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/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_LMV
35 #include <linux/file.h>
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/user_namespace.h>
39 #include <linux/uidgid.h>
40 #include <linux/slab.h>
41 #include <linux/pagemap.h>
43 #include <linux/math64.h>
44 #include <linux/seq_file.h>
45 #include <linux/namei.h>
47 #include <obd_support.h>
48 #include <lustre_lib.h>
49 #include <lustre_net.h>
50 #include <obd_class.h>
51 #include <lustre_lmv.h>
52 #include <lprocfs_status.h>
53 #include <cl_object.h>
54 #include <lustre_fid.h>
55 #include <uapi/linux/lustre/lustre_ioctl.h>
56 #include <lustre_kernelcomm.h>
57 #include "lmv_internal.h"
59 static int lmv_check_connect(struct obd_device *obd);
61 void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt,
64 if (tgt->ltd_active == activate)
67 tgt->ltd_active = activate;
68 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count +=
71 tgt->ltd_exp->exp_obd->obd_inactive = !activate;
77 * -EINVAL : UUID can't be found in the LMV's target list
78 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
79 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
81 static int lmv_set_mdc_active(struct lmv_obd *lmv,
82 const struct obd_uuid *uuid,
85 struct lu_tgt_desc *tgt = NULL;
86 struct obd_device *obd;
91 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
92 lmv, uuid->uuid, activate);
94 spin_lock(&lmv->lmv_lock);
95 lmv_foreach_connected_tgt(lmv, tgt) {
96 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n",
97 tgt->ltd_index, tgt->ltd_uuid.uuid,
98 tgt->ltd_exp->exp_handle.h_cookie);
100 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
105 GOTO(out_lmv_lock, rc = -EINVAL);
107 obd = class_exp2obd(tgt->ltd_exp);
109 GOTO(out_lmv_lock, rc = -ENOTCONN);
111 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
112 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
113 obd->obd_type->typ_name, tgt->ltd_index);
114 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
116 if (tgt->ltd_active == activate) {
117 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
118 activate ? "" : "in");
119 GOTO(out_lmv_lock, rc);
122 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
123 activate ? "" : "in");
124 lmv_activate_target(lmv, tgt, activate);
128 spin_unlock(&lmv->lmv_lock);
132 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
134 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
135 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
137 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
140 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
141 enum obd_notify_event ev)
143 struct obd_connect_data *conn_data;
144 struct lmv_obd *lmv = &obd->u.lmv;
145 struct obd_uuid *uuid;
149 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
150 CERROR("unexpected notification of %s %s!\n",
151 watched->obd_type->typ_name,
156 uuid = &watched->u.cli.cl_target_uuid;
157 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
159 * Set MDC as active before notifying the observer, so the
160 * observer can use the MDC normally.
162 rc = lmv_set_mdc_active(lmv, uuid,
163 ev == OBD_NOTIFY_ACTIVE);
165 CERROR("%sactivation of %s failed: %d\n",
166 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
170 } else if (ev == OBD_NOTIFY_OCD) {
171 conn_data = &watched->u.cli.cl_import->imp_connect_data;
173 * XXX: Make sure that ocd_connect_flags from all targets are
174 * the same. Otherwise one of MDTs runs wrong version or
175 * something like this. --umka
177 obd->obd_self_export->exp_connect_data = *conn_data;
181 * Pass the notification up the chain.
183 if (obd->obd_observer)
184 rc = obd_notify(obd->obd_observer, watched, ev);
189 static int lmv_connect(const struct lu_env *env,
190 struct obd_export **pexp, struct obd_device *obd,
191 struct obd_uuid *cluuid, struct obd_connect_data *data,
194 struct lmv_obd *lmv = &obd->u.lmv;
195 struct lustre_handle conn = { 0 };
196 struct obd_export *exp;
200 rc = class_connect(&conn, obd, cluuid);
202 CERROR("class_connection() returned %d\n", rc);
206 exp = class_conn2export(&conn);
209 lmv->conn_data = *data;
210 lmv->lmv_cache = localdata;
212 lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
213 &obd->obd_kset.kobj);
214 if (!lmv->lmv_tgts_kobj) {
215 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
216 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
219 rc = lmv_check_connect(obd);
228 if (lmv->lmv_tgts_kobj)
229 kobject_put(lmv->lmv_tgts_kobj);
231 class_disconnect(exp);
236 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
239 struct obd_device *obd = exp->exp_obd;
240 struct lmv_obd *lmv = &obd->u.lmv;
241 struct lmv_tgt_desc *tgt;
247 if (lmv->max_easize < easize) {
248 lmv->max_easize = easize;
251 if (lmv->max_def_easize < def_easize) {
252 lmv->max_def_easize = def_easize;
259 if (lmv->connected == 0)
262 lmv_foreach_connected_tgt(lmv, tgt) {
263 if (!tgt->ltd_active)
266 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
268 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
269 " rc = %d\n", obd->obd_name, tgt->ltd_index, rc);
276 #define MAX_STRING_SIZE 128
278 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
280 struct lmv_obd *lmv = &obd->u.lmv;
281 struct obd_device *mdc_obd;
282 struct obd_export *mdc_exp;
283 struct lu_fld_target target;
287 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
290 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
294 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
295 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
296 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
298 if (!mdc_obd->obd_set_up) {
299 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
303 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
304 &lmv->conn_data, lmv->lmv_cache);
306 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
311 * Init fid sequence client for this mdc and add new fld target.
313 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
317 target.ft_srv = NULL;
318 target.ft_exp = mdc_exp;
319 target.ft_idx = tgt->ltd_index;
321 fld_client_add_target(&lmv->lmv_fld, &target);
323 rc = obd_register_observer(mdc_obd, obd);
325 obd_disconnect(mdc_exp);
326 CERROR("target %s register_observer error %d\n",
327 tgt->ltd_uuid.uuid, rc);
331 if (obd->obd_observer) {
333 * Tell the observer about the new target.
335 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
338 obd_disconnect(mdc_exp);
344 tgt->ltd_exp = mdc_exp;
345 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count++;
347 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
349 rc = lu_qos_add_tgt(&lmv->lmv_qos, tgt);
351 obd_disconnect(mdc_exp);
355 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
356 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
357 atomic_read(&obd->obd_refcount));
359 lmv_statfs_check_update(obd, tgt);
361 if (lmv->lmv_tgts_kobj)
362 /* Even if we failed to create the link, that's fine */
363 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
364 &mdc_obd->obd_kset.kobj,
369 static void lmv_del_target(struct lmv_obd *lmv, struct lu_tgt_desc *tgt)
372 ltd_del_tgt(&lmv->lmv_mdt_descs, tgt);
376 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
377 __u32 index, int gen)
379 struct obd_device *mdc_obd;
380 struct lmv_obd *lmv = &obd->u.lmv;
381 struct lmv_tgt_desc *tgt;
382 struct lu_tgt_descs *ltd = &lmv->lmv_mdt_descs;
387 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
388 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
391 CERROR("%s: Target %s not attached: rc = %d\n",
392 obd->obd_name, uuidp->uuid, -EINVAL);
400 mutex_init(&tgt->ltd_fid_mutex);
401 tgt->ltd_index = index;
402 tgt->ltd_uuid = *uuidp;
405 mutex_lock(<d->ltd_mutex);
406 rc = ltd_add_tgt(ltd, tgt);
407 mutex_unlock(<d->ltd_mutex);
413 /* lmv_check_connect() will connect this target. */
416 rc = lmv_connect_mdc(obd, tgt);
418 int easize = sizeof(struct lmv_stripe_md) +
419 lmv->lmv_mdt_count * sizeof(struct lu_fid);
421 lmv_init_ea_size(obd->obd_self_export, easize, 0);
431 static int lmv_check_connect(struct obd_device *obd)
433 struct lmv_obd *lmv = &obd->u.lmv;
434 struct lmv_tgt_desc *tgt;
443 mutex_lock(&lmv->lmv_mdt_descs.ltd_mutex);
445 GOTO(unlock, rc = 0);
447 if (!lmv->lmv_mdt_count) {
448 CERROR("%s: no targets configured: rc = -EINVAL\n",
450 GOTO(unlock, rc = -EINVAL);
453 if (!lmv_mdt0_inited(lmv)) {
454 CERROR("%s: no target configured for index 0: rc = -EINVAL.\n",
456 GOTO(unlock, rc = -EINVAL);
459 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
460 obd->obd_uuid.uuid, obd->obd_name);
462 lmv_foreach_tgt(lmv, tgt) {
463 rc = lmv_connect_mdc(obd, tgt);
469 easize = lmv_mds_md_size(lmv->lmv_mdt_count, LMV_MAGIC);
470 lmv_init_ea_size(obd->obd_self_export, easize, 0);
473 mutex_unlock(&lmv->lmv_mdt_descs.ltd_mutex);
478 lmv_foreach_tgt(lmv, tgt) {
483 --lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count;
484 obd_disconnect(tgt->ltd_exp);
490 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
492 struct lmv_obd *lmv = &obd->u.lmv;
493 struct obd_device *mdc_obd;
497 LASSERT(tgt != NULL);
498 LASSERT(obd != NULL);
500 mdc_obd = class_exp2obd(tgt->ltd_exp);
503 mdc_obd->obd_force = obd->obd_force;
504 mdc_obd->obd_fail = obd->obd_fail;
505 mdc_obd->obd_no_recov = obd->obd_no_recov;
507 if (lmv->lmv_tgts_kobj)
508 sysfs_remove_link(lmv->lmv_tgts_kobj,
512 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
514 CERROR("Can't finanize fids factory\n");
516 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
517 tgt->ltd_exp->exp_obd->obd_name,
518 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
520 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
521 rc = obd_disconnect(tgt->ltd_exp);
523 if (tgt->ltd_active) {
524 CERROR("Target %s disconnect error %d\n",
525 tgt->ltd_uuid.uuid, rc);
529 lmv_activate_target(lmv, tgt, 0);
534 static int lmv_disconnect(struct obd_export *exp)
536 struct obd_device *obd = class_exp2obd(exp);
537 struct lmv_obd *lmv = &obd->u.lmv;
538 struct lmv_tgt_desc *tgt;
543 lmv_foreach_connected_tgt(lmv, tgt)
544 lmv_disconnect_mdc(obd, tgt);
546 if (lmv->lmv_tgts_kobj)
547 kobject_put(lmv->lmv_tgts_kobj);
550 class_export_put(exp);
551 rc = class_disconnect(exp);
557 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
560 struct obd_device *obd = class_exp2obd(exp);
561 struct lmv_obd *lmv = &obd->u.lmv;
562 struct getinfo_fid2path *gf;
563 struct lmv_tgt_desc *tgt;
564 struct getinfo_fid2path *remote_gf = NULL;
565 struct lu_fid root_fid;
566 int remote_gf_size = 0;
570 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
572 RETURN(PTR_ERR(tgt));
574 root_fid = *gf->gf_u.gf_root_fid;
575 LASSERT(fid_is_sane(&root_fid));
578 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
579 if (rc != 0 && rc != -EREMOTE)
580 GOTO(out_fid2path, rc);
582 /* If remote_gf != NULL, it means just building the
583 * path on the remote MDT, copy this path segement to gf */
584 if (remote_gf != NULL) {
585 struct getinfo_fid2path *ori_gf;
589 ori_gf = (struct getinfo_fid2path *)karg;
590 if (strlen(ori_gf->gf_u.gf_path) + 1 +
591 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
592 GOTO(out_fid2path, rc = -EOVERFLOW);
594 ptr = ori_gf->gf_u.gf_path;
596 len = strlen(gf->gf_u.gf_path);
597 /* move the current path to the right to release space
598 * for closer-to-root part */
599 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
600 memcpy(ptr, gf->gf_u.gf_path, len);
604 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
605 tgt->ltd_exp->exp_obd->obd_name,
606 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
610 GOTO(out_fid2path, rc);
612 /* sigh, has to go to another MDT to do path building further */
613 if (remote_gf == NULL) {
614 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
615 OBD_ALLOC(remote_gf, remote_gf_size);
616 if (remote_gf == NULL)
617 GOTO(out_fid2path, rc = -ENOMEM);
618 remote_gf->gf_pathlen = PATH_MAX;
621 if (!fid_is_sane(&gf->gf_fid)) {
622 CERROR("%s: invalid FID "DFID": rc = %d\n",
623 tgt->ltd_exp->exp_obd->obd_name,
624 PFID(&gf->gf_fid), -EINVAL);
625 GOTO(out_fid2path, rc = -EINVAL);
628 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
630 GOTO(out_fid2path, rc = -EINVAL);
632 remote_gf->gf_fid = gf->gf_fid;
633 remote_gf->gf_recno = -1;
634 remote_gf->gf_linkno = -1;
635 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
636 *remote_gf->gf_u.gf_root_fid = root_fid;
638 goto repeat_fid2path;
641 if (remote_gf != NULL)
642 OBD_FREE(remote_gf, remote_gf_size);
646 static int lmv_hsm_req_count(struct lmv_obd *lmv,
647 const struct hsm_user_request *hur,
648 const struct lmv_tgt_desc *tgt_mds)
650 struct lmv_tgt_desc *curr_tgt;
654 /* count how many requests must be sent to the given target */
655 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
656 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
657 if (IS_ERR(curr_tgt))
658 RETURN(PTR_ERR(curr_tgt));
659 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
665 static int lmv_hsm_req_build(struct lmv_obd *lmv,
666 struct hsm_user_request *hur_in,
667 const struct lmv_tgt_desc *tgt_mds,
668 struct hsm_user_request *hur_out)
671 struct lmv_tgt_desc *curr_tgt;
673 /* build the hsm_user_request for the given target */
674 hur_out->hur_request = hur_in->hur_request;
676 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
677 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
678 if (IS_ERR(curr_tgt))
679 RETURN(PTR_ERR(curr_tgt));
680 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
681 hur_out->hur_user_item[nr_out] =
682 hur_in->hur_user_item[i];
686 hur_out->hur_request.hr_itemcount = nr_out;
687 memcpy(hur_data(hur_out), hur_data(hur_in),
688 hur_in->hur_request.hr_data_len);
693 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
694 int len, struct lustre_kernelcomm *lk,
697 struct lmv_obd *lmv = &obd->u.lmv;
698 struct lu_tgt_desc *tgt;
703 /* unregister request (call from llapi_hsm_copytool_fini) */
704 lmv_foreach_connected_tgt(lmv, tgt)
705 /* best effort: try to clean as much as possible
706 * (continue on error) */
707 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
709 /* Whatever the result, remove copytool from kuc groups.
710 * Unreached coordinators will get EPIPE on next requests
711 * and will unregister automatically.
713 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
718 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
719 int len, struct lustre_kernelcomm *lk,
722 struct lmv_obd *lmv = &obd->u.lmv;
724 bool any_set = false;
725 struct kkuc_ct_data *kcd;
727 struct lu_tgt_desc *tgt;
734 filp = fget(lk->lk_wfd);
738 if (lk->lk_flags & LK_FLG_DATANR)
739 kcd_size = offsetof(struct kkuc_ct_data,
740 kcd_archives[lk->lk_data_count]);
742 kcd_size = sizeof(*kcd);
744 OBD_ALLOC(kcd, kcd_size);
746 GOTO(err_fput, rc = -ENOMEM);
748 kcd->kcd_nr_archives = lk->lk_data_count;
749 if (lk->lk_flags & LK_FLG_DATANR) {
750 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
751 if (lk->lk_data_count > 0)
752 memcpy(kcd->kcd_archives, lk->lk_data,
753 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
755 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
758 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
759 lk->lk_group, kcd, kcd_size);
760 OBD_FREE(kcd, kcd_size);
764 /* All or nothing: try to register to all MDS.
765 * In case of failure, unregister from previous MDS,
766 * except if it because of inactive target. */
767 lmv_foreach_connected_tgt(lmv, tgt) {
768 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
770 if (tgt->ltd_active) {
771 /* permanent error */
772 CERROR("%s: iocontrol MDC %s on MDT"
773 " idx %d cmd %x: err = %d\n",
774 lmv2obd_dev(lmv)->obd_name,
775 tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
778 lk->lk_flags |= LK_FLG_STOP;
780 /* unregister from previous MDS */
781 lmv_foreach_connected_tgt(lmv, tgt) {
782 if (tgt->ltd_index >= i)
785 obd_iocontrol(cmd, tgt->ltd_exp, len,
788 GOTO(err_kkuc_rem, rc);
790 /* else: transient error.
791 * kuc will register to the missing MDT
799 /* no registration done: return error */
800 GOTO(err_kkuc_rem, rc = -ENOTCONN);
805 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
812 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
813 int len, void *karg, void __user *uarg)
815 struct obd_device *obd = class_exp2obd(exp);
816 struct lmv_obd *lmv = &obd->u.lmv;
817 struct lu_tgt_desc *tgt = NULL;
819 __u32 count = lmv->lmv_mdt_count;
828 case IOC_OBD_STATFS: {
829 struct obd_ioctl_data *data = karg;
830 struct obd_device *mdc_obd;
831 struct obd_statfs stat_buf = {0};
834 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
836 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
839 tgt = lmv_tgt(lmv, index);
843 if (!tgt->ltd_active)
846 mdc_obd = class_exp2obd(tgt->ltd_exp);
851 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
852 min((int) data->ioc_plen2,
853 (int) sizeof(struct obd_uuid))))
856 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
857 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
861 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
862 min((int) data->ioc_plen1,
863 (int) sizeof(stat_buf))))
867 case OBD_IOC_QUOTACTL: {
868 struct if_quotactl *qctl = karg;
869 struct obd_quotactl *oqctl;
871 if (qctl->qc_valid == QC_MDTIDX) {
872 tgt = lmv_tgt(lmv, qctl->qc_idx);
873 } else if (qctl->qc_valid == QC_UUID) {
874 lmv_foreach_tgt(lmv, tgt) {
875 if (!obd_uuid_equals(&tgt->ltd_uuid,
888 if (!tgt || !tgt->ltd_exp)
891 OBD_ALLOC_PTR(oqctl);
895 QCTL_COPY(oqctl, qctl);
896 rc = obd_quotactl(tgt->ltd_exp, oqctl);
898 QCTL_COPY(qctl, oqctl);
899 qctl->qc_valid = QC_MDTIDX;
900 qctl->obd_uuid = tgt->ltd_uuid;
905 case LL_IOC_GET_CONNECT_FLAGS: {
906 tgt = lmv_tgt(lmv, 0);
908 if (tgt && tgt->ltd_exp)
909 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
912 case LL_IOC_FID2MDTIDX: {
913 struct lu_fid *fid = karg;
916 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
920 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
921 * point to user space memory for FID2MDTIDX. */
922 *(__u32 *)uarg = mdt_index;
925 case OBD_IOC_FID2PATH: {
926 rc = lmv_fid2path(exp, len, karg, uarg);
929 case LL_IOC_HSM_STATE_GET:
930 case LL_IOC_HSM_STATE_SET:
931 case LL_IOC_HSM_ACTION: {
932 struct md_op_data *op_data = karg;
934 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
936 RETURN(PTR_ERR(tgt));
938 if (tgt->ltd_exp == NULL)
941 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
944 case LL_IOC_HSM_PROGRESS: {
945 const struct hsm_progress_kernel *hpk = karg;
947 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
949 RETURN(PTR_ERR(tgt));
950 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
953 case LL_IOC_HSM_REQUEST: {
954 struct hsm_user_request *hur = karg;
955 unsigned int reqcount = hur->hur_request.hr_itemcount;
960 /* if the request is about a single fid
961 * or if there is a single MDS, no need to split
963 if (reqcount == 1 || count == 1) {
964 tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
966 RETURN(PTR_ERR(tgt));
967 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
969 /* split fid list to their respective MDS */
970 lmv_foreach_connected_tgt(lmv, tgt) {
973 struct hsm_user_request *req;
975 nr = lmv_hsm_req_count(lmv, hur, tgt);
978 if (nr == 0) /* nothing for this MDS */
981 /* build a request with fids for this MDS */
982 reqlen = offsetof(typeof(*hur),
984 + hur->hur_request.hr_data_len;
985 OBD_ALLOC_LARGE(req, reqlen);
988 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
990 GOTO(hsm_req_err, rc1);
991 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
994 if (rc1 != 0 && rc == 0)
996 OBD_FREE_LARGE(req, reqlen);
1001 case LL_IOC_LOV_SWAP_LAYOUTS: {
1002 struct md_op_data *op_data = karg;
1003 struct lmv_tgt_desc *tgt1, *tgt2;
1005 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1007 RETURN(PTR_ERR(tgt1));
1009 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1011 RETURN(PTR_ERR(tgt2));
1013 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1016 /* only files on same MDT can have their layouts swapped */
1017 if (tgt1->ltd_index != tgt2->ltd_index)
1020 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1023 case LL_IOC_HSM_CT_START: {
1024 struct lustre_kernelcomm *lk = karg;
1025 if (lk->lk_flags & LK_FLG_STOP)
1026 rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1028 rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1032 lmv_foreach_connected_tgt(lmv, tgt) {
1033 struct obd_device *mdc_obd;
1036 /* ll_umount_begin() sets force flag but for lmv, not
1037 * mdc. Let's pass it through */
1038 mdc_obd = class_exp2obd(tgt->ltd_exp);
1039 mdc_obd->obd_force = obd->obd_force;
1040 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1042 if (tgt->ltd_active) {
1043 CERROR("error: iocontrol MDC %s on MDT"
1044 " idx %d cmd %x: err = %d\n",
1046 tgt->ltd_index, cmd, err);
1059 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1060 struct lu_fid *fid, struct md_op_data *op_data)
1062 struct obd_device *obd = class_exp2obd(exp);
1063 struct lmv_obd *lmv = &obd->u.lmv;
1064 struct lmv_tgt_desc *tgt;
1072 tgt = lmv_tgt(lmv, op_data->op_mds);
1076 if (!tgt->ltd_active || !tgt->ltd_exp)
1080 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1081 * on server that seq in new allocated fid is not yet known.
1083 mutex_lock(&tgt->ltd_fid_mutex);
1084 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1085 mutex_unlock(&tgt->ltd_fid_mutex);
1087 LASSERT(fid_is_sane(fid));
1094 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1096 struct lmv_obd *lmv = &obd->u.lmv;
1097 struct lmv_desc *desc;
1098 struct lnet_process_id lnet_id;
1104 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1105 CERROR("LMV setup requires a descriptor\n");
1109 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1110 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1111 CERROR("Lmv descriptor size wrong: %d > %d\n",
1112 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1116 obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1117 desc->ld_uuid.uuid);
1118 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1119 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1120 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1121 LMV_DESC_QOS_MAXAGE_DEFAULT;
1122 lmv->max_def_easize = 0;
1123 lmv->max_easize = 0;
1125 spin_lock_init(&lmv->lmv_lock);
1128 * initialize rr_index to lower 32bit of netid, so that client
1129 * can distribute subdirs evenly from the beginning.
1131 while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1132 if (lnet_id.nid != LNET_NID_LO_0) {
1133 lmv->lmv_qos_rr_index = (u32)lnet_id.nid;
1138 rc = lmv_tunables_init(obd);
1140 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1143 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1144 LUSTRE_CLI_FLD_HASH_DHT);
1146 CERROR("Can't init FLD, err %d\n", rc);
1148 rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1150 CWARN("%s: error initialize target table: rc = %d\n",
1156 static int lmv_cleanup(struct obd_device *obd)
1158 struct lmv_obd *lmv = &obd->u.lmv;
1159 struct lu_tgt_desc *tgt;
1160 struct lu_tgt_desc *tmp;
1164 fld_client_fini(&lmv->lmv_fld);
1165 lmv_foreach_tgt_safe(lmv, tgt, tmp)
1166 lmv_del_target(lmv, tgt);
1167 lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1172 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1174 struct lustre_cfg *lcfg = buf;
1175 struct obd_uuid obd_uuid;
1181 switch (lcfg->lcfg_command) {
1183 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1184 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1185 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1186 GOTO(out, rc = -EINVAL);
1188 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1190 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1191 GOTO(out, rc = -EINVAL);
1192 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1193 GOTO(out, rc = -EINVAL);
1194 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1197 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1198 GOTO(out, rc = -EINVAL);
1204 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1208 if (flags & OBD_STATFS_FOR_MDT0)
1211 if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1212 return lmv->lmv_statfs_start;
1214 /* choose initial MDT for this client */
1216 struct lnet_process_id lnet_id;
1217 if (LNetGetId(i, &lnet_id) == -ENOENT)
1220 if (lnet_id.nid != LNET_NID_LO_0) {
1221 /* We dont need a full 64-bit modulus, just enough
1222 * to distribute the requests across MDTs evenly.
1224 lmv->lmv_statfs_start = (u32)lnet_id.nid %
1230 return lmv->lmv_statfs_start;
1233 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1234 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1236 struct obd_device *obd = class_exp2obd(exp);
1237 struct lmv_obd *lmv = &obd->u.lmv;
1238 struct obd_statfs *temp;
1239 struct lu_tgt_desc *tgt;
1246 OBD_ALLOC(temp, sizeof(*temp));
1250 /* distribute statfs among MDTs */
1251 idx = lmv_select_statfs_mdt(lmv, flags);
1253 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1254 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1255 tgt = lmv_tgt(lmv, idx);
1256 if (!tgt || !tgt->ltd_exp)
1259 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age,
1260 flags | OBD_STATFS_NESTED);
1262 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1263 tgt->ltd_exp->exp_obd->obd_name, i, rc);
1264 GOTO(out_free_temp, rc);
1267 if (temp->os_state & OS_STATFS_SUM ||
1268 flags == OBD_STATFS_FOR_MDT0) {
1269 /* reset to the last aggregated values
1270 * and don't sum with non-aggrated data */
1271 /* If the statfs is from mount, it needs to retrieve
1272 * necessary information from MDT0. i.e. mount does
1273 * not need the merged osfs from all of MDT. Also
1274 * clients can be mounted as long as MDT0 is in
1283 osfs->os_bavail += temp->os_bavail;
1284 osfs->os_blocks += temp->os_blocks;
1285 osfs->os_ffree += temp->os_ffree;
1286 osfs->os_files += temp->os_files;
1287 osfs->os_granted += temp->os_granted;
1293 OBD_FREE(temp, sizeof(*temp));
1297 static int lmv_statfs_update(void *cookie, int rc)
1299 struct obd_info *oinfo = cookie;
1300 struct obd_device *obd = oinfo->oi_obd;
1301 struct lmv_obd *lmv = &obd->u.lmv;
1302 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1303 struct obd_statfs *osfs = oinfo->oi_osfs;
1306 * NB: don't deactivate TGT upon error, because we may not trigger async
1307 * statfs any longer, then there is no chance to activate TGT.
1310 spin_lock(&lmv->lmv_lock);
1311 tgt->ltd_statfs = *osfs;
1312 tgt->ltd_statfs_age = ktime_get_seconds();
1313 spin_unlock(&lmv->lmv_lock);
1314 lmv->lmv_qos.lq_dirty = 1;
1320 /* update tgt statfs async if it's ld_qos_maxage old */
1321 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1323 struct obd_info oinfo = {
1326 .oi_cb_up = lmv_statfs_update,
1330 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1331 obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1334 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1339 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1342 struct obd_device *obd = exp->exp_obd;
1343 struct lmv_obd *lmv = &obd->u.lmv;
1344 struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1352 rc = md_get_root(tgt->ltd_exp, fileset, fid);
1356 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1357 u64 obd_md_valid, const char *name, size_t buf_size,
1358 struct ptlrpc_request **req)
1360 struct obd_device *obd = exp->exp_obd;
1361 struct lmv_obd *lmv = &obd->u.lmv;
1362 struct lmv_tgt_desc *tgt;
1367 tgt = lmv_fid2tgt(lmv, fid);
1369 RETURN(PTR_ERR(tgt));
1371 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1376 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1377 u64 obd_md_valid, const char *name,
1378 const void *value, size_t value_size,
1379 unsigned int xattr_flags, u32 suppgid,
1380 struct ptlrpc_request **req)
1382 struct obd_device *obd = exp->exp_obd;
1383 struct lmv_obd *lmv = &obd->u.lmv;
1384 struct lmv_tgt_desc *tgt;
1389 tgt = lmv_fid2tgt(lmv, fid);
1391 RETURN(PTR_ERR(tgt));
1393 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1394 value, value_size, xattr_flags, suppgid, req);
1399 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1400 struct ptlrpc_request **request)
1402 struct obd_device *obd = exp->exp_obd;
1403 struct lmv_obd *lmv = &obd->u.lmv;
1404 struct lmv_tgt_desc *tgt;
1409 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1411 RETURN(PTR_ERR(tgt));
1413 if (op_data->op_flags & MF_GET_MDT_IDX) {
1414 op_data->op_mds = tgt->ltd_index;
1418 rc = md_getattr(tgt->ltd_exp, op_data, request);
1423 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1425 struct obd_device *obd = exp->exp_obd;
1426 struct lmv_obd *lmv = &obd->u.lmv;
1427 struct lu_tgt_desc *tgt;
1431 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1434 * With DNE every object can have two locks in different namespaces:
1435 * lookup lock in space of MDT storing direntry and update/open lock in
1436 * space of MDT storing inode.
1438 lmv_foreach_connected_tgt(lmv, tgt)
1439 md_null_inode(tgt->ltd_exp, fid);
1444 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1445 struct md_open_data *mod, struct ptlrpc_request **request)
1447 struct obd_device *obd = exp->exp_obd;
1448 struct lmv_obd *lmv = &obd->u.lmv;
1449 struct lmv_tgt_desc *tgt;
1454 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1456 RETURN(PTR_ERR(tgt));
1458 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1459 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1463 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv, __u32 *mdt)
1465 struct lu_tgt_desc *tgt;
1466 __u64 total_weight = 0;
1467 __u64 cur_weight = 0;
1473 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1474 RETURN(ERR_PTR(-EAGAIN));
1476 down_write(&lmv->lmv_qos.lq_rw_sem);
1478 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1479 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1481 rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1483 GOTO(unlock, tgt = ERR_PTR(rc));
1485 lmv_foreach_tgt(lmv, tgt) {
1486 tgt->ltd_qos.ltq_usable = 0;
1487 if (!tgt->ltd_exp || !tgt->ltd_active)
1490 tgt->ltd_qos.ltq_usable = 1;
1491 lu_tgt_qos_weight_calc(tgt);
1492 total_weight += tgt->ltd_qos.ltq_weight;
1495 rand = lu_prandom_u64_max(total_weight);
1497 lmv_foreach_connected_tgt(lmv, tgt) {
1498 if (!tgt->ltd_qos.ltq_usable)
1501 cur_weight += tgt->ltd_qos.ltq_weight;
1502 if (cur_weight < rand)
1505 *mdt = tgt->ltd_index;
1506 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1507 GOTO(unlock, rc = 0);
1510 /* no proper target found */
1511 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1513 up_write(&lmv->lmv_qos.lq_rw_sem);
1518 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv, __u32 *mdt)
1520 struct lu_tgt_desc *tgt;
1526 spin_lock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1527 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1528 index = (i + lmv->lmv_qos_rr_index) %
1529 lmv->lmv_mdt_descs.ltd_tgts_size;
1530 tgt = lmv_tgt(lmv, index);
1531 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1534 *mdt = tgt->ltd_index;
1535 lmv->lmv_qos_rr_index = (*mdt + 1) %
1536 lmv->lmv_mdt_descs.ltd_tgts_size;
1537 spin_unlock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1541 spin_unlock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1543 RETURN(ERR_PTR(-ENODEV));
1546 static struct lmv_tgt_desc *
1547 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1548 const char *name, int namelen, struct lu_fid *fid,
1549 __u32 *mds, bool new_layout)
1551 struct lmv_tgt_desc *tgt;
1552 const struct lmv_oinfo *oinfo;
1554 if (!lmv_dir_striped(lsm) || !namelen) {
1555 tgt = lmv_fid2tgt(lmv, fid);
1559 *mds = tgt->ltd_index;
1563 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1564 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1565 return ERR_PTR(-EBADF);
1566 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1568 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, new_layout);
1570 return ERR_CAST(oinfo);
1573 /* check stripe FID is sane */
1574 if (!fid_is_sane(&oinfo->lmo_fid))
1575 return ERR_PTR(-ENODEV);
1577 *fid = oinfo->lmo_fid;
1578 *mds = oinfo->lmo_mds;
1579 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1581 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1583 return tgt ? tgt : ERR_PTR(-ENODEV);
1587 * Locate MDT of op_data->op_fid1
1589 * For striped directory, it will locate the stripe by name hash, if hash_type
1590 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1591 * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1592 * indicates whether old or new layout is used to locate.
1594 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1596 * \param[in] lmv LMV device
1597 * \param[in] op_data client MD stack parameters, name, namelen
1600 * retval pointer to the lmv_tgt_desc if succeed.
1601 * ERR_PTR(errno) if failed.
1603 struct lmv_tgt_desc *
1604 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1606 struct lmv_stripe_md *lsm = op_data->op_mea1;
1607 struct lmv_oinfo *oinfo;
1608 struct lmv_tgt_desc *tgt;
1610 if (lmv_dir_foreign(lsm))
1611 return ERR_PTR(-ENODATA);
1613 /* During creating VOLATILE file, it should honor the mdt
1614 * index if the file under striped dir is being restored, see
1616 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1617 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1618 tgt = lmv_tgt(lmv, op_data->op_mds);
1620 return ERR_PTR(-ENODEV);
1622 if (lmv_dir_striped(lsm)) {
1625 /* refill the right parent fid */
1626 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1627 oinfo = &lsm->lsm_md_oinfo[i];
1628 if (oinfo->lmo_mds == op_data->op_mds) {
1629 op_data->op_fid1 = oinfo->lmo_fid;
1634 if (i == lsm->lsm_md_stripe_count)
1635 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1637 } else if (lmv_dir_bad_hash(lsm)) {
1638 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1639 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1641 op_data->op_fid1 = oinfo->lmo_fid;
1642 op_data->op_mds = oinfo->lmo_mds;
1643 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1645 return ERR_PTR(-ENODEV);
1647 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1648 op_data->op_name, op_data->op_namelen,
1649 &op_data->op_fid1, &op_data->op_mds,
1650 op_data->op_new_layout);
1656 /* Locate MDT of op_data->op_fid2 for link/rename */
1657 static struct lmv_tgt_desc *
1658 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1660 struct lmv_tgt_desc *tgt;
1663 LASSERT(op_data->op_name);
1664 if (lmv_dir_layout_changing(op_data->op_mea2)) {
1665 struct lu_fid fid1 = op_data->op_fid1;
1666 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1667 struct ptlrpc_request *request = NULL;
1670 * avoid creating new file under old layout of migrating
1671 * directory, check it here.
1673 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1674 op_data->op_name, op_data->op_namelen,
1675 &op_data->op_fid2, &op_data->op_mds, false);
1679 op_data->op_fid1 = op_data->op_fid2;
1680 op_data->op_mea1 = op_data->op_mea2;
1681 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1682 op_data->op_fid1 = fid1;
1683 op_data->op_mea1 = lsm1;
1685 ptlrpc_req_finished(request);
1686 RETURN(ERR_PTR(-EEXIST));
1690 RETURN(ERR_PTR(rc));
1693 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1694 op_data->op_namelen, &op_data->op_fid2,
1695 &op_data->op_mds, true);
1698 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1700 struct lu_tgt_desc *tgt;
1701 struct ptlrpc_request *request;
1704 LASSERT(lmv_dir_layout_changing(op_data->op_mea1));
1705 LASSERT(!op_data->op_new_layout);
1707 tgt = lmv_locate_tgt(lmv, op_data);
1709 return PTR_ERR(tgt);
1711 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1713 ptlrpc_req_finished(request);
1720 /* mkdir by QoS in two cases:
1721 * 1. 'lfs mkdir -i -1'
1722 * 2. parent default LMV master_mdt_index is -1
1724 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1725 * directories under striped directory.
1727 static inline bool lmv_op_qos_mkdir(const struct md_op_data *op_data)
1729 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1730 const struct lmv_user_md *lum = op_data->op_data;
1732 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1735 if (lmv_dir_striped(op_data->op_mea1))
1738 if (op_data->op_cli_flags & CLI_SET_MEA && lum &&
1739 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1740 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1741 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT)
1744 if (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT)
1750 /* 'lfs mkdir -i <specific_MDT>' */
1751 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1753 const struct lmv_user_md *lum = op_data->op_data;
1755 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1756 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1757 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1758 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1759 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1762 /* parent default LMV master_mdt_index is not -1. */
1764 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1766 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1767 op_data->op_default_mea1 &&
1768 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1771 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1772 const void *data, size_t datalen, umode_t mode, uid_t uid,
1773 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1774 struct ptlrpc_request **request)
1776 struct obd_device *obd = exp->exp_obd;
1777 struct lmv_obd *lmv = &obd->u.lmv;
1778 struct lmv_tgt_desc *tgt;
1779 struct mdt_body *repbody;
1784 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1787 if (lmv_dir_bad_hash(op_data->op_mea1))
1790 if (lmv_dir_layout_changing(op_data->op_mea1)) {
1792 * if parent is migrating, create() needs to lookup existing
1793 * name in both old and new layout, check old layout on client.
1795 rc = lmv_old_layout_lookup(lmv, op_data);
1799 op_data->op_new_layout = true;
1802 tgt = lmv_locate_tgt(lmv, op_data);
1804 RETURN(PTR_ERR(tgt));
1806 if (lmv_op_user_specific_mkdir(op_data)) {
1807 struct lmv_user_md *lum = op_data->op_data;
1809 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
1810 tgt = lmv_tgt(lmv, op_data->op_mds);
1813 } else if (lmv_op_default_specific_mkdir(op_data)) {
1815 op_data->op_default_mea1->lsm_md_master_mdt_index;
1816 tgt = lmv_tgt(lmv, op_data->op_mds);
1819 } else if (lmv_op_qos_mkdir(op_data)) {
1820 tgt = lmv_locate_tgt_qos(lmv, &op_data->op_mds);
1821 if (tgt == ERR_PTR(-EAGAIN))
1822 tgt = lmv_locate_tgt_rr(lmv, &op_data->op_mds);
1824 RETURN(PTR_ERR(tgt));
1826 * only update statfs after QoS mkdir, this means the cached
1827 * statfs may be stale, and current mkdir may not follow QoS
1828 * accurately, but it's not serious, and avoids periodic statfs
1829 * when client doesn't mkdir by QoS.
1831 lmv_statfs_check_update(obd, tgt);
1835 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1839 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
1840 (int)op_data->op_namelen, op_data->op_name,
1841 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
1844 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1845 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1846 cap_effective, rdev, request);
1848 if (*request == NULL)
1850 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1853 /* dir restripe needs to send to MDT where dir is located */
1854 if (rc != -EREMOTE ||
1855 !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
1858 repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
1859 if (repbody == NULL)
1862 /* Not cross-ref case, just get out of here. */
1863 if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
1866 op_data->op_fid2 = repbody->mbo_fid1;
1867 ptlrpc_req_finished(*request);
1870 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
1872 RETURN(PTR_ERR(tgt));
1874 op_data->op_mds = tgt->ltd_index;
1879 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1880 const union ldlm_policy_data *policy, struct md_op_data *op_data,
1881 struct lustre_handle *lockh, __u64 extra_lock_flags)
1883 struct obd_device *obd = exp->exp_obd;
1884 struct lmv_obd *lmv = &obd->u.lmv;
1885 struct lmv_tgt_desc *tgt;
1890 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1892 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1894 RETURN(PTR_ERR(tgt));
1896 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1897 PFID(&op_data->op_fid1), tgt->ltd_index);
1899 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1906 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1907 struct ptlrpc_request **preq)
1909 struct obd_device *obd = exp->exp_obd;
1910 struct lmv_obd *lmv = &obd->u.lmv;
1911 struct lmv_tgt_desc *tgt;
1912 struct mdt_body *body;
1918 tgt = lmv_locate_tgt(lmv, op_data);
1920 RETURN(PTR_ERR(tgt));
1922 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1923 (int)op_data->op_namelen, op_data->op_name,
1924 PFID(&op_data->op_fid1), tgt->ltd_index);
1926 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1927 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
1928 ptlrpc_req_finished(*preq);
1936 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1937 LASSERT(body != NULL);
1939 if (body->mbo_valid & OBD_MD_MDS) {
1940 op_data->op_fid1 = body->mbo_fid1;
1941 op_data->op_valid |= OBD_MD_FLCROSSREF;
1942 op_data->op_namelen = 0;
1943 op_data->op_name = NULL;
1945 ptlrpc_req_finished(*preq);
1954 #define md_op_data_fid(op_data, fl) \
1955 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1956 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1957 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1958 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1961 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1962 struct md_op_data *op_data, __u32 op_tgt,
1963 enum ldlm_mode mode, int bits, int flag)
1965 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1966 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1967 union ldlm_policy_data policy = { { 0 } };
1971 if (!fid_is_sane(fid))
1975 tgt = lmv_fid2tgt(lmv, fid);
1977 RETURN(PTR_ERR(tgt));
1980 if (tgt->ltd_index != op_tgt) {
1981 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1982 policy.l_inodebits.bits = bits;
1983 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1984 mode, LCF_ASYNC, NULL);
1987 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1989 op_data->op_flags |= flag;
1997 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2000 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2001 struct ptlrpc_request **request)
2003 struct obd_device *obd = exp->exp_obd;
2004 struct lmv_obd *lmv = &obd->u.lmv;
2005 struct lmv_tgt_desc *tgt;
2009 LASSERT(op_data->op_namelen != 0);
2011 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2012 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2013 op_data->op_name, PFID(&op_data->op_fid1));
2015 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2016 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2017 op_data->op_cap = cfs_curproc_cap_pack();
2019 tgt = lmv_locate_tgt2(lmv, op_data);
2021 RETURN(PTR_ERR(tgt));
2024 * Cancel UPDATE lock on child (fid1).
2026 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2027 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2028 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2032 rc = md_link(tgt->ltd_exp, op_data, request);
2037 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2038 const char *name, size_t namelen,
2039 struct ptlrpc_request **request)
2041 struct obd_device *obd = exp->exp_obd;
2042 struct lmv_obd *lmv = &obd->u.lmv;
2043 struct lmv_stripe_md *lsm = op_data->op_mea1;
2044 struct lmv_tgt_desc *parent_tgt;
2045 struct lmv_tgt_desc *sp_tgt;
2046 struct lmv_tgt_desc *tp_tgt = NULL;
2047 struct lmv_tgt_desc *child_tgt;
2048 struct lmv_tgt_desc *tgt;
2049 struct lu_fid target_fid = { 0 };
2054 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2056 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2057 PFID(&op_data->op_fid1), (int)namelen, name);
2059 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2060 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2061 op_data->op_cap = cfs_curproc_cap_pack();
2063 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2064 if (IS_ERR(parent_tgt))
2065 RETURN(PTR_ERR(parent_tgt));
2067 if (lmv_dir_striped(lsm)) {
2068 const struct lmv_oinfo *oinfo;
2070 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
2072 RETURN(PTR_ERR(oinfo));
2074 /* save source stripe FID in fid4 temporarily for ELC */
2075 op_data->op_fid4 = oinfo->lmo_fid;
2076 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2081 * if parent is being migrated too, fill op_fid2 with target
2082 * stripe fid, otherwise the target stripe is not created yet.
2084 if (lmv_dir_layout_changing(lsm)) {
2085 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
2088 RETURN(PTR_ERR(oinfo));
2090 op_data->op_fid2 = oinfo->lmo_fid;
2091 tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2096 sp_tgt = parent_tgt;
2099 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2100 if (IS_ERR(child_tgt))
2101 RETURN(PTR_ERR(child_tgt));
2103 /* for directory, migrate to MDT specified by lum_stripe_offset;
2104 * otherwise migrate to the target stripe of parent, but parent
2105 * directory may have finished migration (normally current file too),
2106 * allocate FID on MDT lum_stripe_offset, and server will check
2107 * whether file was migrated already.
2109 if (S_ISDIR(op_data->op_mode) || !tp_tgt) {
2110 struct lmv_user_md *lum = op_data->op_data;
2112 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2114 op_data->op_mds = tp_tgt->ltd_index;
2116 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2121 * for directory, send migrate request to the MDT where the object will
2122 * be migrated to, because we can't create a striped directory remotely.
2124 * otherwise, send to the MDT where source is located because regular
2125 * file may open lease.
2127 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2128 * backward compatibility.
2130 if (S_ISDIR(op_data->op_mode) &&
2131 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2132 tgt = lmv_fid2tgt(lmv, &target_fid);
2134 RETURN(PTR_ERR(tgt));
2139 /* cancel UPDATE lock of parent master object */
2140 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2141 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2145 /* cancel UPDATE lock of source parent */
2146 if (sp_tgt != parent_tgt) {
2148 * migrate RPC packs master object FID, because we can only pack
2149 * two FIDs in reint RPC, but MDS needs to know both source
2150 * parent and target parent, and it will obtain them from master
2151 * FID and LMV, the other FID in RPC is kept for target.
2153 * since this FID is not passed to MDC, cancel it anyway.
2155 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2156 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2160 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2162 op_data->op_fid4 = target_fid;
2164 /* cancel UPDATE locks of target parent */
2165 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2166 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2170 /* cancel LOOKUP lock of source if source is remote object */
2171 if (child_tgt != sp_tgt) {
2172 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2173 LCK_EX, MDS_INODELOCK_LOOKUP,
2174 MF_MDC_CANCEL_FID3);
2179 /* cancel ELC locks of source */
2180 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2181 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2185 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2190 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2191 const char *old, size_t oldlen,
2192 const char *new, size_t newlen,
2193 struct ptlrpc_request **request)
2195 struct obd_device *obd = exp->exp_obd;
2196 struct lmv_obd *lmv = &obd->u.lmv;
2197 struct lmv_tgt_desc *sp_tgt;
2198 struct lmv_tgt_desc *tp_tgt = NULL;
2199 struct lmv_tgt_desc *src_tgt = NULL;
2200 struct lmv_tgt_desc *tgt;
2201 struct mdt_body *body;
2206 LASSERT(oldlen != 0);
2208 if (op_data->op_cli_flags & CLI_MIGRATE) {
2209 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2213 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2214 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2215 op_data->op_cap = cfs_curproc_cap_pack();
2217 op_data->op_name = new;
2218 op_data->op_namelen = newlen;
2220 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2222 RETURN(PTR_ERR(tp_tgt));
2224 /* Since the target child might be destroyed, and it might become
2225 * orphan, and we can only check orphan on the local MDT right now, so
2226 * we send rename request to the MDT where target child is located. If
2227 * target child does not exist, then it will send the request to the
2229 if (fid_is_sane(&op_data->op_fid4)) {
2230 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2232 RETURN(PTR_ERR(tgt));
2237 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2239 /* cancel UPDATE locks of target parent */
2240 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2241 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2245 if (fid_is_sane(&op_data->op_fid4)) {
2246 /* cancel LOOKUP lock of target on target parent */
2247 if (tgt != tp_tgt) {
2248 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2249 tgt->ltd_index, LCK_EX,
2250 MDS_INODELOCK_LOOKUP,
2251 MF_MDC_CANCEL_FID4);
2257 if (fid_is_sane(&op_data->op_fid3)) {
2258 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2259 if (IS_ERR(src_tgt))
2260 RETURN(PTR_ERR(src_tgt));
2262 /* cancel ELC locks of source */
2263 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2264 LCK_EX, MDS_INODELOCK_ELC,
2265 MF_MDC_CANCEL_FID3);
2270 op_data->op_name = old;
2271 op_data->op_namelen = oldlen;
2273 sp_tgt = lmv_locate_tgt(lmv, op_data);
2275 RETURN(PTR_ERR(sp_tgt));
2277 /* cancel UPDATE locks of source parent */
2278 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2279 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2283 if (fid_is_sane(&op_data->op_fid3)) {
2284 /* cancel LOOKUP lock of source on source parent */
2285 if (src_tgt != sp_tgt) {
2286 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2287 tgt->ltd_index, LCK_EX,
2288 MDS_INODELOCK_LOOKUP,
2289 MF_MDC_CANCEL_FID3);
2296 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2297 PFID(&op_data->op_fid1), (int)oldlen, old,
2298 PFID(&op_data->op_fid2), (int)newlen, new);
2300 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2302 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2303 ptlrpc_req_finished(*request);
2308 if (rc && rc != -EXDEV)
2311 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2315 /* Not cross-ref case, just get out of here. */
2316 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2319 op_data->op_fid4 = body->mbo_fid1;
2321 ptlrpc_req_finished(*request);
2324 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2326 RETURN(PTR_ERR(tgt));
2328 if (fid_is_sane(&op_data->op_fid4)) {
2329 /* cancel LOOKUP lock of target on target parent */
2330 if (tgt != tp_tgt) {
2331 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2332 tgt->ltd_index, LCK_EX,
2333 MDS_INODELOCK_LOOKUP,
2334 MF_MDC_CANCEL_FID4);
2343 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2344 void *ea, size_t ealen, struct ptlrpc_request **request)
2346 struct obd_device *obd = exp->exp_obd;
2347 struct lmv_obd *lmv = &obd->u.lmv;
2348 struct lmv_tgt_desc *tgt;
2353 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2354 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2355 op_data->op_xvalid);
2357 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2358 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2360 RETURN(PTR_ERR(tgt));
2362 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2367 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2368 struct ptlrpc_request **request)
2370 struct obd_device *obd = exp->exp_obd;
2371 struct lmv_obd *lmv = &obd->u.lmv;
2372 struct lmv_tgt_desc *tgt;
2377 tgt = lmv_fid2tgt(lmv, fid);
2379 RETURN(PTR_ERR(tgt));
2381 rc = md_fsync(tgt->ltd_exp, fid, request);
2385 struct stripe_dirent {
2386 struct page *sd_page;
2387 struct lu_dirpage *sd_dp;
2388 struct lu_dirent *sd_ent;
2392 struct lmv_dir_ctxt {
2393 struct lmv_obd *ldc_lmv;
2394 struct md_op_data *ldc_op_data;
2395 struct md_callback *ldc_cb_op;
2398 struct stripe_dirent ldc_stripes[0];
2401 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2403 if (stripe->sd_page) {
2404 kunmap(stripe->sd_page);
2405 put_page(stripe->sd_page);
2406 stripe->sd_page = NULL;
2407 stripe->sd_ent = NULL;
2411 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2415 for (i = 0; i < ctxt->ldc_count; i++)
2416 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2419 /* if @ent is dummy, or . .., get next */
2420 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2421 struct lu_dirent *ent,
2424 for (; ent; ent = lu_dirent_next(ent)) {
2425 /* Skip dummy entry */
2426 if (le16_to_cpu(ent->lde_namelen) == 0)
2429 /* skip . and .. for other stripes */
2431 (strncmp(ent->lde_name, ".",
2432 le16_to_cpu(ent->lde_namelen)) == 0 ||
2433 strncmp(ent->lde_name, "..",
2434 le16_to_cpu(ent->lde_namelen)) == 0))
2437 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2444 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2445 struct stripe_dirent *stripe,
2448 struct md_op_data *op_data = ctxt->ldc_op_data;
2449 struct lmv_oinfo *oinfo;
2450 struct lu_fid fid = op_data->op_fid1;
2451 struct inode *inode = op_data->op_data;
2452 struct lmv_tgt_desc *tgt;
2453 struct lu_dirent *ent = stripe->sd_ent;
2454 __u64 hash = ctxt->ldc_hash;
2459 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2463 if (stripe->sd_page) {
2464 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2466 /* @hash should be the last dirent hash */
2467 LASSERTF(hash <= end,
2468 "ctxt@%p stripe@%p hash %llx end %llx\n",
2469 ctxt, stripe, hash, end);
2470 /* unload last page */
2471 stripe_dirent_unload(stripe);
2473 if (end == MDS_DIR_END_OFF) {
2474 stripe->sd_eof = true;
2480 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2481 if (!oinfo->lmo_root) {
2486 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2492 /* op_data is shared by stripes, reset after use */
2493 op_data->op_fid1 = oinfo->lmo_fid;
2494 op_data->op_fid2 = oinfo->lmo_fid;
2495 op_data->op_data = oinfo->lmo_root;
2497 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_cb_op, hash,
2500 op_data->op_fid1 = fid;
2501 op_data->op_fid2 = fid;
2502 op_data->op_data = inode;
2507 stripe->sd_dp = page_address(stripe->sd_page);
2508 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2510 /* in case a page filled with ., .. and dummy, read next */
2513 stripe->sd_ent = ent;
2516 /* treat error as eof, so dir can be partially accessed */
2517 stripe->sd_eof = true;
2518 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2519 "directory is partially accessed!\n",
2520 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2527 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2529 struct obd_device *obd = exp->exp_obd;
2530 struct lmv_obd *lmv = &obd->u.lmv;
2531 struct lmv_tgt_desc *tgt;
2536 rc = lmv_check_connect(obd);
2540 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2542 RETURN(PTR_ERR(tgt));
2544 data->op_flags |= MF_MDC_CANCEL_FID1;
2545 rc = md_file_resync(tgt->ltd_exp, data);
2550 * Get dirent with the closest hash for striped directory
2552 * This function will search the dir entry, whose hash value is the
2553 * closest(>=) to hash from all of sub-stripes, and it is only being called
2554 * for striped directory.
2556 * \param[in] ctxt dir read context
2558 * \retval dirent get the entry successfully
2559 * NULL does not get the entry, normally it means
2560 * it reaches the end of the directory, while read
2561 * stripe dirent error is ignored to allow partial
2564 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2566 struct stripe_dirent *stripe;
2567 struct lu_dirent *ent = NULL;
2571 /* TODO: optimize with k-way merge sort */
2572 for (i = 0; i < ctxt->ldc_count; i++) {
2573 stripe = &ctxt->ldc_stripes[i];
2577 if (!stripe->sd_ent) {
2578 stripe_dirent_load(ctxt, stripe, i);
2579 if (!stripe->sd_ent) {
2580 LASSERT(stripe->sd_eof);
2586 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2587 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2589 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2596 stripe = &ctxt->ldc_stripes[min];
2597 ent = stripe->sd_ent;
2598 /* pop found dirent */
2599 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2607 * Build dir entry page for striped directory
2609 * This function gets one entry by @offset from a striped directory. It will
2610 * read entries from all of stripes, and choose one closest to the required
2611 * offset(&offset). A few notes
2612 * 1. skip . and .. for non-zero stripes, because there can only have one .
2613 * and .. in a directory.
2614 * 2. op_data will be shared by all of stripes, instead of allocating new
2615 * one, so need to restore before reusing.
2617 * \param[in] exp obd export refer to LMV
2618 * \param[in] op_data hold those MD parameters of read_entry
2619 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2620 * \param[in] offset starting hash offset
2621 * \param[out] ppage the page holding the entry. Note: because the entry
2622 * will be accessed in upper layer, so we need hold the
2623 * page until the usages of entry is finished, see
2624 * ll_dir_entry_next.
2626 * retval =0 if get entry successfully
2627 * <0 cannot get entry
2629 static int lmv_striped_read_page(struct obd_export *exp,
2630 struct md_op_data *op_data,
2631 struct md_callback *cb_op,
2632 __u64 offset, struct page **ppage)
2634 struct page *page = NULL;
2635 struct lu_dirpage *dp;
2637 struct lu_dirent *ent;
2638 struct lu_dirent *last_ent;
2640 struct lmv_dir_ctxt *ctxt;
2641 struct lu_dirent *next = NULL;
2647 /* Allocate a page and read entries from all of stripes and fill
2648 * the page by hash order */
2649 page = alloc_page(GFP_KERNEL);
2653 /* Initialize the entry page */
2655 memset(dp, 0, sizeof(*dp));
2656 dp->ldp_hash_start = cpu_to_le64(offset);
2659 left_bytes = PAGE_SIZE - sizeof(*dp);
2663 /* initalize dir read context */
2664 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2665 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2667 GOTO(free_page, rc = -ENOMEM);
2668 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2669 ctxt->ldc_op_data = op_data;
2670 ctxt->ldc_cb_op = cb_op;
2671 ctxt->ldc_hash = offset;
2672 ctxt->ldc_count = stripe_count;
2675 next = lmv_dirent_next(ctxt);
2677 /* end of directory */
2679 ctxt->ldc_hash = MDS_DIR_END_OFF;
2682 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2684 ent_size = le16_to_cpu(next->lde_reclen);
2686 /* the last entry lde_reclen is 0, but it might not be the last
2687 * one of this temporay dir page */
2689 ent_size = lu_dirent_calc_size(
2690 le16_to_cpu(next->lde_namelen),
2691 le32_to_cpu(next->lde_attrs));
2693 if (ent_size > left_bytes)
2696 memcpy(ent, next, ent_size);
2698 /* Replace . with master FID and Replace .. with the parent FID
2699 * of master object */
2700 if (strncmp(ent->lde_name, ".",
2701 le16_to_cpu(ent->lde_namelen)) == 0 &&
2702 le16_to_cpu(ent->lde_namelen) == 1)
2703 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2704 else if (strncmp(ent->lde_name, "..",
2705 le16_to_cpu(ent->lde_namelen)) == 0 &&
2706 le16_to_cpu(ent->lde_namelen) == 2)
2707 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2709 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2710 le16_to_cpu(ent->lde_namelen), ent->lde_name,
2711 le64_to_cpu(ent->lde_hash));
2713 left_bytes -= ent_size;
2714 ent->lde_reclen = cpu_to_le16(ent_size);
2716 ent = (void *)ent + ent_size;
2719 last_ent->lde_reclen = 0;
2722 dp->ldp_flags |= LDF_EMPTY;
2723 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2724 dp->ldp_flags |= LDF_COLLIDE;
2725 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2726 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2728 put_lmv_dir_ctxt(ctxt);
2729 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2742 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2743 struct md_callback *cb_op, __u64 offset,
2744 struct page **ppage)
2746 struct obd_device *obd = exp->exp_obd;
2747 struct lmv_obd *lmv = &obd->u.lmv;
2748 struct lmv_tgt_desc *tgt;
2753 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
2756 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
2757 rc = lmv_striped_read_page(exp, op_data, cb_op, offset, ppage);
2761 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2763 RETURN(PTR_ERR(tgt));
2765 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2771 * Unlink a file/directory
2773 * Unlink a file or directory under the parent dir. The unlink request
2774 * usually will be sent to the MDT where the child is located, but if
2775 * the client does not have the child FID then request will be sent to the
2776 * MDT where the parent is located.
2778 * If the parent is a striped directory then it also needs to locate which
2779 * stripe the name of the child is located, and replace the parent FID
2780 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2781 * it will walk through all of sub-stripes until the child is being
2784 * \param[in] exp export refer to LMV
2785 * \param[in] op_data different parameters transferred beween client
2786 * MD stacks, name, namelen, FIDs etc.
2787 * op_fid1 is the parent FID, op_fid2 is the child
2789 * \param[out] request point to the request of unlink.
2791 * retval 0 if succeed
2792 * negative errno if failed.
2794 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2795 struct ptlrpc_request **request)
2797 struct obd_device *obd = exp->exp_obd;
2798 struct lmv_obd *lmv = &obd->u.lmv;
2799 struct lmv_tgt_desc *tgt;
2800 struct lmv_tgt_desc *parent_tgt;
2801 struct mdt_body *body;
2806 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2807 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2808 op_data->op_cap = cfs_curproc_cap_pack();
2811 parent_tgt = lmv_locate_tgt(lmv, op_data);
2812 if (IS_ERR(parent_tgt))
2813 RETURN(PTR_ERR(parent_tgt));
2815 if (likely(!fid_is_zero(&op_data->op_fid2))) {
2816 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2818 RETURN(PTR_ERR(tgt));
2824 * If child's fid is given, cancel unused locks for it if it is from
2825 * another export than parent.
2827 * LOOKUP lock for child (fid3) should also be cancelled on parent
2828 * tgt_tgt in mdc_unlink().
2830 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2832 if (parent_tgt != tgt)
2833 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
2834 LCK_EX, MDS_INODELOCK_LOOKUP,
2835 MF_MDC_CANCEL_FID3);
2837 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2838 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2842 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2843 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
2846 rc = md_unlink(tgt->ltd_exp, op_data, request);
2847 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2848 ptlrpc_req_finished(*request);
2856 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2860 /* Not cross-ref case, just get out of here. */
2861 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2864 /* This is a remote object, try remote MDT. */
2865 op_data->op_fid2 = body->mbo_fid1;
2866 ptlrpc_req_finished(*request);
2869 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2871 RETURN(PTR_ERR(tgt));
2876 static int lmv_precleanup(struct obd_device *obd)
2879 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
2880 fld_client_debugfs_fini(&obd->u.lmv.lmv_fld);
2881 lprocfs_obd_cleanup(obd);
2882 lprocfs_free_md_stats(obd);
2887 * Get by key a value associated with a LMV device.
2889 * Dispatch request to lower-layer devices as needed.
2891 * \param[in] env execution environment for this thread
2892 * \param[in] exp export for the LMV device
2893 * \param[in] keylen length of key identifier
2894 * \param[in] key identifier of key to get value for
2895 * \param[in] vallen size of \a val
2896 * \param[out] val pointer to storage location for value
2897 * \param[in] lsm optional striping metadata of object
2899 * \retval 0 on success
2900 * \retval negative negated errno on failure
2902 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2903 __u32 keylen, void *key, __u32 *vallen, void *val)
2905 struct obd_device *obd;
2906 struct lmv_obd *lmv;
2907 struct lu_tgt_desc *tgt;
2912 obd = class_exp2obd(exp);
2914 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2915 exp->exp_handle.h_cookie);
2920 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2921 LASSERT(*vallen == sizeof(__u32));
2922 lmv_foreach_connected_tgt(lmv, tgt) {
2923 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2928 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2929 KEY_IS(KEY_DEFAULT_EASIZE) ||
2930 KEY_IS(KEY_CONN_DATA)) {
2932 * Forwarding this request to first MDS, it should know LOV
2935 tgt = lmv_tgt(lmv, 0);
2939 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
2940 if (!rc && KEY_IS(KEY_CONN_DATA))
2941 exp->exp_connect_data = *(struct obd_connect_data *)val;
2943 } else if (KEY_IS(KEY_TGT_COUNT)) {
2944 *((int *)val) = lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count;
2948 CDEBUG(D_IOCTL, "Invalid key\n");
2952 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
2953 int *__rcs, struct ptlrpc_request_set *_set)
2955 struct obd_device *obd = class_exp2obd(exp);
2956 struct ptlrpc_request_set *set = _set;
2957 struct lmv_obd *lmv = &obd->u.lmv;
2958 int tgt_count = lmv->lmv_mdt_count;
2959 struct lu_tgt_desc *tgt;
2960 struct fid_array *fat, **fas = NULL;
2961 int i, rc, **rcs = NULL;
2964 set = ptlrpc_prep_set();
2969 /* split FIDs by targets */
2970 OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
2972 GOTO(out, rc = -ENOMEM);
2973 OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
2975 GOTO(out_fas, rc = -ENOMEM);
2977 for (i = 0; i < fa->fa_nr; i++) {
2980 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
2982 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
2983 PFID(&fa->fa_fids[i]), rc);
2986 LASSERT(idx < tgt_count);
2988 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
2989 fa_fids[fa->fa_nr]));
2991 GOTO(out, rc = -ENOMEM);
2993 OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
2995 GOTO(out, rc = -ENOMEM);
2998 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3001 lmv_foreach_connected_tgt(lmv, tgt) {
3002 fat = fas[tgt->ltd_index];
3003 if (!fat || fat->fa_nr == 0)
3005 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3008 rc = ptlrpc_set_wait(NULL, set);
3011 for (i = 0; i < tgt_count; i++) {
3013 if (!fat || fat->fa_nr == 0)
3015 /* copy FIDs back */
3016 memcpy(fa->fa_fids + j, fat->fa_fids,
3017 fat->fa_nr * sizeof(struct lu_fid));
3019 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3024 ptlrpc_set_destroy(set);
3027 for (i = 0; i < tgt_count; i++) {
3029 OBD_FREE(fas[i], offsetof(struct fid_array,
3030 fa_fids[fa->fa_nr]));
3032 OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3035 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3038 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3044 * Asynchronously set by key a value associated with a LMV device.
3046 * Dispatch request to lower-layer devices as needed.
3048 * \param[in] env execution environment for this thread
3049 * \param[in] exp export for the LMV device
3050 * \param[in] keylen length of key identifier
3051 * \param[in] key identifier of key to store value for
3052 * \param[in] vallen size of value to store
3053 * \param[in] val pointer to data to be stored
3054 * \param[in] set optional list of related ptlrpc requests
3056 * \retval 0 on success
3057 * \retval negative negated errno on failure
3059 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3060 __u32 keylen, void *key, __u32 vallen, void *val,
3061 struct ptlrpc_request_set *set)
3063 struct lmv_tgt_desc *tgt = NULL;
3064 struct obd_device *obd;
3065 struct lmv_obd *lmv;
3069 obd = class_exp2obd(exp);
3071 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3072 exp->exp_handle.h_cookie);
3077 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3078 KEY_IS(KEY_DEFAULT_EASIZE)) {
3081 lmv_foreach_connected_tgt(lmv, tgt) {
3082 err = obd_set_info_async(env, tgt->ltd_exp,
3083 keylen, key, vallen, val, set);
3094 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3095 const struct lmv_mds_md_v1 *lmm1)
3097 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3104 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3105 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3106 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3107 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3108 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3110 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3111 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3112 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3113 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3114 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3115 sizeof(lsm->lsm_md_pool_name));
3117 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3120 CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3121 "layout_version %d\n", lsm->lsm_md_stripe_count,
3122 lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3123 lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3124 lsm->lsm_md_layout_version);
3126 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3127 for (i = 0; i < stripe_count; i++) {
3128 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3129 &lmm1->lmv_stripe_fids[i]);
3131 * set default value -1, so lmv_locate_tgt() knows this stripe
3132 * target is not initialized.
3134 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3135 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3138 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3139 &lsm->lsm_md_oinfo[i].lmo_mds);
3146 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3147 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3153 static inline int lmv_unpack_user_md(struct obd_export *exp,
3154 struct lmv_stripe_md *lsm,
3155 const struct lmv_user_md *lmu)
3157 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3158 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3159 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3160 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3161 lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3166 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3167 const union lmv_mds_md *lmm, size_t lmm_size)
3169 struct lmv_stripe_md *lsm;
3172 bool allocated = false;
3175 LASSERT(lsmp != NULL);
3179 if (lsm != NULL && lmm == NULL) {
3181 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3183 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3186 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3188 OBD_FREE_LARGE(lfm, lfm_size);
3192 if (lmv_dir_striped(lsm)) {
3193 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3194 if (lsm->lsm_md_oinfo[i].lmo_root)
3195 iput(lsm->lsm_md_oinfo[i].lmo_root);
3197 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3199 lsm_size = lmv_stripe_md_size(0);
3201 OBD_FREE(lsm, lsm_size);
3206 /* foreign lmv case */
3207 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3208 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3211 OBD_ALLOC_LARGE(lfm, lmm_size);
3214 *lsmp = (struct lmv_stripe_md *)lfm;
3216 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3217 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3218 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3219 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3220 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3225 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3229 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3230 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3231 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3232 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3237 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3238 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3241 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3242 * stripecount should be 0 then.
3244 lsm_size = lmv_stripe_md_size(0);
3247 OBD_ALLOC(lsm, lsm_size);
3254 switch (le32_to_cpu(lmm->lmv_magic)) {
3256 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3258 case LMV_USER_MAGIC:
3259 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3262 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3263 le32_to_cpu(lmm->lmv_magic));
3268 if (rc != 0 && allocated) {
3269 OBD_FREE(lsm, lsm_size);
3276 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3278 lmv_unpackmd(NULL, &lsm, NULL, 0);
3280 EXPORT_SYMBOL(lmv_free_memmd);
3282 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3283 union ldlm_policy_data *policy,
3284 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3287 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3288 struct lu_tgt_desc *tgt;
3294 LASSERT(fid != NULL);
3296 lmv_foreach_connected_tgt(lmv, tgt) {
3297 if (!tgt->ltd_active)
3300 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3308 static int lmv_set_lock_data(struct obd_export *exp,
3309 const struct lustre_handle *lockh,
3310 void *data, __u64 *bits)
3312 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3313 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3318 if (tgt == NULL || tgt->ltd_exp == NULL)
3320 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3324 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3325 const struct lu_fid *fid, enum ldlm_type type,
3326 union ldlm_policy_data *policy,
3327 enum ldlm_mode mode, struct lustre_handle *lockh)
3329 struct obd_device *obd = exp->exp_obd;
3330 struct lmv_obd *lmv = &obd->u.lmv;
3332 struct lu_tgt_desc *tgt;
3338 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3341 * With DNE every object can have two locks in different namespaces:
3342 * lookup lock in space of MDT storing direntry and update/open lock in
3343 * space of MDT storing inode. Try the MDT that the FID maps to first,
3344 * since this can be easily found, and only try others if that fails.
3346 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3347 i < lmv->lmv_mdt_descs.ltd_tgts_size;
3348 i++, index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3350 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3351 obd->obd_name, PFID(fid), index);
3355 tgt = lmv_tgt(lmv, index);
3356 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3359 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
3368 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3369 struct obd_export *dt_exp, struct obd_export *md_exp,
3370 struct lustre_md *md)
3372 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3373 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3375 if (!tgt || !tgt->ltd_exp)
3378 return md_get_lustre_md(tgt->ltd_exp, req, dt_exp, md_exp, md);
3381 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3383 struct obd_device *obd = exp->exp_obd;
3384 struct lmv_obd *lmv = &obd->u.lmv;
3385 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3389 if (md->default_lmv) {
3390 lmv_free_memmd(md->default_lmv);
3391 md->default_lmv = NULL;
3393 if (md->lmv != NULL) {
3394 lmv_free_memmd(md->lmv);
3397 if (!tgt || !tgt->ltd_exp)
3399 RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3402 int lmv_set_open_replay_data(struct obd_export *exp,
3403 struct obd_client_handle *och,
3404 struct lookup_intent *it)
3406 struct obd_device *obd = exp->exp_obd;
3407 struct lmv_obd *lmv = &obd->u.lmv;
3408 struct lmv_tgt_desc *tgt;
3412 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3414 RETURN(PTR_ERR(tgt));
3416 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3419 int lmv_clear_open_replay_data(struct obd_export *exp,
3420 struct obd_client_handle *och)
3422 struct obd_device *obd = exp->exp_obd;
3423 struct lmv_obd *lmv = &obd->u.lmv;
3424 struct lmv_tgt_desc *tgt;
3428 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3430 RETURN(PTR_ERR(tgt));
3432 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3435 int lmv_intent_getattr_async(struct obd_export *exp,
3436 struct md_enqueue_info *minfo)
3438 struct md_op_data *op_data = &minfo->mi_data;
3439 struct obd_device *obd = exp->exp_obd;
3440 struct lmv_obd *lmv = &obd->u.lmv;
3441 struct lmv_tgt_desc *ptgt;
3442 struct lmv_tgt_desc *ctgt;
3447 if (!fid_is_sane(&op_data->op_fid2))
3450 ptgt = lmv_locate_tgt(lmv, op_data);
3452 RETURN(PTR_ERR(ptgt));
3454 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3456 RETURN(PTR_ERR(ctgt));
3459 * remote object needs two RPCs to lookup and getattr, considering the
3460 * complexity don't support statahead for now.
3465 rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3470 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3471 struct lu_fid *fid, __u64 *bits)
3473 struct obd_device *obd = exp->exp_obd;
3474 struct lmv_obd *lmv = &obd->u.lmv;
3475 struct lmv_tgt_desc *tgt;
3480 tgt = lmv_fid2tgt(lmv, fid);
3482 RETURN(PTR_ERR(tgt));
3484 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3488 int lmv_get_fid_from_lsm(struct obd_export *exp,
3489 const struct lmv_stripe_md *lsm,
3490 const char *name, int namelen, struct lu_fid *fid)
3492 const struct lmv_oinfo *oinfo;
3494 LASSERT(lmv_dir_striped(lsm));
3496 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3498 return PTR_ERR(oinfo);
3500 *fid = oinfo->lmo_fid;
3506 * For lmv, only need to send request to master MDT, and the master MDT will
3507 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3508 * we directly fetch data from the slave MDTs.
3510 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3511 struct obd_quotactl *oqctl)
3513 struct obd_device *obd = class_exp2obd(exp);
3514 struct lmv_obd *lmv = &obd->u.lmv;
3515 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3516 __u64 curspace, curinodes;
3521 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3522 CERROR("master lmv inactive\n");
3526 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3527 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3531 curspace = curinodes = 0;
3532 lmv_foreach_connected_tgt(lmv, tgt) {
3535 if (!tgt->ltd_active)
3538 err = obd_quotactl(tgt->ltd_exp, oqctl);
3540 CERROR("getquota on mdt %d failed. %d\n",
3541 tgt->ltd_index, err);
3545 curspace += oqctl->qc_dqblk.dqb_curspace;
3546 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3549 oqctl->qc_dqblk.dqb_curspace = curspace;
3550 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3555 static int lmv_merge_attr(struct obd_export *exp,
3556 const struct lu_fid *fid,
3557 const struct lmv_stripe_md *lsm,
3558 struct cl_attr *attr,
3559 ldlm_blocking_callback cb_blocking)
3564 if (!lmv_dir_striped(lsm))
3567 rc = lmv_revalidate_slaves(exp, fid, lsm, cb_blocking, 0);
3571 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3572 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3578 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3579 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3580 i_size_read(inode), (unsigned long long)inode->i_blocks,
3581 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3582 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3584 /* for slave stripe, it needs to subtract nlink for . and .. */
3586 attr->cat_nlink += inode->i_nlink - 2;
3588 attr->cat_nlink = inode->i_nlink;
3590 attr->cat_size += i_size_read(inode);
3591 attr->cat_blocks += inode->i_blocks;
3593 if (attr->cat_atime < inode->i_atime.tv_sec)
3594 attr->cat_atime = inode->i_atime.tv_sec;
3596 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3597 attr->cat_ctime = inode->i_ctime.tv_sec;
3599 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3600 attr->cat_mtime = inode->i_mtime.tv_sec;
3605 static const struct obd_ops lmv_obd_ops = {
3606 .o_owner = THIS_MODULE,
3607 .o_setup = lmv_setup,
3608 .o_cleanup = lmv_cleanup,
3609 .o_precleanup = lmv_precleanup,
3610 .o_process_config = lmv_process_config,
3611 .o_connect = lmv_connect,
3612 .o_disconnect = lmv_disconnect,
3613 .o_statfs = lmv_statfs,
3614 .o_get_info = lmv_get_info,
3615 .o_set_info_async = lmv_set_info_async,
3616 .o_notify = lmv_notify,
3617 .o_get_uuid = lmv_get_uuid,
3618 .o_fid_alloc = lmv_fid_alloc,
3619 .o_iocontrol = lmv_iocontrol,
3620 .o_quotactl = lmv_quotactl
3623 static const struct md_ops lmv_md_ops = {
3624 .m_get_root = lmv_get_root,
3625 .m_null_inode = lmv_null_inode,
3626 .m_close = lmv_close,
3627 .m_create = lmv_create,
3628 .m_enqueue = lmv_enqueue,
3629 .m_getattr = lmv_getattr,
3630 .m_getxattr = lmv_getxattr,
3631 .m_getattr_name = lmv_getattr_name,
3632 .m_intent_lock = lmv_intent_lock,
3634 .m_rename = lmv_rename,
3635 .m_setattr = lmv_setattr,
3636 .m_setxattr = lmv_setxattr,
3637 .m_fsync = lmv_fsync,
3638 .m_file_resync = lmv_file_resync,
3639 .m_read_page = lmv_read_page,
3640 .m_unlink = lmv_unlink,
3641 .m_init_ea_size = lmv_init_ea_size,
3642 .m_cancel_unused = lmv_cancel_unused,
3643 .m_set_lock_data = lmv_set_lock_data,
3644 .m_lock_match = lmv_lock_match,
3645 .m_get_lustre_md = lmv_get_lustre_md,
3646 .m_free_lustre_md = lmv_free_lustre_md,
3647 .m_merge_attr = lmv_merge_attr,
3648 .m_set_open_replay_data = lmv_set_open_replay_data,
3649 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3650 .m_intent_getattr_async = lmv_intent_getattr_async,
3651 .m_revalidate_lock = lmv_revalidate_lock,
3652 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3653 .m_unpackmd = lmv_unpackmd,
3654 .m_rmfid = lmv_rmfid,
3657 static int __init lmv_init(void)
3659 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3660 LUSTRE_LMV_NAME, NULL);
3663 static void __exit lmv_exit(void)
3665 class_unregister_type(LUSTRE_LMV_NAME);
3668 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3669 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3670 MODULE_VERSION(LUSTRE_VERSION_STRING);
3671 MODULE_LICENSE("GPL");
3673 module_init(lmv_init);
3674 module_exit(lmv_exit);