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 #ifdef HAVE_UIDGID_HEADER
40 # include <linux/uidgid.h>
42 #include <linux/slab.h>
43 #include <linux/pagemap.h>
45 #include <linux/math64.h>
46 #include <linux/seq_file.h>
47 #include <linux/namei.h>
49 #include <obd_support.h>
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <obd_class.h>
53 #include <lustre_lmv.h>
54 #include <lprocfs_status.h>
55 #include <cl_object.h>
56 #include <lustre_fid.h>
57 #include <uapi/linux/lustre/lustre_ioctl.h>
58 #include <lustre_kernelcomm.h>
59 #include "lmv_internal.h"
61 static int lmv_check_connect(struct obd_device *obd);
63 void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt,
66 if (tgt->ltd_active == activate)
69 tgt->ltd_active = activate;
70 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count +=
73 tgt->ltd_exp->exp_obd->obd_inactive = !activate;
79 * -EINVAL : UUID can't be found in the LMV's target list
80 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
81 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
83 static int lmv_set_mdc_active(struct lmv_obd *lmv,
84 const struct obd_uuid *uuid,
87 struct lu_tgt_desc *tgt = NULL;
88 struct obd_device *obd;
93 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
94 lmv, uuid->uuid, activate);
96 spin_lock(&lmv->lmv_lock);
97 lmv_foreach_connected_tgt(lmv, tgt) {
98 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n",
99 tgt->ltd_index, tgt->ltd_uuid.uuid,
100 tgt->ltd_exp->exp_handle.h_cookie);
102 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
107 GOTO(out_lmv_lock, rc = -EINVAL);
109 obd = class_exp2obd(tgt->ltd_exp);
111 GOTO(out_lmv_lock, rc = -ENOTCONN);
113 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
114 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
115 obd->obd_type->typ_name, tgt->ltd_index);
116 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
118 if (tgt->ltd_active == activate) {
119 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
120 activate ? "" : "in");
121 GOTO(out_lmv_lock, rc);
124 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
125 activate ? "" : "in");
126 lmv_activate_target(lmv, tgt, activate);
130 spin_unlock(&lmv->lmv_lock);
134 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
136 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
137 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
139 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
142 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
143 enum obd_notify_event ev)
145 struct obd_connect_data *conn_data;
146 struct lmv_obd *lmv = &obd->u.lmv;
147 struct obd_uuid *uuid;
151 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
152 CERROR("unexpected notification of %s %s!\n",
153 watched->obd_type->typ_name,
158 uuid = &watched->u.cli.cl_target_uuid;
159 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
161 * Set MDC as active before notifying the observer, so the
162 * observer can use the MDC normally.
164 rc = lmv_set_mdc_active(lmv, uuid,
165 ev == OBD_NOTIFY_ACTIVE);
167 CERROR("%sactivation of %s failed: %d\n",
168 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
172 } else if (ev == OBD_NOTIFY_OCD) {
173 conn_data = &watched->u.cli.cl_import->imp_connect_data;
175 * XXX: Make sure that ocd_connect_flags from all targets are
176 * the same. Otherwise one of MDTs runs wrong version or
177 * something like this. --umka
179 obd->obd_self_export->exp_connect_data = *conn_data;
183 * Pass the notification up the chain.
185 if (obd->obd_observer)
186 rc = obd_notify(obd->obd_observer, watched, ev);
191 static int lmv_connect(const struct lu_env *env,
192 struct obd_export **pexp, struct obd_device *obd,
193 struct obd_uuid *cluuid, struct obd_connect_data *data,
196 struct lmv_obd *lmv = &obd->u.lmv;
197 struct lustre_handle conn = { 0 };
198 struct obd_export *exp;
202 rc = class_connect(&conn, obd, cluuid);
204 CERROR("class_connection() returned %d\n", rc);
208 exp = class_conn2export(&conn);
211 lmv->conn_data = *data;
212 lmv->lmv_cache = localdata;
214 lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
215 &obd->obd_kset.kobj);
216 if (!lmv->lmv_tgts_kobj) {
217 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
218 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
221 rc = lmv_check_connect(obd);
230 if (lmv->lmv_tgts_kobj)
231 kobject_put(lmv->lmv_tgts_kobj);
233 class_disconnect(exp);
238 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
241 struct obd_device *obd = exp->exp_obd;
242 struct lmv_obd *lmv = &obd->u.lmv;
243 struct lmv_tgt_desc *tgt;
249 if (lmv->max_easize < easize) {
250 lmv->max_easize = easize;
253 if (lmv->max_def_easize < def_easize) {
254 lmv->max_def_easize = def_easize;
261 if (lmv->connected == 0)
264 lmv_foreach_connected_tgt(lmv, tgt) {
265 if (!tgt->ltd_active)
268 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
270 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
271 " rc = %d\n", obd->obd_name, tgt->ltd_index, rc);
278 #define MAX_STRING_SIZE 128
280 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
282 struct lmv_obd *lmv = &obd->u.lmv;
283 struct obd_device *mdc_obd;
284 struct obd_export *mdc_exp;
285 struct lu_fld_target target;
289 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
292 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
296 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
297 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
298 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
300 if (!mdc_obd->obd_set_up) {
301 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
305 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
306 &lmv->conn_data, lmv->lmv_cache);
308 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
313 * Init fid sequence client for this mdc and add new fld target.
315 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
319 target.ft_srv = NULL;
320 target.ft_exp = mdc_exp;
321 target.ft_idx = tgt->ltd_index;
323 fld_client_add_target(&lmv->lmv_fld, &target);
325 rc = obd_register_observer(mdc_obd, obd);
327 obd_disconnect(mdc_exp);
328 CERROR("target %s register_observer error %d\n",
329 tgt->ltd_uuid.uuid, rc);
333 if (obd->obd_observer) {
335 * Tell the observer about the new target.
337 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
340 obd_disconnect(mdc_exp);
346 tgt->ltd_exp = mdc_exp;
347 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count++;
349 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
351 rc = lu_qos_add_tgt(&lmv->lmv_qos, tgt);
353 obd_disconnect(mdc_exp);
357 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
358 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
359 atomic_read(&obd->obd_refcount));
361 lmv_statfs_check_update(obd, tgt);
363 if (lmv->lmv_tgts_kobj)
364 /* Even if we failed to create the link, that's fine */
365 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
366 &mdc_obd->obd_kset.kobj,
371 static void lmv_del_target(struct lmv_obd *lmv, struct lu_tgt_desc *tgt)
374 ltd_del_tgt(&lmv->lmv_mdt_descs, tgt);
378 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
379 __u32 index, int gen)
381 struct obd_device *mdc_obd;
382 struct lmv_obd *lmv = &obd->u.lmv;
383 struct lmv_tgt_desc *tgt;
384 struct lu_tgt_descs *ltd = &lmv->lmv_mdt_descs;
389 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
390 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
393 CERROR("%s: Target %s not attached: rc = %d\n",
394 obd->obd_name, uuidp->uuid, -EINVAL);
402 mutex_init(&tgt->ltd_fid_mutex);
403 tgt->ltd_index = index;
404 tgt->ltd_uuid = *uuidp;
407 mutex_lock(<d->ltd_mutex);
408 rc = ltd_add_tgt(ltd, tgt);
409 mutex_unlock(<d->ltd_mutex);
415 /* lmv_check_connect() will connect this target. */
418 rc = lmv_connect_mdc(obd, tgt);
420 int easize = sizeof(struct lmv_stripe_md) +
421 lmv->lmv_mdt_count * sizeof(struct lu_fid);
423 lmv_init_ea_size(obd->obd_self_export, easize, 0);
433 static int lmv_check_connect(struct obd_device *obd)
435 struct lmv_obd *lmv = &obd->u.lmv;
436 struct lmv_tgt_desc *tgt;
445 mutex_lock(&lmv->lmv_mdt_descs.ltd_mutex);
447 GOTO(unlock, rc = 0);
449 if (!lmv->lmv_mdt_count) {
450 CERROR("%s: no targets configured: rc = -EINVAL\n",
452 GOTO(unlock, rc = -EINVAL);
455 if (!lmv_mdt0_inited(lmv)) {
456 CERROR("%s: no target configured for index 0: rc = -EINVAL.\n",
458 GOTO(unlock, rc = -EINVAL);
461 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
462 obd->obd_uuid.uuid, obd->obd_name);
464 lmv_foreach_tgt(lmv, tgt) {
465 rc = lmv_connect_mdc(obd, tgt);
471 easize = lmv_mds_md_size(lmv->lmv_mdt_count, LMV_MAGIC);
472 lmv_init_ea_size(obd->obd_self_export, easize, 0);
475 mutex_unlock(&lmv->lmv_mdt_descs.ltd_mutex);
480 lmv_foreach_tgt(lmv, tgt) {
485 --lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count;
486 obd_disconnect(tgt->ltd_exp);
492 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
494 struct lmv_obd *lmv = &obd->u.lmv;
495 struct obd_device *mdc_obd;
499 LASSERT(tgt != NULL);
500 LASSERT(obd != NULL);
502 mdc_obd = class_exp2obd(tgt->ltd_exp);
505 mdc_obd->obd_force = obd->obd_force;
506 mdc_obd->obd_fail = obd->obd_fail;
507 mdc_obd->obd_no_recov = obd->obd_no_recov;
509 if (lmv->lmv_tgts_kobj)
510 sysfs_remove_link(lmv->lmv_tgts_kobj,
514 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
516 CERROR("Can't finanize fids factory\n");
518 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
519 tgt->ltd_exp->exp_obd->obd_name,
520 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
522 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
523 rc = obd_disconnect(tgt->ltd_exp);
525 if (tgt->ltd_active) {
526 CERROR("Target %s disconnect error %d\n",
527 tgt->ltd_uuid.uuid, rc);
531 lmv_activate_target(lmv, tgt, 0);
536 static int lmv_disconnect(struct obd_export *exp)
538 struct obd_device *obd = class_exp2obd(exp);
539 struct lmv_obd *lmv = &obd->u.lmv;
540 struct lmv_tgt_desc *tgt;
545 lmv_foreach_connected_tgt(lmv, tgt)
546 lmv_disconnect_mdc(obd, tgt);
548 if (lmv->lmv_tgts_kobj)
549 kobject_put(lmv->lmv_tgts_kobj);
552 class_export_put(exp);
553 rc = class_disconnect(exp);
559 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
562 struct obd_device *obddev = class_exp2obd(exp);
563 struct lmv_obd *lmv = &obddev->u.lmv;
564 struct getinfo_fid2path *gf;
565 struct lmv_tgt_desc *tgt;
566 struct getinfo_fid2path *remote_gf = NULL;
567 struct lu_fid root_fid;
568 int remote_gf_size = 0;
572 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
574 RETURN(PTR_ERR(tgt));
576 root_fid = *gf->gf_u.gf_root_fid;
577 LASSERT(fid_is_sane(&root_fid));
580 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
581 if (rc != 0 && rc != -EREMOTE)
582 GOTO(out_fid2path, rc);
584 /* If remote_gf != NULL, it means just building the
585 * path on the remote MDT, copy this path segement to gf */
586 if (remote_gf != NULL) {
587 struct getinfo_fid2path *ori_gf;
591 ori_gf = (struct getinfo_fid2path *)karg;
592 if (strlen(ori_gf->gf_u.gf_path) + 1 +
593 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
594 GOTO(out_fid2path, rc = -EOVERFLOW);
596 ptr = ori_gf->gf_u.gf_path;
598 len = strlen(gf->gf_u.gf_path);
599 /* move the current path to the right to release space
600 * for closer-to-root part */
601 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
602 memcpy(ptr, gf->gf_u.gf_path, len);
606 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
607 tgt->ltd_exp->exp_obd->obd_name,
608 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
612 GOTO(out_fid2path, rc);
614 /* sigh, has to go to another MDT to do path building further */
615 if (remote_gf == NULL) {
616 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
617 OBD_ALLOC(remote_gf, remote_gf_size);
618 if (remote_gf == NULL)
619 GOTO(out_fid2path, rc = -ENOMEM);
620 remote_gf->gf_pathlen = PATH_MAX;
623 if (!fid_is_sane(&gf->gf_fid)) {
624 CERROR("%s: invalid FID "DFID": rc = %d\n",
625 tgt->ltd_exp->exp_obd->obd_name,
626 PFID(&gf->gf_fid), -EINVAL);
627 GOTO(out_fid2path, rc = -EINVAL);
630 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
632 GOTO(out_fid2path, rc = -EINVAL);
634 remote_gf->gf_fid = gf->gf_fid;
635 remote_gf->gf_recno = -1;
636 remote_gf->gf_linkno = -1;
637 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
638 *remote_gf->gf_u.gf_root_fid = root_fid;
640 goto repeat_fid2path;
643 if (remote_gf != NULL)
644 OBD_FREE(remote_gf, remote_gf_size);
648 static int lmv_hsm_req_count(struct lmv_obd *lmv,
649 const struct hsm_user_request *hur,
650 const struct lmv_tgt_desc *tgt_mds)
652 struct lmv_tgt_desc *curr_tgt;
656 /* count how many requests must be sent to the given target */
657 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
658 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
659 if (IS_ERR(curr_tgt))
660 RETURN(PTR_ERR(curr_tgt));
661 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
667 static int lmv_hsm_req_build(struct lmv_obd *lmv,
668 struct hsm_user_request *hur_in,
669 const struct lmv_tgt_desc *tgt_mds,
670 struct hsm_user_request *hur_out)
673 struct lmv_tgt_desc *curr_tgt;
675 /* build the hsm_user_request for the given target */
676 hur_out->hur_request = hur_in->hur_request;
678 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
679 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
680 if (IS_ERR(curr_tgt))
681 RETURN(PTR_ERR(curr_tgt));
682 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
683 hur_out->hur_user_item[nr_out] =
684 hur_in->hur_user_item[i];
688 hur_out->hur_request.hr_itemcount = nr_out;
689 memcpy(hur_data(hur_out), hur_data(hur_in),
690 hur_in->hur_request.hr_data_len);
695 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
696 int len, struct lustre_kernelcomm *lk,
699 struct lmv_obd *lmv = &obd->u.lmv;
700 struct lu_tgt_desc *tgt;
705 /* unregister request (call from llapi_hsm_copytool_fini) */
706 lmv_foreach_connected_tgt(lmv, tgt)
707 /* best effort: try to clean as much as possible
708 * (continue on error) */
709 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
711 /* Whatever the result, remove copytool from kuc groups.
712 * Unreached coordinators will get EPIPE on next requests
713 * and will unregister automatically.
715 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
720 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
721 int len, struct lustre_kernelcomm *lk,
724 struct lmv_obd *lmv = &obd->u.lmv;
726 bool any_set = false;
727 struct kkuc_ct_data *kcd;
729 struct lu_tgt_desc *tgt;
736 filp = fget(lk->lk_wfd);
740 if (lk->lk_flags & LK_FLG_DATANR)
741 kcd_size = offsetof(struct kkuc_ct_data,
742 kcd_archives[lk->lk_data_count]);
744 kcd_size = sizeof(*kcd);
746 OBD_ALLOC(kcd, kcd_size);
748 GOTO(err_fput, rc = -ENOMEM);
750 kcd->kcd_nr_archives = lk->lk_data_count;
751 if (lk->lk_flags & LK_FLG_DATANR) {
752 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
753 if (lk->lk_data_count > 0)
754 memcpy(kcd->kcd_archives, lk->lk_data,
755 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
757 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
760 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
761 lk->lk_group, kcd, kcd_size);
762 OBD_FREE(kcd, kcd_size);
766 /* All or nothing: try to register to all MDS.
767 * In case of failure, unregister from previous MDS,
768 * except if it because of inactive target. */
769 lmv_foreach_connected_tgt(lmv, tgt) {
770 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
772 if (tgt->ltd_active) {
773 /* permanent error */
774 CERROR("%s: iocontrol MDC %s on MDT"
775 " idx %d cmd %x: err = %d\n",
776 lmv2obd_dev(lmv)->obd_name,
777 tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
780 lk->lk_flags |= LK_FLG_STOP;
782 /* unregister from previous MDS */
783 lmv_foreach_connected_tgt(lmv, tgt) {
784 if (tgt->ltd_index >= i)
787 obd_iocontrol(cmd, tgt->ltd_exp, len,
790 GOTO(err_kkuc_rem, rc);
792 /* else: transient error.
793 * kuc will register to the missing MDT
801 /* no registration done: return error */
802 GOTO(err_kkuc_rem, rc = -ENOTCONN);
807 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
814 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
815 int len, void *karg, void __user *uarg)
817 struct obd_device *obddev = class_exp2obd(exp);
818 struct lmv_obd *lmv = &obddev->u.lmv;
819 struct lu_tgt_desc *tgt = NULL;
821 __u32 count = lmv->lmv_mdt_count;
830 case IOC_OBD_STATFS: {
831 struct obd_ioctl_data *data = karg;
832 struct obd_device *mdc_obd;
833 struct obd_statfs stat_buf = {0};
836 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
838 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
841 tgt = lmv_tgt(lmv, index);
842 if (!tgt || !tgt->ltd_active)
845 mdc_obd = class_exp2obd(tgt->ltd_exp);
850 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
851 min((int) data->ioc_plen2,
852 (int) sizeof(struct obd_uuid))))
855 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
856 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
860 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
861 min((int) data->ioc_plen1,
862 (int) sizeof(stat_buf))))
866 case OBD_IOC_QUOTACTL: {
867 struct if_quotactl *qctl = karg;
868 struct obd_quotactl *oqctl;
870 if (qctl->qc_valid == QC_MDTIDX) {
871 tgt = lmv_tgt(lmv, qctl->qc_idx);
872 } else if (qctl->qc_valid == QC_UUID) {
873 lmv_foreach_tgt(lmv, tgt) {
874 if (!obd_uuid_equals(&tgt->ltd_uuid,
887 if (!tgt || !tgt->ltd_exp)
890 OBD_ALLOC_PTR(oqctl);
894 QCTL_COPY(oqctl, qctl);
895 rc = obd_quotactl(tgt->ltd_exp, oqctl);
897 QCTL_COPY(qctl, oqctl);
898 qctl->qc_valid = QC_MDTIDX;
899 qctl->obd_uuid = tgt->ltd_uuid;
904 case LL_IOC_GET_CONNECT_FLAGS: {
905 tgt = lmv_tgt(lmv, 0);
907 if (tgt && tgt->ltd_exp)
908 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
911 case LL_IOC_FID2MDTIDX: {
912 struct lu_fid *fid = karg;
915 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
919 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
920 * point to user space memory for FID2MDTIDX. */
921 *(__u32 *)uarg = mdt_index;
924 case OBD_IOC_FID2PATH: {
925 rc = lmv_fid2path(exp, len, karg, uarg);
928 case LL_IOC_HSM_STATE_GET:
929 case LL_IOC_HSM_STATE_SET:
930 case LL_IOC_HSM_ACTION: {
931 struct md_op_data *op_data = karg;
933 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
935 RETURN(PTR_ERR(tgt));
937 if (tgt->ltd_exp == NULL)
940 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
943 case LL_IOC_HSM_PROGRESS: {
944 const struct hsm_progress_kernel *hpk = karg;
946 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
948 RETURN(PTR_ERR(tgt));
949 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
952 case LL_IOC_HSM_REQUEST: {
953 struct hsm_user_request *hur = karg;
954 unsigned int reqcount = hur->hur_request.hr_itemcount;
959 /* if the request is about a single fid
960 * or if there is a single MDS, no need to split
962 if (reqcount == 1 || count == 1) {
963 tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
965 RETURN(PTR_ERR(tgt));
966 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
968 /* split fid list to their respective MDS */
969 lmv_foreach_connected_tgt(lmv, tgt) {
972 struct hsm_user_request *req;
974 nr = lmv_hsm_req_count(lmv, hur, tgt);
977 if (nr == 0) /* nothing for this MDS */
980 /* build a request with fids for this MDS */
981 reqlen = offsetof(typeof(*hur),
983 + hur->hur_request.hr_data_len;
984 OBD_ALLOC_LARGE(req, reqlen);
987 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
989 GOTO(hsm_req_err, rc1);
990 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
993 if (rc1 != 0 && rc == 0)
995 OBD_FREE_LARGE(req, reqlen);
1000 case LL_IOC_LOV_SWAP_LAYOUTS: {
1001 struct md_op_data *op_data = karg;
1002 struct lmv_tgt_desc *tgt1, *tgt2;
1004 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1006 RETURN(PTR_ERR(tgt1));
1008 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1010 RETURN(PTR_ERR(tgt2));
1012 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1015 /* only files on same MDT can have their layouts swapped */
1016 if (tgt1->ltd_index != tgt2->ltd_index)
1019 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1022 case LL_IOC_HSM_CT_START: {
1023 struct lustre_kernelcomm *lk = karg;
1024 if (lk->lk_flags & LK_FLG_STOP)
1025 rc = lmv_hsm_ct_unregister(obddev, cmd, len, lk, uarg);
1027 rc = lmv_hsm_ct_register(obddev, cmd, len, lk, uarg);
1031 lmv_foreach_connected_tgt(lmv, tgt) {
1032 struct obd_device *mdc_obd;
1035 /* ll_umount_begin() sets force flag but for lmv, not
1036 * mdc. Let's pass it through */
1037 mdc_obd = class_exp2obd(tgt->ltd_exp);
1038 mdc_obd->obd_force = obddev->obd_force;
1039 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1041 if (tgt->ltd_active) {
1042 CERROR("error: iocontrol MDC %s on MDT"
1043 " idx %d cmd %x: err = %d\n",
1045 tgt->ltd_index, cmd, err);
1058 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1059 struct lu_fid *fid, struct md_op_data *op_data)
1061 struct obd_device *obd = class_exp2obd(exp);
1062 struct lmv_obd *lmv = &obd->u.lmv;
1063 struct lmv_tgt_desc *tgt;
1071 tgt = lmv_tgt(lmv, op_data->op_mds);
1075 if (!tgt->ltd_active || !tgt->ltd_exp)
1079 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1080 * on server that seq in new allocated fid is not yet known.
1082 mutex_lock(&tgt->ltd_fid_mutex);
1083 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1084 mutex_unlock(&tgt->ltd_fid_mutex);
1086 LASSERT(fid_is_sane(fid));
1093 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1095 struct lmv_obd *lmv = &obd->u.lmv;
1096 struct lmv_desc *desc;
1097 struct lnet_process_id lnet_id;
1103 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1104 CERROR("LMV setup requires a descriptor\n");
1108 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1109 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1110 CERROR("Lmv descriptor size wrong: %d > %d\n",
1111 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1115 obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1116 desc->ld_uuid.uuid);
1117 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1118 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1119 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1120 LMV_DESC_QOS_MAXAGE_DEFAULT;
1121 lmv->max_def_easize = 0;
1122 lmv->max_easize = 0;
1124 spin_lock_init(&lmv->lmv_lock);
1127 * initialize rr_index to lower 32bit of netid, so that client
1128 * can distribute subdirs evenly from the beginning.
1130 while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1131 if (LNET_NETTYP(LNET_NIDNET(lnet_id.nid)) != LOLND) {
1132 lmv->lmv_qos_rr_index = (u32)lnet_id.nid;
1137 rc = lmv_tunables_init(obd);
1139 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1142 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1143 LUSTRE_CLI_FLD_HASH_DHT);
1145 CERROR("Can't init FLD, err %d\n", rc);
1147 rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1149 CWARN("%s: error initialize target table: rc = %d\n",
1155 static int lmv_cleanup(struct obd_device *obd)
1157 struct lmv_obd *lmv = &obd->u.lmv;
1158 struct lu_tgt_desc *tgt;
1159 struct lu_tgt_desc *tmp;
1163 fld_client_fini(&lmv->lmv_fld);
1164 lmv_foreach_tgt_safe(lmv, tgt, tmp)
1165 lmv_del_target(lmv, tgt);
1166 lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1171 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1173 struct lustre_cfg *lcfg = buf;
1174 struct obd_uuid obd_uuid;
1180 switch (lcfg->lcfg_command) {
1182 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1183 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1184 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1185 GOTO(out, rc = -EINVAL);
1187 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1189 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1190 GOTO(out, rc = -EINVAL);
1191 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1192 GOTO(out, rc = -EINVAL);
1193 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1196 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1197 GOTO(out, rc = -EINVAL);
1203 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1207 if (flags & OBD_STATFS_FOR_MDT0)
1210 if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1211 return lmv->lmv_statfs_start;
1213 /* choose initial MDT for this client */
1215 struct lnet_process_id lnet_id;
1216 if (LNetGetId(i, &lnet_id) == -ENOENT)
1219 if (LNET_NETTYP(LNET_NIDNET(lnet_id.nid)) != LOLND) {
1220 /* We dont need a full 64-bit modulus, just enough
1221 * to distribute the requests across MDTs evenly.
1223 lmv->lmv_statfs_start = (u32)lnet_id.nid %
1229 return lmv->lmv_statfs_start;
1232 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1233 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1235 struct obd_device *obd = class_exp2obd(exp);
1236 struct lmv_obd *lmv = &obd->u.lmv;
1237 struct obd_statfs *temp;
1238 struct lu_tgt_desc *tgt;
1245 OBD_ALLOC(temp, sizeof(*temp));
1249 /* distribute statfs among MDTs */
1250 idx = lmv_select_statfs_mdt(lmv, flags);
1252 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1253 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1254 tgt = lmv_tgt(lmv, idx);
1255 if (!tgt || !tgt->ltd_exp)
1258 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age, flags);
1260 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1261 tgt->ltd_exp->exp_obd->obd_name, i, rc);
1262 GOTO(out_free_temp, rc);
1265 if (temp->os_state & OS_STATE_SUM ||
1266 flags == OBD_STATFS_FOR_MDT0) {
1267 /* reset to the last aggregated values
1268 * and don't sum with non-aggrated data */
1269 /* If the statfs is from mount, it needs to retrieve
1270 * necessary information from MDT0. i.e. mount does
1271 * not need the merged osfs from all of MDT. Also
1272 * clients can be mounted as long as MDT0 is in
1281 osfs->os_bavail += temp->os_bavail;
1282 osfs->os_blocks += temp->os_blocks;
1283 osfs->os_ffree += temp->os_ffree;
1284 osfs->os_files += temp->os_files;
1285 osfs->os_granted += temp->os_granted;
1291 OBD_FREE(temp, sizeof(*temp));
1295 static int lmv_statfs_update(void *cookie, int rc)
1297 struct obd_info *oinfo = cookie;
1298 struct obd_device *obd = oinfo->oi_obd;
1299 struct lmv_obd *lmv = &obd->u.lmv;
1300 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1301 struct obd_statfs *osfs = oinfo->oi_osfs;
1304 * NB: don't deactivate TGT upon error, because we may not trigger async
1305 * statfs any longer, then there is no chance to activate TGT.
1308 spin_lock(&lmv->lmv_lock);
1309 tgt->ltd_statfs = *osfs;
1310 tgt->ltd_statfs_age = ktime_get_seconds();
1311 spin_unlock(&lmv->lmv_lock);
1312 lmv->lmv_qos.lq_dirty = 1;
1318 /* update tgt statfs async if it's ld_qos_maxage old */
1319 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1321 struct obd_info oinfo = {
1324 .oi_cb_up = lmv_statfs_update,
1328 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1329 obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1332 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1337 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1340 struct obd_device *obd = exp->exp_obd;
1341 struct lmv_obd *lmv = &obd->u.lmv;
1342 struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1350 rc = md_get_root(tgt->ltd_exp, fileset, fid);
1354 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1355 u64 obd_md_valid, const char *name, size_t buf_size,
1356 struct ptlrpc_request **req)
1358 struct obd_device *obd = exp->exp_obd;
1359 struct lmv_obd *lmv = &obd->u.lmv;
1360 struct lmv_tgt_desc *tgt;
1365 tgt = lmv_fid2tgt(lmv, fid);
1367 RETURN(PTR_ERR(tgt));
1369 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1374 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1375 u64 obd_md_valid, const char *name,
1376 const void *value, size_t value_size,
1377 unsigned int xattr_flags, u32 suppgid,
1378 struct ptlrpc_request **req)
1380 struct obd_device *obd = exp->exp_obd;
1381 struct lmv_obd *lmv = &obd->u.lmv;
1382 struct lmv_tgt_desc *tgt;
1387 tgt = lmv_fid2tgt(lmv, fid);
1389 RETURN(PTR_ERR(tgt));
1391 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1392 value, value_size, xattr_flags, suppgid, req);
1397 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1398 struct ptlrpc_request **request)
1400 struct obd_device *obd = exp->exp_obd;
1401 struct lmv_obd *lmv = &obd->u.lmv;
1402 struct lmv_tgt_desc *tgt;
1407 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1409 RETURN(PTR_ERR(tgt));
1411 if (op_data->op_flags & MF_GET_MDT_IDX) {
1412 op_data->op_mds = tgt->ltd_index;
1416 rc = md_getattr(tgt->ltd_exp, op_data, request);
1421 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1423 struct obd_device *obd = exp->exp_obd;
1424 struct lmv_obd *lmv = &obd->u.lmv;
1425 struct lu_tgt_desc *tgt;
1429 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1432 * With DNE every object can have two locks in different namespaces:
1433 * lookup lock in space of MDT storing direntry and update/open lock in
1434 * space of MDT storing inode.
1436 lmv_foreach_connected_tgt(lmv, tgt)
1437 md_null_inode(tgt->ltd_exp, fid);
1442 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1443 struct md_open_data *mod, struct ptlrpc_request **request)
1445 struct obd_device *obd = exp->exp_obd;
1446 struct lmv_obd *lmv = &obd->u.lmv;
1447 struct lmv_tgt_desc *tgt;
1452 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1454 RETURN(PTR_ERR(tgt));
1456 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1457 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1461 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv, __u32 *mdt)
1463 struct lu_tgt_desc *tgt;
1464 __u64 total_weight = 0;
1465 __u64 cur_weight = 0;
1471 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1472 RETURN(ERR_PTR(-EAGAIN));
1474 down_write(&lmv->lmv_qos.lq_rw_sem);
1476 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1477 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1479 rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1481 GOTO(unlock, tgt = ERR_PTR(rc));
1483 lmv_foreach_tgt(lmv, tgt) {
1484 tgt->ltd_qos.ltq_usable = 0;
1485 if (!tgt->ltd_exp || !tgt->ltd_active)
1488 tgt->ltd_qos.ltq_usable = 1;
1489 lu_tgt_qos_weight_calc(tgt);
1490 total_weight += tgt->ltd_qos.ltq_weight;
1493 rand = lu_prandom_u64_max(total_weight);
1495 lmv_foreach_connected_tgt(lmv, tgt) {
1496 if (!tgt->ltd_qos.ltq_usable)
1499 cur_weight += tgt->ltd_qos.ltq_weight;
1500 if (cur_weight < rand)
1503 *mdt = tgt->ltd_index;
1504 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1505 GOTO(unlock, rc = 0);
1508 /* no proper target found */
1509 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1511 up_write(&lmv->lmv_qos.lq_rw_sem);
1516 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv, __u32 *mdt)
1518 struct lu_tgt_desc *tgt;
1524 spin_lock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1525 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1526 index = (i + lmv->lmv_qos_rr_index) %
1527 lmv->lmv_mdt_descs.ltd_tgts_size;
1528 tgt = lmv_tgt(lmv, index);
1529 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1532 *mdt = tgt->ltd_index;
1533 lmv->lmv_qos_rr_index = (*mdt + 1) %
1534 lmv->lmv_mdt_descs.ltd_tgts_size;
1535 spin_unlock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1539 spin_unlock(&lmv->lmv_qos.lq_rr.lqr_alloc);
1541 RETURN(ERR_PTR(-ENODEV));
1544 static struct lmv_tgt_desc *
1545 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1546 const char *name, int namelen, struct lu_fid *fid,
1547 __u32 *mds, bool post_migrate)
1549 struct lmv_tgt_desc *tgt;
1550 const struct lmv_oinfo *oinfo;
1552 if (!lmv_dir_striped(lsm) || !namelen) {
1553 tgt = lmv_fid2tgt(lmv, fid);
1557 *mds = tgt->ltd_index;
1561 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1562 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1563 return ERR_PTR(-EBADF);
1564 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1566 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
1569 return ERR_CAST(oinfo);
1572 *fid = oinfo->lmo_fid;
1573 *mds = oinfo->lmo_mds;
1574 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1576 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1578 return tgt ? tgt : ERR_PTR(-ENODEV);
1582 * Locate MDT of op_data->op_fid1
1584 * For striped directory, it will locate the stripe by name hash, if hash_type
1585 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1586 * which is set outside, and if dir is migrating, 'op_data->op_post_migrate'
1587 * indicates whether old or new layout is used to locate.
1589 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1591 * \param[in] lmv LMV device
1592 * \param[in] op_data client MD stack parameters, name, namelen
1595 * retval pointer to the lmv_tgt_desc if succeed.
1596 * ERR_PTR(errno) if failed.
1598 struct lmv_tgt_desc *
1599 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1601 struct lmv_stripe_md *lsm = op_data->op_mea1;
1602 struct lmv_oinfo *oinfo;
1603 struct lmv_tgt_desc *tgt;
1605 if (lmv_dir_foreign(lsm))
1606 return ERR_PTR(-ENODATA);
1608 /* During creating VOLATILE file, it should honor the mdt
1609 * index if the file under striped dir is being restored, see
1611 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1612 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1613 tgt = lmv_tgt(lmv, op_data->op_mds);
1615 return ERR_PTR(-ENODEV);
1617 if (lmv_dir_striped(lsm)) {
1620 /* refill the right parent fid */
1621 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1622 oinfo = &lsm->lsm_md_oinfo[i];
1623 if (oinfo->lmo_mds == op_data->op_mds) {
1624 op_data->op_fid1 = oinfo->lmo_fid;
1629 if (i == lsm->lsm_md_stripe_count)
1630 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1632 } else if (lmv_dir_bad_hash(lsm)) {
1633 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1634 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1636 op_data->op_fid1 = oinfo->lmo_fid;
1637 op_data->op_mds = oinfo->lmo_mds;
1638 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1640 return ERR_PTR(-ENODEV);
1642 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1643 op_data->op_name, op_data->op_namelen,
1644 &op_data->op_fid1, &op_data->op_mds,
1645 op_data->op_post_migrate);
1651 /* Locate MDT of op_data->op_fid2 for link/rename */
1652 static struct lmv_tgt_desc *
1653 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1655 struct lmv_tgt_desc *tgt;
1658 LASSERT(op_data->op_name);
1659 if (lmv_dir_migrating(op_data->op_mea2)) {
1660 struct lu_fid fid1 = op_data->op_fid1;
1661 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1662 struct ptlrpc_request *request = NULL;
1665 * avoid creating new file under old layout of migrating
1666 * directory, check it here.
1668 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1669 op_data->op_name, op_data->op_namelen,
1670 &op_data->op_fid2, &op_data->op_mds, false);
1674 op_data->op_fid1 = op_data->op_fid2;
1675 op_data->op_mea1 = op_data->op_mea2;
1676 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1677 op_data->op_fid1 = fid1;
1678 op_data->op_mea1 = lsm1;
1680 ptlrpc_req_finished(request);
1681 RETURN(ERR_PTR(-EEXIST));
1685 RETURN(ERR_PTR(rc));
1688 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1689 op_data->op_namelen, &op_data->op_fid2,
1690 &op_data->op_mds, true);
1693 int lmv_migrate_existence_check(struct lmv_obd *lmv, struct md_op_data *op_data)
1695 struct lu_tgt_desc *tgt;
1696 struct ptlrpc_request *request;
1699 LASSERT(lmv_dir_migrating(op_data->op_mea1));
1701 tgt = lmv_locate_tgt(lmv, op_data);
1703 return PTR_ERR(tgt);
1705 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1707 ptlrpc_req_finished(request);
1714 /* mkdir by QoS in two cases:
1715 * 1. 'lfs mkdir -i -1'
1716 * 2. parent default LMV master_mdt_index is -1
1718 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1719 * directories under striped directory.
1721 static inline bool lmv_op_qos_mkdir(const struct md_op_data *op_data)
1723 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1724 const struct lmv_user_md *lum = op_data->op_data;
1726 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1729 if (lmv_dir_striped(op_data->op_mea1))
1732 if (op_data->op_cli_flags & CLI_SET_MEA && lum &&
1733 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1734 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1735 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT)
1738 if (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT)
1744 /* 'lfs mkdir -i <specific_MDT>' */
1745 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1747 const struct lmv_user_md *lum = op_data->op_data;
1749 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1750 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1751 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1752 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1753 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1756 /* parent default LMV master_mdt_index is not -1. */
1758 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1760 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1761 op_data->op_default_mea1 &&
1762 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1765 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1766 const void *data, size_t datalen, umode_t mode, uid_t uid,
1767 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1768 struct ptlrpc_request **request)
1770 struct obd_device *obd = exp->exp_obd;
1771 struct lmv_obd *lmv = &obd->u.lmv;
1772 struct lmv_tgt_desc *tgt;
1777 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1780 if (lmv_dir_bad_hash(op_data->op_mea1))
1783 if (lmv_dir_migrating(op_data->op_mea1)) {
1785 * if parent is migrating, create() needs to lookup existing
1786 * name in both old and new layout, check old layout on client.
1788 rc = lmv_migrate_existence_check(lmv, op_data);
1792 op_data->op_post_migrate = true;
1795 tgt = lmv_locate_tgt(lmv, op_data);
1797 RETURN(PTR_ERR(tgt));
1799 if (lmv_op_qos_mkdir(op_data)) {
1800 tgt = lmv_locate_tgt_qos(lmv, &op_data->op_mds);
1801 if (tgt == ERR_PTR(-EAGAIN))
1802 tgt = lmv_locate_tgt_rr(lmv, &op_data->op_mds);
1804 * only update statfs after QoS mkdir, this means the cached
1805 * statfs may be stale, and current mkdir may not follow QoS
1806 * accurately, but it's not serious, and avoids periodic statfs
1807 * when client doesn't mkdir by QoS.
1810 lmv_statfs_check_update(obd, tgt);
1811 } else if (lmv_op_user_specific_mkdir(op_data)) {
1812 struct lmv_user_md *lum = op_data->op_data;
1814 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
1815 tgt = lmv_tgt(lmv, op_data->op_mds);
1818 } else if (lmv_op_default_specific_mkdir(op_data)) {
1820 op_data->op_default_mea1->lsm_md_master_mdt_index;
1821 tgt = lmv_tgt(lmv, op_data->op_mds);
1827 RETURN(PTR_ERR(tgt));
1829 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1833 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
1834 (int)op_data->op_namelen, op_data->op_name,
1835 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
1838 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1839 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1840 cap_effective, rdev, request);
1842 if (*request == NULL)
1844 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1850 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1851 const union ldlm_policy_data *policy, struct md_op_data *op_data,
1852 struct lustre_handle *lockh, __u64 extra_lock_flags)
1854 struct obd_device *obd = exp->exp_obd;
1855 struct lmv_obd *lmv = &obd->u.lmv;
1856 struct lmv_tgt_desc *tgt;
1861 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1863 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1865 RETURN(PTR_ERR(tgt));
1867 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1868 PFID(&op_data->op_fid1), tgt->ltd_index);
1870 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1877 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1878 struct ptlrpc_request **preq)
1880 struct obd_device *obd = exp->exp_obd;
1881 struct lmv_obd *lmv = &obd->u.lmv;
1882 struct lmv_tgt_desc *tgt;
1883 struct mdt_body *body;
1889 tgt = lmv_locate_tgt(lmv, op_data);
1891 RETURN(PTR_ERR(tgt));
1893 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1894 (int)op_data->op_namelen, op_data->op_name,
1895 PFID(&op_data->op_fid1), tgt->ltd_index);
1897 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1898 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
1899 ptlrpc_req_finished(*preq);
1907 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1908 LASSERT(body != NULL);
1910 if (body->mbo_valid & OBD_MD_MDS) {
1911 op_data->op_fid1 = body->mbo_fid1;
1912 op_data->op_valid |= OBD_MD_FLCROSSREF;
1913 op_data->op_namelen = 0;
1914 op_data->op_name = NULL;
1916 ptlrpc_req_finished(*preq);
1925 #define md_op_data_fid(op_data, fl) \
1926 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1927 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1928 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1929 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1932 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1933 struct md_op_data *op_data, __u32 op_tgt,
1934 enum ldlm_mode mode, int bits, int flag)
1936 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1937 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1938 union ldlm_policy_data policy = { { 0 } };
1942 if (!fid_is_sane(fid))
1946 tgt = lmv_fid2tgt(lmv, fid);
1948 RETURN(PTR_ERR(tgt));
1951 if (tgt->ltd_index != op_tgt) {
1952 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1953 policy.l_inodebits.bits = bits;
1954 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1955 mode, LCF_ASYNC, NULL);
1958 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1960 op_data->op_flags |= flag;
1968 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1971 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1972 struct ptlrpc_request **request)
1974 struct obd_device *obd = exp->exp_obd;
1975 struct lmv_obd *lmv = &obd->u.lmv;
1976 struct lmv_tgt_desc *tgt;
1980 LASSERT(op_data->op_namelen != 0);
1982 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1983 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1984 op_data->op_name, PFID(&op_data->op_fid1));
1986 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1987 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1988 op_data->op_cap = cfs_curproc_cap_pack();
1990 tgt = lmv_locate_tgt2(lmv, op_data);
1992 RETURN(PTR_ERR(tgt));
1995 * Cancel UPDATE lock on child (fid1).
1997 op_data->op_flags |= MF_MDC_CANCEL_FID2;
1998 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
1999 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2003 rc = md_link(tgt->ltd_exp, op_data, request);
2008 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2009 const char *name, size_t namelen,
2010 struct ptlrpc_request **request)
2012 struct obd_device *obd = exp->exp_obd;
2013 struct lmv_obd *lmv = &obd->u.lmv;
2014 struct lmv_stripe_md *lsm = op_data->op_mea1;
2015 struct lmv_tgt_desc *parent_tgt;
2016 struct lmv_tgt_desc *sp_tgt;
2017 struct lmv_tgt_desc *tp_tgt = NULL;
2018 struct lmv_tgt_desc *child_tgt;
2019 struct lmv_tgt_desc *tgt;
2020 struct lu_fid target_fid;
2025 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2027 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2028 PFID(&op_data->op_fid1), (int)namelen, name);
2030 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2031 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2032 op_data->op_cap = cfs_curproc_cap_pack();
2034 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2035 if (IS_ERR(parent_tgt))
2036 RETURN(PTR_ERR(parent_tgt));
2038 if (lmv_dir_striped(lsm)) {
2039 __u32 hash_type = lsm->lsm_md_hash_type;
2040 __u32 stripe_count = lsm->lsm_md_stripe_count;
2043 * old stripes are appended after new stripes for migrating
2046 if (lmv_dir_migrating(lsm)) {
2047 hash_type = lsm->lsm_md_migrate_hash;
2048 stripe_count -= lsm->lsm_md_migrate_offset;
2051 rc = lmv_name_to_stripe_index(hash_type, stripe_count, name,
2056 if (lmv_dir_migrating(lsm))
2057 rc += lsm->lsm_md_migrate_offset;
2059 /* save it in fid4 temporarily for early cancel */
2060 op_data->op_fid4 = lsm->lsm_md_oinfo[rc].lmo_fid;
2061 sp_tgt = lmv_tgt(lmv, lsm->lsm_md_oinfo[rc].lmo_mds);
2066 * if parent is being migrated too, fill op_fid2 with target
2067 * stripe fid, otherwise the target stripe is not created yet.
2069 if (lmv_dir_migrating(lsm)) {
2070 hash_type = lsm->lsm_md_hash_type &
2071 ~LMV_HASH_FLAG_MIGRATION;
2072 stripe_count = lsm->lsm_md_migrate_offset;
2074 rc = lmv_name_to_stripe_index(hash_type, stripe_count,
2079 op_data->op_fid2 = lsm->lsm_md_oinfo[rc].lmo_fid;
2080 tp_tgt = lmv_tgt(lmv, lsm->lsm_md_oinfo[rc].lmo_mds);
2085 sp_tgt = parent_tgt;
2088 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2089 if (IS_ERR(child_tgt))
2090 RETURN(PTR_ERR(child_tgt));
2092 /* for directory, migrate to MDT specified by lum_stripe_offset;
2093 * otherwise migrate to the target stripe of parent, but parent
2094 * directory may have finished migration (normally current file too),
2095 * allocate FID on MDT lum_stripe_offset, and server will check
2096 * whether file was migrated already.
2098 if (S_ISDIR(op_data->op_mode) || !tp_tgt) {
2099 struct lmv_user_md *lum = op_data->op_data;
2101 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2103 op_data->op_mds = tp_tgt->ltd_index;
2105 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2110 * for directory, send migrate request to the MDT where the object will
2111 * be migrated to, because we can't create a striped directory remotely.
2113 * otherwise, send to the MDT where source is located because regular
2114 * file may open lease.
2116 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2117 * backward compatibility.
2119 if (S_ISDIR(op_data->op_mode) &&
2120 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2121 tgt = lmv_fid2tgt(lmv, &target_fid);
2123 RETURN(PTR_ERR(tgt));
2128 /* cancel UPDATE lock of parent master object */
2129 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2130 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2134 /* cancel UPDATE lock of source parent */
2135 if (sp_tgt != parent_tgt) {
2137 * migrate RPC packs master object FID, because we can only pack
2138 * two FIDs in reint RPC, but MDS needs to know both source
2139 * parent and target parent, and it will obtain them from master
2140 * FID and LMV, the other FID in RPC is kept for target.
2142 * since this FID is not passed to MDC, cancel it anyway.
2144 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2145 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2149 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2151 op_data->op_fid4 = target_fid;
2153 /* cancel UPDATE locks of target parent */
2154 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2155 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2159 /* cancel LOOKUP lock of source if source is remote object */
2160 if (child_tgt != sp_tgt) {
2161 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2162 LCK_EX, MDS_INODELOCK_LOOKUP,
2163 MF_MDC_CANCEL_FID3);
2168 /* cancel ELC locks of source */
2169 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2170 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2174 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2179 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2180 const char *old, size_t oldlen,
2181 const char *new, size_t newlen,
2182 struct ptlrpc_request **request)
2184 struct obd_device *obd = exp->exp_obd;
2185 struct lmv_obd *lmv = &obd->u.lmv;
2186 struct lmv_tgt_desc *sp_tgt;
2187 struct lmv_tgt_desc *tp_tgt = NULL;
2188 struct lmv_tgt_desc *src_tgt = NULL;
2189 struct lmv_tgt_desc *tgt;
2190 struct mdt_body *body;
2195 LASSERT(oldlen != 0);
2197 if (op_data->op_cli_flags & CLI_MIGRATE) {
2198 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2202 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2203 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2204 op_data->op_cap = cfs_curproc_cap_pack();
2206 op_data->op_name = new;
2207 op_data->op_namelen = newlen;
2209 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2211 RETURN(PTR_ERR(tp_tgt));
2213 /* Since the target child might be destroyed, and it might become
2214 * orphan, and we can only check orphan on the local MDT right now, so
2215 * we send rename request to the MDT where target child is located. If
2216 * target child does not exist, then it will send the request to the
2218 if (fid_is_sane(&op_data->op_fid4)) {
2219 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2221 RETURN(PTR_ERR(tgt));
2226 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2228 /* cancel UPDATE locks of target parent */
2229 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2230 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2234 if (fid_is_sane(&op_data->op_fid4)) {
2235 /* cancel LOOKUP lock of target on target parent */
2236 if (tgt != tp_tgt) {
2237 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2238 tgt->ltd_index, LCK_EX,
2239 MDS_INODELOCK_LOOKUP,
2240 MF_MDC_CANCEL_FID4);
2246 if (fid_is_sane(&op_data->op_fid3)) {
2247 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2248 if (IS_ERR(src_tgt))
2249 RETURN(PTR_ERR(src_tgt));
2251 /* cancel ELC locks of source */
2252 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2253 LCK_EX, MDS_INODELOCK_ELC,
2254 MF_MDC_CANCEL_FID3);
2259 op_data->op_name = old;
2260 op_data->op_namelen = oldlen;
2262 sp_tgt = lmv_locate_tgt(lmv, op_data);
2264 RETURN(PTR_ERR(sp_tgt));
2266 /* cancel UPDATE locks of source parent */
2267 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2268 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2272 if (fid_is_sane(&op_data->op_fid3)) {
2273 /* cancel LOOKUP lock of source on source parent */
2274 if (src_tgt != sp_tgt) {
2275 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2276 tgt->ltd_index, LCK_EX,
2277 MDS_INODELOCK_LOOKUP,
2278 MF_MDC_CANCEL_FID3);
2285 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2286 PFID(&op_data->op_fid1), (int)oldlen, old,
2287 PFID(&op_data->op_fid2), (int)newlen, new);
2289 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2291 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2292 ptlrpc_req_finished(*request);
2297 if (rc && rc != -EXDEV)
2300 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2304 /* Not cross-ref case, just get out of here. */
2305 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2308 op_data->op_fid4 = body->mbo_fid1;
2310 ptlrpc_req_finished(*request);
2313 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2315 RETURN(PTR_ERR(tgt));
2317 if (fid_is_sane(&op_data->op_fid4)) {
2318 /* cancel LOOKUP lock of target on target parent */
2319 if (tgt != tp_tgt) {
2320 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2321 tgt->ltd_index, LCK_EX,
2322 MDS_INODELOCK_LOOKUP,
2323 MF_MDC_CANCEL_FID4);
2332 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2333 void *ea, size_t ealen, struct ptlrpc_request **request)
2335 struct obd_device *obd = exp->exp_obd;
2336 struct lmv_obd *lmv = &obd->u.lmv;
2337 struct lmv_tgt_desc *tgt;
2342 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2343 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2344 op_data->op_xvalid);
2346 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2347 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2349 RETURN(PTR_ERR(tgt));
2351 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2356 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2357 struct ptlrpc_request **request)
2359 struct obd_device *obd = exp->exp_obd;
2360 struct lmv_obd *lmv = &obd->u.lmv;
2361 struct lmv_tgt_desc *tgt;
2366 tgt = lmv_fid2tgt(lmv, fid);
2368 RETURN(PTR_ERR(tgt));
2370 rc = md_fsync(tgt->ltd_exp, fid, request);
2374 struct stripe_dirent {
2375 struct page *sd_page;
2376 struct lu_dirpage *sd_dp;
2377 struct lu_dirent *sd_ent;
2381 struct lmv_dir_ctxt {
2382 struct lmv_obd *ldc_lmv;
2383 struct md_op_data *ldc_op_data;
2384 struct md_callback *ldc_cb_op;
2387 struct stripe_dirent ldc_stripes[0];
2390 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2392 if (stripe->sd_page) {
2393 kunmap(stripe->sd_page);
2394 put_page(stripe->sd_page);
2395 stripe->sd_page = NULL;
2396 stripe->sd_ent = NULL;
2400 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2404 for (i = 0; i < ctxt->ldc_count; i++)
2405 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2408 /* if @ent is dummy, or . .., get next */
2409 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2410 struct lu_dirent *ent,
2413 for (; ent; ent = lu_dirent_next(ent)) {
2414 /* Skip dummy entry */
2415 if (le16_to_cpu(ent->lde_namelen) == 0)
2418 /* skip . and .. for other stripes */
2420 (strncmp(ent->lde_name, ".",
2421 le16_to_cpu(ent->lde_namelen)) == 0 ||
2422 strncmp(ent->lde_name, "..",
2423 le16_to_cpu(ent->lde_namelen)) == 0))
2426 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2433 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2434 struct stripe_dirent *stripe,
2437 struct md_op_data *op_data = ctxt->ldc_op_data;
2438 struct lmv_oinfo *oinfo;
2439 struct lu_fid fid = op_data->op_fid1;
2440 struct inode *inode = op_data->op_data;
2441 struct lmv_tgt_desc *tgt;
2442 struct lu_dirent *ent = stripe->sd_ent;
2443 __u64 hash = ctxt->ldc_hash;
2448 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2452 if (stripe->sd_page) {
2453 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2455 /* @hash should be the last dirent hash */
2456 LASSERTF(hash <= end,
2457 "ctxt@%p stripe@%p hash %llx end %llx\n",
2458 ctxt, stripe, hash, end);
2459 /* unload last page */
2460 stripe_dirent_unload(stripe);
2462 if (end == MDS_DIR_END_OFF) {
2463 stripe->sd_eof = true;
2469 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2470 if (!oinfo->lmo_root) {
2475 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2481 /* op_data is shared by stripes, reset after use */
2482 op_data->op_fid1 = oinfo->lmo_fid;
2483 op_data->op_fid2 = oinfo->lmo_fid;
2484 op_data->op_data = oinfo->lmo_root;
2486 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_cb_op, hash,
2489 op_data->op_fid1 = fid;
2490 op_data->op_fid2 = fid;
2491 op_data->op_data = inode;
2496 stripe->sd_dp = page_address(stripe->sd_page);
2497 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2499 /* in case a page filled with ., .. and dummy, read next */
2502 stripe->sd_ent = ent;
2505 /* treat error as eof, so dir can be partially accessed */
2506 stripe->sd_eof = true;
2507 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2508 "directory is partially accessed!\n",
2509 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2516 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2518 struct obd_device *obd = exp->exp_obd;
2519 struct lmv_obd *lmv = &obd->u.lmv;
2520 struct lmv_tgt_desc *tgt;
2525 rc = lmv_check_connect(obd);
2529 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2531 RETURN(PTR_ERR(tgt));
2533 data->op_flags |= MF_MDC_CANCEL_FID1;
2534 rc = md_file_resync(tgt->ltd_exp, data);
2539 * Get dirent with the closest hash for striped directory
2541 * This function will search the dir entry, whose hash value is the
2542 * closest(>=) to hash from all of sub-stripes, and it is only being called
2543 * for striped directory.
2545 * \param[in] ctxt dir read context
2547 * \retval dirent get the entry successfully
2548 * NULL does not get the entry, normally it means
2549 * it reaches the end of the directory, while read
2550 * stripe dirent error is ignored to allow partial
2553 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2555 struct stripe_dirent *stripe;
2556 struct lu_dirent *ent = NULL;
2560 /* TODO: optimize with k-way merge sort */
2561 for (i = 0; i < ctxt->ldc_count; i++) {
2562 stripe = &ctxt->ldc_stripes[i];
2566 if (!stripe->sd_ent) {
2567 stripe_dirent_load(ctxt, stripe, i);
2568 if (!stripe->sd_ent) {
2569 LASSERT(stripe->sd_eof);
2575 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2576 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2578 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2585 stripe = &ctxt->ldc_stripes[min];
2586 ent = stripe->sd_ent;
2587 /* pop found dirent */
2588 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2596 * Build dir entry page for striped directory
2598 * This function gets one entry by @offset from a striped directory. It will
2599 * read entries from all of stripes, and choose one closest to the required
2600 * offset(&offset). A few notes
2601 * 1. skip . and .. for non-zero stripes, because there can only have one .
2602 * and .. in a directory.
2603 * 2. op_data will be shared by all of stripes, instead of allocating new
2604 * one, so need to restore before reusing.
2606 * \param[in] exp obd export refer to LMV
2607 * \param[in] op_data hold those MD parameters of read_entry
2608 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2609 * \param[in] offset starting hash offset
2610 * \param[out] ppage the page holding the entry. Note: because the entry
2611 * will be accessed in upper layer, so we need hold the
2612 * page until the usages of entry is finished, see
2613 * ll_dir_entry_next.
2615 * retval =0 if get entry successfully
2616 * <0 cannot get entry
2618 static int lmv_striped_read_page(struct obd_export *exp,
2619 struct md_op_data *op_data,
2620 struct md_callback *cb_op,
2621 __u64 offset, struct page **ppage)
2623 struct page *page = NULL;
2624 struct lu_dirpage *dp;
2626 struct lu_dirent *ent;
2627 struct lu_dirent *last_ent;
2629 struct lmv_dir_ctxt *ctxt;
2630 struct lu_dirent *next = NULL;
2636 /* Allocate a page and read entries from all of stripes and fill
2637 * the page by hash order */
2638 page = alloc_page(GFP_KERNEL);
2642 /* Initialize the entry page */
2644 memset(dp, 0, sizeof(*dp));
2645 dp->ldp_hash_start = cpu_to_le64(offset);
2648 left_bytes = PAGE_SIZE - sizeof(*dp);
2652 /* initalize dir read context */
2653 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2654 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2656 GOTO(free_page, rc = -ENOMEM);
2657 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2658 ctxt->ldc_op_data = op_data;
2659 ctxt->ldc_cb_op = cb_op;
2660 ctxt->ldc_hash = offset;
2661 ctxt->ldc_count = stripe_count;
2664 next = lmv_dirent_next(ctxt);
2666 /* end of directory */
2668 ctxt->ldc_hash = MDS_DIR_END_OFF;
2671 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2673 ent_size = le16_to_cpu(next->lde_reclen);
2675 /* the last entry lde_reclen is 0, but it might not be the last
2676 * one of this temporay dir page */
2678 ent_size = lu_dirent_calc_size(
2679 le16_to_cpu(next->lde_namelen),
2680 le32_to_cpu(next->lde_attrs));
2682 if (ent_size > left_bytes)
2685 memcpy(ent, next, ent_size);
2687 /* Replace . with master FID and Replace .. with the parent FID
2688 * of master object */
2689 if (strncmp(ent->lde_name, ".",
2690 le16_to_cpu(ent->lde_namelen)) == 0 &&
2691 le16_to_cpu(ent->lde_namelen) == 1)
2692 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2693 else if (strncmp(ent->lde_name, "..",
2694 le16_to_cpu(ent->lde_namelen)) == 0 &&
2695 le16_to_cpu(ent->lde_namelen) == 2)
2696 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2698 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2699 le16_to_cpu(ent->lde_namelen), ent->lde_name,
2700 le64_to_cpu(ent->lde_hash));
2702 left_bytes -= ent_size;
2703 ent->lde_reclen = cpu_to_le16(ent_size);
2705 ent = (void *)ent + ent_size;
2708 last_ent->lde_reclen = 0;
2711 dp->ldp_flags |= LDF_EMPTY;
2712 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2713 dp->ldp_flags |= LDF_COLLIDE;
2714 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2715 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2717 put_lmv_dir_ctxt(ctxt);
2718 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2731 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2732 struct md_callback *cb_op, __u64 offset,
2733 struct page **ppage)
2735 struct obd_device *obd = exp->exp_obd;
2736 struct lmv_obd *lmv = &obd->u.lmv;
2737 struct lmv_tgt_desc *tgt;
2742 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
2745 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
2746 rc = lmv_striped_read_page(exp, op_data, cb_op, offset, ppage);
2750 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2752 RETURN(PTR_ERR(tgt));
2754 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2760 * Unlink a file/directory
2762 * Unlink a file or directory under the parent dir. The unlink request
2763 * usually will be sent to the MDT where the child is located, but if
2764 * the client does not have the child FID then request will be sent to the
2765 * MDT where the parent is located.
2767 * If the parent is a striped directory then it also needs to locate which
2768 * stripe the name of the child is located, and replace the parent FID
2769 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2770 * it will walk through all of sub-stripes until the child is being
2773 * \param[in] exp export refer to LMV
2774 * \param[in] op_data different parameters transferred beween client
2775 * MD stacks, name, namelen, FIDs etc.
2776 * op_fid1 is the parent FID, op_fid2 is the child
2778 * \param[out] request point to the request of unlink.
2780 * retval 0 if succeed
2781 * negative errno if failed.
2783 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2784 struct ptlrpc_request **request)
2786 struct obd_device *obd = exp->exp_obd;
2787 struct lmv_obd *lmv = &obd->u.lmv;
2788 struct lmv_tgt_desc *tgt;
2789 struct lmv_tgt_desc *parent_tgt;
2790 struct mdt_body *body;
2795 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2796 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2797 op_data->op_cap = cfs_curproc_cap_pack();
2800 parent_tgt = lmv_locate_tgt(lmv, op_data);
2801 if (IS_ERR(parent_tgt))
2802 RETURN(PTR_ERR(parent_tgt));
2804 if (likely(!fid_is_zero(&op_data->op_fid2))) {
2805 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2807 RETURN(PTR_ERR(tgt));
2813 * If child's fid is given, cancel unused locks for it if it is from
2814 * another export than parent.
2816 * LOOKUP lock for child (fid3) should also be cancelled on parent
2817 * tgt_tgt in mdc_unlink().
2819 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2821 if (parent_tgt != tgt)
2822 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
2823 LCK_EX, MDS_INODELOCK_LOOKUP,
2824 MF_MDC_CANCEL_FID3);
2826 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2827 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2831 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2832 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
2835 rc = md_unlink(tgt->ltd_exp, op_data, request);
2836 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2837 ptlrpc_req_finished(*request);
2845 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2849 /* Not cross-ref case, just get out of here. */
2850 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2853 /* This is a remote object, try remote MDT. */
2854 op_data->op_fid2 = body->mbo_fid1;
2855 ptlrpc_req_finished(*request);
2858 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2860 RETURN(PTR_ERR(tgt));
2865 static int lmv_precleanup(struct obd_device *obd)
2868 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
2869 fld_client_debugfs_fini(&obd->u.lmv.lmv_fld);
2870 lprocfs_obd_cleanup(obd);
2871 lprocfs_free_md_stats(obd);
2876 * Get by key a value associated with a LMV device.
2878 * Dispatch request to lower-layer devices as needed.
2880 * \param[in] env execution environment for this thread
2881 * \param[in] exp export for the LMV device
2882 * \param[in] keylen length of key identifier
2883 * \param[in] key identifier of key to get value for
2884 * \param[in] vallen size of \a val
2885 * \param[out] val pointer to storage location for value
2886 * \param[in] lsm optional striping metadata of object
2888 * \retval 0 on success
2889 * \retval negative negated errno on failure
2891 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2892 __u32 keylen, void *key, __u32 *vallen, void *val)
2894 struct obd_device *obd;
2895 struct lmv_obd *lmv;
2896 struct lu_tgt_desc *tgt;
2901 obd = class_exp2obd(exp);
2903 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2904 exp->exp_handle.h_cookie);
2909 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2910 LASSERT(*vallen == sizeof(__u32));
2911 lmv_foreach_connected_tgt(lmv, tgt) {
2912 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2917 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2918 KEY_IS(KEY_DEFAULT_EASIZE) ||
2919 KEY_IS(KEY_CONN_DATA)) {
2921 * Forwarding this request to first MDS, it should know LOV
2924 tgt = lmv_tgt(lmv, 0);
2928 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
2929 if (!rc && KEY_IS(KEY_CONN_DATA))
2930 exp->exp_connect_data = *(struct obd_connect_data *)val;
2932 } else if (KEY_IS(KEY_TGT_COUNT)) {
2933 *((int *)val) = lmv->lmv_mdt_descs.ltd_tgts_size;
2937 CDEBUG(D_IOCTL, "Invalid key\n");
2941 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
2942 int *__rcs, struct ptlrpc_request_set *_set)
2944 struct obd_device *obddev = class_exp2obd(exp);
2945 struct ptlrpc_request_set *set = _set;
2946 struct lmv_obd *lmv = &obddev->u.lmv;
2947 int tgt_count = lmv->lmv_mdt_count;
2948 struct lu_tgt_desc *tgt;
2949 struct fid_array *fat, **fas = NULL;
2950 int i, rc, **rcs = NULL;
2953 set = ptlrpc_prep_set();
2958 /* split FIDs by targets */
2959 OBD_ALLOC(fas, sizeof(fas) * tgt_count);
2961 GOTO(out, rc = -ENOMEM);
2962 OBD_ALLOC(rcs, sizeof(int *) * tgt_count);
2964 GOTO(out_fas, rc = -ENOMEM);
2966 for (i = 0; i < fa->fa_nr; i++) {
2969 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
2971 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
2972 PFID(&fa->fa_fids[i]), rc);
2975 LASSERT(idx < tgt_count);
2977 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
2978 fa_fids[fa->fa_nr]));
2980 GOTO(out, rc = -ENOMEM);
2982 OBD_ALLOC(rcs[idx], sizeof(int) * fa->fa_nr);
2984 GOTO(out, rc = -ENOMEM);
2987 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
2990 lmv_foreach_connected_tgt(lmv, tgt) {
2991 fat = fas[tgt->ltd_index];
2992 if (!fat || fat->fa_nr == 0)
2994 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
2997 rc = ptlrpc_set_wait(NULL, set);
3000 for (i = 0; i < tgt_count; i++) {
3002 if (!fat || fat->fa_nr == 0)
3004 /* copy FIDs back */
3005 memcpy(fa->fa_fids + j, fat->fa_fids,
3006 fat->fa_nr * sizeof(struct lu_fid));
3008 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3013 ptlrpc_set_destroy(set);
3016 for (i = 0; i < tgt_count; i++) {
3018 OBD_FREE(fas[i], offsetof(struct fid_array,
3019 fa_fids[fa->fa_nr]));
3021 OBD_FREE(rcs[i], sizeof(int) * fa->fa_nr);
3024 OBD_FREE(rcs, sizeof(int *) * tgt_count);
3027 OBD_FREE(fas, sizeof(fas) * tgt_count);
3033 * Asynchronously set by key a value associated with a LMV device.
3035 * Dispatch request to lower-layer devices as needed.
3037 * \param[in] env execution environment for this thread
3038 * \param[in] exp export for the LMV device
3039 * \param[in] keylen length of key identifier
3040 * \param[in] key identifier of key to store value for
3041 * \param[in] vallen size of value to store
3042 * \param[in] val pointer to data to be stored
3043 * \param[in] set optional list of related ptlrpc requests
3045 * \retval 0 on success
3046 * \retval negative negated errno on failure
3048 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3049 __u32 keylen, void *key, __u32 vallen, void *val,
3050 struct ptlrpc_request_set *set)
3052 struct lmv_tgt_desc *tgt = NULL;
3053 struct obd_device *obd;
3054 struct lmv_obd *lmv;
3058 obd = class_exp2obd(exp);
3060 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3061 exp->exp_handle.h_cookie);
3066 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3067 KEY_IS(KEY_DEFAULT_EASIZE)) {
3070 lmv_foreach_connected_tgt(lmv, tgt) {
3071 err = obd_set_info_async(env, tgt->ltd_exp,
3072 keylen, key, vallen, val, set);
3083 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3084 const struct lmv_mds_md_v1 *lmm1)
3086 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3093 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3094 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3095 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3096 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3097 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3099 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3100 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3101 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3102 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3103 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3104 sizeof(lsm->lsm_md_pool_name));
3106 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3109 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %#x "
3110 "layout_version %d\n", lsm->lsm_md_stripe_count,
3111 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3112 lsm->lsm_md_layout_version);
3114 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3115 for (i = 0; i < stripe_count; i++) {
3116 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3117 &lmm1->lmv_stripe_fids[i]);
3119 * set default value -1, so lmv_locate_tgt() knows this stripe
3120 * target is not initialized.
3122 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3123 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3126 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3127 &lsm->lsm_md_oinfo[i].lmo_mds);
3134 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3135 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3141 static inline int lmv_unpack_user_md(struct obd_export *exp,
3142 struct lmv_stripe_md *lsm,
3143 const struct lmv_user_md *lmu)
3145 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3146 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3147 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3148 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3153 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3154 const union lmv_mds_md *lmm, size_t lmm_size)
3156 struct lmv_stripe_md *lsm;
3159 bool allocated = false;
3162 LASSERT(lsmp != NULL);
3166 if (lsm != NULL && lmm == NULL) {
3168 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3170 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3173 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3175 OBD_FREE_LARGE(lfm, lfm_size);
3179 if (lmv_dir_striped(lsm)) {
3180 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3181 if (lsm->lsm_md_oinfo[i].lmo_root)
3182 iput(lsm->lsm_md_oinfo[i].lmo_root);
3184 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3186 lsm_size = lmv_stripe_md_size(0);
3188 OBD_FREE(lsm, lsm_size);
3193 /* foreign lmv case */
3194 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3195 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3198 OBD_ALLOC_LARGE(lfm, lmm_size);
3201 *lsmp = (struct lmv_stripe_md *)lfm;
3203 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3204 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3205 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3206 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3207 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3212 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3216 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3217 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3218 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3219 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3224 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3225 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3228 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3229 * stripecount should be 0 then.
3231 lsm_size = lmv_stripe_md_size(0);
3234 OBD_ALLOC(lsm, lsm_size);
3241 switch (le32_to_cpu(lmm->lmv_magic)) {
3243 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3245 case LMV_USER_MAGIC:
3246 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3249 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3250 le32_to_cpu(lmm->lmv_magic));
3255 if (rc != 0 && allocated) {
3256 OBD_FREE(lsm, lsm_size);
3263 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3265 lmv_unpackmd(NULL, &lsm, NULL, 0);
3267 EXPORT_SYMBOL(lmv_free_memmd);
3269 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3270 union ldlm_policy_data *policy,
3271 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3274 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3275 struct lu_tgt_desc *tgt;
3281 LASSERT(fid != NULL);
3283 lmv_foreach_connected_tgt(lmv, tgt) {
3284 if (!tgt->ltd_active)
3287 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3295 static int lmv_set_lock_data(struct obd_export *exp,
3296 const struct lustre_handle *lockh,
3297 void *data, __u64 *bits)
3299 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3300 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3305 if (tgt == NULL || tgt->ltd_exp == NULL)
3307 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3311 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3312 const struct lu_fid *fid, enum ldlm_type type,
3313 union ldlm_policy_data *policy,
3314 enum ldlm_mode mode, struct lustre_handle *lockh)
3316 struct obd_device *obd = exp->exp_obd;
3317 struct lmv_obd *lmv = &obd->u.lmv;
3319 struct lu_tgt_desc *tgt;
3325 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3328 * With DNE every object can have two locks in different namespaces:
3329 * lookup lock in space of MDT storing direntry and update/open lock in
3330 * space of MDT storing inode. Try the MDT that the FID maps to first,
3331 * since this can be easily found, and only try others if that fails.
3333 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3334 i < lmv->lmv_mdt_descs.ltd_tgts_size;
3335 i++, index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3337 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3338 obd->obd_name, PFID(fid), index);
3342 tgt = lmv_tgt(lmv, index);
3343 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3346 rc = md_lock_match(tgt->ltd_exp, flags, fid, type, policy, mode,
3355 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3356 struct obd_export *dt_exp, struct obd_export *md_exp,
3357 struct lustre_md *md)
3359 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3360 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3362 if (!tgt || !tgt->ltd_exp)
3365 return md_get_lustre_md(tgt->ltd_exp, req, dt_exp, md_exp, md);
3368 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3370 struct obd_device *obd = exp->exp_obd;
3371 struct lmv_obd *lmv = &obd->u.lmv;
3372 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3376 if (md->default_lmv) {
3377 lmv_free_memmd(md->default_lmv);
3378 md->default_lmv = NULL;
3380 if (md->lmv != NULL) {
3381 lmv_free_memmd(md->lmv);
3384 if (!tgt || !tgt->ltd_exp)
3386 RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3389 int lmv_set_open_replay_data(struct obd_export *exp,
3390 struct obd_client_handle *och,
3391 struct lookup_intent *it)
3393 struct obd_device *obd = exp->exp_obd;
3394 struct lmv_obd *lmv = &obd->u.lmv;
3395 struct lmv_tgt_desc *tgt;
3399 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3401 RETURN(PTR_ERR(tgt));
3403 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3406 int lmv_clear_open_replay_data(struct obd_export *exp,
3407 struct obd_client_handle *och)
3409 struct obd_device *obd = exp->exp_obd;
3410 struct lmv_obd *lmv = &obd->u.lmv;
3411 struct lmv_tgt_desc *tgt;
3415 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3417 RETURN(PTR_ERR(tgt));
3419 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3422 int lmv_intent_getattr_async(struct obd_export *exp,
3423 struct md_enqueue_info *minfo)
3425 struct md_op_data *op_data = &minfo->mi_data;
3426 struct obd_device *obd = exp->exp_obd;
3427 struct lmv_obd *lmv = &obd->u.lmv;
3428 struct lmv_tgt_desc *ptgt;
3429 struct lmv_tgt_desc *ctgt;
3434 if (!fid_is_sane(&op_data->op_fid2))
3437 ptgt = lmv_locate_tgt(lmv, op_data);
3439 RETURN(PTR_ERR(ptgt));
3441 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
3443 RETURN(PTR_ERR(ctgt));
3446 * remote object needs two RPCs to lookup and getattr, considering the
3447 * complexity don't support statahead for now.
3452 rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3457 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3458 struct lu_fid *fid, __u64 *bits)
3460 struct obd_device *obd = exp->exp_obd;
3461 struct lmv_obd *lmv = &obd->u.lmv;
3462 struct lmv_tgt_desc *tgt;
3467 tgt = lmv_fid2tgt(lmv, fid);
3469 RETURN(PTR_ERR(tgt));
3471 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3475 int lmv_get_fid_from_lsm(struct obd_export *exp,
3476 const struct lmv_stripe_md *lsm,
3477 const char *name, int namelen, struct lu_fid *fid)
3479 const struct lmv_oinfo *oinfo;
3481 LASSERT(lmv_dir_striped(lsm));
3483 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3485 return PTR_ERR(oinfo);
3487 *fid = oinfo->lmo_fid;
3493 * For lmv, only need to send request to master MDT, and the master MDT will
3494 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3495 * we directly fetch data from the slave MDTs.
3497 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3498 struct obd_quotactl *oqctl)
3500 struct obd_device *obd = class_exp2obd(exp);
3501 struct lmv_obd *lmv = &obd->u.lmv;
3502 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3503 __u64 curspace, curinodes;
3508 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3509 CERROR("master lmv inactive\n");
3513 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3514 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3518 curspace = curinodes = 0;
3519 lmv_foreach_connected_tgt(lmv, tgt) {
3522 if (!tgt->ltd_active)
3525 err = obd_quotactl(tgt->ltd_exp, oqctl);
3527 CERROR("getquota on mdt %d failed. %d\n",
3528 tgt->ltd_index, err);
3532 curspace += oqctl->qc_dqblk.dqb_curspace;
3533 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3536 oqctl->qc_dqblk.dqb_curspace = curspace;
3537 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3542 static int lmv_merge_attr(struct obd_export *exp,
3543 const struct lmv_stripe_md *lsm,
3544 struct cl_attr *attr,
3545 ldlm_blocking_callback cb_blocking)
3550 if (!lmv_dir_striped(lsm))
3553 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3557 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3558 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3564 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3565 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3566 i_size_read(inode), (unsigned long long)inode->i_blocks,
3567 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3568 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3570 /* for slave stripe, it needs to subtract nlink for . and .. */
3572 attr->cat_nlink += inode->i_nlink - 2;
3574 attr->cat_nlink = inode->i_nlink;
3576 attr->cat_size += i_size_read(inode);
3577 attr->cat_blocks += inode->i_blocks;
3579 if (attr->cat_atime < inode->i_atime.tv_sec)
3580 attr->cat_atime = inode->i_atime.tv_sec;
3582 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3583 attr->cat_ctime = inode->i_ctime.tv_sec;
3585 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3586 attr->cat_mtime = inode->i_mtime.tv_sec;
3591 struct obd_ops lmv_obd_ops = {
3592 .o_owner = THIS_MODULE,
3593 .o_setup = lmv_setup,
3594 .o_cleanup = lmv_cleanup,
3595 .o_precleanup = lmv_precleanup,
3596 .o_process_config = lmv_process_config,
3597 .o_connect = lmv_connect,
3598 .o_disconnect = lmv_disconnect,
3599 .o_statfs = lmv_statfs,
3600 .o_get_info = lmv_get_info,
3601 .o_set_info_async = lmv_set_info_async,
3602 .o_notify = lmv_notify,
3603 .o_get_uuid = lmv_get_uuid,
3604 .o_fid_alloc = lmv_fid_alloc,
3605 .o_iocontrol = lmv_iocontrol,
3606 .o_quotactl = lmv_quotactl
3609 struct md_ops lmv_md_ops = {
3610 .m_get_root = lmv_get_root,
3611 .m_null_inode = lmv_null_inode,
3612 .m_close = lmv_close,
3613 .m_create = lmv_create,
3614 .m_enqueue = lmv_enqueue,
3615 .m_getattr = lmv_getattr,
3616 .m_getxattr = lmv_getxattr,
3617 .m_getattr_name = lmv_getattr_name,
3618 .m_intent_lock = lmv_intent_lock,
3620 .m_rename = lmv_rename,
3621 .m_setattr = lmv_setattr,
3622 .m_setxattr = lmv_setxattr,
3623 .m_fsync = lmv_fsync,
3624 .m_file_resync = lmv_file_resync,
3625 .m_read_page = lmv_read_page,
3626 .m_unlink = lmv_unlink,
3627 .m_init_ea_size = lmv_init_ea_size,
3628 .m_cancel_unused = lmv_cancel_unused,
3629 .m_set_lock_data = lmv_set_lock_data,
3630 .m_lock_match = lmv_lock_match,
3631 .m_get_lustre_md = lmv_get_lustre_md,
3632 .m_free_lustre_md = lmv_free_lustre_md,
3633 .m_merge_attr = lmv_merge_attr,
3634 .m_set_open_replay_data = lmv_set_open_replay_data,
3635 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3636 .m_intent_getattr_async = lmv_intent_getattr_async,
3637 .m_revalidate_lock = lmv_revalidate_lock,
3638 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3639 .m_unpackmd = lmv_unpackmd,
3640 .m_rmfid = lmv_rmfid,
3643 static int __init lmv_init(void)
3645 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3646 LUSTRE_LMV_NAME, NULL);
3649 static void __exit lmv_exit(void)
3651 class_unregister_type(LUSTRE_LMV_NAME);
3654 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3655 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3656 MODULE_VERSION(LUSTRE_VERSION_STRING);
3657 MODULE_LICENSE("GPL");
3659 module_init(lmv_init);
3660 module_exit(lmv_exit);