(lmv, tgt) {
462 rc = lmv_connect_mdc(obd, tgt);
468 easize = lmv_mds_md_size(lmv->lmv_mdt_count, LMV_MAGIC);
469 lmv_init_ea_size(obd->obd_self_export, easize, 0);
472 mutex_unlock(&lmv->lmv_mdt_descs.ltd_mutex);
477 lmv_foreach_tgt(lmv, tgt) {
482 --lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count;
483 obd_disconnect(tgt->ltd_exp);
489 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
491 struct lmv_obd *lmv = &obd->u.lmv;
492 struct obd_device *mdc_obd;
496 LASSERT(tgt != NULL);
497 LASSERT(obd != NULL);
499 mdc_obd = class_exp2obd(tgt->ltd_exp);
502 mdc_obd->obd_force = obd->obd_force;
503 mdc_obd->obd_fail = obd->obd_fail;
504 mdc_obd->obd_no_recov = obd->obd_no_recov;
506 if (lmv->lmv_tgts_kobj)
507 sysfs_remove_link(lmv->lmv_tgts_kobj,
511 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
513 CERROR("Can't finalize fids factory\n");
515 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
516 tgt->ltd_exp->exp_obd->obd_name,
517 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
519 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
520 rc = obd_disconnect(tgt->ltd_exp);
522 if (tgt->ltd_active) {
523 CERROR("Target %s disconnect error %d\n",
524 tgt->ltd_uuid.uuid, rc);
528 lmv_activate_target(lmv, tgt, 0);
533 static int lmv_disconnect(struct obd_export *exp)
535 struct obd_device *obd = class_exp2obd(exp);
536 struct lmv_obd *lmv = &obd->u.lmv;
537 struct lmv_tgt_desc *tgt;
542 lmv_foreach_connected_tgt(lmv, tgt)
543 lmv_disconnect_mdc(obd, tgt);
545 if (lmv->lmv_tgts_kobj)
546 kobject_put(lmv->lmv_tgts_kobj);
549 class_export_put(exp);
550 rc = class_disconnect(exp);
556 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
559 struct obd_device *obd = class_exp2obd(exp);
560 struct lmv_obd *lmv = &obd->u.lmv;
561 struct getinfo_fid2path *gf;
562 struct lmv_tgt_desc *tgt;
563 struct getinfo_fid2path *remote_gf = NULL;
564 struct lu_fid root_fid;
565 int remote_gf_size = 0;
569 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
571 RETURN(PTR_ERR(tgt));
573 root_fid = *gf->gf_u.gf_root_fid;
574 LASSERT(fid_is_sane(&root_fid));
577 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
578 if (rc != 0 && rc != -EREMOTE)
579 GOTO(out_fid2path, rc);
581 /* If remote_gf != NULL, it means just building the
582 * path on the remote MDT, copy this path segement to gf */
583 if (remote_gf != NULL) {
584 struct getinfo_fid2path *ori_gf;
588 ori_gf = (struct getinfo_fid2path *)karg;
589 if (strlen(ori_gf->gf_u.gf_path) + 1 +
590 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
591 GOTO(out_fid2path, rc = -EOVERFLOW);
593 ptr = ori_gf->gf_u.gf_path;
595 len = strlen(gf->gf_u.gf_path);
596 /* move the current path to the right to release space
597 * for closer-to-root part */
598 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
599 memcpy(ptr, gf->gf_u.gf_path, len);
603 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
604 tgt->ltd_exp->exp_obd->obd_name,
605 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
609 GOTO(out_fid2path, rc);
611 /* sigh, has to go to another MDT to do path building further */
612 if (remote_gf == NULL) {
613 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
614 OBD_ALLOC(remote_gf, remote_gf_size);
615 if (remote_gf == NULL)
616 GOTO(out_fid2path, rc = -ENOMEM);
617 remote_gf->gf_pathlen = PATH_MAX;
620 if (!fid_is_sane(&gf->gf_fid)) {
621 CERROR("%s: invalid FID "DFID": rc = %d\n",
622 tgt->ltd_exp->exp_obd->obd_name,
623 PFID(&gf->gf_fid), -EINVAL);
624 GOTO(out_fid2path, rc = -EINVAL);
627 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
629 GOTO(out_fid2path, rc = -EINVAL);
631 remote_gf->gf_fid = gf->gf_fid;
632 remote_gf->gf_recno = -1;
633 remote_gf->gf_linkno = -1;
634 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
635 *remote_gf->gf_u.gf_root_fid = root_fid;
637 goto repeat_fid2path;
640 if (remote_gf != NULL)
641 OBD_FREE(remote_gf, remote_gf_size);
645 static int lmv_hsm_req_count(struct lmv_obd *lmv,
646 const struct hsm_user_request *hur,
647 const struct lmv_tgt_desc *tgt_mds)
649 struct lmv_tgt_desc *curr_tgt;
653 /* count how many requests must be sent to the given target */
654 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
655 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
656 if (IS_ERR(curr_tgt))
657 RETURN(PTR_ERR(curr_tgt));
658 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
664 static int lmv_hsm_req_build(struct lmv_obd *lmv,
665 struct hsm_user_request *hur_in,
666 const struct lmv_tgt_desc *tgt_mds,
667 struct hsm_user_request *hur_out)
670 struct lmv_tgt_desc *curr_tgt;
672 /* build the hsm_user_request for the given target */
673 hur_out->hur_request = hur_in->hur_request;
675 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
676 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
677 if (IS_ERR(curr_tgt))
678 RETURN(PTR_ERR(curr_tgt));
679 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
680 hur_out->hur_user_item[nr_out] =
681 hur_in->hur_user_item[i];
685 hur_out->hur_request.hr_itemcount = nr_out;
686 memcpy(hur_data(hur_out), hur_data(hur_in),
687 hur_in->hur_request.hr_data_len);
692 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
693 int len, struct lustre_kernelcomm *lk,
696 struct lmv_obd *lmv = &obd->u.lmv;
697 struct lu_tgt_desc *tgt;
702 /* unregister request (call from llapi_hsm_copytool_fini) */
703 lmv_foreach_connected_tgt(lmv, tgt)
704 /* best effort: try to clean as much as possible
705 * (continue on error) */
706 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
708 /* Whatever the result, remove copytool from kuc groups.
709 * Unreached coordinators will get EPIPE on next requests
710 * and will unregister automatically.
712 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
717 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
718 int len, struct lustre_kernelcomm *lk,
721 struct lmv_obd *lmv = &obd->u.lmv;
723 bool any_set = false;
724 struct kkuc_ct_data *kcd;
726 struct lu_tgt_desc *tgt;
733 filp = fget(lk->lk_wfd);
737 if (lk->lk_flags & LK_FLG_DATANR)
738 kcd_size = offsetof(struct kkuc_ct_data,
739 kcd_archives[lk->lk_data_count]);
741 kcd_size = sizeof(*kcd);
743 OBD_ALLOC(kcd, kcd_size);
745 GOTO(err_fput, rc = -ENOMEM);
747 kcd->kcd_nr_archives = lk->lk_data_count;
748 if (lk->lk_flags & LK_FLG_DATANR) {
749 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
750 if (lk->lk_data_count > 0)
751 memcpy(kcd->kcd_archives, lk->lk_data,
752 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
754 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
757 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
758 lk->lk_group, kcd, kcd_size);
759 OBD_FREE(kcd, kcd_size);
763 /* All or nothing: try to register to all MDS.
764 * In case of failure, unregister from previous MDS,
765 * except if it because of inactive target. */
766 lmv_foreach_connected_tgt(lmv, tgt) {
767 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
769 if (tgt->ltd_active) {
770 /* permanent error */
771 CERROR("%s: iocontrol MDC %s on MDT"
772 " idx %d cmd %x: err = %d\n",
773 lmv2obd_dev(lmv)->obd_name,
774 tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
777 lk->lk_flags |= LK_FLG_STOP;
779 /* unregister from previous MDS */
780 lmv_foreach_connected_tgt(lmv, tgt) {
781 if (tgt->ltd_index >= i)
784 obd_iocontrol(cmd, tgt->ltd_exp, len,
787 GOTO(err_kkuc_rem, rc);
789 /* else: transient error.
790 * kuc will register to the missing MDT
798 /* no registration done: return error */
799 GOTO(err_kkuc_rem, rc = -ENOTCONN);
804 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
811 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
812 int len, void *karg, void __user *uarg)
814 struct obd_device *obd = class_exp2obd(exp);
815 struct lmv_obd *lmv = &obd->u.lmv;
816 struct lu_tgt_desc *tgt = NULL;
818 __u32 count = lmv->lmv_mdt_count;
827 case IOC_OBD_STATFS: {
828 struct obd_ioctl_data *data = karg;
829 struct obd_device *mdc_obd;
830 struct obd_statfs stat_buf = {0};
833 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
835 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
838 tgt = lmv_tgt(lmv, index);
842 if (!tgt->ltd_active)
845 mdc_obd = class_exp2obd(tgt->ltd_exp);
850 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
851 min((int) data->ioc_plen2,
852 (int) sizeof(struct obd_uuid))))
855 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
856 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
860 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
861 min((int) data->ioc_plen1,
862 (int) sizeof(stat_buf))))
866 case OBD_IOC_QUOTACTL: {
867 struct if_quotactl *qctl = karg;
868 struct obd_quotactl *oqctl;
870 if (qctl->qc_valid == QC_MDTIDX) {
871 tgt = lmv_tgt(lmv, qctl->qc_idx);
872 } else if (qctl->qc_valid == QC_UUID) {
873 lmv_foreach_tgt(lmv, tgt) {
874 if (!obd_uuid_equals(&tgt->ltd_uuid,
887 if (!tgt || !tgt->ltd_exp)
890 OBD_ALLOC_PTR(oqctl);
894 QCTL_COPY(oqctl, qctl);
895 rc = obd_quotactl(tgt->ltd_exp, oqctl);
897 QCTL_COPY(qctl, oqctl);
898 qctl->qc_valid = QC_MDTIDX;
899 qctl->obd_uuid = tgt->ltd_uuid;
904 case LL_IOC_GET_CONNECT_FLAGS: {
905 tgt = lmv_tgt(lmv, 0);
907 if (tgt && tgt->ltd_exp)
908 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
911 case LL_IOC_FID2MDTIDX: {
912 struct lu_fid *fid = karg;
915 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
919 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
920 * point to user space memory for FID2MDTIDX. */
921 *(__u32 *)uarg = mdt_index;
924 case OBD_IOC_FID2PATH: {
925 rc = lmv_fid2path(exp, len, karg, uarg);
928 case LL_IOC_HSM_STATE_GET:
929 case LL_IOC_HSM_STATE_SET:
930 case LL_IOC_HSM_ACTION: {
931 struct md_op_data *op_data = karg;
933 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
935 RETURN(PTR_ERR(tgt));
937 if (tgt->ltd_exp == NULL)
940 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
943 case LL_IOC_HSM_PROGRESS: {
944 const struct hsm_progress_kernel *hpk = karg;
946 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
948 RETURN(PTR_ERR(tgt));
949 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
952 case LL_IOC_HSM_REQUEST: {
953 struct hsm_user_request *hur = karg;
954 unsigned int reqcount = hur->hur_request.hr_itemcount;
959 /* if the request is about a single fid
960 * or if there is a single MDS, no need to split
962 if (reqcount == 1 || count == 1) {
963 tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
965 RETURN(PTR_ERR(tgt));
966 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
968 /* split fid list to their respective MDS */
969 lmv_foreach_connected_tgt(lmv, tgt) {
972 struct hsm_user_request *req;
974 nr = lmv_hsm_req_count(lmv, hur, tgt);
977 if (nr == 0) /* nothing for this MDS */
980 /* build a request with fids for this MDS */
981 reqlen = offsetof(typeof(*hur),
983 + hur->hur_request.hr_data_len;
984 OBD_ALLOC_LARGE(req, reqlen);
987 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
989 GOTO(hsm_req_err, rc1);
990 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
993 if (rc1 != 0 && rc == 0)
995 OBD_FREE_LARGE(req, reqlen);
1000 case LL_IOC_LOV_SWAP_LAYOUTS: {
1001 struct md_op_data *op_data = karg;
1002 struct lmv_tgt_desc *tgt1, *tgt2;
1004 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1006 RETURN(PTR_ERR(tgt1));
1008 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1010 RETURN(PTR_ERR(tgt2));
1012 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1015 /* only files on same MDT can have their layouts swapped */
1016 if (tgt1->ltd_index != tgt2->ltd_index)
1019 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1022 case LL_IOC_HSM_CT_START: {
1023 struct lustre_kernelcomm *lk = karg;
1024 if (lk->lk_flags & LK_FLG_STOP)
1025 rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1027 rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1031 lmv_foreach_connected_tgt(lmv, tgt) {
1032 struct obd_device *mdc_obd;
1035 /* ll_umount_begin() sets force flag but for lmv, not
1036 * mdc. Let's pass it through */
1037 mdc_obd = class_exp2obd(tgt->ltd_exp);
1038 mdc_obd->obd_force = obd->obd_force;
1039 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1041 if (tgt->ltd_active) {
1042 CERROR("error: iocontrol MDC %s on MDT"
1043 " idx %d cmd %x: err = %d\n",
1045 tgt->ltd_index, cmd, err);
1058 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1059 struct lu_fid *fid, struct md_op_data *op_data)
1061 struct obd_device *obd = class_exp2obd(exp);
1062 struct lmv_obd *lmv = &obd->u.lmv;
1063 struct lmv_tgt_desc *tgt;
1071 tgt = lmv_tgt(lmv, op_data->op_mds);
1075 if (!tgt->ltd_active || !tgt->ltd_exp)
1079 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1080 * on server that seq in new allocated fid is not yet known.
1082 mutex_lock(&tgt->ltd_fid_mutex);
1083 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1084 mutex_unlock(&tgt->ltd_fid_mutex);
1086 LASSERT(fid_is_sane(fid));
1093 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1095 struct lmv_obd *lmv = &obd->u.lmv;
1096 struct lmv_desc *desc;
1097 struct lnet_processid lnet_id;
1103 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1104 CERROR("LMV setup requires a descriptor\n");
1108 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1109 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1110 CERROR("Lmv descriptor size wrong: %d > %d\n",
1111 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1115 obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1116 desc->ld_uuid.uuid);
1117 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1118 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1119 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1120 LMV_DESC_QOS_MAXAGE_DEFAULT;
1121 lmv->max_def_easize = 0;
1122 lmv->max_easize = 0;
1124 spin_lock_init(&lmv->lmv_lock);
1127 * initialize rr_index to lower 32bit of netid, so that client
1128 * can distribute subdirs evenly from the beginning.
1130 while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1131 if (!nid_is_lo0(&lnet_id.nid)) {
1132 lmv->lmv_qos_rr_index = ntohl(lnet_id.nid.nid_addr[0]);
1137 rc = lmv_tunables_init(obd);
1139 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1142 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1143 LUSTRE_CLI_FLD_HASH_DHT);
1145 CERROR("Can't init FLD, err %d\n", rc);
1147 rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1149 CWARN("%s: error initialize target table: rc = %d\n",
1155 static int lmv_cleanup(struct obd_device *obd)
1157 struct lmv_obd *lmv = &obd->u.lmv;
1158 struct lu_tgt_desc *tgt;
1159 struct lu_tgt_desc *tmp;
1163 fld_client_fini(&lmv->lmv_fld);
1164 lmv_foreach_tgt_safe(lmv, tgt, tmp)
1165 lmv_del_target(lmv, tgt);
1166 lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1171 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1173 struct lustre_cfg *lcfg = buf;
1174 struct obd_uuid obd_uuid;
1180 switch (lcfg->lcfg_command) {
1182 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1183 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1184 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1185 GOTO(out, rc = -EINVAL);
1187 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1189 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1190 GOTO(out, rc = -EINVAL);
1191 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1192 GOTO(out, rc = -EINVAL);
1193 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1196 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1197 GOTO(out, rc = -EINVAL);
1203 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1207 if (flags & OBD_STATFS_FOR_MDT0)
1210 if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1211 return lmv->lmv_statfs_start;
1213 /* choose initial MDT for this client */
1215 struct lnet_processid lnet_id;
1216 if (LNetGetId(i, &lnet_id) == -ENOENT)
1219 if (!nid_is_lo0(&lnet_id.nid)) {
1220 /* We dont need a full 64-bit modulus, just enough
1221 * to distribute the requests across MDTs evenly.
1223 lmv->lmv_statfs_start = nidhash(&lnet_id.nid) %
1229 return lmv->lmv_statfs_start;
1232 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1233 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1235 struct obd_device *obd = class_exp2obd(exp);
1236 struct lmv_obd *lmv = &obd->u.lmv;
1237 struct obd_statfs *temp;
1238 struct lu_tgt_desc *tgt;
1245 OBD_ALLOC(temp, sizeof(*temp));
1249 /* distribute statfs among MDTs */
1250 idx = lmv_select_statfs_mdt(lmv, flags);
1252 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1253 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1254 tgt = lmv_tgt(lmv, idx);
1255 if (!tgt || !tgt->ltd_exp)
1258 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age,
1259 flags | OBD_STATFS_NESTED);
1261 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1262 tgt->ltd_exp->exp_obd->obd_name, i, rc);
1263 GOTO(out_free_temp, rc);
1266 if (temp->os_state & OS_STATFS_SUM ||
1267 flags == OBD_STATFS_FOR_MDT0) {
1268 /* reset to the last aggregated values
1269 * and don't sum with non-aggrated data */
1270 /* If the statfs is from mount, it needs to retrieve
1271 * necessary information from MDT0. i.e. mount does
1272 * not need the merged osfs from all of MDT. Also
1273 * clients can be mounted as long as MDT0 is in
1282 osfs->os_bavail += temp->os_bavail;
1283 osfs->os_blocks += temp->os_blocks;
1284 osfs->os_ffree += temp->os_ffree;
1285 osfs->os_files += temp->os_files;
1286 osfs->os_granted += temp->os_granted;
1292 OBD_FREE(temp, sizeof(*temp));
1296 static int lmv_statfs_update(void *cookie, int rc)
1298 struct obd_info *oinfo = cookie;
1299 struct obd_device *obd = oinfo->oi_obd;
1300 struct lmv_obd *lmv = &obd->u.lmv;
1301 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1302 struct obd_statfs *osfs = oinfo->oi_osfs;
1305 * NB: don't deactivate TGT upon error, because we may not trigger async
1306 * statfs any longer, then there is no chance to activate TGT.
1309 spin_lock(&lmv->lmv_lock);
1310 tgt->ltd_statfs = *osfs;
1311 tgt->ltd_statfs_age = ktime_get_seconds();
1312 spin_unlock(&lmv->lmv_lock);
1313 set_bit(LQ_DIRTY, &lmv->lmv_qos.lq_flags);
1319 /* update tgt statfs async if it's ld_qos_maxage old */
1320 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1322 struct obd_info oinfo = {
1325 .oi_cb_up = lmv_statfs_update,
1329 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1330 obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1333 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1338 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1341 struct obd_device *obd = exp->exp_obd;
1342 struct lmv_obd *lmv = &obd->u.lmv;
1343 struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1351 rc = md_get_root(tgt->ltd_exp, fileset, fid);
1355 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1356 u64 obd_md_valid, const char *name, size_t buf_size,
1357 struct ptlrpc_request **req)
1359 struct obd_device *obd = exp->exp_obd;
1360 struct lmv_obd *lmv = &obd->u.lmv;
1361 struct lmv_tgt_desc *tgt;
1366 tgt = lmv_fid2tgt(lmv, fid);
1368 RETURN(PTR_ERR(tgt));
1370 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1375 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1376 u64 obd_md_valid, const char *name,
1377 const void *value, size_t value_size,
1378 unsigned int xattr_flags, u32 suppgid,
1379 struct ptlrpc_request **req)
1381 struct obd_device *obd = exp->exp_obd;
1382 struct lmv_obd *lmv = &obd->u.lmv;
1383 struct lmv_tgt_desc *tgt;
1388 tgt = lmv_fid2tgt(lmv, fid);
1390 RETURN(PTR_ERR(tgt));
1392 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1393 value, value_size, xattr_flags, suppgid, req);
1398 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1399 struct ptlrpc_request **request)
1401 struct obd_device *obd = exp->exp_obd;
1402 struct lmv_obd *lmv = &obd->u.lmv;
1403 struct lmv_tgt_desc *tgt;
1408 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1410 RETURN(PTR_ERR(tgt));
1412 if (op_data->op_flags & MF_GET_MDT_IDX) {
1413 op_data->op_mds = tgt->ltd_index;
1417 rc = md_getattr(tgt->ltd_exp, op_data, request);
1422 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1424 struct obd_device *obd = exp->exp_obd;
1425 struct lmv_obd *lmv = &obd->u.lmv;
1426 struct lu_tgt_desc *tgt;
1430 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1433 * With DNE every object can have two locks in different namespaces:
1434 * lookup lock in space of MDT storing direntry and update/open lock in
1435 * space of MDT storing inode.
1437 lmv_foreach_connected_tgt(lmv, tgt)
1438 md_null_inode(tgt->ltd_exp, fid);
1443 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1444 struct md_open_data *mod, struct ptlrpc_request **request)
1446 struct obd_device *obd = exp->exp_obd;
1447 struct lmv_obd *lmv = &obd->u.lmv;
1448 struct lmv_tgt_desc *tgt;
1453 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1455 RETURN(PTR_ERR(tgt));
1457 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1458 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1462 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv, __u32 mdt,
1463 unsigned short dir_depth)
1465 struct lu_tgt_desc *tgt, *cur = NULL;
1466 __u64 total_avail = 0;
1467 __u64 total_weight = 0;
1468 __u64 cur_weight = 0;
1469 int total_usable = 0;
1475 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1476 RETURN(ERR_PTR(-EAGAIN));
1478 down_write(&lmv->lmv_qos.lq_rw_sem);
1480 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1481 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1483 rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1485 GOTO(unlock, tgt = ERR_PTR(rc));
1487 lmv_foreach_tgt(lmv, tgt) {
1488 if (!tgt->ltd_exp || !tgt->ltd_active) {
1489 tgt->ltd_qos.ltq_usable = 0;
1493 tgt->ltd_qos.ltq_usable = 1;
1494 lu_tgt_qos_weight_calc(tgt);
1495 if (tgt->ltd_index == mdt)
1497 total_avail += tgt->ltd_qos.ltq_avail;
1498 total_weight += tgt->ltd_qos.ltq_weight;
1502 /* if current MDT has above-average space, within range of the QOS
1503 * threshold, stay on the same MDT to avoid creating needless remote
1504 * MDT directories. It's more likely for low level directories
1505 * "16 / (dir_depth + 10)" is the factor to make it more unlikely for
1506 * top level directories, while more likely for low levels.
1508 rand = total_avail * 16 / (total_usable * (dir_depth + 10));
1509 if (cur && cur->ltd_qos.ltq_avail >= rand) {
1514 rand = lu_prandom_u64_max(total_weight);
1516 lmv_foreach_connected_tgt(lmv, tgt) {
1517 if (!tgt->ltd_qos.ltq_usable)
1520 cur_weight += tgt->ltd_qos.ltq_weight;
1521 if (cur_weight < rand)
1524 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1528 /* no proper target found */
1529 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1531 up_write(&lmv->lmv_qos.lq_rw_sem);
1536 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv)
1538 struct lu_tgt_desc *tgt;
1544 spin_lock(&lmv->lmv_lock);
1545 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1546 index = (i + lmv->lmv_qos_rr_index) %
1547 lmv->lmv_mdt_descs.ltd_tgts_size;
1548 tgt = lmv_tgt(lmv, index);
1549 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1552 lmv->lmv_qos_rr_index = (tgt->ltd_index + 1) %
1553 lmv->lmv_mdt_descs.ltd_tgts_size;
1554 spin_unlock(&lmv->lmv_lock);
1558 spin_unlock(&lmv->lmv_lock);
1560 RETURN(ERR_PTR(-ENODEV));
1563 /* locate MDT which is less full (avoid the most full MDT) */
1564 static struct lu_tgt_desc *lmv_locate_tgt_lf(struct lmv_obd *lmv)
1566 struct lu_tgt_desc *min = NULL;
1567 struct lu_tgt_desc *tgt;
1573 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1574 RETURN(ERR_PTR(-EAGAIN));
1576 down_write(&lmv->lmv_qos.lq_rw_sem);
1578 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1579 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1581 lmv_foreach_tgt(lmv, tgt) {
1582 if (!tgt->ltd_exp || !tgt->ltd_active) {
1583 tgt->ltd_qos.ltq_usable = 0;
1587 tgt->ltd_qos.ltq_usable = 1;
1588 lu_tgt_qos_weight_calc(tgt);
1589 avail += tgt->ltd_qos.ltq_avail;
1590 if (!min || min->ltd_qos.ltq_avail > tgt->ltd_qos.ltq_avail)
1594 /* avoid the most full MDT */
1596 avail -= min->ltd_qos.ltq_avail;
1598 rand = lu_prandom_u64_max(avail);
1600 lmv_foreach_connected_tgt(lmv, tgt) {
1601 if (!tgt->ltd_qos.ltq_usable)
1607 avail += tgt->ltd_qos.ltq_avail;
1614 /* no proper target found */
1615 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1617 up_write(&lmv->lmv_qos.lq_rw_sem);
1622 /* locate MDT by file name, for striped directory, the file name hash decides
1623 * which stripe its dirent is stored.
1625 static struct lmv_tgt_desc *
1626 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1627 const char *name, int namelen, struct lu_fid *fid,
1628 __u32 *mds, bool new_layout)
1630 struct lmv_tgt_desc *tgt;
1631 const struct lmv_oinfo *oinfo;
1633 if (!lmv_dir_striped(lsm) || !namelen) {
1634 tgt = lmv_fid2tgt(lmv, fid);
1638 *mds = tgt->ltd_index;
1642 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1643 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1644 return ERR_PTR(-EBADF);
1645 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1647 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, new_layout);
1649 return ERR_CAST(oinfo);
1652 /* check stripe FID is sane */
1653 if (!fid_is_sane(&oinfo->lmo_fid))
1654 return ERR_PTR(-ENODEV);
1656 *fid = oinfo->lmo_fid;
1657 *mds = oinfo->lmo_mds;
1658 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1660 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1662 return tgt ? tgt : ERR_PTR(-ENODEV);
1666 * Locate MDT of op_data->op_fid1
1668 * For striped directory, it will locate the stripe by name hash, if hash_type
1669 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1670 * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1671 * indicates whether old or new layout is used to locate.
1673 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1675 * \param[in] lmv LMV device
1676 * \param[in/out] op_data client MD stack parameters, name, namelen etc,
1677 * op_mds and op_fid1 will be updated if op_mea1
1678 * indicates fid1 represents a striped directory.
1680 * retval pointer to the lmv_tgt_desc if succeed.
1681 * ERR_PTR(errno) if failed.
1683 struct lmv_tgt_desc *
1684 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1686 struct lmv_stripe_md *lsm = op_data->op_mea1;
1687 struct lmv_oinfo *oinfo;
1688 struct lmv_tgt_desc *tgt;
1690 if (lmv_dir_foreign(lsm))
1691 return ERR_PTR(-ENODATA);
1693 /* During creating VOLATILE file, it should honor the mdt
1694 * index if the file under striped dir is being restored, see
1696 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1697 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1698 tgt = lmv_tgt(lmv, op_data->op_mds);
1700 return ERR_PTR(-ENODEV);
1702 if (lmv_dir_striped(lsm)) {
1705 /* refill the right parent fid */
1706 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1707 oinfo = &lsm->lsm_md_oinfo[i];
1708 if (oinfo->lmo_mds == op_data->op_mds) {
1709 op_data->op_fid1 = oinfo->lmo_fid;
1714 if (i == lsm->lsm_md_stripe_count)
1715 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1717 } else if (lmv_dir_bad_hash(lsm)) {
1718 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1719 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1721 op_data->op_fid1 = oinfo->lmo_fid;
1722 op_data->op_mds = oinfo->lmo_mds;
1723 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1725 return ERR_PTR(-ENODEV);
1727 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1728 op_data->op_name, op_data->op_namelen,
1729 &op_data->op_fid1, &op_data->op_mds,
1730 op_data->op_new_layout);
1736 /* Locate MDT of op_data->op_fid2 for link/rename */
1737 static struct lmv_tgt_desc *
1738 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1740 struct lmv_tgt_desc *tgt;
1743 LASSERT(op_data->op_name);
1744 if (lmv_dir_layout_changing(op_data->op_mea2)) {
1745 struct lu_fid fid1 = op_data->op_fid1;
1746 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1747 struct ptlrpc_request *request = NULL;
1750 * avoid creating new file under old layout of migrating
1751 * directory, check it here.
1753 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1754 op_data->op_name, op_data->op_namelen,
1755 &op_data->op_fid2, &op_data->op_mds, false);
1759 op_data->op_fid1 = op_data->op_fid2;
1760 op_data->op_mea1 = op_data->op_mea2;
1761 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1762 op_data->op_fid1 = fid1;
1763 op_data->op_mea1 = lsm1;
1765 ptlrpc_req_finished(request);
1766 RETURN(ERR_PTR(-EEXIST));
1770 RETURN(ERR_PTR(rc));
1773 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1774 op_data->op_namelen, &op_data->op_fid2,
1775 &op_data->op_mds, true);
1778 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1780 struct lu_tgt_desc *tgt;
1781 struct ptlrpc_request *request;
1784 LASSERT(lmv_dir_layout_changing(op_data->op_mea1));
1785 LASSERT(!op_data->op_new_layout);
1787 tgt = lmv_locate_tgt(lmv, op_data);
1789 return PTR_ERR(tgt);
1791 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1793 ptlrpc_req_finished(request);
1800 /* mkdir by QoS upon 'lfs mkdir -i -1'.
1802 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1803 * directories under striped directory.
1805 static inline bool lmv_op_user_qos_mkdir(const struct md_op_data *op_data)
1807 const struct lmv_user_md *lum = op_data->op_data;
1809 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1812 if (lmv_dir_striped(op_data->op_mea1))
1815 return (op_data->op_cli_flags & CLI_SET_MEA) && lum &&
1816 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
1817 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
1820 /* mkdir by QoS if either ROOT or parent default LMV is space balanced. */
1821 static inline bool lmv_op_default_qos_mkdir(const struct md_op_data *op_data)
1823 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1825 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1828 if (lmv_dir_striped(op_data->op_mea1))
1831 return (op_data->op_flags & MF_QOS_MKDIR) ||
1832 (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT);
1835 /* if parent default LMV is space balanced, and
1836 * 1. max_inherit_rr is set
1837 * 2. or parent is ROOT
1838 * mkdir roundrobin. Or if parent doesn't have default LMV, while ROOT default
1839 * LMV requests roundrobin mkdir, do the same.
1840 * NB, this needs to check server is balanced, which is done by caller.
1842 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data)
1844 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1846 if (!lmv_op_default_qos_mkdir(op_data))
1849 return (op_data->op_flags & MF_RR_MKDIR) ||
1850 (lsm && lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE) ||
1851 fid_is_root(&op_data->op_fid1);
1854 /* 'lfs mkdir -i <specific_MDT>' */
1855 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1857 const struct lmv_user_md *lum = op_data->op_data;
1859 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1860 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1861 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1862 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1863 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1866 /* parent default LMV master_mdt_index is not -1. */
1868 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1870 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1871 op_data->op_default_mea1 &&
1872 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1876 /* locate MDT by space usage */
1877 static struct lu_tgt_desc *lmv_locate_tgt_by_space(struct lmv_obd *lmv,
1878 struct md_op_data *op_data,
1879 struct lmv_tgt_desc *tgt)
1881 struct lmv_tgt_desc *tmp = tgt;
1883 tgt = lmv_locate_tgt_qos(lmv, op_data->op_mds, op_data->op_dir_depth);
1884 if (tgt == ERR_PTR(-EAGAIN)) {
1885 if (ltd_qos_is_balanced(&lmv->lmv_mdt_descs) &&
1886 !lmv_op_default_rr_mkdir(op_data) &&
1887 !lmv_op_user_qos_mkdir(op_data))
1888 /* if not necessary, don't create remote directory. */
1891 tgt = lmv_locate_tgt_rr(lmv);
1895 * only update statfs after QoS mkdir, this means the cached statfs may
1896 * be stale, and current mkdir may not follow QoS accurately, but it's
1897 * not serious, and avoids periodic statfs when client doesn't mkdir by
1901 op_data->op_mds = tgt->ltd_index;
1902 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1908 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1909 const void *data, size_t datalen, umode_t mode, uid_t uid,
1910 gid_t gid, kernel_cap_t cap_effective, __u64 rdev,
1911 struct ptlrpc_request **request)
1913 struct obd_device *obd = exp->exp_obd;
1914 struct lmv_obd *lmv = &obd->u.lmv;
1915 struct lmv_tgt_desc *tgt;
1916 struct mdt_body *repbody;
1921 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1924 if (lmv_dir_bad_hash(op_data->op_mea1))
1927 if (lmv_dir_layout_changing(op_data->op_mea1)) {
1929 * if parent is migrating, create() needs to lookup existing
1930 * name in both old and new layout, check old layout on client.
1932 rc = lmv_old_layout_lookup(lmv, op_data);
1936 op_data->op_new_layout = true;
1939 tgt = lmv_locate_tgt(lmv, op_data);
1941 RETURN(PTR_ERR(tgt));
1943 /* the order to apply policy in mkdir:
1944 * 1. is "lfs mkdir -i N"? mkdir on MDT N.
1945 * 2. is "lfs mkdir -i -1"? mkdir by space usage.
1946 * 3. is starting MDT specified in default LMV? mkdir on MDT N.
1947 * 4. is default LMV space balanced? mkdir by space usage.
1949 if (lmv_op_user_specific_mkdir(op_data)) {
1950 struct lmv_user_md *lum = op_data->op_data;
1952 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
1953 tgt = lmv_tgt(lmv, op_data->op_mds);
1956 } else if (lmv_op_user_qos_mkdir(op_data)) {
1957 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
1959 RETURN(PTR_ERR(tgt));
1960 } else if (lmv_op_default_specific_mkdir(op_data)) {
1962 op_data->op_default_mea1->lsm_md_master_mdt_index;
1963 tgt = lmv_tgt(lmv, op_data->op_mds);
1966 } else if (lmv_op_default_qos_mkdir(op_data)) {
1967 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
1969 RETURN(PTR_ERR(tgt));
1973 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1977 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
1978 (int)op_data->op_namelen, op_data->op_name,
1979 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
1982 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1983 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1984 cap_effective, rdev, request);
1986 if (*request == NULL)
1988 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1991 /* dir restripe needs to send to MDT where dir is located */
1992 if (rc != -EREMOTE ||
1993 !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
1996 repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
1997 if (repbody == NULL)
2000 /* Not cross-ref case, just get out of here. */
2001 if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
2004 op_data->op_fid2 = repbody->mbo_fid1;
2005 ptlrpc_req_finished(*request);
2008 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2010 RETURN(PTR_ERR(tgt));
2012 op_data->op_mds = tgt->ltd_index;
2017 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
2018 const union ldlm_policy_data *policy, struct md_op_data *op_data,
2019 struct lustre_handle *lockh, __u64 extra_lock_flags)
2021 struct obd_device *obd = exp->exp_obd;
2022 struct lmv_obd *lmv = &obd->u.lmv;
2023 struct lmv_tgt_desc *tgt;
2028 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
2030 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2032 RETURN(PTR_ERR(tgt));
2034 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
2035 PFID(&op_data->op_fid1), tgt->ltd_index);
2037 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
2044 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
2045 struct ptlrpc_request **preq)
2047 struct obd_device *obd = exp->exp_obd;
2048 struct lmv_obd *lmv = &obd->u.lmv;
2049 struct lmv_tgt_desc *tgt;
2050 struct mdt_body *body;
2056 if (op_data->op_namelen == 2 &&
2057 op_data->op_name[0] == '.' && op_data->op_name[1] == '.')
2058 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2060 tgt = lmv_locate_tgt(lmv, op_data);
2062 RETURN(PTR_ERR(tgt));
2064 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
2065 (int)op_data->op_namelen, op_data->op_name,
2066 PFID(&op_data->op_fid1), tgt->ltd_index);
2068 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2069 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2070 ptlrpc_req_finished(*preq);
2078 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2079 LASSERT(body != NULL);
2081 if (body->mbo_valid & OBD_MD_MDS) {
2082 op_data->op_fid1 = body->mbo_fid1;
2083 op_data->op_valid |= OBD_MD_FLCROSSREF;
2084 op_data->op_namelen = 0;
2085 op_data->op_name = NULL;
2087 ptlrpc_req_finished(*preq);
2096 #define md_op_data_fid(op_data, fl) \
2097 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2098 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2099 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2100 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2103 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2104 struct md_op_data *op_data, __u32 op_tgt,
2105 enum ldlm_mode mode, int bits, int flag)
2107 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2108 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2109 union ldlm_policy_data policy = { { 0 } };
2113 if (!fid_is_sane(fid))
2117 tgt = lmv_fid2tgt(lmv, fid);
2119 RETURN(PTR_ERR(tgt));
2122 if (tgt->ltd_index != op_tgt) {
2123 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2124 policy.l_inodebits.bits = bits;
2125 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2126 mode, LCF_ASYNC, NULL);
2129 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2131 op_data->op_flags |= flag;
2139 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2142 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2143 struct ptlrpc_request **request)
2145 struct obd_device *obd = exp->exp_obd;
2146 struct lmv_obd *lmv = &obd->u.lmv;
2147 struct lmv_tgt_desc *tgt;
2151 LASSERT(op_data->op_namelen != 0);
2153 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2154 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2155 op_data->op_name, PFID(&op_data->op_fid1));
2157 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2158 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2159 op_data->op_cap = current_cap();
2161 tgt = lmv_locate_tgt2(lmv, op_data);
2163 RETURN(PTR_ERR(tgt));
2166 * Cancel UPDATE lock on child (fid1).
2168 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2169 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2170 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2174 rc = md_link(tgt->ltd_exp, op_data, request);
2179 /* migrate the top directory */
2180 static inline bool lmv_op_topdir_migrate(const struct md_op_data *op_data)
2182 if (!S_ISDIR(op_data->op_mode))
2185 if (lmv_dir_layout_changing(op_data->op_mea1))
2191 /* migrate top dir to specific MDTs */
2192 static inline bool lmv_topdir_specific_migrate(const struct md_op_data *op_data)
2194 const struct lmv_user_md *lum = op_data->op_data;
2196 if (!lmv_op_topdir_migrate(op_data))
2199 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2202 /* migrate top dir in QoS mode if user issued "lfs migrate -m -1..." */
2203 static inline bool lmv_topdir_qos_migrate(const struct md_op_data *op_data)
2205 const struct lmv_user_md *lum = op_data->op_data;
2207 if (!lmv_op_topdir_migrate(op_data))
2210 return le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
2213 static inline bool lmv_subdir_specific_migrate(const struct md_op_data *op_data)
2215 const struct lmv_user_md *lum = op_data->op_data;
2217 if (!S_ISDIR(op_data->op_mode))
2220 if (!lmv_dir_layout_changing(op_data->op_mea1))
2223 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2226 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2227 const char *name, size_t namelen,
2228 struct ptlrpc_request **request)
2230 struct obd_device *obd = exp->exp_obd;
2231 struct lmv_obd *lmv = &obd->u.lmv;
2232 struct lmv_stripe_md *lsm = op_data->op_mea1;
2233 struct lmv_tgt_desc *parent_tgt;
2234 struct lmv_tgt_desc *sp_tgt;
2235 struct lmv_tgt_desc *tp_tgt = NULL;
2236 struct lmv_tgt_desc *child_tgt;
2237 struct lmv_tgt_desc *tgt;
2238 struct lu_fid target_fid = { 0 };
2243 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2245 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2246 PFID(&op_data->op_fid1), (int)namelen, name);
2248 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2249 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2250 op_data->op_cap = current_cap();
2252 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2253 if (IS_ERR(parent_tgt))
2254 RETURN(PTR_ERR(parent_tgt));
2256 if (lmv_dir_striped(lsm)) {
2257 const struct lmv_oinfo *oinfo;
2259 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
2261 RETURN(PTR_ERR(oinfo));
2263 /* save source stripe FID in fid4 temporarily for ELC */
2264 op_data->op_fid4 = oinfo->lmo_fid;
2265 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2270 * if parent is being migrated too, fill op_fid2 with target
2271 * stripe fid, otherwise the target stripe is not created yet.
2273 if (lmv_dir_layout_changing(lsm)) {
2274 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
2277 RETURN(PTR_ERR(oinfo));
2279 op_data->op_fid2 = oinfo->lmo_fid;
2280 tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2284 /* parent unchanged and update namespace only */
2285 if (lu_fid_eq(&op_data->op_fid4, &op_data->op_fid2) &&
2286 op_data->op_bias & MDS_MIGRATE_NSONLY)
2290 sp_tgt = parent_tgt;
2293 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2294 if (IS_ERR(child_tgt))
2295 RETURN(PTR_ERR(child_tgt));
2297 if (lmv_topdir_specific_migrate(op_data)) {
2298 struct lmv_user_md *lum = op_data->op_data;
2300 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2301 } else if (lmv_topdir_qos_migrate(op_data)) {
2302 tgt = lmv_locate_tgt_lf(lmv);
2303 if (tgt == ERR_PTR(-EAGAIN))
2304 tgt = lmv_locate_tgt_rr(lmv);
2306 RETURN(PTR_ERR(tgt));
2308 op_data->op_mds = tgt->ltd_index;
2309 } else if (lmv_subdir_specific_migrate(op_data)) {
2310 struct lmv_user_md *lum = op_data->op_data;
2314 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2315 /* adjust MDTs in lum, since subdir is located on where
2316 * its parent stripe is, not the first specified MDT.
2318 for (i = 0; i < le32_to_cpu(lum->lum_stripe_count);
2320 if (le32_to_cpu(lum->lum_objects[i].lum_mds) ==
2325 if (i == le32_to_cpu(lum->lum_stripe_count))
2328 lum->lum_objects[i].lum_mds =
2329 lum->lum_objects[0].lum_mds;
2330 lum->lum_objects[0].lum_mds =
2331 cpu_to_le32(tp_tgt->ltd_index);
2333 /* NB, the above adjusts subdir migration for command like
2334 * "lfs migrate -m 0,1,2 ...", but for migration like
2335 * "lfs migrate -m 0 -c 2 ...", the top dir is migrated to MDT0
2336 * and MDT1, however its subdir may be migrated to MDT1 and MDT2
2339 lum->lum_stripe_offset = cpu_to_le32(tp_tgt->ltd_index);
2340 op_data->op_mds = tp_tgt->ltd_index;
2341 } else if (tp_tgt) {
2342 op_data->op_mds = tp_tgt->ltd_index;
2344 op_data->op_mds = sp_tgt->ltd_index;
2347 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2352 * for directory, send migrate request to the MDT where the object will
2353 * be migrated to, because we can't create a striped directory remotely.
2355 * otherwise, send to the MDT where source is located because regular
2356 * file may open lease.
2358 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2359 * backward compatibility.
2361 if (S_ISDIR(op_data->op_mode) &&
2362 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2363 tgt = lmv_fid2tgt(lmv, &target_fid);
2365 RETURN(PTR_ERR(tgt));
2370 /* cancel UPDATE lock of parent master object */
2371 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2372 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2376 /* cancel UPDATE lock of source parent */
2377 if (sp_tgt != parent_tgt) {
2379 * migrate RPC packs master object FID, because we can only pack
2380 * two FIDs in reint RPC, but MDS needs to know both source
2381 * parent and target parent, and it will obtain them from master
2382 * FID and LMV, the other FID in RPC is kept for target.
2384 * since this FID is not passed to MDC, cancel it anyway.
2386 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2387 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2391 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2393 op_data->op_fid4 = target_fid;
2395 /* cancel UPDATE locks of target parent */
2396 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2397 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2401 /* cancel LOOKUP lock of source if source is remote object */
2402 if (child_tgt != sp_tgt) {
2403 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2404 LCK_EX, MDS_INODELOCK_LOOKUP,
2405 MF_MDC_CANCEL_FID3);
2410 /* cancel ELC locks of source */
2411 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2412 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2416 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2421 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2422 const char *old, size_t oldlen,
2423 const char *new, size_t newlen,
2424 struct ptlrpc_request **request)
2426 struct obd_device *obd = exp->exp_obd;
2427 struct lmv_obd *lmv = &obd->u.lmv;
2428 struct lmv_tgt_desc *sp_tgt;
2429 struct lmv_tgt_desc *tp_tgt = NULL;
2430 struct lmv_tgt_desc *src_tgt = NULL;
2431 struct lmv_tgt_desc *tgt;
2432 struct mdt_body *body;
2437 LASSERT(oldlen != 0);
2439 if (op_data->op_cli_flags & CLI_MIGRATE) {
2440 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2444 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2445 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2446 op_data->op_cap = current_cap();
2448 op_data->op_name = new;
2449 op_data->op_namelen = newlen;
2451 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2453 RETURN(PTR_ERR(tp_tgt));
2455 /* Since the target child might be destroyed, and it might become
2456 * orphan, and we can only check orphan on the local MDT right now, so
2457 * we send rename request to the MDT where target child is located. If
2458 * target child does not exist, then it will send the request to the
2460 if (fid_is_sane(&op_data->op_fid4)) {
2461 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2463 RETURN(PTR_ERR(tgt));
2468 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2470 /* cancel UPDATE locks of target parent */
2471 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2472 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2476 if (fid_is_sane(&op_data->op_fid4)) {
2477 /* cancel LOOKUP lock of target on target parent */
2478 if (tgt != tp_tgt) {
2479 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2480 tgt->ltd_index, LCK_EX,
2481 MDS_INODELOCK_LOOKUP,
2482 MF_MDC_CANCEL_FID4);
2488 if (fid_is_sane(&op_data->op_fid3)) {
2489 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2490 if (IS_ERR(src_tgt))
2491 RETURN(PTR_ERR(src_tgt));
2493 /* cancel ELC locks of source */
2494 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2495 LCK_EX, MDS_INODELOCK_ELC,
2496 MF_MDC_CANCEL_FID3);
2501 op_data->op_name = old;
2502 op_data->op_namelen = oldlen;
2504 sp_tgt = lmv_locate_tgt(lmv, op_data);
2506 RETURN(PTR_ERR(sp_tgt));
2508 /* cancel UPDATE locks of source parent */
2509 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2510 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2514 if (fid_is_sane(&op_data->op_fid3)) {
2515 /* cancel LOOKUP lock of source on source parent */
2516 if (src_tgt != sp_tgt) {
2517 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2518 tgt->ltd_index, LCK_EX,
2519 MDS_INODELOCK_LOOKUP,
2520 MF_MDC_CANCEL_FID3);
2527 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2528 PFID(&op_data->op_fid1), (int)oldlen, old,
2529 PFID(&op_data->op_fid2), (int)newlen, new);
2531 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2533 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2534 ptlrpc_req_finished(*request);
2539 if (rc && rc != -EXDEV)
2542 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2546 /* Not cross-ref case, just get out of here. */
2547 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2550 op_data->op_fid4 = body->mbo_fid1;
2552 ptlrpc_req_finished(*request);
2555 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2557 RETURN(PTR_ERR(tgt));
2559 if (fid_is_sane(&op_data->op_fid4)) {
2560 /* cancel LOOKUP lock of target on target parent */
2561 if (tgt != tp_tgt) {
2562 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2563 tgt->ltd_index, LCK_EX,
2564 MDS_INODELOCK_LOOKUP,
2565 MF_MDC_CANCEL_FID4);
2574 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2575 void *ea, size_t ealen, struct ptlrpc_request **request)
2577 struct obd_device *obd = exp->exp_obd;
2578 struct lmv_obd *lmv = &obd->u.lmv;
2579 struct lmv_tgt_desc *tgt;
2584 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2585 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2586 op_data->op_xvalid);
2588 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2589 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2591 RETURN(PTR_ERR(tgt));
2593 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2598 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2599 struct ptlrpc_request **request)
2601 struct obd_device *obd = exp->exp_obd;
2602 struct lmv_obd *lmv = &obd->u.lmv;
2603 struct lmv_tgt_desc *tgt;
2608 tgt = lmv_fid2tgt(lmv, fid);
2610 RETURN(PTR_ERR(tgt));
2612 rc = md_fsync(tgt->ltd_exp, fid, request);
2616 struct stripe_dirent {
2617 struct page *sd_page;
2618 struct lu_dirpage *sd_dp;
2619 struct lu_dirent *sd_ent;
2623 struct lmv_dir_ctxt {
2624 struct lmv_obd *ldc_lmv;
2625 struct md_op_data *ldc_op_data;
2626 struct md_readdir_info *ldc_mrinfo;
2629 struct stripe_dirent ldc_stripes[0];
2632 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2634 if (stripe->sd_page) {
2635 kunmap(stripe->sd_page);
2636 put_page(stripe->sd_page);
2637 stripe->sd_page = NULL;
2638 stripe->sd_ent = NULL;
2642 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2646 for (i = 0; i < ctxt->ldc_count; i++)
2647 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2650 /* if @ent is dummy, or . .., get next */
2651 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2652 struct lu_dirent *ent,
2655 for (; ent; ent = lu_dirent_next(ent)) {
2656 /* Skip dummy entry */
2657 if (le16_to_cpu(ent->lde_namelen) == 0)
2660 /* skip . and .. for other stripes */
2662 (strncmp(ent->lde_name, ".",
2663 le16_to_cpu(ent->lde_namelen)) == 0 ||
2664 strncmp(ent->lde_name, "..",
2665 le16_to_cpu(ent->lde_namelen)) == 0))
2668 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2675 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2676 struct stripe_dirent *stripe,
2679 struct md_op_data *op_data = ctxt->ldc_op_data;
2680 struct lmv_oinfo *oinfo;
2681 struct lu_fid fid = op_data->op_fid1;
2682 struct inode *inode = op_data->op_data;
2683 struct lmv_tgt_desc *tgt;
2684 struct lu_dirent *ent = stripe->sd_ent;
2685 __u64 hash = ctxt->ldc_hash;
2690 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2694 if (stripe->sd_page) {
2695 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2697 /* @hash should be the last dirent hash */
2698 LASSERTF(hash <= end,
2699 "ctxt@%p stripe@%p hash %llx end %llx\n",
2700 ctxt, stripe, hash, end);
2701 /* unload last page */
2702 stripe_dirent_unload(stripe);
2704 if (end == MDS_DIR_END_OFF) {
2705 stripe->sd_eof = true;
2711 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2712 if (!oinfo->lmo_root) {
2717 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2723 /* op_data is shared by stripes, reset after use */
2724 op_data->op_fid1 = oinfo->lmo_fid;
2725 op_data->op_fid2 = oinfo->lmo_fid;
2726 op_data->op_data = oinfo->lmo_root;
2728 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_mrinfo, hash,
2731 op_data->op_fid1 = fid;
2732 op_data->op_fid2 = fid;
2733 op_data->op_data = inode;
2738 stripe->sd_dp = page_address(stripe->sd_page);
2739 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2741 /* in case a page filled with ., .. and dummy, read next */
2744 stripe->sd_ent = ent;
2747 /* treat error as eof, so dir can be partially accessed */
2748 stripe->sd_eof = true;
2749 ctxt->ldc_mrinfo->mr_partial_readdir_rc = rc;
2750 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2751 "directory is partially accessed!\n",
2752 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2759 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2761 struct obd_device *obd = exp->exp_obd;
2762 struct lmv_obd *lmv = &obd->u.lmv;
2763 struct lmv_tgt_desc *tgt;
2768 rc = lmv_check_connect(obd);
2772 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2774 RETURN(PTR_ERR(tgt));
2776 data->op_flags |= MF_MDC_CANCEL_FID1;
2777 rc = md_file_resync(tgt->ltd_exp, data);
2782 * Get dirent with the closest hash for striped directory
2784 * This function will search the dir entry, whose hash value is the
2785 * closest(>=) to hash from all of sub-stripes, and it is only being called
2786 * for striped directory.
2788 * \param[in] ctxt dir read context
2790 * \retval dirent get the entry successfully
2791 * NULL does not get the entry, normally it means
2792 * it reaches the end of the directory, while read
2793 * stripe dirent error is ignored to allow partial
2796 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2798 struct stripe_dirent *stripe;
2799 struct lu_dirent *ent = NULL;
2803 /* TODO: optimize with k-way merge sort */
2804 for (i = 0; i < ctxt->ldc_count; i++) {
2805 stripe = &ctxt->ldc_stripes[i];
2809 if (!stripe->sd_ent) {
2810 stripe_dirent_load(ctxt, stripe, i);
2811 if (!stripe->sd_ent) {
2812 LASSERT(stripe->sd_eof);
2818 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2819 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2821 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2828 stripe = &ctxt->ldc_stripes[min];
2829 ent = stripe->sd_ent;
2830 /* pop found dirent */
2831 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2839 * Build dir entry page for striped directory
2841 * This function gets one entry by @offset from a striped directory. It will
2842 * read entries from all of stripes, and choose one closest to the required
2843 * offset(&offset). A few notes
2844 * 1. skip . and .. for non-zero stripes, because there can only have one .
2845 * and .. in a directory.
2846 * 2. op_data will be shared by all of stripes, instead of allocating new
2847 * one, so need to restore before reusing.
2849 * \param[in] exp obd export refer to LMV
2850 * \param[in] op_data hold those MD parameters of read_entry
2851 * \param[in] mrinfo ldlm callback being used in enqueue in mdc_read_entry,
2852 * and partial readdir result will be stored in it.
2853 * \param[in] offset starting hash offset
2854 * \param[out] ppage the page holding the entry. Note: because the entry
2855 * will be accessed in upper layer, so we need hold the
2856 * page until the usages of entry is finished, see
2857 * ll_dir_entry_next.
2859 * retval =0 if get entry successfully
2860 * <0 cannot get entry
2862 static int lmv_striped_read_page(struct obd_export *exp,
2863 struct md_op_data *op_data,
2864 struct md_readdir_info *mrinfo, __u64 offset,
2865 struct page **ppage)
2867 struct page *page = NULL;
2868 struct lu_dirpage *dp;
2870 struct lu_dirent *ent;
2871 struct lu_dirent *last_ent;
2873 struct lmv_dir_ctxt *ctxt;
2874 struct lu_dirent *next = NULL;
2880 /* Allocate a page and read entries from all of stripes and fill
2881 * the page by hash order */
2882 page = alloc_page(GFP_KERNEL);
2886 /* Initialize the entry page */
2888 memset(dp, 0, sizeof(*dp));
2889 dp->ldp_hash_start = cpu_to_le64(offset);
2892 left_bytes = PAGE_SIZE - sizeof(*dp);
2896 /* initalize dir read context */
2897 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2898 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2900 GOTO(free_page, rc = -ENOMEM);
2901 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2902 ctxt->ldc_op_data = op_data;
2903 ctxt->ldc_mrinfo = mrinfo;
2904 ctxt->ldc_hash = offset;
2905 ctxt->ldc_count = stripe_count;
2908 next = lmv_dirent_next(ctxt);
2910 /* end of directory */
2912 ctxt->ldc_hash = MDS_DIR_END_OFF;
2915 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2917 ent_size = le16_to_cpu(next->lde_reclen);
2919 /* the last entry lde_reclen is 0, but it might not be the last
2920 * one of this temporay dir page */
2922 ent_size = lu_dirent_calc_size(
2923 le16_to_cpu(next->lde_namelen),
2924 le32_to_cpu(next->lde_attrs));
2926 if (ent_size > left_bytes)
2929 memcpy(ent, next, ent_size);
2931 /* Replace . with master FID and Replace .. with the parent FID
2932 * of master object */
2933 if (strncmp(ent->lde_name, ".",
2934 le16_to_cpu(ent->lde_namelen)) == 0 &&
2935 le16_to_cpu(ent->lde_namelen) == 1)
2936 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2937 else if (strncmp(ent->lde_name, "..",
2938 le16_to_cpu(ent->lde_namelen)) == 0 &&
2939 le16_to_cpu(ent->lde_namelen) == 2)
2940 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2942 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2943 le16_to_cpu(ent->lde_namelen), ent->lde_name,
2944 le64_to_cpu(ent->lde_hash));
2946 left_bytes -= ent_size;
2947 ent->lde_reclen = cpu_to_le16(ent_size);
2949 ent = (void *)ent + ent_size;
2952 last_ent->lde_reclen = 0;
2955 dp->ldp_flags |= LDF_EMPTY;
2956 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2957 dp->ldp_flags |= LDF_COLLIDE;
2958 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2959 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2961 put_lmv_dir_ctxt(ctxt);
2962 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2975 static int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2976 struct md_readdir_info *mrinfo, __u64 offset,
2977 struct page **ppage)
2979 struct obd_device *obd = exp->exp_obd;
2980 struct lmv_obd *lmv = &obd->u.lmv;
2981 struct lmv_tgt_desc *tgt;
2986 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
2989 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
2990 rc = lmv_striped_read_page(exp, op_data, mrinfo, offset, ppage);
2994 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2996 RETURN(PTR_ERR(tgt));
2998 rc = md_read_page(tgt->ltd_exp, op_data, mrinfo, offset, ppage);
3004 * Unlink a file/directory
3006 * Unlink a file or directory under the parent dir. The unlink request
3007 * usually will be sent to the MDT where the child is located, but if
3008 * the client does not have the child FID then request will be sent to the
3009 * MDT where the parent is located.
3011 * If the parent is a striped directory then it also needs to locate which
3012 * stripe the name of the child is located, and replace the parent FID
3013 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
3014 * it will walk through all of sub-stripes until the child is being
3017 * \param[in] exp export refer to LMV
3018 * \param[in] op_data different parameters transferred beween client
3019 * MD stacks, name, namelen, FIDs etc.
3020 * op_fid1 is the parent FID, op_fid2 is the child
3022 * \param[out] request point to the request of unlink.
3024 * retval 0 if succeed
3025 * negative errno if failed.
3027 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
3028 struct ptlrpc_request **request)
3030 struct obd_device *obd = exp->exp_obd;
3031 struct lmv_obd *lmv = &obd->u.lmv;
3032 struct lmv_tgt_desc *tgt;
3033 struct lmv_tgt_desc *parent_tgt;
3034 struct mdt_body *body;
3039 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3040 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3041 op_data->op_cap = current_cap();
3044 parent_tgt = lmv_locate_tgt(lmv, op_data);
3045 if (IS_ERR(parent_tgt))
3046 RETURN(PTR_ERR(parent_tgt));
3048 if (likely(!fid_is_zero(&op_data->op_fid2))) {
3049 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3051 RETURN(PTR_ERR(tgt));
3057 * If child's fid is given, cancel unused locks for it if it is from
3058 * another export than parent.
3060 * LOOKUP lock for child (fid3) should also be cancelled on parent
3061 * tgt_tgt in mdc_unlink().
3063 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
3065 if (parent_tgt != tgt)
3066 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
3067 LCK_EX, MDS_INODELOCK_LOOKUP,
3068 MF_MDC_CANCEL_FID3);
3070 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
3071 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
3075 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
3076 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
3079 rc = md_unlink(tgt->ltd_exp, op_data, request);
3080 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
3081 ptlrpc_req_finished(*request);
3089 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
3093 /* Not cross-ref case, just get out of here. */
3094 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
3097 /* This is a remote object, try remote MDT. */
3098 op_data->op_fid2 = body->mbo_fid1;
3099 ptlrpc_req_finished(*request);
3102 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3104 RETURN(PTR_ERR(tgt));
3109 static int lmv_precleanup(struct obd_device *obd)
3112 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
3113 fld_client_debugfs_fini(&obd->u.lmv.lmv_fld);
3114 lprocfs_obd_cleanup(obd);
3115 lprocfs_free_md_stats(obd);
3120 * Get by key a value associated with a LMV device.
3122 * Dispatch request to lower-layer devices as needed.
3124 * \param[in] env execution environment for this thread
3125 * \param[in] exp export for the LMV device
3126 * \param[in] keylen length of key identifier
3127 * \param[in] key identifier of key to get value for
3128 * \param[in] vallen size of \a val
3129 * \param[out] val pointer to storage location for value
3130 * \param[in] lsm optional striping metadata of object
3132 * \retval 0 on success
3133 * \retval negative negated errno on failure
3135 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
3136 __u32 keylen, void *key, __u32 *vallen, void *val)
3138 struct obd_device *obd;
3139 struct lmv_obd *lmv;
3140 struct lu_tgt_desc *tgt;
3145 obd = class_exp2obd(exp);
3147 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3148 exp->exp_handle.h_cookie);
3153 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
3154 LASSERT(*vallen == sizeof(__u32));
3155 lmv_foreach_connected_tgt(lmv, tgt) {
3156 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
3161 } else if (KEY_IS(KEY_MAX_EASIZE) ||
3162 KEY_IS(KEY_DEFAULT_EASIZE) ||
3163 KEY_IS(KEY_CONN_DATA)) {
3165 * Forwarding this request to first MDS, it should know LOV
3168 tgt = lmv_tgt(lmv, 0);
3172 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
3173 if (!rc && KEY_IS(KEY_CONN_DATA))
3174 exp->exp_connect_data = *(struct obd_connect_data *)val;
3176 } else if (KEY_IS(KEY_TGT_COUNT)) {
3177 *((int *)val) = lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count;
3181 CDEBUG(D_IOCTL, "Invalid key\n");
3185 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
3186 int *__rcs, struct ptlrpc_request_set *_set)
3188 struct obd_device *obd = class_exp2obd(exp);
3189 struct ptlrpc_request_set *set = _set;
3190 struct lmv_obd *lmv = &obd->u.lmv;
3191 int tgt_count = lmv->lmv_mdt_count;
3192 struct lu_tgt_desc *tgt;
3193 struct fid_array *fat, **fas = NULL;
3194 int i, rc, **rcs = NULL;
3197 set = ptlrpc_prep_set();
3202 /* split FIDs by targets */
3203 OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
3205 GOTO(out, rc = -ENOMEM);
3206 OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
3208 GOTO(out_fas, rc = -ENOMEM);
3210 for (i = 0; i < fa->fa_nr; i++) {
3213 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
3215 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
3216 PFID(&fa->fa_fids[i]), rc);
3219 LASSERT(idx < tgt_count);
3221 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
3222 fa_fids[fa->fa_nr]));
3224 GOTO(out, rc = -ENOMEM);
3226 OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
3228 GOTO(out, rc = -ENOMEM);
3231 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3234 lmv_foreach_connected_tgt(lmv, tgt) {
3235 fat = fas[tgt->ltd_index];
3236 if (!fat || fat->fa_nr == 0)
3238 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3241 rc = ptlrpc_set_wait(NULL, set);
3244 for (i = 0; i < tgt_count; i++) {
3246 if (!fat || fat->fa_nr == 0)
3248 /* copy FIDs back */
3249 memcpy(fa->fa_fids + j, fat->fa_fids,
3250 fat->fa_nr * sizeof(struct lu_fid));
3252 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3257 ptlrpc_set_destroy(set);
3260 for (i = 0; i < tgt_count; i++) {
3262 OBD_FREE(fas[i], offsetof(struct fid_array,
3263 fa_fids[fa->fa_nr]));
3265 OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3268 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3271 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3277 * Asynchronously set by key a value associated with a LMV device.
3279 * Dispatch request to lower-layer devices as needed.
3281 * \param[in] env execution environment for this thread
3282 * \param[in] exp export for the LMV device
3283 * \param[in] keylen length of key identifier
3284 * \param[in] key identifier of key to store value for
3285 * \param[in] vallen size of value to store
3286 * \param[in] val pointer to data to be stored
3287 * \param[in] set optional list of related ptlrpc requests
3289 * \retval 0 on success
3290 * \retval negative negated errno on failure
3292 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3293 __u32 keylen, void *key, __u32 vallen, void *val,
3294 struct ptlrpc_request_set *set)
3296 struct lmv_tgt_desc *tgt;
3297 struct obd_device *obd;
3298 struct lmv_obd *lmv;
3302 obd = class_exp2obd(exp);
3304 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3305 exp->exp_handle.h_cookie);
3310 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3311 KEY_IS(KEY_DEFAULT_EASIZE)) {
3314 lmv_foreach_connected_tgt(lmv, tgt) {
3315 err = obd_set_info_async(env, tgt->ltd_exp,
3316 keylen, key, vallen, val, set);
3327 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3328 const struct lmv_mds_md_v1 *lmm1)
3330 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3337 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3338 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3339 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3340 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3341 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3343 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3344 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3345 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3346 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3347 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3348 sizeof(lsm->lsm_md_pool_name));
3350 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3353 CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3354 "layout_version %d\n", lsm->lsm_md_stripe_count,
3355 lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3356 lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3357 lsm->lsm_md_layout_version);
3359 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3360 for (i = 0; i < stripe_count; i++) {
3361 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3362 &lmm1->lmv_stripe_fids[i]);
3364 * set default value -1, so lmv_locate_tgt() knows this stripe
3365 * target is not initialized.
3367 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3368 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3371 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3372 &lsm->lsm_md_oinfo[i].lmo_mds);
3379 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3380 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3386 static inline int lmv_unpack_user_md(struct obd_export *exp,
3387 struct lmv_stripe_md *lsm,
3388 const struct lmv_user_md *lmu)
3390 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3391 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3392 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3393 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3394 lsm->lsm_md_max_inherit = lmu->lum_max_inherit;
3395 lsm->lsm_md_max_inherit_rr = lmu->lum_max_inherit_rr;
3396 lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3401 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3402 const union lmv_mds_md *lmm, size_t lmm_size)
3404 struct lmv_stripe_md *lsm;
3407 bool allocated = false;
3410 LASSERT(lsmp != NULL);
3414 if (lsm != NULL && lmm == NULL) {
3416 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3418 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3421 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3423 OBD_FREE_LARGE(lfm, lfm_size);
3427 if (lmv_dir_striped(lsm)) {
3428 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3429 iput(lsm->lsm_md_oinfo[i].lmo_root);
3430 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3432 lsm_size = lmv_stripe_md_size(0);
3434 OBD_FREE(lsm, lsm_size);
3439 /* foreign lmv case */
3440 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3441 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3444 OBD_ALLOC_LARGE(lfm, lmm_size);
3447 *lsmp = (struct lmv_stripe_md *)lfm;
3449 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3450 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3451 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3452 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3453 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3458 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3462 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3463 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3464 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3465 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3470 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3471 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3474 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3475 * stripecount should be 0 then.
3477 lsm_size = lmv_stripe_md_size(0);
3480 OBD_ALLOC(lsm, lsm_size);
3487 switch (le32_to_cpu(lmm->lmv_magic)) {
3489 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3491 case LMV_USER_MAGIC:
3492 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3495 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3496 le32_to_cpu(lmm->lmv_magic));
3501 if (rc != 0 && allocated) {
3502 OBD_FREE(lsm, lsm_size);
3509 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3511 lmv_unpackmd(NULL, &lsm, NULL, 0);
3513 EXPORT_SYMBOL(lmv_free_memmd);
3515 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3516 union ldlm_policy_data *policy,
3517 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3520 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3521 struct lu_tgt_desc *tgt;
3527 LASSERT(fid != NULL);
3529 lmv_foreach_connected_tgt(lmv, tgt) {
3530 if (!tgt->ltd_active)
3533 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3541 static int lmv_set_lock_data(struct obd_export *exp,
3542 const struct lustre_handle *lockh,
3543 void *data, __u64 *bits)
3545 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3546 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3551 if (tgt == NULL || tgt->ltd_exp == NULL)
3553 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3557 static enum ldlm_mode
3558 lmv_lock_match(struct obd_export *exp, __u64 flags,
3559 const struct lu_fid *fid, enum ldlm_type type,
3560 union ldlm_policy_data *policy,
3561 enum ldlm_mode mode, struct lustre_handle *lockh)
3563 struct obd_device *obd = exp->exp_obd;
3564 struct lmv_obd *lmv = &obd->u.lmv;
3566 struct lu_tgt_desc *tgt;
3572 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3575 * With DNE every object can have two locks in different namespaces:
3576 * lookup lock in space of MDT storing direntry and update/open lock in
3577 * space of MDT storing inode. Try the MDT that the FID maps to first,
3578 * since this can be easily found, and only try others if that fails.
3580 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3581 i < lmv->lmv_mdt_descs.ltd_tgts_size;
3582 i++, index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3584 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3585 obd->obd_name, PFID(fid), index);
3589 tgt = lmv_tgt(lmv, index);
3590 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3593 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
3603 lmv_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
3604 struct obd_export *dt_exp, struct obd_export *md_exp,
3605 struct lustre_md *md)
3607 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3608 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3610 if (!tgt || !tgt->ltd_exp)
3613 return md_get_lustre_md(tgt->ltd_exp, pill, dt_exp, md_exp, md);
3616 static int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3618 struct obd_device *obd = exp->exp_obd;
3619 struct lmv_obd *lmv = &obd->u.lmv;
3620 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3624 if (md->default_lmv) {
3625 lmv_free_memmd(md->default_lmv);
3626 md->default_lmv = NULL;
3628 if (md->lmv != NULL) {
3629 lmv_free_memmd(md->lmv);
3632 if (!tgt || !tgt->ltd_exp)
3634 RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3637 static int lmv_set_open_replay_data(struct obd_export *exp,
3638 struct obd_client_handle *och,
3639 struct lookup_intent *it)
3641 struct obd_device *obd = exp->exp_obd;
3642 struct lmv_obd *lmv = &obd->u.lmv;
3643 struct lmv_tgt_desc *tgt;
3647 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3649 RETURN(PTR_ERR(tgt));
3651 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3654 static int lmv_clear_open_replay_data(struct obd_export *exp,
3655 struct obd_client_handle *och)
3657 struct obd_device *obd = exp->exp_obd;
3658 struct lmv_obd *lmv = &obd->u.lmv;
3659 struct lmv_tgt_desc *tgt;
3663 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3665 RETURN(PTR_ERR(tgt));
3667 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3670 static int lmv_intent_getattr_async(struct obd_export *exp,
3671 struct md_enqueue_info *minfo)
3673 struct md_op_data *op_data = &minfo->mi_data;
3674 struct obd_device *obd = exp->exp_obd;
3675 struct lmv_obd *lmv = &obd->u.lmv;
3676 struct lmv_tgt_desc *ptgt;
3677 struct lmv_tgt_desc *ctgt;
3682 if (!fid_is_sane(&op_data->op_fid2))
3685 ptgt = lmv_locate_tgt(lmv, op_data);
3687 RETURN(PTR_ERR(ptgt));
3689 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3691 RETURN(PTR_ERR(ctgt));
3694 * remote object needs two RPCs to lookup and getattr, considering the
3695 * complexity don't support statahead for now.
3700 rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3705 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3706 struct lu_fid *fid, __u64 *bits)
3708 struct obd_device *obd = exp->exp_obd;
3709 struct lmv_obd *lmv = &obd->u.lmv;
3710 struct lmv_tgt_desc *tgt;
3715 tgt = lmv_fid2tgt(lmv, fid);
3717 RETURN(PTR_ERR(tgt));
3719 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3723 static int lmv_get_fid_from_lsm(struct obd_export *exp,
3724 const struct lmv_stripe_md *lsm,
3725 const char *name, int namelen,
3728 const struct lmv_oinfo *oinfo;
3730 if (!lmv_dir_striped(lsm))
3733 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3735 return PTR_ERR(oinfo);
3737 *fid = oinfo->lmo_fid;
3743 * For lmv, only need to send request to master MDT, and the master MDT will
3744 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3745 * we directly fetch data from the slave MDTs.
3747 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3748 struct obd_quotactl *oqctl)
3750 struct obd_device *obd = class_exp2obd(exp);
3751 struct lmv_obd *lmv = &obd->u.lmv;
3752 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3753 __u64 curspace, curinodes;
3758 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3759 CERROR("master lmv inactive\n");
3763 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3764 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3768 curspace = curinodes = 0;
3769 lmv_foreach_connected_tgt(lmv, tgt) {
3772 if (!tgt->ltd_active)
3775 err = obd_quotactl(tgt->ltd_exp, oqctl);
3777 CERROR("getquota on mdt %d failed. %d\n",
3778 tgt->ltd_index, err);
3782 curspace += oqctl->qc_dqblk.dqb_curspace;
3783 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3786 oqctl->qc_dqblk.dqb_curspace = curspace;
3787 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3792 static int lmv_merge_attr(struct obd_export *exp,
3793 const struct lmv_stripe_md *lsm,
3794 struct cl_attr *attr,
3795 ldlm_blocking_callback cb_blocking)
3800 if (!lmv_dir_striped(lsm))
3803 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3807 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3808 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3814 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3815 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3816 i_size_read(inode), (unsigned long long)inode->i_blocks,
3817 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3818 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3820 /* for slave stripe, it needs to subtract nlink for . and .. */
3822 attr->cat_nlink += inode->i_nlink - 2;
3824 attr->cat_nlink = inode->i_nlink;
3826 attr->cat_size += i_size_read(inode);
3827 attr->cat_blocks += inode->i_blocks;
3829 if (attr->cat_atime < inode->i_atime.tv_sec)
3830 attr->cat_atime = inode->i_atime.tv_sec;
3832 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3833 attr->cat_ctime = inode->i_ctime.tv_sec;
3835 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3836 attr->cat_mtime = inode->i_mtime.tv_sec;
3841 static const struct obd_ops lmv_obd_ops = {
3842 .o_owner = THIS_MODULE,
3843 .o_setup = lmv_setup,
3844 .o_cleanup = lmv_cleanup,
3845 .o_precleanup = lmv_precleanup,
3846 .o_process_config = lmv_process_config,
3847 .o_connect = lmv_connect,
3848 .o_disconnect = lmv_disconnect,
3849 .o_statfs = lmv_statfs,
3850 .o_get_info = lmv_get_info,
3851 .o_set_info_async = lmv_set_info_async,
3852 .o_notify = lmv_notify,
3853 .o_get_uuid = lmv_get_uuid,
3854 .o_fid_alloc = lmv_fid_alloc,
3855 .o_iocontrol = lmv_iocontrol,
3856 .o_quotactl = lmv_quotactl
3859 static const struct md_ops lmv_md_ops = {
3860 .m_get_root = lmv_get_root,
3861 .m_null_inode = lmv_null_inode,
3862 .m_close = lmv_close,
3863 .m_create = lmv_create,
3864 .m_enqueue = lmv_enqueue,
3865 .m_getattr = lmv_getattr,
3866 .m_getxattr = lmv_getxattr,
3867 .m_getattr_name = lmv_getattr_name,
3868 .m_intent_lock = lmv_intent_lock,
3870 .m_rename = lmv_rename,
3871 .m_setattr = lmv_setattr,
3872 .m_setxattr = lmv_setxattr,
3873 .m_fsync = lmv_fsync,
3874 .m_file_resync = lmv_file_resync,
3875 .m_read_page = lmv_read_page,
3876 .m_unlink = lmv_unlink,
3877 .m_init_ea_size = lmv_init_ea_size,
3878 .m_cancel_unused = lmv_cancel_unused,
3879 .m_set_lock_data = lmv_set_lock_data,
3880 .m_lock_match = lmv_lock_match,
3881 .m_get_lustre_md = lmv_get_lustre_md,
3882 .m_free_lustre_md = lmv_free_lustre_md,
3883 .m_merge_attr = lmv_merge_attr,
3884 .m_set_open_replay_data = lmv_set_open_replay_data,
3885 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3886 .m_intent_getattr_async = lmv_intent_getattr_async,
3887 .m_revalidate_lock = lmv_revalidate_lock,
3888 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3889 .m_unpackmd = lmv_unpackmd,
3890 .m_rmfid = lmv_rmfid,
3893 static int __init lmv_init(void)
3895 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true,
3896 LUSTRE_LMV_NAME, NULL);
3899 static void __exit lmv_exit(void)
3901 class_unregister_type(LUSTRE_LMV_NAME);
3904 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3905 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3906 MODULE_VERSION(LUSTRE_VERSION_STRING);
3907 MODULE_LICENSE("GPL");
3909 module_init(lmv_init);
3910 module_exit(lmv_exit);