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 rc = class_disconnect(exp);
554 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
557 struct obd_device *obd = class_exp2obd(exp);
558 struct lmv_obd *lmv = &obd->u.lmv;
559 struct getinfo_fid2path *gf;
560 struct lmv_tgt_desc *tgt;
561 struct getinfo_fid2path *remote_gf = NULL;
562 struct lu_fid root_fid;
563 int remote_gf_size = 0;
567 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
569 RETURN(PTR_ERR(tgt));
571 root_fid = *gf->gf_u.gf_root_fid;
572 LASSERT(fid_is_sane(&root_fid));
575 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
576 if (rc != 0 && rc != -EREMOTE)
577 GOTO(out_fid2path, rc);
579 /* If remote_gf != NULL, it means just building the
580 * path on the remote MDT, copy this path segement to gf */
581 if (remote_gf != NULL) {
582 struct getinfo_fid2path *ori_gf;
586 ori_gf = (struct getinfo_fid2path *)karg;
587 if (strlen(ori_gf->gf_u.gf_path) + 1 +
588 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
589 GOTO(out_fid2path, rc = -EOVERFLOW);
591 ptr = ori_gf->gf_u.gf_path;
593 len = strlen(gf->gf_u.gf_path);
594 /* move the current path to the right to release space
595 * for closer-to-root part */
596 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
597 memcpy(ptr, gf->gf_u.gf_path, len);
601 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
602 tgt->ltd_exp->exp_obd->obd_name,
603 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
607 GOTO(out_fid2path, rc);
609 /* sigh, has to go to another MDT to do path building further */
610 if (remote_gf == NULL) {
611 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
612 OBD_ALLOC(remote_gf, remote_gf_size);
613 if (remote_gf == NULL)
614 GOTO(out_fid2path, rc = -ENOMEM);
615 remote_gf->gf_pathlen = PATH_MAX;
618 if (!fid_is_sane(&gf->gf_fid)) {
619 CERROR("%s: invalid FID "DFID": rc = %d\n",
620 tgt->ltd_exp->exp_obd->obd_name,
621 PFID(&gf->gf_fid), -EINVAL);
622 GOTO(out_fid2path, rc = -EINVAL);
625 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
627 GOTO(out_fid2path, rc = -EINVAL);
629 remote_gf->gf_fid = gf->gf_fid;
630 remote_gf->gf_recno = -1;
631 remote_gf->gf_linkno = -1;
632 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
633 *remote_gf->gf_u.gf_root_fid = root_fid;
635 goto repeat_fid2path;
638 if (remote_gf != NULL)
639 OBD_FREE(remote_gf, remote_gf_size);
643 static int lmv_hsm_req_count(struct lmv_obd *lmv,
644 const struct hsm_user_request *hur,
645 const struct lmv_tgt_desc *tgt_mds)
647 struct lmv_tgt_desc *curr_tgt;
651 /* count how many requests must be sent to the given target */
652 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
653 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
654 if (IS_ERR(curr_tgt))
655 RETURN(PTR_ERR(curr_tgt));
656 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
662 static int lmv_hsm_req_build(struct lmv_obd *lmv,
663 struct hsm_user_request *hur_in,
664 const struct lmv_tgt_desc *tgt_mds,
665 struct hsm_user_request *hur_out)
668 struct lmv_tgt_desc *curr_tgt;
670 /* build the hsm_user_request for the given target */
671 hur_out->hur_request = hur_in->hur_request;
673 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
674 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
675 if (IS_ERR(curr_tgt))
676 RETURN(PTR_ERR(curr_tgt));
677 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
678 hur_out->hur_user_item[nr_out] =
679 hur_in->hur_user_item[i];
683 hur_out->hur_request.hr_itemcount = nr_out;
684 memcpy(hur_data(hur_out), hur_data(hur_in),
685 hur_in->hur_request.hr_data_len);
690 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
691 int len, struct lustre_kernelcomm *lk,
694 struct lmv_obd *lmv = &obd->u.lmv;
695 struct lu_tgt_desc *tgt;
700 /* unregister request (call from llapi_hsm_copytool_fini) */
701 lmv_foreach_connected_tgt(lmv, tgt)
702 /* best effort: try to clean as much as possible
703 * (continue on error) */
704 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
706 /* Whatever the result, remove copytool from kuc groups.
707 * Unreached coordinators will get EPIPE on next requests
708 * and will unregister automatically.
710 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
715 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
716 int len, struct lustre_kernelcomm *lk,
719 struct lmv_obd *lmv = &obd->u.lmv;
721 bool any_set = false;
722 struct kkuc_ct_data *kcd;
724 struct lu_tgt_desc *tgt;
731 filp = fget(lk->lk_wfd);
735 if (lk->lk_flags & LK_FLG_DATANR)
736 kcd_size = offsetof(struct kkuc_ct_data,
737 kcd_archives[lk->lk_data_count]);
739 kcd_size = sizeof(*kcd);
741 OBD_ALLOC(kcd, kcd_size);
743 GOTO(err_fput, rc = -ENOMEM);
745 kcd->kcd_nr_archives = lk->lk_data_count;
746 if (lk->lk_flags & LK_FLG_DATANR) {
747 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
748 if (lk->lk_data_count > 0)
749 memcpy(kcd->kcd_archives, lk->lk_data,
750 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
752 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
755 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
756 lk->lk_group, kcd, kcd_size);
757 OBD_FREE(kcd, kcd_size);
761 /* All or nothing: try to register to all MDS.
762 * In case of failure, unregister from previous MDS,
763 * except if it because of inactive target. */
764 lmv_foreach_connected_tgt(lmv, tgt) {
765 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
767 if (tgt->ltd_active) {
768 /* permanent error */
769 CERROR("%s: iocontrol MDC %s on MDT"
770 " idx %d cmd %x: err = %d\n",
771 lmv2obd_dev(lmv)->obd_name,
772 tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
775 lk->lk_flags |= LK_FLG_STOP;
777 /* unregister from previous MDS */
778 lmv_foreach_connected_tgt(lmv, tgt) {
779 if (tgt->ltd_index >= i)
782 obd_iocontrol(cmd, tgt->ltd_exp, len,
785 GOTO(err_kkuc_rem, rc);
787 /* else: transient error.
788 * kuc will register to the missing MDT
796 /* no registration done: return error */
797 GOTO(err_kkuc_rem, rc = -ENOTCONN);
802 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
809 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
810 int len, void *karg, void __user *uarg)
812 struct obd_device *obd = class_exp2obd(exp);
813 struct lmv_obd *lmv = &obd->u.lmv;
814 struct lu_tgt_desc *tgt = NULL;
816 __u32 count = lmv->lmv_mdt_count;
825 case IOC_OBD_STATFS: {
826 struct obd_ioctl_data *data = karg;
827 struct obd_device *mdc_obd;
828 struct obd_statfs stat_buf = {0};
831 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
833 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
836 tgt = lmv_tgt(lmv, index);
840 if (!tgt->ltd_active)
843 mdc_obd = class_exp2obd(tgt->ltd_exp);
848 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
849 min((int) data->ioc_plen2,
850 (int) sizeof(struct obd_uuid))))
853 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
854 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
858 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
859 min((int) data->ioc_plen1,
860 (int) sizeof(stat_buf))))
864 case OBD_IOC_QUOTACTL: {
865 struct if_quotactl *qctl = karg;
866 struct obd_quotactl *oqctl;
867 struct obd_import *imp;
869 if (qctl->qc_valid == QC_MDTIDX) {
870 tgt = lmv_tgt(lmv, qctl->qc_idx);
871 } else if (qctl->qc_valid == QC_UUID) {
872 lmv_foreach_tgt(lmv, tgt) {
873 if (!obd_uuid_equals(&tgt->ltd_uuid,
892 imp = class_exp2cliimp(tgt->ltd_exp);
893 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
894 qctl->qc_valid = QC_MDTIDX;
895 qctl->obd_uuid = tgt->ltd_uuid;
899 OBD_ALLOC_PTR(oqctl);
903 QCTL_COPY(oqctl, qctl);
904 rc = obd_quotactl(tgt->ltd_exp, oqctl);
906 QCTL_COPY(qctl, oqctl);
907 qctl->qc_valid = QC_MDTIDX;
908 qctl->obd_uuid = tgt->ltd_uuid;
913 case LL_IOC_GET_CONNECT_FLAGS: {
914 tgt = lmv_tgt(lmv, 0);
916 if (tgt && tgt->ltd_exp)
917 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
920 case LL_IOC_FID2MDTIDX: {
921 struct lu_fid *fid = karg;
924 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
928 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
929 * point to user space memory for FID2MDTIDX. */
930 *(__u32 *)uarg = mdt_index;
933 case OBD_IOC_FID2PATH: {
934 rc = lmv_fid2path(exp, len, karg, uarg);
937 case LL_IOC_HSM_STATE_GET:
938 case LL_IOC_HSM_STATE_SET:
939 case LL_IOC_HSM_ACTION: {
940 struct md_op_data *op_data = karg;
942 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
944 RETURN(PTR_ERR(tgt));
946 if (tgt->ltd_exp == NULL)
949 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
952 case LL_IOC_HSM_PROGRESS: {
953 const struct hsm_progress_kernel *hpk = karg;
955 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
957 RETURN(PTR_ERR(tgt));
958 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
961 case LL_IOC_HSM_REQUEST: {
962 struct hsm_user_request *hur = karg;
963 unsigned int reqcount = hur->hur_request.hr_itemcount;
968 /* if the request is about a single fid
969 * or if there is a single MDS, no need to split
971 if (reqcount == 1 || count == 1) {
972 tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
974 RETURN(PTR_ERR(tgt));
975 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
977 /* split fid list to their respective MDS */
978 lmv_foreach_connected_tgt(lmv, tgt) {
981 struct hsm_user_request *req;
983 nr = lmv_hsm_req_count(lmv, hur, tgt);
986 if (nr == 0) /* nothing for this MDS */
989 /* build a request with fids for this MDS */
990 reqlen = offsetof(typeof(*hur),
992 + hur->hur_request.hr_data_len;
993 OBD_ALLOC_LARGE(req, reqlen);
996 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
998 GOTO(hsm_req_err, rc1);
999 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1002 if (rc1 != 0 && rc == 0)
1004 OBD_FREE_LARGE(req, reqlen);
1009 case LL_IOC_LOV_SWAP_LAYOUTS: {
1010 struct md_op_data *op_data = karg;
1011 struct lmv_tgt_desc *tgt1, *tgt2;
1013 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1015 RETURN(PTR_ERR(tgt1));
1017 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1019 RETURN(PTR_ERR(tgt2));
1021 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1024 /* only files on same MDT can have their layouts swapped */
1025 if (tgt1->ltd_index != tgt2->ltd_index)
1028 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1031 case LL_IOC_HSM_CT_START: {
1032 struct lustre_kernelcomm *lk = karg;
1033 if (lk->lk_flags & LK_FLG_STOP)
1034 rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1036 rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1040 lmv_foreach_connected_tgt(lmv, tgt) {
1041 struct obd_device *mdc_obd;
1044 /* ll_umount_begin() sets force flag but for lmv, not
1045 * mdc. Let's pass it through */
1046 mdc_obd = class_exp2obd(tgt->ltd_exp);
1047 mdc_obd->obd_force = obd->obd_force;
1048 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1050 if (tgt->ltd_active) {
1051 CERROR("error: iocontrol MDC %s on MDT"
1052 " idx %d cmd %x: err = %d\n",
1054 tgt->ltd_index, cmd, err);
1067 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1068 struct lu_fid *fid, struct md_op_data *op_data)
1070 struct obd_device *obd = class_exp2obd(exp);
1071 struct lmv_obd *lmv = &obd->u.lmv;
1072 struct lmv_tgt_desc *tgt;
1080 tgt = lmv_tgt(lmv, op_data->op_mds);
1084 if (!tgt->ltd_active || !tgt->ltd_exp)
1088 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1089 * on server that seq in new allocated fid is not yet known.
1091 mutex_lock(&tgt->ltd_fid_mutex);
1092 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1093 mutex_unlock(&tgt->ltd_fid_mutex);
1095 LASSERT(fid_is_sane(fid));
1102 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1104 struct lmv_obd *lmv = &obd->u.lmv;
1105 struct lmv_desc *desc;
1106 struct lnet_processid lnet_id;
1112 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1113 CERROR("LMV setup requires a descriptor\n");
1117 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1118 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1119 CERROR("Lmv descriptor size wrong: %d > %d\n",
1120 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1124 obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1125 desc->ld_uuid.uuid);
1126 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1127 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1128 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1129 LMV_DESC_QOS_MAXAGE_DEFAULT;
1130 lmv->max_def_easize = 0;
1131 lmv->max_easize = 0;
1133 spin_lock_init(&lmv->lmv_lock);
1136 * initialize rr_index to lower 32bit of netid, so that client
1137 * can distribute subdirs evenly from the beginning.
1139 while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1140 if (!nid_is_lo0(&lnet_id.nid)) {
1141 lmv->lmv_qos_rr_index = ntohl(lnet_id.nid.nid_addr[0]);
1146 rc = lmv_tunables_init(obd);
1148 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1151 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1152 LUSTRE_CLI_FLD_HASH_DHT);
1154 CERROR("Can't init FLD, err %d\n", rc);
1156 rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1158 CWARN("%s: error initialize target table: rc = %d\n",
1164 static int lmv_cleanup(struct obd_device *obd)
1166 struct lmv_obd *lmv = &obd->u.lmv;
1167 struct lu_tgt_desc *tgt;
1168 struct lu_tgt_desc *tmp;
1172 fld_client_fini(&lmv->lmv_fld);
1173 fld_client_debugfs_fini(&lmv->lmv_fld);
1175 lprocfs_obd_cleanup(obd);
1176 lprocfs_free_md_stats(obd);
1178 lmv_foreach_tgt_safe(lmv, tgt, tmp)
1179 lmv_del_target(lmv, tgt);
1180 lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1185 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1187 struct lustre_cfg *lcfg = buf;
1188 struct obd_uuid obd_uuid;
1194 switch (lcfg->lcfg_command) {
1196 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1197 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1198 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1199 GOTO(out, rc = -EINVAL);
1201 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1203 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1204 GOTO(out, rc = -EINVAL);
1205 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1206 GOTO(out, rc = -EINVAL);
1207 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1210 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1211 GOTO(out, rc = -EINVAL);
1217 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1221 if (flags & OBD_STATFS_FOR_MDT0)
1224 if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1225 return lmv->lmv_statfs_start;
1227 /* choose initial MDT for this client */
1229 struct lnet_processid lnet_id;
1230 if (LNetGetId(i, &lnet_id) == -ENOENT)
1233 if (!nid_is_lo0(&lnet_id.nid)) {
1234 /* We dont need a full 64-bit modulus, just enough
1235 * to distribute the requests across MDTs evenly.
1237 lmv->lmv_statfs_start = nidhash(&lnet_id.nid) %
1243 return lmv->lmv_statfs_start;
1246 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1247 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1249 struct obd_device *obd = class_exp2obd(exp);
1250 struct lmv_obd *lmv = &obd->u.lmv;
1251 struct obd_statfs *temp;
1252 struct lu_tgt_desc *tgt;
1260 OBD_ALLOC(temp, sizeof(*temp));
1264 /* distribute statfs among MDTs */
1265 idx = lmv_select_statfs_mdt(lmv, flags);
1267 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1268 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1269 tgt = lmv_tgt(lmv, idx);
1270 if (!tgt || !tgt->ltd_exp)
1273 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age,
1274 flags | OBD_STATFS_NESTED);
1276 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1277 tgt->ltd_exp->exp_obd->obd_name, i, rc);
1279 /* Try another MDT */
1280 if (flags & OBD_STATFS_SUM)
1282 GOTO(out_free_temp, rc);
1285 if (temp->os_state & OS_STATFS_SUM ||
1286 flags == OBD_STATFS_FOR_MDT0) {
1287 /* reset to the last aggregated values
1288 * and don't sum with non-aggrated data */
1289 /* If the statfs is from mount, it needs to retrieve
1290 * necessary information from MDT0. i.e. mount does
1291 * not need the merged osfs from all of MDT. Also
1292 * clients can be mounted as long as MDT0 is in
1295 GOTO(out_free_temp, rc);
1301 osfs->os_bavail += temp->os_bavail;
1302 osfs->os_blocks += temp->os_blocks;
1303 osfs->os_ffree += temp->os_ffree;
1304 osfs->os_files += temp->os_files;
1305 osfs->os_granted += temp->os_granted;
1308 /* There is no stats from some MDTs, data incomplete */
1312 OBD_FREE(temp, sizeof(*temp));
1316 static int lmv_statfs_update(void *cookie, int rc)
1318 struct obd_info *oinfo = cookie;
1319 struct obd_device *obd = oinfo->oi_obd;
1320 struct lmv_obd *lmv = &obd->u.lmv;
1321 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1322 struct obd_statfs *osfs = oinfo->oi_osfs;
1325 * NB: don't deactivate TGT upon error, because we may not trigger async
1326 * statfs any longer, then there is no chance to activate TGT.
1329 spin_lock(&lmv->lmv_lock);
1330 tgt->ltd_statfs = *osfs;
1331 tgt->ltd_statfs_age = ktime_get_seconds();
1332 spin_unlock(&lmv->lmv_lock);
1333 set_bit(LQ_DIRTY, &lmv->lmv_qos.lq_flags);
1339 /* update tgt statfs async if it's ld_qos_maxage old */
1340 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1342 struct obd_info oinfo = {
1345 .oi_cb_up = lmv_statfs_update,
1349 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1350 obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1353 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1358 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1361 struct obd_device *obd = exp->exp_obd;
1362 struct lmv_obd *lmv = &obd->u.lmv;
1363 struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1371 rc = md_get_root(tgt->ltd_exp, fileset, fid);
1375 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1376 u64 obd_md_valid, const char *name, size_t buf_size,
1377 struct ptlrpc_request **req)
1379 struct obd_device *obd = exp->exp_obd;
1380 struct lmv_obd *lmv = &obd->u.lmv;
1381 struct lmv_tgt_desc *tgt;
1386 tgt = lmv_fid2tgt(lmv, fid);
1388 RETURN(PTR_ERR(tgt));
1390 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1395 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1396 u64 obd_md_valid, const char *name,
1397 const void *value, size_t value_size,
1398 unsigned int xattr_flags, u32 suppgid,
1399 struct ptlrpc_request **req)
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, fid);
1410 RETURN(PTR_ERR(tgt));
1412 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1413 value, value_size, xattr_flags, suppgid, req);
1418 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1419 struct ptlrpc_request **request)
1421 struct obd_device *obd = exp->exp_obd;
1422 struct lmv_obd *lmv = &obd->u.lmv;
1423 struct lmv_tgt_desc *tgt;
1428 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1430 RETURN(PTR_ERR(tgt));
1432 if (op_data->op_flags & MF_GET_MDT_IDX) {
1433 op_data->op_mds = tgt->ltd_index;
1437 rc = md_getattr(tgt->ltd_exp, op_data, request);
1442 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1444 struct obd_device *obd = exp->exp_obd;
1445 struct lmv_obd *lmv = &obd->u.lmv;
1446 struct lu_tgt_desc *tgt;
1450 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1453 * With DNE every object can have two locks in different namespaces:
1454 * lookup lock in space of MDT storing direntry and update/open lock in
1455 * space of MDT storing inode.
1457 lmv_foreach_connected_tgt(lmv, tgt)
1458 md_null_inode(tgt->ltd_exp, fid);
1463 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1464 struct md_open_data *mod, struct ptlrpc_request **request)
1466 struct obd_device *obd = exp->exp_obd;
1467 struct lmv_obd *lmv = &obd->u.lmv;
1468 struct lmv_tgt_desc *tgt;
1473 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1475 RETURN(PTR_ERR(tgt));
1477 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1478 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1482 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv, __u32 mdt,
1483 unsigned short dir_depth)
1485 struct lu_tgt_desc *tgt, *cur = NULL;
1486 __u64 total_avail = 0;
1487 __u64 total_weight = 0;
1488 __u64 cur_weight = 0;
1489 int total_usable = 0;
1495 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1496 RETURN(ERR_PTR(-EAGAIN));
1498 down_write(&lmv->lmv_qos.lq_rw_sem);
1500 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1501 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1503 rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1505 GOTO(unlock, tgt = ERR_PTR(rc));
1507 lmv_foreach_tgt(lmv, tgt) {
1508 if (!tgt->ltd_exp || !tgt->ltd_active) {
1509 tgt->ltd_qos.ltq_usable = 0;
1513 tgt->ltd_qos.ltq_usable = 1;
1514 lu_tgt_qos_weight_calc(tgt);
1515 if (tgt->ltd_index == mdt)
1517 total_avail += tgt->ltd_qos.ltq_avail;
1518 total_weight += tgt->ltd_qos.ltq_weight;
1522 /* if current MDT has above-average space, within range of the QOS
1523 * threshold, stay on the same MDT to avoid creating needless remote
1524 * MDT directories. It's more likely for low level directories
1525 * "16 / (dir_depth + 10)" is the factor to make it more unlikely for
1526 * top level directories, while more likely for low levels.
1528 rand = total_avail * 16 / (total_usable * (dir_depth + 10));
1529 if (cur && cur->ltd_qos.ltq_avail >= rand) {
1534 rand = lu_prandom_u64_max(total_weight);
1536 lmv_foreach_connected_tgt(lmv, tgt) {
1537 if (!tgt->ltd_qos.ltq_usable)
1540 cur_weight += tgt->ltd_qos.ltq_weight;
1541 if (cur_weight < rand)
1544 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1548 /* no proper target found */
1549 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1551 up_write(&lmv->lmv_qos.lq_rw_sem);
1556 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv)
1558 struct lu_tgt_desc *tgt;
1564 spin_lock(&lmv->lmv_lock);
1565 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1566 index = (i + lmv->lmv_qos_rr_index) %
1567 lmv->lmv_mdt_descs.ltd_tgts_size;
1568 tgt = lmv_tgt(lmv, index);
1569 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1572 lmv->lmv_qos_rr_index = (tgt->ltd_index + 1) %
1573 lmv->lmv_mdt_descs.ltd_tgts_size;
1574 spin_unlock(&lmv->lmv_lock);
1578 spin_unlock(&lmv->lmv_lock);
1580 RETURN(ERR_PTR(-ENODEV));
1583 /* locate MDT which is less full (avoid the most full MDT) */
1584 static struct lu_tgt_desc *lmv_locate_tgt_lf(struct lmv_obd *lmv)
1586 struct lu_tgt_desc *min = NULL;
1587 struct lu_tgt_desc *tgt;
1593 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1594 RETURN(ERR_PTR(-EAGAIN));
1596 down_write(&lmv->lmv_qos.lq_rw_sem);
1598 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1599 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1601 lmv_foreach_tgt(lmv, tgt) {
1602 if (!tgt->ltd_exp || !tgt->ltd_active) {
1603 tgt->ltd_qos.ltq_usable = 0;
1607 tgt->ltd_qos.ltq_usable = 1;
1608 lu_tgt_qos_weight_calc(tgt);
1609 avail += tgt->ltd_qos.ltq_avail;
1610 if (!min || min->ltd_qos.ltq_avail > tgt->ltd_qos.ltq_avail)
1614 /* avoid the most full MDT */
1616 avail -= min->ltd_qos.ltq_avail;
1618 rand = lu_prandom_u64_max(avail);
1620 lmv_foreach_connected_tgt(lmv, tgt) {
1621 if (!tgt->ltd_qos.ltq_usable)
1627 avail += tgt->ltd_qos.ltq_avail;
1634 /* no proper target found */
1635 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1637 up_write(&lmv->lmv_qos.lq_rw_sem);
1642 /* locate MDT by file name, for striped directory, the file name hash decides
1643 * which stripe its dirent is stored.
1645 static struct lmv_tgt_desc *
1646 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1647 const char *name, int namelen, struct lu_fid *fid,
1648 __u32 *mds, bool new_layout)
1650 struct lmv_tgt_desc *tgt;
1651 const struct lmv_oinfo *oinfo;
1653 if (!lmv_dir_striped(lsm) || !namelen) {
1654 tgt = lmv_fid2tgt(lmv, fid);
1658 *mds = tgt->ltd_index;
1662 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1663 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1664 return ERR_PTR(-EBADF);
1665 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1667 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, new_layout);
1669 return ERR_CAST(oinfo);
1672 /* check stripe FID is sane */
1673 if (!fid_is_sane(&oinfo->lmo_fid))
1674 return ERR_PTR(-ENODEV);
1676 *fid = oinfo->lmo_fid;
1677 *mds = oinfo->lmo_mds;
1678 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1680 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1682 return tgt ? tgt : ERR_PTR(-ENODEV);
1686 * Locate MDT of op_data->op_fid1
1688 * For striped directory, it will locate the stripe by name hash, if hash_type
1689 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1690 * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1691 * indicates whether old or new layout is used to locate.
1693 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1695 * \param[in] lmv LMV device
1696 * \param[in/out] op_data client MD stack parameters, name, namelen etc,
1697 * op_mds and op_fid1 will be updated if op_mea1
1698 * indicates fid1 represents a striped directory.
1700 * retval pointer to the lmv_tgt_desc if succeed.
1701 * ERR_PTR(errno) if failed.
1703 struct lmv_tgt_desc *
1704 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1706 struct lmv_stripe_md *lsm = op_data->op_mea1;
1707 struct lmv_oinfo *oinfo;
1708 struct lmv_tgt_desc *tgt;
1710 if (lmv_dir_foreign(lsm))
1711 return ERR_PTR(-ENODATA);
1713 /* During creating VOLATILE file, it should honor the mdt
1714 * index if the file under striped dir is being restored, see
1716 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1717 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1718 tgt = lmv_tgt(lmv, op_data->op_mds);
1720 return ERR_PTR(-ENODEV);
1722 if (lmv_dir_striped(lsm)) {
1725 /* refill the right parent fid */
1726 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1727 oinfo = &lsm->lsm_md_oinfo[i];
1728 if (oinfo->lmo_mds == op_data->op_mds) {
1729 op_data->op_fid1 = oinfo->lmo_fid;
1734 if (i == lsm->lsm_md_stripe_count)
1735 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1737 } else if (lmv_dir_bad_hash(lsm)) {
1738 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1739 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1741 op_data->op_fid1 = oinfo->lmo_fid;
1742 op_data->op_mds = oinfo->lmo_mds;
1743 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1745 return ERR_PTR(-ENODEV);
1747 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1748 op_data->op_name, op_data->op_namelen,
1749 &op_data->op_fid1, &op_data->op_mds,
1750 op_data->op_new_layout);
1756 /* Locate MDT of op_data->op_fid2 for link/rename */
1757 static struct lmv_tgt_desc *
1758 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1760 struct lmv_tgt_desc *tgt;
1763 LASSERT(op_data->op_name);
1764 if (lmv_dir_layout_changing(op_data->op_mea2)) {
1765 struct lu_fid fid1 = op_data->op_fid1;
1766 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1767 struct ptlrpc_request *request = NULL;
1770 * avoid creating new file under old layout of migrating
1771 * directory, check it here.
1773 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1774 op_data->op_name, op_data->op_namelen,
1775 &op_data->op_fid2, &op_data->op_mds, false);
1779 op_data->op_fid1 = op_data->op_fid2;
1780 op_data->op_mea1 = op_data->op_mea2;
1781 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1782 op_data->op_fid1 = fid1;
1783 op_data->op_mea1 = lsm1;
1785 ptlrpc_req_finished(request);
1786 RETURN(ERR_PTR(-EEXIST));
1790 RETURN(ERR_PTR(rc));
1793 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1794 op_data->op_namelen, &op_data->op_fid2,
1795 &op_data->op_mds, true);
1798 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1800 struct lu_tgt_desc *tgt;
1801 struct ptlrpc_request *request;
1804 LASSERT(lmv_dir_layout_changing(op_data->op_mea1));
1805 LASSERT(!op_data->op_new_layout);
1807 tgt = lmv_locate_tgt(lmv, op_data);
1809 return PTR_ERR(tgt);
1811 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1813 ptlrpc_req_finished(request);
1820 /* mkdir by QoS upon 'lfs mkdir -i -1'.
1822 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1823 * directories under striped directory.
1825 static inline bool lmv_op_user_qos_mkdir(const struct md_op_data *op_data)
1827 const struct lmv_user_md *lum = op_data->op_data;
1829 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1832 if (lmv_dir_striped(op_data->op_mea1))
1835 return (op_data->op_cli_flags & CLI_SET_MEA) && lum &&
1836 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
1837 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
1840 /* mkdir by QoS if either ROOT or parent default LMV is space balanced. */
1841 static inline bool lmv_op_default_qos_mkdir(const struct md_op_data *op_data)
1843 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1845 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1848 if (lmv_dir_striped(op_data->op_mea1))
1851 return (op_data->op_flags & MF_QOS_MKDIR) ||
1852 (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT);
1855 /* if parent default LMV is space balanced, and
1856 * 1. max_inherit_rr is set
1857 * 2. or parent is ROOT
1858 * mkdir roundrobin. Or if parent doesn't have default LMV, while ROOT default
1859 * LMV requests roundrobin mkdir, do the same.
1860 * NB, this needs to check server is balanced, which is done by caller.
1862 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data)
1864 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1866 if (!lmv_op_default_qos_mkdir(op_data))
1869 return (op_data->op_flags & MF_RR_MKDIR) ||
1870 (lsm && lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE) ||
1871 fid_is_root(&op_data->op_fid1);
1874 /* 'lfs mkdir -i <specific_MDT>' */
1875 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1877 const struct lmv_user_md *lum = op_data->op_data;
1879 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1880 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1881 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1882 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1883 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1886 /* parent default LMV master_mdt_index is not -1. */
1888 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1890 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1891 op_data->op_default_mea1 &&
1892 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1896 /* locate MDT by space usage */
1897 static struct lu_tgt_desc *lmv_locate_tgt_by_space(struct lmv_obd *lmv,
1898 struct md_op_data *op_data,
1899 struct lmv_tgt_desc *tgt)
1901 struct lmv_tgt_desc *tmp = tgt;
1903 tgt = lmv_locate_tgt_qos(lmv, op_data->op_mds, op_data->op_dir_depth);
1904 if (tgt == ERR_PTR(-EAGAIN)) {
1905 if (ltd_qos_is_balanced(&lmv->lmv_mdt_descs) &&
1906 !lmv_op_default_rr_mkdir(op_data) &&
1907 !lmv_op_user_qos_mkdir(op_data))
1908 /* if not necessary, don't create remote directory. */
1911 tgt = lmv_locate_tgt_rr(lmv);
1915 * only update statfs after QoS mkdir, this means the cached statfs may
1916 * be stale, and current mkdir may not follow QoS accurately, but it's
1917 * not serious, and avoids periodic statfs when client doesn't mkdir by
1921 op_data->op_mds = tgt->ltd_index;
1922 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1928 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1929 const void *data, size_t datalen, umode_t mode, uid_t uid,
1930 gid_t gid, kernel_cap_t cap_effective, __u64 rdev,
1931 struct ptlrpc_request **request)
1933 struct obd_device *obd = exp->exp_obd;
1934 struct lmv_obd *lmv = &obd->u.lmv;
1935 struct lmv_tgt_desc *tgt;
1936 struct mdt_body *repbody;
1941 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1944 if (lmv_dir_bad_hash(op_data->op_mea1))
1947 if (lmv_dir_layout_changing(op_data->op_mea1)) {
1949 * if parent is migrating, create() needs to lookup existing
1950 * name in both old and new layout, check old layout on client.
1952 rc = lmv_old_layout_lookup(lmv, op_data);
1956 op_data->op_new_layout = true;
1959 tgt = lmv_locate_tgt(lmv, op_data);
1961 RETURN(PTR_ERR(tgt));
1963 /* the order to apply policy in mkdir:
1964 * 1. is "lfs mkdir -i N"? mkdir on MDT N.
1965 * 2. is "lfs mkdir -i -1"? mkdir by space usage.
1966 * 3. is starting MDT specified in default LMV? mkdir on MDT N.
1967 * 4. is default LMV space balanced? mkdir by space usage.
1969 if (lmv_op_user_specific_mkdir(op_data)) {
1970 struct lmv_user_md *lum = op_data->op_data;
1972 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
1973 tgt = lmv_tgt(lmv, op_data->op_mds);
1976 } else if (lmv_op_user_qos_mkdir(op_data)) {
1977 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
1979 RETURN(PTR_ERR(tgt));
1980 } else if (lmv_op_default_specific_mkdir(op_data)) {
1982 op_data->op_default_mea1->lsm_md_master_mdt_index;
1983 tgt = lmv_tgt(lmv, op_data->op_mds);
1986 } else if (lmv_op_default_qos_mkdir(op_data)) {
1987 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
1989 RETURN(PTR_ERR(tgt));
1993 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1997 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
1998 (int)op_data->op_namelen, op_data->op_name,
1999 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
2002 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2003 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
2004 cap_effective, rdev, request);
2006 if (*request == NULL)
2008 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
2011 /* dir restripe needs to send to MDT where dir is located */
2012 if (rc != -EREMOTE ||
2013 !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
2016 repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2017 if (repbody == NULL)
2020 /* Not cross-ref case, just get out of here. */
2021 if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
2024 op_data->op_fid2 = repbody->mbo_fid1;
2025 ptlrpc_req_finished(*request);
2028 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2030 RETURN(PTR_ERR(tgt));
2032 op_data->op_mds = tgt->ltd_index;
2037 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
2038 const union ldlm_policy_data *policy, struct md_op_data *op_data,
2039 struct lustre_handle *lockh, __u64 extra_lock_flags)
2041 struct obd_device *obd = exp->exp_obd;
2042 struct lmv_obd *lmv = &obd->u.lmv;
2043 struct lmv_tgt_desc *tgt;
2048 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
2050 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2052 RETURN(PTR_ERR(tgt));
2054 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
2055 PFID(&op_data->op_fid1), tgt->ltd_index);
2057 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
2064 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
2065 struct ptlrpc_request **preq)
2067 struct obd_device *obd = exp->exp_obd;
2068 struct lmv_obd *lmv = &obd->u.lmv;
2069 struct lmv_tgt_desc *tgt;
2070 struct mdt_body *body;
2076 if (op_data->op_namelen == 2 &&
2077 op_data->op_name[0] == '.' && op_data->op_name[1] == '.')
2078 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2080 tgt = lmv_locate_tgt(lmv, op_data);
2082 RETURN(PTR_ERR(tgt));
2084 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
2085 (int)op_data->op_namelen, op_data->op_name,
2086 PFID(&op_data->op_fid1), tgt->ltd_index);
2088 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2089 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2090 ptlrpc_req_finished(*preq);
2098 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2099 LASSERT(body != NULL);
2101 if (body->mbo_valid & OBD_MD_MDS) {
2102 op_data->op_fid1 = body->mbo_fid1;
2103 op_data->op_valid |= OBD_MD_FLCROSSREF;
2104 op_data->op_namelen = 0;
2105 op_data->op_name = NULL;
2107 ptlrpc_req_finished(*preq);
2116 #define md_op_data_fid(op_data, fl) \
2117 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2118 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2119 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2120 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2123 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2124 struct md_op_data *op_data, __u32 op_tgt,
2125 enum ldlm_mode mode, int bits, int flag)
2127 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2128 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2129 union ldlm_policy_data policy = { { 0 } };
2133 if (!fid_is_sane(fid))
2137 tgt = lmv_fid2tgt(lmv, fid);
2139 RETURN(PTR_ERR(tgt));
2142 if (tgt->ltd_index != op_tgt) {
2143 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2144 policy.l_inodebits.bits = bits;
2145 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2146 mode, LCF_ASYNC, NULL);
2149 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2151 op_data->op_flags |= flag;
2159 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2162 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2163 struct ptlrpc_request **request)
2165 struct obd_device *obd = exp->exp_obd;
2166 struct lmv_obd *lmv = &obd->u.lmv;
2167 struct lmv_tgt_desc *tgt;
2171 LASSERT(op_data->op_namelen != 0);
2173 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2174 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2175 op_data->op_name, PFID(&op_data->op_fid1));
2177 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2178 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2179 op_data->op_cap = current_cap();
2181 tgt = lmv_locate_tgt2(lmv, op_data);
2183 RETURN(PTR_ERR(tgt));
2186 * Cancel UPDATE lock on child (fid1).
2188 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2189 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2190 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2194 rc = md_link(tgt->ltd_exp, op_data, request);
2199 /* migrate the top directory */
2200 static inline bool lmv_op_topdir_migrate(const struct md_op_data *op_data)
2202 if (!S_ISDIR(op_data->op_mode))
2205 if (lmv_dir_layout_changing(op_data->op_mea1))
2211 /* migrate top dir to specific MDTs */
2212 static inline bool lmv_topdir_specific_migrate(const struct md_op_data *op_data)
2214 const struct lmv_user_md *lum = op_data->op_data;
2216 if (!lmv_op_topdir_migrate(op_data))
2219 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2222 /* migrate top dir in QoS mode if user issued "lfs migrate -m -1..." */
2223 static inline bool lmv_topdir_qos_migrate(const struct md_op_data *op_data)
2225 const struct lmv_user_md *lum = op_data->op_data;
2227 if (!lmv_op_topdir_migrate(op_data))
2230 return le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
2233 static inline bool lmv_subdir_specific_migrate(const struct md_op_data *op_data)
2235 const struct lmv_user_md *lum = op_data->op_data;
2237 if (!S_ISDIR(op_data->op_mode))
2240 if (!lmv_dir_layout_changing(op_data->op_mea1))
2243 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2246 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2247 const char *name, size_t namelen,
2248 struct ptlrpc_request **request)
2250 struct obd_device *obd = exp->exp_obd;
2251 struct lmv_obd *lmv = &obd->u.lmv;
2252 struct lmv_stripe_md *lsm = op_data->op_mea1;
2253 struct lmv_tgt_desc *parent_tgt;
2254 struct lmv_tgt_desc *sp_tgt;
2255 struct lmv_tgt_desc *tp_tgt = NULL;
2256 struct lmv_tgt_desc *child_tgt;
2257 struct lmv_tgt_desc *tgt;
2258 struct lu_fid target_fid = { 0 };
2263 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2265 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2266 PFID(&op_data->op_fid1), (int)namelen, name);
2268 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2269 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2270 op_data->op_cap = current_cap();
2272 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2273 if (IS_ERR(parent_tgt))
2274 RETURN(PTR_ERR(parent_tgt));
2276 if (lmv_dir_striped(lsm)) {
2277 const struct lmv_oinfo *oinfo;
2279 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
2281 RETURN(PTR_ERR(oinfo));
2283 /* save source stripe FID in fid4 temporarily for ELC */
2284 op_data->op_fid4 = oinfo->lmo_fid;
2285 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2290 * if parent is being migrated too, fill op_fid2 with target
2291 * stripe fid, otherwise the target stripe is not created yet.
2293 if (lmv_dir_layout_changing(lsm)) {
2294 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
2297 RETURN(PTR_ERR(oinfo));
2299 op_data->op_fid2 = oinfo->lmo_fid;
2300 tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2304 /* parent unchanged and update namespace only */
2305 if (lu_fid_eq(&op_data->op_fid4, &op_data->op_fid2) &&
2306 op_data->op_bias & MDS_MIGRATE_NSONLY)
2310 sp_tgt = parent_tgt;
2313 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2314 if (IS_ERR(child_tgt))
2315 RETURN(PTR_ERR(child_tgt));
2317 if (lmv_topdir_specific_migrate(op_data)) {
2318 struct lmv_user_md *lum = op_data->op_data;
2320 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2321 } else if (lmv_topdir_qos_migrate(op_data)) {
2322 tgt = lmv_locate_tgt_lf(lmv);
2323 if (tgt == ERR_PTR(-EAGAIN))
2324 tgt = lmv_locate_tgt_rr(lmv);
2326 RETURN(PTR_ERR(tgt));
2328 op_data->op_mds = tgt->ltd_index;
2329 } else if (lmv_subdir_specific_migrate(op_data)) {
2330 struct lmv_user_md *lum = op_data->op_data;
2334 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2335 /* adjust MDTs in lum, since subdir is located on where
2336 * its parent stripe is, not the first specified MDT.
2338 for (i = 0; i < le32_to_cpu(lum->lum_stripe_count);
2340 if (le32_to_cpu(lum->lum_objects[i].lum_mds) ==
2345 if (i == le32_to_cpu(lum->lum_stripe_count))
2348 lum->lum_objects[i].lum_mds =
2349 lum->lum_objects[0].lum_mds;
2350 lum->lum_objects[0].lum_mds =
2351 cpu_to_le32(tp_tgt->ltd_index);
2353 /* NB, the above adjusts subdir migration for command like
2354 * "lfs migrate -m 0,1,2 ...", but for migration like
2355 * "lfs migrate -m 0 -c 2 ...", the top dir is migrated to MDT0
2356 * and MDT1, however its subdir may be migrated to MDT1 and MDT2
2359 lum->lum_stripe_offset = cpu_to_le32(tp_tgt->ltd_index);
2360 op_data->op_mds = tp_tgt->ltd_index;
2361 } else if (tp_tgt) {
2362 op_data->op_mds = tp_tgt->ltd_index;
2364 op_data->op_mds = sp_tgt->ltd_index;
2367 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2372 * for directory, send migrate request to the MDT where the object will
2373 * be migrated to, because we can't create a striped directory remotely.
2375 * otherwise, send to the MDT where source is located because regular
2376 * file may open lease.
2378 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2379 * backward compatibility.
2381 if (S_ISDIR(op_data->op_mode) &&
2382 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2383 tgt = lmv_fid2tgt(lmv, &target_fid);
2385 RETURN(PTR_ERR(tgt));
2390 /* cancel UPDATE lock of parent master object */
2391 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2392 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2396 /* cancel UPDATE lock of source parent */
2397 if (sp_tgt != parent_tgt) {
2399 * migrate RPC packs master object FID, because we can only pack
2400 * two FIDs in reint RPC, but MDS needs to know both source
2401 * parent and target parent, and it will obtain them from master
2402 * FID and LMV, the other FID in RPC is kept for target.
2404 * since this FID is not passed to MDC, cancel it anyway.
2406 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2407 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2411 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2413 op_data->op_fid4 = target_fid;
2415 /* cancel UPDATE locks of target parent */
2416 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2417 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2421 /* cancel LOOKUP lock of source if source is remote object */
2422 if (child_tgt != sp_tgt) {
2423 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2424 LCK_EX, MDS_INODELOCK_LOOKUP,
2425 MF_MDC_CANCEL_FID3);
2430 /* cancel ELC locks of source */
2431 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2432 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2436 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2441 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2442 const char *old, size_t oldlen,
2443 const char *new, size_t newlen,
2444 struct ptlrpc_request **request)
2446 struct obd_device *obd = exp->exp_obd;
2447 struct lmv_obd *lmv = &obd->u.lmv;
2448 struct lmv_tgt_desc *sp_tgt;
2449 struct lmv_tgt_desc *tp_tgt = NULL;
2450 struct lmv_tgt_desc *src_tgt = NULL;
2451 struct lmv_tgt_desc *tgt;
2452 struct mdt_body *body;
2457 LASSERT(oldlen != 0);
2459 if (op_data->op_cli_flags & CLI_MIGRATE) {
2460 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2464 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2465 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2466 op_data->op_cap = current_cap();
2468 op_data->op_name = new;
2469 op_data->op_namelen = newlen;
2471 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2473 RETURN(PTR_ERR(tp_tgt));
2475 /* Since the target child might be destroyed, and it might become
2476 * orphan, and we can only check orphan on the local MDT right now, so
2477 * we send rename request to the MDT where target child is located. If
2478 * target child does not exist, then it will send the request to the
2480 if (fid_is_sane(&op_data->op_fid4)) {
2481 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2483 RETURN(PTR_ERR(tgt));
2488 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2490 /* cancel UPDATE locks of target parent */
2491 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2492 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2496 if (fid_is_sane(&op_data->op_fid4)) {
2497 /* cancel LOOKUP lock of target on target parent */
2498 if (tgt != tp_tgt) {
2499 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2500 tgt->ltd_index, LCK_EX,
2501 MDS_INODELOCK_LOOKUP,
2502 MF_MDC_CANCEL_FID4);
2508 if (fid_is_sane(&op_data->op_fid3)) {
2509 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2510 if (IS_ERR(src_tgt))
2511 RETURN(PTR_ERR(src_tgt));
2513 /* cancel ELC locks of source */
2514 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2515 LCK_EX, MDS_INODELOCK_ELC,
2516 MF_MDC_CANCEL_FID3);
2521 op_data->op_name = old;
2522 op_data->op_namelen = oldlen;
2524 sp_tgt = lmv_locate_tgt(lmv, op_data);
2526 RETURN(PTR_ERR(sp_tgt));
2528 /* cancel UPDATE locks of source parent */
2529 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2530 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2534 if (fid_is_sane(&op_data->op_fid3)) {
2535 /* cancel LOOKUP lock of source on source parent */
2536 if (src_tgt != sp_tgt) {
2537 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2538 tgt->ltd_index, LCK_EX,
2539 MDS_INODELOCK_LOOKUP,
2540 MF_MDC_CANCEL_FID3);
2547 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2548 PFID(&op_data->op_fid1), (int)oldlen, old,
2549 PFID(&op_data->op_fid2), (int)newlen, new);
2551 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2553 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2554 ptlrpc_req_finished(*request);
2559 if (rc && rc != -EXDEV)
2562 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2566 /* Not cross-ref case, just get out of here. */
2567 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2570 op_data->op_fid4 = body->mbo_fid1;
2572 ptlrpc_req_finished(*request);
2575 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2577 RETURN(PTR_ERR(tgt));
2579 if (fid_is_sane(&op_data->op_fid4)) {
2580 /* cancel LOOKUP lock of target on target parent */
2581 if (tgt != tp_tgt) {
2582 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2583 tgt->ltd_index, LCK_EX,
2584 MDS_INODELOCK_LOOKUP,
2585 MF_MDC_CANCEL_FID4);
2594 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2595 void *ea, size_t ealen, struct ptlrpc_request **request)
2597 struct obd_device *obd = exp->exp_obd;
2598 struct lmv_obd *lmv = &obd->u.lmv;
2599 struct lmv_tgt_desc *tgt;
2604 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2605 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2606 op_data->op_xvalid);
2608 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2609 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2611 RETURN(PTR_ERR(tgt));
2613 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2618 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2619 struct ptlrpc_request **request)
2621 struct obd_device *obd = exp->exp_obd;
2622 struct lmv_obd *lmv = &obd->u.lmv;
2623 struct lmv_tgt_desc *tgt;
2628 tgt = lmv_fid2tgt(lmv, fid);
2630 RETURN(PTR_ERR(tgt));
2632 rc = md_fsync(tgt->ltd_exp, fid, request);
2636 struct stripe_dirent {
2637 struct page *sd_page;
2638 struct lu_dirpage *sd_dp;
2639 struct lu_dirent *sd_ent;
2643 struct lmv_dir_ctxt {
2644 struct lmv_obd *ldc_lmv;
2645 struct md_op_data *ldc_op_data;
2646 struct md_readdir_info *ldc_mrinfo;
2649 struct stripe_dirent ldc_stripes[0];
2652 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2654 if (stripe->sd_page) {
2655 kunmap(stripe->sd_page);
2656 put_page(stripe->sd_page);
2657 stripe->sd_page = NULL;
2658 stripe->sd_ent = NULL;
2662 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2666 for (i = 0; i < ctxt->ldc_count; i++)
2667 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2670 /* if @ent is dummy, or . .., get next */
2671 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2672 struct lu_dirent *ent,
2675 for (; ent; ent = lu_dirent_next(ent)) {
2676 /* Skip dummy entry */
2677 if (le16_to_cpu(ent->lde_namelen) == 0)
2680 /* skip . and .. for other stripes */
2682 (strncmp(ent->lde_name, ".",
2683 le16_to_cpu(ent->lde_namelen)) == 0 ||
2684 strncmp(ent->lde_name, "..",
2685 le16_to_cpu(ent->lde_namelen)) == 0))
2688 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2695 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2696 struct stripe_dirent *stripe,
2699 struct md_op_data *op_data = ctxt->ldc_op_data;
2700 struct lmv_oinfo *oinfo;
2701 struct lu_fid fid = op_data->op_fid1;
2702 struct inode *inode = op_data->op_data;
2703 struct lmv_tgt_desc *tgt;
2704 struct lu_dirent *ent = stripe->sd_ent;
2705 __u64 hash = ctxt->ldc_hash;
2710 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2714 if (stripe->sd_page) {
2715 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2717 /* @hash should be the last dirent hash */
2718 LASSERTF(hash <= end,
2719 "ctxt@%p stripe@%p hash %llx end %llx\n",
2720 ctxt, stripe, hash, end);
2721 /* unload last page */
2722 stripe_dirent_unload(stripe);
2724 if (end == MDS_DIR_END_OFF) {
2725 stripe->sd_eof = true;
2731 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2732 if (!oinfo->lmo_root) {
2737 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2743 /* op_data is shared by stripes, reset after use */
2744 op_data->op_fid1 = oinfo->lmo_fid;
2745 op_data->op_fid2 = oinfo->lmo_fid;
2746 op_data->op_data = oinfo->lmo_root;
2748 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_mrinfo, hash,
2751 op_data->op_fid1 = fid;
2752 op_data->op_fid2 = fid;
2753 op_data->op_data = inode;
2758 stripe->sd_dp = page_address(stripe->sd_page);
2759 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2761 /* in case a page filled with ., .. and dummy, read next */
2764 stripe->sd_ent = ent;
2767 /* treat error as eof, so dir can be partially accessed */
2768 stripe->sd_eof = true;
2769 ctxt->ldc_mrinfo->mr_partial_readdir_rc = rc;
2770 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2771 "directory is partially accessed!\n",
2772 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2779 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2781 struct obd_device *obd = exp->exp_obd;
2782 struct lmv_obd *lmv = &obd->u.lmv;
2783 struct lmv_tgt_desc *tgt;
2788 rc = lmv_check_connect(obd);
2792 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2794 RETURN(PTR_ERR(tgt));
2796 data->op_flags |= MF_MDC_CANCEL_FID1;
2797 rc = md_file_resync(tgt->ltd_exp, data);
2802 * Get dirent with the closest hash for striped directory
2804 * This function will search the dir entry, whose hash value is the
2805 * closest(>=) to hash from all of sub-stripes, and it is only being called
2806 * for striped directory.
2808 * \param[in] ctxt dir read context
2810 * \retval dirent get the entry successfully
2811 * NULL does not get the entry, normally it means
2812 * it reaches the end of the directory, while read
2813 * stripe dirent error is ignored to allow partial
2816 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2818 struct stripe_dirent *stripe;
2819 struct lu_dirent *ent = NULL;
2823 /* TODO: optimize with k-way merge sort */
2824 for (i = 0; i < ctxt->ldc_count; i++) {
2825 stripe = &ctxt->ldc_stripes[i];
2829 if (!stripe->sd_ent) {
2830 stripe_dirent_load(ctxt, stripe, i);
2831 if (!stripe->sd_ent) {
2832 LASSERT(stripe->sd_eof);
2838 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2839 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2841 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2848 stripe = &ctxt->ldc_stripes[min];
2849 ent = stripe->sd_ent;
2850 /* pop found dirent */
2851 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2859 * Build dir entry page for striped directory
2861 * This function gets one entry by @offset from a striped directory. It will
2862 * read entries from all of stripes, and choose one closest to the required
2863 * offset(&offset). A few notes
2864 * 1. skip . and .. for non-zero stripes, because there can only have one .
2865 * and .. in a directory.
2866 * 2. op_data will be shared by all of stripes, instead of allocating new
2867 * one, so need to restore before reusing.
2869 * \param[in] exp obd export refer to LMV
2870 * \param[in] op_data hold those MD parameters of read_entry
2871 * \param[in] mrinfo ldlm callback being used in enqueue in mdc_read_entry,
2872 * and partial readdir result will be stored in it.
2873 * \param[in] offset starting hash offset
2874 * \param[out] ppage the page holding the entry. Note: because the entry
2875 * will be accessed in upper layer, so we need hold the
2876 * page until the usages of entry is finished, see
2877 * ll_dir_entry_next.
2879 * retval =0 if get entry successfully
2880 * <0 cannot get entry
2882 static int lmv_striped_read_page(struct obd_export *exp,
2883 struct md_op_data *op_data,
2884 struct md_readdir_info *mrinfo, __u64 offset,
2885 struct page **ppage)
2887 struct page *page = NULL;
2888 struct lu_dirpage *dp;
2890 struct lu_dirent *ent;
2891 struct lu_dirent *last_ent;
2893 struct lmv_dir_ctxt *ctxt;
2894 struct lu_dirent *next = NULL;
2900 /* Allocate a page and read entries from all of stripes and fill
2901 * the page by hash order */
2902 page = alloc_page(GFP_KERNEL);
2906 /* Initialize the entry page */
2908 memset(dp, 0, sizeof(*dp));
2909 dp->ldp_hash_start = cpu_to_le64(offset);
2912 left_bytes = PAGE_SIZE - sizeof(*dp);
2916 /* initalize dir read context */
2917 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2918 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2920 GOTO(free_page, rc = -ENOMEM);
2921 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2922 ctxt->ldc_op_data = op_data;
2923 ctxt->ldc_mrinfo = mrinfo;
2924 ctxt->ldc_hash = offset;
2925 ctxt->ldc_count = stripe_count;
2928 next = lmv_dirent_next(ctxt);
2930 /* end of directory */
2932 ctxt->ldc_hash = MDS_DIR_END_OFF;
2935 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2937 ent_size = le16_to_cpu(next->lde_reclen);
2939 /* the last entry lde_reclen is 0, but it might not be the last
2940 * one of this temporay dir page */
2942 ent_size = lu_dirent_calc_size(
2943 le16_to_cpu(next->lde_namelen),
2944 le32_to_cpu(next->lde_attrs));
2946 if (ent_size > left_bytes)
2949 memcpy(ent, next, ent_size);
2951 /* Replace . with master FID and Replace .. with the parent FID
2952 * of master object */
2953 if (strncmp(ent->lde_name, ".",
2954 le16_to_cpu(ent->lde_namelen)) == 0 &&
2955 le16_to_cpu(ent->lde_namelen) == 1)
2956 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2957 else if (strncmp(ent->lde_name, "..",
2958 le16_to_cpu(ent->lde_namelen)) == 0 &&
2959 le16_to_cpu(ent->lde_namelen) == 2)
2960 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2962 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2963 le16_to_cpu(ent->lde_namelen), ent->lde_name,
2964 le64_to_cpu(ent->lde_hash));
2966 left_bytes -= ent_size;
2967 ent->lde_reclen = cpu_to_le16(ent_size);
2969 ent = (void *)ent + ent_size;
2972 last_ent->lde_reclen = 0;
2975 dp->ldp_flags |= LDF_EMPTY;
2976 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2977 dp->ldp_flags |= LDF_COLLIDE;
2978 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2979 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2981 put_lmv_dir_ctxt(ctxt);
2982 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2995 static int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2996 struct md_readdir_info *mrinfo, __u64 offset,
2997 struct page **ppage)
2999 struct obd_device *obd = exp->exp_obd;
3000 struct lmv_obd *lmv = &obd->u.lmv;
3001 struct lmv_tgt_desc *tgt;
3006 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
3009 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
3010 rc = lmv_striped_read_page(exp, op_data, mrinfo, offset, ppage);
3014 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
3016 RETURN(PTR_ERR(tgt));
3018 rc = md_read_page(tgt->ltd_exp, op_data, mrinfo, offset, ppage);
3024 * Unlink a file/directory
3026 * Unlink a file or directory under the parent dir. The unlink request
3027 * usually will be sent to the MDT where the child is located, but if
3028 * the client does not have the child FID then request will be sent to the
3029 * MDT where the parent is located.
3031 * If the parent is a striped directory then it also needs to locate which
3032 * stripe the name of the child is located, and replace the parent FID
3033 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
3034 * it will walk through all of sub-stripes until the child is being
3037 * \param[in] exp export refer to LMV
3038 * \param[in] op_data different parameters transferred beween client
3039 * MD stacks, name, namelen, FIDs etc.
3040 * op_fid1 is the parent FID, op_fid2 is the child
3042 * \param[out] request point to the request of unlink.
3044 * retval 0 if succeed
3045 * negative errno if failed.
3047 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
3048 struct ptlrpc_request **request)
3050 struct obd_device *obd = exp->exp_obd;
3051 struct lmv_obd *lmv = &obd->u.lmv;
3052 struct lmv_tgt_desc *tgt;
3053 struct lmv_tgt_desc *parent_tgt;
3054 struct mdt_body *body;
3059 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3060 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3061 op_data->op_cap = current_cap();
3064 parent_tgt = lmv_locate_tgt(lmv, op_data);
3065 if (IS_ERR(parent_tgt))
3066 RETURN(PTR_ERR(parent_tgt));
3068 if (likely(!fid_is_zero(&op_data->op_fid2))) {
3069 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3071 RETURN(PTR_ERR(tgt));
3077 * If child's fid is given, cancel unused locks for it if it is from
3078 * another export than parent.
3080 * LOOKUP lock for child (fid3) should also be cancelled on parent
3081 * tgt_tgt in mdc_unlink().
3083 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
3085 if (parent_tgt != tgt)
3086 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
3087 LCK_EX, MDS_INODELOCK_LOOKUP,
3088 MF_MDC_CANCEL_FID3);
3090 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
3091 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
3095 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
3096 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
3099 rc = md_unlink(tgt->ltd_exp, op_data, request);
3100 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
3101 ptlrpc_req_finished(*request);
3109 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
3113 /* Not cross-ref case, just get out of here. */
3114 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
3117 /* This is a remote object, try remote MDT. */
3118 op_data->op_fid2 = body->mbo_fid1;
3119 ptlrpc_req_finished(*request);
3122 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3124 RETURN(PTR_ERR(tgt));
3129 static int lmv_precleanup(struct obd_device *obd)
3132 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
3137 * Get by key a value associated with a LMV device.
3139 * Dispatch request to lower-layer devices as needed.
3141 * \param[in] env execution environment for this thread
3142 * \param[in] exp export for the LMV device
3143 * \param[in] keylen length of key identifier
3144 * \param[in] key identifier of key to get value for
3145 * \param[in] vallen size of \a val
3146 * \param[out] val pointer to storage location for value
3147 * \param[in] lsm optional striping metadata of object
3149 * \retval 0 on success
3150 * \retval negative negated errno on failure
3152 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
3153 __u32 keylen, void *key, __u32 *vallen, void *val)
3155 struct obd_device *obd;
3156 struct lmv_obd *lmv;
3157 struct lu_tgt_desc *tgt;
3162 obd = class_exp2obd(exp);
3164 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3165 exp->exp_handle.h_cookie);
3170 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
3171 LASSERT(*vallen == sizeof(__u32));
3172 lmv_foreach_connected_tgt(lmv, tgt) {
3173 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
3178 } else if (KEY_IS(KEY_MAX_EASIZE) ||
3179 KEY_IS(KEY_DEFAULT_EASIZE) ||
3180 KEY_IS(KEY_CONN_DATA)) {
3182 * Forwarding this request to first MDS, it should know LOV
3185 tgt = lmv_tgt(lmv, 0);
3189 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
3190 if (!rc && KEY_IS(KEY_CONN_DATA))
3191 exp->exp_connect_data = *(struct obd_connect_data *)val;
3193 } else if (KEY_IS(KEY_TGT_COUNT)) {
3194 *((int *)val) = lmv->lmv_mdt_descs.ltd_tgts_size;
3198 CDEBUG(D_IOCTL, "Invalid key\n");
3202 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
3203 int *__rcs, struct ptlrpc_request_set *_set)
3205 struct obd_device *obd = class_exp2obd(exp);
3206 struct ptlrpc_request_set *set = _set;
3207 struct lmv_obd *lmv = &obd->u.lmv;
3208 int tgt_count = lmv->lmv_mdt_count;
3209 struct lu_tgt_desc *tgt;
3210 struct fid_array *fat, **fas = NULL;
3211 int i, rc, **rcs = NULL;
3214 set = ptlrpc_prep_set();
3219 /* split FIDs by targets */
3220 OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
3222 GOTO(out, rc = -ENOMEM);
3223 OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
3225 GOTO(out_fas, rc = -ENOMEM);
3227 for (i = 0; i < fa->fa_nr; i++) {
3230 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
3232 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
3233 PFID(&fa->fa_fids[i]), rc);
3236 LASSERT(idx < tgt_count);
3238 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
3239 fa_fids[fa->fa_nr]));
3241 GOTO(out, rc = -ENOMEM);
3243 OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
3245 GOTO(out, rc = -ENOMEM);
3248 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3251 lmv_foreach_connected_tgt(lmv, tgt) {
3252 fat = fas[tgt->ltd_index];
3253 if (!fat || fat->fa_nr == 0)
3255 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3258 rc = ptlrpc_set_wait(NULL, set);
3261 for (i = 0; i < tgt_count; i++) {
3263 if (!fat || fat->fa_nr == 0)
3265 /* copy FIDs back */
3266 memcpy(fa->fa_fids + j, fat->fa_fids,
3267 fat->fa_nr * sizeof(struct lu_fid));
3269 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3274 ptlrpc_set_destroy(set);
3277 for (i = 0; i < tgt_count; i++) {
3279 OBD_FREE(fas[i], offsetof(struct fid_array,
3280 fa_fids[fa->fa_nr]));
3282 OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3285 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3288 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3294 * Asynchronously set by key a value associated with a LMV device.
3296 * Dispatch request to lower-layer devices as needed.
3298 * \param[in] env execution environment for this thread
3299 * \param[in] exp export for the LMV device
3300 * \param[in] keylen length of key identifier
3301 * \param[in] key identifier of key to store value for
3302 * \param[in] vallen size of value to store
3303 * \param[in] val pointer to data to be stored
3304 * \param[in] set optional list of related ptlrpc requests
3306 * \retval 0 on success
3307 * \retval negative negated errno on failure
3309 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3310 __u32 keylen, void *key, __u32 vallen, void *val,
3311 struct ptlrpc_request_set *set)
3313 struct lmv_tgt_desc *tgt;
3314 struct obd_device *obd;
3315 struct lmv_obd *lmv;
3319 obd = class_exp2obd(exp);
3321 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3322 exp->exp_handle.h_cookie);
3327 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3328 KEY_IS(KEY_DEFAULT_EASIZE)) {
3331 lmv_foreach_connected_tgt(lmv, tgt) {
3332 err = obd_set_info_async(env, tgt->ltd_exp,
3333 keylen, key, vallen, val, set);
3344 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3345 const struct lmv_mds_md_v1 *lmm1)
3347 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3354 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3355 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3356 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3357 if (CFS_FAIL_CHECK(OBD_FAIL_LMV_UNKNOWN_STRIPE))
3358 lsm->lsm_md_hash_type = cfs_fail_val ?: LMV_HASH_TYPE_UNKNOWN;
3360 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3361 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3362 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3363 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3364 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3365 sizeof(lsm->lsm_md_pool_name));
3367 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3370 CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3371 "layout_version %d\n", lsm->lsm_md_stripe_count,
3372 lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3373 lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3374 lsm->lsm_md_layout_version);
3376 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3377 for (i = 0; i < stripe_count; i++) {
3378 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3379 &lmm1->lmv_stripe_fids[i]);
3381 * set default value -1, so lmv_locate_tgt() knows this stripe
3382 * target is not initialized.
3384 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3385 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3388 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3389 &lsm->lsm_md_oinfo[i].lmo_mds);
3396 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3397 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3403 static inline int lmv_unpack_user_md(struct obd_export *exp,
3404 struct lmv_stripe_md *lsm,
3405 const struct lmv_user_md *lmu)
3407 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3408 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3409 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3410 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3411 lsm->lsm_md_max_inherit = lmu->lum_max_inherit;
3412 lsm->lsm_md_max_inherit_rr = lmu->lum_max_inherit_rr;
3413 lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3418 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3419 const union lmv_mds_md *lmm, size_t lmm_size)
3421 struct lmv_stripe_md *lsm;
3424 bool allocated = false;
3427 LASSERT(lsmp != NULL);
3431 if (lsm != NULL && lmm == NULL) {
3433 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3435 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3438 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3440 OBD_FREE_LARGE(lfm, lfm_size);
3444 if (lmv_dir_striped(lsm)) {
3445 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3446 iput(lsm->lsm_md_oinfo[i].lmo_root);
3447 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3449 lsm_size = lmv_stripe_md_size(0);
3451 OBD_FREE(lsm, lsm_size);
3456 /* foreign lmv case */
3457 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3458 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3461 OBD_ALLOC_LARGE(lfm, lmm_size);
3464 *lsmp = (struct lmv_stripe_md *)lfm;
3466 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3467 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3468 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3469 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3470 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3475 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3479 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3480 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3481 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3482 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3487 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3488 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3491 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3492 * stripecount should be 0 then.
3494 lsm_size = lmv_stripe_md_size(0);
3497 OBD_ALLOC(lsm, lsm_size);
3504 switch (le32_to_cpu(lmm->lmv_magic)) {
3506 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3508 case LMV_USER_MAGIC:
3509 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3512 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3513 le32_to_cpu(lmm->lmv_magic));
3518 if (rc != 0 && allocated) {
3519 OBD_FREE(lsm, lsm_size);
3526 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3528 lmv_unpackmd(NULL, &lsm, NULL, 0);
3530 EXPORT_SYMBOL(lmv_free_memmd);
3532 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3533 union ldlm_policy_data *policy,
3534 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3537 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3538 struct lu_tgt_desc *tgt;
3544 LASSERT(fid != NULL);
3546 lmv_foreach_connected_tgt(lmv, tgt) {
3547 if (!tgt->ltd_active)
3550 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3558 static int lmv_set_lock_data(struct obd_export *exp,
3559 const struct lustre_handle *lockh,
3560 void *data, __u64 *bits)
3562 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3563 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3568 if (tgt == NULL || tgt->ltd_exp == NULL)
3570 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3574 static enum ldlm_mode
3575 lmv_lock_match(struct obd_export *exp, __u64 flags,
3576 const struct lu_fid *fid, enum ldlm_type type,
3577 union ldlm_policy_data *policy,
3578 enum ldlm_mode mode, struct lustre_handle *lockh)
3580 struct obd_device *obd = exp->exp_obd;
3581 struct lmv_obd *lmv = &obd->u.lmv;
3583 struct lu_tgt_desc *tgt;
3589 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3592 * With DNE every object can have two locks in different namespaces:
3593 * lookup lock in space of MDT storing direntry and update/open lock in
3594 * space of MDT storing inode. Try the MDT that the FID maps to first,
3595 * since this can be easily found, and only try others if that fails.
3597 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3598 i < lmv->lmv_mdt_descs.ltd_tgts_size;
3599 i++, index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3601 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3602 obd->obd_name, PFID(fid), index);
3606 tgt = lmv_tgt(lmv, index);
3607 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3610 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
3620 lmv_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
3621 struct obd_export *dt_exp, struct obd_export *md_exp,
3622 struct lustre_md *md)
3624 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3625 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3627 if (!tgt || !tgt->ltd_exp)
3630 return md_get_lustre_md(tgt->ltd_exp, pill, dt_exp, md_exp, md);
3633 static int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3635 struct obd_device *obd = exp->exp_obd;
3636 struct lmv_obd *lmv = &obd->u.lmv;
3637 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3641 if (md->default_lmv) {
3642 lmv_free_memmd(md->default_lmv);
3643 md->default_lmv = NULL;
3645 if (md->lmv != NULL) {
3646 lmv_free_memmd(md->lmv);
3649 if (!tgt || !tgt->ltd_exp)
3651 RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3654 static int lmv_set_open_replay_data(struct obd_export *exp,
3655 struct obd_client_handle *och,
3656 struct lookup_intent *it)
3658 struct obd_device *obd = exp->exp_obd;
3659 struct lmv_obd *lmv = &obd->u.lmv;
3660 struct lmv_tgt_desc *tgt;
3664 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3666 RETURN(PTR_ERR(tgt));
3668 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3671 static int lmv_clear_open_replay_data(struct obd_export *exp,
3672 struct obd_client_handle *och)
3674 struct obd_device *obd = exp->exp_obd;
3675 struct lmv_obd *lmv = &obd->u.lmv;
3676 struct lmv_tgt_desc *tgt;
3680 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3682 RETURN(PTR_ERR(tgt));
3684 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3687 static int lmv_intent_getattr_async(struct obd_export *exp,
3688 struct md_enqueue_info *minfo)
3690 struct md_op_data *op_data = &minfo->mi_data;
3691 struct obd_device *obd = exp->exp_obd;
3692 struct lmv_obd *lmv = &obd->u.lmv;
3693 struct lmv_tgt_desc *ptgt;
3694 struct lmv_tgt_desc *ctgt;
3699 if (!fid_is_sane(&op_data->op_fid2))
3702 ptgt = lmv_locate_tgt(lmv, op_data);
3704 RETURN(PTR_ERR(ptgt));
3706 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3708 RETURN(PTR_ERR(ctgt));
3711 * remote object needs two RPCs to lookup and getattr, considering the
3712 * complexity don't support statahead for now.
3717 rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3722 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3723 struct lu_fid *fid, __u64 *bits)
3725 struct obd_device *obd = exp->exp_obd;
3726 struct lmv_obd *lmv = &obd->u.lmv;
3727 struct lmv_tgt_desc *tgt;
3732 tgt = lmv_fid2tgt(lmv, fid);
3734 RETURN(PTR_ERR(tgt));
3736 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3740 static int lmv_get_fid_from_lsm(struct obd_export *exp,
3741 const struct lmv_stripe_md *lsm,
3742 const char *name, int namelen,
3745 const struct lmv_oinfo *oinfo;
3747 if (!lmv_dir_striped(lsm))
3750 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3752 return PTR_ERR(oinfo);
3754 *fid = oinfo->lmo_fid;
3760 * For lmv, only need to send request to master MDT, and the master MDT will
3761 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3762 * we directly fetch data from the slave MDTs.
3764 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3765 struct obd_quotactl *oqctl)
3767 struct obd_device *obd = class_exp2obd(exp);
3768 struct lmv_obd *lmv = &obd->u.lmv;
3769 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3770 __u64 curspace, curinodes;
3775 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3776 CERROR("master lmv inactive\n");
3780 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3781 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3785 curspace = curinodes = 0;
3786 lmv_foreach_connected_tgt(lmv, tgt) {
3789 if (!tgt->ltd_active)
3792 err = obd_quotactl(tgt->ltd_exp, oqctl);
3794 CERROR("getquota on mdt %d failed. %d\n",
3795 tgt->ltd_index, err);
3799 curspace += oqctl->qc_dqblk.dqb_curspace;
3800 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3803 oqctl->qc_dqblk.dqb_curspace = curspace;
3804 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3809 static int lmv_merge_attr(struct obd_export *exp,
3810 const struct lmv_stripe_md *lsm,
3811 struct cl_attr *attr,
3812 ldlm_blocking_callback cb_blocking)
3817 if (!lmv_dir_striped(lsm))
3820 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3824 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3825 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3831 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3832 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3833 i_size_read(inode), (unsigned long long)inode->i_blocks,
3834 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3835 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3837 /* for slave stripe, it needs to subtract nlink for . and .. */
3839 attr->cat_nlink += inode->i_nlink - 2;
3841 attr->cat_nlink = inode->i_nlink;
3843 attr->cat_size += i_size_read(inode);
3844 attr->cat_blocks += inode->i_blocks;
3846 if (attr->cat_atime < inode->i_atime.tv_sec)
3847 attr->cat_atime = inode->i_atime.tv_sec;
3849 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3850 attr->cat_ctime = inode->i_ctime.tv_sec;
3852 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3853 attr->cat_mtime = inode->i_mtime.tv_sec;
3858 static const struct obd_ops lmv_obd_ops = {
3859 .o_owner = THIS_MODULE,
3860 .o_setup = lmv_setup,
3861 .o_cleanup = lmv_cleanup,
3862 .o_precleanup = lmv_precleanup,
3863 .o_process_config = lmv_process_config,
3864 .o_connect = lmv_connect,
3865 .o_disconnect = lmv_disconnect,
3866 .o_statfs = lmv_statfs,
3867 .o_get_info = lmv_get_info,
3868 .o_set_info_async = lmv_set_info_async,
3869 .o_notify = lmv_notify,
3870 .o_get_uuid = lmv_get_uuid,
3871 .o_fid_alloc = lmv_fid_alloc,
3872 .o_iocontrol = lmv_iocontrol,
3873 .o_quotactl = lmv_quotactl
3876 static const struct md_ops lmv_md_ops = {
3877 .m_get_root = lmv_get_root,
3878 .m_null_inode = lmv_null_inode,
3879 .m_close = lmv_close,
3880 .m_create = lmv_create,
3881 .m_enqueue = lmv_enqueue,
3882 .m_getattr = lmv_getattr,
3883 .m_getxattr = lmv_getxattr,
3884 .m_getattr_name = lmv_getattr_name,
3885 .m_intent_lock = lmv_intent_lock,
3887 .m_rename = lmv_rename,
3888 .m_setattr = lmv_setattr,
3889 .m_setxattr = lmv_setxattr,
3890 .m_fsync = lmv_fsync,
3891 .m_file_resync = lmv_file_resync,
3892 .m_read_page = lmv_read_page,
3893 .m_unlink = lmv_unlink,
3894 .m_init_ea_size = lmv_init_ea_size,
3895 .m_cancel_unused = lmv_cancel_unused,
3896 .m_set_lock_data = lmv_set_lock_data,
3897 .m_lock_match = lmv_lock_match,
3898 .m_get_lustre_md = lmv_get_lustre_md,
3899 .m_free_lustre_md = lmv_free_lustre_md,
3900 .m_merge_attr = lmv_merge_attr,
3901 .m_set_open_replay_data = lmv_set_open_replay_data,
3902 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3903 .m_intent_getattr_async = lmv_intent_getattr_async,
3904 .m_revalidate_lock = lmv_revalidate_lock,
3905 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3906 .m_unpackmd = lmv_unpackmd,
3907 .m_rmfid = lmv_rmfid,
3910 static int __init lmv_init(void)
3912 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true,
3913 LUSTRE_LMV_NAME, NULL);
3916 static void __exit lmv_exit(void)
3918 class_unregister_type(LUSTRE_LMV_NAME);
3921 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3922 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3923 MODULE_VERSION(LUSTRE_VERSION_STRING);
3924 MODULE_LICENSE("GPL");
3926 module_init(lmv_init);
3927 module_exit(lmv_exit);