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_ioctl_old.h>
56 #include <lustre_kernelcomm.h>
57 #include "lmv_internal.h"
59 static int lmv_check_connect(struct obd_device *obd);
60 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data);
62 void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt,
65 if (tgt->ltd_active == activate)
68 tgt->ltd_active = activate;
69 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count +=
72 tgt->ltd_exp->exp_obd->obd_inactive = !activate;
78 * -EINVAL : UUID can't be found in the LMV's target list
79 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
80 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
82 static int lmv_set_mdc_active(struct lmv_obd *lmv,
83 const struct obd_uuid *uuid,
86 struct lu_tgt_desc *tgt = NULL;
87 struct obd_device *obd;
92 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
93 lmv, uuid->uuid, activate);
95 spin_lock(&lmv->lmv_lock);
96 lmv_foreach_connected_tgt(lmv, tgt) {
97 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n",
98 tgt->ltd_index, tgt->ltd_uuid.uuid,
99 tgt->ltd_exp->exp_handle.h_cookie);
101 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
106 GOTO(out_lmv_lock, rc = -EINVAL);
108 obd = class_exp2obd(tgt->ltd_exp);
110 GOTO(out_lmv_lock, rc = -ENOTCONN);
112 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
113 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
114 obd->obd_type->typ_name, tgt->ltd_index);
115 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
117 if (tgt->ltd_active == activate) {
118 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
119 activate ? "" : "in");
120 GOTO(out_lmv_lock, rc);
123 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
124 activate ? "" : "in");
125 lmv_activate_target(lmv, tgt, activate);
129 spin_unlock(&lmv->lmv_lock);
133 static struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
135 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
136 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
138 return tgt ? obd_get_uuid(tgt->ltd_exp) : NULL;
141 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
142 enum obd_notify_event ev)
144 struct obd_connect_data *conn_data;
145 struct lmv_obd *lmv = &obd->u.lmv;
146 struct obd_uuid *uuid;
150 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
151 CERROR("unexpected notification of %s %s!\n",
152 watched->obd_type->typ_name,
157 uuid = &watched->u.cli.cl_target_uuid;
158 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
160 * Set MDC as active before notifying the observer, so the
161 * observer can use the MDC normally.
163 rc = lmv_set_mdc_active(lmv, uuid,
164 ev == OBD_NOTIFY_ACTIVE);
166 CERROR("%sactivation of %s failed: %d\n",
167 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
171 } else if (ev == OBD_NOTIFY_OCD) {
172 conn_data = &watched->u.cli.cl_import->imp_connect_data;
174 * XXX: Make sure that ocd_connect_flags from all targets are
175 * the same. Otherwise one of MDTs runs wrong version or
176 * something like this. --umka
178 obd->obd_self_export->exp_connect_data = *conn_data;
182 * Pass the notification up the chain.
184 if (obd->obd_observer)
185 rc = obd_notify(obd->obd_observer, watched, ev);
190 static int lmv_connect(const struct lu_env *env,
191 struct obd_export **pexp, struct obd_device *obd,
192 struct obd_uuid *cluuid, struct obd_connect_data *data,
195 struct lmv_obd *lmv = &obd->u.lmv;
196 struct lustre_handle conn = { 0 };
197 struct obd_export *exp;
201 rc = class_connect(&conn, obd, cluuid);
203 CERROR("class_connection() returned %d\n", rc);
207 exp = class_conn2export(&conn);
210 lmv->conn_data = *data;
211 lmv->lmv_cache = localdata;
213 lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
214 &obd->obd_kset.kobj);
215 if (!lmv->lmv_tgts_kobj) {
216 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
217 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
220 rc = lmv_check_connect(obd);
229 if (lmv->lmv_tgts_kobj)
230 kobject_put(lmv->lmv_tgts_kobj);
232 class_disconnect(exp);
237 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
240 struct obd_device *obd = exp->exp_obd;
241 struct lmv_obd *lmv = &obd->u.lmv;
242 struct lmv_tgt_desc *tgt;
248 if (lmv->max_easize < easize) {
249 lmv->max_easize = easize;
252 if (lmv->max_def_easize < def_easize) {
253 lmv->max_def_easize = def_easize;
260 if (lmv->connected == 0)
263 lmv_foreach_connected_tgt(lmv, tgt) {
264 if (!tgt->ltd_active)
267 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
269 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
270 " rc = %d\n", obd->obd_name, tgt->ltd_index, rc);
277 #define MAX_STRING_SIZE 128
279 static int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
281 struct lmv_obd *lmv = &obd->u.lmv;
282 struct obd_device *mdc_obd;
283 struct obd_export *mdc_exp;
284 struct lu_fld_target target;
288 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
291 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
295 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
296 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
297 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
299 if (!mdc_obd->obd_set_up) {
300 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
304 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
305 &lmv->conn_data, lmv->lmv_cache);
307 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
312 * Init fid sequence client for this mdc and add new fld target.
314 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
318 target.ft_srv = NULL;
319 target.ft_exp = mdc_exp;
320 target.ft_idx = tgt->ltd_index;
322 fld_client_add_target(&lmv->lmv_fld, &target);
324 rc = obd_register_observer(mdc_obd, obd);
326 obd_disconnect(mdc_exp);
327 CERROR("target %s register_observer error %d\n",
328 tgt->ltd_uuid.uuid, rc);
332 if (obd->obd_observer) {
334 * Tell the observer about the new target.
336 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
339 obd_disconnect(mdc_exp);
345 tgt->ltd_exp = mdc_exp;
346 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count++;
348 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
350 rc = lu_qos_add_tgt(&lmv->lmv_qos, tgt);
352 obd_disconnect(mdc_exp);
356 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
357 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
358 kref_read(&obd->obd_refcount));
360 lmv_statfs_check_update(obd, tgt);
362 if (lmv->lmv_tgts_kobj)
363 /* Even if we failed to create the link, that's fine */
364 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
365 &mdc_obd->obd_kset.kobj,
370 static void lmv_del_target(struct lmv_obd *lmv, struct lu_tgt_desc *tgt)
373 ltd_del_tgt(&lmv->lmv_mdt_descs, tgt);
377 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
378 __u32 index, int gen)
380 struct obd_device *mdc_obd;
381 struct lmv_obd *lmv = &obd->u.lmv;
382 struct lmv_tgt_desc *tgt;
383 struct lu_tgt_descs *ltd = &lmv->lmv_mdt_descs;
388 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
389 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
392 CERROR("%s: Target %s not attached: rc = %d\n",
393 obd->obd_name, uuidp->uuid, -EINVAL);
401 mutex_init(&tgt->ltd_fid_mutex);
402 tgt->ltd_index = index;
403 tgt->ltd_uuid = *uuidp;
406 mutex_lock(<d->ltd_mutex);
407 rc = ltd_add_tgt(ltd, tgt);
408 mutex_unlock(<d->ltd_mutex);
414 /* lmv_check_connect() will connect this target. */
417 rc = lmv_connect_mdc(obd, tgt);
419 int easize = sizeof(struct lmv_stripe_md) +
420 lmv->lmv_mdt_count * sizeof(struct lu_fid);
422 lmv_init_ea_size(obd->obd_self_export, easize, 0);
432 static int lmv_check_connect(struct obd_device *obd)
434 struct lmv_obd *lmv = &obd->u.lmv;
435 struct lmv_tgt_desc *tgt;
444 mutex_lock(&lmv->lmv_mdt_descs.ltd_mutex);
446 GOTO(unlock, rc = 0);
448 if (!lmv->lmv_mdt_count) {
449 CERROR("%s: no targets configured: rc = -EINVAL\n",
451 GOTO(unlock, rc = -EINVAL);
454 if (!lmv_mdt0_inited(lmv)) {
455 CERROR("%s: no target configured for index 0: rc = -EINVAL.\n",
457 GOTO(unlock, rc = -EINVAL);
460 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
461 obd->obd_uuid.uuid, obd->obd_name);
463 lmv_foreach_tgt(lmv, tgt) {
464 rc = lmv_connect_mdc(obd, tgt);
470 easize = lmv_mds_md_size(lmv->lmv_mdt_count, LMV_MAGIC);
471 lmv_init_ea_size(obd->obd_self_export, easize, 0);
474 mutex_unlock(&lmv->lmv_mdt_descs.ltd_mutex);
479 lmv_foreach_tgt(lmv, tgt) {
484 --lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count;
485 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
486 obd_disconnect(tgt->ltd_exp);
492 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
494 struct lmv_obd *lmv = &obd->u.lmv;
495 struct obd_device *mdc_obd;
499 LASSERT(tgt != NULL);
500 LASSERT(obd != NULL);
502 mdc_obd = class_exp2obd(tgt->ltd_exp);
505 mdc_obd->obd_force = obd->obd_force;
506 mdc_obd->obd_fail = obd->obd_fail;
507 mdc_obd->obd_no_recov = obd->obd_no_recov;
509 if (lmv->lmv_tgts_kobj)
510 sysfs_remove_link(lmv->lmv_tgts_kobj,
514 rc = lu_qos_del_tgt(&lmv->lmv_qos, tgt);
516 CERROR("%s: Can't del target from QoS table: rc = %d\n",
517 tgt->ltd_exp->exp_obd->obd_name, rc);
519 rc = fld_client_del_target(&lmv->lmv_fld, tgt->ltd_index);
521 CERROR("%s: Can't del fld targets: rc = %d\n",
522 tgt->ltd_exp->exp_obd->obd_name, rc);
524 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
526 CERROR("%s: Can't finalize fids factory: rc = %d\n",
527 tgt->ltd_exp->exp_obd->obd_name, rc);
529 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
530 tgt->ltd_exp->exp_obd->obd_name,
531 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
533 lmv_activate_target(lmv, tgt, 0);
534 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
535 rc = obd_disconnect(tgt->ltd_exp);
537 CERROR("%s: Target %s disconnect error: rc = %d\n",
538 tgt->ltd_exp->exp_obd->obd_name,
539 tgt->ltd_uuid.uuid, rc);
545 static int lmv_disconnect(struct obd_export *exp)
547 struct obd_device *obd = class_exp2obd(exp);
548 struct lmv_obd *lmv = &obd->u.lmv;
549 struct lmv_tgt_desc *tgt;
554 lmv_foreach_connected_tgt(lmv, tgt)
555 lmv_disconnect_mdc(obd, tgt);
557 if (lmv->lmv_tgts_kobj)
558 kobject_put(lmv->lmv_tgts_kobj);
561 rc = class_disconnect(exp);
566 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
569 struct obd_device *obd = class_exp2obd(exp);
570 struct lmv_obd *lmv = &obd->u.lmv;
571 struct getinfo_fid2path *gf;
572 struct lmv_tgt_desc *tgt;
573 struct getinfo_fid2path *remote_gf = NULL;
574 struct lu_fid root_fid;
575 int remote_gf_size = 0;
576 int currentisenc = 0;
581 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
583 RETURN(PTR_ERR(tgt));
585 root_fid = *gf->gf_u.gf_root_fid;
586 LASSERT(fid_is_sane(&root_fid));
589 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
590 if (rc != 0 && rc != -EREMOTE)
591 GOTO(out_fid2path, rc);
593 if (gf->gf_u.gf_path[0] == '/') {
594 /* by convention, server side (mdt_path_current()) puts
595 * a leading '/' to tell client that we are dealing with
604 /* If remote_gf != NULL, it means just building the
605 * path on the remote MDT, copy this path segment to gf.
607 if (remote_gf != NULL) {
608 struct getinfo_fid2path *ori_gf;
613 ori_gf = (struct getinfo_fid2path *)karg;
614 if (strlen(ori_gf->gf_u.gf_path) + 1 +
615 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
616 GOTO(out_fid2path, rc = -EOVERFLOW);
618 ptr = ori_gf->gf_u.gf_path;
619 oldisenc = ptr[0] == '/';
621 len = strlen(gf->gf_u.gf_path);
623 /* move the current path to the right to release space
624 * for closer-to-root part
626 memmove(ptr + len - currentisenc + 1 + globalisenc,
628 strlen(ori_gf->gf_u.gf_path) - oldisenc + 1);
631 memcpy(ptr, gf->gf_u.gf_path + currentisenc,
633 ptr[len - currentisenc] = '/';
637 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
638 tgt->ltd_exp->exp_obd->obd_name,
639 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
643 GOTO(out_fid2path, rc);
645 /* sigh, has to go to another MDT to do path building further */
646 if (remote_gf == NULL) {
647 remote_gf_size = sizeof(*remote_gf) + len - sizeof(*gf);
648 OBD_ALLOC(remote_gf, remote_gf_size);
649 if (remote_gf == NULL)
650 GOTO(out_fid2path, rc = -ENOMEM);
651 remote_gf->gf_pathlen = len - sizeof(*gf);
654 if (!fid_is_sane(&gf->gf_fid)) {
655 CERROR("%s: invalid FID "DFID": rc = %d\n",
656 tgt->ltd_exp->exp_obd->obd_name,
657 PFID(&gf->gf_fid), -EINVAL);
658 GOTO(out_fid2path, rc = -EINVAL);
661 tgt = lmv_fid2tgt(lmv, &gf->gf_fid);
663 GOTO(out_fid2path, rc = -EINVAL);
665 remote_gf->gf_fid = gf->gf_fid;
666 remote_gf->gf_recno = -1;
667 remote_gf->gf_linkno = -1;
668 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
669 *remote_gf->gf_u.gf_root_fid = root_fid;
671 goto repeat_fid2path;
674 if (remote_gf != NULL)
675 OBD_FREE(remote_gf, remote_gf_size);
679 static int lmv_hsm_req_count(struct lmv_obd *lmv,
680 const struct hsm_user_request *hur,
681 const struct lmv_tgt_desc *tgt_mds)
683 struct lmv_tgt_desc *curr_tgt;
687 /* count how many requests must be sent to the given target */
688 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
689 curr_tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[i].hui_fid);
690 if (IS_ERR(curr_tgt))
691 RETURN(PTR_ERR(curr_tgt));
692 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
698 static int lmv_hsm_req_build(struct lmv_obd *lmv,
699 struct hsm_user_request *hur_in,
700 const struct lmv_tgt_desc *tgt_mds,
701 struct hsm_user_request *hur_out)
704 struct lmv_tgt_desc *curr_tgt;
706 /* build the hsm_user_request for the given target */
707 hur_out->hur_request = hur_in->hur_request;
709 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
710 curr_tgt = lmv_fid2tgt(lmv, &hur_in->hur_user_item[i].hui_fid);
711 if (IS_ERR(curr_tgt))
712 RETURN(PTR_ERR(curr_tgt));
713 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
714 hur_out->hur_user_item[nr_out] =
715 hur_in->hur_user_item[i];
719 hur_out->hur_request.hr_itemcount = nr_out;
720 memcpy(hur_data(hur_out), hur_data(hur_in),
721 hur_in->hur_request.hr_data_len);
726 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
727 int len, struct lustre_kernelcomm *lk,
730 struct lmv_obd *lmv = &obd->u.lmv;
731 struct lu_tgt_desc *tgt;
736 /* unregister request (call from llapi_hsm_copytool_fini) */
737 lmv_foreach_connected_tgt(lmv, tgt)
738 /* best effort: try to clean as much as possible
739 * (continue on error) */
740 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
742 /* Whatever the result, remove copytool from kuc groups.
743 * Unreached coordinators will get EPIPE on next requests
744 * and will unregister automatically.
746 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
751 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
752 int len, struct lustre_kernelcomm *lk,
755 struct lmv_obd *lmv = &obd->u.lmv;
757 bool any_set = false;
758 struct kkuc_ct_data *kcd;
760 struct lu_tgt_desc *tgt;
767 filp = fget(lk->lk_wfd);
771 if (lk->lk_flags & LK_FLG_DATANR)
772 kcd_size = offsetof(struct kkuc_ct_data,
773 kcd_archives[lk->lk_data_count]);
775 kcd_size = sizeof(*kcd);
777 OBD_ALLOC(kcd, kcd_size);
779 GOTO(err_fput, rc = -ENOMEM);
781 kcd->kcd_nr_archives = lk->lk_data_count;
782 if (lk->lk_flags & LK_FLG_DATANR) {
783 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
784 if (lk->lk_data_count > 0)
785 memcpy(kcd->kcd_archives, lk->lk_data,
786 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
788 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
791 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
792 lk->lk_group, kcd, kcd_size);
793 OBD_FREE(kcd, kcd_size);
797 /* All or nothing: try to register to all MDS.
798 * In case of failure, unregister from previous MDS,
799 * except if it because of inactive target. */
800 lmv_foreach_connected_tgt(lmv, tgt) {
801 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
803 if (tgt->ltd_active) {
804 /* permanent error */
805 CERROR("%s: iocontrol MDC %s on MDT"
806 " idx %d cmd %x: err = %d\n",
807 lmv2obd_dev(lmv)->obd_name,
808 tgt->ltd_uuid.uuid, tgt->ltd_index, cmd,
811 lk->lk_flags |= LK_FLG_STOP;
813 /* unregister from previous MDS */
814 lmv_foreach_connected_tgt(lmv, tgt) {
815 if (tgt->ltd_index >= i)
818 obd_iocontrol(cmd, tgt->ltd_exp, len,
821 GOTO(err_kkuc_rem, rc);
823 /* else: transient error.
824 * kuc will register to the missing MDT
832 /* no registration done: return error */
833 GOTO(err_kkuc_rem, rc = -ENOTCONN);
838 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
845 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
846 int len, void *karg, void __user *uarg)
848 struct obd_device *obd = class_exp2obd(exp);
849 struct lmv_obd *lmv = &obd->u.lmv;
850 struct lu_tgt_desc *tgt = NULL;
852 __u32 count = lmv->lmv_mdt_count;
856 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
857 exp->exp_obd->obd_name, cmd, len, karg, uarg);
861 /* exit early for unknown ioctl types */
862 if (unlikely(_IOC_TYPE(cmd) != 'f' && !IOC_OSC_SET_ACTIVE_ALLOW(cmd)))
863 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "unknown", -ENOTTY));
865 /* handle commands that don't use @karg first */
867 case LL_IOC_GET_CONNECT_FLAGS:
868 tgt = lmv_tgt(lmv, 0);
870 if (tgt && tgt->ltd_exp)
871 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, NULL, uarg);
875 if (unlikely(karg == NULL))
876 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", -EINVAL));
879 case IOC_OBD_STATFS: {
880 struct obd_ioctl_data *data = karg;
881 struct obd_device *mdc_obd;
882 struct obd_statfs stat_buf = {0};
885 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
887 if (index >= lmv->lmv_mdt_descs.ltd_tgts_size)
890 tgt = lmv_tgt(lmv, index);
894 if (!tgt->ltd_active)
897 mdc_obd = class_exp2obd(tgt->ltd_exp);
902 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
903 min((int) data->ioc_plen2,
904 (int) sizeof(struct obd_uuid))))
907 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
908 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
912 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
913 min((int) data->ioc_plen1,
914 (int) sizeof(stat_buf))))
918 case OBD_IOC_QUOTACTL: {
919 struct if_quotactl *qctl = karg;
920 struct obd_quotactl *oqctl;
921 struct obd_import *imp;
923 if (qctl->qc_valid == QC_MDTIDX) {
924 tgt = lmv_tgt(lmv, qctl->qc_idx);
925 } else if (qctl->qc_valid == QC_UUID) {
926 lmv_foreach_tgt(lmv, tgt) {
927 if (!obd_uuid_equals(&tgt->ltd_uuid,
946 imp = class_exp2cliimp(tgt->ltd_exp);
947 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
948 qctl->qc_valid = QC_MDTIDX;
949 qctl->obd_uuid = tgt->ltd_uuid;
953 OBD_ALLOC_PTR(oqctl);
957 QCTL_COPY(oqctl, qctl);
958 rc = obd_quotactl(tgt->ltd_exp, oqctl);
960 QCTL_COPY_NO_PNAME(qctl, oqctl);
961 qctl->qc_valid = QC_MDTIDX;
962 qctl->obd_uuid = tgt->ltd_uuid;
967 case LL_IOC_FID2MDTIDX: {
968 struct lu_fid *fid = karg;
971 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
975 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
976 * point to user space memory for FID2MDTIDX. */
977 *(__u32 *)uarg = mdt_index;
980 case OBD_IOC_FID2PATH: {
981 rc = lmv_fid2path(exp, len, karg, uarg);
984 case LL_IOC_HSM_STATE_GET:
985 case LL_IOC_HSM_STATE_SET:
986 case LL_IOC_HSM_ACTION: {
987 struct md_op_data *op_data = karg;
989 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
991 RETURN(PTR_ERR(tgt));
993 if (tgt->ltd_exp == NULL)
996 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
999 case LL_IOC_HSM_PROGRESS: {
1000 const struct hsm_progress_kernel *hpk = karg;
1002 tgt = lmv_fid2tgt(lmv, &hpk->hpk_fid);
1004 RETURN(PTR_ERR(tgt));
1005 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1008 case LL_IOC_HSM_REQUEST: {
1009 struct hsm_user_request *hur = karg;
1010 unsigned int reqcount = hur->hur_request.hr_itemcount;
1015 /* if the request is about a single fid
1016 * or if there is a single MDS, no need to split
1018 if (reqcount == 1 || count == 1) {
1019 tgt = lmv_fid2tgt(lmv, &hur->hur_user_item[0].hui_fid);
1021 RETURN(PTR_ERR(tgt));
1022 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1024 /* split fid list to their respective MDS */
1025 lmv_foreach_connected_tgt(lmv, tgt) {
1028 struct hsm_user_request *req;
1030 nr = lmv_hsm_req_count(lmv, hur, tgt);
1033 if (nr == 0) /* nothing for this MDS */
1036 /* build a request with fids for this MDS */
1037 reqlen = offsetof(typeof(*hur),
1039 + hur->hur_request.hr_data_len;
1040 OBD_ALLOC_LARGE(req, reqlen);
1043 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1045 GOTO(hsm_req_err, rc1);
1046 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1049 if (rc1 != 0 && rc == 0)
1051 OBD_FREE_LARGE(req, reqlen);
1056 case LL_IOC_LOV_SWAP_LAYOUTS: {
1057 struct md_op_data *op_data = karg;
1058 struct lmv_tgt_desc *tgt1, *tgt2;
1060 tgt1 = lmv_fid2tgt(lmv, &op_data->op_fid1);
1062 RETURN(PTR_ERR(tgt1));
1064 tgt2 = lmv_fid2tgt(lmv, &op_data->op_fid2);
1066 RETURN(PTR_ERR(tgt2));
1068 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1071 /* only files on same MDT can have their layouts swapped */
1072 if (tgt1->ltd_index != tgt2->ltd_index)
1075 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1078 case LL_IOC_HSM_CT_START: {
1079 struct lustre_kernelcomm *lk = karg;
1081 if (lk->lk_flags & LK_FLG_STOP)
1082 rc = lmv_hsm_ct_unregister(obd, cmd, len, lk, uarg);
1084 rc = lmv_hsm_ct_register(obd, cmd, len, lk, uarg);
1088 lmv_foreach_connected_tgt(lmv, tgt) {
1089 struct obd_device *mdc_obd;
1092 /* ll_umount_begin() sets force flag but for lmv, not
1093 * mdc. Let's pass it through */
1094 mdc_obd = class_exp2obd(tgt->ltd_exp);
1095 mdc_obd->obd_force = obd->obd_force;
1096 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1098 if (tgt->ltd_active) {
1099 OBD_IOC_ERROR(obd->obd_name, cmd,
1100 tgt->ltd_uuid.uuid, err);
1103 if (unlikely(err == -ENOTTY))
1117 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1118 struct lu_fid *fid, struct md_op_data *op_data)
1120 struct obd_device *obd = class_exp2obd(exp);
1121 struct lmv_obd *lmv = &obd->u.lmv;
1122 struct lmv_tgt_desc *tgt;
1130 tgt = lmv_tgt(lmv, op_data->op_mds);
1134 if (!tgt->ltd_active || !tgt->ltd_exp)
1138 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1139 * on server that seq in new allocated fid is not yet known.
1141 mutex_lock(&tgt->ltd_fid_mutex);
1142 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1143 mutex_unlock(&tgt->ltd_fid_mutex);
1145 LASSERT(fid_is_sane(fid));
1152 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1154 struct lmv_obd *lmv = &obd->u.lmv;
1155 struct lmv_desc *desc;
1156 struct lnet_processid lnet_id;
1162 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1163 CERROR("LMV setup requires a descriptor\n");
1167 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1168 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1169 CERROR("Lmv descriptor size wrong: %d > %d\n",
1170 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1174 obd_str2uuid(&lmv->lmv_mdt_descs.ltd_lmv_desc.ld_uuid,
1175 desc->ld_uuid.uuid);
1176 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_tgt_count = 0;
1177 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count = 0;
1178 lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage =
1179 LMV_DESC_QOS_MAXAGE_DEFAULT;
1180 lmv->max_def_easize = 0;
1181 lmv->max_easize = 0;
1183 spin_lock_init(&lmv->lmv_lock);
1186 * initialize rr_index to lower 32bit of netid, so that client
1187 * can distribute subdirs evenly from the beginning.
1189 while (LNetGetId(i++, &lnet_id, false) != -ENOENT) {
1190 if (!nid_is_lo0(&lnet_id.nid)) {
1191 lmv->lmv_qos_rr_index = ntohl(lnet_id.nid.nid_addr[0]);
1196 rc = lmv_tunables_init(obd);
1198 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1201 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1202 LUSTRE_CLI_FLD_HASH_DHT);
1204 CERROR("Can't init FLD, err %d\n", rc);
1206 rc = lu_tgt_descs_init(&lmv->lmv_mdt_descs, true);
1208 CWARN("%s: error initialize target table: rc = %d\n",
1214 static int lmv_cleanup(struct obd_device *obd)
1216 struct lmv_obd *lmv = &obd->u.lmv;
1217 struct lu_tgt_desc *tgt;
1218 struct lu_tgt_desc *tmp;
1222 fld_client_fini(&lmv->lmv_fld);
1223 fld_client_debugfs_fini(&lmv->lmv_fld);
1225 lprocfs_obd_cleanup(obd);
1226 lprocfs_free_md_stats(obd);
1228 lmv_foreach_tgt_safe(lmv, tgt, tmp)
1229 lmv_del_target(lmv, tgt);
1230 lu_tgt_descs_fini(&lmv->lmv_mdt_descs);
1235 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1237 struct lustre_cfg *lcfg = buf;
1238 struct obd_uuid obd_uuid;
1244 switch (lcfg->lcfg_command) {
1246 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1247 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1248 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1249 GOTO(out, rc = -EINVAL);
1251 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1253 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1254 GOTO(out, rc = -EINVAL);
1255 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1256 GOTO(out, rc = -EINVAL);
1257 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1260 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1261 GOTO(out, rc = -EINVAL);
1267 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1271 if (flags & OBD_STATFS_FOR_MDT0)
1274 if (lmv->lmv_statfs_start || lmv->lmv_mdt_count == 1)
1275 return lmv->lmv_statfs_start;
1277 /* choose initial MDT for this client */
1279 struct lnet_processid lnet_id;
1280 if (LNetGetId(i, &lnet_id, false) == -ENOENT)
1283 if (!nid_is_lo0(&lnet_id.nid)) {
1284 /* We dont need a full 64-bit modulus, just enough
1285 * to distribute the requests across MDTs evenly.
1287 lmv->lmv_statfs_start = nidhash(&lnet_id.nid) %
1293 return lmv->lmv_statfs_start;
1296 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1297 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1299 struct obd_device *obd = class_exp2obd(exp);
1300 struct lmv_obd *lmv = &obd->u.lmv;
1301 struct obd_statfs *temp;
1302 struct lu_tgt_desc *tgt;
1310 OBD_ALLOC(temp, sizeof(*temp));
1314 /* distribute statfs among MDTs */
1315 idx = lmv_select_statfs_mdt(lmv, flags);
1317 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++, idx++) {
1318 idx = idx % lmv->lmv_mdt_descs.ltd_tgts_size;
1319 tgt = lmv_tgt(lmv, idx);
1320 if (!tgt || !tgt->ltd_exp)
1323 rc = obd_statfs(env, tgt->ltd_exp, temp, max_age,
1324 flags | OBD_STATFS_NESTED);
1326 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1327 tgt->ltd_exp->exp_obd->obd_name, i, rc);
1329 /* Try another MDT */
1330 if (flags & OBD_STATFS_SUM)
1332 GOTO(out_free_temp, rc);
1335 if (temp->os_state & OS_STATFS_SUM ||
1336 flags == OBD_STATFS_FOR_MDT0) {
1337 /* reset to the last aggregated values
1338 * and don't sum with non-aggrated data */
1339 /* If the statfs is from mount, it needs to retrieve
1340 * necessary information from MDT0. i.e. mount does
1341 * not need the merged osfs from all of MDT. Also
1342 * clients can be mounted as long as MDT0 is in
1345 GOTO(out_free_temp, rc);
1351 osfs->os_bavail += temp->os_bavail;
1352 osfs->os_blocks += temp->os_blocks;
1353 osfs->os_ffree += temp->os_ffree;
1354 osfs->os_files += temp->os_files;
1355 osfs->os_granted += temp->os_granted;
1358 /* There is no stats from some MDTs, data incomplete */
1362 OBD_FREE(temp, sizeof(*temp));
1366 static int lmv_statfs_update(void *cookie, int rc)
1368 struct obd_info *oinfo = cookie;
1369 struct obd_device *obd = oinfo->oi_obd;
1370 struct lmv_obd *lmv = &obd->u.lmv;
1371 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1372 struct obd_statfs *osfs = oinfo->oi_osfs;
1375 * NB: don't deactivate TGT upon error, because we may not trigger async
1376 * statfs any longer, then there is no chance to activate TGT.
1379 spin_lock(&lmv->lmv_lock);
1380 tgt->ltd_statfs = *osfs;
1381 tgt->ltd_statfs_age = ktime_get_seconds();
1382 spin_unlock(&lmv->lmv_lock);
1383 set_bit(LQ_DIRTY, &lmv->lmv_qos.lq_flags);
1389 /* update tgt statfs async if it's ld_qos_maxage old */
1390 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1392 struct obd_info oinfo = {
1395 .oi_cb_up = lmv_statfs_update,
1399 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1400 obd->u.lmv.lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1403 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1408 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1411 struct obd_device *obd = exp->exp_obd;
1412 struct lmv_obd *lmv = &obd->u.lmv;
1413 struct lu_tgt_desc *tgt = lmv_tgt(lmv, 0);
1421 rc = md_get_root(tgt->ltd_exp, fileset, fid);
1425 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1426 u64 obd_md_valid, const char *name, size_t buf_size,
1427 struct ptlrpc_request **req)
1429 struct obd_device *obd = exp->exp_obd;
1430 struct lmv_obd *lmv = &obd->u.lmv;
1431 struct lmv_tgt_desc *tgt;
1436 tgt = lmv_fid2tgt(lmv, fid);
1438 RETURN(PTR_ERR(tgt));
1440 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1445 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1446 u64 obd_md_valid, const char *name,
1447 const void *value, size_t value_size,
1448 unsigned int xattr_flags, u32 suppgid,
1449 struct ptlrpc_request **req)
1451 struct obd_device *obd = exp->exp_obd;
1452 struct lmv_obd *lmv = &obd->u.lmv;
1453 struct lmv_tgt_desc *tgt;
1458 tgt = lmv_fid2tgt(lmv, fid);
1460 RETURN(PTR_ERR(tgt));
1462 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1463 value, value_size, xattr_flags, suppgid, req);
1468 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1469 struct ptlrpc_request **request)
1471 struct obd_device *obd = exp->exp_obd;
1472 struct lmv_obd *lmv = &obd->u.lmv;
1473 struct lmv_tgt_desc *tgt;
1478 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1480 RETURN(PTR_ERR(tgt));
1482 if (op_data->op_flags & MF_GET_MDT_IDX) {
1483 op_data->op_mds = tgt->ltd_index;
1487 rc = md_getattr(tgt->ltd_exp, op_data, request);
1492 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1494 struct obd_device *obd = exp->exp_obd;
1495 struct lmv_obd *lmv = &obd->u.lmv;
1496 struct lu_tgt_desc *tgt;
1500 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1503 * With DNE every object can have two locks in different namespaces:
1504 * lookup lock in space of MDT storing direntry and update/open lock in
1505 * space of MDT storing inode.
1507 lmv_foreach_connected_tgt(lmv, tgt)
1508 md_null_inode(tgt->ltd_exp, fid);
1513 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1514 struct md_open_data *mod, struct ptlrpc_request **request)
1516 struct obd_device *obd = exp->exp_obd;
1517 struct lmv_obd *lmv = &obd->u.lmv;
1518 struct lmv_tgt_desc *tgt;
1523 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
1525 RETURN(PTR_ERR(tgt));
1527 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1528 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1532 static struct lu_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv,
1533 struct md_op_data *op_data)
1535 struct lu_tgt_desc *tgt, *cur = NULL;
1536 __u64 total_avail = 0;
1537 __u64 total_weight = 0;
1538 __u64 cur_weight = 0;
1539 int total_usable = 0;
1545 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1546 RETURN(ERR_PTR(-EAGAIN));
1548 down_write(&lmv->lmv_qos.lq_rw_sem);
1550 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1551 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1553 rc = ltd_qos_penalties_calc(&lmv->lmv_mdt_descs);
1555 GOTO(unlock, tgt = ERR_PTR(rc));
1557 lmv_foreach_tgt(lmv, tgt) {
1558 if (!tgt->ltd_exp || !tgt->ltd_active) {
1559 tgt->ltd_qos.ltq_usable = 0;
1562 /* update one hour overdue statfs */
1563 if (ktime_get_seconds() - tgt->ltd_statfs_age >
1564 60 * lmv->lmv_mdt_descs.ltd_lmv_desc.ld_qos_maxage)
1565 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1566 tgt->ltd_qos.ltq_usable = 1;
1567 lu_tgt_qos_weight_calc(tgt, true);
1568 if (tgt->ltd_index == op_data->op_mds)
1570 total_avail += tgt->ltd_qos.ltq_avail;
1571 total_weight += tgt->ltd_qos.ltq_weight;
1575 /* If current MDT has above-average space and dir is not aleady using
1576 * round-robin to spread across more MDTs, stay on the parent MDT
1577 * to avoid creating needless remote MDT directories. Remote dirs
1578 * close to the root balance space more effectively than bottom dirs,
1579 * so prefer to create remote dirs at top level of directory tree.
1580 * "16 / (dir_depth + 10)" is the factor to make it less likely
1581 * for top-level directories to stay local unless they have more than
1582 * average free space, while deep dirs prefer local until more full.
1583 * depth=0 -> 160%, depth=3 -> 123%, depth=6 -> 100%,
1584 * depth=9 -> 84%, depth=12 -> 73%, depth=15 -> 64%
1586 if (!lmv_op_default_rr_mkdir(op_data)) {
1587 rand = total_avail * 16 /
1588 (total_usable * (op_data->op_dir_depth + 10));
1589 if (cur && cur->ltd_qos.ltq_avail >= rand) {
1595 rand = lu_prandom_u64_max(total_weight);
1597 lmv_foreach_connected_tgt(lmv, tgt) {
1598 if (!tgt->ltd_qos.ltq_usable)
1601 cur_weight += tgt->ltd_qos.ltq_weight;
1602 if (cur_weight < rand)
1605 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1609 /* no proper target found */
1610 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1612 up_write(&lmv->lmv_qos.lq_rw_sem);
1617 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv)
1619 struct lu_tgt_desc *tgt;
1625 spin_lock(&lmv->lmv_lock);
1626 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1627 index = (i + lmv->lmv_qos_rr_index) %
1628 lmv->lmv_mdt_descs.ltd_tgts_size;
1629 tgt = lmv_tgt(lmv, index);
1630 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1633 lmv->lmv_qos_rr_index = (tgt->ltd_index + 1) %
1634 lmv->lmv_mdt_descs.ltd_tgts_size;
1635 spin_unlock(&lmv->lmv_lock);
1639 spin_unlock(&lmv->lmv_lock);
1641 RETURN(ERR_PTR(-ENODEV));
1644 /* locate MDT which is less full (avoid the most full MDT) */
1645 static struct lu_tgt_desc *lmv_locate_tgt_lf(struct lmv_obd *lmv)
1647 struct lu_tgt_desc *min = NULL;
1648 struct lu_tgt_desc *tgt;
1654 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1655 RETURN(ERR_PTR(-EAGAIN));
1657 down_write(&lmv->lmv_qos.lq_rw_sem);
1659 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1660 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1662 lmv_foreach_tgt(lmv, tgt) {
1663 if (!tgt->ltd_exp || !tgt->ltd_active) {
1664 tgt->ltd_qos.ltq_usable = 0;
1668 tgt->ltd_qos.ltq_usable = 1;
1669 lu_tgt_qos_weight_calc(tgt, true);
1670 avail += tgt->ltd_qos.ltq_avail;
1671 if (!min || min->ltd_qos.ltq_avail > tgt->ltd_qos.ltq_avail)
1675 /* avoid the most full MDT */
1677 avail -= min->ltd_qos.ltq_avail;
1679 rand = lu_prandom_u64_max(avail);
1681 lmv_foreach_connected_tgt(lmv, tgt) {
1682 if (!tgt->ltd_qos.ltq_usable)
1688 avail += tgt->ltd_qos.ltq_avail;
1695 /* no proper target found */
1696 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1698 up_write(&lmv->lmv_qos.lq_rw_sem);
1703 /* locate MDT by file name, for striped directory, the file name hash decides
1704 * which stripe its dirent is stored.
1706 static struct lmv_tgt_desc *
1707 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_object *lso,
1708 const char *name, int namelen, struct lu_fid *fid,
1709 __u32 *mds, bool new_layout)
1711 struct lmv_tgt_desc *tgt;
1712 const struct lmv_oinfo *oinfo;
1714 if (!lmv_dir_striped(lso) || !namelen) {
1715 tgt = lmv_fid2tgt(lmv, fid);
1719 *mds = tgt->ltd_index;
1723 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1724 if (cfs_fail_val >= lso->lso_lsm.lsm_md_stripe_count)
1725 return ERR_PTR(-EBADF);
1726 oinfo = &lso->lso_lsm.lsm_md_oinfo[cfs_fail_val];
1728 oinfo = lsm_name_to_stripe_info(lso, name, namelen, new_layout);
1730 return ERR_CAST(oinfo);
1733 /* check stripe FID is sane */
1734 if (!fid_is_sane(&oinfo->lmo_fid))
1735 return ERR_PTR(-ENODEV);
1737 *fid = oinfo->lmo_fid;
1738 *mds = oinfo->lmo_mds;
1739 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1741 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1743 return tgt ? tgt : ERR_PTR(-ENODEV);
1747 * Locate MDT of op_data->op_fid1
1749 * For striped directory, it will locate the stripe by name hash, if hash_type
1750 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1751 * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1752 * indicates whether old or new layout is used to locate.
1754 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1756 * \param[in] lmv LMV device
1757 * \param[in/out] op_data client MD stack parameters, name, namelen etc,
1758 * op_mds and op_fid1 will be updated if op_lso1
1759 * indicates fid1 represents a striped directory.
1761 * retval pointer to the lmv_tgt_desc if succeed.
1762 * ERR_PTR(errno) if failed.
1764 struct lmv_tgt_desc *
1765 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1767 struct lmv_stripe_md *lsm;
1768 struct lmv_oinfo *oinfo;
1769 struct lmv_tgt_desc *tgt;
1771 if (lmv_dir_foreign(op_data->op_lso1))
1772 return ERR_PTR(-ENODATA);
1774 /* During creating VOLATILE file, it should honor the mdt
1775 * index if the file under striped dir is being restored, see
1777 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1778 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1779 tgt = lmv_tgt(lmv, op_data->op_mds);
1781 return ERR_PTR(-ENODEV);
1783 if (lmv_dir_striped(op_data->op_lso1)) {
1786 /* refill the right parent fid */
1787 lsm = &op_data->op_lso1->lso_lsm;
1788 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1789 oinfo = &lsm->lsm_md_oinfo[i];
1790 if (oinfo->lmo_mds == op_data->op_mds) {
1791 op_data->op_fid1 = oinfo->lmo_fid;
1796 if (i == lsm->lsm_md_stripe_count)
1797 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1799 } else if (lmv_dir_bad_hash(op_data->op_lso1)) {
1800 lsm = &op_data->op_lso1->lso_lsm;
1802 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1803 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1805 op_data->op_fid1 = oinfo->lmo_fid;
1806 op_data->op_mds = oinfo->lmo_mds;
1807 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1809 return ERR_PTR(-ENODEV);
1811 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_lso1,
1812 op_data->op_name, op_data->op_namelen,
1813 &op_data->op_fid1, &op_data->op_mds,
1814 op_data->op_new_layout);
1820 /* Locate MDT of op_data->op_fid2 for link/rename */
1821 static struct lmv_tgt_desc *
1822 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1824 struct lmv_tgt_desc *tgt;
1827 LASSERT(op_data->op_name);
1828 if (lmv_dir_layout_changing(op_data->op_lso2)) {
1829 struct lu_fid fid1 = op_data->op_fid1;
1830 struct lmv_stripe_object *lso1 = op_data->op_lso1;
1831 struct ptlrpc_request *request = NULL;
1834 * avoid creating new file under old layout of migrating
1835 * directory, check it here.
1837 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_lso2,
1838 op_data->op_name, op_data->op_namelen,
1839 &op_data->op_fid2, &op_data->op_mds, false);
1843 op_data->op_fid1 = op_data->op_fid2;
1844 op_data->op_lso1 = op_data->op_lso2;
1845 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1846 op_data->op_fid1 = fid1;
1847 op_data->op_lso1 = lso1;
1849 ptlrpc_req_finished(request);
1850 RETURN(ERR_PTR(-EEXIST));
1854 RETURN(ERR_PTR(rc));
1857 return lmv_locate_tgt_by_name(lmv, op_data->op_lso2,
1858 op_data->op_name, op_data->op_namelen,
1859 &op_data->op_fid2, &op_data->op_mds,
1863 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1865 struct lu_tgt_desc *tgt;
1866 struct ptlrpc_request *request;
1869 LASSERT(lmv_dir_layout_changing(op_data->op_lso1));
1870 LASSERT(!op_data->op_new_layout);
1872 tgt = lmv_locate_tgt(lmv, op_data);
1874 return PTR_ERR(tgt);
1876 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1878 ptlrpc_req_finished(request);
1885 /* mkdir by QoS upon 'lfs mkdir -i -1'.
1887 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1888 * directories under striped directory.
1890 static inline bool lmv_op_user_qos_mkdir(const struct md_op_data *op_data)
1892 const struct lmv_user_md *lum = op_data->op_data;
1894 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1897 if (lmv_dir_striped(op_data->op_lso1))
1900 return (op_data->op_cli_flags & CLI_SET_MEA) && lum &&
1901 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
1902 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
1905 /* mkdir by QoS if either ROOT or parent default LMV is space balanced. */
1906 static inline bool lmv_op_default_qos_mkdir(const struct md_op_data *op_data)
1908 const struct lmv_stripe_object *lso = op_data->op_default_lso1;
1910 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1913 if (lmv_dir_striped(op_data->op_lso1))
1916 return (op_data->op_flags & MF_QOS_MKDIR) ||
1917 (lso && lso->lso_lsm.lsm_md_master_mdt_index ==
1918 LMV_OFFSET_DEFAULT);
1921 /* if parent default LMV is space balanced, and
1922 * 1. max_inherit_rr is set
1923 * 2. or parent is ROOT
1924 * mkdir roundrobin. Or if parent doesn't have default LMV, while ROOT default
1925 * LMV requests roundrobin mkdir, do the same.
1926 * NB, this needs to check server is balanced, which is done by caller.
1928 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data)
1930 const struct lmv_stripe_object *lso = op_data->op_default_lso1;
1932 return (op_data->op_flags & MF_RR_MKDIR) ||
1933 (lso && lso->lso_lsm.lsm_md_max_inherit_rr !=
1934 LMV_INHERIT_RR_NONE) || fid_is_root(&op_data->op_fid1);
1937 /* 'lfs mkdir -i <specific_MDT>' */
1938 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1940 const struct lmv_user_md *lum = op_data->op_data;
1942 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1943 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1944 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1945 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1946 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1949 /* parent default LMV master_mdt_index is not -1. */
1951 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1953 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1954 op_data->op_default_lso1 &&
1955 op_data->op_default_lso1->lso_lsm.lsm_md_master_mdt_index !=
1959 /* locate MDT by space usage */
1960 static struct lu_tgt_desc *lmv_locate_tgt_by_space(struct lmv_obd *lmv,
1961 struct md_op_data *op_data,
1962 struct lmv_tgt_desc *tgt)
1964 struct lmv_tgt_desc *tmp = tgt;
1966 tgt = lmv_locate_tgt_qos(lmv, op_data);
1967 if (tgt == ERR_PTR(-EAGAIN)) {
1968 if (ltd_qos_is_balanced(&lmv->lmv_mdt_descs) &&
1969 !lmv_op_default_rr_mkdir(op_data) &&
1970 !lmv_op_user_qos_mkdir(op_data))
1971 /* if not necessary, don't create remote directory. */
1974 tgt = lmv_locate_tgt_rr(lmv);
1976 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1980 op_data->op_mds = tgt->ltd_index;
1985 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1986 const void *data, size_t datalen, umode_t mode, uid_t uid,
1987 gid_t gid, kernel_cap_t cap_effective, __u64 rdev,
1988 struct ptlrpc_request **request)
1990 struct obd_device *obd = exp->exp_obd;
1991 struct lmv_obd *lmv = &obd->u.lmv;
1992 struct lmv_tgt_desc *tgt;
1993 struct mdt_body *repbody;
1998 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
2001 if (lmv_dir_bad_hash(op_data->op_lso1))
2004 if (lmv_dir_layout_changing(op_data->op_lso1)) {
2006 * if parent is migrating, create() needs to lookup existing
2007 * name in both old and new layout, check old layout on client.
2009 rc = lmv_old_layout_lookup(lmv, op_data);
2013 op_data->op_new_layout = true;
2016 tgt = lmv_locate_tgt(lmv, op_data);
2018 RETURN(PTR_ERR(tgt));
2020 /* the order to apply policy in mkdir:
2021 * 1. is "lfs mkdir -i N"? mkdir on MDT N.
2022 * 2. is "lfs mkdir -i -1"? mkdir by space usage.
2023 * 3. is starting MDT specified in default LMV? mkdir on MDT N.
2024 * 4. is default LMV space balanced? mkdir by space usage.
2026 if (lmv_op_user_specific_mkdir(op_data)) {
2027 struct lmv_user_md *lum = op_data->op_data;
2029 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2030 tgt = lmv_tgt(lmv, op_data->op_mds);
2033 } else if (lmv_op_user_qos_mkdir(op_data)) {
2034 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2036 RETURN(PTR_ERR(tgt));
2037 } else if (lmv_op_default_specific_mkdir(op_data)) {
2038 struct lmv_stripe_md *lsm = &op_data->op_default_lso1->lso_lsm;
2040 op_data->op_mds = lsm->lsm_md_master_mdt_index;
2041 tgt = lmv_tgt(lmv, op_data->op_mds);
2044 } else if (lmv_op_default_qos_mkdir(op_data)) {
2045 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2047 RETURN(PTR_ERR(tgt));
2051 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2055 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
2056 (int)op_data->op_namelen, op_data->op_name,
2057 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
2060 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2061 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
2062 cap_effective, rdev, request);
2064 if (*request == NULL)
2066 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
2069 /* dir restripe needs to send to MDT where dir is located */
2070 if (rc != -EREMOTE ||
2071 !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
2074 repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2075 if (repbody == NULL)
2078 /* Not cross-ref case, just get out of here. */
2079 if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
2082 op_data->op_fid2 = repbody->mbo_fid1;
2083 ptlrpc_req_finished(*request);
2086 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2088 RETURN(PTR_ERR(tgt));
2090 op_data->op_mds = tgt->ltd_index;
2095 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
2096 const union ldlm_policy_data *policy, struct md_op_data *op_data,
2097 struct lustre_handle *lockh, __u64 extra_lock_flags)
2099 struct obd_device *obd = exp->exp_obd;
2100 struct lmv_obd *lmv = &obd->u.lmv;
2101 struct lmv_tgt_desc *tgt;
2106 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
2108 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2110 RETURN(PTR_ERR(tgt));
2112 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
2113 PFID(&op_data->op_fid1), tgt->ltd_index);
2115 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
2122 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
2123 struct ptlrpc_request **preq)
2125 struct obd_device *obd = exp->exp_obd;
2126 struct lmv_obd *lmv = &obd->u.lmv;
2127 struct lmv_tgt_desc *tgt;
2128 struct mdt_body *body;
2134 if (op_data->op_namelen == 2 &&
2135 op_data->op_name[0] == '.' && op_data->op_name[1] == '.')
2136 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2138 tgt = lmv_locate_tgt(lmv, op_data);
2140 RETURN(PTR_ERR(tgt));
2142 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
2143 (int)op_data->op_namelen, op_data->op_name,
2144 PFID(&op_data->op_fid1), tgt->ltd_index);
2146 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2147 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2148 ptlrpc_req_finished(*preq);
2156 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2157 LASSERT(body != NULL);
2159 if (body->mbo_valid & OBD_MD_MDS) {
2160 op_data->op_fid1 = body->mbo_fid1;
2161 op_data->op_valid |= OBD_MD_FLCROSSREF;
2162 op_data->op_namelen = 0;
2163 op_data->op_name = NULL;
2165 ptlrpc_req_finished(*preq);
2174 #define md_op_data_fid(op_data, fl) \
2175 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2176 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2177 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2178 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2181 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2182 struct md_op_data *op_data, __u32 op_tgt,
2183 enum ldlm_mode mode, int bits, int flag)
2185 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2186 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2187 union ldlm_policy_data policy = { { 0 } };
2191 if (!fid_is_sane(fid))
2195 tgt = lmv_fid2tgt(lmv, fid);
2197 RETURN(PTR_ERR(tgt));
2200 if (tgt->ltd_index != op_tgt) {
2201 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2202 policy.l_inodebits.bits = bits;
2203 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2204 mode, LCF_ASYNC, NULL);
2207 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2209 op_data->op_flags |= flag;
2217 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2220 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2221 struct ptlrpc_request **request)
2223 struct obd_device *obd = exp->exp_obd;
2224 struct lmv_obd *lmv = &obd->u.lmv;
2225 struct lmv_tgt_desc *tgt;
2229 LASSERT(op_data->op_namelen != 0);
2231 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2232 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2233 op_data->op_name, PFID(&op_data->op_fid1));
2235 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2236 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2237 op_data->op_cap = current_cap();
2239 tgt = lmv_locate_tgt2(lmv, op_data);
2241 RETURN(PTR_ERR(tgt));
2244 * Cancel UPDATE lock on child (fid1).
2246 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2247 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2248 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2252 rc = md_link(tgt->ltd_exp, op_data, request);
2257 /* migrate the top directory */
2258 static inline bool lmv_op_topdir_migrate(const struct md_op_data *op_data)
2260 if (!S_ISDIR(op_data->op_mode))
2263 if (lmv_dir_layout_changing(op_data->op_lso1))
2269 /* migrate top dir to specific MDTs */
2270 static inline bool lmv_topdir_specific_migrate(const struct md_op_data *op_data)
2272 const struct lmv_user_md *lum = op_data->op_data;
2274 if (!lmv_op_topdir_migrate(op_data))
2277 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2280 /* migrate top dir in QoS mode if user issued "lfs migrate -m -1..." */
2281 static inline bool lmv_topdir_qos_migrate(const struct md_op_data *op_data)
2283 const struct lmv_user_md *lum = op_data->op_data;
2285 if (!lmv_op_topdir_migrate(op_data))
2288 return le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
2291 static inline bool lmv_subdir_specific_migrate(const struct md_op_data *op_data)
2293 const struct lmv_user_md *lum = op_data->op_data;
2295 if (!S_ISDIR(op_data->op_mode))
2298 if (!lmv_dir_layout_changing(op_data->op_lso1))
2301 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2304 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2305 const char *name, size_t namelen,
2306 struct ptlrpc_request **request)
2308 struct obd_device *obd = exp->exp_obd;
2309 struct lmv_obd *lmv = &obd->u.lmv;
2310 struct lmv_stripe_object *lso = op_data->op_lso1;
2311 struct lmv_tgt_desc *parent_tgt;
2312 struct lmv_tgt_desc *sp_tgt;
2313 struct lmv_tgt_desc *tp_tgt = NULL;
2314 struct lmv_tgt_desc *child_tgt;
2315 struct lmv_tgt_desc *tgt;
2316 struct lu_fid target_fid = { 0 };
2321 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2323 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2324 PFID(&op_data->op_fid1), (int)namelen, name);
2326 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2327 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2328 op_data->op_cap = current_cap();
2330 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2331 if (IS_ERR(parent_tgt))
2332 RETURN(PTR_ERR(parent_tgt));
2334 if (lmv_dir_striped(lso)) {
2335 const struct lmv_oinfo *oinfo;
2337 oinfo = lsm_name_to_stripe_info(lso, name, namelen, false);
2339 RETURN(PTR_ERR(oinfo));
2341 /* save source stripe FID in fid4 temporarily for ELC */
2342 op_data->op_fid4 = oinfo->lmo_fid;
2343 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2348 * if parent is being migrated too, fill op_fid2 with target
2349 * stripe fid, otherwise the target stripe is not created yet.
2351 if (lmv_dir_layout_changing(lso)) {
2352 oinfo = lsm_name_to_stripe_info(lso, name, namelen,
2355 RETURN(PTR_ERR(oinfo));
2357 op_data->op_fid2 = oinfo->lmo_fid;
2358 tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2362 /* parent unchanged and update namespace only */
2363 if (lu_fid_eq(&op_data->op_fid4, &op_data->op_fid2) &&
2364 op_data->op_bias & MDS_MIGRATE_NSONLY)
2368 sp_tgt = parent_tgt;
2371 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2372 if (IS_ERR(child_tgt))
2373 RETURN(PTR_ERR(child_tgt));
2375 if (lmv_topdir_specific_migrate(op_data)) {
2376 struct lmv_user_md *lum = op_data->op_data;
2378 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2379 } else if (lmv_topdir_qos_migrate(op_data)) {
2380 tgt = lmv_locate_tgt_lf(lmv);
2381 if (tgt == ERR_PTR(-EAGAIN))
2382 tgt = lmv_locate_tgt_rr(lmv);
2384 RETURN(PTR_ERR(tgt));
2386 op_data->op_mds = tgt->ltd_index;
2387 } else if (lmv_subdir_specific_migrate(op_data)) {
2388 struct lmv_user_md *lum = op_data->op_data;
2392 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2393 /* adjust MDTs in lum, since subdir is located on where
2394 * its parent stripe is, not the first specified MDT.
2396 for (i = 0; i < le32_to_cpu(lum->lum_stripe_count);
2398 if (le32_to_cpu(lum->lum_objects[i].lum_mds) ==
2403 if (i == le32_to_cpu(lum->lum_stripe_count))
2406 lum->lum_objects[i].lum_mds =
2407 lum->lum_objects[0].lum_mds;
2408 lum->lum_objects[0].lum_mds =
2409 cpu_to_le32(tp_tgt->ltd_index);
2411 /* NB, the above adjusts subdir migration for command like
2412 * "lfs migrate -m 0,1,2 ...", but for migration like
2413 * "lfs migrate -m 0 -c 2 ...", the top dir is migrated to MDT0
2414 * and MDT1, however its subdir may be migrated to MDT1 and MDT2
2417 lum->lum_stripe_offset = cpu_to_le32(tp_tgt->ltd_index);
2418 op_data->op_mds = tp_tgt->ltd_index;
2419 } else if (tp_tgt) {
2420 op_data->op_mds = tp_tgt->ltd_index;
2422 op_data->op_mds = sp_tgt->ltd_index;
2425 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2430 * for directory, send migrate request to the MDT where the object will
2431 * be migrated to, because we can't create a striped directory remotely.
2433 * otherwise, send to the MDT where source is located because regular
2434 * file may open lease.
2436 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2437 * backward compatibility.
2439 if (S_ISDIR(op_data->op_mode) &&
2440 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2441 tgt = lmv_fid2tgt(lmv, &target_fid);
2443 RETURN(PTR_ERR(tgt));
2448 /* cancel UPDATE lock of parent master object */
2449 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2450 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2454 /* cancel UPDATE lock of source parent */
2455 if (sp_tgt != parent_tgt) {
2457 * migrate RPC packs master object FID, because we can only pack
2458 * two FIDs in reint RPC, but MDS needs to know both source
2459 * parent and target parent, and it will obtain them from master
2460 * FID and LMV, the other FID in RPC is kept for target.
2462 * since this FID is not passed to MDC, cancel it anyway.
2464 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2465 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2469 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2471 op_data->op_fid4 = target_fid;
2473 /* cancel UPDATE locks of target parent */
2474 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2475 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2479 /* cancel LOOKUP lock of source if source is remote object */
2480 if (child_tgt != sp_tgt) {
2481 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2482 LCK_EX, MDS_INODELOCK_LOOKUP,
2483 MF_MDC_CANCEL_FID3);
2488 /* cancel ELC locks of source */
2489 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2490 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2494 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2499 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2500 const char *old, size_t oldlen,
2501 const char *new, size_t newlen,
2502 struct ptlrpc_request **request)
2504 struct obd_device *obd = exp->exp_obd;
2505 struct lmv_obd *lmv = &obd->u.lmv;
2506 struct lmv_tgt_desc *sp_tgt;
2507 struct lmv_tgt_desc *tp_tgt = NULL;
2508 struct lmv_tgt_desc *src_tgt = NULL;
2509 struct lmv_tgt_desc *tgt;
2510 struct mdt_body *body;
2515 LASSERT(oldlen != 0);
2517 if (op_data->op_cli_flags & CLI_MIGRATE) {
2518 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2522 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2523 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2524 op_data->op_cap = current_cap();
2526 op_data->op_name = new;
2527 op_data->op_namelen = newlen;
2529 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2531 RETURN(PTR_ERR(tp_tgt));
2533 /* Since the target child might be destroyed, and it might become
2534 * orphan, and we can only check orphan on the local MDT right now, so
2535 * we send rename request to the MDT where target child is located. If
2536 * target child does not exist, then it will send the request to the
2538 if (fid_is_sane(&op_data->op_fid4)) {
2539 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2541 RETURN(PTR_ERR(tgt));
2546 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2548 /* cancel UPDATE locks of target parent */
2549 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2550 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2554 if (fid_is_sane(&op_data->op_fid4)) {
2555 /* cancel LOOKUP lock of target on target parent */
2556 if (tgt != tp_tgt) {
2557 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2558 tgt->ltd_index, LCK_EX,
2559 MDS_INODELOCK_LOOKUP,
2560 MF_MDC_CANCEL_FID4);
2566 if (fid_is_sane(&op_data->op_fid3)) {
2567 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2568 if (IS_ERR(src_tgt))
2569 RETURN(PTR_ERR(src_tgt));
2571 /* cancel ELC locks of source */
2572 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2573 LCK_EX, MDS_INODELOCK_ELC,
2574 MF_MDC_CANCEL_FID3);
2579 op_data->op_name = old;
2580 op_data->op_namelen = oldlen;
2582 sp_tgt = lmv_locate_tgt(lmv, op_data);
2584 RETURN(PTR_ERR(sp_tgt));
2586 /* cancel UPDATE locks of source parent */
2587 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2588 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2592 if (fid_is_sane(&op_data->op_fid3)) {
2593 /* cancel LOOKUP lock of source on source parent */
2594 if (src_tgt != sp_tgt) {
2595 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2596 tgt->ltd_index, LCK_EX,
2597 MDS_INODELOCK_LOOKUP,
2598 MF_MDC_CANCEL_FID3);
2605 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2606 PFID(&op_data->op_fid1), (int)oldlen, old,
2607 PFID(&op_data->op_fid2), (int)newlen, new);
2609 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2611 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2612 ptlrpc_req_finished(*request);
2617 if (rc && rc != -EXDEV)
2620 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2624 /* Not cross-ref case, just get out of here. */
2625 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2628 op_data->op_fid4 = body->mbo_fid1;
2630 ptlrpc_req_finished(*request);
2633 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2635 RETURN(PTR_ERR(tgt));
2637 if (fid_is_sane(&op_data->op_fid4)) {
2638 /* cancel LOOKUP lock of target on target parent */
2639 if (tgt != tp_tgt) {
2640 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2641 tgt->ltd_index, LCK_EX,
2642 MDS_INODELOCK_LOOKUP,
2643 MF_MDC_CANCEL_FID4);
2652 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2653 void *ea, size_t ealen, struct ptlrpc_request **request)
2655 struct obd_device *obd = exp->exp_obd;
2656 struct lmv_obd *lmv = &obd->u.lmv;
2657 struct lmv_tgt_desc *tgt;
2662 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2663 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2664 op_data->op_xvalid);
2666 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2667 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2669 RETURN(PTR_ERR(tgt));
2671 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2676 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2677 struct ptlrpc_request **request)
2679 struct obd_device *obd = exp->exp_obd;
2680 struct lmv_obd *lmv = &obd->u.lmv;
2681 struct lmv_tgt_desc *tgt;
2686 tgt = lmv_fid2tgt(lmv, fid);
2688 RETURN(PTR_ERR(tgt));
2690 rc = md_fsync(tgt->ltd_exp, fid, request);
2694 struct stripe_dirent {
2695 struct page *sd_page;
2696 struct lu_dirpage *sd_dp;
2697 struct lu_dirent *sd_ent;
2701 struct lmv_dir_ctxt {
2702 struct lmv_obd *ldc_lmv;
2703 struct md_op_data *ldc_op_data;
2704 struct md_readdir_info *ldc_mrinfo;
2707 struct stripe_dirent ldc_stripes[0];
2710 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2712 if (stripe->sd_page) {
2713 kunmap(stripe->sd_page);
2714 put_page(stripe->sd_page);
2715 stripe->sd_page = NULL;
2716 stripe->sd_ent = NULL;
2720 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2724 for (i = 0; i < ctxt->ldc_count; i++)
2725 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2728 /* if @ent is dummy, or . .., get next */
2729 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2730 struct lu_dirent *ent,
2733 for (; ent; ent = lu_dirent_next(ent)) {
2734 /* Skip dummy entry */
2735 if (le16_to_cpu(ent->lde_namelen) == 0)
2738 /* skip . and .. for other stripes */
2740 (strncmp(ent->lde_name, ".",
2741 le16_to_cpu(ent->lde_namelen)) == 0 ||
2742 strncmp(ent->lde_name, "..",
2743 le16_to_cpu(ent->lde_namelen)) == 0))
2746 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2753 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2754 struct stripe_dirent *stripe,
2757 struct md_op_data *op_data = ctxt->ldc_op_data;
2758 struct lmv_oinfo *oinfo;
2759 struct lu_fid fid = op_data->op_fid1;
2760 struct inode *inode = op_data->op_data;
2761 struct lmv_tgt_desc *tgt;
2762 struct lu_dirent *ent = stripe->sd_ent;
2763 __u64 hash = ctxt->ldc_hash;
2768 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2772 if (stripe->sd_page) {
2773 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2775 /* @hash should be the last dirent hash */
2776 LASSERTF(hash <= end,
2777 "ctxt@%p stripe@%p hash %llx end %llx\n",
2778 ctxt, stripe, hash, end);
2779 /* unload last page */
2780 stripe_dirent_unload(stripe);
2782 if (end == MDS_DIR_END_OFF) {
2783 stripe->sd_eof = true;
2789 oinfo = &op_data->op_lso1->lso_lsm.lsm_md_oinfo[stripe_index];
2790 if (!oinfo->lmo_root) {
2795 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2801 /* op_data is shared by stripes, reset after use */
2802 op_data->op_fid1 = oinfo->lmo_fid;
2803 op_data->op_fid2 = oinfo->lmo_fid;
2804 op_data->op_data = oinfo->lmo_root;
2806 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_mrinfo, hash,
2809 op_data->op_fid1 = fid;
2810 op_data->op_fid2 = fid;
2811 op_data->op_data = inode;
2816 stripe->sd_dp = page_address(stripe->sd_page);
2817 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2819 /* in case a page filled with ., .. and dummy, read next */
2822 stripe->sd_ent = ent;
2825 /* treat error as eof, so dir can be partially accessed */
2826 stripe->sd_eof = true;
2827 ctxt->ldc_mrinfo->mr_partial_readdir_rc = rc;
2828 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2829 "directory is partially accessed!\n",
2830 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2837 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2839 struct obd_device *obd = exp->exp_obd;
2840 struct lmv_obd *lmv = &obd->u.lmv;
2841 struct lmv_tgt_desc *tgt;
2846 rc = lmv_check_connect(obd);
2850 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2852 RETURN(PTR_ERR(tgt));
2854 data->op_flags |= MF_MDC_CANCEL_FID1;
2855 rc = md_file_resync(tgt->ltd_exp, data);
2860 * Get dirent with the closest hash for striped directory
2862 * This function will search the dir entry, whose hash value is the
2863 * closest(>=) to hash from all of sub-stripes, and it is only being called
2864 * for striped directory.
2866 * \param[in] ctxt dir read context
2868 * \retval dirent get the entry successfully
2869 * NULL does not get the entry, normally it means
2870 * it reaches the end of the directory, while read
2871 * stripe dirent error is ignored to allow partial
2874 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2876 struct stripe_dirent *stripe;
2877 struct lu_dirent *ent = NULL;
2881 /* TODO: optimize with k-way merge sort */
2882 for (i = 0; i < ctxt->ldc_count; i++) {
2883 stripe = &ctxt->ldc_stripes[i];
2887 if (!stripe->sd_ent) {
2888 stripe_dirent_load(ctxt, stripe, i);
2889 if (!stripe->sd_ent) {
2890 LASSERT(stripe->sd_eof);
2896 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2897 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2899 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2906 stripe = &ctxt->ldc_stripes[min];
2907 ent = stripe->sd_ent;
2908 /* pop found dirent */
2909 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2917 * Build dir entry page for striped directory
2919 * This function gets one entry by @offset from a striped directory. It will
2920 * read entries from all of stripes, and choose one closest to the required
2921 * offset(&offset). A few notes
2922 * 1. skip . and .. for non-zero stripes, because there can only have one .
2923 * and .. in a directory.
2924 * 2. op_data will be shared by all of stripes, instead of allocating new
2925 * one, so need to restore before reusing.
2927 * \param[in] exp obd export refer to LMV
2928 * \param[in] op_data hold those MD parameters of read_entry
2929 * \param[in] mrinfo ldlm callback being used in enqueue in mdc_read_entry,
2930 * and partial readdir result will be stored in it.
2931 * \param[in] offset starting hash offset
2932 * \param[out] ppage the page holding the entry. Note: because the entry
2933 * will be accessed in upper layer, so we need hold the
2934 * page until the usages of entry is finished, see
2935 * ll_dir_entry_next.
2937 * retval =0 if get entry successfully
2938 * <0 cannot get entry
2940 static int lmv_striped_read_page(struct obd_export *exp,
2941 struct md_op_data *op_data,
2942 struct md_readdir_info *mrinfo, __u64 offset,
2943 struct page **ppage)
2945 struct page *page = NULL;
2946 struct lu_dirpage *dp;
2948 struct lu_dirent *ent;
2949 struct lu_dirent *last_ent;
2951 struct lmv_dir_ctxt *ctxt;
2952 struct lu_dirent *next = NULL;
2958 /* Allocate a page and read entries from all of stripes and fill
2959 * the page by hash order */
2960 page = alloc_page(GFP_KERNEL);
2964 /* Initialize the entry page */
2966 memset(dp, 0, sizeof(*dp));
2967 dp->ldp_hash_start = cpu_to_le64(offset);
2970 left_bytes = PAGE_SIZE - sizeof(*dp);
2974 /* initalize dir read context */
2975 stripe_count = op_data->op_lso1->lso_lsm.lsm_md_stripe_count;
2976 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2978 GOTO(free_page, rc = -ENOMEM);
2979 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2980 ctxt->ldc_op_data = op_data;
2981 ctxt->ldc_mrinfo = mrinfo;
2982 ctxt->ldc_hash = offset;
2983 ctxt->ldc_count = stripe_count;
2986 next = lmv_dirent_next(ctxt);
2988 /* end of directory */
2990 ctxt->ldc_hash = MDS_DIR_END_OFF;
2993 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2995 ent_size = le16_to_cpu(next->lde_reclen);
2997 /* the last entry lde_reclen is 0, but it might not be the last
2998 * one of this temporay dir page */
3000 ent_size = lu_dirent_calc_size(
3001 le16_to_cpu(next->lde_namelen),
3002 le32_to_cpu(next->lde_attrs));
3004 if (ent_size > left_bytes)
3007 memcpy(ent, next, ent_size);
3009 /* Replace . with master FID and Replace .. with the parent FID
3010 * of master object */
3011 if (strncmp(ent->lde_name, ".",
3012 le16_to_cpu(ent->lde_namelen)) == 0 &&
3013 le16_to_cpu(ent->lde_namelen) == 1)
3014 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
3015 else if (strncmp(ent->lde_name, "..",
3016 le16_to_cpu(ent->lde_namelen)) == 0 &&
3017 le16_to_cpu(ent->lde_namelen) == 2)
3018 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
3020 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
3021 le16_to_cpu(ent->lde_namelen), ent->lde_name,
3022 le64_to_cpu(ent->lde_hash));
3024 left_bytes -= ent_size;
3025 ent->lde_reclen = cpu_to_le16(ent_size);
3027 ent = (void *)ent + ent_size;
3030 last_ent->lde_reclen = 0;
3033 dp->ldp_flags |= LDF_EMPTY;
3034 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
3035 dp->ldp_flags |= LDF_COLLIDE;
3036 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
3037 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
3039 put_lmv_dir_ctxt(ctxt);
3040 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
3053 static int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
3054 struct md_readdir_info *mrinfo, __u64 offset,
3055 struct page **ppage)
3057 struct obd_device *obd = exp->exp_obd;
3058 struct lmv_obd *lmv = &obd->u.lmv;
3059 struct lmv_tgt_desc *tgt;
3064 if (unlikely(lmv_dir_foreign(op_data->op_lso1)))
3067 if (unlikely(lmv_dir_striped(op_data->op_lso1))) {
3068 rc = lmv_striped_read_page(exp, op_data, mrinfo, offset, ppage);
3072 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
3074 RETURN(PTR_ERR(tgt));
3076 rc = md_read_page(tgt->ltd_exp, op_data, mrinfo, offset, ppage);
3082 * Unlink a file/directory
3084 * Unlink a file or directory under the parent dir. The unlink request
3085 * usually will be sent to the MDT where the child is located, but if
3086 * the client does not have the child FID then request will be sent to the
3087 * MDT where the parent is located.
3089 * If the parent is a striped directory then it also needs to locate which
3090 * stripe the name of the child is located, and replace the parent FID
3091 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
3092 * it will walk through all of sub-stripes until the child is being
3095 * \param[in] exp export refer to LMV
3096 * \param[in] op_data different parameters transferred beween client
3097 * MD stacks, name, namelen, FIDs etc.
3098 * op_fid1 is the parent FID, op_fid2 is the child
3100 * \param[out] request point to the request of unlink.
3102 * retval 0 if succeed
3103 * negative errno if failed.
3105 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
3106 struct ptlrpc_request **request)
3108 struct obd_device *obd = exp->exp_obd;
3109 struct lmv_obd *lmv = &obd->u.lmv;
3110 struct lmv_tgt_desc *tgt;
3111 struct lmv_tgt_desc *parent_tgt;
3112 struct mdt_body *body;
3117 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3118 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3119 op_data->op_cap = current_cap();
3122 parent_tgt = lmv_locate_tgt(lmv, op_data);
3123 if (IS_ERR(parent_tgt))
3124 RETURN(PTR_ERR(parent_tgt));
3126 if (likely(!fid_is_zero(&op_data->op_fid2))) {
3127 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3129 RETURN(PTR_ERR(tgt));
3135 * If child's fid is given, cancel unused locks for it if it is from
3136 * another export than parent.
3138 * LOOKUP lock for child (fid3) should also be cancelled on parent
3139 * tgt_tgt in mdc_unlink().
3141 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
3143 if (parent_tgt != tgt)
3144 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
3145 LCK_EX, MDS_INODELOCK_LOOKUP,
3146 MF_MDC_CANCEL_FID3);
3148 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
3149 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
3153 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
3154 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
3157 rc = md_unlink(tgt->ltd_exp, op_data, request);
3158 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
3159 ptlrpc_req_finished(*request);
3167 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
3171 /* Not cross-ref case, just get out of here. */
3172 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
3175 /* This is a remote object, try remote MDT. */
3176 op_data->op_fid2 = body->mbo_fid1;
3177 ptlrpc_req_finished(*request);
3180 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3182 RETURN(PTR_ERR(tgt));
3187 static int lmv_precleanup(struct obd_device *obd)
3190 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
3195 * Get by key a value associated with a LMV device.
3197 * Dispatch request to lower-layer devices as needed.
3199 * \param[in] env execution environment for this thread
3200 * \param[in] exp export for the LMV device
3201 * \param[in] keylen length of key identifier
3202 * \param[in] key identifier of key to get value for
3203 * \param[in] vallen size of \a val
3204 * \param[out] val pointer to storage location for value
3205 * \param[in] lsm optional striping metadata of object
3207 * \retval 0 on success
3208 * \retval negative negated errno on failure
3210 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
3211 __u32 keylen, void *key, __u32 *vallen, void *val)
3213 struct obd_device *obd;
3214 struct lmv_obd *lmv;
3215 struct lu_tgt_desc *tgt;
3220 obd = class_exp2obd(exp);
3222 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3223 exp->exp_handle.h_cookie);
3228 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
3229 LASSERT(*vallen == sizeof(__u32));
3230 lmv_foreach_connected_tgt(lmv, tgt) {
3231 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
3236 } else if (KEY_IS(KEY_MAX_EASIZE) ||
3237 KEY_IS(KEY_DEFAULT_EASIZE) ||
3238 KEY_IS(KEY_CONN_DATA)) {
3240 * Forwarding this request to first MDS, it should know LOV
3243 tgt = lmv_tgt(lmv, 0);
3247 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
3248 if (!rc && KEY_IS(KEY_CONN_DATA))
3249 exp->exp_connect_data = *(struct obd_connect_data *)val;
3251 } else if (KEY_IS(KEY_TGT_COUNT)) {
3252 *((int *)val) = lmv->lmv_mdt_descs.ltd_tgts_size;
3256 CDEBUG(D_IOCTL, "Invalid key\n");
3260 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
3261 int *__rcs, struct ptlrpc_request_set *_set)
3263 struct obd_device *obd = class_exp2obd(exp);
3264 struct ptlrpc_request_set *set = _set;
3265 struct lmv_obd *lmv = &obd->u.lmv;
3266 int tgt_count = lmv->lmv_mdt_count;
3267 struct lu_tgt_desc *tgt;
3268 struct fid_array *fat, **fas = NULL;
3269 int i, rc, **rcs = NULL;
3272 set = ptlrpc_prep_set();
3277 /* split FIDs by targets */
3278 OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
3280 GOTO(out, rc = -ENOMEM);
3281 OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
3283 GOTO(out_fas, rc = -ENOMEM);
3285 for (i = 0; i < fa->fa_nr; i++) {
3288 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
3290 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
3291 PFID(&fa->fa_fids[i]), rc);
3294 LASSERT(idx < tgt_count);
3296 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
3297 fa_fids[fa->fa_nr]));
3299 GOTO(out, rc = -ENOMEM);
3301 OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
3303 GOTO(out, rc = -ENOMEM);
3306 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3309 lmv_foreach_connected_tgt(lmv, tgt) {
3310 fat = fas[tgt->ltd_index];
3311 if (!fat || fat->fa_nr == 0)
3313 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3316 rc = ptlrpc_set_wait(NULL, set);
3319 for (i = 0; i < tgt_count; i++) {
3321 if (!fat || fat->fa_nr == 0)
3323 /* copy FIDs back */
3324 memcpy(fa->fa_fids + j, fat->fa_fids,
3325 fat->fa_nr * sizeof(struct lu_fid));
3327 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3332 ptlrpc_set_destroy(set);
3335 for (i = 0; i < tgt_count; i++) {
3337 OBD_FREE(fas[i], offsetof(struct fid_array,
3338 fa_fids[fa->fa_nr]));
3340 OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3343 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3346 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3352 * Asynchronously set by key a value associated with a LMV device.
3354 * Dispatch request to lower-layer devices as needed.
3356 * \param[in] env execution environment for this thread
3357 * \param[in] exp export for the LMV device
3358 * \param[in] keylen length of key identifier
3359 * \param[in] key identifier of key to store value for
3360 * \param[in] vallen size of value to store
3361 * \param[in] val pointer to data to be stored
3362 * \param[in] set optional list of related ptlrpc requests
3364 * \retval 0 on success
3365 * \retval negative negated errno on failure
3367 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3368 __u32 keylen, void *key, __u32 vallen, void *val,
3369 struct ptlrpc_request_set *set)
3371 struct lmv_tgt_desc *tgt;
3372 struct obd_device *obd;
3373 struct lmv_obd *lmv;
3377 obd = class_exp2obd(exp);
3379 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3380 exp->exp_handle.h_cookie);
3385 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3386 KEY_IS(KEY_DEFAULT_EASIZE)) {
3389 lmv_foreach_connected_tgt(lmv, tgt) {
3390 err = obd_set_info_async(env, tgt->ltd_exp,
3391 keylen, key, vallen, val, set);
3402 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3403 const struct lmv_mds_md_v1 *lmm1)
3405 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3412 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3413 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3414 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3415 if (CFS_FAIL_CHECK(OBD_FAIL_LMV_UNKNOWN_STRIPE))
3416 lsm->lsm_md_hash_type = cfs_fail_val ?: LMV_HASH_TYPE_UNKNOWN;
3418 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3419 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3420 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3421 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3422 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3423 sizeof(lsm->lsm_md_pool_name));
3425 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3428 CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3429 "layout_version %d\n", lsm->lsm_md_stripe_count,
3430 lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3431 lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3432 lsm->lsm_md_layout_version);
3434 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3435 for (i = 0; i < stripe_count; i++) {
3436 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3437 &lmm1->lmv_stripe_fids[i]);
3439 * set default value -1, so lmv_locate_tgt() knows this stripe
3440 * target is not initialized.
3442 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3443 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3446 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3447 &lsm->lsm_md_oinfo[i].lmo_mds);
3454 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3455 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3461 static inline int lmv_unpack_user_md(struct obd_export *exp,
3462 struct lmv_stripe_md *lsm,
3463 const struct lmv_user_md *lmu)
3465 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3466 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3467 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3468 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3469 lsm->lsm_md_max_inherit = lmu->lum_max_inherit;
3470 lsm->lsm_md_max_inherit_rr = lmu->lum_max_inherit_rr;
3471 lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3476 struct lmv_stripe_object *lmv_stripe_object_alloc(__u32 magic,
3477 const union lmv_mds_md *lmm,
3480 struct lmv_stripe_object *lsm_obj;
3483 if (magic == LMV_MAGIC_FOREIGN) {
3484 struct lmv_foreign_md *lfm;
3486 size = offsetof(typeof(*lfm), lfm_value[0]);
3487 if (lmm_size < size)
3488 RETURN(ERR_PTR(-EPROTO));
3490 size += le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3491 if (lmm_size < size)
3492 RETURN(ERR_PTR(-EPROTO));
3494 OBD_ALLOC_LARGE(lsm_obj, lmm_size +
3495 offsetof(typeof(*lsm_obj), lso_lfm));
3497 if (magic == LMV_MAGIC_V1) {
3500 size = offsetof(struct lmv_mds_md_v1,
3501 lmv_stripe_fids[0]);
3502 if (lmm_size < size)
3503 RETURN(ERR_PTR(-EPROTO));
3505 count = lmv_mds_md_stripe_count_get(lmm);
3506 size += count * sizeof(struct lu_fid);
3507 if (lmm_size < size)
3508 RETURN(ERR_PTR(-EPROTO));
3510 size = lmv_stripe_md_size(count);
3512 if (lmm && lmm_size < sizeof(struct lmv_user_md))
3513 RETURN(ERR_PTR(-EPROTO));
3516 * Unpack default dirstripe(lmv_user_md) to
3517 * lmv_stripe_md, stripecount should be 0 then.
3519 size = lmv_stripe_md_size(0);
3521 size += offsetof(typeof(*lsm_obj), lso_lsm);
3522 OBD_ALLOC(lsm_obj, size);
3526 atomic_set(&lsm_obj->lso_refs, 1);
3530 RETURN(ERR_PTR(-ENOMEM));
3532 EXPORT_SYMBOL(lmv_stripe_object_alloc);
3534 static int lmv_stripe_object_create(struct obd_export *exp,
3535 struct lmv_stripe_object **lsop,
3536 const union lmv_mds_md *lmm,
3539 struct lmv_stripe_object *lsm_obj;
3544 LASSERT(lsop != NULL && *lsop == NULL);
3549 magic = le32_to_cpu(lmm->lmv_magic);
3550 if (magic == LMV_MAGIC_STRIPE)
3553 if (magic != LMV_MAGIC_V1 && magic != LMV_USER_MAGIC &&
3554 magic != LMV_MAGIC_FOREIGN) {
3555 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3556 exp->exp_obd->obd_name, magic, -EIO);
3560 /* foreign lmv case */
3561 if (magic == LMV_MAGIC_FOREIGN) {
3562 struct lmv_foreign_md *lfm;
3564 lsm_obj = lmv_stripe_object_alloc(magic, lmm, lmm_size);
3565 if (IS_ERR(lsm_obj))
3566 RETURN(PTR_ERR(lsm_obj));
3569 lfm = &lsm_obj->lso_lfm;
3570 lfm->lfm_magic = magic;
3571 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3572 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3573 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3574 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3580 lsm_obj = lmv_stripe_object_alloc(magic, lmm, lmm_size);
3581 if (IS_ERR(lsm_obj))
3582 RETURN(PTR_ERR(lsm_obj));
3586 rc = lmv_unpack_md_v1(exp, &lsm_obj->lso_lsm, &lmm->lmv_md_v1);
3588 case LMV_USER_MAGIC:
3589 rc = lmv_unpack_user_md(exp, &lsm_obj->lso_lsm,
3593 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3600 lmv_stripe_object_put(&lsm_obj);
3606 struct lmv_stripe_object *
3607 lmv_stripe_object_get(struct lmv_stripe_object *lsm_obj)
3609 if (lsm_obj == NULL)
3612 atomic_inc(&lsm_obj->lso_refs);
3613 CDEBUG(D_INODE, "get %p %u\n", lsm_obj,
3614 atomic_read(&lsm_obj->lso_refs));
3617 EXPORT_SYMBOL(lmv_stripe_object_get);
3619 void lmv_stripe_object_put(struct lmv_stripe_object **lsop)
3621 struct lmv_stripe_object *lsm_obj;
3625 LASSERT(lsop != NULL);
3628 if (lsm_obj == NULL)
3632 CDEBUG(D_INODE, "put %p %u\n", lsm_obj,
3633 atomic_read(&lsm_obj->lso_refs) - 1);
3635 if (!atomic_dec_and_test(&lsm_obj->lso_refs))
3638 if (lmv_dir_foreign(lsm_obj)) {
3639 size = lsm_obj->lso_lfm.lfm_length +
3640 offsetof(typeof(lsm_obj->lso_lfm), lfm_value[0]) +
3641 offsetof(typeof(*lsm_obj), lso_lsm);
3642 OBD_FREE_LARGE(lsm_obj, size);
3646 if (lmv_dir_striped(lsm_obj)) {
3647 struct lmv_stripe_md *lsm = &lsm_obj->lso_lsm;
3649 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3650 iput(lsm->lsm_md_oinfo[i].lmo_root);
3651 size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3653 size = lmv_stripe_md_size(0);
3655 OBD_FREE(lsm_obj, size + offsetof(typeof(*lsm_obj), lso_lsm));
3657 EXPORT_SYMBOL(lmv_stripe_object_put);
3659 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3660 union ldlm_policy_data *policy,
3661 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3664 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3665 struct lu_tgt_desc *tgt;
3671 LASSERT(fid != NULL);
3673 lmv_foreach_connected_tgt(lmv, tgt) {
3674 if (!tgt->ltd_active)
3677 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3685 static int lmv_set_lock_data(struct obd_export *exp,
3686 const struct lustre_handle *lockh,
3687 void *data, __u64 *bits)
3689 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3690 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3695 if (tgt == NULL || tgt->ltd_exp == NULL)
3697 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3701 static enum ldlm_mode
3702 lmv_lock_match(struct obd_export *exp, __u64 flags,
3703 const struct lu_fid *fid, enum ldlm_type type,
3704 union ldlm_policy_data *policy,
3705 enum ldlm_mode mode, struct lustre_handle *lockh)
3707 struct obd_device *obd = exp->exp_obd;
3708 struct lmv_obd *lmv = &obd->u.lmv;
3709 struct lu_tgt_desc *tgt;
3710 __u64 bits = policy->l_inodebits.bits;
3711 enum ldlm_mode rc = LCK_MINMODE;
3715 /* only one bit is set */
3716 LASSERT(bits && !(bits & (bits - 1)));
3717 /* With DNE every object can have two locks in different namespaces:
3718 * lookup lock in space of MDT storing direntry and update/open lock in
3719 * space of MDT storing inode. Try the MDT that the FID maps to first,
3720 * since this can be easily found, and only try others if that fails.
3722 if (bits == MDS_INODELOCK_LOOKUP) {
3723 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3724 i < lmv->lmv_mdt_descs.ltd_tgts_size; i++,
3725 index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3728 "%s: "DFID" is inaccessible: rc = %d\n",
3729 obd->obd_name, PFID(fid), index);
3732 tgt = lmv_tgt(lmv, index);
3733 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3735 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3736 policy, mode, lockh);
3741 tgt = lmv_fid2tgt(lmv, fid);
3742 if (!IS_ERR(tgt) && tgt->ltd_exp && tgt->ltd_active)
3743 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3744 policy, mode, lockh);
3747 CDEBUG(D_INODE, "Lock match for "DFID": %d\n", PFID(fid), rc);
3753 lmv_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
3754 struct obd_export *dt_exp, struct obd_export *md_exp,
3755 struct lustre_md *md)
3757 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3758 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3760 if (!tgt || !tgt->ltd_exp)
3763 return md_get_lustre_md(tgt->ltd_exp, pill, dt_exp, md_exp, md);
3766 static int lmv_put_lustre_md(struct obd_export *exp, struct lustre_md *md)
3768 struct obd_device *obd = exp->exp_obd;
3769 struct lmv_obd *lmv = &obd->u.lmv;
3770 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3773 lmv_stripe_object_put(&md->def_lsm_obj);
3774 lmv_stripe_object_put(&md->lsm_obj);
3776 if (!tgt || !tgt->ltd_exp)
3781 static int lmv_set_open_replay_data(struct obd_export *exp,
3782 struct obd_client_handle *och,
3783 struct lookup_intent *it)
3785 struct obd_device *obd = exp->exp_obd;
3786 struct lmv_obd *lmv = &obd->u.lmv;
3787 struct lmv_tgt_desc *tgt;
3791 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3793 RETURN(PTR_ERR(tgt));
3795 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3798 static int lmv_clear_open_replay_data(struct obd_export *exp,
3799 struct obd_client_handle *och)
3801 struct obd_device *obd = exp->exp_obd;
3802 struct lmv_obd *lmv = &obd->u.lmv;
3803 struct lmv_tgt_desc *tgt;
3807 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3809 RETURN(PTR_ERR(tgt));
3811 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3814 static int lmv_intent_getattr_async(struct obd_export *exp,
3815 struct md_op_item *item)
3817 struct md_op_data *op_data = &item->mop_data;
3818 struct obd_device *obd = exp->exp_obd;
3819 struct lmv_obd *lmv = &obd->u.lmv;
3820 struct lmv_tgt_desc *ptgt;
3821 struct lmv_tgt_desc *ctgt;
3826 if (!(fid_is_sane(&op_data->op_fid2) ||
3827 fid_is_zero(&op_data->op_fid2)))
3830 ptgt = lmv_locate_tgt(lmv, op_data);
3832 RETURN(PTR_ERR(ptgt));
3835 * Zeroed FID @op_fid2 means that the intent getattr() comes from
3836 * statahead by regularized file names. Currently only do statahead
3837 * for the children files located same as the parent directory.
3839 if (!fid_is_zero(&op_data->op_fid2)) {
3840 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3842 RETURN(PTR_ERR(ctgt));
3845 * remote object needs two RPCs to lookup and getattr,
3846 * considering the complexity don't support statahead for now.
3852 rc = md_intent_getattr_async(ptgt->ltd_exp, item);
3857 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3858 struct lu_fid *fid, __u64 *bits)
3860 struct obd_device *obd = exp->exp_obd;
3861 struct lmv_obd *lmv = &obd->u.lmv;
3862 struct lmv_tgt_desc *tgt;
3867 tgt = lmv_fid2tgt(lmv, fid);
3869 RETURN(PTR_ERR(tgt));
3871 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3875 static int lmv_get_fid_from_lsm(struct obd_export *exp,
3876 const struct lmv_stripe_object *lso,
3877 const char *name, int namelen,
3880 const struct lmv_oinfo *oinfo;
3882 LASSERT(lmv_dir_striped(lso));
3883 oinfo = lsm_name_to_stripe_info(lso, name, namelen, false);
3885 return PTR_ERR(oinfo);
3887 *fid = oinfo->lmo_fid;
3893 * For lmv, only need to send request to master MDT, and the master MDT will
3894 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3895 * we directly fetch data from the slave MDTs.
3897 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3898 struct obd_quotactl *oqctl)
3900 struct obd_device *obd = class_exp2obd(exp);
3901 struct lmv_obd *lmv = &obd->u.lmv;
3902 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3903 __u64 curspace, curinodes;
3908 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3909 CERROR("master lmv inactive\n");
3913 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3914 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3918 curspace = curinodes = 0;
3919 lmv_foreach_connected_tgt(lmv, tgt) {
3922 if (!tgt->ltd_active)
3925 err = obd_quotactl(tgt->ltd_exp, oqctl);
3927 CERROR("getquota on mdt %d failed. %d\n",
3928 tgt->ltd_index, err);
3932 curspace += oqctl->qc_dqblk.dqb_curspace;
3933 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3936 oqctl->qc_dqblk.dqb_curspace = curspace;
3937 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3942 static int lmv_merge_attr(struct obd_export *exp,
3943 const struct lmv_stripe_object *lso,
3944 struct cl_attr *attr,
3945 ldlm_blocking_callback cb_blocking)
3947 const struct lmv_stripe_md *lsm = &lso->lso_lsm;
3951 if (!lmv_dir_striped(lso))
3954 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3958 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3959 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3965 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3966 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3967 i_size_read(inode), (unsigned long long)inode->i_blocks,
3968 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3969 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3971 /* for slave stripe, it needs to subtract nlink for . and .. */
3973 attr->cat_nlink += inode->i_nlink - 2;
3975 attr->cat_nlink = inode->i_nlink;
3977 attr->cat_size += i_size_read(inode);
3978 attr->cat_blocks += inode->i_blocks;
3980 if (attr->cat_atime < inode->i_atime.tv_sec)
3981 attr->cat_atime = inode->i_atime.tv_sec;
3983 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3984 attr->cat_ctime = inode->i_ctime.tv_sec;
3986 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3987 attr->cat_mtime = inode->i_mtime.tv_sec;
3992 static struct lu_batch *lmv_batch_create(struct obd_export *exp,
3993 enum lu_batch_flags flags,
3996 struct lu_batch *bh;
3997 struct lmv_batch *lbh;
4002 RETURN(ERR_PTR(-ENOMEM));
4004 bh = &lbh->lbh_super;
4005 bh->lbt_flags = flags;
4006 bh->lbt_max_count = max_count;
4008 if (flags & BATCH_FL_RQSET) {
4009 bh->lbt_rqset = ptlrpc_prep_set();
4010 if (bh->lbt_rqset == NULL) {
4012 RETURN(ERR_PTR(-ENOMEM));
4016 INIT_LIST_HEAD(&lbh->lbh_sub_batch_list);
4020 static int lmv_batch_stop(struct obd_export *exp, struct lu_batch *bh)
4022 struct lmv_batch *lbh;
4023 struct lmvsub_batch *sub;
4024 struct lmvsub_batch *tmp;
4029 lbh = container_of(bh, struct lmv_batch, lbh_super);
4030 list_for_each_entry_safe(sub, tmp, &lbh->lbh_sub_batch_list,
4032 list_del(&sub->sbh_sub_item);
4033 rc = md_batch_stop(sub->sbh_tgt->ltd_exp, sub->sbh_sub);
4035 CERROR("%s: stop batch processing failed: rc = %d\n",
4036 exp->exp_obd->obd_name, rc);
4037 if (bh->lbt_result == 0)
4038 bh->lbt_result = rc;
4043 if (bh->lbt_flags & BATCH_FL_RQSET) {
4044 rc = ptlrpc_set_wait(NULL, bh->lbt_rqset);
4045 ptlrpc_set_destroy(bh->lbt_rqset);
4052 static int lmv_batch_flush(struct obd_export *exp, struct lu_batch *bh,
4055 struct lmv_batch *lbh;
4056 struct lmvsub_batch *sub;
4062 lbh = container_of(bh, struct lmv_batch, lbh_super);
4063 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
4064 rc1 = md_batch_flush(sub->sbh_tgt->ltd_exp, sub->sbh_sub, wait);
4066 CERROR("%s: stop batch processing failed: rc = %d\n",
4067 exp->exp_obd->obd_name, rc);
4068 if (bh->lbt_result == 0)
4069 bh->lbt_result = rc;
4076 if (wait && bh->lbt_flags & BATCH_FL_RQSET) {
4077 rc1 = ptlrpc_set_wait(NULL, bh->lbt_rqset);
4085 static inline struct lmv_tgt_desc *
4086 lmv_batch_locate_tgt(struct lmv_obd *lmv, struct md_op_item *item)
4088 struct md_op_data *op_data = &item->mop_data;
4089 struct lmv_tgt_desc *tgt;
4091 switch (item->mop_opc) {
4092 case MD_OP_GETATTR: {
4093 struct lmv_tgt_desc *ptgt;
4095 if (!(fid_is_sane(&op_data->op_fid2) ||
4096 fid_is_zero(&op_data->op_fid2)))
4097 RETURN(ERR_PTR(-EINVAL));
4099 ptgt = lmv_locate_tgt(lmv, op_data);
4104 * Zeroed @op_fid2 means that it is a statahead populating call
4105 * in the file name pattern which is using file name format to
4106 * prefetch the attributes. Thus it has no idea about the FID of
4107 * the children file. The children file is considered to be
4108 * located on the same storage target with the parent directory
4109 * or the stripped directory.
4111 if (fid_is_zero(&op_data->op_fid2)) {
4116 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
4121 * Remote object needs two RPCs to lookup and getattr,
4122 * considering the complexity don't support statahead for now.
4125 RETURN(ERR_PTR(-EREMOTE));
4130 tgt = ERR_PTR(-ENOTSUPP);
4136 struct lu_batch *lmv_batch_lookup_sub(struct lmv_batch *lbh,
4137 struct lmv_tgt_desc *tgt)
4139 struct lmvsub_batch *sub;
4141 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
4142 if (sub->sbh_tgt == tgt)
4143 return sub->sbh_sub;
4149 struct lu_batch *lmv_batch_get_sub(struct lmv_batch *lbh,
4150 struct lmv_tgt_desc *tgt)
4152 struct lmvsub_batch *sbh;
4153 struct lu_batch *child_bh;
4154 struct lu_batch *bh;
4158 child_bh = lmv_batch_lookup_sub(lbh, tgt);
4159 if (child_bh != NULL)
4164 RETURN(ERR_PTR(-ENOMEM));
4166 INIT_LIST_HEAD(&sbh->sbh_sub_item);
4169 bh = &lbh->lbh_super;
4170 child_bh = md_batch_create(tgt->ltd_exp, bh->lbt_flags,
4172 if (IS_ERR(child_bh)) {
4177 child_bh->lbt_rqset = bh->lbt_rqset;
4178 sbh->sbh_sub = child_bh;
4179 list_add(&sbh->sbh_sub_item, &lbh->lbh_sub_batch_list);
4183 static int lmv_batch_add(struct obd_export *exp, struct lu_batch *bh,
4184 struct md_op_item *item)
4186 struct obd_device *obd = exp->exp_obd;
4187 struct lmv_obd *lmv = &obd->u.lmv;
4188 struct lmv_tgt_desc *tgt;
4189 struct lmv_batch *lbh;
4190 struct lu_batch *child_bh;
4195 tgt = lmv_batch_locate_tgt(lmv, item);
4197 RETURN(PTR_ERR(tgt));
4199 lbh = container_of(bh, struct lmv_batch, lbh_super);
4200 child_bh = lmv_batch_get_sub(lbh, tgt);
4201 if (IS_ERR(child_bh))
4202 RETURN(PTR_ERR(child_bh));
4204 rc = md_batch_add(tgt->ltd_exp, child_bh, item);
4208 static const struct obd_ops lmv_obd_ops = {
4209 .o_owner = THIS_MODULE,
4210 .o_setup = lmv_setup,
4211 .o_cleanup = lmv_cleanup,
4212 .o_precleanup = lmv_precleanup,
4213 .o_process_config = lmv_process_config,
4214 .o_connect = lmv_connect,
4215 .o_disconnect = lmv_disconnect,
4216 .o_statfs = lmv_statfs,
4217 .o_get_info = lmv_get_info,
4218 .o_set_info_async = lmv_set_info_async,
4219 .o_notify = lmv_notify,
4220 .o_get_uuid = lmv_get_uuid,
4221 .o_fid_alloc = lmv_fid_alloc,
4222 .o_iocontrol = lmv_iocontrol,
4223 .o_quotactl = lmv_quotactl
4226 static const struct md_ops lmv_md_ops = {
4227 .m_get_root = lmv_get_root,
4228 .m_null_inode = lmv_null_inode,
4229 .m_close = lmv_close,
4230 .m_create = lmv_create,
4231 .m_enqueue = lmv_enqueue,
4232 .m_getattr = lmv_getattr,
4233 .m_getxattr = lmv_getxattr,
4234 .m_getattr_name = lmv_getattr_name,
4235 .m_intent_lock = lmv_intent_lock,
4237 .m_rename = lmv_rename,
4238 .m_setattr = lmv_setattr,
4239 .m_setxattr = lmv_setxattr,
4240 .m_fsync = lmv_fsync,
4241 .m_file_resync = lmv_file_resync,
4242 .m_read_page = lmv_read_page,
4243 .m_unlink = lmv_unlink,
4244 .m_init_ea_size = lmv_init_ea_size,
4245 .m_cancel_unused = lmv_cancel_unused,
4246 .m_set_lock_data = lmv_set_lock_data,
4247 .m_lock_match = lmv_lock_match,
4248 .m_get_lustre_md = lmv_get_lustre_md,
4249 .m_put_lustre_md = lmv_put_lustre_md,
4250 .m_merge_attr = lmv_merge_attr,
4251 .m_set_open_replay_data = lmv_set_open_replay_data,
4252 .m_clear_open_replay_data = lmv_clear_open_replay_data,
4253 .m_intent_getattr_async = lmv_intent_getattr_async,
4254 .m_revalidate_lock = lmv_revalidate_lock,
4255 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
4256 .m_stripe_object_create = lmv_stripe_object_create,
4257 .m_rmfid = lmv_rmfid,
4258 .m_batch_create = lmv_batch_create,
4259 .m_batch_add = lmv_batch_add,
4260 .m_batch_stop = lmv_batch_stop,
4261 .m_batch_flush = lmv_batch_flush,
4264 static int __init lmv_init(void)
4266 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true,
4267 LUSTRE_LMV_NAME, NULL);
4270 static void __exit lmv_exit(void)
4272 class_unregister_type(LUSTRE_LMV_NAME);
4275 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4276 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
4277 MODULE_VERSION(LUSTRE_VERSION_STRING);
4278 MODULE_LICENSE("GPL");
4280 module_init(lmv_init);
4281 module_exit(lmv_exit);