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(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;
1563 tgt->ltd_qos.ltq_usable = 1;
1564 lu_tgt_qos_weight_calc(tgt, true);
1565 if (tgt->ltd_index == op_data->op_mds)
1567 total_avail += tgt->ltd_qos.ltq_avail;
1568 total_weight += tgt->ltd_qos.ltq_weight;
1572 /* If current MDT has above-average space and dir is not aleady using
1573 * round-robin to spread across more MDTs, stay on the parent MDT
1574 * to avoid creating needless remote MDT directories. Remote dirs
1575 * close to the root balance space more effectively than bottom dirs,
1576 * so prefer to create remote dirs at top level of directory tree.
1577 * "16 / (dir_depth + 10)" is the factor to make it less likely
1578 * for top-level directories to stay local unless they have more than
1579 * average free space, while deep dirs prefer local until more full.
1580 * depth=0 -> 160%, depth=3 -> 123%, depth=6 -> 100%,
1581 * depth=9 -> 84%, depth=12 -> 73%, depth=15 -> 64%
1583 if (!lmv_op_default_rr_mkdir(op_data)) {
1584 rand = total_avail * 16 /
1585 (total_usable * (op_data->op_dir_depth + 10));
1586 if (cur && cur->ltd_qos.ltq_avail >= rand) {
1592 rand = lu_prandom_u64_max(total_weight);
1594 lmv_foreach_connected_tgt(lmv, tgt) {
1595 if (!tgt->ltd_qos.ltq_usable)
1598 cur_weight += tgt->ltd_qos.ltq_weight;
1599 if (cur_weight < rand)
1602 ltd_qos_update(&lmv->lmv_mdt_descs, tgt, &total_weight);
1606 /* no proper target found */
1607 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1609 up_write(&lmv->lmv_qos.lq_rw_sem);
1614 static struct lu_tgt_desc *lmv_locate_tgt_rr(struct lmv_obd *lmv)
1616 struct lu_tgt_desc *tgt;
1622 spin_lock(&lmv->lmv_lock);
1623 for (i = 0; i < lmv->lmv_mdt_descs.ltd_tgts_size; i++) {
1624 index = (i + lmv->lmv_qos_rr_index) %
1625 lmv->lmv_mdt_descs.ltd_tgts_size;
1626 tgt = lmv_tgt(lmv, index);
1627 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
1630 lmv->lmv_qos_rr_index = (tgt->ltd_index + 1) %
1631 lmv->lmv_mdt_descs.ltd_tgts_size;
1632 spin_unlock(&lmv->lmv_lock);
1636 spin_unlock(&lmv->lmv_lock);
1638 RETURN(ERR_PTR(-ENODEV));
1641 /* locate MDT which is less full (avoid the most full MDT) */
1642 static struct lu_tgt_desc *lmv_locate_tgt_lf(struct lmv_obd *lmv)
1644 struct lu_tgt_desc *min = NULL;
1645 struct lu_tgt_desc *tgt;
1651 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1652 RETURN(ERR_PTR(-EAGAIN));
1654 down_write(&lmv->lmv_qos.lq_rw_sem);
1656 if (!ltd_qos_is_usable(&lmv->lmv_mdt_descs))
1657 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1659 lmv_foreach_tgt(lmv, tgt) {
1660 if (!tgt->ltd_exp || !tgt->ltd_active) {
1661 tgt->ltd_qos.ltq_usable = 0;
1665 tgt->ltd_qos.ltq_usable = 1;
1666 lu_tgt_qos_weight_calc(tgt, true);
1667 avail += tgt->ltd_qos.ltq_avail;
1668 if (!min || min->ltd_qos.ltq_avail > tgt->ltd_qos.ltq_avail)
1672 /* avoid the most full MDT */
1674 avail -= min->ltd_qos.ltq_avail;
1676 rand = lu_prandom_u64_max(avail);
1678 lmv_foreach_connected_tgt(lmv, tgt) {
1679 if (!tgt->ltd_qos.ltq_usable)
1685 avail += tgt->ltd_qos.ltq_avail;
1692 /* no proper target found */
1693 GOTO(unlock, tgt = ERR_PTR(-EAGAIN));
1695 up_write(&lmv->lmv_qos.lq_rw_sem);
1700 /* locate MDT by file name, for striped directory, the file name hash decides
1701 * which stripe its dirent is stored.
1703 static struct lmv_tgt_desc *
1704 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_object *lso,
1705 const char *name, int namelen, struct lu_fid *fid,
1706 __u32 *mds, bool new_layout)
1708 struct lmv_tgt_desc *tgt;
1709 const struct lmv_oinfo *oinfo;
1711 if (!lmv_dir_striped(lso) || !namelen) {
1712 tgt = lmv_fid2tgt(lmv, fid);
1716 *mds = tgt->ltd_index;
1720 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1721 if (cfs_fail_val >= lso->lso_lsm.lsm_md_stripe_count)
1722 return ERR_PTR(-EBADF);
1723 oinfo = &lso->lso_lsm.lsm_md_oinfo[cfs_fail_val];
1725 oinfo = lsm_name_to_stripe_info(lso, name, namelen, new_layout);
1727 return ERR_CAST(oinfo);
1730 /* check stripe FID is sane */
1731 if (!fid_is_sane(&oinfo->lmo_fid))
1732 return ERR_PTR(-ENODEV);
1734 *fid = oinfo->lmo_fid;
1735 *mds = oinfo->lmo_mds;
1736 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1738 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1740 return tgt ? tgt : ERR_PTR(-ENODEV);
1744 * Locate MDT of op_data->op_fid1
1746 * For striped directory, it will locate the stripe by name hash, if hash_type
1747 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1748 * which is set outside, and if dir is migrating, 'op_data->op_new_layout'
1749 * indicates whether old or new layout is used to locate.
1751 * For plain direcotry, it just locate the MDT of op_data->op_fid1.
1753 * \param[in] lmv LMV device
1754 * \param[in/out] op_data client MD stack parameters, name, namelen etc,
1755 * op_mds and op_fid1 will be updated if op_lso1
1756 * indicates fid1 represents a striped directory.
1758 * retval pointer to the lmv_tgt_desc if succeed.
1759 * ERR_PTR(errno) if failed.
1761 struct lmv_tgt_desc *
1762 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1764 struct lmv_stripe_md *lsm;
1765 struct lmv_oinfo *oinfo;
1766 struct lmv_tgt_desc *tgt;
1768 if (lmv_dir_foreign(op_data->op_lso1))
1769 return ERR_PTR(-ENODATA);
1771 /* During creating VOLATILE file, it should honor the mdt
1772 * index if the file under striped dir is being restored, see
1774 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1775 op_data->op_mds != LMV_OFFSET_DEFAULT) {
1776 tgt = lmv_tgt(lmv, op_data->op_mds);
1778 return ERR_PTR(-ENODEV);
1780 if (lmv_dir_striped(op_data->op_lso1)) {
1783 /* refill the right parent fid */
1784 lsm = &op_data->op_lso1->lso_lsm;
1785 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1786 oinfo = &lsm->lsm_md_oinfo[i];
1787 if (oinfo->lmo_mds == op_data->op_mds) {
1788 op_data->op_fid1 = oinfo->lmo_fid;
1793 if (i == lsm->lsm_md_stripe_count)
1794 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1796 } else if (lmv_dir_bad_hash(op_data->op_lso1)) {
1797 lsm = &op_data->op_lso1->lso_lsm;
1799 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1800 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1802 op_data->op_fid1 = oinfo->lmo_fid;
1803 op_data->op_mds = oinfo->lmo_mds;
1804 tgt = lmv_tgt(lmv, oinfo->lmo_mds);
1806 return ERR_PTR(-ENODEV);
1808 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_lso1,
1809 op_data->op_name, op_data->op_namelen,
1810 &op_data->op_fid1, &op_data->op_mds,
1811 op_data->op_new_layout);
1817 /* Locate MDT of op_data->op_fid2 for link/rename */
1818 static struct lmv_tgt_desc *
1819 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1821 struct lmv_tgt_desc *tgt;
1824 LASSERT(op_data->op_name);
1825 if (lmv_dir_layout_changing(op_data->op_lso2)) {
1826 struct lu_fid fid1 = op_data->op_fid1;
1827 struct lmv_stripe_object *lso1 = op_data->op_lso1;
1828 struct ptlrpc_request *request = NULL;
1831 * avoid creating new file under old layout of migrating
1832 * directory, check it here.
1834 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_lso2,
1835 op_data->op_name, op_data->op_namelen,
1836 &op_data->op_fid2, &op_data->op_mds, false);
1840 op_data->op_fid1 = op_data->op_fid2;
1841 op_data->op_lso1 = op_data->op_lso2;
1842 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1843 op_data->op_fid1 = fid1;
1844 op_data->op_lso1 = lso1;
1846 ptlrpc_req_finished(request);
1847 RETURN(ERR_PTR(-EEXIST));
1851 RETURN(ERR_PTR(rc));
1854 return lmv_locate_tgt_by_name(lmv, op_data->op_lso2,
1855 op_data->op_name, op_data->op_namelen,
1856 &op_data->op_fid2, &op_data->op_mds,
1860 int lmv_old_layout_lookup(struct lmv_obd *lmv, struct md_op_data *op_data)
1862 struct lu_tgt_desc *tgt;
1863 struct ptlrpc_request *request;
1866 LASSERT(lmv_dir_layout_changing(op_data->op_lso1));
1867 LASSERT(!op_data->op_new_layout);
1869 tgt = lmv_locate_tgt(lmv, op_data);
1871 return PTR_ERR(tgt);
1873 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1875 ptlrpc_req_finished(request);
1882 /* mkdir by QoS upon 'lfs mkdir -i -1'.
1884 * NB, mkdir by QoS only if parent is not striped, this is to avoid remote
1885 * directories under striped directory.
1887 static inline bool lmv_op_user_qos_mkdir(const struct md_op_data *op_data)
1889 const struct lmv_user_md *lum = op_data->op_data;
1891 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1894 if (lmv_dir_striped(op_data->op_lso1))
1897 return (op_data->op_cli_flags & CLI_SET_MEA) && lum &&
1898 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC &&
1899 le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
1902 /* mkdir by QoS if either ROOT or parent default LMV is space balanced. */
1903 static inline bool lmv_op_default_qos_mkdir(const struct md_op_data *op_data)
1905 const struct lmv_stripe_object *lso = op_data->op_default_lso1;
1907 if (op_data->op_code != LUSTRE_OPC_MKDIR)
1910 if (lmv_dir_striped(op_data->op_lso1))
1913 return (op_data->op_flags & MF_QOS_MKDIR) ||
1914 (lso && lso->lso_lsm.lsm_md_master_mdt_index ==
1915 LMV_OFFSET_DEFAULT);
1918 /* if parent default LMV is space balanced, and
1919 * 1. max_inherit_rr is set
1920 * 2. or parent is ROOT
1921 * mkdir roundrobin. Or if parent doesn't have default LMV, while ROOT default
1922 * LMV requests roundrobin mkdir, do the same.
1923 * NB, this needs to check server is balanced, which is done by caller.
1925 static inline bool lmv_op_default_rr_mkdir(const struct md_op_data *op_data)
1927 const struct lmv_stripe_object *lso = op_data->op_default_lso1;
1929 return (op_data->op_flags & MF_RR_MKDIR) ||
1930 (lso && lso->lso_lsm.lsm_md_max_inherit_rr !=
1931 LMV_INHERIT_RR_NONE) || fid_is_root(&op_data->op_fid1);
1934 /* 'lfs mkdir -i <specific_MDT>' */
1935 static inline bool lmv_op_user_specific_mkdir(const struct md_op_data *op_data)
1937 const struct lmv_user_md *lum = op_data->op_data;
1939 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1940 op_data->op_cli_flags & CLI_SET_MEA && lum &&
1941 (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC ||
1942 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) &&
1943 le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
1946 /* parent default LMV master_mdt_index is not -1. */
1948 lmv_op_default_specific_mkdir(const struct md_op_data *op_data)
1950 return op_data->op_code == LUSTRE_OPC_MKDIR &&
1951 op_data->op_default_lso1 &&
1952 op_data->op_default_lso1->lso_lsm.lsm_md_master_mdt_index !=
1956 /* locate MDT by space usage */
1957 static struct lu_tgt_desc *lmv_locate_tgt_by_space(struct lmv_obd *lmv,
1958 struct md_op_data *op_data,
1959 struct lmv_tgt_desc *tgt)
1961 struct lmv_tgt_desc *tmp = tgt;
1963 tgt = lmv_locate_tgt_qos(lmv, op_data);
1964 if (tgt == ERR_PTR(-EAGAIN)) {
1965 if (ltd_qos_is_balanced(&lmv->lmv_mdt_descs) &&
1966 !lmv_op_default_rr_mkdir(op_data) &&
1967 !lmv_op_user_qos_mkdir(op_data))
1968 /* if not necessary, don't create remote directory. */
1971 tgt = lmv_locate_tgt_rr(lmv);
1975 * only update statfs after QoS mkdir, this means the cached statfs may
1976 * be stale, and current mkdir may not follow QoS accurately, but it's
1977 * not serious, and avoids periodic statfs when client doesn't mkdir by
1981 op_data->op_mds = tgt->ltd_index;
1982 lmv_statfs_check_update(lmv2obd_dev(lmv), tgt);
1988 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1989 const void *data, size_t datalen, umode_t mode, uid_t uid,
1990 gid_t gid, kernel_cap_t cap_effective, __u64 rdev,
1991 struct ptlrpc_request **request)
1993 struct obd_device *obd = exp->exp_obd;
1994 struct lmv_obd *lmv = &obd->u.lmv;
1995 struct lmv_tgt_desc *tgt;
1996 struct mdt_body *repbody;
2001 if (!lmv->lmv_mdt_descs.ltd_lmv_desc.ld_active_tgt_count)
2004 if (lmv_dir_bad_hash(op_data->op_lso1))
2007 if (lmv_dir_layout_changing(op_data->op_lso1)) {
2009 * if parent is migrating, create() needs to lookup existing
2010 * name in both old and new layout, check old layout on client.
2012 rc = lmv_old_layout_lookup(lmv, op_data);
2016 op_data->op_new_layout = true;
2019 tgt = lmv_locate_tgt(lmv, op_data);
2021 RETURN(PTR_ERR(tgt));
2023 /* the order to apply policy in mkdir:
2024 * 1. is "lfs mkdir -i N"? mkdir on MDT N.
2025 * 2. is "lfs mkdir -i -1"? mkdir by space usage.
2026 * 3. is starting MDT specified in default LMV? mkdir on MDT N.
2027 * 4. is default LMV space balanced? mkdir by space usage.
2029 if (lmv_op_user_specific_mkdir(op_data)) {
2030 struct lmv_user_md *lum = op_data->op_data;
2032 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2033 tgt = lmv_tgt(lmv, op_data->op_mds);
2036 } else if (lmv_op_user_qos_mkdir(op_data)) {
2037 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2039 RETURN(PTR_ERR(tgt));
2040 } else if (lmv_op_default_specific_mkdir(op_data)) {
2041 struct lmv_stripe_md *lsm = &op_data->op_default_lso1->lso_lsm;
2043 op_data->op_mds = lsm->lsm_md_master_mdt_index;
2044 tgt = lmv_tgt(lmv, op_data->op_mds);
2047 } else if (lmv_op_default_qos_mkdir(op_data)) {
2048 tgt = lmv_locate_tgt_by_space(lmv, op_data, tgt);
2050 RETURN(PTR_ERR(tgt));
2054 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2058 CDEBUG(D_INODE, "CREATE name '%.*s' "DFID" on "DFID" -> mds #%x\n",
2059 (int)op_data->op_namelen, op_data->op_name,
2060 PFID(&op_data->op_fid2), PFID(&op_data->op_fid1),
2063 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2064 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
2065 cap_effective, rdev, request);
2067 if (*request == NULL)
2069 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
2072 /* dir restripe needs to send to MDT where dir is located */
2073 if (rc != -EREMOTE ||
2074 !(exp_connect_flags2(exp) & OBD_CONNECT2_CRUSH))
2077 repbody = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2078 if (repbody == NULL)
2081 /* Not cross-ref case, just get out of here. */
2082 if (likely(!(repbody->mbo_valid & OBD_MD_MDS)))
2085 op_data->op_fid2 = repbody->mbo_fid1;
2086 ptlrpc_req_finished(*request);
2089 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
2091 RETURN(PTR_ERR(tgt));
2093 op_data->op_mds = tgt->ltd_index;
2098 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
2099 const union ldlm_policy_data *policy, struct md_op_data *op_data,
2100 struct lustre_handle *lockh, __u64 extra_lock_flags)
2102 struct obd_device *obd = exp->exp_obd;
2103 struct lmv_obd *lmv = &obd->u.lmv;
2104 struct lmv_tgt_desc *tgt;
2109 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
2111 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2113 RETURN(PTR_ERR(tgt));
2115 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
2116 PFID(&op_data->op_fid1), tgt->ltd_index);
2118 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
2125 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
2126 struct ptlrpc_request **preq)
2128 struct obd_device *obd = exp->exp_obd;
2129 struct lmv_obd *lmv = &obd->u.lmv;
2130 struct lmv_tgt_desc *tgt;
2131 struct mdt_body *body;
2137 if (op_data->op_namelen == 2 &&
2138 op_data->op_name[0] == '.' && op_data->op_name[1] == '.')
2139 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2141 tgt = lmv_locate_tgt(lmv, op_data);
2143 RETURN(PTR_ERR(tgt));
2145 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
2146 (int)op_data->op_namelen, op_data->op_name,
2147 PFID(&op_data->op_fid1), tgt->ltd_index);
2149 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
2150 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2151 ptlrpc_req_finished(*preq);
2159 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
2160 LASSERT(body != NULL);
2162 if (body->mbo_valid & OBD_MD_MDS) {
2163 op_data->op_fid1 = body->mbo_fid1;
2164 op_data->op_valid |= OBD_MD_FLCROSSREF;
2165 op_data->op_namelen = 0;
2166 op_data->op_name = NULL;
2168 ptlrpc_req_finished(*preq);
2177 #define md_op_data_fid(op_data, fl) \
2178 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
2179 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
2180 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
2181 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
2184 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
2185 struct md_op_data *op_data, __u32 op_tgt,
2186 enum ldlm_mode mode, int bits, int flag)
2188 struct lu_fid *fid = md_op_data_fid(op_data, flag);
2189 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2190 union ldlm_policy_data policy = { { 0 } };
2194 if (!fid_is_sane(fid))
2198 tgt = lmv_fid2tgt(lmv, fid);
2200 RETURN(PTR_ERR(tgt));
2203 if (tgt->ltd_index != op_tgt) {
2204 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
2205 policy.l_inodebits.bits = bits;
2206 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
2207 mode, LCF_ASYNC, NULL);
2210 "EARLY_CANCEL skip operation target %d on "DFID"\n",
2212 op_data->op_flags |= flag;
2220 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
2223 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
2224 struct ptlrpc_request **request)
2226 struct obd_device *obd = exp->exp_obd;
2227 struct lmv_obd *lmv = &obd->u.lmv;
2228 struct lmv_tgt_desc *tgt;
2232 LASSERT(op_data->op_namelen != 0);
2234 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2235 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2236 op_data->op_name, PFID(&op_data->op_fid1));
2238 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2239 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2240 op_data->op_cap = current_cap();
2242 tgt = lmv_locate_tgt2(lmv, op_data);
2244 RETURN(PTR_ERR(tgt));
2247 * Cancel UPDATE lock on child (fid1).
2249 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2250 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
2251 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2255 rc = md_link(tgt->ltd_exp, op_data, request);
2260 /* migrate the top directory */
2261 static inline bool lmv_op_topdir_migrate(const struct md_op_data *op_data)
2263 if (!S_ISDIR(op_data->op_mode))
2266 if (lmv_dir_layout_changing(op_data->op_lso1))
2272 /* migrate top dir to specific MDTs */
2273 static inline bool lmv_topdir_specific_migrate(const struct md_op_data *op_data)
2275 const struct lmv_user_md *lum = op_data->op_data;
2277 if (!lmv_op_topdir_migrate(op_data))
2280 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2283 /* migrate top dir in QoS mode if user issued "lfs migrate -m -1..." */
2284 static inline bool lmv_topdir_qos_migrate(const struct md_op_data *op_data)
2286 const struct lmv_user_md *lum = op_data->op_data;
2288 if (!lmv_op_topdir_migrate(op_data))
2291 return le32_to_cpu(lum->lum_stripe_offset) == LMV_OFFSET_DEFAULT;
2294 static inline bool lmv_subdir_specific_migrate(const struct md_op_data *op_data)
2296 const struct lmv_user_md *lum = op_data->op_data;
2298 if (!S_ISDIR(op_data->op_mode))
2301 if (!lmv_dir_layout_changing(op_data->op_lso1))
2304 return le32_to_cpu(lum->lum_stripe_offset) != LMV_OFFSET_DEFAULT;
2307 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2308 const char *name, size_t namelen,
2309 struct ptlrpc_request **request)
2311 struct obd_device *obd = exp->exp_obd;
2312 struct lmv_obd *lmv = &obd->u.lmv;
2313 struct lmv_stripe_object *lso = op_data->op_lso1;
2314 struct lmv_tgt_desc *parent_tgt;
2315 struct lmv_tgt_desc *sp_tgt;
2316 struct lmv_tgt_desc *tp_tgt = NULL;
2317 struct lmv_tgt_desc *child_tgt;
2318 struct lmv_tgt_desc *tgt;
2319 struct lu_fid target_fid = { 0 };
2324 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2326 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2327 PFID(&op_data->op_fid1), (int)namelen, name);
2329 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2330 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2331 op_data->op_cap = current_cap();
2333 parent_tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2334 if (IS_ERR(parent_tgt))
2335 RETURN(PTR_ERR(parent_tgt));
2337 if (lmv_dir_striped(lso)) {
2338 const struct lmv_oinfo *oinfo;
2340 oinfo = lsm_name_to_stripe_info(lso, name, namelen, false);
2342 RETURN(PTR_ERR(oinfo));
2344 /* save source stripe FID in fid4 temporarily for ELC */
2345 op_data->op_fid4 = oinfo->lmo_fid;
2346 sp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2351 * if parent is being migrated too, fill op_fid2 with target
2352 * stripe fid, otherwise the target stripe is not created yet.
2354 if (lmv_dir_layout_changing(lso)) {
2355 oinfo = lsm_name_to_stripe_info(lso, name, namelen,
2358 RETURN(PTR_ERR(oinfo));
2360 op_data->op_fid2 = oinfo->lmo_fid;
2361 tp_tgt = lmv_tgt(lmv, oinfo->lmo_mds);
2365 /* parent unchanged and update namespace only */
2366 if (lu_fid_eq(&op_data->op_fid4, &op_data->op_fid2) &&
2367 op_data->op_bias & MDS_MIGRATE_NSONLY)
2371 sp_tgt = parent_tgt;
2374 child_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2375 if (IS_ERR(child_tgt))
2376 RETURN(PTR_ERR(child_tgt));
2378 if (lmv_topdir_specific_migrate(op_data)) {
2379 struct lmv_user_md *lum = op_data->op_data;
2381 op_data->op_mds = le32_to_cpu(lum->lum_stripe_offset);
2382 } else if (lmv_topdir_qos_migrate(op_data)) {
2383 tgt = lmv_locate_tgt_lf(lmv);
2384 if (tgt == ERR_PTR(-EAGAIN))
2385 tgt = lmv_locate_tgt_rr(lmv);
2387 RETURN(PTR_ERR(tgt));
2389 op_data->op_mds = tgt->ltd_index;
2390 } else if (lmv_subdir_specific_migrate(op_data)) {
2391 struct lmv_user_md *lum = op_data->op_data;
2395 if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC_SPECIFIC) {
2396 /* adjust MDTs in lum, since subdir is located on where
2397 * its parent stripe is, not the first specified MDT.
2399 for (i = 0; i < le32_to_cpu(lum->lum_stripe_count);
2401 if (le32_to_cpu(lum->lum_objects[i].lum_mds) ==
2406 if (i == le32_to_cpu(lum->lum_stripe_count))
2409 lum->lum_objects[i].lum_mds =
2410 lum->lum_objects[0].lum_mds;
2411 lum->lum_objects[0].lum_mds =
2412 cpu_to_le32(tp_tgt->ltd_index);
2414 /* NB, the above adjusts subdir migration for command like
2415 * "lfs migrate -m 0,1,2 ...", but for migration like
2416 * "lfs migrate -m 0 -c 2 ...", the top dir is migrated to MDT0
2417 * and MDT1, however its subdir may be migrated to MDT1 and MDT2
2420 lum->lum_stripe_offset = cpu_to_le32(tp_tgt->ltd_index);
2421 op_data->op_mds = tp_tgt->ltd_index;
2422 } else if (tp_tgt) {
2423 op_data->op_mds = tp_tgt->ltd_index;
2425 op_data->op_mds = sp_tgt->ltd_index;
2428 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2433 * for directory, send migrate request to the MDT where the object will
2434 * be migrated to, because we can't create a striped directory remotely.
2436 * otherwise, send to the MDT where source is located because regular
2437 * file may open lease.
2439 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2440 * backward compatibility.
2442 if (S_ISDIR(op_data->op_mode) &&
2443 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2444 tgt = lmv_fid2tgt(lmv, &target_fid);
2446 RETURN(PTR_ERR(tgt));
2451 /* cancel UPDATE lock of parent master object */
2452 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index, LCK_EX,
2453 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2457 /* cancel UPDATE lock of source parent */
2458 if (sp_tgt != parent_tgt) {
2460 * migrate RPC packs master object FID, because we can only pack
2461 * two FIDs in reint RPC, but MDS needs to know both source
2462 * parent and target parent, and it will obtain them from master
2463 * FID and LMV, the other FID in RPC is kept for target.
2465 * since this FID is not passed to MDC, cancel it anyway.
2467 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2468 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2472 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2474 op_data->op_fid4 = target_fid;
2476 /* cancel UPDATE locks of target parent */
2477 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2478 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2482 /* cancel LOOKUP lock of source if source is remote object */
2483 if (child_tgt != sp_tgt) {
2484 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index,
2485 LCK_EX, MDS_INODELOCK_LOOKUP,
2486 MF_MDC_CANCEL_FID3);
2491 /* cancel ELC locks of source */
2492 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_index, LCK_EX,
2493 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2497 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2502 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2503 const char *old, size_t oldlen,
2504 const char *new, size_t newlen,
2505 struct ptlrpc_request **request)
2507 struct obd_device *obd = exp->exp_obd;
2508 struct lmv_obd *lmv = &obd->u.lmv;
2509 struct lmv_tgt_desc *sp_tgt;
2510 struct lmv_tgt_desc *tp_tgt = NULL;
2511 struct lmv_tgt_desc *src_tgt = NULL;
2512 struct lmv_tgt_desc *tgt;
2513 struct mdt_body *body;
2518 LASSERT(oldlen != 0);
2520 if (op_data->op_cli_flags & CLI_MIGRATE) {
2521 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2525 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2526 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2527 op_data->op_cap = current_cap();
2529 op_data->op_name = new;
2530 op_data->op_namelen = newlen;
2532 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2534 RETURN(PTR_ERR(tp_tgt));
2536 /* Since the target child might be destroyed, and it might become
2537 * orphan, and we can only check orphan on the local MDT right now, so
2538 * we send rename request to the MDT where target child is located. If
2539 * target child does not exist, then it will send the request to the
2541 if (fid_is_sane(&op_data->op_fid4)) {
2542 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2544 RETURN(PTR_ERR(tgt));
2549 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2551 /* cancel UPDATE locks of target parent */
2552 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_index, LCK_EX,
2553 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2557 if (fid_is_sane(&op_data->op_fid4)) {
2558 /* cancel LOOKUP lock of target on target parent */
2559 if (tgt != tp_tgt) {
2560 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2561 tgt->ltd_index, LCK_EX,
2562 MDS_INODELOCK_LOOKUP,
2563 MF_MDC_CANCEL_FID4);
2569 if (fid_is_sane(&op_data->op_fid3)) {
2570 src_tgt = lmv_fid2tgt(lmv, &op_data->op_fid3);
2571 if (IS_ERR(src_tgt))
2572 RETURN(PTR_ERR(src_tgt));
2574 /* cancel ELC locks of source */
2575 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_index,
2576 LCK_EX, MDS_INODELOCK_ELC,
2577 MF_MDC_CANCEL_FID3);
2582 op_data->op_name = old;
2583 op_data->op_namelen = oldlen;
2585 sp_tgt = lmv_locate_tgt(lmv, op_data);
2587 RETURN(PTR_ERR(sp_tgt));
2589 /* cancel UPDATE locks of source parent */
2590 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_index, LCK_EX,
2591 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2595 if (fid_is_sane(&op_data->op_fid3)) {
2596 /* cancel LOOKUP lock of source on source parent */
2597 if (src_tgt != sp_tgt) {
2598 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2599 tgt->ltd_index, LCK_EX,
2600 MDS_INODELOCK_LOOKUP,
2601 MF_MDC_CANCEL_FID3);
2608 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2609 PFID(&op_data->op_fid1), (int)oldlen, old,
2610 PFID(&op_data->op_fid2), (int)newlen, new);
2612 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2614 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2615 ptlrpc_req_finished(*request);
2620 if (rc && rc != -EXDEV)
2623 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2627 /* Not cross-ref case, just get out of here. */
2628 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2631 op_data->op_fid4 = body->mbo_fid1;
2633 ptlrpc_req_finished(*request);
2636 tgt = lmv_fid2tgt(lmv, &op_data->op_fid4);
2638 RETURN(PTR_ERR(tgt));
2640 if (fid_is_sane(&op_data->op_fid4)) {
2641 /* cancel LOOKUP lock of target on target parent */
2642 if (tgt != tp_tgt) {
2643 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2644 tgt->ltd_index, LCK_EX,
2645 MDS_INODELOCK_LOOKUP,
2646 MF_MDC_CANCEL_FID4);
2655 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2656 void *ea, size_t ealen, struct ptlrpc_request **request)
2658 struct obd_device *obd = exp->exp_obd;
2659 struct lmv_obd *lmv = &obd->u.lmv;
2660 struct lmv_tgt_desc *tgt;
2665 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2666 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2667 op_data->op_xvalid);
2669 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2670 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
2672 RETURN(PTR_ERR(tgt));
2674 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2679 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2680 struct ptlrpc_request **request)
2682 struct obd_device *obd = exp->exp_obd;
2683 struct lmv_obd *lmv = &obd->u.lmv;
2684 struct lmv_tgt_desc *tgt;
2689 tgt = lmv_fid2tgt(lmv, fid);
2691 RETURN(PTR_ERR(tgt));
2693 rc = md_fsync(tgt->ltd_exp, fid, request);
2697 struct stripe_dirent {
2698 struct page *sd_page;
2699 struct lu_dirpage *sd_dp;
2700 struct lu_dirent *sd_ent;
2704 struct lmv_dir_ctxt {
2705 struct lmv_obd *ldc_lmv;
2706 struct md_op_data *ldc_op_data;
2707 struct md_readdir_info *ldc_mrinfo;
2710 struct stripe_dirent ldc_stripes[0];
2713 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2715 if (stripe->sd_page) {
2716 kunmap(stripe->sd_page);
2717 put_page(stripe->sd_page);
2718 stripe->sd_page = NULL;
2719 stripe->sd_ent = NULL;
2723 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2727 for (i = 0; i < ctxt->ldc_count; i++)
2728 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2731 /* if @ent is dummy, or . .., get next */
2732 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2733 struct lu_dirent *ent,
2736 for (; ent; ent = lu_dirent_next(ent)) {
2737 /* Skip dummy entry */
2738 if (le16_to_cpu(ent->lde_namelen) == 0)
2741 /* skip . and .. for other stripes */
2743 (strncmp(ent->lde_name, ".",
2744 le16_to_cpu(ent->lde_namelen)) == 0 ||
2745 strncmp(ent->lde_name, "..",
2746 le16_to_cpu(ent->lde_namelen)) == 0))
2749 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2756 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2757 struct stripe_dirent *stripe,
2760 struct md_op_data *op_data = ctxt->ldc_op_data;
2761 struct lmv_oinfo *oinfo;
2762 struct lu_fid fid = op_data->op_fid1;
2763 struct inode *inode = op_data->op_data;
2764 struct lmv_tgt_desc *tgt;
2765 struct lu_dirent *ent = stripe->sd_ent;
2766 __u64 hash = ctxt->ldc_hash;
2771 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2775 if (stripe->sd_page) {
2776 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2778 /* @hash should be the last dirent hash */
2779 LASSERTF(hash <= end,
2780 "ctxt@%p stripe@%p hash %llx end %llx\n",
2781 ctxt, stripe, hash, end);
2782 /* unload last page */
2783 stripe_dirent_unload(stripe);
2785 if (end == MDS_DIR_END_OFF) {
2786 stripe->sd_eof = true;
2792 oinfo = &op_data->op_lso1->lso_lsm.lsm_md_oinfo[stripe_index];
2793 if (!oinfo->lmo_root) {
2798 tgt = lmv_tgt(ctxt->ldc_lmv, oinfo->lmo_mds);
2804 /* op_data is shared by stripes, reset after use */
2805 op_data->op_fid1 = oinfo->lmo_fid;
2806 op_data->op_fid2 = oinfo->lmo_fid;
2807 op_data->op_data = oinfo->lmo_root;
2809 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_mrinfo, hash,
2812 op_data->op_fid1 = fid;
2813 op_data->op_fid2 = fid;
2814 op_data->op_data = inode;
2819 stripe->sd_dp = page_address(stripe->sd_page);
2820 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2822 /* in case a page filled with ., .. and dummy, read next */
2825 stripe->sd_ent = ent;
2828 /* treat error as eof, so dir can be partially accessed */
2829 stripe->sd_eof = true;
2830 ctxt->ldc_mrinfo->mr_partial_readdir_rc = rc;
2831 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2832 "directory is partially accessed!\n",
2833 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2840 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2842 struct obd_device *obd = exp->exp_obd;
2843 struct lmv_obd *lmv = &obd->u.lmv;
2844 struct lmv_tgt_desc *tgt;
2849 rc = lmv_check_connect(obd);
2853 tgt = lmv_fid2tgt(lmv, &data->op_fid1);
2855 RETURN(PTR_ERR(tgt));
2857 data->op_flags |= MF_MDC_CANCEL_FID1;
2858 rc = md_file_resync(tgt->ltd_exp, data);
2863 * Get dirent with the closest hash for striped directory
2865 * This function will search the dir entry, whose hash value is the
2866 * closest(>=) to hash from all of sub-stripes, and it is only being called
2867 * for striped directory.
2869 * \param[in] ctxt dir read context
2871 * \retval dirent get the entry successfully
2872 * NULL does not get the entry, normally it means
2873 * it reaches the end of the directory, while read
2874 * stripe dirent error is ignored to allow partial
2877 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2879 struct stripe_dirent *stripe;
2880 struct lu_dirent *ent = NULL;
2884 /* TODO: optimize with k-way merge sort */
2885 for (i = 0; i < ctxt->ldc_count; i++) {
2886 stripe = &ctxt->ldc_stripes[i];
2890 if (!stripe->sd_ent) {
2891 stripe_dirent_load(ctxt, stripe, i);
2892 if (!stripe->sd_ent) {
2893 LASSERT(stripe->sd_eof);
2899 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2900 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2902 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2909 stripe = &ctxt->ldc_stripes[min];
2910 ent = stripe->sd_ent;
2911 /* pop found dirent */
2912 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2920 * Build dir entry page for striped directory
2922 * This function gets one entry by @offset from a striped directory. It will
2923 * read entries from all of stripes, and choose one closest to the required
2924 * offset(&offset). A few notes
2925 * 1. skip . and .. for non-zero stripes, because there can only have one .
2926 * and .. in a directory.
2927 * 2. op_data will be shared by all of stripes, instead of allocating new
2928 * one, so need to restore before reusing.
2930 * \param[in] exp obd export refer to LMV
2931 * \param[in] op_data hold those MD parameters of read_entry
2932 * \param[in] mrinfo ldlm callback being used in enqueue in mdc_read_entry,
2933 * and partial readdir result will be stored in it.
2934 * \param[in] offset starting hash offset
2935 * \param[out] ppage the page holding the entry. Note: because the entry
2936 * will be accessed in upper layer, so we need hold the
2937 * page until the usages of entry is finished, see
2938 * ll_dir_entry_next.
2940 * retval =0 if get entry successfully
2941 * <0 cannot get entry
2943 static int lmv_striped_read_page(struct obd_export *exp,
2944 struct md_op_data *op_data,
2945 struct md_readdir_info *mrinfo, __u64 offset,
2946 struct page **ppage)
2948 struct page *page = NULL;
2949 struct lu_dirpage *dp;
2951 struct lu_dirent *ent;
2952 struct lu_dirent *last_ent;
2954 struct lmv_dir_ctxt *ctxt;
2955 struct lu_dirent *next = NULL;
2961 /* Allocate a page and read entries from all of stripes and fill
2962 * the page by hash order */
2963 page = alloc_page(GFP_KERNEL);
2967 /* Initialize the entry page */
2969 memset(dp, 0, sizeof(*dp));
2970 dp->ldp_hash_start = cpu_to_le64(offset);
2973 left_bytes = PAGE_SIZE - sizeof(*dp);
2977 /* initalize dir read context */
2978 stripe_count = op_data->op_lso1->lso_lsm.lsm_md_stripe_count;
2979 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2981 GOTO(free_page, rc = -ENOMEM);
2982 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2983 ctxt->ldc_op_data = op_data;
2984 ctxt->ldc_mrinfo = mrinfo;
2985 ctxt->ldc_hash = offset;
2986 ctxt->ldc_count = stripe_count;
2989 next = lmv_dirent_next(ctxt);
2991 /* end of directory */
2993 ctxt->ldc_hash = MDS_DIR_END_OFF;
2996 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2998 ent_size = le16_to_cpu(next->lde_reclen);
3000 /* the last entry lde_reclen is 0, but it might not be the last
3001 * one of this temporay dir page */
3003 ent_size = lu_dirent_calc_size(
3004 le16_to_cpu(next->lde_namelen),
3005 le32_to_cpu(next->lde_attrs));
3007 if (ent_size > left_bytes)
3010 memcpy(ent, next, ent_size);
3012 /* Replace . with master FID and Replace .. with the parent FID
3013 * of master object */
3014 if (strncmp(ent->lde_name, ".",
3015 le16_to_cpu(ent->lde_namelen)) == 0 &&
3016 le16_to_cpu(ent->lde_namelen) == 1)
3017 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
3018 else if (strncmp(ent->lde_name, "..",
3019 le16_to_cpu(ent->lde_namelen)) == 0 &&
3020 le16_to_cpu(ent->lde_namelen) == 2)
3021 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
3023 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
3024 le16_to_cpu(ent->lde_namelen), ent->lde_name,
3025 le64_to_cpu(ent->lde_hash));
3027 left_bytes -= ent_size;
3028 ent->lde_reclen = cpu_to_le16(ent_size);
3030 ent = (void *)ent + ent_size;
3033 last_ent->lde_reclen = 0;
3036 dp->ldp_flags |= LDF_EMPTY;
3037 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
3038 dp->ldp_flags |= LDF_COLLIDE;
3039 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
3040 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
3042 put_lmv_dir_ctxt(ctxt);
3043 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
3056 static int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
3057 struct md_readdir_info *mrinfo, __u64 offset,
3058 struct page **ppage)
3060 struct obd_device *obd = exp->exp_obd;
3061 struct lmv_obd *lmv = &obd->u.lmv;
3062 struct lmv_tgt_desc *tgt;
3067 if (unlikely(lmv_dir_foreign(op_data->op_lso1)))
3070 if (unlikely(lmv_dir_striped(op_data->op_lso1))) {
3071 rc = lmv_striped_read_page(exp, op_data, mrinfo, offset, ppage);
3075 tgt = lmv_fid2tgt(lmv, &op_data->op_fid1);
3077 RETURN(PTR_ERR(tgt));
3079 rc = md_read_page(tgt->ltd_exp, op_data, mrinfo, offset, ppage);
3085 * Unlink a file/directory
3087 * Unlink a file or directory under the parent dir. The unlink request
3088 * usually will be sent to the MDT where the child is located, but if
3089 * the client does not have the child FID then request will be sent to the
3090 * MDT where the parent is located.
3092 * If the parent is a striped directory then it also needs to locate which
3093 * stripe the name of the child is located, and replace the parent FID
3094 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
3095 * it will walk through all of sub-stripes until the child is being
3098 * \param[in] exp export refer to LMV
3099 * \param[in] op_data different parameters transferred beween client
3100 * MD stacks, name, namelen, FIDs etc.
3101 * op_fid1 is the parent FID, op_fid2 is the child
3103 * \param[out] request point to the request of unlink.
3105 * retval 0 if succeed
3106 * negative errno if failed.
3108 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
3109 struct ptlrpc_request **request)
3111 struct obd_device *obd = exp->exp_obd;
3112 struct lmv_obd *lmv = &obd->u.lmv;
3113 struct lmv_tgt_desc *tgt;
3114 struct lmv_tgt_desc *parent_tgt;
3115 struct mdt_body *body;
3120 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3121 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3122 op_data->op_cap = current_cap();
3125 parent_tgt = lmv_locate_tgt(lmv, op_data);
3126 if (IS_ERR(parent_tgt))
3127 RETURN(PTR_ERR(parent_tgt));
3129 if (likely(!fid_is_zero(&op_data->op_fid2))) {
3130 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3132 RETURN(PTR_ERR(tgt));
3138 * If child's fid is given, cancel unused locks for it if it is from
3139 * another export than parent.
3141 * LOOKUP lock for child (fid3) should also be cancelled on parent
3142 * tgt_tgt in mdc_unlink().
3144 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
3146 if (parent_tgt != tgt)
3147 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_index,
3148 LCK_EX, MDS_INODELOCK_LOOKUP,
3149 MF_MDC_CANCEL_FID3);
3151 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_index, LCK_EX,
3152 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
3156 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
3157 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
3160 rc = md_unlink(tgt->ltd_exp, op_data, request);
3161 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
3162 ptlrpc_req_finished(*request);
3170 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
3174 /* Not cross-ref case, just get out of here. */
3175 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
3178 /* This is a remote object, try remote MDT. */
3179 op_data->op_fid2 = body->mbo_fid1;
3180 ptlrpc_req_finished(*request);
3183 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3185 RETURN(PTR_ERR(tgt));
3190 static int lmv_precleanup(struct obd_device *obd)
3193 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
3198 * Get by key a value associated with a LMV device.
3200 * Dispatch request to lower-layer devices as needed.
3202 * \param[in] env execution environment for this thread
3203 * \param[in] exp export for the LMV device
3204 * \param[in] keylen length of key identifier
3205 * \param[in] key identifier of key to get value for
3206 * \param[in] vallen size of \a val
3207 * \param[out] val pointer to storage location for value
3208 * \param[in] lsm optional striping metadata of object
3210 * \retval 0 on success
3211 * \retval negative negated errno on failure
3213 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
3214 __u32 keylen, void *key, __u32 *vallen, void *val)
3216 struct obd_device *obd;
3217 struct lmv_obd *lmv;
3218 struct lu_tgt_desc *tgt;
3223 obd = class_exp2obd(exp);
3225 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3226 exp->exp_handle.h_cookie);
3231 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
3232 LASSERT(*vallen == sizeof(__u32));
3233 lmv_foreach_connected_tgt(lmv, tgt) {
3234 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
3239 } else if (KEY_IS(KEY_MAX_EASIZE) ||
3240 KEY_IS(KEY_DEFAULT_EASIZE) ||
3241 KEY_IS(KEY_CONN_DATA)) {
3243 * Forwarding this request to first MDS, it should know LOV
3246 tgt = lmv_tgt(lmv, 0);
3250 rc = obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val);
3251 if (!rc && KEY_IS(KEY_CONN_DATA))
3252 exp->exp_connect_data = *(struct obd_connect_data *)val;
3254 } else if (KEY_IS(KEY_TGT_COUNT)) {
3255 *((int *)val) = lmv->lmv_mdt_descs.ltd_tgts_size;
3259 CDEBUG(D_IOCTL, "Invalid key\n");
3263 static int lmv_rmfid(struct obd_export *exp, struct fid_array *fa,
3264 int *__rcs, struct ptlrpc_request_set *_set)
3266 struct obd_device *obd = class_exp2obd(exp);
3267 struct ptlrpc_request_set *set = _set;
3268 struct lmv_obd *lmv = &obd->u.lmv;
3269 int tgt_count = lmv->lmv_mdt_count;
3270 struct lu_tgt_desc *tgt;
3271 struct fid_array *fat, **fas = NULL;
3272 int i, rc, **rcs = NULL;
3275 set = ptlrpc_prep_set();
3280 /* split FIDs by targets */
3281 OBD_ALLOC_PTR_ARRAY(fas, tgt_count);
3283 GOTO(out, rc = -ENOMEM);
3284 OBD_ALLOC_PTR_ARRAY(rcs, tgt_count);
3286 GOTO(out_fas, rc = -ENOMEM);
3288 for (i = 0; i < fa->fa_nr; i++) {
3291 rc = lmv_fld_lookup(lmv, &fa->fa_fids[i], &idx);
3293 CDEBUG(D_OTHER, "can't lookup "DFID": rc = %d\n",
3294 PFID(&fa->fa_fids[i]), rc);
3297 LASSERT(idx < tgt_count);
3299 OBD_ALLOC(fas[idx], offsetof(struct fid_array,
3300 fa_fids[fa->fa_nr]));
3302 GOTO(out, rc = -ENOMEM);
3304 OBD_ALLOC_PTR_ARRAY(rcs[idx], fa->fa_nr);
3306 GOTO(out, rc = -ENOMEM);
3309 fat->fa_fids[fat->fa_nr++] = fa->fa_fids[i];
3312 lmv_foreach_connected_tgt(lmv, tgt) {
3313 fat = fas[tgt->ltd_index];
3314 if (!fat || fat->fa_nr == 0)
3316 rc = md_rmfid(tgt->ltd_exp, fat, rcs[tgt->ltd_index], set);
3319 rc = ptlrpc_set_wait(NULL, set);
3322 for (i = 0; i < tgt_count; i++) {
3324 if (!fat || fat->fa_nr == 0)
3326 /* copy FIDs back */
3327 memcpy(fa->fa_fids + j, fat->fa_fids,
3328 fat->fa_nr * sizeof(struct lu_fid));
3330 memcpy(__rcs + j, rcs[i], fat->fa_nr * sizeof(**rcs));
3335 ptlrpc_set_destroy(set);
3338 for (i = 0; i < tgt_count; i++) {
3340 OBD_FREE(fas[i], offsetof(struct fid_array,
3341 fa_fids[fa->fa_nr]));
3343 OBD_FREE_PTR_ARRAY(rcs[i], fa->fa_nr);
3346 OBD_FREE_PTR_ARRAY(rcs, tgt_count);
3349 OBD_FREE_PTR_ARRAY(fas, tgt_count);
3355 * Asynchronously set by key a value associated with a LMV device.
3357 * Dispatch request to lower-layer devices as needed.
3359 * \param[in] env execution environment for this thread
3360 * \param[in] exp export for the LMV device
3361 * \param[in] keylen length of key identifier
3362 * \param[in] key identifier of key to store value for
3363 * \param[in] vallen size of value to store
3364 * \param[in] val pointer to data to be stored
3365 * \param[in] set optional list of related ptlrpc requests
3367 * \retval 0 on success
3368 * \retval negative negated errno on failure
3370 static int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
3371 __u32 keylen, void *key, __u32 vallen, void *val,
3372 struct ptlrpc_request_set *set)
3374 struct lmv_tgt_desc *tgt;
3375 struct obd_device *obd;
3376 struct lmv_obd *lmv;
3380 obd = class_exp2obd(exp);
3382 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
3383 exp->exp_handle.h_cookie);
3388 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
3389 KEY_IS(KEY_DEFAULT_EASIZE)) {
3392 lmv_foreach_connected_tgt(lmv, tgt) {
3393 err = obd_set_info_async(env, tgt->ltd_exp,
3394 keylen, key, vallen, val, set);
3405 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3406 const struct lmv_mds_md_v1 *lmm1)
3408 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3415 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3416 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3417 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3418 if (CFS_FAIL_CHECK(OBD_FAIL_LMV_UNKNOWN_STRIPE))
3419 lsm->lsm_md_hash_type = cfs_fail_val ?: LMV_HASH_TYPE_UNKNOWN;
3421 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3422 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3423 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3424 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3425 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3426 sizeof(lsm->lsm_md_pool_name));
3428 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3431 CDEBUG(D_INFO, "unpack lsm count %d/%d, master %d hash_type %#x/%#x "
3432 "layout_version %d\n", lsm->lsm_md_stripe_count,
3433 lsm->lsm_md_migrate_offset, lsm->lsm_md_master_mdt_index,
3434 lsm->lsm_md_hash_type, lsm->lsm_md_migrate_hash,
3435 lsm->lsm_md_layout_version);
3437 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3438 for (i = 0; i < stripe_count; i++) {
3439 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3440 &lmm1->lmv_stripe_fids[i]);
3442 * set default value -1, so lmv_locate_tgt() knows this stripe
3443 * target is not initialized.
3445 lsm->lsm_md_oinfo[i].lmo_mds = LMV_OFFSET_DEFAULT;
3446 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3449 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3450 &lsm->lsm_md_oinfo[i].lmo_mds);
3457 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3458 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3464 static inline int lmv_unpack_user_md(struct obd_export *exp,
3465 struct lmv_stripe_md *lsm,
3466 const struct lmv_user_md *lmu)
3468 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3469 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3470 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3471 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3472 lsm->lsm_md_max_inherit = lmu->lum_max_inherit;
3473 lsm->lsm_md_max_inherit_rr = lmu->lum_max_inherit_rr;
3474 lsm->lsm_md_pool_name[LOV_MAXPOOLNAME] = 0;
3479 struct lmv_stripe_object *lmv_stripe_object_alloc(__u32 magic,
3480 const union lmv_mds_md *lmm,
3483 struct lmv_stripe_object *lsm_obj;
3486 if (magic == LMV_MAGIC_FOREIGN) {
3487 struct lmv_foreign_md *lfm;
3489 size = offsetof(typeof(*lfm), lfm_value[0]);
3490 if (lmm_size < size)
3491 RETURN(ERR_PTR(-EPROTO));
3493 size += le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3494 if (lmm_size < size)
3495 RETURN(ERR_PTR(-EPROTO));
3497 OBD_ALLOC_LARGE(lsm_obj, lmm_size +
3498 offsetof(typeof(*lsm_obj), lso_lfm));
3500 if (magic == LMV_MAGIC_V1) {
3503 size = offsetof(struct lmv_mds_md_v1,
3504 lmv_stripe_fids[0]);
3505 if (lmm_size < size)
3506 RETURN(ERR_PTR(-EPROTO));
3508 count = lmv_mds_md_stripe_count_get(lmm);
3509 size += count * sizeof(struct lu_fid);
3510 if (lmm_size < size)
3511 RETURN(ERR_PTR(-EPROTO));
3513 size = lmv_stripe_md_size(count);
3515 if (lmm && lmm_size < sizeof(struct lmv_user_md))
3516 RETURN(ERR_PTR(-EPROTO));
3519 * Unpack default dirstripe(lmv_user_md) to
3520 * lmv_stripe_md, stripecount should be 0 then.
3522 size = lmv_stripe_md_size(0);
3524 size += offsetof(typeof(*lsm_obj), lso_lsm);
3525 OBD_ALLOC(lsm_obj, size);
3529 atomic_set(&lsm_obj->lso_refs, 1);
3533 RETURN(ERR_PTR(-ENOMEM));
3535 EXPORT_SYMBOL(lmv_stripe_object_alloc);
3537 static int lmv_stripe_object_create(struct obd_export *exp,
3538 struct lmv_stripe_object **lsop,
3539 const union lmv_mds_md *lmm,
3542 struct lmv_stripe_object *lsm_obj;
3547 LASSERT(lsop != NULL && *lsop == NULL);
3552 magic = le32_to_cpu(lmm->lmv_magic);
3553 if (magic == LMV_MAGIC_STRIPE)
3556 if (magic != LMV_MAGIC_V1 && magic != LMV_USER_MAGIC &&
3557 magic != LMV_MAGIC_FOREIGN) {
3558 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3559 exp->exp_obd->obd_name, magic, -EIO);
3563 /* foreign lmv case */
3564 if (magic == LMV_MAGIC_FOREIGN) {
3565 struct lmv_foreign_md *lfm;
3567 lsm_obj = lmv_stripe_object_alloc(magic, lmm, lmm_size);
3568 if (IS_ERR(lsm_obj))
3569 RETURN(PTR_ERR(lsm_obj));
3572 lfm = &lsm_obj->lso_lfm;
3573 lfm->lfm_magic = magic;
3574 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3575 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3576 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3577 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3583 lsm_obj = lmv_stripe_object_alloc(magic, lmm, lmm_size);
3584 if (IS_ERR(lsm_obj))
3585 RETURN(PTR_ERR(lsm_obj));
3589 rc = lmv_unpack_md_v1(exp, &lsm_obj->lso_lsm, &lmm->lmv_md_v1);
3591 case LMV_USER_MAGIC:
3592 rc = lmv_unpack_user_md(exp, &lsm_obj->lso_lsm,
3596 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3603 lmv_stripe_object_put(&lsm_obj);
3609 struct lmv_stripe_object *
3610 lmv_stripe_object_get(struct lmv_stripe_object *lsm_obj)
3612 if (lsm_obj == NULL)
3615 atomic_inc(&lsm_obj->lso_refs);
3616 CDEBUG(D_INODE, "get %p %u\n", lsm_obj,
3617 atomic_read(&lsm_obj->lso_refs));
3620 EXPORT_SYMBOL(lmv_stripe_object_get);
3622 void lmv_stripe_object_put(struct lmv_stripe_object **lsop)
3624 struct lmv_stripe_object *lsm_obj;
3628 LASSERT(lsop != NULL);
3631 if (lsm_obj == NULL)
3635 CDEBUG(D_INODE, "put %p %u\n", lsm_obj,
3636 atomic_read(&lsm_obj->lso_refs) - 1);
3638 if (!atomic_dec_and_test(&lsm_obj->lso_refs))
3641 if (lmv_dir_foreign(lsm_obj)) {
3642 size = lsm_obj->lso_lfm.lfm_length +
3643 offsetof(typeof(lsm_obj->lso_lfm), lfm_value[0]) +
3644 offsetof(typeof(*lsm_obj), lso_lsm);
3645 OBD_FREE_LARGE(lsm_obj, size);
3649 if (lmv_dir_striped(lsm_obj)) {
3650 struct lmv_stripe_md *lsm = &lsm_obj->lso_lsm;
3652 for (i = 0; i < lsm->lsm_md_stripe_count; i++)
3653 iput(lsm->lsm_md_oinfo[i].lmo_root);
3654 size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3656 size = lmv_stripe_md_size(0);
3658 OBD_FREE(lsm_obj, size + offsetof(typeof(*lsm_obj), lso_lsm));
3660 EXPORT_SYMBOL(lmv_stripe_object_put);
3662 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3663 union ldlm_policy_data *policy,
3664 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3667 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3668 struct lu_tgt_desc *tgt;
3674 LASSERT(fid != NULL);
3676 lmv_foreach_connected_tgt(lmv, tgt) {
3677 if (!tgt->ltd_active)
3680 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3688 static int lmv_set_lock_data(struct obd_export *exp,
3689 const struct lustre_handle *lockh,
3690 void *data, __u64 *bits)
3692 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3693 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3698 if (tgt == NULL || tgt->ltd_exp == NULL)
3700 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3704 static enum ldlm_mode
3705 lmv_lock_match(struct obd_export *exp, __u64 flags,
3706 const struct lu_fid *fid, enum ldlm_type type,
3707 union ldlm_policy_data *policy,
3708 enum ldlm_mode mode, struct lustre_handle *lockh)
3710 struct obd_device *obd = exp->exp_obd;
3711 struct lmv_obd *lmv = &obd->u.lmv;
3712 struct lu_tgt_desc *tgt;
3713 __u64 bits = policy->l_inodebits.bits;
3714 enum ldlm_mode rc = LCK_MINMODE;
3718 /* only one bit is set */
3719 LASSERT(bits && !(bits & (bits - 1)));
3720 /* With DNE every object can have two locks in different namespaces:
3721 * lookup lock in space of MDT storing direntry and update/open lock in
3722 * space of MDT storing inode. Try the MDT that the FID maps to first,
3723 * since this can be easily found, and only try others if that fails.
3725 if (bits == MDS_INODELOCK_LOOKUP) {
3726 for (i = 0, index = lmv_fid2tgt_index(lmv, fid);
3727 i < lmv->lmv_mdt_descs.ltd_tgts_size; i++,
3728 index = (index + 1) % lmv->lmv_mdt_descs.ltd_tgts_size) {
3731 "%s: "DFID" is inaccessible: rc = %d\n",
3732 obd->obd_name, PFID(fid), index);
3735 tgt = lmv_tgt(lmv, index);
3736 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active)
3738 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3739 policy, mode, lockh);
3744 tgt = lmv_fid2tgt(lmv, fid);
3745 if (!IS_ERR(tgt) && tgt->ltd_exp && tgt->ltd_active)
3746 rc = md_lock_match(tgt->ltd_exp, flags, fid, type,
3747 policy, mode, lockh);
3750 CDEBUG(D_INODE, "Lock match for "DFID": %d\n", PFID(fid), rc);
3756 lmv_get_lustre_md(struct obd_export *exp, struct req_capsule *pill,
3757 struct obd_export *dt_exp, struct obd_export *md_exp,
3758 struct lustre_md *md)
3760 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3761 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3763 if (!tgt || !tgt->ltd_exp)
3766 return md_get_lustre_md(tgt->ltd_exp, pill, dt_exp, md_exp, md);
3769 static int lmv_put_lustre_md(struct obd_export *exp, struct lustre_md *md)
3771 struct obd_device *obd = exp->exp_obd;
3772 struct lmv_obd *lmv = &obd->u.lmv;
3773 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3776 lmv_stripe_object_put(&md->def_lsm_obj);
3777 lmv_stripe_object_put(&md->lsm_obj);
3779 if (!tgt || !tgt->ltd_exp)
3784 static int lmv_set_open_replay_data(struct obd_export *exp,
3785 struct obd_client_handle *och,
3786 struct lookup_intent *it)
3788 struct obd_device *obd = exp->exp_obd;
3789 struct lmv_obd *lmv = &obd->u.lmv;
3790 struct lmv_tgt_desc *tgt;
3794 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3796 RETURN(PTR_ERR(tgt));
3798 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3801 static int lmv_clear_open_replay_data(struct obd_export *exp,
3802 struct obd_client_handle *och)
3804 struct obd_device *obd = exp->exp_obd;
3805 struct lmv_obd *lmv = &obd->u.lmv;
3806 struct lmv_tgt_desc *tgt;
3810 tgt = lmv_fid2tgt(lmv, &och->och_fid);
3812 RETURN(PTR_ERR(tgt));
3814 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3817 static int lmv_intent_getattr_async(struct obd_export *exp,
3818 struct md_op_item *item)
3820 struct md_op_data *op_data = &item->mop_data;
3821 struct obd_device *obd = exp->exp_obd;
3822 struct lmv_obd *lmv = &obd->u.lmv;
3823 struct lmv_tgt_desc *ptgt;
3824 struct lmv_tgt_desc *ctgt;
3829 if (!(fid_is_sane(&op_data->op_fid2) ||
3830 fid_is_zero(&op_data->op_fid2)))
3833 ptgt = lmv_locate_tgt(lmv, op_data);
3835 RETURN(PTR_ERR(ptgt));
3838 * Zeroed FID @op_fid2 means that the intent getattr() comes from
3839 * statahead by regularized file names. Currently only do statahead
3840 * for the children files located same as the parent directory.
3842 if (!fid_is_zero(&op_data->op_fid2)) {
3843 ctgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
3845 RETURN(PTR_ERR(ctgt));
3848 * remote object needs two RPCs to lookup and getattr,
3849 * considering the complexity don't support statahead for now.
3855 rc = md_intent_getattr_async(ptgt->ltd_exp, item);
3860 static int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3861 struct lu_fid *fid, __u64 *bits)
3863 struct obd_device *obd = exp->exp_obd;
3864 struct lmv_obd *lmv = &obd->u.lmv;
3865 struct lmv_tgt_desc *tgt;
3870 tgt = lmv_fid2tgt(lmv, fid);
3872 RETURN(PTR_ERR(tgt));
3874 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3878 static int lmv_get_fid_from_lsm(struct obd_export *exp,
3879 const struct lmv_stripe_object *lso,
3880 const char *name, int namelen,
3883 const struct lmv_oinfo *oinfo;
3885 LASSERT(lmv_dir_striped(lso));
3886 oinfo = lsm_name_to_stripe_info(lso, name, namelen, false);
3888 return PTR_ERR(oinfo);
3890 *fid = oinfo->lmo_fid;
3896 * For lmv, only need to send request to master MDT, and the master MDT will
3897 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3898 * we directly fetch data from the slave MDTs.
3900 static int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3901 struct obd_quotactl *oqctl)
3903 struct obd_device *obd = class_exp2obd(exp);
3904 struct lmv_obd *lmv = &obd->u.lmv;
3905 struct lmv_tgt_desc *tgt = lmv_tgt(lmv, 0);
3906 __u64 curspace, curinodes;
3911 if (!tgt || !tgt->ltd_exp || !tgt->ltd_active) {
3912 CERROR("master lmv inactive\n");
3916 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3917 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3921 curspace = curinodes = 0;
3922 lmv_foreach_connected_tgt(lmv, tgt) {
3925 if (!tgt->ltd_active)
3928 err = obd_quotactl(tgt->ltd_exp, oqctl);
3930 CERROR("getquota on mdt %d failed. %d\n",
3931 tgt->ltd_index, err);
3935 curspace += oqctl->qc_dqblk.dqb_curspace;
3936 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3939 oqctl->qc_dqblk.dqb_curspace = curspace;
3940 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3945 static int lmv_merge_attr(struct obd_export *exp,
3946 const struct lmv_stripe_object *lso,
3947 struct cl_attr *attr,
3948 ldlm_blocking_callback cb_blocking)
3950 const struct lmv_stripe_md *lsm = &lso->lso_lsm;
3954 if (!lmv_dir_striped(lso))
3957 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3961 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3962 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3968 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3969 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3970 i_size_read(inode), (unsigned long long)inode->i_blocks,
3971 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3972 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3974 /* for slave stripe, it needs to subtract nlink for . and .. */
3976 attr->cat_nlink += inode->i_nlink - 2;
3978 attr->cat_nlink = inode->i_nlink;
3980 attr->cat_size += i_size_read(inode);
3981 attr->cat_blocks += inode->i_blocks;
3983 if (attr->cat_atime < inode->i_atime.tv_sec)
3984 attr->cat_atime = inode->i_atime.tv_sec;
3986 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3987 attr->cat_ctime = inode->i_ctime.tv_sec;
3989 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3990 attr->cat_mtime = inode->i_mtime.tv_sec;
3995 static struct lu_batch *lmv_batch_create(struct obd_export *exp,
3996 enum lu_batch_flags flags,
3999 struct lu_batch *bh;
4000 struct lmv_batch *lbh;
4005 RETURN(ERR_PTR(-ENOMEM));
4007 bh = &lbh->lbh_super;
4008 bh->lbt_flags = flags;
4009 bh->lbt_max_count = max_count;
4011 if (flags & BATCH_FL_RQSET) {
4012 bh->lbt_rqset = ptlrpc_prep_set();
4013 if (bh->lbt_rqset == NULL) {
4015 RETURN(ERR_PTR(-ENOMEM));
4019 INIT_LIST_HEAD(&lbh->lbh_sub_batch_list);
4023 static int lmv_batch_stop(struct obd_export *exp, struct lu_batch *bh)
4025 struct lmv_batch *lbh;
4026 struct lmvsub_batch *sub;
4027 struct lmvsub_batch *tmp;
4032 lbh = container_of(bh, struct lmv_batch, lbh_super);
4033 list_for_each_entry_safe(sub, tmp, &lbh->lbh_sub_batch_list,
4035 list_del(&sub->sbh_sub_item);
4036 rc = md_batch_stop(sub->sbh_tgt->ltd_exp, sub->sbh_sub);
4038 CERROR("%s: stop batch processing failed: rc = %d\n",
4039 exp->exp_obd->obd_name, rc);
4040 if (bh->lbt_result == 0)
4041 bh->lbt_result = rc;
4046 if (bh->lbt_flags & BATCH_FL_RQSET) {
4047 rc = ptlrpc_set_wait(NULL, bh->lbt_rqset);
4048 ptlrpc_set_destroy(bh->lbt_rqset);
4055 static int lmv_batch_flush(struct obd_export *exp, struct lu_batch *bh,
4058 struct lmv_batch *lbh;
4059 struct lmvsub_batch *sub;
4065 lbh = container_of(bh, struct lmv_batch, lbh_super);
4066 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
4067 rc1 = md_batch_flush(sub->sbh_tgt->ltd_exp, sub->sbh_sub, wait);
4069 CERROR("%s: stop batch processing failed: rc = %d\n",
4070 exp->exp_obd->obd_name, rc);
4071 if (bh->lbt_result == 0)
4072 bh->lbt_result = rc;
4079 if (wait && bh->lbt_flags & BATCH_FL_RQSET) {
4080 rc1 = ptlrpc_set_wait(NULL, bh->lbt_rqset);
4088 static inline struct lmv_tgt_desc *
4089 lmv_batch_locate_tgt(struct lmv_obd *lmv, struct md_op_item *item)
4091 struct md_op_data *op_data = &item->mop_data;
4092 struct lmv_tgt_desc *tgt;
4094 switch (item->mop_opc) {
4095 case MD_OP_GETATTR: {
4096 struct lmv_tgt_desc *ptgt;
4098 if (!(fid_is_sane(&op_data->op_fid2) ||
4099 fid_is_zero(&op_data->op_fid2)))
4100 RETURN(ERR_PTR(-EINVAL));
4102 ptgt = lmv_locate_tgt(lmv, op_data);
4107 * Zeroed @op_fid2 means that it is a statahead populating call
4108 * in the file name pattern which is using file name format to
4109 * prefetch the attributes. Thus it has no idea about the FID of
4110 * the children file. The children file is considered to be
4111 * located on the same storage target with the parent directory
4112 * or the stripped directory.
4114 if (fid_is_zero(&op_data->op_fid2)) {
4119 tgt = lmv_fid2tgt(lmv, &op_data->op_fid2);
4124 * Remote object needs two RPCs to lookup and getattr,
4125 * considering the complexity don't support statahead for now.
4128 RETURN(ERR_PTR(-EREMOTE));
4133 tgt = ERR_PTR(-ENOTSUPP);
4139 struct lu_batch *lmv_batch_lookup_sub(struct lmv_batch *lbh,
4140 struct lmv_tgt_desc *tgt)
4142 struct lmvsub_batch *sub;
4144 list_for_each_entry(sub, &lbh->lbh_sub_batch_list, sbh_sub_item) {
4145 if (sub->sbh_tgt == tgt)
4146 return sub->sbh_sub;
4152 struct lu_batch *lmv_batch_get_sub(struct lmv_batch *lbh,
4153 struct lmv_tgt_desc *tgt)
4155 struct lmvsub_batch *sbh;
4156 struct lu_batch *child_bh;
4157 struct lu_batch *bh;
4161 child_bh = lmv_batch_lookup_sub(lbh, tgt);
4162 if (child_bh != NULL)
4167 RETURN(ERR_PTR(-ENOMEM));
4169 INIT_LIST_HEAD(&sbh->sbh_sub_item);
4172 bh = &lbh->lbh_super;
4173 child_bh = md_batch_create(tgt->ltd_exp, bh->lbt_flags,
4175 if (IS_ERR(child_bh)) {
4180 child_bh->lbt_rqset = bh->lbt_rqset;
4181 sbh->sbh_sub = child_bh;
4182 list_add(&sbh->sbh_sub_item, &lbh->lbh_sub_batch_list);
4186 static int lmv_batch_add(struct obd_export *exp, struct lu_batch *bh,
4187 struct md_op_item *item)
4189 struct obd_device *obd = exp->exp_obd;
4190 struct lmv_obd *lmv = &obd->u.lmv;
4191 struct lmv_tgt_desc *tgt;
4192 struct lmv_batch *lbh;
4193 struct lu_batch *child_bh;
4198 tgt = lmv_batch_locate_tgt(lmv, item);
4200 RETURN(PTR_ERR(tgt));
4202 lbh = container_of(bh, struct lmv_batch, lbh_super);
4203 child_bh = lmv_batch_get_sub(lbh, tgt);
4204 if (IS_ERR(child_bh))
4205 RETURN(PTR_ERR(child_bh));
4207 rc = md_batch_add(tgt->ltd_exp, child_bh, item);
4211 static const struct obd_ops lmv_obd_ops = {
4212 .o_owner = THIS_MODULE,
4213 .o_setup = lmv_setup,
4214 .o_cleanup = lmv_cleanup,
4215 .o_precleanup = lmv_precleanup,
4216 .o_process_config = lmv_process_config,
4217 .o_connect = lmv_connect,
4218 .o_disconnect = lmv_disconnect,
4219 .o_statfs = lmv_statfs,
4220 .o_get_info = lmv_get_info,
4221 .o_set_info_async = lmv_set_info_async,
4222 .o_notify = lmv_notify,
4223 .o_get_uuid = lmv_get_uuid,
4224 .o_fid_alloc = lmv_fid_alloc,
4225 .o_iocontrol = lmv_iocontrol,
4226 .o_quotactl = lmv_quotactl
4229 static const struct md_ops lmv_md_ops = {
4230 .m_get_root = lmv_get_root,
4231 .m_null_inode = lmv_null_inode,
4232 .m_close = lmv_close,
4233 .m_create = lmv_create,
4234 .m_enqueue = lmv_enqueue,
4235 .m_getattr = lmv_getattr,
4236 .m_getxattr = lmv_getxattr,
4237 .m_getattr_name = lmv_getattr_name,
4238 .m_intent_lock = lmv_intent_lock,
4240 .m_rename = lmv_rename,
4241 .m_setattr = lmv_setattr,
4242 .m_setxattr = lmv_setxattr,
4243 .m_fsync = lmv_fsync,
4244 .m_file_resync = lmv_file_resync,
4245 .m_read_page = lmv_read_page,
4246 .m_unlink = lmv_unlink,
4247 .m_init_ea_size = lmv_init_ea_size,
4248 .m_cancel_unused = lmv_cancel_unused,
4249 .m_set_lock_data = lmv_set_lock_data,
4250 .m_lock_match = lmv_lock_match,
4251 .m_get_lustre_md = lmv_get_lustre_md,
4252 .m_put_lustre_md = lmv_put_lustre_md,
4253 .m_merge_attr = lmv_merge_attr,
4254 .m_set_open_replay_data = lmv_set_open_replay_data,
4255 .m_clear_open_replay_data = lmv_clear_open_replay_data,
4256 .m_intent_getattr_async = lmv_intent_getattr_async,
4257 .m_revalidate_lock = lmv_revalidate_lock,
4258 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
4259 .m_stripe_object_create = lmv_stripe_object_create,
4260 .m_rmfid = lmv_rmfid,
4261 .m_batch_create = lmv_batch_create,
4262 .m_batch_add = lmv_batch_add,
4263 .m_batch_stop = lmv_batch_stop,
4264 .m_batch_flush = lmv_batch_flush,
4267 static int __init lmv_init(void)
4269 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true,
4270 LUSTRE_LMV_NAME, NULL);
4273 static void __exit lmv_exit(void)
4275 class_unregister_type(LUSTRE_LMV_NAME);
4278 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4279 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
4280 MODULE_VERSION(LUSTRE_VERSION_STRING);
4281 MODULE_LICENSE("GPL");
4283 module_init(lmv_init);
4284 module_exit(lmv_exit);