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);
859 /* exit early for unknown ioctl types */
860 if (unlikely(_IOC_TYPE(cmd) != 'f' && cmd != IOC_OSC_SET_ACTIVE))
861 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "unknown", -ENOTTY));
863 /* handle commands that don't use @karg first */
865 case LL_IOC_GET_CONNECT_FLAGS:
866 tgt = lmv_tgt(lmv, 0);
868 if (tgt && tgt->ltd_exp)
869 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, NULL, uarg);
873 if (unlikely(karg == NULL))
874 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", -EINVAL));
877 case IOC_OBD_STATFS: {
878 struct obd_ioctl_data *data = karg;
879 struct obd_device *mdc_obd;
880 struct obd_statfs stat_buf = {0};
883 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
885 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
888 tgt = lmv_tgt(lmv, index);
892 if (!tgt->ltd_active)
895 mdc_obd = class_exp2obd(tgt->ltd_exp);
900 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
901 min((int) data->ioc_plen2,
902 (int) sizeof(struct obd_uuid))))
905 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
906 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
910 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
911 min((int) data->ioc_plen1,
912 (int) sizeof(stat_buf))))
916 case OBD_IOC_QUOTACTL: {
917 struct if_quotactl *qctl = karg;
918 struct obd_quotactl *oqctl;
919 struct obd_import *imp;
921 if (qctl->qc_valid == QC_MDTIDX) {
922 tgt = lmv_tgt(lmv, qctl->qc_idx);
923 } else if (qctl->qc_valid == QC_UUID) {
924 lmv_foreach_tgt(lmv, tgt) {
925 if (!obd_uuid_equals(&tgt->ltd_uuid,
944 imp = class_exp2cliimp(tgt->ltd_exp);
945 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
946 qctl->qc_valid = QC_MDTIDX;
947 qctl->obd_uuid = tgt->ltd_uuid;
951 OBD_ALLOC_PTR(oqctl);
955 QCTL_COPY(oqctl, qctl);
956 rc = obd_quotactl(tgt->ltd_exp, oqctl);
958 QCTL_COPY(qctl, oqctl);
959 qctl->qc_valid = QC_MDTIDX;
960 qctl->obd_uuid = tgt->ltd_uuid;
965 case LL_IOC_FID2MDTIDX: {
966 struct lu_fid *fid = karg;
969 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
973 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
974 * point to user space memory for FID2MDTIDX. */
975 *(__u32 *)uarg = mdt_index;
978 case OBD_IOC_FID2PATH: {
979 rc = lmv_fid2path(exp, len, karg, uarg);
982 case LL_IOC_HSM_STATE_GET:
983 case LL_IOC_HSM_STATE_SET:
984 case LL_IOC_HSM_ACTION: {
985 struct md_op_data *op_data = karg;
987 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
989 RETURN(PTR_ERR(tgt));
991 if (tgt->ltd_exp == NULL)
994 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
997 case LL_IOC_HSM_PROGRESS: {
998 const struct hsm_progress_kernel *hpk = karg;
1000 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
1002 RETURN(PTR_ERR(tgt));
1003 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1006 case LL_IOC_HSM_REQUEST: {
1007 struct hsm_user_request *hur = karg;
1008 unsigned int reqcount = hur->hur_request.hr_itemcount;
1013 /* if the request is about a single fid
1014 * or if there is a single MDS, no need to split
1016 if (reqcount == 1 || count == 1) {
1017 tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
1019 RETURN(PTR_ERR(tgt));
1020 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1022 /* split fid list to their respective MDS */
1023 lmv_foreach_connected_tgt(lmv, tgt) {
1026 struct hsm_user_request *req;
1028 nr = lmv_hsm_req_count(lmv, hur, tgt);
1031 if (nr == 0) /* nothing for this MDS */
1034 /* build a request with fids for this MDS */
1035 reqlen = offsetof(typeof(*hur),
1037 + hur->hur_request.hr_data_len;
1038 OBD_ALLOC_LARGE(req, reqlen);
1041 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1043 GOTO(hsm_req_err, rc1);
1044 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1047 if (rc1 != 0 && rc == 0)
1049 OBD_FREE_LARGE(req, reqlen);
1054 case LL_IOC_LOV_SWAP_LAYOUTS: {
1055 struct md_op_data *op_data = karg;
1056 struct lmv_tgt_desc *tgt1, *tgt2;
1058 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1060 RETURN(PTR_ERR(tgt1));
1062 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1064 RETURN(PTR_ERR(tgt2));
1066 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1069 /* only files on same MDT can have their layouts swapped */
1070 if (tgt1->ltd_index != tgt2->ltd_index)
1073 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1076 case LL_IOC_HSM_CT_START: {
1077 struct lustre_kernelcomm *lk = karg;
1079 if (lk->lk_flags & LK_FLG_STOP)
1080 rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1082 rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1086 lmv_foreach_connected_tgt(lmv, tgt) {
1087 struct obd_device *mdc_obd;
1090 /* ll_umount_begin() sets force flag but for lmv, not
1091 * mdc. Let's pass it through */
1092 mdc_obd = class_exp2obd(tgt->ltd_exp);
1093 mdc_obd->obd_force = obd->obd_force;
1094 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1096 if (tgt->ltd_active) {
1097 OBD_IOC_ERROR(obd->obd_name, cmd,
1098 tgt->ltd_uuid.uuid, err);
1101 if (unlikely(err == -ENOTTY))
1115 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1116 struct lu_fid *fid, struct md_op_data *op_data)
1118 struct obd_device *obd = class_exp2obd(exp);
1119 struct lmv_obd *lmv = &obd->u.lmv;
1120 struct lmv_tgt_desc *tgt;
1128 tgt = lmv_tgt(lmv, op_data->op_mds);
1132 if (!tgt->ltd_active || !tgt->ltd_exp)
1136 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1137 * on server that seq in new allocated fid is not yet known.
1139 mutex_lock(&tgt->ltd_fid_mutex);
1140 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1141 mutex_unlock(&tgt->ltd_fid_mutex);
1143 LASSERT(fid_is_sane(fid));
1150 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1152 struct lmv_obd *lmv = &obd->u.lmv;
1153 struct lmv_desc *desc;
1154 struct lnet_processid lnet_id;
1160 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1161 CERROR("LMV setup requires a descriptor\n");
1165 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1166 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1167 CERROR("Lmv descriptor size wrong: %d > %d\n",
1168 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1172 obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1173 desc->ld_uuid.uuid);
1174 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1175 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1176 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1177 LMV_DESC_QOS_MAXAGE_DEFAULT;
1178 lmv->max_def_easize = 0;
1179 lmv->max_easize = 0;
1181 spin_lock_init(&lmv->lmv_lock);
1184 * initialize rr_index to lower 32bit of netid, so that client
1185 * can distribute subdirs evenly from the beginning.
1187 while (LNetGetId(i++, &lnet_id) != -ENOENT) {
1188 if (!nid_is_lo0(&lnet_id.nid)) {
1189 lmv->lmv_qos_rr_index = ntohl(lnet_id.nid.nid_addr[0]);
1194 rc = lmv_tunables_init(obd);
1196 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1199 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1200 LUSTRE_CLI_FLD_HASH_DHT);
1202 CERROR("Can't init FLD, err %d\n", rc);
1204 rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1206 CWARN("%s: error initialize target table: rc = %d\n",
1212 static int lmv_cleanup(struct obd_device *obd)
1214 struct lmv_obd *lmv = &obd->u.lmv;
1215 struct lu_tgt_desc *tgt;
1216 struct lu_tgt_desc *tmp;
1220 fld_client_fini(&lmv->lmv_fld);
1221 fld_client_debugfs_fini(&lmv->lmv_fld);
1223 lprocfs_obd_cleanup(obd);
1224 lprocfs_free_md_stats(obd);
1226 lmv_foreach_tgt_safe(lmv, tgt, tmp)
1227 lmv_del_target(lmv, tgt);
1228 lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1233 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1235 struct lustre_cfg *lcfg = buf;
1236 struct obd_uuid obd_uuid;
1242 switch (lcfg->lcfg_command) {
1244 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1245 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1246 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1247 GOTO(out, rc = -EINVAL);
1249 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1251 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1252 GOTO(out, rc = -EINVAL);
1253 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1254 GOTO(out, rc = -EINVAL);
1255 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1258 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1259 GOTO(out, rc = -EINVAL);
1265 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1269 if (flags & OBD_STATFS_FOR_MDT0)
1272 if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1273 return lmv->lmv_statfs_start;
1275 /* choose initial MDT for this client */
1277 struct lnet_processid lnet_id;
1278 if (LNetGetId(i, &lnet_id) == -ENOENT)
1281 if (!nid_is_lo0(&lnet_id.nid)) {
1282 /* We dont need a full 64-bit modulus, just enough
1283 * to distribute the requests across MDTs evenly.
1285 lmv->lmv_statfs_start = nidhash(&lnet_id.nid) %
1291 return lmv->lmv_statfs_start;
1294 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1295 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1297 struct obd_device *obd = class_exp2obd(exp);
1298 struct lmv_obd *lmv = &obd->u.lmv;
1299 struct obd_statfs *temp;
1300 struct lu_tgt_desc *tgt;
1308 OBD_ALLOC(temp, sizeof(*temp));
1312 /* distribute statfs among MDTs */
1313 idx = lmv_select_statfs_mdt(lmv, flags);
1315 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1316 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1317 tgt = lmv_tgt(lmv, idx);
1318 if (!tgt || !tgt->ltd_exp)
1321 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age,
1322 flags | OBD_STATFS_NESTED);
1324 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1325 tgt->ltd_exp->exp_obd->obd_name, i, rc);
1327 /* Try another MDT */
1328 if (flags & OBD_STATFS_SUM)
1330 GOTO(out_free_temp, rc);
1333 if (temp->os_state & OS_STATFS_SUM ||
1334 flags == OBD_STATFS_FOR_MDT0) {
1335 /* reset to the last aggregated values
1336 * and don't sum with non-aggrated data */
1337 /* If the statfs is from mount, it needs to retrieve
1338 * necessary information from MDT0. i.e. mount does
1339 * not need the merged osfs from all of MDT. Also
1340 * clients can be mounted as long as MDT0 is in
1343 GOTO(out_free_temp, rc);
1349 osfs->os_bavail += temp->os_bavail;
1350 osfs->os_blocks += temp->os_blocks;
1351 osfs->os_ffree += temp->os_ffree;
1352 osfs->os_files += temp->os_files;
1353 osfs->os_granted += temp->os_granted;
1356 /* There is no stats from some MDTs, data incomplete */
1360 OBD_FREE(temp, sizeof(*temp));
1364 static int lmv_statfs_update(void *cookie, int rc)
1366 struct obd_info *oinfo = cookie;
1367 struct obd_device *obd = oinfo->oi_obd;
1368 struct lmv_obd *lmv = &obd->u.lmv;
1369 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1370 struct obd_statfs *osfs = oinfo->oi_osfs;
1373 * NB: don't deactivate TGT upon error, because we may not trigger async
1374 * statfs any longer, then there is no chance to activate TGT.
1377 spin_lock(&lmv->lmv_lock);
1378 tgt->ltd_statfs = *osfs;
1379 tgt->ltd_statfs_age = ktime_get_seconds();
1380 spin_unlock(&lmv->lmv_lock);
1381 set_bit(LQ_DIRTY, &lmv->lmv_qos.lq_flags);
1387 /* update tgt statfs async if it's ld_qos_maxage old */
1388 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1390 struct obd_info oinfo = {
1393 .oi_cb_up = lmv_statfs_update,
1397 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1398 obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1401 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1406 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1409 struct obd_device *obd = exp->exp_obd;
1410 struct lmv_obd *lmv = &obd->u.lmv;
1411 struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1419 rc = md_get_root(tgt->ltd_exp, fileset, fid);
1423 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1424 u64 obd_md_valid, const char *name, size_t buf_size,
1425 struct ptlrpc_request **req)
1427 struct obd_device *obd = exp->exp_obd;
1428 struct lmv_obd *lmv = &obd->u.lmv;
1429 struct lmv_tgt_desc *tgt;
1434 tgt = lmv_fid2tgt(lmv, fid);
1436 RETURN(PTR_ERR(tgt));
1438 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1443 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1444 u64 obd_md_valid, const char *name,
1445 const void *value, size_t value_size,
1446 unsigned int xattr_flags, u32 suppgid,
1447 struct ptlrpc_request **req)
1449 struct obd_device *obd = exp->exp_obd;
1450 struct lmv_obd *lmv = &obd->u.lmv;
1451 struct lmv_tgt_desc *tgt;
1456 tgt = lmv_fid2tgt(lmv, fid);
1458 RETURN(PTR_ERR(tgt));
1460 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1461 value, value_size, xattr_flags, suppgid, req);
1466 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1467 struct ptlrpc_request **request)
1469 struct obd_device *obd = exp->exp_obd;
1470 struct lmv_obd *lmv = &obd->u.lmv;
1471 struct lmv_tgt_desc *tgt;
1476 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1478 RETURN(PTR_ERR(tgt));
1480 if (op_data->op_flags & MF_GET_MDT_IDX) {
1481 op_data->op_mds = tgt->ltd_index;
1485 rc = md_getattr(tgt->ltd_exp, op_data, request);
1490 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1492 struct obd_device *obd = exp->exp_obd;
1493 struct lmv_obd *lmv = &obd->u.lmv;
1494 struct lu_tgt_desc *tgt;
1498 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1501 * With DNE every object can have two locks in different namespaces:
1502 * lookup lock in space of MDT storing direntry and update/open lock in
1503 * space of MDT storing inode.
1505 lmv_foreach_connected_tgt(lmv, tgt)
1506 md_null_inode(tgt->ltd_exp, fid);
1511 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1512 struct md_open_data *mod, struct ptlrpc_request **request)
1514 struct obd_device *obd = exp->exp_obd;
1515 struct lmv_obd *lmv = &obd->u.lmv;
1516 struct lmv_tgt_desc *tgt;
1521 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1523 RETURN(PTR_ERR(tgt));
1525 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1526 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1530 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv,
1531 struct md_op_data *op_data)
1533 struct lu_tgt_desc *tgt, *cur = NULL;
1534 __u64 total_avail = 0;
1535 __u64 total_weight = 0;
1536 __u64 cur_weight = 0;
1537 int total_usable = 0;
1543 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1544 RETURN(ERR_PTR(-EAGAIN));
1546 down_write(&lmv->lmv_qos.lq_rw_sem);
1548 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1549 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1551 rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1553 GOTO(unlock, tgt = ERR_PTR(rc));
1555 lmv_foreach_tgt(lmv, tgt) {
1556 if (!tgt->ltd_exp || !tgt->ltd_active) {
1557 tgt->ltd_qos.ltq_usable = 0;
1561 tgt->ltd_qos.ltq_usable = 1;
1562 lu_tgt_qos_weight_calc(tgt, true);
1563 if (tgt->ltd_index == op_data->op_mds)
1565 total_avail += tgt->ltd_qos.ltq_avail;
1566 total_weight += tgt->ltd_qos.ltq_weight;
1570 /* If current MDT has above-average space and dir is not aleady using
1571 * round-robin to spread across more MDTs, stay on the parent MDT
1572 * to avoid creating needless remote MDT directories. Remote dirs
1573 * close to the root balance space more effectively than bottom dirs,
1574 * so prefer to create remote dirs at top level of directory tree.
1575 * "16 / (dir_depth + 10)" is the factor to make it less likely
1576 * for top-level directories to stay local unless they have more than
1577 * average free space, while deep dirs prefer local until more full.
1578 * depth=0 -> 160%, depth=3 -> 123%, depth=6 -> 100%,
1579 * depth=9 -> 84%, depth=12 -> 73%, depth=15 -> 64%
1581 if (!lmv_op_default_rr_mkdir(op_data)) {
1582 rand = total_avail * 16 /
1583 (total_usable * (op_data->op_dir_depth + 10));
1584 if (cur && cur->ltd_qos.ltq_avail >= rand) {
1590 rand = lu_prandom_u64_max(total_weight);
1592 lmv_foreach_connected_tgt(lmv, tgt) {
1593 if (!tgt->ltd_qos.ltq_usable)
1596 cur_weight += tgt->ltd_qos.ltq_weight;
1597 if (cur_weight < rand)
1600 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1604 /* no proper target found */
1605 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1607 up_write(&lmv->lmv_qos.lq_rw_sem);
1612 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv)
1614 struct lu_tgt_desc *tgt;
1620 spin_lock(&lmv->lmv_lock);
1621 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1622 index = (i + lmv->lmv_qos_rr_index) %
1623 lmv->lmv_mdt_descs.ltd_tgts_size;
1624 tgt = lmv_tgt(lmv, index);
1625 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1628 lmv->lmv_qos_rr_index = (tgt->ltd_index + 1) %
1629 lmv->lmv_mdt_descs.ltd_tgts_size;
1630 spin_unlock(&lmv->lmv_lock);
1634 spin_unlock(&lmv->lmv_lock);
1636 RETURN(ERR_PTR(-ENODEV));
1639 /* locate MDT which is less full (avoid the most full MDT) */
1640 static struct lu_tgt_desc *lmv_locate_tgt_lf(struct lmv_obd *lmv)
1642 struct lu_tgt_desc *min = NULL;
1643 struct lu_tgt_desc *tgt;
1649 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1650 RETURN(ERR_PTR(-EAGAIN));
1652 down_write(&lmv->lmv_qos.lq_rw_sem);
1654 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1655 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1657 lmv_foreach_tgt(lmv, tgt) {
1658 if (!tgt->ltd_exp || !tgt->ltd_active) {
1659 tgt->ltd_qos.ltq_usable = 0;
1663 tgt->ltd_qos.ltq_usable = 1;
1664 lu_tgt_qos_weight_calc(tgt, true);
1665 avail += tgt->ltd_qos.ltq_avail;
1666 if (!min || min->ltd_qos.ltq_avail > tgt->ltd_qos.ltq_avail)
1670 /* avoid the most full MDT */
1672 avail -= min->ltd_qos.ltq_avail;
1674 rand = lu_prandom_u64_max(avail);
1676 lmv_foreach_connected_tgt(lmv, tgt) {
1677 if (!tgt->ltd_qos.ltq_usable)
1683 avail += tgt->ltd_qos.ltq_avail;
1690 /* no proper target found */
1691 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1693 up_write(&lmv->lmv_qos.lq_rw_sem);
1698 /* locate MDT by file name, for striped directory, the file name hash decides
1699 * which stripe its dirent is stored.
1701 static struct lmv_tgt_desc *
1702 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1703 const char *name, int namelen, struct lu_fid *fid,
1704 __u32 *mds, bool new_layout)
1706 struct lmv_tgt_desc *tgt;
1707 const struct lmv_oinfo *oinfo;
1709 if (!lmv_dir_striped(lsm) || !namelen) {
1710 tgt = lmv_fid2tgt(lmv, fid);
1714 *mds = tgt->ltd_index;
1718 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1719 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1720 return ERR_PTR(-EBADF);
1721 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1723 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, new_layout);
1725 return ERR_CAST(oinfo);
1728 /* check stripe FID is sane */
1729 if (!fid_is_sane(&oinfo->lmo_fid))
1730 return ERR_PTR(-ENODEV);
1732 *fid = oinfo->lmo_fid;
1733 *mds = oinfo->lmo_mds;
1734 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1736 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1738 return tgt ? tgt : ERR_PTR(-ENODEV);
1742 * Locate MDT of op_data->op_fid1
1744 * For striped directory, it will locate the stripe by name hash, if hash_type
1745 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1746 * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1747 * indicates whether old or new layout is used to locate.
1749 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1751 * \param[in] lmv LMV device
1752 * \param[in/out] op_data client MD stack parameters, name, namelen etc,
1753 * op_mds and op_fid1 will be updated if op_mea1
1754 * indicates fid1 represents a striped directory.
1756 * retval pointer to the lmv_tgt_desc if succeed.
1757 * ERR_PTR(errno) if failed.
1759 struct lmv_tgt_desc *
1760 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1762 struct lmv_stripe_md *lsm = op_data->op_mea1;
1763 struct lmv_oinfo *oinfo;
1764 struct lmv_tgt_desc *tgt;
1766 if (lmv_dir_foreign(lsm))
1767 return ERR_PTR(-ENODATA);
1769 /* During creating VOLATILE file, it should honor the mdt
1770 * index if the file under striped dir is being restored, see
1772 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1773 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1774 tgt = lmv_tgt(lmv, op_data->op_mds);
1776 return ERR_PTR(-ENODEV);
1778 if (lmv_dir_striped(lsm)) {
1781 /* refill the right parent fid */
1782 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1783 oinfo = &lsm->lsm_md_oinfo[i];
1784 if (oinfo->lmo_mds == op_data->op_mds) {
1785 op_data->op_fid1 = oinfo->lmo_fid;
1790 if (i == lsm->lsm_md_stripe_count)
1791 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1793 } else if (lmv_dir_bad_hash(lsm)) {
1794 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1795 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1797 op_data->op_fid1 = oinfo->lmo_fid;
1798 op_data->op_mds = oinfo->lmo_mds;
1799 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1801 return ERR_PTR(-ENODEV);
1803 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1804 op_data->op_name, op_data->op_namelen,
1805 &op_data->op_fid1, &op_data->op_mds,
1806 op_data->op_new_layout);
1812 /* Locate MDT of op_data->op_fid2 for link/rename */
1813 static struct lmv_tgt_desc *
1814 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1816 struct lmv_tgt_desc *tgt;
1819 LASSERT(op_data->op_name);
1820 if (lmv_dir_layout_changing(op_data->op_mea2)) {
1821 struct lu_fid fid1 = op_data->op_fid1;
1822 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1823 struct ptlrpc_request *request = NULL;
1826 * avoid creating new file under old layout of migrating
1827 * directory, check it here.
1829 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1830 op_data->op_name, op_data->op_namelen,
1831 &op_data->op_fid2, &op_data->op_mds, false);
1835 op_data->op_fid1 = op_data->op_fid2;
1836 op_data->op_mea1 = op_data->op_mea2;
1837 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1838 op_data->op_fid1 = fid1;
1839 op_data->op_mea1 = lsm1;
1841 ptlrpc_req_finished(request);
1842 RETURN(ERR_PTR(-EEXIST));
1846 RETURN(ERR_PTR(rc));
1849 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1850 op_data->op_namelen, &op_data->op_fid2,
1851 &op_data->op_mds, true);
1854 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1856 struct lu_tgt_desc *tgt;
1857 struct ptlrpc_request *request;
1860 LASSERT(lmv_dir_layout_changing(op_data->op_mea1));
1861 LASSERT(!op_data->op_new_layout);
1863 tgt = lmv_locate_tgt(lmv, op_data);
1865 return PTR_ERR(tgt);
1867 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1869 ptlrpc_req_finished(request);
1876 /* mkdir by QoS upon 'lfs mkdir -i -1'.
1878 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1879 * directories under striped directory.
1881 static inline bool lmv_op_user_qos_mkdir(const struct md_op_data *op_data)
1883 const struct lmv_user_md *lum = op_data->op_data;
1885 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1888 if (lmv_dir_striped(op_data->op_mea1))
1891 return (op_data->op_cli_flags & CLI_SET_MEA) && lum &&
1892 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
1893 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
1896 /* mkdir by QoS if either ROOT or parent default LMV is space balanced. */
1897 static inline bool lmv_op_default_qos_mkdir(const struct md_op_data *op_data)
1899 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1901 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1904 if (lmv_dir_striped(op_data->op_mea1))
1907 return (op_data->op_flags & MF_QOS_MKDIR) ||
1908 (lsm && lsm->lsm_md_master_mdt_index == LMV_OFFSET_DEFAULT);
1911 /* if parent default LMV is space balanced, and
1912 * 1. max_inherit_rr is set
1913 * 2. or parent is ROOT
1914 * mkdir roundrobin. Or if parent doesn't have default LMV, while ROOT default
1915 * LMV requests roundrobin mkdir, do the same.
1916 * NB, this needs to check server is balanced, which is done by caller.
1918 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data)
1920 const struct lmv_stripe_md *lsm = op_data->op_default_mea1;
1922 return (op_data->op_flags & MF_RR_MKDIR) ||
1923 (lsm && lsm->lsm_md_max_inherit_rr != LMV_INHERIT_RR_NONE) ||
1924 fid_is_root(&op_data->op_fid1);
1927 /* 'lfs mkdir -i <specific_MDT>' */
1928 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1930 const struct lmv_user_md *lum = op_data->op_data;
1932 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1933 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1934 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1935 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1936 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1939 /* parent default LMV master_mdt_index is not -1. */
1941 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1943 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1944 op_data->op_default_mea1 &&
1945 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1949 /* locate MDT by space usage */
1950 static struct lu_tgt_desc *lmv_locate_tgt_by_space(struct lmv_obd *lmv,
1951 struct md_op_data *op_data,
1952 struct lmv_tgt_desc *tgt)
1954 struct lmv_tgt_desc *tmp = tgt;
1956 tgt = lmv_locate_tgt_qos(lmv, op_data);
1957 if (tgt == ERR_PTR(-EAGAIN)) {
1958 if (ltd_qos_is_balanced(&lmv->lmv_mdt_descs) &&
1959 !lmv_op_default_rr_mkdir(op_data) &&
1960 !lmv_op_user_qos_mkdir(op_data))
1961 /* if not necessary, don't create remote directory. */
1964 tgt = lmv_locate_tgt_rr(lmv);
1968 * only update statfs after QoS mkdir, this means the cached statfs may
1969 * be stale, and current mkdir may not follow QoS accurately, but it's
1970 * not serious, and avoids periodic statfs when client doesn't mkdir by
1974 op_data->op_mds = tgt->ltd_index;
1975 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1981 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1982 const void *data, size_t datalen, umode_t mode, uid_t uid,
1983 gid_t gid, kernel_cap_t cap_effective, __u64 rdev,
1984 struct ptlrpc_request **request)
1986 struct obd_device *obd = exp->exp_obd;
1987 struct lmv_obd *lmv = &obd->u.lmv;
1988 struct lmv_tgt_desc *tgt;
1989 struct mdt_body *repbody;
1994 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
1997 if (lmv_dir_bad_hash(op_data->op_mea1))
2000 if (lmv_dir_layout_changing(op_data->op_mea1)) {
2002 * if parent is migrating, create() needs to lookup existing
2003 * name in both old and new layout, check old layout on client.
2005 rc = lmv_old_layout_lookup(lmv, op_data);
2009 op_data->op_new_layout = true;
2012 tgt = lmv_locate_tgt(lmv, op_data);
2014 RETURN(PTR_ERR(tgt));
2016 /* the order to apply policy in mkdir:
2017 * 1. is "lfs mkdir -i N"? mkdir on MDT N.
2018 * 2. is "lfs mkdir -i -1"? mkdir by space usage.
2019 * 3. is starting MDT specified in default LMV? mkdir on MDT N.
2020 * 4. is default LMV space balanced? mkdir by space usage.
2022 if (lmv_op_user_specific_mkdir(op_data)) {
2023 struct lmv_user_md *lum = op_data->op_data;
2025 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2026 tgt = lmv_tgt(lmv, op_data->op_mds);
2029 } else if (lmv_op_user_qos_mkdir(op_data)) {
2030 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2032 RETURN(PTR_ERR(tgt));
2033 } else if (lmv_op_default_specific_mkdir(op_data)) {
2035 op_data->op_default_mea1->lsm_md_master_mdt_index;
2036 tgt = lmv_tgt(lmv, op_data->op_mds);
2039 } else if (lmv_op_default_qos_mkdir(op_data)) {
2040 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2042 RETURN(PTR_ERR(tgt));
2046 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2050 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
2051 (int)op_data->op_namelen, op_data->op_name,
2052 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
2055 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2056 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
2057 cap_effective, rdev, request);
2059 if (*request == NULL)
2061 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
2064 /* dir restripe needs to send to MDT where dir is located */
2065 if (rc != -EREMOTE ||
2066 !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
2069 repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2070 if (repbody == NULL)
2073 /* Not cross-ref case, just get out of here. */
2074 if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
2077 op_data->op_fid2 = repbody->mbo_fid1;
2078 ptlrpc_req_finished(*request);
2081 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2083 RETURN(PTR_ERR(tgt));
2085 op_data->op_mds = tgt->ltd_index;
2090 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
2091 const union ldlm_policy_data *policy, struct md_op_data *op_data,
2092 struct lustre_handle *lockh, __u64 extra_lock_flags)
2094 struct obd_device *obd = exp->exp_obd;
2095 struct lmv_obd *lmv = &obd->u.lmv;
2096 struct lmv_tgt_desc *tgt;
2101 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
2103 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2105 RETURN(PTR_ERR(tgt));
2107 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
2108 PFID(&op_data->op_fid1), tgt->ltd_index);
2110 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
2117 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
2118 struct ptlrpc_request **preq)
2120 struct obd_device *obd = exp->exp_obd;
2121 struct lmv_obd *lmv = &obd->u.lmv;
2122 struct lmv_tgt_desc *tgt;
2123 struct mdt_body *body;
2129 if (op_data->op_namelen == 2 &&
2130 op_data->op_name[0] == '.' && op_data->op_name[1] == '.')
2131 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2133 tgt = lmv_locate_tgt(lmv, op_data);
2135 RETURN(PTR_ERR(tgt));
2137 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
2138 (int)op_data->op_namelen, op_data->op_name,
2139 PFID(&op_data->op_fid1), tgt->ltd_index);
2141 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2142 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2143 ptlrpc_req_finished(*preq);
2151 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2152 LASSERT(body != NULL);
2154 if (body->mbo_valid & OBD_MD_MDS) {
2155 op_data->op_fid1 = body->mbo_fid1;
2156 op_data->op_valid |= OBD_MD_FLCROSSREF;
2157 op_data->op_namelen = 0;
2158 op_data->op_name = NULL;
2160 ptlrpc_req_finished(*preq);
2169 #define md_op_data_fid(op_data, fl) \
2170 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2171 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2172 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2173 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2176 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2177 struct md_op_data *op_data, __u32 op_tgt,
2178 enum ldlm_mode mode, int bits, int flag)
2180 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2181 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2182 union ldlm_policy_data policy = { { 0 } };
2186 if (!fid_is_sane(fid))
2190 tgt = lmv_fid2tgt(lmv, fid);
2192 RETURN(PTR_ERR(tgt));
2195 if (tgt->ltd_index != op_tgt) {
2196 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2197 policy.l_inodebits.bits = bits;
2198 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2199 mode, LCF_ASYNC, NULL);
2202 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2204 op_data->op_flags |= flag;
2212 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2215 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2216 struct ptlrpc_request **request)
2218 struct obd_device *obd = exp->exp_obd;
2219 struct lmv_obd *lmv = &obd->u.lmv;
2220 struct lmv_tgt_desc *tgt;
2224 LASSERT(op_data->op_namelen != 0);
2226 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2227 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2228 op_data->op_name, PFID(&op_data->op_fid1));
2230 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2231 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2232 op_data->op_cap = current_cap();
2234 tgt = lmv_locate_tgt2(lmv, op_data);
2236 RETURN(PTR_ERR(tgt));
2239 * Cancel UPDATE lock on child (fid1).
2241 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2242 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2243 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2247 rc = md_link(tgt->ltd_exp, op_data, request);
2252 /* migrate the top directory */
2253 static inline bool lmv_op_topdir_migrate(const struct md_op_data *op_data)
2255 if (!S_ISDIR(op_data->op_mode))
2258 if (lmv_dir_layout_changing(op_data->op_mea1))
2264 /* migrate top dir to specific MDTs */
2265 static inline bool lmv_topdir_specific_migrate(const struct md_op_data *op_data)
2267 const struct lmv_user_md *lum = op_data->op_data;
2269 if (!lmv_op_topdir_migrate(op_data))
2272 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2275 /* migrate top dir in QoS mode if user issued "lfs migrate -m -1..." */
2276 static inline bool lmv_topdir_qos_migrate(const struct md_op_data *op_data)
2278 const struct lmv_user_md *lum = op_data->op_data;
2280 if (!lmv_op_topdir_migrate(op_data))
2283 return le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
2286 static inline bool lmv_subdir_specific_migrate(const struct md_op_data *op_data)
2288 const struct lmv_user_md *lum = op_data->op_data;
2290 if (!S_ISDIR(op_data->op_mode))
2293 if (!lmv_dir_layout_changing(op_data->op_mea1))
2296 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2299 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2300 const char *name, size_t namelen,
2301 struct ptlrpc_request **request)
2303 struct obd_device *obd = exp->exp_obd;
2304 struct lmv_obd *lmv = &obd->u.lmv;
2305 struct lmv_stripe_md *lsm = op_data->op_mea1;
2306 struct lmv_tgt_desc *parent_tgt;
2307 struct lmv_tgt_desc *sp_tgt;
2308 struct lmv_tgt_desc *tp_tgt = NULL;
2309 struct lmv_tgt_desc *child_tgt;
2310 struct lmv_tgt_desc *tgt;
2311 struct lu_fid target_fid = { 0 };
2316 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2318 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2319 PFID(&op_data->op_fid1), (int)namelen, name);
2321 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2322 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2323 op_data->op_cap = current_cap();
2325 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2326 if (IS_ERR(parent_tgt))
2327 RETURN(PTR_ERR(parent_tgt));
2329 if (lmv_dir_striped(lsm)) {
2330 const struct lmv_oinfo *oinfo;
2332 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
2334 RETURN(PTR_ERR(oinfo));
2336 /* save source stripe FID in fid4 temporarily for ELC */
2337 op_data->op_fid4 = oinfo->lmo_fid;
2338 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2343 * if parent is being migrated too, fill op_fid2 with target
2344 * stripe fid, otherwise the target stripe is not created yet.
2346 if (lmv_dir_layout_changing(lsm)) {
2347 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
2350 RETURN(PTR_ERR(oinfo));
2352 op_data->op_fid2 = oinfo->lmo_fid;
2353 tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2357 /* parent unchanged and update namespace only */
2358 if (lu_fid_eq(&op_data->op_fid4, &op_data->op_fid2) &&
2359 op_data->op_bias & MDS_MIGRATE_NSONLY)
2363 sp_tgt = parent_tgt;
2366 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2367 if (IS_ERR(child_tgt))
2368 RETURN(PTR_ERR(child_tgt));
2370 if (lmv_topdir_specific_migrate(op_data)) {
2371 struct lmv_user_md *lum = op_data->op_data;
2373 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2374 } else if (lmv_topdir_qos_migrate(op_data)) {
2375 tgt = lmv_locate_tgt_lf(lmv);
2376 if (tgt == ERR_PTR(-EAGAIN))
2377 tgt = lmv_locate_tgt_rr(lmv);
2379 RETURN(PTR_ERR(tgt));
2381 op_data->op_mds = tgt->ltd_index;
2382 } else if (lmv_subdir_specific_migrate(op_data)) {
2383 struct lmv_user_md *lum = op_data->op_data;
2387 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2388 /* adjust MDTs in lum, since subdir is located on where
2389 * its parent stripe is, not the first specified MDT.
2391 for (i = 0; i < le32_to_cpu(lum->lum_stripe_count);
2393 if (le32_to_cpu(lum->lum_objects[i].lum_mds) ==
2398 if (i == le32_to_cpu(lum->lum_stripe_count))
2401 lum->lum_objects[i].lum_mds =
2402 lum->lum_objects[0].lum_mds;
2403 lum->lum_objects[0].lum_mds =
2404 cpu_to_le32(tp_tgt->ltd_index);
2406 /* NB, the above adjusts subdir migration for command like
2407 * "lfs migrate -m 0,1,2 ...", but for migration like
2408 * "lfs migrate -m 0 -c 2 ...", the top dir is migrated to MDT0
2409 * and MDT1, however its subdir may be migrated to MDT1 and MDT2
2412 lum->lum_stripe_offset = cpu_to_le32(tp_tgt->ltd_index);
2413 op_data->op_mds = tp_tgt->ltd_index;
2414 } else if (tp_tgt) {
2415 op_data->op_mds = tp_tgt->ltd_index;
2417 op_data->op_mds = sp_tgt->ltd_index;
2420 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2425 * for directory, send migrate request to the MDT where the object will
2426 * be migrated to, because we can't create a striped directory remotely.
2428 * otherwise, send to the MDT where source is located because regular
2429 * file may open lease.
2431 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2432 * backward compatibility.
2434 if (S_ISDIR(op_data->op_mode) &&
2435 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2436 tgt = lmv_fid2tgt(lmv, &target_fid);
2438 RETURN(PTR_ERR(tgt));
2443 /* cancel UPDATE lock of parent master object */
2444 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2445 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2449 /* cancel UPDATE lock of source parent */
2450 if (sp_tgt != parent_tgt) {
2452 * migrate RPC packs master object FID, because we can only pack
2453 * two FIDs in reint RPC, but MDS needs to know both source
2454 * parent and target parent, and it will obtain them from master
2455 * FID and LMV, the other FID in RPC is kept for target.
2457 * since this FID is not passed to MDC, cancel it anyway.
2459 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2460 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2464 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2466 op_data->op_fid4 = target_fid;
2468 /* cancel UPDATE locks of target parent */
2469 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2470 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2474 /* cancel LOOKUP lock of source if source is remote object */
2475 if (child_tgt != sp_tgt) {
2476 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2477 LCK_EX, MDS_INODELOCK_LOOKUP,
2478 MF_MDC_CANCEL_FID3);
2483 /* cancel ELC locks of source */
2484 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2485 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2489 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2494 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2495 const char *old, size_t oldlen,
2496 const char *new, size_t newlen,
2497 struct ptlrpc_request **request)
2499 struct obd_device *obd = exp->exp_obd;
2500 struct lmv_obd *lmv = &obd->u.lmv;
2501 struct lmv_tgt_desc *sp_tgt;
2502 struct lmv_tgt_desc *tp_tgt = NULL;
2503 struct lmv_tgt_desc *src_tgt = NULL;
2504 struct lmv_tgt_desc *tgt;
2505 struct mdt_body *body;
2510 LASSERT(oldlen != 0);
2512 if (op_data->op_cli_flags & CLI_MIGRATE) {
2513 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2517 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2518 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2519 op_data->op_cap = current_cap();
2521 op_data->op_name = new;
2522 op_data->op_namelen = newlen;
2524 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2526 RETURN(PTR_ERR(tp_tgt));
2528 /* Since the target child might be destroyed, and it might become
2529 * orphan, and we can only check orphan on the local MDT right now, so
2530 * we send rename request to the MDT where target child is located. If
2531 * target child does not exist, then it will send the request to the
2533 if (fid_is_sane(&op_data->op_fid4)) {
2534 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2536 RETURN(PTR_ERR(tgt));
2541 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2543 /* cancel UPDATE locks of target parent */
2544 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2545 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2549 if (fid_is_sane(&op_data->op_fid4)) {
2550 /* cancel LOOKUP lock of target on target parent */
2551 if (tgt != tp_tgt) {
2552 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2553 tgt->ltd_index, LCK_EX,
2554 MDS_INODELOCK_LOOKUP,
2555 MF_MDC_CANCEL_FID4);
2561 if (fid_is_sane(&op_data->op_fid3)) {
2562 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2563 if (IS_ERR(src_tgt))
2564 RETURN(PTR_ERR(src_tgt));
2566 /* cancel ELC locks of source */
2567 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2568 LCK_EX, MDS_INODELOCK_ELC,
2569 MF_MDC_CANCEL_FID3);
2574 op_data->op_name = old;
2575 op_data->op_namelen = oldlen;
2577 sp_tgt = lmv_locate_tgt(lmv, op_data);
2579 RETURN(PTR_ERR(sp_tgt));
2581 /* cancel UPDATE locks of source parent */
2582 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2583 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2587 if (fid_is_sane(&op_data->op_fid3)) {
2588 /* cancel LOOKUP lock of source on source parent */
2589 if (src_tgt != sp_tgt) {
2590 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2591 tgt->ltd_index, LCK_EX,
2592 MDS_INODELOCK_LOOKUP,
2593 MF_MDC_CANCEL_FID3);
2600 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2601 PFID(&op_data->op_fid1), (int)oldlen, old,
2602 PFID(&op_data->op_fid2), (int)newlen, new);
2604 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2606 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2607 ptlrpc_req_finished(*request);
2612 if (rc && rc != -EXDEV)
2615 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2619 /* Not cross-ref case, just get out of here. */
2620 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2623 op_data->op_fid4 = body->mbo_fid1;
2625 ptlrpc_req_finished(*request);
2628 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2630 RETURN(PTR_ERR(tgt));
2632 if (fid_is_sane(&op_data->op_fid4)) {
2633 /* cancel LOOKUP lock of target on target parent */
2634 if (tgt != tp_tgt) {
2635 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2636 tgt->ltd_index, LCK_EX,
2637 MDS_INODELOCK_LOOKUP,
2638 MF_MDC_CANCEL_FID4);
2647 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2648 void *ea, size_t ealen, struct ptlrpc_request **request)
2650 struct obd_device *obd = exp->exp_obd;
2651 struct lmv_obd *lmv = &obd->u.lmv;
2652 struct lmv_tgt_desc *tgt;
2657 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2658 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2659 op_data->op_xvalid);
2661 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2662 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2664 RETURN(PTR_ERR(tgt));
2666 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2671 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2672 struct ptlrpc_request **request)
2674 struct obd_device *obd = exp->exp_obd;
2675 struct lmv_obd *lmv = &obd->u.lmv;
2676 struct lmv_tgt_desc *tgt;
2681 tgt = lmv_fid2tgt(lmv, fid);
2683 RETURN(PTR_ERR(tgt));
2685 rc = md_fsync(tgt->ltd_exp, fid, request);
2689 struct stripe_dirent {
2690 struct page *sd_page;
2691 struct lu_dirpage *sd_dp;
2692 struct lu_dirent *sd_ent;
2696 struct lmv_dir_ctxt {
2697 struct lmv_obd *ldc_lmv;
2698 struct md_op_data *ldc_op_data;
2699 struct md_readdir_info *ldc_mrinfo;
2702 struct stripe_dirent ldc_stripes[0];
2705 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2707 if (stripe->sd_page) {
2708 kunmap(stripe->sd_page);
2709 put_page(stripe->sd_page);
2710 stripe->sd_page = NULL;
2711 stripe->sd_ent = NULL;
2715 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2719 for (i = 0; i < ctxt->ldc_count; i++)
2720 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2723 /* if @ent is dummy, or . .., get next */
2724 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2725 struct lu_dirent *ent,
2728 for (; ent; ent = lu_dirent_next(ent)) {
2729 /* Skip dummy entry */
2730 if (le16_to_cpu(ent->lde_namelen) == 0)
2733 /* skip . and .. for other stripes */
2735 (strncmp(ent->lde_name, ".",
2736 le16_to_cpu(ent->lde_namelen)) == 0 ||
2737 strncmp(ent->lde_name, "..",
2738 le16_to_cpu(ent->lde_namelen)) == 0))
2741 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2748 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2749 struct stripe_dirent *stripe,
2752 struct md_op_data *op_data = ctxt->ldc_op_data;
2753 struct lmv_oinfo *oinfo;
2754 struct lu_fid fid = op_data->op_fid1;
2755 struct inode *inode = op_data->op_data;
2756 struct lmv_tgt_desc *tgt;
2757 struct lu_dirent *ent = stripe->sd_ent;
2758 __u64 hash = ctxt->ldc_hash;
2763 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2767 if (stripe->sd_page) {
2768 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2770 /* @hash should be the last dirent hash */
2771 LASSERTF(hash <= end,
2772 "ctxt@%p stripe@%p hash %llx end %llx\n",
2773 ctxt, stripe, hash, end);
2774 /* unload last page */
2775 stripe_dirent_unload(stripe);
2777 if (end == MDS_DIR_END_OFF) {
2778 stripe->sd_eof = true;
2784 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2785 if (!oinfo->lmo_root) {
2790 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2796 /* op_data is shared by stripes, reset after use */
2797 op_data->op_fid1 = oinfo->lmo_fid;
2798 op_data->op_fid2 = oinfo->lmo_fid;
2799 op_data->op_data = oinfo->lmo_root;
2801 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_mrinfo, hash,
2804 op_data->op_fid1 = fid;
2805 op_data->op_fid2 = fid;
2806 op_data->op_data = inode;
2811 stripe->sd_dp = page_address(stripe->sd_page);
2812 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2814 /* in case a page filled with ., .. and dummy, read next */
2817 stripe->sd_ent = ent;
2820 /* treat error as eof, so dir can be partially accessed */
2821 stripe->sd_eof = true;
2822 ctxt->ldc_mrinfo->mr_partial_readdir_rc = rc;
2823 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2824 "directory is partially accessed!\n",
2825 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2832 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2834 struct obd_device *obd = exp->exp_obd;
2835 struct lmv_obd *lmv = &obd->u.lmv;
2836 struct lmv_tgt_desc *tgt;
2841 rc = lmv_check_connect(obd);
2845 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2847 RETURN(PTR_ERR(tgt));
2849 data->op_flags |= MF_MDC_CANCEL_FID1;
2850 rc = md_file_resync(tgt->ltd_exp, data);
2855 * Get dirent with the closest hash for striped directory
2857 * This function will search the dir entry, whose hash value is the
2858 * closest(>=) to hash from all of sub-stripes, and it is only being called
2859 * for striped directory.
2861 * \param[in] ctxt dir read context
2863 * \retval dirent get the entry successfully
2864 * NULL does not get the entry, normally it means
2865 * it reaches the end of the directory, while read
2866 * stripe dirent error is ignored to allow partial
2869 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2871 struct stripe_dirent *stripe;
2872 struct lu_dirent *ent = NULL;
2876 /* TODO: optimize with k-way merge sort */
2877 for (i = 0; i < ctxt->ldc_count; i++) {
2878 stripe = &ctxt->ldc_stripes[i];
2882 if (!stripe->sd_ent) {
2883 stripe_dirent_load(ctxt, stripe, i);
2884 if (!stripe->sd_ent) {
2885 LASSERT(stripe->sd_eof);
2891 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2892 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2894 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2901 stripe = &ctxt->ldc_stripes[min];
2902 ent = stripe->sd_ent;
2903 /* pop found dirent */
2904 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2912 * Build dir entry page for striped directory
2914 * This function gets one entry by @offset from a striped directory. It will
2915 * read entries from all of stripes, and choose one closest to the required
2916 * offset(&offset). A few notes
2917 * 1. skip . and .. for non-zero stripes, because there can only have one .
2918 * and .. in a directory.
2919 * 2. op_data will be shared by all of stripes, instead of allocating new
2920 * one, so need to restore before reusing.
2922 * \param[in] exp obd export refer to LMV
2923 * \param[in] op_data hold those MD parameters of read_entry
2924 * \param[in] mrinfo ldlm callback being used in enqueue in mdc_read_entry,
2925 * and partial readdir result will be stored in it.
2926 * \param[in] offset starting hash offset
2927 * \param[out] ppage the page holding the entry. Note: because the entry
2928 * will be accessed in upper layer, so we need hold the
2929 * page until the usages of entry is finished, see
2930 * ll_dir_entry_next.
2932 * retval =0 if get entry successfully
2933 * <0 cannot get entry
2935 static int lmv_striped_read_page(struct obd_export *exp,
2936 struct md_op_data *op_data,
2937 struct md_readdir_info *mrinfo, __u64 offset,
2938 struct page **ppage)
2940 struct page *page = NULL;
2941 struct lu_dirpage *dp;
2943 struct lu_dirent *ent;
2944 struct lu_dirent *last_ent;
2946 struct lmv_dir_ctxt *ctxt;
2947 struct lu_dirent *next = NULL;
2953 /* Allocate a page and read entries from all of stripes and fill
2954 * the page by hash order */
2955 page = alloc_page(GFP_KERNEL);
2959 /* Initialize the entry page */
2961 memset(dp, 0, sizeof(*dp));
2962 dp->ldp_hash_start = cpu_to_le64(offset);
2965 left_bytes = PAGE_SIZE - sizeof(*dp);
2969 /* initalize dir read context */
2970 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2971 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2973 GOTO(free_page, rc = -ENOMEM);
2974 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2975 ctxt->ldc_op_data = op_data;
2976 ctxt->ldc_mrinfo = mrinfo;
2977 ctxt->ldc_hash = offset;
2978 ctxt->ldc_count = stripe_count;
2981 next = lmv_dirent_next(ctxt);
2983 /* end of directory */
2985 ctxt->ldc_hash = MDS_DIR_END_OFF;
2988 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2990 ent_size = le16_to_cpu(next->lde_reclen);
2992 /* the last entry lde_reclen is 0, but it might not be the last
2993 * one of this temporay dir page */
2995 ent_size = lu_dirent_calc_size(
2996 le16_to_cpu(next->lde_namelen),
2997 le32_to_cpu(next->lde_attrs));
2999 if (ent_size > left_bytes)
3002 memcpy(ent, next, ent_size);
3004 /* Replace . with master FID and Replace .. with the parent FID
3005 * of master object */
3006 if (strncmp(ent->lde_name, ".",
3007 le16_to_cpu(ent->lde_namelen)) == 0 &&
3008 le16_to_cpu(ent->lde_namelen) == 1)
3009 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
3010 else if (strncmp(ent->lde_name, "..",
3011 le16_to_cpu(ent->lde_namelen)) == 0 &&
3012 le16_to_cpu(ent->lde_namelen) == 2)
3013 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
3015 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
3016 le16_to_cpu(ent->lde_namelen), ent->lde_name,
3017 le64_to_cpu(ent->lde_hash));
3019 left_bytes -= ent_size;
3020 ent->lde_reclen = cpu_to_le16(ent_size);
3022 ent = (void *)ent + ent_size;
3025 last_ent->lde_reclen = 0;
3028 dp->ldp_flags |= LDF_EMPTY;
3029 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
3030 dp->ldp_flags |= LDF_COLLIDE;
3031 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
3032 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
3034 put_lmv_dir_ctxt(ctxt);
3035 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
3048 static int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
3049 struct md_readdir_info *mrinfo, __u64 offset,
3050 struct page **ppage)
3052 struct obd_device *obd = exp->exp_obd;
3053 struct lmv_obd *lmv = &obd->u.lmv;
3054 struct lmv_tgt_desc *tgt;
3059 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
3062 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
3063 rc = lmv_striped_read_page(exp, op_data, mrinfo, offset, ppage);
3067 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
3069 RETURN(PTR_ERR(tgt));
3071 rc = md_read_page(tgt->ltd_exp, op_data, mrinfo, offset, ppage);
3077 * Unlink a file/directory
3079 * Unlink a file or directory under the parent dir. The unlink request
3080 * usually will be sent to the MDT where the child is located, but if
3081 * the client does not have the child FID then request will be sent to the
3082 * MDT where the parent is located.
3084 * If the parent is a striped directory then it also needs to locate which
3085 * stripe the name of the child is located, and replace the parent FID
3086 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
3087 * it will walk through all of sub-stripes until the child is being
3090 * \param[in] exp export refer to LMV
3091 * \param[in] op_data different parameters transferred beween client
3092 * MD stacks, name, namelen, FIDs etc.
3093 * op_fid1 is the parent FID, op_fid2 is the child
3095 * \param[out] request point to the request of unlink.
3097 * retval 0 if succeed
3098 * negative errno if failed.
3100 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
3101 struct ptlrpc_request **request)
3103 struct obd_device *obd = exp->exp_obd;
3104 struct lmv_obd *lmv = &obd->u.lmv;
3105 struct lmv_tgt_desc *tgt;
3106 struct lmv_tgt_desc *parent_tgt;
3107 struct mdt_body *body;
3112 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3113 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3114 op_data->op_cap = current_cap();
3117 parent_tgt = lmv_locate_tgt(lmv, op_data);
3118 if (IS_ERR(parent_tgt))
3119 RETURN(PTR_ERR(parent_tgt));
3121 if (likely(!fid_is_zero(&op_data->op_fid2))) {
3122 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3124 RETURN(PTR_ERR(tgt));
3130 * If child's fid is given, cancel unused locks for it if it is from
3131 * another export than parent.
3133 * LOOKUP lock for child (fid3) should also be cancelled on parent
3134 * tgt_tgt in mdc_unlink().
3136 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
3138 if (parent_tgt != tgt)
3139 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
3140 LCK_EX, MDS_INODELOCK_LOOKUP,
3141 MF_MDC_CANCEL_FID3);
3143 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
3144 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
3148 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
3149 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
3152 rc = md_unlink(tgt->ltd_exp, op_data, request);
3153 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
3154 ptlrpc_req_finished(*request);
3162 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
3166 /* Not cross-ref case, just get out of here. */
3167 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
3170 /* This is a remote object, try remote MDT. */
3171 op_data->op_fid2 = body->mbo_fid1;
3172 ptlrpc_req_finished(*request);
3175 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3177 RETURN(PTR_ERR(tgt));
3182 static int lmv_precleanup(struct obd_device *obd)
3185 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
3190 * Get by key a value associated with a LMV device.
3192 * Dispatch request to lower-layer devices as needed.
3194 * \param[in] env execution environment for this thread
3195 * \param[in] exp export for the LMV device
3196 * \param[in] keylen length of key identifier
3197 * \param[in] key identifier of key to get value for
3198 * \param[in] vallen size of \a val
3199 * \param[out] val pointer to storage location for value
3200 * \param[in] lsm optional striping metadata of object
3202 * \retval 0 on success
3203 * \retval negative negated errno on failure
3205 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
3206 __u32 keylen, void *key, __u32 *vallen, void *val)
3208 struct obd_device *obd;
3209 struct lmv_obd *lmv;
3210 struct lu_tgt_desc *tgt;
3215 obd = class_exp2obd(exp);
3217 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3218 exp->exp_handle.h_cookie);
3223 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
3224 LASSERT(*vallen == sizeof(__u32));
3225 lmv_foreach_connected_tgt(lmv, tgt) {
3226 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
3231 } else if (KEY_IS(KEY_MAX_EASIZE) ||
3232 KEY_IS(KEY_DEFAULT_EASIZE) ||
3233 KEY_IS(KEY_CONN_DATA)) {
3235 * Forwarding this request to first MDS, it should know LOV
3238 tgt = lmv_tgt(lmv, 0);
3242 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
3243 if (!rc && KEY_IS(KEY_CONN_DATA))
3244 exp->exp_connect_data = *(struct obd_connect_data *)val;
3246 } else if (KEY_IS(KEY_TGT_COUNT)) {
3247 *((int *)val) = lmv->lmv_mdt_descs.ltd_tgts_size;
3251 CDEBUG(D_IOCTL, "Invalid key\n");
3255 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
3256 int *__rcs, struct ptlrpc_request_set *_set)
3258 struct obd_device *obd = class_exp2obd(exp);
3259 struct ptlrpc_request_set *set = _set;
3260 struct lmv_obd *lmv = &obd->u.lmv;
3261 int tgt_count = lmv->lmv_mdt_count;
3262 struct lu_tgt_desc *tgt;
3263 struct fid_array *fat, **fas = NULL;
3264 int i, rc, **rcs = NULL;
3267 set = ptlrpc_prep_set();
3272 /* split FIDs by targets */
3273 OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
3275 GOTO(out, rc = -ENOMEM);
3276 OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
3278 GOTO(out_fas, rc = -ENOMEM);
3280 for (i = 0; i < fa->fa_nr; i++) {
3283 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
3285 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
3286 PFID(&fa->fa_fids[i]), rc);
3289 LASSERT(idx < tgt_count);
3291 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
3292 fa_fids[fa->fa_nr]));
3294 GOTO(out, rc = -ENOMEM);
3296 OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
3298 GOTO(out, rc = -ENOMEM);
3301 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3304 lmv_foreach_connected_tgt(lmv, tgt) {
3305 fat = fas[tgt->ltd_index];
3306 if (!fat || fat->fa_nr == 0)
3308 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3311 rc = ptlrpc_set_wait(NULL, set);
3314 for (i = 0; i < tgt_count; i++) {
3316 if (!fat || fat->fa_nr == 0)
3318 /* copy FIDs back */
3319 memcpy(fa->fa_fids + j, fat->fa_fids,
3320 fat->fa_nr * sizeof(struct lu_fid));
3322 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3327 ptlrpc_set_destroy(set);
3330 for (i = 0; i < tgt_count; i++) {
3332 OBD_FREE(fas[i], offsetof(struct fid_array,
3333 fa_fids[fa->fa_nr]));
3335 OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3338 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3341 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3347 * Asynchronously set by key a value associated with a LMV device.
3349 * Dispatch request to lower-layer devices as needed.
3351 * \param[in] env execution environment for this thread
3352 * \param[in] exp export for the LMV device
3353 * \param[in] keylen length of key identifier
3354 * \param[in] key identifier of key to store value for
3355 * \param[in] vallen size of value to store
3356 * \param[in] val pointer to data to be stored
3357 * \param[in] set optional list of related ptlrpc requests
3359 * \retval 0 on success
3360 * \retval negative negated errno on failure
3362 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3363 __u32 keylen, void *key, __u32 vallen, void *val,
3364 struct ptlrpc_request_set *set)
3366 struct lmv_tgt_desc *tgt;
3367 struct obd_device *obd;
3368 struct lmv_obd *lmv;
3372 obd = class_exp2obd(exp);
3374 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3375 exp->exp_handle.h_cookie);
3380 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3381 KEY_IS(KEY_DEFAULT_EASIZE)) {
3384 lmv_foreach_connected_tgt(lmv, tgt) {
3385 err = obd_set_info_async(env, tgt->ltd_exp,
3386 keylen, key, vallen, val, set);
3397 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3398 const struct lmv_mds_md_v1 *lmm1)
3400 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3407 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3408 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3409 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3410 if (CFS_FAIL_CHECK(OBD_FAIL_LMV_UNKNOWN_STRIPE))
3411 lsm->lsm_md_hash_type = cfs_fail_val ?: LMV_HASH_TYPE_UNKNOWN;
3413 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3414 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3415 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3416 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3417 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3418 sizeof(lsm->lsm_md_pool_name));
3420 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3423 CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3424 "layout_version %d\n", lsm->lsm_md_stripe_count,
3425 lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3426 lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3427 lsm->lsm_md_layout_version);
3429 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3430 for (i = 0; i < stripe_count; i++) {
3431 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3432 &lmm1->lmv_stripe_fids[i]);
3434 * set default value -1, so lmv_locate_tgt() knows this stripe
3435 * target is not initialized.
3437 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3438 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3441 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3442 &lsm->lsm_md_oinfo[i].lmo_mds);
3449 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3450 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3456 static inline int lmv_unpack_user_md(struct obd_export *exp,
3457 struct lmv_stripe_md *lsm,
3458 const struct lmv_user_md *lmu)
3460 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3461 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3462 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3463 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3464 lsm->lsm_md_max_inherit = lmu->lum_max_inherit;
3465 lsm->lsm_md_max_inherit_rr = lmu->lum_max_inherit_rr;
3466 lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3471 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3472 const union lmv_mds_md *lmm, size_t lmm_size)
3474 struct lmv_stripe_md *lsm;
3477 bool allocated = false;
3480 LASSERT(lsmp != NULL);
3484 if (lsm != NULL && lmm == NULL) {
3486 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3488 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3491 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3493 OBD_FREE_LARGE(lfm, lfm_size);
3497 if (lmv_dir_striped(lsm)) {
3498 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3499 iput(lsm->lsm_md_oinfo[i].lmo_root);
3500 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3502 lsm_size = lmv_stripe_md_size(0);
3504 OBD_FREE(lsm, lsm_size);
3509 /* foreign lmv case */
3510 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3511 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3514 OBD_ALLOC_LARGE(lfm, lmm_size);
3517 *lsmp = (struct lmv_stripe_md *)lfm;
3519 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3520 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3521 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3522 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3523 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3528 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3532 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3533 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3534 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3535 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3540 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3541 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3544 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3545 * stripecount should be 0 then.
3547 lsm_size = lmv_stripe_md_size(0);
3550 OBD_ALLOC(lsm, lsm_size);
3557 switch (le32_to_cpu(lmm->lmv_magic)) {
3559 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3561 case LMV_USER_MAGIC:
3562 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3565 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3566 le32_to_cpu(lmm->lmv_magic));
3571 if (rc != 0 && allocated) {
3572 OBD_FREE(lsm, lsm_size);
3579 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3581 lmv_unpackmd(NULL, &lsm, NULL, 0);
3583 EXPORT_SYMBOL(lmv_free_memmd);
3585 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3586 union ldlm_policy_data *policy,
3587 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3590 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3591 struct lu_tgt_desc *tgt;
3597 LASSERT(fid != NULL);
3599 lmv_foreach_connected_tgt(lmv, tgt) {
3600 if (!tgt->ltd_active)
3603 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3611 static int lmv_set_lock_data(struct obd_export *exp,
3612 const struct lustre_handle *lockh,
3613 void *data, __u64 *bits)
3615 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3616 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3621 if (tgt == NULL || tgt->ltd_exp == NULL)
3623 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3627 static enum ldlm_mode
3628 lmv_lock_match(struct obd_export *exp, __u64 flags,
3629 const struct lu_fid *fid, enum ldlm_type type,
3630 union ldlm_policy_data *policy,
3631 enum ldlm_mode mode, struct lustre_handle *lockh)
3633 struct obd_device *obd = exp->exp_obd;
3634 struct lmv_obd *lmv = &obd->u.lmv;
3635 struct lu_tgt_desc *tgt;
3636 __u64 bits = policy->l_inodebits.bits;
3637 enum ldlm_mode rc = LCK_MINMODE;
3641 /* only one bit is set */
3642 LASSERT(bits && !(bits & (bits - 1)));
3643 /* With DNE every object can have two locks in different namespaces:
3644 * lookup lock in space of MDT storing direntry and update/open lock in
3645 * space of MDT storing inode. Try the MDT that the FID maps to first,
3646 * since this can be easily found, and only try others if that fails.
3648 if (bits == MDS_INODELOCK_LOOKUP) {
3649 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3650 i < lmv->lmv_mdt_descs.ltd_tgts_size; i++,
3651 index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3654 "%s: "DFID" is inaccessible: rc = %d\n",
3655 obd->obd_name, PFID(fid), index);
3658 tgt = lmv_tgt(lmv, index);
3659 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3661 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3662 policy, mode, lockh);
3667 tgt = lmv_fid2tgt(lmv, fid);
3668 if (!IS_ERR(tgt) && tgt->ltd_exp && tgt->ltd_active)
3669 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3670 policy, mode, lockh);
3673 CDEBUG(D_INODE, "Lock match for "DFID": %d\n", PFID(fid), rc);
3679 lmv_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
3680 struct obd_export *dt_exp, struct obd_export *md_exp,
3681 struct lustre_md *md)
3683 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3684 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3686 if (!tgt || !tgt->ltd_exp)
3689 return md_get_lustre_md(tgt->ltd_exp, pill, dt_exp, md_exp, md);
3692 static int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3694 struct obd_device *obd = exp->exp_obd;
3695 struct lmv_obd *lmv = &obd->u.lmv;
3696 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3700 if (md->default_lmv) {
3701 lmv_free_memmd(md->default_lmv);
3702 md->default_lmv = NULL;
3704 if (md->lmv != NULL) {
3705 lmv_free_memmd(md->lmv);
3708 if (!tgt || !tgt->ltd_exp)
3710 RETURN(md_free_lustre_md(tgt->ltd_exp, md));
3713 static int lmv_set_open_replay_data(struct obd_export *exp,
3714 struct obd_client_handle *och,
3715 struct lookup_intent *it)
3717 struct obd_device *obd = exp->exp_obd;
3718 struct lmv_obd *lmv = &obd->u.lmv;
3719 struct lmv_tgt_desc *tgt;
3723 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3725 RETURN(PTR_ERR(tgt));
3727 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3730 static int lmv_clear_open_replay_data(struct obd_export *exp,
3731 struct obd_client_handle *och)
3733 struct obd_device *obd = exp->exp_obd;
3734 struct lmv_obd *lmv = &obd->u.lmv;
3735 struct lmv_tgt_desc *tgt;
3739 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3741 RETURN(PTR_ERR(tgt));
3743 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3746 static int lmv_intent_getattr_async(struct obd_export *exp,
3747 struct md_op_item *item)
3749 struct md_op_data *op_data = &item->mop_data;
3750 struct obd_device *obd = exp->exp_obd;
3751 struct lmv_obd *lmv = &obd->u.lmv;
3752 struct lmv_tgt_desc *ptgt;
3753 struct lmv_tgt_desc *ctgt;
3758 if (!fid_is_sane(&op_data->op_fid2))
3761 ptgt = lmv_locate_tgt(lmv, op_data);
3763 RETURN(PTR_ERR(ptgt));
3765 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3767 RETURN(PTR_ERR(ctgt));
3770 * remote object needs two RPCs to lookup and getattr, considering the
3771 * complexity don't support statahead for now.
3776 rc = md_intent_getattr_async(ptgt->ltd_exp, item);
3781 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3782 struct lu_fid *fid, __u64 *bits)
3784 struct obd_device *obd = exp->exp_obd;
3785 struct lmv_obd *lmv = &obd->u.lmv;
3786 struct lmv_tgt_desc *tgt;
3791 tgt = lmv_fid2tgt(lmv, fid);
3793 RETURN(PTR_ERR(tgt));
3795 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3799 static int lmv_get_fid_from_lsm(struct obd_export *exp,
3800 const struct lmv_stripe_md *lsm,
3801 const char *name, int namelen,
3804 const struct lmv_oinfo *oinfo;
3806 if (!lmv_dir_striped(lsm))
3809 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3811 return PTR_ERR(oinfo);
3813 *fid = oinfo->lmo_fid;
3819 * For lmv, only need to send request to master MDT, and the master MDT will
3820 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3821 * we directly fetch data from the slave MDTs.
3823 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3824 struct obd_quotactl *oqctl)
3826 struct obd_device *obd = class_exp2obd(exp);
3827 struct lmv_obd *lmv = &obd->u.lmv;
3828 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3829 __u64 curspace, curinodes;
3834 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3835 CERROR("master lmv inactive\n");
3839 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3840 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3844 curspace = curinodes = 0;
3845 lmv_foreach_connected_tgt(lmv, tgt) {
3848 if (!tgt->ltd_active)
3851 err = obd_quotactl(tgt->ltd_exp, oqctl);
3853 CERROR("getquota on mdt %d failed. %d\n",
3854 tgt->ltd_index, err);
3858 curspace += oqctl->qc_dqblk.dqb_curspace;
3859 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3862 oqctl->qc_dqblk.dqb_curspace = curspace;
3863 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3868 static int lmv_merge_attr(struct obd_export *exp,
3869 const struct lmv_stripe_md *lsm,
3870 struct cl_attr *attr,
3871 ldlm_blocking_callback cb_blocking)
3876 if (!lmv_dir_striped(lsm))
3879 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3883 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3884 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3890 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3891 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3892 i_size_read(inode), (unsigned long long)inode->i_blocks,
3893 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3894 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3896 /* for slave stripe, it needs to subtract nlink for . and .. */
3898 attr->cat_nlink += inode->i_nlink - 2;
3900 attr->cat_nlink = inode->i_nlink;
3902 attr->cat_size += i_size_read(inode);
3903 attr->cat_blocks += inode->i_blocks;
3905 if (attr->cat_atime < inode->i_atime.tv_sec)
3906 attr->cat_atime = inode->i_atime.tv_sec;
3908 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3909 attr->cat_ctime = inode->i_ctime.tv_sec;
3911 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3912 attr->cat_mtime = inode->i_mtime.tv_sec;
3917 static struct lu_batch *lmv_batch_create(struct obd_export *exp,
3918 enum lu_batch_flags flags,
3921 struct lu_batch *bh;
3922 struct lmv_batch *lbh;
3927 RETURN(ERR_PTR(-ENOMEM));
3929 bh = &lbh->lbh_super;
3930 bh->lbt_flags = flags;
3931 bh->lbt_max_count = max_count;
3933 if (flags & BATCH_FL_RQSET) {
3934 bh->lbt_rqset = ptlrpc_prep_set();
3935 if (bh->lbt_rqset == NULL) {
3937 RETURN(ERR_PTR(-ENOMEM));
3941 INIT_LIST_HEAD(&lbh->lbh_sub_batch_list);
3945 static int lmv_batch_stop(struct obd_export *exp, struct lu_batch *bh)
3947 struct lmv_batch *lbh;
3948 struct lmvsub_batch *sub;
3949 struct lmvsub_batch *tmp;
3954 lbh = container_of(bh, struct lmv_batch, lbh_super);
3955 list_for_each_entry_safe(sub, tmp, &lbh->lbh_sub_batch_list,
3957 list_del(&sub->sbh_sub_item);
3958 rc = md_batch_stop(sub->sbh_tgt->ltd_exp, sub->sbh_sub);
3960 CERROR("%s: stop batch processing failed: rc = %d\n",
3961 exp->exp_obd->obd_name, rc);
3962 if (bh->lbt_result == 0)
3963 bh->lbt_result = rc;
3968 if (bh->lbt_flags & BATCH_FL_RQSET) {
3969 rc = ptlrpc_set_wait(NULL, bh->lbt_rqset);
3970 ptlrpc_set_destroy(bh->lbt_rqset);
3977 static int lmv_batch_flush(struct obd_export *exp, struct lu_batch *bh,
3980 struct lmv_batch *lbh;
3981 struct lmvsub_batch *sub;
3987 lbh = container_of(bh, struct lmv_batch, lbh_super);
3988 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
3989 rc1 = md_batch_flush(sub->sbh_tgt->ltd_exp, sub->sbh_sub, wait);
3991 CERROR("%s: stop batch processing failed: rc = %d\n",
3992 exp->exp_obd->obd_name, rc);
3993 if (bh->lbt_result == 0)
3994 bh->lbt_result = rc;
4001 if (wait && bh->lbt_flags & BATCH_FL_RQSET) {
4002 rc1 = ptlrpc_set_wait(NULL, bh->lbt_rqset);
4010 static inline struct lmv_tgt_desc *
4011 lmv_batch_locate_tgt(struct lmv_obd *lmv, struct md_op_item *item)
4013 struct md_op_data *op_data = &item->mop_data;
4014 struct lmv_tgt_desc *tgt;
4016 switch (item->mop_opc) {
4017 case MD_OP_GETATTR: {
4018 struct lmv_tgt_desc *ptgt;
4020 if (!fid_is_sane(&op_data->op_fid2))
4021 RETURN(ERR_PTR(-EINVAL));
4023 ptgt = lmv_locate_tgt(lmv, op_data);
4027 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
4032 * Remote object needs two RPCs to lookup and getattr,
4033 * considering the complexity don't support statahead for now.
4036 RETURN(ERR_PTR(-EREMOTE));
4041 tgt = ERR_PTR(-ENOTSUPP);
4047 struct lu_batch *lmv_batch_lookup_sub(struct lmv_batch *lbh,
4048 struct lmv_tgt_desc *tgt)
4050 struct lmvsub_batch *sub;
4052 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
4053 if (sub->sbh_tgt == tgt)
4054 return sub->sbh_sub;
4060 struct lu_batch *lmv_batch_get_sub(struct lmv_batch *lbh,
4061 struct lmv_tgt_desc *tgt)
4063 struct lmvsub_batch *sbh;
4064 struct lu_batch *child_bh;
4065 struct lu_batch *bh;
4069 child_bh = lmv_batch_lookup_sub(lbh, tgt);
4070 if (child_bh != NULL)
4075 RETURN(ERR_PTR(-ENOMEM));
4077 INIT_LIST_HEAD(&sbh->sbh_sub_item);
4080 bh = &lbh->lbh_super;
4081 child_bh = md_batch_create(tgt->ltd_exp, bh->lbt_flags,
4083 if (IS_ERR(child_bh)) {
4088 child_bh->lbt_rqset = bh->lbt_rqset;
4089 sbh->sbh_sub = child_bh;
4090 list_add(&sbh->sbh_sub_item, &lbh->lbh_sub_batch_list);
4094 static int lmv_batch_add(struct obd_export *exp, struct lu_batch *bh,
4095 struct md_op_item *item)
4097 struct obd_device *obd = exp->exp_obd;
4098 struct lmv_obd *lmv = &obd->u.lmv;
4099 struct lmv_tgt_desc *tgt;
4100 struct lmv_batch *lbh;
4101 struct lu_batch *child_bh;
4106 tgt = lmv_batch_locate_tgt(lmv, item);
4108 RETURN(PTR_ERR(tgt));
4110 lbh = container_of(bh, struct lmv_batch, lbh_super);
4111 child_bh = lmv_batch_get_sub(lbh, tgt);
4112 if (IS_ERR(child_bh))
4113 RETURN(PTR_ERR(child_bh));
4115 rc = md_batch_add(tgt->ltd_exp, child_bh, item);
4119 static const struct obd_ops lmv_obd_ops = {
4120 .o_owner = THIS_MODULE,
4121 .o_setup = lmv_setup,
4122 .o_cleanup = lmv_cleanup,
4123 .o_precleanup = lmv_precleanup,
4124 .o_process_config = lmv_process_config,
4125 .o_connect = lmv_connect,
4126 .o_disconnect = lmv_disconnect,
4127 .o_statfs = lmv_statfs,
4128 .o_get_info = lmv_get_info,
4129 .o_set_info_async = lmv_set_info_async,
4130 .o_notify = lmv_notify,
4131 .o_get_uuid = lmv_get_uuid,
4132 .o_fid_alloc = lmv_fid_alloc,
4133 .o_iocontrol = lmv_iocontrol,
4134 .o_quotactl = lmv_quotactl
4137 static const struct md_ops lmv_md_ops = {
4138 .m_get_root = lmv_get_root,
4139 .m_null_inode = lmv_null_inode,
4140 .m_close = lmv_close,
4141 .m_create = lmv_create,
4142 .m_enqueue = lmv_enqueue,
4143 .m_getattr = lmv_getattr,
4144 .m_getxattr = lmv_getxattr,
4145 .m_getattr_name = lmv_getattr_name,
4146 .m_intent_lock = lmv_intent_lock,
4148 .m_rename = lmv_rename,
4149 .m_setattr = lmv_setattr,
4150 .m_setxattr = lmv_setxattr,
4151 .m_fsync = lmv_fsync,
4152 .m_file_resync = lmv_file_resync,
4153 .m_read_page = lmv_read_page,
4154 .m_unlink = lmv_unlink,
4155 .m_init_ea_size = lmv_init_ea_size,
4156 .m_cancel_unused = lmv_cancel_unused,
4157 .m_set_lock_data = lmv_set_lock_data,
4158 .m_lock_match = lmv_lock_match,
4159 .m_get_lustre_md = lmv_get_lustre_md,
4160 .m_free_lustre_md = lmv_free_lustre_md,
4161 .m_merge_attr = lmv_merge_attr,
4162 .m_set_open_replay_data = lmv_set_open_replay_data,
4163 .m_clear_open_replay_data = lmv_clear_open_replay_data,
4164 .m_intent_getattr_async = lmv_intent_getattr_async,
4165 .m_revalidate_lock = lmv_revalidate_lock,
4166 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
4167 .m_unpackmd = lmv_unpackmd,
4168 .m_rmfid = lmv_rmfid,
4169 .m_batch_create = lmv_batch_create,
4170 .m_batch_add = lmv_batch_add,
4171 .m_batch_stop = lmv_batch_stop,
4172 .m_batch_flush = lmv_batch_flush,
4175 static int __init lmv_init(void)
4177 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true,
4178 LUSTRE_LMV_NAME, NULL);
4181 static void __exit lmv_exit(void)
4183 class_unregister_type(LUSTRE_LMV_NAME);
4186 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4187 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
4188 MODULE_VERSION(LUSTRE_VERSION_STRING);
4189 MODULE_LICENSE("GPL");
4191 module_init(lmv_init);
4192 module_exit(lmv_exit);