4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
32 #define DEBUG_SUBSYSTEM S_LMV
34 #include <linux/file.h>
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/user_namespace.h>
38 #include <linux/uidgid.h>
39 #include <linux/slab.h>
40 #include <linux/pagemap.h>
42 #include <linux/math64.h>
43 #include <linux/seq_file.h>
44 #include <linux/namei.h>
46 #include <obd_support.h>
47 #include <lustre_lib.h>
48 #include <lustre_net.h>
49 #include <obd_class.h>
50 #include <lustre_lmv.h>
51 #include <lprocfs_status.h>
52 #include <cl_object.h>
53 #include <lustre_fid.h>
54 #include <uapi/linux/lustre/lustre_ioctl.h>
55 #include <lustre_kernelcomm.h>
56 #include "lmv_internal.h"
58 static int lmv_check_connect(struct obd_device *obd);
59 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data);
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 static 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 ? obd_get_uuid(tgt->ltd_exp) : NULL;
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 static 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 = lu_qos_del_tgt(&lmv->lmv_qos, tgt);
514 CERROR("%s: Can't del target from QoS table: rc = %d\n",
515 tgt->ltd_exp->exp_obd->obd_name, rc);
517 rc = fld_client_del_target(&lmv->lmv_fld, tgt->ltd_index);
519 CERROR("%s: Can't del fld targets: rc = %d\n",
520 tgt->ltd_exp->exp_obd->obd_name, rc);
522 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
524 CERROR("%s: Can't finalize fids factory: rc = %d\n",
525 tgt->ltd_exp->exp_obd->obd_name, rc);
527 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
528 tgt->ltd_exp->exp_obd->obd_name,
529 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
531 lmv_activate_target(lmv, tgt, 0);
532 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
533 rc = obd_disconnect(tgt->ltd_exp);
535 CERROR("%s: Target %s disconnect error: rc = %d\n",
536 tgt->ltd_exp->exp_obd->obd_name,
537 tgt->ltd_uuid.uuid, rc);
543 static int lmv_disconnect(struct obd_export *exp)
545 struct obd_device *obd = class_exp2obd(exp);
546 struct lmv_obd *lmv = &obd->u.lmv;
547 struct lmv_tgt_desc *tgt;
552 lmv_foreach_connected_tgt(lmv, tgt)
553 lmv_disconnect_mdc(obd, tgt);
555 if (lmv->lmv_tgts_kobj)
556 kobject_put(lmv->lmv_tgts_kobj);
559 rc = class_disconnect(exp);
564 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
567 struct obd_device *obd = class_exp2obd(exp);
568 struct lmv_obd *lmv = &obd->u.lmv;
569 struct getinfo_fid2path *gf;
570 struct lmv_tgt_desc *tgt;
571 struct getinfo_fid2path *remote_gf = NULL;
572 struct lu_fid root_fid;
573 int remote_gf_size = 0;
574 int currentisenc = 0;
579 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
581 RETURN(PTR_ERR(tgt));
583 root_fid = *gf->gf_u.gf_root_fid;
584 LASSERT(fid_is_sane(&root_fid));
587 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
588 if (rc != 0 && rc != -EREMOTE)
589 GOTO(out_fid2path, rc);
591 if (gf->gf_u.gf_path[0] == '/') {
592 /* by convention, server side (mdt_path_current()) puts
593 * a leading '/' to tell client that we are dealing with
602 /* If remote_gf != NULL, it means just building the
603 * path on the remote MDT, copy this path segment to gf.
605 if (remote_gf != NULL) {
606 struct getinfo_fid2path *ori_gf;
611 ori_gf = (struct getinfo_fid2path *)karg;
612 if (strlen(ori_gf->gf_u.gf_path) + 1 +
613 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
614 GOTO(out_fid2path, rc = -EOVERFLOW);
616 ptr = ori_gf->gf_u.gf_path;
617 oldisenc = ptr[0] == '/';
619 len = strlen(gf->gf_u.gf_path);
621 /* move the current path to the right to release space
622 * for closer-to-root part
624 memmove(ptr + len - currentisenc + 1 + globalisenc,
626 strlen(ori_gf->gf_u.gf_path) - oldisenc + 1);
629 memcpy(ptr, gf->gf_u.gf_path + currentisenc,
631 ptr[len - currentisenc] = '/';
635 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
636 tgt->ltd_exp->exp_obd->obd_name,
637 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
641 GOTO(out_fid2path, rc);
643 /* sigh, has to go to another MDT to do path building further */
644 if (remote_gf == NULL) {
645 remote_gf_size = sizeof(*remote_gf) + len - sizeof(*gf);
646 OBD_ALLOC(remote_gf, remote_gf_size);
647 if (remote_gf == NULL)
648 GOTO(out_fid2path, rc = -ENOMEM);
649 remote_gf->gf_pathlen = len - sizeof(*gf);
652 if (!fid_is_sane(&gf->gf_fid)) {
653 CERROR("%s: invalid FID "DFID": rc = %d\n",
654 tgt->ltd_exp->exp_obd->obd_name,
655 PFID(&gf->gf_fid), -EINVAL);
656 GOTO(out_fid2path, rc = -EINVAL);
659 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
661 GOTO(out_fid2path, rc = -EINVAL);
663 remote_gf->gf_fid = gf->gf_fid;
664 remote_gf->gf_recno = -1;
665 remote_gf->gf_linkno = -1;
666 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
667 *remote_gf->gf_u.gf_root_fid = root_fid;
669 goto repeat_fid2path;
672 if (remote_gf != NULL)
673 OBD_FREE(remote_gf, remote_gf_size);
677 static int lmv_hsm_req_count(struct lmv_obd *lmv,
678 const struct hsm_user_request *hur,
679 const struct lmv_tgt_desc *tgt_mds)
681 struct lmv_tgt_desc *curr_tgt;
685 /* count how many requests must be sent to the given target */
686 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
687 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
688 if (IS_ERR(curr_tgt))
689 RETURN(PTR_ERR(curr_tgt));
690 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
696 static int lmv_hsm_req_build(struct lmv_obd *lmv,
697 struct hsm_user_request *hur_in,
698 const struct lmv_tgt_desc *tgt_mds,
699 struct hsm_user_request *hur_out)
702 struct lmv_tgt_desc *curr_tgt;
704 /* build the hsm_user_request for the given target */
705 hur_out->hur_request = hur_in->hur_request;
707 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
708 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
709 if (IS_ERR(curr_tgt))
710 RETURN(PTR_ERR(curr_tgt));
711 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
712 hur_out->hur_user_item[nr_out] =
713 hur_in->hur_user_item[i];
717 hur_out->hur_request.hr_itemcount = nr_out;
718 memcpy(hur_data(hur_out), hur_data(hur_in),
719 hur_in->hur_request.hr_data_len);
724 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
725 int len, struct lustre_kernelcomm *lk,
728 struct lmv_obd *lmv = &obd->u.lmv;
729 struct lu_tgt_desc *tgt;
734 /* unregister request (call from llapi_hsm_copytool_fini) */
735 lmv_foreach_connected_tgt(lmv, tgt)
736 /* best effort: try to clean as much as possible
737 * (continue on error) */
738 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
740 /* Whatever the result, remove copytool from kuc groups.
741 * Unreached coordinators will get EPIPE on next requests
742 * and will unregister automatically.
744 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
749 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
750 int len, struct lustre_kernelcomm *lk,
753 struct lmv_obd *lmv = &obd->u.lmv;
755 bool any_set = false;
756 struct kkuc_ct_data *kcd;
758 struct lu_tgt_desc *tgt;
765 filp = fget(lk->lk_wfd);
769 if (lk->lk_flags & LK_FLG_DATANR)
770 kcd_size = offsetof(struct kkuc_ct_data,
771 kcd_archives[lk->lk_data_count]);
773 kcd_size = sizeof(*kcd);
775 OBD_ALLOC(kcd, kcd_size);
777 GOTO(err_fput, rc = -ENOMEM);
779 kcd->kcd_nr_archives = lk->lk_data_count;
780 if (lk->lk_flags & LK_FLG_DATANR) {
781 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
782 if (lk->lk_data_count > 0)
783 memcpy(kcd->kcd_archives, lk->lk_data,
784 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
786 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
789 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
790 lk->lk_group, kcd, kcd_size);
791 OBD_FREE(kcd, kcd_size);
795 /* All or nothing: try to register to all MDS.
796 * In case of failure, unregister from previous MDS,
797 * except if it because of inactive target. */
798 lmv_foreach_connected_tgt(lmv, tgt) {
799 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
801 if (tgt->ltd_active) {
802 /* permanent error */
803 CERROR("%s: iocontrol MDC %s on MDT"
804 " idx %d cmd %x: err = %d\n",
805 lmv2obd_dev(lmv)->obd_name,
806 tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
809 lk->lk_flags |= LK_FLG_STOP;
811 /* unregister from previous MDS */
812 lmv_foreach_connected_tgt(lmv, tgt) {
813 if (tgt->ltd_index >= i)
816 obd_iocontrol(cmd, tgt->ltd_exp, len,
819 GOTO(err_kkuc_rem, rc);
821 /* else: transient error.
822 * kuc will register to the missing MDT
830 /* no registration done: return error */
831 GOTO(err_kkuc_rem, rc = -ENOTCONN);
836 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
843 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
844 int len, void *karg, void __user *uarg)
846 struct obd_device *obd = class_exp2obd(exp);
847 struct lmv_obd *lmv = &obd->u.lmv;
848 struct lu_tgt_desc *tgt = NULL;
850 __u32 count = lmv->lmv_mdt_count;
854 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
855 exp->exp_obd->obd_name, cmd, len, karg, uarg);
861 case IOC_OBD_STATFS: {
862 struct obd_ioctl_data *data = karg;
863 struct obd_device *mdc_obd;
864 struct obd_statfs stat_buf = {0};
867 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
869 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
872 tgt = lmv_tgt(lmv, index);
876 if (!tgt->ltd_active)
879 mdc_obd = class_exp2obd(tgt->ltd_exp);
884 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
885 min((int) data->ioc_plen2,
886 (int) sizeof(struct obd_uuid))))
889 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
890 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
894 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
895 min((int) data->ioc_plen1,
896 (int) sizeof(stat_buf))))
900 case OBD_IOC_QUOTACTL: {
901 struct if_quotactl *qctl = karg;
902 struct obd_quotactl *oqctl;
903 struct obd_import *imp;
905 if (qctl->qc_valid == QC_MDTIDX) {
906 tgt = lmv_tgt(lmv, qctl->qc_idx);
907 } else if (qctl->qc_valid == QC_UUID) {
908 lmv_foreach_tgt(lmv, tgt) {
909 if (!obd_uuid_equals(&tgt->ltd_uuid,
928 imp = class_exp2cliimp(tgt->ltd_exp);
929 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
930 qctl->qc_valid = QC_MDTIDX;
931 qctl->obd_uuid = tgt->ltd_uuid;
935 OBD_ALLOC_PTR(oqctl);
939 QCTL_COPY(oqctl, qctl);
940 rc = obd_quotactl(tgt->ltd_exp, oqctl);
942 QCTL_COPY(qctl, oqctl);
943 qctl->qc_valid = QC_MDTIDX;
944 qctl->obd_uuid = tgt->ltd_uuid;
949 case LL_IOC_GET_CONNECT_FLAGS: {
950 tgt = lmv_tgt(lmv, 0);
952 if (tgt && tgt->ltd_exp)
953 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
956 case LL_IOC_FID2MDTIDX: {
957 struct lu_fid *fid = karg;
960 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
964 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
965 * point to user space memory for FID2MDTIDX. */
966 *(__u32 *)uarg = mdt_index;
969 case OBD_IOC_FID2PATH: {
970 rc = lmv_fid2path(exp, len, karg, uarg);
973 case LL_IOC_HSM_STATE_GET:
974 case LL_IOC_HSM_STATE_SET:
975 case LL_IOC_HSM_ACTION: {
976 struct md_op_data *op_data = karg;
978 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
980 RETURN(PTR_ERR(tgt));
982 if (tgt->ltd_exp == NULL)
985 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
988 case LL_IOC_HSM_PROGRESS: {
989 const struct hsm_progress_kernel *hpk = karg;
991 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
993 RETURN(PTR_ERR(tgt));
994 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
997 case LL_IOC_HSM_REQUEST: {
998 struct hsm_user_request *hur = karg;
999 unsigned int reqcount = hur->hur_request.hr_itemcount;
1004 /* if the request is about a single fid
1005 * or if there is a single MDS, no need to split
1007 if (reqcount == 1 || count == 1) {
1008 tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
1010 RETURN(PTR_ERR(tgt));
1011 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1013 /* split fid list to their respective MDS */
1014 lmv_foreach_connected_tgt(lmv, tgt) {
1017 struct hsm_user_request *req;
1019 nr = lmv_hsm_req_count(lmv, hur, tgt);
1022 if (nr == 0) /* nothing for this MDS */
1025 /* build a request with fids for this MDS */
1026 reqlen = offsetof(typeof(*hur),
1028 + hur->hur_request.hr_data_len;
1029 OBD_ALLOC_LARGE(req, reqlen);
1032 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1034 GOTO(hsm_req_err, rc1);
1035 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1038 if (rc1 != 0 && rc == 0)
1040 OBD_FREE_LARGE(req, reqlen);
1045 case LL_IOC_LOV_SWAP_LAYOUTS: {
1046 struct md_op_data *op_data = karg;
1047 struct lmv_tgt_desc *tgt1, *tgt2;
1049 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1051 RETURN(PTR_ERR(tgt1));
1053 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1055 RETURN(PTR_ERR(tgt2));
1057 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1060 /* only files on same MDT can have their layouts swapped */
1061 if (tgt1->ltd_index != tgt2->ltd_index)
1064 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1067 case LL_IOC_HSM_CT_START: {
1068 struct lustre_kernelcomm *lk = karg;
1070 if (lk->lk_flags & LK_FLG_STOP)
1071 rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1073 rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1077 lmv_foreach_connected_tgt(lmv, tgt) {
1078 struct obd_device *mdc_obd;
1081 /* ll_umount_begin() sets force flag but for lmv, not
1082 * mdc. Let's pass it through */
1083 mdc_obd = class_exp2obd(tgt->ltd_exp);
1084 mdc_obd->obd_force = obd->obd_force;
1085 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1087 if (tgt->ltd_active) {
1088 OBD_IOC_ERROR(obd->obd_name, cmd,
1089 tgt->ltd_uuid.uuid, err);
1102 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1103 struct lu_fid *fid, struct md_op_data *op_data)
1105 struct obd_device *obd = class_exp2obd(exp);
1106 struct lmv_obd *lmv = &obd->u.lmv;
1107 struct lmv_tgt_desc *tgt;
1115 tgt = lmv_tgt(lmv, op_data->op_mds);
1119 if (!tgt->ltd_active || !tgt->ltd_exp)
1123 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1124 * on server that seq in new allocated fid is not yet known.
1126 mutex_lock(&tgt->ltd_fid_mutex);
1127 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1128 mutex_unlock(&tgt->ltd_fid_mutex);
1130 LASSERT(fid_is_sane(fid));
1137 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1139 struct lmv_obd *lmv = &obd->u.lmv;
1140 struct lmv_desc *desc;
1141 struct lnet_processid lnet_id;
1147 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1148 CERROR("LMV setup requires a descriptor\n");
1152 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1153 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1154 CERROR("Lmv descriptor size wrong: %d > %d\n",
1155 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1159 obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1160 desc->ld_uuid.uuid);
1161 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1162 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1163 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1164 LMV_DESC_QOS_MAXAGE_DEFAULT;
1165 lmv->max_def_easize = 0;
1166 lmv->max_easize = 0;
1168 spin_lock_init(&lmv->lmv_lock);
1171 * initialize rr_index to lower 32bit of netid, so that client
1172 * can distribute subdirs evenly from the beginning.
1174 while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1175 if (!nid_is_lo0(&lnet_id.nid)) {
1176 lmv->lmv_qos_rr_index = ntohl(lnet_id.nid.nid_addr[0]);
1181 rc = lmv_tunables_init(obd);
1183 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1186 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1187 LUSTRE_CLI_FLD_HASH_DHT);
1189 CERROR("Can't init FLD, err %d\n", rc);
1191 rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1193 CWARN("%s: error initialize target table: rc = %d\n",
1199 static int lmv_cleanup(struct obd_device *obd)
1201 struct lmv_obd *lmv = &obd->u.lmv;
1202 struct lu_tgt_desc *tgt;
1203 struct lu_tgt_desc *tmp;
1207 fld_client_fini(&lmv->lmv_fld);
1208 fld_client_debugfs_fini(&lmv->lmv_fld);
1210 lprocfs_obd_cleanup(obd);
1211 lprocfs_free_md_stats(obd);
1213 lmv_foreach_tgt_safe(lmv, tgt, tmp)
1214 lmv_del_target(lmv, tgt);
1215 lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1220 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1222 struct lustre_cfg *lcfg = buf;
1223 struct obd_uuid obd_uuid;
1229 switch (lcfg->lcfg_command) {
1231 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1232 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1233 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1234 GOTO(out, rc = -EINVAL);
1236 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1238 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1239 GOTO(out, rc = -EINVAL);
1240 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1241 GOTO(out, rc = -EINVAL);
1242 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1245 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1246 GOTO(out, rc = -EINVAL);
1252 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1256 if (flags & OBD_STATFS_FOR_MDT0)
1259 if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1260 return lmv->lmv_statfs_start;
1262 /* choose initial MDT for this client */
1264 struct lnet_processid lnet_id;
1265 if (LNetGetId(i, &lnet_id) == -ENOENT)
1268 if (!nid_is_lo0(&lnet_id.nid)) {
1269 /* We dont need a full 64-bit modulus, just enough
1270 * to distribute the requests across MDTs evenly.
1272 lmv->lmv_statfs_start = nidhash(&lnet_id.nid) %
1278 return lmv->lmv_statfs_start;
1281 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1282 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1284 struct obd_device *obd = class_exp2obd(exp);
1285 struct lmv_obd *lmv = &obd->u.lmv;
1286 struct obd_statfs *temp;
1287 struct lu_tgt_desc *tgt;
1295 OBD_ALLOC(temp, sizeof(*temp));
1299 /* distribute statfs among MDTs */
1300 idx = lmv_select_statfs_mdt(lmv, flags);
1302 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1303 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1304 tgt = lmv_tgt(lmv, idx);
1305 if (!tgt || !tgt->ltd_exp)
1308 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age,
1309 flags | OBD_STATFS_NESTED);
1311 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1312 tgt->ltd_exp->exp_obd->obd_name, i, rc);
1314 /* Try another MDT */
1315 if (flags & OBD_STATFS_SUM)
1317 GOTO(out_free_temp, rc);
1320 if (temp->os_state & OS_STATFS_SUM ||
1321 flags == OBD_STATFS_FOR_MDT0) {
1322 /* reset to the last aggregated values
1323 * and don't sum with non-aggrated data */
1324 /* If the statfs is from mount, it needs to retrieve
1325 * necessary information from MDT0. i.e. mount does
1326 * not need the merged osfs from all of MDT. Also
1327 * clients can be mounted as long as MDT0 is in
1330 GOTO(out_free_temp, rc);
1336 osfs->os_bavail += temp->os_bavail;
1337 osfs->os_blocks += temp->os_blocks;
1338 osfs->os_ffree += temp->os_ffree;
1339 osfs->os_files += temp->os_files;
1340 osfs->os_granted += temp->os_granted;
1343 /* There is no stats from some MDTs, data incomplete */
1347 OBD_FREE(temp, sizeof(*temp));
1351 static int lmv_statfs_update(void *cookie, int rc)
1353 struct obd_info *oinfo = cookie;
1354 struct obd_device *obd = oinfo->oi_obd;
1355 struct lmv_obd *lmv = &obd->u.lmv;
1356 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1357 struct obd_statfs *osfs = oinfo->oi_osfs;
1360 * NB: don't deactivate TGT upon error, because we may not trigger async
1361 * statfs any longer, then there is no chance to activate TGT.
1364 spin_lock(&lmv->lmv_lock);
1365 tgt->ltd_statfs = *osfs;
1366 tgt->ltd_statfs_age = ktime_get_seconds();
1367 spin_unlock(&lmv->lmv_lock);
1368 set_bit(LQ_DIRTY, &lmv->lmv_qos.lq_flags);
1374 /* update tgt statfs async if it's ld_qos_maxage old */
1375 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1377 struct obd_info oinfo = {
1380 .oi_cb_up = lmv_statfs_update,
1384 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1385 obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1388 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1393 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1396 struct obd_device *obd = exp->exp_obd;
1397 struct lmv_obd *lmv = &obd->u.lmv;
1398 struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1406 rc = md_get_root(tgt->ltd_exp, fileset, fid);
1410 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1411 u64 obd_md_valid, const char *name, size_t buf_size,
1412 struct ptlrpc_request **req)
1414 struct obd_device *obd = exp->exp_obd;
1415 struct lmv_obd *lmv = &obd->u.lmv;
1416 struct lmv_tgt_desc *tgt;
1421 tgt = lmv_fid2tgt(lmv, fid);
1423 RETURN(PTR_ERR(tgt));
1425 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1430 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1431 u64 obd_md_valid, const char *name,
1432 const void *value, size_t value_size,
1433 unsigned int xattr_flags, u32 suppgid,
1434 struct ptlrpc_request **req)
1436 struct obd_device *obd = exp->exp_obd;
1437 struct lmv_obd *lmv = &obd->u.lmv;
1438 struct lmv_tgt_desc *tgt;
1443 tgt = lmv_fid2tgt(lmv, fid);
1445 RETURN(PTR_ERR(tgt));
1447 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1448 value, value_size, xattr_flags, suppgid, req);
1453 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1454 struct ptlrpc_request **request)
1456 struct obd_device *obd = exp->exp_obd;
1457 struct lmv_obd *lmv = &obd->u.lmv;
1458 struct lmv_tgt_desc *tgt;
1463 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1465 RETURN(PTR_ERR(tgt));
1467 if (op_data->op_flags & MF_GET_MDT_IDX) {
1468 op_data->op_mds = tgt->ltd_index;
1472 rc = md_getattr(tgt->ltd_exp, op_data, request);
1477 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1479 struct obd_device *obd = exp->exp_obd;
1480 struct lmv_obd *lmv = &obd->u.lmv;
1481 struct lu_tgt_desc *tgt;
1485 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1488 * With DNE every object can have two locks in different namespaces:
1489 * lookup lock in space of MDT storing direntry and update/open lock in
1490 * space of MDT storing inode.
1492 lmv_foreach_connected_tgt(lmv, tgt)
1493 md_null_inode(tgt->ltd_exp, fid);
1498 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1499 struct md_open_data *mod, struct ptlrpc_request **request)
1501 struct obd_device *obd = exp->exp_obd;
1502 struct lmv_obd *lmv = &obd->u.lmv;
1503 struct lmv_tgt_desc *tgt;
1508 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1510 RETURN(PTR_ERR(tgt));
1512 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1513 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1517 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv,
1518 struct md_op_data *op_data)
1520 struct lu_tgt_desc *tgt, *cur = NULL;
1521 __u64 total_avail = 0;
1522 __u64 total_weight = 0;
1523 __u64 cur_weight = 0;
1524 int total_usable = 0;
1530 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1531 RETURN(ERR_PTR(-EAGAIN));
1533 down_write(&lmv->lmv_qos.lq_rw_sem);
1535 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1536 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1538 rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1540 GOTO(unlock, tgt = ERR_PTR(rc));
1542 lmv_foreach_tgt(lmv, tgt) {
1543 if (!tgt->ltd_exp || !tgt->ltd_active) {
1544 tgt->ltd_qos.ltq_usable = 0;
1548 tgt->ltd_qos.ltq_usable = 1;
1549 lu_tgt_qos_weight_calc(tgt, true);
1550 if (tgt->ltd_index == op_data->op_mds)
1552 total_avail += tgt->ltd_qos.ltq_avail;
1553 total_weight += tgt->ltd_qos.ltq_weight;
1557 /* If current MDT has above-average space and dir is not aleady using
1558 * round-robin to spread across more MDTs, stay on the parent MDT
1559 * to avoid creating needless remote MDT directories. Remote dirs
1560 * close to the root balance space more effectively than bottom dirs,
1561 * so prefer to create remote dirs at top level of directory tree.
1562 * "16 / (dir_depth + 10)" is the factor to make it less likely
1563 * for top-level directories to stay local unless they have more than
1564 * average free space, while deep dirs prefer local until more full.
1565 * depth=0 -> 160%, depth=3 -> 123%, depth=6 -> 100%,
1566 * depth=9 -> 84%, depth=12 -> 73%, depth=15 -> 64%
1568 if (!lmv_op_default_rr_mkdir(op_data)) {
1569 rand = total_avail * 16 /
1570 (total_usable * (op_data->op_dir_depth + 10));
1571 if (cur && cur->ltd_qos.ltq_avail >= rand) {
1577 rand = lu_prandom_u64_max(total_weight);
1579 lmv_foreach_connected_tgt(lmv, tgt) {
1580 if (!tgt->ltd_qos.ltq_usable)
1583 cur_weight += tgt->ltd_qos.ltq_weight;
1584 if (cur_weight < rand)
1587 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1591 /* no proper target found */
1592 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1594 up_write(&lmv->lmv_qos.lq_rw_sem);
1599 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv)
1601 struct lu_tgt_desc *tgt;
1607 spin_lock(&lmv->lmv_lock);
1608 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1609 index = (i + lmv->lmv_qos_rr_index) %
1610 lmv->lmv_mdt_descs.ltd_tgts_size;
1611 tgt = lmv_tgt(lmv, index);
1612 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1615 lmv->lmv_qos_rr_index = (tgt->ltd_index + 1) %
1616 lmv->lmv_mdt_descs.ltd_tgts_size;
1617 spin_unlock(&lmv->lmv_lock);
1621 spin_unlock(&lmv->lmv_lock);
1623 RETURN(ERR_PTR(-ENODEV));
1626 /* locate MDT which is less full (avoid the most full MDT) */
1627 static struct lu_tgt_desc *lmv_locate_tgt_lf(struct lmv_obd *lmv)
1629 struct lu_tgt_desc *min = NULL;
1630 struct lu_tgt_desc *tgt;
1636 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1637 RETURN(ERR_PTR(-EAGAIN));
1639 down_write(&lmv->lmv_qos.lq_rw_sem);
1641 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1642 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1644 lmv_foreach_tgt(lmv, tgt) {
1645 if (!tgt->ltd_exp || !tgt->ltd_active) {
1646 tgt->ltd_qos.ltq_usable = 0;
1650 tgt->ltd_qos.ltq_usable = 1;
1651 lu_tgt_qos_weight_calc(tgt, true);
1652 avail += tgt->ltd_qos.ltq_avail;
1653 if (!min || min->ltd_qos.ltq_avail > tgt->ltd_qos.ltq_avail)
1657 /* avoid the most full MDT */
1659 avail -= min->ltd_qos.ltq_avail;
1661 rand = lu_prandom_u64_max(avail);
1663 lmv_foreach_connected_tgt(lmv, tgt) {
1664 if (!tgt->ltd_qos.ltq_usable)
1670 avail += tgt->ltd_qos.ltq_avail;
1677 /* no proper target found */
1678 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1680 up_write(&lmv->lmv_qos.lq_rw_sem);
1685 /* locate MDT by file name, for striped directory, the file name hash decides
1686 * which stripe its dirent is stored.
1688 static struct lmv_tgt_desc *
1689 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1690 const char *name, int namelen, struct lu_fid *fid,
1691 __u32 *mds, bool new_layout)
1693 struct lmv_tgt_desc *tgt;
1694 const struct lmv_oinfo *oinfo;
1696 if (!lmv_dir_striped(lsm) || !namelen) {
1697 tgt = lmv_fid2tgt(lmv, fid);
1701 *mds = tgt->ltd_index;
1705 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1706 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1707 return ERR_PTR(-EBADF);
1708 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1710 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, new_layout);
1712 return ERR_CAST(oinfo);
1715 /* check stripe FID is sane */
1716 if (!fid_is_sane(&oinfo->lmo_fid))
1717 return ERR_PTR(-ENODEV);
1719 *fid = oinfo->lmo_fid;
1720 *mds = oinfo->lmo_mds;
1721 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1723 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1725 return tgt ? tgt : ERR_PTR(-ENODEV);
1729 * Locate MDT of op_data->op_fid1
1731 * For striped directory, it will locate the stripe by name hash, if hash_type
1732 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1733 * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1734 * indicates whether old or new layout is used to locate.
1736 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1738 * \param[in] lmv LMV device
1739 * \param[in/out] op_data client MD stack parameters, name, namelen etc,
1740 * op_mds and op_fid1 will be updated if op_mea1
1741 * indicates fid1 represents a striped directory.
1743 * retval pointer to the lmv_tgt_desc if succeed.
1744 * ERR_PTR(errno) if failed.
1746 struct lmv_tgt_desc *
1747 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1749 struct lmv_stripe_md *lsm = op_data->op_mea1;
1750 struct lmv_oinfo *oinfo;
1751 struct lmv_tgt_desc *tgt;
1753 if (lmv_dir_foreign(lsm))
1754 return ERR_PTR(-ENODATA);
1756 /* During creating VOLATILE file, it should honor the mdt
1757 * index if the file under striped dir is being restored, see
1759 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1760 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1761 tgt = lmv_tgt(lmv, op_data->op_mds);
1763 return ERR_PTR(-ENODEV);
1765 if (lmv_dir_striped(lsm)) {
1768 /* refill the right parent fid */
1769 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1770 oinfo = &lsm->lsm_md_oinfo[i];
1771 if (oinfo->lmo_mds == op_data->op_mds) {
1772 op_data->op_fid1 = oinfo->lmo_fid;
1777 if (i == lsm->lsm_md_stripe_count)
1778 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1780 } else if (lmv_dir_bad_hash(lsm)) {
1781 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1782 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1784 op_data->op_fid1 = oinfo->lmo_fid;
1785 op_data->op_mds = oinfo->lmo_mds;
1786 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1788 return ERR_PTR(-ENODEV);
1790 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1791 op_data->op_name, op_data->op_namelen,
1792 &op_data->op_fid1, &op_data->op_mds,
1793 op_data->op_new_layout);
1799 /* Locate MDT of op_data->op_fid2 for link/rename */
1800 static struct lmv_tgt_desc *
1801 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1803 struct lmv_tgt_desc *tgt;
1806 LASSERT(op_data->op_name);
1807 if (lmv_dir_layout_changing(op_data->op_mea2)) {
1808 struct lu_fid fid1 = op_data->op_fid1;
1809 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1810 struct ptlrpc_request *request = NULL;
1813 * avoid creating new file under old layout of migrating
1814 * directory, check it here.
1816 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1817 op_data->op_name, op_data->op_namelen,
1818 &op_data->op_fid2, &op_data->op_mds, false);
1822 op_data->op_fid1 = op_data->op_fid2;
1823 op_data->op_mea1 = op_data->op_mea2;
1824 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1825 op_data->op_fid1 = fid1;
1826 op_data->op_mea1 = lsm1;
1828 ptlrpc_req_finished(request);
1829 RETURN(ERR_PTR(-EEXIST));
1833 RETURN(ERR_PTR(rc));
1836 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1837 op_data->op_namelen, &op_data->op_fid2,
1838 &op_data->op_mds, true);
1841 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1843 struct lu_tgt_desc *tgt;
1844 struct ptlrpc_request *request;
1847 LASSERT(lmv_dir_layout_changing(op_data->op_mea1));
1848 LASSERT(!op_data->op_new_layout);
1850 tgt = lmv_locate_tgt(lmv, op_data);
1852 return PTR_ERR(tgt);
1854 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1856 ptlrpc_req_finished(request);
1863 /* mkdir by QoS upon 'lfs mkdir -i -1'.
1865 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1866 * directories under striped directory.
1868 static inline bool lmv_op_user_qos_mkdir(const struct md_op_data *op_data)
1870 const struct lmv_user_md *lum = op_data->op_data;
1872 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1875 if (lmv_dir_striped(op_data->op_mea1))
1878 return (op_data->op_cli_flags & CLI_SET_MEA) && lum &&
1879 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
1880 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
1883 /* mkdir by QoS if either ROOT or parent default LMV is space balanced. */
1884 static inline bool lmv_op_default_qos_mkdir(const struct md_op_data *op_data)
1886 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1888 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1891 if (lmv_dir_striped(op_data->op_mea1))
1894 return (op_data->op_flags & MF_QOS_MKDIR) ||
1895 (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT);
1898 /* if parent default LMV is space balanced, and
1899 * 1. max_inherit_rr is set
1900 * 2. or parent is ROOT
1901 * mkdir roundrobin. Or if parent doesn't have default LMV, while ROOT default
1902 * LMV requests roundrobin mkdir, do the same.
1903 * NB, this needs to check server is balanced, which is done by caller.
1905 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data)
1907 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1909 return (op_data->op_flags & MF_RR_MKDIR) ||
1910 (lsm && lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE) ||
1911 fid_is_root(&op_data->op_fid1);
1914 /* 'lfs mkdir -i <specific_MDT>' */
1915 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1917 const struct lmv_user_md *lum = op_data->op_data;
1919 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1920 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1921 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1922 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1923 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1926 /* parent default LMV master_mdt_index is not -1. */
1928 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1930 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1931 op_data->op_default_mea1 &&
1932 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1936 /* locate MDT by space usage */
1937 static struct lu_tgt_desc *lmv_locate_tgt_by_space(struct lmv_obd *lmv,
1938 struct md_op_data *op_data,
1939 struct lmv_tgt_desc *tgt)
1941 struct lmv_tgt_desc *tmp = tgt;
1943 tgt = lmv_locate_tgt_qos(lmv, op_data);
1944 if (tgt == ERR_PTR(-EAGAIN)) {
1945 if (ltd_qos_is_balanced(&lmv->lmv_mdt_descs) &&
1946 !lmv_op_default_rr_mkdir(op_data) &&
1947 !lmv_op_user_qos_mkdir(op_data))
1948 /* if not necessary, don't create remote directory. */
1951 tgt = lmv_locate_tgt_rr(lmv);
1955 * only update statfs after QoS mkdir, this means the cached statfs may
1956 * be stale, and current mkdir may not follow QoS accurately, but it's
1957 * not serious, and avoids periodic statfs when client doesn't mkdir by
1961 op_data->op_mds = tgt->ltd_index;
1962 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1968 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1969 const void *data, size_t datalen, umode_t mode, uid_t uid,
1970 gid_t gid, kernel_cap_t cap_effective, __u64 rdev,
1971 struct ptlrpc_request **request)
1973 struct obd_device *obd = exp->exp_obd;
1974 struct lmv_obd *lmv = &obd->u.lmv;
1975 struct lmv_tgt_desc *tgt;
1976 struct mdt_body *repbody;
1981 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1984 if (lmv_dir_bad_hash(op_data->op_mea1))
1987 if (lmv_dir_layout_changing(op_data->op_mea1)) {
1989 * if parent is migrating, create() needs to lookup existing
1990 * name in both old and new layout, check old layout on client.
1992 rc = lmv_old_layout_lookup(lmv, op_data);
1996 op_data->op_new_layout = true;
1999 tgt = lmv_locate_tgt(lmv, op_data);
2001 RETURN(PTR_ERR(tgt));
2003 /* the order to apply policy in mkdir:
2004 * 1. is "lfs mkdir -i N"? mkdir on MDT N.
2005 * 2. is "lfs mkdir -i -1"? mkdir by space usage.
2006 * 3. is starting MDT specified in default LMV? mkdir on MDT N.
2007 * 4. is default LMV space balanced? mkdir by space usage.
2009 if (lmv_op_user_specific_mkdir(op_data)) {
2010 struct lmv_user_md *lum = op_data->op_data;
2012 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2013 tgt = lmv_tgt(lmv, op_data->op_mds);
2016 } else if (lmv_op_user_qos_mkdir(op_data)) {
2017 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2019 RETURN(PTR_ERR(tgt));
2020 } else if (lmv_op_default_specific_mkdir(op_data)) {
2022 op_data->op_default_mea1->lsm_md_master_mdt_index;
2023 tgt = lmv_tgt(lmv, op_data->op_mds);
2026 } else if (lmv_op_default_qos_mkdir(op_data)) {
2027 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2029 RETURN(PTR_ERR(tgt));
2033 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2037 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
2038 (int)op_data->op_namelen, op_data->op_name,
2039 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
2042 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2043 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
2044 cap_effective, rdev, request);
2046 if (*request == NULL)
2048 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
2051 /* dir restripe needs to send to MDT where dir is located */
2052 if (rc != -EREMOTE ||
2053 !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
2056 repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2057 if (repbody == NULL)
2060 /* Not cross-ref case, just get out of here. */
2061 if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
2064 op_data->op_fid2 = repbody->mbo_fid1;
2065 ptlrpc_req_finished(*request);
2068 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2070 RETURN(PTR_ERR(tgt));
2072 op_data->op_mds = tgt->ltd_index;
2077 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
2078 const union ldlm_policy_data *policy, struct md_op_data *op_data,
2079 struct lustre_handle *lockh, __u64 extra_lock_flags)
2081 struct obd_device *obd = exp->exp_obd;
2082 struct lmv_obd *lmv = &obd->u.lmv;
2083 struct lmv_tgt_desc *tgt;
2088 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
2090 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2092 RETURN(PTR_ERR(tgt));
2094 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
2095 PFID(&op_data->op_fid1), tgt->ltd_index);
2097 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
2104 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
2105 struct ptlrpc_request **preq)
2107 struct obd_device *obd = exp->exp_obd;
2108 struct lmv_obd *lmv = &obd->u.lmv;
2109 struct lmv_tgt_desc *tgt;
2110 struct mdt_body *body;
2116 if (op_data->op_namelen == 2 &&
2117 op_data->op_name[0] == '.' && op_data->op_name[1] == '.')
2118 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2120 tgt = lmv_locate_tgt(lmv, op_data);
2122 RETURN(PTR_ERR(tgt));
2124 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
2125 (int)op_data->op_namelen, op_data->op_name,
2126 PFID(&op_data->op_fid1), tgt->ltd_index);
2128 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2129 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2130 ptlrpc_req_finished(*preq);
2138 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2139 LASSERT(body != NULL);
2141 if (body->mbo_valid & OBD_MD_MDS) {
2142 op_data->op_fid1 = body->mbo_fid1;
2143 op_data->op_valid |= OBD_MD_FLCROSSREF;
2144 op_data->op_namelen = 0;
2145 op_data->op_name = NULL;
2147 ptlrpc_req_finished(*preq);
2156 #define md_op_data_fid(op_data, fl) \
2157 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2158 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2159 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2160 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2163 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2164 struct md_op_data *op_data, __u32 op_tgt,
2165 enum ldlm_mode mode, int bits, int flag)
2167 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2168 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2169 union ldlm_policy_data policy = { { 0 } };
2173 if (!fid_is_sane(fid))
2177 tgt = lmv_fid2tgt(lmv, fid);
2179 RETURN(PTR_ERR(tgt));
2182 if (tgt->ltd_index != op_tgt) {
2183 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2184 policy.l_inodebits.bits = bits;
2185 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2186 mode, LCF_ASYNC, NULL);
2189 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2191 op_data->op_flags |= flag;
2199 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2202 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2203 struct ptlrpc_request **request)
2205 struct obd_device *obd = exp->exp_obd;
2206 struct lmv_obd *lmv = &obd->u.lmv;
2207 struct lmv_tgt_desc *tgt;
2211 LASSERT(op_data->op_namelen != 0);
2213 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2214 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2215 op_data->op_name, PFID(&op_data->op_fid1));
2217 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2218 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2219 op_data->op_cap = current_cap();
2221 tgt = lmv_locate_tgt2(lmv, op_data);
2223 RETURN(PTR_ERR(tgt));
2226 * Cancel UPDATE lock on child (fid1).
2228 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2229 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2230 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2234 rc = md_link(tgt->ltd_exp, op_data, request);
2239 /* migrate the top directory */
2240 static inline bool lmv_op_topdir_migrate(const struct md_op_data *op_data)
2242 if (!S_ISDIR(op_data->op_mode))
2245 if (lmv_dir_layout_changing(op_data->op_mea1))
2251 /* migrate top dir to specific MDTs */
2252 static inline bool lmv_topdir_specific_migrate(const struct md_op_data *op_data)
2254 const struct lmv_user_md *lum = op_data->op_data;
2256 if (!lmv_op_topdir_migrate(op_data))
2259 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2262 /* migrate top dir in QoS mode if user issued "lfs migrate -m -1..." */
2263 static inline bool lmv_topdir_qos_migrate(const struct md_op_data *op_data)
2265 const struct lmv_user_md *lum = op_data->op_data;
2267 if (!lmv_op_topdir_migrate(op_data))
2270 return le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
2273 static inline bool lmv_subdir_specific_migrate(const struct md_op_data *op_data)
2275 const struct lmv_user_md *lum = op_data->op_data;
2277 if (!S_ISDIR(op_data->op_mode))
2280 if (!lmv_dir_layout_changing(op_data->op_mea1))
2283 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2286 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2287 const char *name, size_t namelen,
2288 struct ptlrpc_request **request)
2290 struct obd_device *obd = exp->exp_obd;
2291 struct lmv_obd *lmv = &obd->u.lmv;
2292 struct lmv_stripe_md *lsm = op_data->op_mea1;
2293 struct lmv_tgt_desc *parent_tgt;
2294 struct lmv_tgt_desc *sp_tgt;
2295 struct lmv_tgt_desc *tp_tgt = NULL;
2296 struct lmv_tgt_desc *child_tgt;
2297 struct lmv_tgt_desc *tgt;
2298 struct lu_fid target_fid = { 0 };
2303 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2305 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2306 PFID(&op_data->op_fid1), (int)namelen, name);
2308 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2309 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2310 op_data->op_cap = current_cap();
2312 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2313 if (IS_ERR(parent_tgt))
2314 RETURN(PTR_ERR(parent_tgt));
2316 if (lmv_dir_striped(lsm)) {
2317 const struct lmv_oinfo *oinfo;
2319 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
2321 RETURN(PTR_ERR(oinfo));
2323 /* save source stripe FID in fid4 temporarily for ELC */
2324 op_data->op_fid4 = oinfo->lmo_fid;
2325 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2330 * if parent is being migrated too, fill op_fid2 with target
2331 * stripe fid, otherwise the target stripe is not created yet.
2333 if (lmv_dir_layout_changing(lsm)) {
2334 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
2337 RETURN(PTR_ERR(oinfo));
2339 op_data->op_fid2 = oinfo->lmo_fid;
2340 tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2344 /* parent unchanged and update namespace only */
2345 if (lu_fid_eq(&op_data->op_fid4, &op_data->op_fid2) &&
2346 op_data->op_bias & MDS_MIGRATE_NSONLY)
2350 sp_tgt = parent_tgt;
2353 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2354 if (IS_ERR(child_tgt))
2355 RETURN(PTR_ERR(child_tgt));
2357 if (lmv_topdir_specific_migrate(op_data)) {
2358 struct lmv_user_md *lum = op_data->op_data;
2360 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2361 } else if (lmv_topdir_qos_migrate(op_data)) {
2362 tgt = lmv_locate_tgt_lf(lmv);
2363 if (tgt == ERR_PTR(-EAGAIN))
2364 tgt = lmv_locate_tgt_rr(lmv);
2366 RETURN(PTR_ERR(tgt));
2368 op_data->op_mds = tgt->ltd_index;
2369 } else if (lmv_subdir_specific_migrate(op_data)) {
2370 struct lmv_user_md *lum = op_data->op_data;
2374 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2375 /* adjust MDTs in lum, since subdir is located on where
2376 * its parent stripe is, not the first specified MDT.
2378 for (i = 0; i < le32_to_cpu(lum->lum_stripe_count);
2380 if (le32_to_cpu(lum->lum_objects[i].lum_mds) ==
2385 if (i == le32_to_cpu(lum->lum_stripe_count))
2388 lum->lum_objects[i].lum_mds =
2389 lum->lum_objects[0].lum_mds;
2390 lum->lum_objects[0].lum_mds =
2391 cpu_to_le32(tp_tgt->ltd_index);
2393 /* NB, the above adjusts subdir migration for command like
2394 * "lfs migrate -m 0,1,2 ...", but for migration like
2395 * "lfs migrate -m 0 -c 2 ...", the top dir is migrated to MDT0
2396 * and MDT1, however its subdir may be migrated to MDT1 and MDT2
2399 lum->lum_stripe_offset = cpu_to_le32(tp_tgt->ltd_index);
2400 op_data->op_mds = tp_tgt->ltd_index;
2401 } else if (tp_tgt) {
2402 op_data->op_mds = tp_tgt->ltd_index;
2404 op_data->op_mds = sp_tgt->ltd_index;
2407 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2412 * for directory, send migrate request to the MDT where the object will
2413 * be migrated to, because we can't create a striped directory remotely.
2415 * otherwise, send to the MDT where source is located because regular
2416 * file may open lease.
2418 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2419 * backward compatibility.
2421 if (S_ISDIR(op_data->op_mode) &&
2422 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2423 tgt = lmv_fid2tgt(lmv, &target_fid);
2425 RETURN(PTR_ERR(tgt));
2430 /* cancel UPDATE lock of parent master object */
2431 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2432 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2436 /* cancel UPDATE lock of source parent */
2437 if (sp_tgt != parent_tgt) {
2439 * migrate RPC packs master object FID, because we can only pack
2440 * two FIDs in reint RPC, but MDS needs to know both source
2441 * parent and target parent, and it will obtain them from master
2442 * FID and LMV, the other FID in RPC is kept for target.
2444 * since this FID is not passed to MDC, cancel it anyway.
2446 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2447 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2451 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2453 op_data->op_fid4 = target_fid;
2455 /* cancel UPDATE locks of target parent */
2456 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2457 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2461 /* cancel LOOKUP lock of source if source is remote object */
2462 if (child_tgt != sp_tgt) {
2463 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2464 LCK_EX, MDS_INODELOCK_LOOKUP,
2465 MF_MDC_CANCEL_FID3);
2470 /* cancel ELC locks of source */
2471 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2472 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2476 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2481 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2482 const char *old, size_t oldlen,
2483 const char *new, size_t newlen,
2484 struct ptlrpc_request **request)
2486 struct obd_device *obd = exp->exp_obd;
2487 struct lmv_obd *lmv = &obd->u.lmv;
2488 struct lmv_tgt_desc *sp_tgt;
2489 struct lmv_tgt_desc *tp_tgt = NULL;
2490 struct lmv_tgt_desc *src_tgt = NULL;
2491 struct lmv_tgt_desc *tgt;
2492 struct mdt_body *body;
2497 LASSERT(oldlen != 0);
2499 if (op_data->op_cli_flags & CLI_MIGRATE) {
2500 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2504 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2505 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2506 op_data->op_cap = current_cap();
2508 op_data->op_name = new;
2509 op_data->op_namelen = newlen;
2511 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2513 RETURN(PTR_ERR(tp_tgt));
2515 /* Since the target child might be destroyed, and it might become
2516 * orphan, and we can only check orphan on the local MDT right now, so
2517 * we send rename request to the MDT where target child is located. If
2518 * target child does not exist, then it will send the request to the
2520 if (fid_is_sane(&op_data->op_fid4)) {
2521 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2523 RETURN(PTR_ERR(tgt));
2528 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2530 /* cancel UPDATE locks of target parent */
2531 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2532 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2536 if (fid_is_sane(&op_data->op_fid4)) {
2537 /* cancel LOOKUP lock of target on target parent */
2538 if (tgt != tp_tgt) {
2539 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2540 tgt->ltd_index, LCK_EX,
2541 MDS_INODELOCK_LOOKUP,
2542 MF_MDC_CANCEL_FID4);
2548 if (fid_is_sane(&op_data->op_fid3)) {
2549 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2550 if (IS_ERR(src_tgt))
2551 RETURN(PTR_ERR(src_tgt));
2553 /* cancel ELC locks of source */
2554 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2555 LCK_EX, MDS_INODELOCK_ELC,
2556 MF_MDC_CANCEL_FID3);
2561 op_data->op_name = old;
2562 op_data->op_namelen = oldlen;
2564 sp_tgt = lmv_locate_tgt(lmv, op_data);
2566 RETURN(PTR_ERR(sp_tgt));
2568 /* cancel UPDATE locks of source parent */
2569 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2570 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2574 if (fid_is_sane(&op_data->op_fid3)) {
2575 /* cancel LOOKUP lock of source on source parent */
2576 if (src_tgt != sp_tgt) {
2577 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2578 tgt->ltd_index, LCK_EX,
2579 MDS_INODELOCK_LOOKUP,
2580 MF_MDC_CANCEL_FID3);
2587 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2588 PFID(&op_data->op_fid1), (int)oldlen, old,
2589 PFID(&op_data->op_fid2), (int)newlen, new);
2591 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2593 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2594 ptlrpc_req_finished(*request);
2599 if (rc && rc != -EXDEV)
2602 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2606 /* Not cross-ref case, just get out of here. */
2607 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2610 op_data->op_fid4 = body->mbo_fid1;
2612 ptlrpc_req_finished(*request);
2615 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2617 RETURN(PTR_ERR(tgt));
2619 if (fid_is_sane(&op_data->op_fid4)) {
2620 /* cancel LOOKUP lock of target on target parent */
2621 if (tgt != tp_tgt) {
2622 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2623 tgt->ltd_index, LCK_EX,
2624 MDS_INODELOCK_LOOKUP,
2625 MF_MDC_CANCEL_FID4);
2634 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2635 void *ea, size_t ealen, struct ptlrpc_request **request)
2637 struct obd_device *obd = exp->exp_obd;
2638 struct lmv_obd *lmv = &obd->u.lmv;
2639 struct lmv_tgt_desc *tgt;
2644 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2645 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2646 op_data->op_xvalid);
2648 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2649 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2651 RETURN(PTR_ERR(tgt));
2653 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2658 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2659 struct ptlrpc_request **request)
2661 struct obd_device *obd = exp->exp_obd;
2662 struct lmv_obd *lmv = &obd->u.lmv;
2663 struct lmv_tgt_desc *tgt;
2668 tgt = lmv_fid2tgt(lmv, fid);
2670 RETURN(PTR_ERR(tgt));
2672 rc = md_fsync(tgt->ltd_exp, fid, request);
2676 struct stripe_dirent {
2677 struct page *sd_page;
2678 struct lu_dirpage *sd_dp;
2679 struct lu_dirent *sd_ent;
2683 struct lmv_dir_ctxt {
2684 struct lmv_obd *ldc_lmv;
2685 struct md_op_data *ldc_op_data;
2686 struct md_readdir_info *ldc_mrinfo;
2689 struct stripe_dirent ldc_stripes[0];
2692 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2694 if (stripe->sd_page) {
2695 kunmap(stripe->sd_page);
2696 put_page(stripe->sd_page);
2697 stripe->sd_page = NULL;
2698 stripe->sd_ent = NULL;
2702 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2706 for (i = 0; i < ctxt->ldc_count; i++)
2707 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2710 /* if @ent is dummy, or . .., get next */
2711 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2712 struct lu_dirent *ent,
2715 for (; ent; ent = lu_dirent_next(ent)) {
2716 /* Skip dummy entry */
2717 if (le16_to_cpu(ent->lde_namelen) == 0)
2720 /* skip . and .. for other stripes */
2722 (strncmp(ent->lde_name, ".",
2723 le16_to_cpu(ent->lde_namelen)) == 0 ||
2724 strncmp(ent->lde_name, "..",
2725 le16_to_cpu(ent->lde_namelen)) == 0))
2728 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2735 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2736 struct stripe_dirent *stripe,
2739 struct md_op_data *op_data = ctxt->ldc_op_data;
2740 struct lmv_oinfo *oinfo;
2741 struct lu_fid fid = op_data->op_fid1;
2742 struct inode *inode = op_data->op_data;
2743 struct lmv_tgt_desc *tgt;
2744 struct lu_dirent *ent = stripe->sd_ent;
2745 __u64 hash = ctxt->ldc_hash;
2750 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2754 if (stripe->sd_page) {
2755 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2757 /* @hash should be the last dirent hash */
2758 LASSERTF(hash <= end,
2759 "ctxt@%p stripe@%p hash %llx end %llx\n",
2760 ctxt, stripe, hash, end);
2761 /* unload last page */
2762 stripe_dirent_unload(stripe);
2764 if (end == MDS_DIR_END_OFF) {
2765 stripe->sd_eof = true;
2771 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2772 if (!oinfo->lmo_root) {
2777 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2783 /* op_data is shared by stripes, reset after use */
2784 op_data->op_fid1 = oinfo->lmo_fid;
2785 op_data->op_fid2 = oinfo->lmo_fid;
2786 op_data->op_data = oinfo->lmo_root;
2788 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_mrinfo, hash,
2791 op_data->op_fid1 = fid;
2792 op_data->op_fid2 = fid;
2793 op_data->op_data = inode;
2798 stripe->sd_dp = page_address(stripe->sd_page);
2799 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2801 /* in case a page filled with ., .. and dummy, read next */
2804 stripe->sd_ent = ent;
2807 /* treat error as eof, so dir can be partially accessed */
2808 stripe->sd_eof = true;
2809 ctxt->ldc_mrinfo->mr_partial_readdir_rc = rc;
2810 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2811 "directory is partially accessed!\n",
2812 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2819 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2821 struct obd_device *obd = exp->exp_obd;
2822 struct lmv_obd *lmv = &obd->u.lmv;
2823 struct lmv_tgt_desc *tgt;
2828 rc = lmv_check_connect(obd);
2832 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2834 RETURN(PTR_ERR(tgt));
2836 data->op_flags |= MF_MDC_CANCEL_FID1;
2837 rc = md_file_resync(tgt->ltd_exp, data);
2842 * Get dirent with the closest hash for striped directory
2844 * This function will search the dir entry, whose hash value is the
2845 * closest(>=) to hash from all of sub-stripes, and it is only being called
2846 * for striped directory.
2848 * \param[in] ctxt dir read context
2850 * \retval dirent get the entry successfully
2851 * NULL does not get the entry, normally it means
2852 * it reaches the end of the directory, while read
2853 * stripe dirent error is ignored to allow partial
2856 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2858 struct stripe_dirent *stripe;
2859 struct lu_dirent *ent = NULL;
2863 /* TODO: optimize with k-way merge sort */
2864 for (i = 0; i < ctxt->ldc_count; i++) {
2865 stripe = &ctxt->ldc_stripes[i];
2869 if (!stripe->sd_ent) {
2870 stripe_dirent_load(ctxt, stripe, i);
2871 if (!stripe->sd_ent) {
2872 LASSERT(stripe->sd_eof);
2878 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2879 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2881 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2888 stripe = &ctxt->ldc_stripes[min];
2889 ent = stripe->sd_ent;
2890 /* pop found dirent */
2891 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2899 * Build dir entry page for striped directory
2901 * This function gets one entry by @offset from a striped directory. It will
2902 * read entries from all of stripes, and choose one closest to the required
2903 * offset(&offset). A few notes
2904 * 1. skip . and .. for non-zero stripes, because there can only have one .
2905 * and .. in a directory.
2906 * 2. op_data will be shared by all of stripes, instead of allocating new
2907 * one, so need to restore before reusing.
2909 * \param[in] exp obd export refer to LMV
2910 * \param[in] op_data hold those MD parameters of read_entry
2911 * \param[in] mrinfo ldlm callback being used in enqueue in mdc_read_entry,
2912 * and partial readdir result will be stored in it.
2913 * \param[in] offset starting hash offset
2914 * \param[out] ppage the page holding the entry. Note: because the entry
2915 * will be accessed in upper layer, so we need hold the
2916 * page until the usages of entry is finished, see
2917 * ll_dir_entry_next.
2919 * retval =0 if get entry successfully
2920 * <0 cannot get entry
2922 static int lmv_striped_read_page(struct obd_export *exp,
2923 struct md_op_data *op_data,
2924 struct md_readdir_info *mrinfo, __u64 offset,
2925 struct page **ppage)
2927 struct page *page = NULL;
2928 struct lu_dirpage *dp;
2930 struct lu_dirent *ent;
2931 struct lu_dirent *last_ent;
2933 struct lmv_dir_ctxt *ctxt;
2934 struct lu_dirent *next = NULL;
2940 /* Allocate a page and read entries from all of stripes and fill
2941 * the page by hash order */
2942 page = alloc_page(GFP_KERNEL);
2946 /* Initialize the entry page */
2948 memset(dp, 0, sizeof(*dp));
2949 dp->ldp_hash_start = cpu_to_le64(offset);
2952 left_bytes = PAGE_SIZE - sizeof(*dp);
2956 /* initalize dir read context */
2957 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2958 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2960 GOTO(free_page, rc = -ENOMEM);
2961 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2962 ctxt->ldc_op_data = op_data;
2963 ctxt->ldc_mrinfo = mrinfo;
2964 ctxt->ldc_hash = offset;
2965 ctxt->ldc_count = stripe_count;
2968 next = lmv_dirent_next(ctxt);
2970 /* end of directory */
2972 ctxt->ldc_hash = MDS_DIR_END_OFF;
2975 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2977 ent_size = le16_to_cpu(next->lde_reclen);
2979 /* the last entry lde_reclen is 0, but it might not be the last
2980 * one of this temporay dir page */
2982 ent_size = lu_dirent_calc_size(
2983 le16_to_cpu(next->lde_namelen),
2984 le32_to_cpu(next->lde_attrs));
2986 if (ent_size > left_bytes)
2989 memcpy(ent, next, ent_size);
2991 /* Replace . with master FID and Replace .. with the parent FID
2992 * of master object */
2993 if (strncmp(ent->lde_name, ".",
2994 le16_to_cpu(ent->lde_namelen)) == 0 &&
2995 le16_to_cpu(ent->lde_namelen) == 1)
2996 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2997 else if (strncmp(ent->lde_name, "..",
2998 le16_to_cpu(ent->lde_namelen)) == 0 &&
2999 le16_to_cpu(ent->lde_namelen) == 2)
3000 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
3002 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
3003 le16_to_cpu(ent->lde_namelen), ent->lde_name,
3004 le64_to_cpu(ent->lde_hash));
3006 left_bytes -= ent_size;
3007 ent->lde_reclen = cpu_to_le16(ent_size);
3009 ent = (void *)ent + ent_size;
3012 last_ent->lde_reclen = 0;
3015 dp->ldp_flags |= LDF_EMPTY;
3016 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
3017 dp->ldp_flags |= LDF_COLLIDE;
3018 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
3019 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
3021 put_lmv_dir_ctxt(ctxt);
3022 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
3035 static int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
3036 struct md_readdir_info *mrinfo, __u64 offset,
3037 struct page **ppage)
3039 struct obd_device *obd = exp->exp_obd;
3040 struct lmv_obd *lmv = &obd->u.lmv;
3041 struct lmv_tgt_desc *tgt;
3046 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
3049 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
3050 rc = lmv_striped_read_page(exp, op_data, mrinfo, offset, ppage);
3054 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
3056 RETURN(PTR_ERR(tgt));
3058 rc = md_read_page(tgt->ltd_exp, op_data, mrinfo, offset, ppage);
3064 * Unlink a file/directory
3066 * Unlink a file or directory under the parent dir. The unlink request
3067 * usually will be sent to the MDT where the child is located, but if
3068 * the client does not have the child FID then request will be sent to the
3069 * MDT where the parent is located.
3071 * If the parent is a striped directory then it also needs to locate which
3072 * stripe the name of the child is located, and replace the parent FID
3073 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
3074 * it will walk through all of sub-stripes until the child is being
3077 * \param[in] exp export refer to LMV
3078 * \param[in] op_data different parameters transferred beween client
3079 * MD stacks, name, namelen, FIDs etc.
3080 * op_fid1 is the parent FID, op_fid2 is the child
3082 * \param[out] request point to the request of unlink.
3084 * retval 0 if succeed
3085 * negative errno if failed.
3087 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
3088 struct ptlrpc_request **request)
3090 struct obd_device *obd = exp->exp_obd;
3091 struct lmv_obd *lmv = &obd->u.lmv;
3092 struct lmv_tgt_desc *tgt;
3093 struct lmv_tgt_desc *parent_tgt;
3094 struct mdt_body *body;
3099 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3100 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3101 op_data->op_cap = current_cap();
3104 parent_tgt = lmv_locate_tgt(lmv, op_data);
3105 if (IS_ERR(parent_tgt))
3106 RETURN(PTR_ERR(parent_tgt));
3108 if (likely(!fid_is_zero(&op_data->op_fid2))) {
3109 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3111 RETURN(PTR_ERR(tgt));
3117 * If child's fid is given, cancel unused locks for it if it is from
3118 * another export than parent.
3120 * LOOKUP lock for child (fid3) should also be cancelled on parent
3121 * tgt_tgt in mdc_unlink().
3123 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
3125 if (parent_tgt != tgt)
3126 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
3127 LCK_EX, MDS_INODELOCK_LOOKUP,
3128 MF_MDC_CANCEL_FID3);
3130 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
3131 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
3135 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
3136 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
3139 rc = md_unlink(tgt->ltd_exp, op_data, request);
3140 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
3141 ptlrpc_req_finished(*request);
3149 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
3153 /* Not cross-ref case, just get out of here. */
3154 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
3157 /* This is a remote object, try remote MDT. */
3158 op_data->op_fid2 = body->mbo_fid1;
3159 ptlrpc_req_finished(*request);
3162 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3164 RETURN(PTR_ERR(tgt));
3169 static int lmv_precleanup(struct obd_device *obd)
3172 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
3177 * Get by key a value associated with a LMV device.
3179 * Dispatch request to lower-layer devices as needed.
3181 * \param[in] env execution environment for this thread
3182 * \param[in] exp export for the LMV device
3183 * \param[in] keylen length of key identifier
3184 * \param[in] key identifier of key to get value for
3185 * \param[in] vallen size of \a val
3186 * \param[out] val pointer to storage location for value
3187 * \param[in] lsm optional striping metadata of object
3189 * \retval 0 on success
3190 * \retval negative negated errno on failure
3192 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
3193 __u32 keylen, void *key, __u32 *vallen, void *val)
3195 struct obd_device *obd;
3196 struct lmv_obd *lmv;
3197 struct lu_tgt_desc *tgt;
3202 obd = class_exp2obd(exp);
3204 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3205 exp->exp_handle.h_cookie);
3210 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
3211 LASSERT(*vallen == sizeof(__u32));
3212 lmv_foreach_connected_tgt(lmv, tgt) {
3213 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
3218 } else if (KEY_IS(KEY_MAX_EASIZE) ||
3219 KEY_IS(KEY_DEFAULT_EASIZE) ||
3220 KEY_IS(KEY_CONN_DATA)) {
3222 * Forwarding this request to first MDS, it should know LOV
3225 tgt = lmv_tgt(lmv, 0);
3229 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
3230 if (!rc && KEY_IS(KEY_CONN_DATA))
3231 exp->exp_connect_data = *(struct obd_connect_data *)val;
3233 } else if (KEY_IS(KEY_TGT_COUNT)) {
3234 *((int *)val) = lmv->lmv_mdt_descs.ltd_tgts_size;
3238 CDEBUG(D_IOCTL, "Invalid key\n");
3242 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
3243 int *__rcs, struct ptlrpc_request_set *_set)
3245 struct obd_device *obd = class_exp2obd(exp);
3246 struct ptlrpc_request_set *set = _set;
3247 struct lmv_obd *lmv = &obd->u.lmv;
3248 int tgt_count = lmv->lmv_mdt_count;
3249 struct lu_tgt_desc *tgt;
3250 struct fid_array *fat, **fas = NULL;
3251 int i, rc, **rcs = NULL;
3254 set = ptlrpc_prep_set();
3259 /* split FIDs by targets */
3260 OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
3262 GOTO(out, rc = -ENOMEM);
3263 OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
3265 GOTO(out_fas, rc = -ENOMEM);
3267 for (i = 0; i < fa->fa_nr; i++) {
3270 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
3272 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
3273 PFID(&fa->fa_fids[i]), rc);
3276 LASSERT(idx < tgt_count);
3278 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
3279 fa_fids[fa->fa_nr]));
3281 GOTO(out, rc = -ENOMEM);
3283 OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
3285 GOTO(out, rc = -ENOMEM);
3288 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3291 lmv_foreach_connected_tgt(lmv, tgt) {
3292 fat = fas[tgt->ltd_index];
3293 if (!fat || fat->fa_nr == 0)
3295 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3298 rc = ptlrpc_set_wait(NULL, set);
3301 for (i = 0; i < tgt_count; i++) {
3303 if (!fat || fat->fa_nr == 0)
3305 /* copy FIDs back */
3306 memcpy(fa->fa_fids + j, fat->fa_fids,
3307 fat->fa_nr * sizeof(struct lu_fid));
3309 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3314 ptlrpc_set_destroy(set);
3317 for (i = 0; i < tgt_count; i++) {
3319 OBD_FREE(fas[i], offsetof(struct fid_array,
3320 fa_fids[fa->fa_nr]));
3322 OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3325 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3328 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3334 * Asynchronously set by key a value associated with a LMV device.
3336 * Dispatch request to lower-layer devices as needed.
3338 * \param[in] env execution environment for this thread
3339 * \param[in] exp export for the LMV device
3340 * \param[in] keylen length of key identifier
3341 * \param[in] key identifier of key to store value for
3342 * \param[in] vallen size of value to store
3343 * \param[in] val pointer to data to be stored
3344 * \param[in] set optional list of related ptlrpc requests
3346 * \retval 0 on success
3347 * \retval negative negated errno on failure
3349 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3350 __u32 keylen, void *key, __u32 vallen, void *val,
3351 struct ptlrpc_request_set *set)
3353 struct lmv_tgt_desc *tgt;
3354 struct obd_device *obd;
3355 struct lmv_obd *lmv;
3359 obd = class_exp2obd(exp);
3361 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3362 exp->exp_handle.h_cookie);
3367 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3368 KEY_IS(KEY_DEFAULT_EASIZE)) {
3371 lmv_foreach_connected_tgt(lmv, tgt) {
3372 err = obd_set_info_async(env, tgt->ltd_exp,
3373 keylen, key, vallen, val, set);
3384 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3385 const struct lmv_mds_md_v1 *lmm1)
3387 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3394 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3395 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3396 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3397 if (CFS_FAIL_CHECK(OBD_FAIL_LMV_UNKNOWN_STRIPE))
3398 lsm->lsm_md_hash_type = cfs_fail_val ?: LMV_HASH_TYPE_UNKNOWN;
3400 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3401 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3402 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3403 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3404 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3405 sizeof(lsm->lsm_md_pool_name));
3407 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3410 CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3411 "layout_version %d\n", lsm->lsm_md_stripe_count,
3412 lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3413 lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3414 lsm->lsm_md_layout_version);
3416 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3417 for (i = 0; i < stripe_count; i++) {
3418 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3419 &lmm1->lmv_stripe_fids[i]);
3421 * set default value -1, so lmv_locate_tgt() knows this stripe
3422 * target is not initialized.
3424 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3425 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3428 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3429 &lsm->lsm_md_oinfo[i].lmo_mds);
3436 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3437 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3443 static inline int lmv_unpack_user_md(struct obd_export *exp,
3444 struct lmv_stripe_md *lsm,
3445 const struct lmv_user_md *lmu)
3447 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3448 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3449 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3450 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3451 lsm->lsm_md_max_inherit = lmu->lum_max_inherit;
3452 lsm->lsm_md_max_inherit_rr = lmu->lum_max_inherit_rr;
3453 lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3458 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3459 const union lmv_mds_md *lmm, size_t lmm_size)
3461 struct lmv_stripe_md *lsm;
3464 bool allocated = false;
3467 LASSERT(lsmp != NULL);
3471 if (lsm != NULL && lmm == NULL) {
3473 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3475 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3478 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3480 OBD_FREE_LARGE(lfm, lfm_size);
3484 if (lmv_dir_striped(lsm)) {
3485 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3486 iput(lsm->lsm_md_oinfo[i].lmo_root);
3487 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3489 lsm_size = lmv_stripe_md_size(0);
3491 OBD_FREE(lsm, lsm_size);
3496 /* foreign lmv case */
3497 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3498 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3501 OBD_ALLOC_LARGE(lfm, lmm_size);
3504 *lsmp = (struct lmv_stripe_md *)lfm;
3506 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3507 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3508 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3509 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3510 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3515 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3519 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3520 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3521 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3522 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3527 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3528 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3531 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3532 * stripecount should be 0 then.
3534 lsm_size = lmv_stripe_md_size(0);
3537 OBD_ALLOC(lsm, lsm_size);
3544 switch (le32_to_cpu(lmm->lmv_magic)) {
3546 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3548 case LMV_USER_MAGIC:
3549 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3552 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3553 le32_to_cpu(lmm->lmv_magic));
3558 if (rc != 0 && allocated) {
3559 OBD_FREE(lsm, lsm_size);
3566 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3568 lmv_unpackmd(NULL, &lsm, NULL, 0);
3570 EXPORT_SYMBOL(lmv_free_memmd);
3572 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3573 union ldlm_policy_data *policy,
3574 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3577 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3578 struct lu_tgt_desc *tgt;
3584 LASSERT(fid != NULL);
3586 lmv_foreach_connected_tgt(lmv, tgt) {
3587 if (!tgt->ltd_active)
3590 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3598 static int lmv_set_lock_data(struct obd_export *exp,
3599 const struct lustre_handle *lockh,
3600 void *data, __u64 *bits)
3602 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3603 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3608 if (tgt == NULL || tgt->ltd_exp == NULL)
3610 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3614 static enum ldlm_mode
3615 lmv_lock_match(struct obd_export *exp, __u64 flags,
3616 const struct lu_fid *fid, enum ldlm_type type,
3617 union ldlm_policy_data *policy,
3618 enum ldlm_mode mode, struct lustre_handle *lockh)
3620 struct obd_device *obd = exp->exp_obd;
3621 struct lmv_obd *lmv = &obd->u.lmv;
3622 struct lu_tgt_desc *tgt;
3623 __u64 bits = policy->l_inodebits.bits;
3624 enum ldlm_mode rc = LCK_MINMODE;
3628 /* only one bit is set */
3629 LASSERT(bits && !(bits & (bits - 1)));
3630 /* With DNE every object can have two locks in different namespaces:
3631 * lookup lock in space of MDT storing direntry and update/open lock in
3632 * space of MDT storing inode. Try the MDT that the FID maps to first,
3633 * since this can be easily found, and only try others if that fails.
3635 if (bits == MDS_INODELOCK_LOOKUP) {
3636 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3637 i < lmv->lmv_mdt_descs.ltd_tgts_size; i++,
3638 index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3641 "%s: "DFID" is inaccessible: rc = %d\n",
3642 obd->obd_name, PFID(fid), index);
3645 tgt = lmv_tgt(lmv, index);
3646 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3648 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3649 policy, mode, lockh);
3654 tgt = lmv_fid2tgt(lmv, fid);
3655 if (!IS_ERR(tgt) && tgt->ltd_exp && tgt->ltd_active)
3656 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3657 policy, mode, lockh);
3660 CDEBUG(D_INODE, "Lock match for "DFID": %d\n", PFID(fid), rc);
3666 lmv_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
3667 struct obd_export *dt_exp, struct obd_export *md_exp,
3668 struct lustre_md *md)
3670 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3671 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3673 if (!tgt || !tgt->ltd_exp)
3676 return md_get_lustre_md(tgt->ltd_exp, pill, dt_exp, md_exp, md);
3679 static int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3681 struct obd_device *obd = exp->exp_obd;
3682 struct lmv_obd *lmv = &obd->u.lmv;
3683 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3687 if (md->default_lmv) {
3688 lmv_free_memmd(md->default_lmv);
3689 md->default_lmv = NULL;
3691 if (md->lmv != NULL) {
3692 lmv_free_memmd(md->lmv);
3695 if (!tgt || !tgt->ltd_exp)
3697 RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3700 static int lmv_set_open_replay_data(struct obd_export *exp,
3701 struct obd_client_handle *och,
3702 struct lookup_intent *it)
3704 struct obd_device *obd = exp->exp_obd;
3705 struct lmv_obd *lmv = &obd->u.lmv;
3706 struct lmv_tgt_desc *tgt;
3710 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3712 RETURN(PTR_ERR(tgt));
3714 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3717 static int lmv_clear_open_replay_data(struct obd_export *exp,
3718 struct obd_client_handle *och)
3720 struct obd_device *obd = exp->exp_obd;
3721 struct lmv_obd *lmv = &obd->u.lmv;
3722 struct lmv_tgt_desc *tgt;
3726 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3728 RETURN(PTR_ERR(tgt));
3730 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3733 static int lmv_intent_getattr_async(struct obd_export *exp,
3734 struct md_op_item *item)
3736 struct md_op_data *op_data = &item->mop_data;
3737 struct obd_device *obd = exp->exp_obd;
3738 struct lmv_obd *lmv = &obd->u.lmv;
3739 struct lmv_tgt_desc *ptgt;
3740 struct lmv_tgt_desc *ctgt;
3745 if (!fid_is_sane(&op_data->op_fid2))
3748 ptgt = lmv_locate_tgt(lmv, op_data);
3750 RETURN(PTR_ERR(ptgt));
3752 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3754 RETURN(PTR_ERR(ctgt));
3757 * remote object needs two RPCs to lookup and getattr, considering the
3758 * complexity don't support statahead for now.
3763 rc = md_intent_getattr_async(ptgt->ltd_exp, item);
3768 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3769 struct lu_fid *fid, __u64 *bits)
3771 struct obd_device *obd = exp->exp_obd;
3772 struct lmv_obd *lmv = &obd->u.lmv;
3773 struct lmv_tgt_desc *tgt;
3778 tgt = lmv_fid2tgt(lmv, fid);
3780 RETURN(PTR_ERR(tgt));
3782 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3786 static int lmv_get_fid_from_lsm(struct obd_export *exp,
3787 const struct lmv_stripe_md *lsm,
3788 const char *name, int namelen,
3791 const struct lmv_oinfo *oinfo;
3793 if (!lmv_dir_striped(lsm))
3796 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3798 return PTR_ERR(oinfo);
3800 *fid = oinfo->lmo_fid;
3806 * For lmv, only need to send request to master MDT, and the master MDT will
3807 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3808 * we directly fetch data from the slave MDTs.
3810 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3811 struct obd_quotactl *oqctl)
3813 struct obd_device *obd = class_exp2obd(exp);
3814 struct lmv_obd *lmv = &obd->u.lmv;
3815 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3816 __u64 curspace, curinodes;
3821 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3822 CERROR("master lmv inactive\n");
3826 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3827 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3831 curspace = curinodes = 0;
3832 lmv_foreach_connected_tgt(lmv, tgt) {
3835 if (!tgt->ltd_active)
3838 err = obd_quotactl(tgt->ltd_exp, oqctl);
3840 CERROR("getquota on mdt %d failed. %d\n",
3841 tgt->ltd_index, err);
3845 curspace += oqctl->qc_dqblk.dqb_curspace;
3846 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3849 oqctl->qc_dqblk.dqb_curspace = curspace;
3850 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3855 static int lmv_merge_attr(struct obd_export *exp,
3856 const struct lmv_stripe_md *lsm,
3857 struct cl_attr *attr,
3858 ldlm_blocking_callback cb_blocking)
3863 if (!lmv_dir_striped(lsm))
3866 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3870 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3871 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3877 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3878 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3879 i_size_read(inode), (unsigned long long)inode->i_blocks,
3880 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3881 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3883 /* for slave stripe, it needs to subtract nlink for . and .. */
3885 attr->cat_nlink += inode->i_nlink - 2;
3887 attr->cat_nlink = inode->i_nlink;
3889 attr->cat_size += i_size_read(inode);
3890 attr->cat_blocks += inode->i_blocks;
3892 if (attr->cat_atime < inode->i_atime.tv_sec)
3893 attr->cat_atime = inode->i_atime.tv_sec;
3895 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3896 attr->cat_ctime = inode->i_ctime.tv_sec;
3898 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3899 attr->cat_mtime = inode->i_mtime.tv_sec;
3904 static struct lu_batch *lmv_batch_create(struct obd_export *exp,
3905 enum lu_batch_flags flags,
3908 struct lu_batch *bh;
3909 struct lmv_batch *lbh;
3914 RETURN(ERR_PTR(-ENOMEM));
3916 bh = &lbh->lbh_super;
3917 bh->lbt_flags = flags;
3918 bh->lbt_max_count = max_count;
3920 if (flags & BATCH_FL_RQSET) {
3921 bh->lbt_rqset = ptlrpc_prep_set();
3922 if (bh->lbt_rqset == NULL) {
3924 RETURN(ERR_PTR(-ENOMEM));
3928 INIT_LIST_HEAD(&lbh->lbh_sub_batch_list);
3932 static int lmv_batch_stop(struct obd_export *exp, struct lu_batch *bh)
3934 struct lmv_batch *lbh;
3935 struct lmvsub_batch *sub;
3936 struct lmvsub_batch *tmp;
3941 lbh = container_of(bh, struct lmv_batch, lbh_super);
3942 list_for_each_entry_safe(sub, tmp, &lbh->lbh_sub_batch_list,
3944 list_del(&sub->sbh_sub_item);
3945 rc = md_batch_stop(sub->sbh_tgt->ltd_exp, sub->sbh_sub);
3947 CERROR("%s: stop batch processing failed: rc = %d\n",
3948 exp->exp_obd->obd_name, rc);
3949 if (bh->lbt_result == 0)
3950 bh->lbt_result = rc;
3955 if (bh->lbt_flags & BATCH_FL_RQSET) {
3956 rc = ptlrpc_set_wait(NULL, bh->lbt_rqset);
3957 ptlrpc_set_destroy(bh->lbt_rqset);
3964 static int lmv_batch_flush(struct obd_export *exp, struct lu_batch *bh,
3967 struct lmv_batch *lbh;
3968 struct lmvsub_batch *sub;
3974 lbh = container_of(bh, struct lmv_batch, lbh_super);
3975 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
3976 rc1 = md_batch_flush(sub->sbh_tgt->ltd_exp, sub->sbh_sub, wait);
3978 CERROR("%s: stop batch processing failed: rc = %d\n",
3979 exp->exp_obd->obd_name, rc);
3980 if (bh->lbt_result == 0)
3981 bh->lbt_result = rc;
3988 if (wait && bh->lbt_flags & BATCH_FL_RQSET) {
3989 rc1 = ptlrpc_set_wait(NULL, bh->lbt_rqset);
3997 static inline struct lmv_tgt_desc *
3998 lmv_batch_locate_tgt(struct lmv_obd *lmv, struct md_op_item *item)
4000 struct lmv_tgt_desc *tgt;
4002 switch (item->mop_opc) {
4004 tgt = ERR_PTR(-ENOTSUPP);
4010 struct lu_batch *lmv_batch_lookup_sub(struct lmv_batch *lbh,
4011 struct lmv_tgt_desc *tgt)
4013 struct lmvsub_batch *sub;
4015 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
4016 if (sub->sbh_tgt == tgt)
4017 return sub->sbh_sub;
4023 struct lu_batch *lmv_batch_get_sub(struct lmv_batch *lbh,
4024 struct lmv_tgt_desc *tgt)
4026 struct lmvsub_batch *sbh;
4027 struct lu_batch *child_bh;
4028 struct lu_batch *bh;
4032 child_bh = lmv_batch_lookup_sub(lbh, tgt);
4033 if (child_bh != NULL)
4038 RETURN(ERR_PTR(-ENOMEM));
4040 INIT_LIST_HEAD(&sbh->sbh_sub_item);
4043 bh = &lbh->lbh_super;
4044 child_bh = md_batch_create(tgt->ltd_exp, bh->lbt_flags,
4046 if (IS_ERR(child_bh)) {
4051 child_bh->lbt_rqset = bh->lbt_rqset;
4052 sbh->sbh_sub = child_bh;
4053 list_add(&sbh->sbh_sub_item, &lbh->lbh_sub_batch_list);
4057 static int lmv_batch_add(struct obd_export *exp, struct lu_batch *bh,
4058 struct md_op_item *item)
4060 struct obd_device *obd = exp->exp_obd;
4061 struct lmv_obd *lmv = &obd->u.lmv;
4062 struct lmv_tgt_desc *tgt;
4063 struct lmv_batch *lbh;
4064 struct lu_batch *child_bh;
4069 tgt = lmv_batch_locate_tgt(lmv, item);
4071 RETURN(PTR_ERR(tgt));
4073 lbh = container_of(bh, struct lmv_batch, lbh_super);
4074 child_bh = lmv_batch_get_sub(lbh, tgt);
4075 if (IS_ERR(child_bh))
4076 RETURN(PTR_ERR(child_bh));
4078 rc = md_batch_add(tgt->ltd_exp, child_bh, item);
4082 static const struct obd_ops lmv_obd_ops = {
4083 .o_owner = THIS_MODULE,
4084 .o_setup = lmv_setup,
4085 .o_cleanup = lmv_cleanup,
4086 .o_precleanup = lmv_precleanup,
4087 .o_process_config = lmv_process_config,
4088 .o_connect = lmv_connect,
4089 .o_disconnect = lmv_disconnect,
4090 .o_statfs = lmv_statfs,
4091 .o_get_info = lmv_get_info,
4092 .o_set_info_async = lmv_set_info_async,
4093 .o_notify = lmv_notify,
4094 .o_get_uuid = lmv_get_uuid,
4095 .o_fid_alloc = lmv_fid_alloc,
4096 .o_iocontrol = lmv_iocontrol,
4097 .o_quotactl = lmv_quotactl
4100 static const struct md_ops lmv_md_ops = {
4101 .m_get_root = lmv_get_root,
4102 .m_null_inode = lmv_null_inode,
4103 .m_close = lmv_close,
4104 .m_create = lmv_create,
4105 .m_enqueue = lmv_enqueue,
4106 .m_getattr = lmv_getattr,
4107 .m_getxattr = lmv_getxattr,
4108 .m_getattr_name = lmv_getattr_name,
4109 .m_intent_lock = lmv_intent_lock,
4111 .m_rename = lmv_rename,
4112 .m_setattr = lmv_setattr,
4113 .m_setxattr = lmv_setxattr,
4114 .m_fsync = lmv_fsync,
4115 .m_file_resync = lmv_file_resync,
4116 .m_read_page = lmv_read_page,
4117 .m_unlink = lmv_unlink,
4118 .m_init_ea_size = lmv_init_ea_size,
4119 .m_cancel_unused = lmv_cancel_unused,
4120 .m_set_lock_data = lmv_set_lock_data,
4121 .m_lock_match = lmv_lock_match,
4122 .m_get_lustre_md = lmv_get_lustre_md,
4123 .m_free_lustre_md = lmv_free_lustre_md,
4124 .m_merge_attr = lmv_merge_attr,
4125 .m_set_open_replay_data = lmv_set_open_replay_data,
4126 .m_clear_open_replay_data = lmv_clear_open_replay_data,
4127 .m_intent_getattr_async = lmv_intent_getattr_async,
4128 .m_revalidate_lock = lmv_revalidate_lock,
4129 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
4130 .m_unpackmd = lmv_unpackmd,
4131 .m_rmfid = lmv_rmfid,
4132 .m_batch_create = lmv_batch_create,
4133 .m_batch_add = lmv_batch_add,
4134 .m_batch_stop = lmv_batch_stop,
4135 .m_batch_flush = lmv_batch_flush,
4138 static int __init lmv_init(void)
4140 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true,
4141 LUSTRE_LMV_NAME, NULL);
4144 static void __exit lmv_exit(void)
4146 class_unregister_type(LUSTRE_LMV_NAME);
4149 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4150 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
4151 MODULE_VERSION(LUSTRE_VERSION_STRING);
4152 MODULE_LICENSE("GPL");
4154 module_init(lmv_init);
4155 module_exit(lmv_exit);