4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 #define DEBUG_SUBSYSTEM S_LMV
35 #include <linux/file.h>
36 #include <linux/module.h>
37 #include <linux/init.h>
38 #include <linux/user_namespace.h>
39 #ifdef HAVE_UIDGID_HEADER
40 # include <linux/uidgid.h>
42 #include <linux/slab.h>
43 #include <linux/pagemap.h>
45 #include <linux/math64.h>
46 #include <linux/seq_file.h>
47 #include <linux/namei.h>
49 #include <obd_support.h>
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <obd_class.h>
53 #include <lustre_lmv.h>
54 #include <lprocfs_status.h>
55 #include <cl_object.h>
56 #include <lustre_fid.h>
57 #include <uapi/linux/lustre/lustre_ioctl.h>
58 #include <lustre_kernelcomm.h>
59 #include "lmv_internal.h"
61 static int lmv_check_connect(struct obd_device *obd);
63 static void lmv_activate_target(struct lmv_obd *lmv,
64 struct lmv_tgt_desc *tgt,
67 if (tgt->ltd_active == activate)
70 tgt->ltd_active = activate;
71 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
73 tgt->ltd_exp->exp_obd->obd_inactive = !activate;
79 * -EINVAL : UUID can't be found in the LMV's target list
80 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
81 * -EBADF : The UUID is found, but the OBD of the wrong type (!)
83 static int lmv_set_mdc_active(struct lmv_obd *lmv,
84 const struct obd_uuid *uuid,
87 struct lmv_tgt_desc *tgt = NULL;
88 struct obd_device *obd;
93 CDEBUG(D_INFO, "Searching in lmv %p for uuid %s (activate=%d)\n",
94 lmv, uuid->uuid, activate);
96 spin_lock(&lmv->lmv_lock);
97 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
99 if (tgt == NULL || tgt->ltd_exp == NULL)
102 CDEBUG(D_INFO, "Target idx %d is %s conn %#llx\n", i,
103 tgt->ltd_uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie);
105 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
109 if (i == lmv->desc.ld_tgt_count)
110 GOTO(out_lmv_lock, rc = -EINVAL);
112 obd = class_exp2obd(tgt->ltd_exp);
114 GOTO(out_lmv_lock, rc = -ENOTCONN);
116 CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n",
117 obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd,
118 obd->obd_type->typ_name, i);
119 LASSERT(strcmp(obd->obd_type->typ_name, LUSTRE_MDC_NAME) == 0);
121 if (tgt->ltd_active == activate) {
122 CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd,
123 activate ? "" : "in");
124 GOTO(out_lmv_lock, rc);
127 CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd,
128 activate ? "" : "in");
129 lmv_activate_target(lmv, tgt, activate);
133 spin_unlock(&lmv->lmv_lock);
137 struct obd_uuid *lmv_get_uuid(struct obd_export *exp)
139 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
140 struct lmv_tgt_desc *tgt = lmv->tgts[0];
142 return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp);
145 static int lmv_notify(struct obd_device *obd, struct obd_device *watched,
146 enum obd_notify_event ev)
148 struct obd_connect_data *conn_data;
149 struct lmv_obd *lmv = &obd->u.lmv;
150 struct obd_uuid *uuid;
154 if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) {
155 CERROR("unexpected notification of %s %s!\n",
156 watched->obd_type->typ_name,
161 uuid = &watched->u.cli.cl_target_uuid;
162 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE) {
164 * Set MDC as active before notifying the observer, so the
165 * observer can use the MDC normally.
167 rc = lmv_set_mdc_active(lmv, uuid,
168 ev == OBD_NOTIFY_ACTIVE);
170 CERROR("%sactivation of %s failed: %d\n",
171 ev == OBD_NOTIFY_ACTIVE ? "" : "de",
175 } else if (ev == OBD_NOTIFY_OCD) {
176 conn_data = &watched->u.cli.cl_import->imp_connect_data;
178 * XXX: Make sure that ocd_connect_flags from all targets are
179 * the same. Otherwise one of MDTs runs wrong version or
180 * something like this. --umka
182 obd->obd_self_export->exp_connect_data = *conn_data;
186 * Pass the notification up the chain.
188 if (obd->obd_observer)
189 rc = obd_notify(obd->obd_observer, watched, ev);
194 static int lmv_connect(const struct lu_env *env,
195 struct obd_export **pexp, struct obd_device *obd,
196 struct obd_uuid *cluuid, struct obd_connect_data *data,
199 struct lmv_obd *lmv = &obd->u.lmv;
200 struct lustre_handle conn = { 0 };
201 struct obd_export *exp;
205 rc = class_connect(&conn, obd, cluuid);
207 CERROR("class_connection() returned %d\n", rc);
211 exp = class_conn2export(&conn);
214 lmv->conn_data = *data;
215 lmv->lmv_cache = localdata;
217 lmv->lmv_tgts_kobj = kobject_create_and_add("target_obds",
218 &obd->obd_kset.kobj);
219 if (!lmv->lmv_tgts_kobj) {
220 CERROR("%s: cannot create /sys/fs/lustre/%s/%s/target_obds\n",
221 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
224 rc = lmv_check_connect(obd);
233 if (lmv->lmv_tgts_kobj)
234 kobject_put(lmv->lmv_tgts_kobj);
236 class_disconnect(exp);
241 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
244 struct obd_device *obd = exp->exp_obd;
245 struct lmv_obd *lmv = &obd->u.lmv;
251 if (lmv->max_easize < easize) {
252 lmv->max_easize = easize;
255 if (lmv->max_def_easize < def_easize) {
256 lmv->max_def_easize = def_easize;
263 if (lmv->connected == 0)
266 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
267 struct lmv_tgt_desc *tgt = lmv->tgts[i];
269 if (tgt == NULL || tgt->ltd_exp == NULL) {
270 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
273 if (!tgt->ltd_active)
276 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
278 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
279 " rc = %d\n", obd->obd_name, i, rc);
286 #define MAX_STRING_SIZE 128
288 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
290 struct lmv_obd *lmv = &obd->u.lmv;
291 struct obd_device *mdc_obd;
292 struct obd_export *mdc_exp;
293 struct lu_fld_target target;
297 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
300 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
304 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s\n",
305 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
306 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid);
308 if (!mdc_obd->obd_set_up) {
309 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
313 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &obd->obd_uuid,
314 &lmv->conn_data, lmv->lmv_cache);
316 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
321 * Init fid sequence client for this mdc and add new fld target.
323 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
327 target.ft_srv = NULL;
328 target.ft_exp = mdc_exp;
329 target.ft_idx = tgt->ltd_idx;
331 fld_client_add_target(&lmv->lmv_fld, &target);
333 rc = obd_register_observer(mdc_obd, obd);
335 obd_disconnect(mdc_exp);
336 CERROR("target %s register_observer error %d\n",
337 tgt->ltd_uuid.uuid, rc);
341 if (obd->obd_observer) {
343 * Tell the observer about the new target.
345 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
348 obd_disconnect(mdc_exp);
354 tgt->ltd_exp = mdc_exp;
355 lmv->desc.ld_active_tgt_count++;
357 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
359 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
360 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
361 atomic_read(&obd->obd_refcount));
363 lmv_statfs_check_update(obd, tgt);
365 if (lmv->lmv_tgts_kobj)
366 /* Even if we failed to create the link, that's fine */
367 rc = sysfs_create_link(lmv->lmv_tgts_kobj,
368 &mdc_obd->obd_kset.kobj,
373 static void lmv_del_target(struct lmv_obd *lmv, int index)
375 if (lmv->tgts[index] == NULL)
378 OBD_FREE_PTR(lmv->tgts[index]);
379 lmv->tgts[index] = NULL;
383 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
384 __u32 index, int gen)
386 struct obd_device *mdc_obd;
387 struct lmv_obd *lmv = &obd->u.lmv;
388 struct lmv_tgt_desc *tgt;
389 int orig_tgt_count = 0;
393 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
394 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
397 CERROR("%s: Target %s not attached: rc = %d\n",
398 obd->obd_name, uuidp->uuid, -EINVAL);
402 mutex_lock(&lmv->lmv_init_mutex);
403 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
404 tgt = lmv->tgts[index];
405 CERROR("%s: UUID %s already assigned at LMV target index %d:"
406 " rc = %d\n", obd->obd_name,
407 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
408 mutex_unlock(&lmv->lmv_init_mutex);
412 if (index >= lmv->tgts_size) {
413 /* We need to reallocate the lmv target array. */
414 struct lmv_tgt_desc **newtgts, **old = NULL;
418 while (newsize < index + 1)
419 newsize = newsize << 1;
420 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
421 if (newtgts == NULL) {
422 mutex_unlock(&lmv->lmv_init_mutex);
426 if (lmv->tgts_size) {
427 memcpy(newtgts, lmv->tgts,
428 sizeof(*newtgts) * lmv->tgts_size);
430 oldsize = lmv->tgts_size;
434 lmv->tgts_size = newsize;
437 OBD_FREE(old, sizeof(*old) * oldsize);
439 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
445 mutex_unlock(&lmv->lmv_init_mutex);
449 mutex_init(&tgt->ltd_fid_mutex);
450 tgt->ltd_idx = index;
451 tgt->ltd_uuid = *uuidp;
453 lmv->tgts[index] = tgt;
454 if (index >= lmv->desc.ld_tgt_count) {
455 orig_tgt_count = lmv->desc.ld_tgt_count;
456 lmv->desc.ld_tgt_count = index + 1;
459 if (lmv->connected == 0) {
460 /* lmv_check_connect() will connect this target. */
461 mutex_unlock(&lmv->lmv_init_mutex);
465 /* Otherwise let's connect it ourselves */
466 mutex_unlock(&lmv->lmv_init_mutex);
467 rc = lmv_connect_mdc(obd, tgt);
469 spin_lock(&lmv->lmv_lock);
470 if (lmv->desc.ld_tgt_count == index + 1)
471 lmv->desc.ld_tgt_count = orig_tgt_count;
472 memset(tgt, 0, sizeof(*tgt));
473 spin_unlock(&lmv->lmv_lock);
475 int easize = sizeof(struct lmv_stripe_md) +
476 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
477 lmv_init_ea_size(obd->obd_self_export, easize, 0);
483 static int lmv_check_connect(struct obd_device *obd)
485 struct lmv_obd *lmv = &obd->u.lmv;
486 struct lmv_tgt_desc *tgt;
495 mutex_lock(&lmv->lmv_init_mutex);
496 if (lmv->connected) {
497 mutex_unlock(&lmv->lmv_init_mutex);
501 if (lmv->desc.ld_tgt_count == 0) {
502 mutex_unlock(&lmv->lmv_init_mutex);
503 CERROR("%s: no targets configured.\n", obd->obd_name);
507 LASSERT(lmv->tgts != NULL);
509 if (lmv->tgts[0] == NULL) {
510 mutex_unlock(&lmv->lmv_init_mutex);
511 CERROR("%s: no target configured for index 0.\n",
516 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
517 obd->obd_uuid.uuid, obd->obd_name);
519 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
523 rc = lmv_connect_mdc(obd, tgt);
529 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
530 lmv_init_ea_size(obd->obd_self_export, easize, 0);
531 mutex_unlock(&lmv->lmv_init_mutex);
542 --lmv->desc.ld_active_tgt_count;
543 rc2 = obd_disconnect(tgt->ltd_exp);
545 CERROR("LMV target %s disconnect on "
546 "MDC idx %d: error %d\n",
547 tgt->ltd_uuid.uuid, i, rc2);
552 mutex_unlock(&lmv->lmv_init_mutex);
557 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
559 struct lmv_obd *lmv = &obd->u.lmv;
560 struct obd_device *mdc_obd;
564 LASSERT(tgt != NULL);
565 LASSERT(obd != NULL);
567 mdc_obd = class_exp2obd(tgt->ltd_exp);
570 mdc_obd->obd_force = obd->obd_force;
571 mdc_obd->obd_fail = obd->obd_fail;
572 mdc_obd->obd_no_recov = obd->obd_no_recov;
574 if (lmv->lmv_tgts_kobj)
575 sysfs_remove_link(lmv->lmv_tgts_kobj,
579 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
581 CERROR("Can't finanize fids factory\n");
583 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
584 tgt->ltd_exp->exp_obd->obd_name,
585 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
587 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
588 rc = obd_disconnect(tgt->ltd_exp);
590 if (tgt->ltd_active) {
591 CERROR("Target %s disconnect error %d\n",
592 tgt->ltd_uuid.uuid, rc);
596 lmv_activate_target(lmv, tgt, 0);
601 static int lmv_disconnect(struct obd_export *exp)
603 struct obd_device *obd = class_exp2obd(exp);
604 struct lmv_obd *lmv = &obd->u.lmv;
612 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
613 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
616 lmv_disconnect_mdc(obd, lmv->tgts[i]);
619 if (lmv->lmv_tgts_kobj)
620 kobject_put(lmv->lmv_tgts_kobj);
624 * This is the case when no real connection is established by
625 * lmv_check_connect().
628 class_export_put(exp);
629 rc = class_disconnect(exp);
635 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
638 struct obd_device *obddev = class_exp2obd(exp);
639 struct lmv_obd *lmv = &obddev->u.lmv;
640 struct getinfo_fid2path *gf;
641 struct lmv_tgt_desc *tgt;
642 struct getinfo_fid2path *remote_gf = NULL;
643 struct lu_fid root_fid;
644 int remote_gf_size = 0;
648 tgt = lmv_find_target(lmv, &gf->gf_fid);
650 RETURN(PTR_ERR(tgt));
652 root_fid = *gf->gf_u.gf_root_fid;
653 LASSERT(fid_is_sane(&root_fid));
656 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
657 if (rc != 0 && rc != -EREMOTE)
658 GOTO(out_fid2path, rc);
660 /* If remote_gf != NULL, it means just building the
661 * path on the remote MDT, copy this path segement to gf */
662 if (remote_gf != NULL) {
663 struct getinfo_fid2path *ori_gf;
667 ori_gf = (struct getinfo_fid2path *)karg;
668 if (strlen(ori_gf->gf_u.gf_path) + 1 +
669 strlen(gf->gf_u.gf_path) + 1 > ori_gf->gf_pathlen)
670 GOTO(out_fid2path, rc = -EOVERFLOW);
672 ptr = ori_gf->gf_u.gf_path;
674 len = strlen(gf->gf_u.gf_path);
675 /* move the current path to the right to release space
676 * for closer-to-root part */
677 memmove(ptr + len + 1, ptr, strlen(ori_gf->gf_u.gf_path));
678 memcpy(ptr, gf->gf_u.gf_path, len);
682 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
683 tgt->ltd_exp->exp_obd->obd_name,
684 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
688 GOTO(out_fid2path, rc);
690 /* sigh, has to go to another MDT to do path building further */
691 if (remote_gf == NULL) {
692 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
693 OBD_ALLOC(remote_gf, remote_gf_size);
694 if (remote_gf == NULL)
695 GOTO(out_fid2path, rc = -ENOMEM);
696 remote_gf->gf_pathlen = PATH_MAX;
699 if (!fid_is_sane(&gf->gf_fid)) {
700 CERROR("%s: invalid FID "DFID": rc = %d\n",
701 tgt->ltd_exp->exp_obd->obd_name,
702 PFID(&gf->gf_fid), -EINVAL);
703 GOTO(out_fid2path, rc = -EINVAL);
706 tgt = lmv_find_target(lmv, &gf->gf_fid);
708 GOTO(out_fid2path, rc = -EINVAL);
710 remote_gf->gf_fid = gf->gf_fid;
711 remote_gf->gf_recno = -1;
712 remote_gf->gf_linkno = -1;
713 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
714 *remote_gf->gf_u.gf_root_fid = root_fid;
716 goto repeat_fid2path;
719 if (remote_gf != NULL)
720 OBD_FREE(remote_gf, remote_gf_size);
724 static int lmv_hsm_req_count(struct lmv_obd *lmv,
725 const struct hsm_user_request *hur,
726 const struct lmv_tgt_desc *tgt_mds)
730 struct lmv_tgt_desc *curr_tgt;
732 /* count how many requests must be sent to the given target */
733 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
734 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
735 if (IS_ERR(curr_tgt))
736 RETURN(PTR_ERR(curr_tgt));
737 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
743 static int lmv_hsm_req_build(struct lmv_obd *lmv,
744 struct hsm_user_request *hur_in,
745 const struct lmv_tgt_desc *tgt_mds,
746 struct hsm_user_request *hur_out)
749 struct lmv_tgt_desc *curr_tgt;
751 /* build the hsm_user_request for the given target */
752 hur_out->hur_request = hur_in->hur_request;
754 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
755 curr_tgt = lmv_find_target(lmv,
756 &hur_in->hur_user_item[i].hui_fid);
757 if (IS_ERR(curr_tgt))
758 RETURN(PTR_ERR(curr_tgt));
759 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
760 hur_out->hur_user_item[nr_out] =
761 hur_in->hur_user_item[i];
765 hur_out->hur_request.hr_itemcount = nr_out;
766 memcpy(hur_data(hur_out), hur_data(hur_in),
767 hur_in->hur_request.hr_data_len);
772 static int lmv_hsm_ct_unregister(struct obd_device *obd, unsigned int cmd,
773 int len, struct lustre_kernelcomm *lk,
776 struct lmv_obd *lmv = &obd->u.lmv;
781 /* unregister request (call from llapi_hsm_copytool_fini) */
782 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
783 struct lmv_tgt_desc *tgt = lmv->tgts[i];
785 if (tgt == NULL || tgt->ltd_exp == NULL)
787 /* best effort: try to clean as much as possible
788 * (continue on error) */
789 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
792 /* Whatever the result, remove copytool from kuc groups.
793 * Unreached coordinators will get EPIPE on next requests
794 * and will unregister automatically.
796 rc = libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
801 static int lmv_hsm_ct_register(struct obd_device *obd, unsigned int cmd,
802 int len, struct lustre_kernelcomm *lk,
805 struct lmv_obd *lmv = &obd->u.lmv;
809 bool any_set = false;
810 struct kkuc_ct_data *kcd;
815 filp = fget(lk->lk_wfd);
819 if (lk->lk_flags & LK_FLG_DATANR)
820 kcd_size = offsetof(struct kkuc_ct_data,
821 kcd_archives[lk->lk_data_count]);
823 kcd_size = sizeof(*kcd);
825 OBD_ALLOC(kcd, kcd_size);
827 GOTO(err_fput, rc = -ENOMEM);
829 kcd->kcd_nr_archives = lk->lk_data_count;
830 if (lk->lk_flags & LK_FLG_DATANR) {
831 kcd->kcd_magic = KKUC_CT_DATA_ARRAY_MAGIC;
832 if (lk->lk_data_count > 0)
833 memcpy(kcd->kcd_archives, lk->lk_data,
834 sizeof(*kcd->kcd_archives) * lk->lk_data_count);
836 kcd->kcd_magic = KKUC_CT_DATA_BITMAP_MAGIC;
839 rc = libcfs_kkuc_group_add(filp, &obd->obd_uuid, lk->lk_uid,
840 lk->lk_group, kcd, kcd_size);
841 OBD_FREE(kcd, kcd_size);
845 /* All or nothing: try to register to all MDS.
846 * In case of failure, unregister from previous MDS,
847 * except if it because of inactive target. */
848 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
849 struct lmv_tgt_desc *tgt = lmv->tgts[i];
851 if (tgt == NULL || tgt->ltd_exp == NULL)
854 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
856 if (tgt->ltd_active) {
857 /* permanent error */
858 CERROR("%s: iocontrol MDC %s on MDT"
859 " idx %d cmd %x: err = %d\n",
860 lmv2obd_dev(lmv)->obd_name,
861 tgt->ltd_uuid.uuid, i, cmd, err);
863 lk->lk_flags |= LK_FLG_STOP;
864 /* unregister from previous MDS */
865 for (j = 0; j < i; j++) {
867 if (tgt == NULL || tgt->ltd_exp == NULL)
869 obd_iocontrol(cmd, tgt->ltd_exp, len,
872 GOTO(err_kkuc_rem, rc);
874 /* else: transient error.
875 * kuc will register to the missing MDT
883 /* no registration done: return error */
884 GOTO(err_kkuc_rem, rc = -ENOTCONN);
889 libcfs_kkuc_group_rem(&obd->obd_uuid, lk->lk_uid, lk->lk_group);
899 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
900 int len, void *karg, void __user *uarg)
902 struct obd_device *obddev = class_exp2obd(exp);
903 struct lmv_obd *lmv = &obddev->u.lmv;
904 struct lmv_tgt_desc *tgt = NULL;
908 __u32 count = lmv->desc.ld_tgt_count;
915 case IOC_OBD_STATFS: {
916 struct obd_ioctl_data *data = karg;
917 struct obd_device *mdc_obd;
918 struct obd_statfs stat_buf = {0};
921 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
922 if ((index >= count))
925 tgt = lmv->tgts[index];
926 if (tgt == NULL || !tgt->ltd_active)
929 mdc_obd = class_exp2obd(tgt->ltd_exp);
934 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
935 min((int) data->ioc_plen2,
936 (int) sizeof(struct obd_uuid))))
939 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
940 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
944 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
945 min((int) data->ioc_plen1,
946 (int) sizeof(stat_buf))))
950 case OBD_IOC_QUOTACTL: {
951 struct if_quotactl *qctl = karg;
952 struct obd_quotactl *oqctl;
954 if (qctl->qc_valid == QC_MDTIDX) {
955 if (count <= qctl->qc_idx)
958 tgt = lmv->tgts[qctl->qc_idx];
959 if (tgt == NULL || tgt->ltd_exp == NULL)
961 } else if (qctl->qc_valid == QC_UUID) {
962 for (i = 0; i < count; i++) {
966 if (!obd_uuid_equals(&tgt->ltd_uuid,
970 if (tgt->ltd_exp == NULL)
982 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
983 OBD_ALLOC_PTR(oqctl);
987 QCTL_COPY(oqctl, qctl);
988 rc = obd_quotactl(tgt->ltd_exp, oqctl);
990 QCTL_COPY(qctl, oqctl);
991 qctl->qc_valid = QC_MDTIDX;
992 qctl->obd_uuid = tgt->ltd_uuid;
997 case LL_IOC_GET_CONNECT_FLAGS: {
999 if (tgt == NULL || tgt->ltd_exp == NULL)
1001 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1004 case LL_IOC_FID2MDTIDX: {
1005 struct lu_fid *fid = karg;
1008 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1012 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1013 * point to user space memory for FID2MDTIDX. */
1014 *(__u32 *)uarg = mdt_index;
1017 case OBD_IOC_FID2PATH: {
1018 rc = lmv_fid2path(exp, len, karg, uarg);
1021 case LL_IOC_HSM_STATE_GET:
1022 case LL_IOC_HSM_STATE_SET:
1023 case LL_IOC_HSM_ACTION: {
1024 struct md_op_data *op_data = karg;
1026 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1028 RETURN(PTR_ERR(tgt));
1030 if (tgt->ltd_exp == NULL)
1033 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1036 case LL_IOC_HSM_PROGRESS: {
1037 const struct hsm_progress_kernel *hpk = karg;
1039 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1041 RETURN(PTR_ERR(tgt));
1042 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1045 case LL_IOC_HSM_REQUEST: {
1046 struct hsm_user_request *hur = karg;
1047 unsigned int reqcount = hur->hur_request.hr_itemcount;
1052 /* if the request is about a single fid
1053 * or if there is a single MDS, no need to split
1055 if (reqcount == 1 || count == 1) {
1056 tgt = lmv_find_target(lmv,
1057 &hur->hur_user_item[0].hui_fid);
1059 RETURN(PTR_ERR(tgt));
1060 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1062 /* split fid list to their respective MDS */
1063 for (i = 0; i < count; i++) {
1066 struct hsm_user_request *req;
1069 if (tgt == NULL || tgt->ltd_exp == NULL)
1072 nr = lmv_hsm_req_count(lmv, hur, tgt);
1075 if (nr == 0) /* nothing for this MDS */
1078 /* build a request with fids for this MDS */
1079 reqlen = offsetof(typeof(*hur),
1081 + hur->hur_request.hr_data_len;
1082 OBD_ALLOC_LARGE(req, reqlen);
1085 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1087 GOTO(hsm_req_err, rc1);
1088 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1091 if (rc1 != 0 && rc == 0)
1093 OBD_FREE_LARGE(req, reqlen);
1098 case LL_IOC_LOV_SWAP_LAYOUTS: {
1099 struct md_op_data *op_data = karg;
1100 struct lmv_tgt_desc *tgt1, *tgt2;
1102 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1104 RETURN(PTR_ERR(tgt1));
1106 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1108 RETURN(PTR_ERR(tgt2));
1110 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1113 /* only files on same MDT can have their layouts swapped */
1114 if (tgt1->ltd_idx != tgt2->ltd_idx)
1117 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1120 case LL_IOC_HSM_CT_START: {
1121 struct lustre_kernelcomm *lk = karg;
1122 if (lk->lk_flags & LK_FLG_STOP)
1123 rc = lmv_hsm_ct_unregister(obddev, cmd, len, lk, uarg);
1125 rc = lmv_hsm_ct_register(obddev, cmd, len, lk, uarg);
1129 for (i = 0; i < count; i++) {
1130 struct obd_device *mdc_obd;
1134 if (tgt == NULL || tgt->ltd_exp == NULL)
1136 /* ll_umount_begin() sets force flag but for lmv, not
1137 * mdc. Let's pass it through */
1138 mdc_obd = class_exp2obd(tgt->ltd_exp);
1139 mdc_obd->obd_force = obddev->obd_force;
1140 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1142 if (tgt->ltd_active) {
1143 CERROR("error: iocontrol MDC %s on MDT"
1144 " idx %d cmd %x: err = %d\n",
1145 tgt->ltd_uuid.uuid, i, cmd, err);
1159 * This is _inode_ placement policy function (not name).
1161 static u32 lmv_placement_policy(struct obd_device *obd,
1162 struct md_op_data *op_data)
1164 struct lmv_obd *lmv = &obd->u.lmv;
1165 struct lmv_user_md *lum;
1170 if (lmv->desc.ld_tgt_count == 1)
1173 lum = op_data->op_data;
1176 * 1. See if the stripe offset is specified by lum.
1177 * 2. If parent has default LMV, and its hash type is "space", choose
1178 * MDT with QoS. (see lmv_locate_tgt_qos()).
1179 * 3. Then check if default LMV stripe offset is not -1.
1180 * 4. Finally choose MDS by name hash if the parent
1181 * is striped directory. (see lmv_locate_tgt()).
1183 * presently explicit MDT location is not supported
1184 * for foreign dirs (as it can't be embedded into free
1185 * format LMV, like with lum_stripe_offset), so we only
1186 * rely on default stripe offset or then name hashing.
1188 if (op_data->op_cli_flags & CLI_SET_MEA && lum != NULL &&
1189 le32_to_cpu(lum->lum_magic != LMV_MAGIC_FOREIGN) &&
1190 le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1191 mdt = le32_to_cpu(lum->lum_stripe_offset);
1192 } else if (op_data->op_code == LUSTRE_OPC_MKDIR &&
1193 !lmv_dir_striped(op_data->op_mea1) &&
1194 lmv_dir_space_hashed(op_data->op_default_mea1)) {
1195 mdt = op_data->op_mds;
1196 } else if (op_data->op_code == LUSTRE_OPC_MKDIR &&
1197 op_data->op_default_mea1 &&
1198 op_data->op_default_mea1->lsm_md_master_mdt_index !=
1200 mdt = op_data->op_default_mea1->lsm_md_master_mdt_index;
1201 op_data->op_mds = mdt;
1203 mdt = op_data->op_mds;
1209 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1211 struct lmv_tgt_desc *tgt;
1215 tgt = lmv_get_target(lmv, mds, NULL);
1217 RETURN(PTR_ERR(tgt));
1220 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1221 * on server that seq in new allocated fid is not yet known.
1223 mutex_lock(&tgt->ltd_fid_mutex);
1225 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1226 GOTO(out, rc = -ENODEV);
1229 * Asking underlying tgt layer to allocate new fid.
1231 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1233 LASSERT(fid_is_sane(fid));
1239 mutex_unlock(&tgt->ltd_fid_mutex);
1243 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1244 struct lu_fid *fid, struct md_op_data *op_data)
1246 struct obd_device *obd = class_exp2obd(exp);
1247 struct lmv_obd *lmv = &obd->u.lmv;
1253 LASSERT(op_data != NULL);
1254 LASSERT(fid != NULL);
1256 mds = lmv_placement_policy(obd, op_data);
1258 rc = __lmv_fid_alloc(lmv, fid, mds);
1260 CERROR("Can't alloc new fid, rc %d\n", rc);
1265 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1267 struct lmv_obd *lmv = &obd->u.lmv;
1268 struct lmv_desc *desc;
1272 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1273 CERROR("LMV setup requires a descriptor\n");
1277 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1278 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1279 CERROR("Lmv descriptor size wrong: %d > %d\n",
1280 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1284 lmv->tgts_size = 32U;
1285 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1286 if (lmv->tgts == NULL)
1289 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1290 lmv->desc.ld_tgt_count = 0;
1291 lmv->desc.ld_active_tgt_count = 0;
1292 lmv->desc.ld_qos_maxage = 60;
1293 lmv->max_def_easize = 0;
1294 lmv->max_easize = 0;
1296 spin_lock_init(&lmv->lmv_lock);
1297 mutex_init(&lmv->lmv_init_mutex);
1299 rc = lmv_tunables_init(obd);
1301 CWARN("%s: error adding LMV sysfs/debugfs files: rc = %d\n",
1304 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1305 LUSTRE_CLI_FLD_HASH_DHT);
1307 CERROR("Can't init FLD, err %d\n", rc);
1317 static int lmv_cleanup(struct obd_device *obd)
1319 struct lmv_obd *lmv = &obd->u.lmv;
1322 fld_client_fini(&lmv->lmv_fld);
1323 if (lmv->tgts != NULL) {
1325 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1326 if (lmv->tgts[i] == NULL)
1328 lmv_del_target(lmv, i);
1330 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1336 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1338 struct lustre_cfg *lcfg = buf;
1339 struct obd_uuid obd_uuid;
1345 switch (lcfg->lcfg_command) {
1347 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1348 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1349 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1350 GOTO(out, rc = -EINVAL);
1352 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1354 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1355 GOTO(out, rc = -EINVAL);
1356 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1357 GOTO(out, rc = -EINVAL);
1358 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1361 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1362 GOTO(out, rc = -EINVAL);
1368 static int lmv_select_statfs_mdt(struct lmv_obd *lmv, __u32 flags)
1372 if (flags & OBD_STATFS_FOR_MDT0)
1375 if (lmv->lmv_statfs_start || lmv->desc.ld_tgt_count == 1)
1376 return lmv->lmv_statfs_start;
1378 /* choose initial MDT for this client */
1380 struct lnet_process_id lnet_id;
1381 if (LNetGetId(i, &lnet_id) == -ENOENT)
1384 if (LNET_NETTYP(LNET_NIDNET(lnet_id.nid)) != LOLND) {
1385 /* We dont need a full 64-bit modulus, just enough
1386 * to distribute the requests across MDTs evenly.
1388 lmv->lmv_statfs_start =
1389 (u32)lnet_id.nid % lmv->desc.ld_tgt_count;
1394 return lmv->lmv_statfs_start;
1397 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1398 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
1400 struct obd_device *obd = class_exp2obd(exp);
1401 struct lmv_obd *lmv = &obd->u.lmv;
1402 struct obd_statfs *temp;
1407 OBD_ALLOC(temp, sizeof(*temp));
1411 /* distribute statfs among MDTs */
1412 idx = lmv_select_statfs_mdt(lmv, flags);
1414 for (i = 0; i < lmv->desc.ld_tgt_count; i++, idx++) {
1415 idx = idx % lmv->desc.ld_tgt_count;
1416 if (lmv->tgts[idx] == NULL || lmv->tgts[idx]->ltd_exp == NULL)
1419 rc = obd_statfs(env, lmv->tgts[idx]->ltd_exp, temp,
1422 CERROR("%s: can't stat MDS #%d: rc = %d\n",
1423 lmv->tgts[idx]->ltd_exp->exp_obd->obd_name, i,
1425 GOTO(out_free_temp, rc);
1428 if (temp->os_state & OS_STATE_SUM ||
1429 flags == OBD_STATFS_FOR_MDT0) {
1430 /* reset to the last aggregated values
1431 * and don't sum with non-aggrated data */
1432 /* If the statfs is from mount, it needs to retrieve
1433 * necessary information from MDT0. i.e. mount does
1434 * not need the merged osfs from all of MDT. Also
1435 * clients can be mounted as long as MDT0 is in
1444 osfs->os_bavail += temp->os_bavail;
1445 osfs->os_blocks += temp->os_blocks;
1446 osfs->os_ffree += temp->os_ffree;
1447 osfs->os_files += temp->os_files;
1448 osfs->os_granted += temp->os_granted;
1454 OBD_FREE(temp, sizeof(*temp));
1458 static int lmv_statfs_update(void *cookie, int rc)
1460 struct obd_info *oinfo = cookie;
1461 struct obd_device *obd = oinfo->oi_obd;
1462 struct lmv_obd *lmv = &obd->u.lmv;
1463 struct lmv_tgt_desc *tgt = oinfo->oi_tgt;
1464 struct obd_statfs *osfs = oinfo->oi_osfs;
1467 * NB: don't deactivate TGT upon error, because we may not trigger async
1468 * statfs any longer, then there is no chance to activate TGT.
1471 spin_lock(&lmv->lmv_lock);
1472 tgt->ltd_statfs = *osfs;
1473 tgt->ltd_statfs_age = ktime_get_seconds();
1474 spin_unlock(&lmv->lmv_lock);
1480 /* update tgt statfs async if it's ld_qos_maxage old */
1481 int lmv_statfs_check_update(struct obd_device *obd, struct lmv_tgt_desc *tgt)
1483 struct obd_info oinfo = {
1486 .oi_cb_up = lmv_statfs_update,
1490 if (ktime_get_seconds() - tgt->ltd_statfs_age <
1491 obd->u.lmv.desc.ld_qos_maxage)
1494 rc = obd_statfs_async(tgt->ltd_exp, &oinfo, 0, NULL);
1499 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1502 struct obd_device *obd = exp->exp_obd;
1503 struct lmv_obd *lmv = &obd->u.lmv;
1507 rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1511 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1512 u64 obd_md_valid, const char *name, size_t buf_size,
1513 struct ptlrpc_request **req)
1515 struct obd_device *obd = exp->exp_obd;
1516 struct lmv_obd *lmv = &obd->u.lmv;
1517 struct lmv_tgt_desc *tgt;
1521 tgt = lmv_find_target(lmv, fid);
1523 RETURN(PTR_ERR(tgt));
1525 rc = md_getxattr(tgt->ltd_exp, fid, obd_md_valid, name, buf_size, req);
1530 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1531 u64 obd_md_valid, const char *name,
1532 const void *value, size_t value_size,
1533 unsigned int xattr_flags, u32 suppgid,
1534 struct ptlrpc_request **req)
1536 struct obd_device *obd = exp->exp_obd;
1537 struct lmv_obd *lmv = &obd->u.lmv;
1538 struct lmv_tgt_desc *tgt;
1542 tgt = lmv_find_target(lmv, fid);
1544 RETURN(PTR_ERR(tgt));
1546 rc = md_setxattr(tgt->ltd_exp, fid, obd_md_valid, name,
1547 value, value_size, xattr_flags, suppgid, req);
1552 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1553 struct ptlrpc_request **request)
1555 struct obd_device *obd = exp->exp_obd;
1556 struct lmv_obd *lmv = &obd->u.lmv;
1557 struct lmv_tgt_desc *tgt;
1561 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1563 RETURN(PTR_ERR(tgt));
1565 if (op_data->op_flags & MF_GET_MDT_IDX) {
1566 op_data->op_mds = tgt->ltd_idx;
1570 rc = md_getattr(tgt->ltd_exp, op_data, request);
1575 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1577 struct obd_device *obd = exp->exp_obd;
1578 struct lmv_obd *lmv = &obd->u.lmv;
1582 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1585 * With DNE every object can have two locks in different namespaces:
1586 * lookup lock in space of MDT storing direntry and update/open lock in
1587 * space of MDT storing inode.
1589 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1590 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1592 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1598 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1599 struct md_open_data *mod, struct ptlrpc_request **request)
1601 struct obd_device *obd = exp->exp_obd;
1602 struct lmv_obd *lmv = &obd->u.lmv;
1603 struct lmv_tgt_desc *tgt;
1607 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1609 RETURN(PTR_ERR(tgt));
1611 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1612 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1616 static struct lmv_tgt_desc *lmv_locate_tgt_qos(struct lmv_obd *lmv, __u32 *mdt)
1618 static unsigned int rr_index;
1620 /* locate MDT round-robin is the first step */
1621 *mdt = rr_index % lmv->tgts_size;
1624 return lmv->tgts[*mdt];
1627 static struct lmv_tgt_desc *
1628 lmv_locate_tgt_by_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1629 const char *name, int namelen, struct lu_fid *fid,
1630 __u32 *mds, bool post_migrate)
1632 struct lmv_tgt_desc *tgt;
1633 const struct lmv_oinfo *oinfo;
1635 if (!lmv_dir_striped(lsm) || !namelen) {
1636 tgt = lmv_find_target(lmv, fid);
1640 *mds = tgt->ltd_idx;
1644 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1645 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1646 return ERR_PTR(-EBADF);
1647 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1649 oinfo = lsm_name_to_stripe_info(lsm, name, namelen,
1652 return ERR_CAST(oinfo);
1655 *fid = oinfo->lmo_fid;
1656 *mds = oinfo->lmo_mds;
1657 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1659 CDEBUG(D_INODE, "locate MDT %u parent "DFID"\n", *mds, PFID(fid));
1665 * Locate MDT of op_data->op_fid1
1667 * For striped directory, it will locate the stripe by name hash, if hash_type
1668 * is unknown, it will return the stripe specified by 'op_data->op_stripe_index'
1669 * which is set outside, and if dir is migrating, 'op_data->op_post_migrate'
1670 * indicates whether old or new layout is used to locate.
1672 * For plain direcotry, normally it will locate MDT by FID, but if this
1673 * directory has default LMV, and its hash type is "space", locate MDT with QoS.
1675 * \param[in] lmv LMV device
1676 * \param[in] op_data client MD stack parameters, name, namelen
1679 * retval pointer to the lmv_tgt_desc if succeed.
1680 * ERR_PTR(errno) if failed.
1682 struct lmv_tgt_desc *
1683 lmv_locate_tgt(struct lmv_obd *lmv, struct md_op_data *op_data)
1685 struct lmv_stripe_md *lsm = op_data->op_mea1;
1686 struct lmv_oinfo *oinfo;
1687 struct lmv_tgt_desc *tgt;
1689 if (lmv_dir_foreign(lsm))
1690 return ERR_PTR(-ENODATA);
1692 /* During creating VOLATILE file, it should honor the mdt
1693 * index if the file under striped dir is being restored, see
1695 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1696 (int)op_data->op_mds != -1) {
1697 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1701 if (lmv_dir_striped(lsm)) {
1704 /* refill the right parent fid */
1705 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1706 oinfo = &lsm->lsm_md_oinfo[i];
1707 if (oinfo->lmo_mds == op_data->op_mds) {
1708 op_data->op_fid1 = oinfo->lmo_fid;
1713 if (i == lsm->lsm_md_stripe_count)
1714 op_data->op_fid1 = lsm->lsm_md_oinfo[0].lmo_fid;
1716 } else if (lmv_dir_bad_hash(lsm)) {
1717 LASSERT(op_data->op_stripe_index < lsm->lsm_md_stripe_count);
1718 oinfo = &lsm->lsm_md_oinfo[op_data->op_stripe_index];
1720 op_data->op_fid1 = oinfo->lmo_fid;
1721 op_data->op_mds = oinfo->lmo_mds;
1722 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1723 } else if (op_data->op_code == LUSTRE_OPC_MKDIR &&
1724 lmv_dir_space_hashed(op_data->op_default_mea1) &&
1725 !lmv_dir_striped(lsm)) {
1726 tgt = lmv_locate_tgt_qos(lmv, &op_data->op_mds);
1728 * only update statfs when mkdir under dir with "space" hash,
1729 * this means the cached statfs may be stale, and current mkdir
1730 * may not follow QoS accurately, but it's not serious, and it
1731 * avoids periodic statfs when client doesn't mkdir under
1732 * "space" hashed directories.
1735 struct obd_device *obd;
1737 obd = container_of(lmv, struct obd_device, u.lmv);
1738 lmv_statfs_check_update(obd, tgt);
1741 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea1,
1742 op_data->op_name, op_data->op_namelen,
1743 &op_data->op_fid1, &op_data->op_mds,
1744 op_data->op_post_migrate);
1750 /* Locate MDT of op_data->op_fid2 for link/rename */
1751 static struct lmv_tgt_desc *
1752 lmv_locate_tgt2(struct lmv_obd *lmv, struct md_op_data *op_data)
1754 struct lmv_tgt_desc *tgt;
1757 LASSERT(op_data->op_name);
1758 if (lmv_dir_migrating(op_data->op_mea2)) {
1759 struct lu_fid fid1 = op_data->op_fid1;
1760 struct lmv_stripe_md *lsm1 = op_data->op_mea1;
1761 struct ptlrpc_request *request = NULL;
1764 * avoid creating new file under old layout of migrating
1765 * directory, check it here.
1767 tgt = lmv_locate_tgt_by_name(lmv, op_data->op_mea2,
1768 op_data->op_name, op_data->op_namelen,
1769 &op_data->op_fid2, &op_data->op_mds, false);
1773 op_data->op_fid1 = op_data->op_fid2;
1774 op_data->op_mea1 = op_data->op_mea2;
1775 rc = md_getattr_name(tgt->ltd_exp, op_data, &request);
1776 op_data->op_fid1 = fid1;
1777 op_data->op_mea1 = lsm1;
1779 ptlrpc_req_finished(request);
1780 RETURN(ERR_PTR(-EEXIST));
1784 RETURN(ERR_PTR(rc));
1787 return lmv_locate_tgt_by_name(lmv, op_data->op_mea2, op_data->op_name,
1788 op_data->op_namelen, &op_data->op_fid2,
1789 &op_data->op_mds, true);
1792 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1793 const void *data, size_t datalen, umode_t mode, uid_t uid,
1794 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1795 struct ptlrpc_request **request)
1797 struct obd_device *obd = exp->exp_obd;
1798 struct lmv_obd *lmv = &obd->u.lmv;
1799 struct lmv_tgt_desc *tgt;
1804 if (!lmv->desc.ld_active_tgt_count)
1807 if (lmv_dir_bad_hash(op_data->op_mea1))
1810 if (lmv_dir_migrating(op_data->op_mea1)) {
1812 * if parent is migrating, create() needs to lookup existing
1813 * name, to avoid creating new file under old layout of
1814 * migrating directory, check old layout here.
1816 tgt = lmv_locate_tgt(lmv, op_data);
1818 RETURN(PTR_ERR(tgt));
1820 rc = md_getattr_name(tgt->ltd_exp, op_data, request);
1822 ptlrpc_req_finished(*request);
1830 op_data->op_post_migrate = true;
1833 tgt = lmv_locate_tgt(lmv, op_data);
1835 RETURN(PTR_ERR(tgt));
1837 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1838 (int)op_data->op_namelen, op_data->op_name,
1839 PFID(&op_data->op_fid1), op_data->op_mds);
1841 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1845 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1846 /* Send the create request to the MDT where the object
1847 * will be located */
1848 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1850 RETURN(PTR_ERR(tgt));
1852 op_data->op_mds = tgt->ltd_idx;
1855 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1856 PFID(&op_data->op_fid2), op_data->op_mds);
1858 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1859 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1860 cap_effective, rdev, request);
1862 if (*request == NULL)
1864 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1870 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1871 const union ldlm_policy_data *policy, struct md_op_data *op_data,
1872 struct lustre_handle *lockh, __u64 extra_lock_flags)
1874 struct obd_device *obd = exp->exp_obd;
1875 struct lmv_obd *lmv = &obd->u.lmv;
1876 struct lmv_tgt_desc *tgt;
1880 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1882 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1884 RETURN(PTR_ERR(tgt));
1886 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1887 PFID(&op_data->op_fid1), tgt->ltd_idx);
1889 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1896 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1897 struct ptlrpc_request **preq)
1899 struct obd_device *obd = exp->exp_obd;
1900 struct lmv_obd *lmv = &obd->u.lmv;
1901 struct lmv_tgt_desc *tgt;
1902 struct mdt_body *body;
1908 tgt = lmv_locate_tgt(lmv, op_data);
1910 RETURN(PTR_ERR(tgt));
1912 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1913 (int)op_data->op_namelen, op_data->op_name,
1914 PFID(&op_data->op_fid1), tgt->ltd_idx);
1916 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1917 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
1918 ptlrpc_req_finished(*preq);
1926 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1927 LASSERT(body != NULL);
1929 if (body->mbo_valid & OBD_MD_MDS) {
1930 op_data->op_fid1 = body->mbo_fid1;
1931 op_data->op_valid |= OBD_MD_FLCROSSREF;
1932 op_data->op_namelen = 0;
1933 op_data->op_name = NULL;
1935 ptlrpc_req_finished(*preq);
1944 #define md_op_data_fid(op_data, fl) \
1945 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1946 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1947 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1948 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1951 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1952 struct md_op_data *op_data, __u32 op_tgt,
1953 enum ldlm_mode mode, int bits, int flag)
1955 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1956 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1957 union ldlm_policy_data policy = { { 0 } };
1961 if (!fid_is_sane(fid))
1965 tgt = lmv_find_target(lmv, fid);
1967 RETURN(PTR_ERR(tgt));
1970 if (tgt->ltd_idx != op_tgt) {
1971 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1972 policy.l_inodebits.bits = bits;
1973 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1974 mode, LCF_ASYNC, NULL);
1977 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1979 op_data->op_flags |= flag;
1987 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1990 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
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;
1999 LASSERT(op_data->op_namelen != 0);
2001 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
2002 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
2003 op_data->op_name, PFID(&op_data->op_fid1));
2005 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2006 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2007 op_data->op_cap = cfs_curproc_cap_pack();
2009 tgt = lmv_locate_tgt2(lmv, op_data);
2011 RETURN(PTR_ERR(tgt));
2014 * Cancel UPDATE lock on child (fid1).
2016 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2017 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2018 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2022 rc = md_link(tgt->ltd_exp, op_data, request);
2027 static int lmv_migrate(struct obd_export *exp, struct md_op_data *op_data,
2028 const char *name, size_t namelen,
2029 struct ptlrpc_request **request)
2031 struct obd_device *obd = exp->exp_obd;
2032 struct lmv_obd *lmv = &obd->u.lmv;
2033 struct lmv_stripe_md *lsm = op_data->op_mea1;
2034 struct lmv_tgt_desc *parent_tgt;
2035 struct lmv_tgt_desc *sp_tgt;
2036 struct lmv_tgt_desc *tp_tgt = NULL;
2037 struct lmv_tgt_desc *child_tgt;
2038 struct lmv_tgt_desc *tgt;
2039 struct lu_fid target_fid;
2044 LASSERT(op_data->op_cli_flags & CLI_MIGRATE);
2046 CDEBUG(D_INODE, "MIGRATE "DFID"/%.*s\n",
2047 PFID(&op_data->op_fid1), (int)namelen, name);
2049 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2050 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2051 op_data->op_cap = cfs_curproc_cap_pack();
2053 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2054 if (IS_ERR(parent_tgt))
2055 RETURN(PTR_ERR(parent_tgt));
2057 if (lmv_dir_striped(lsm)) {
2058 __u32 hash_type = lsm->lsm_md_hash_type;
2059 __u32 stripe_count = lsm->lsm_md_stripe_count;
2062 * old stripes are appended after new stripes for migrating
2065 if (lmv_dir_migrating(lsm)) {
2066 hash_type = lsm->lsm_md_migrate_hash;
2067 stripe_count -= lsm->lsm_md_migrate_offset;
2070 rc = lmv_name_to_stripe_index(hash_type, stripe_count, name,
2075 if (lmv_dir_migrating(lsm))
2076 rc += lsm->lsm_md_migrate_offset;
2078 /* save it in fid4 temporarily for early cancel */
2079 op_data->op_fid4 = lsm->lsm_md_oinfo[rc].lmo_fid;
2080 sp_tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[rc].lmo_mds,
2083 RETURN(PTR_ERR(sp_tgt));
2086 * if parent is being migrated too, fill op_fid2 with target
2087 * stripe fid, otherwise the target stripe is not created yet.
2089 if (lmv_dir_migrating(lsm)) {
2090 hash_type = lsm->lsm_md_hash_type &
2091 ~LMV_HASH_FLAG_MIGRATION;
2092 stripe_count = lsm->lsm_md_migrate_offset;
2094 rc = lmv_name_to_stripe_index(hash_type, stripe_count,
2099 op_data->op_fid2 = lsm->lsm_md_oinfo[rc].lmo_fid;
2100 tp_tgt = lmv_get_target(lmv,
2101 lsm->lsm_md_oinfo[rc].lmo_mds,
2104 RETURN(PTR_ERR(tp_tgt));
2107 sp_tgt = parent_tgt;
2110 child_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2111 if (IS_ERR(child_tgt))
2112 RETURN(PTR_ERR(child_tgt));
2114 if (!S_ISDIR(op_data->op_mode) && tp_tgt)
2115 rc = __lmv_fid_alloc(lmv, &target_fid, tp_tgt->ltd_idx);
2117 rc = lmv_fid_alloc(NULL, exp, &target_fid, op_data);
2122 * for directory, send migrate request to the MDT where the object will
2123 * be migrated to, because we can't create a striped directory remotely.
2125 * otherwise, send to the MDT where source is located because regular
2126 * file may open lease.
2128 * NB. if MDT doesn't support DIR_MIGRATE, send to source MDT too for
2129 * backward compatibility.
2131 if (S_ISDIR(op_data->op_mode) &&
2132 (exp_connect_flags2(exp) & OBD_CONNECT2_DIR_MIGRATE)) {
2133 tgt = lmv_find_target(lmv, &target_fid);
2135 RETURN(PTR_ERR(tgt));
2140 /* cancel UPDATE lock of parent master object */
2141 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx, LCK_EX,
2142 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2146 /* cancel UPDATE lock of source parent */
2147 if (sp_tgt != parent_tgt) {
2149 * migrate RPC packs master object FID, because we can only pack
2150 * two FIDs in reint RPC, but MDS needs to know both source
2151 * parent and target parent, and it will obtain them from master
2152 * FID and LMV, the other FID in RPC is kept for target.
2154 * since this FID is not passed to MDC, cancel it anyway.
2156 rc = lmv_early_cancel(exp, sp_tgt, op_data, -1, LCK_EX,
2157 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID4);
2161 op_data->op_flags &= ~MF_MDC_CANCEL_FID4;
2163 op_data->op_fid4 = target_fid;
2165 /* cancel UPDATE locks of target parent */
2166 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_idx, LCK_EX,
2167 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2171 /* cancel LOOKUP lock of source if source is remote object */
2172 if (child_tgt != sp_tgt) {
2173 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_idx,
2174 LCK_EX, MDS_INODELOCK_LOOKUP,
2175 MF_MDC_CANCEL_FID3);
2180 /* cancel ELC locks of source */
2181 rc = lmv_early_cancel(exp, child_tgt, op_data, tgt->ltd_idx, LCK_EX,
2182 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2186 rc = md_rename(tgt->ltd_exp, op_data, name, namelen, NULL, 0, request);
2191 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2192 const char *old, size_t oldlen,
2193 const char *new, size_t newlen,
2194 struct ptlrpc_request **request)
2196 struct obd_device *obd = exp->exp_obd;
2197 struct lmv_obd *lmv = &obd->u.lmv;
2198 struct lmv_tgt_desc *sp_tgt;
2199 struct lmv_tgt_desc *tp_tgt = NULL;
2200 struct lmv_tgt_desc *src_tgt = NULL;
2201 struct lmv_tgt_desc *tgt;
2202 struct mdt_body *body;
2207 LASSERT(oldlen != 0);
2209 if (op_data->op_cli_flags & CLI_MIGRATE) {
2210 rc = lmv_migrate(exp, op_data, old, oldlen, request);
2214 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2215 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2216 op_data->op_cap = cfs_curproc_cap_pack();
2218 op_data->op_name = new;
2219 op_data->op_namelen = newlen;
2221 tp_tgt = lmv_locate_tgt2(lmv, op_data);
2223 RETURN(PTR_ERR(tp_tgt));
2225 /* Since the target child might be destroyed, and it might become
2226 * orphan, and we can only check orphan on the local MDT right now, so
2227 * we send rename request to the MDT where target child is located. If
2228 * target child does not exist, then it will send the request to the
2230 if (fid_is_sane(&op_data->op_fid4)) {
2231 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2233 RETURN(PTR_ERR(tgt));
2238 op_data->op_flags |= MF_MDC_CANCEL_FID4;
2240 /* cancel UPDATE locks of target parent */
2241 rc = lmv_early_cancel(exp, tp_tgt, op_data, tgt->ltd_idx, LCK_EX,
2242 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2);
2246 if (fid_is_sane(&op_data->op_fid4)) {
2247 /* cancel LOOKUP lock of target on target parent */
2248 if (tgt != tp_tgt) {
2249 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2250 tgt->ltd_idx, LCK_EX,
2251 MDS_INODELOCK_LOOKUP,
2252 MF_MDC_CANCEL_FID4);
2258 if (fid_is_sane(&op_data->op_fid3)) {
2259 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2260 if (IS_ERR(src_tgt))
2261 RETURN(PTR_ERR(src_tgt));
2263 /* cancel ELC locks of source */
2264 rc = lmv_early_cancel(exp, src_tgt, op_data, tgt->ltd_idx,
2265 LCK_EX, MDS_INODELOCK_ELC,
2266 MF_MDC_CANCEL_FID3);
2271 op_data->op_name = old;
2272 op_data->op_namelen = oldlen;
2274 sp_tgt = lmv_locate_tgt(lmv, op_data);
2276 RETURN(PTR_ERR(sp_tgt));
2278 /* cancel UPDATE locks of source parent */
2279 rc = lmv_early_cancel(exp, sp_tgt, op_data, tgt->ltd_idx, LCK_EX,
2280 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2284 if (fid_is_sane(&op_data->op_fid3)) {
2285 /* cancel LOOKUP lock of source on source parent */
2286 if (src_tgt != sp_tgt) {
2287 rc = lmv_early_cancel(exp, sp_tgt, op_data,
2288 tgt->ltd_idx, LCK_EX,
2289 MDS_INODELOCK_LOOKUP,
2290 MF_MDC_CANCEL_FID3);
2297 CDEBUG(D_INODE, "RENAME "DFID"/%.*s to "DFID"/%.*s\n",
2298 PFID(&op_data->op_fid1), (int)oldlen, old,
2299 PFID(&op_data->op_fid2), (int)newlen, new);
2301 rc = md_rename(tgt->ltd_exp, op_data, old, oldlen, new, newlen,
2303 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2304 ptlrpc_req_finished(*request);
2309 if (rc && rc != -EXDEV)
2312 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2316 /* Not cross-ref case, just get out of here. */
2317 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2320 op_data->op_fid4 = body->mbo_fid1;
2322 ptlrpc_req_finished(*request);
2325 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2327 RETURN(PTR_ERR(tgt));
2329 if (fid_is_sane(&op_data->op_fid4)) {
2330 /* cancel LOOKUP lock of target on target parent */
2331 if (tgt != tp_tgt) {
2332 rc = lmv_early_cancel(exp, tp_tgt, op_data,
2333 tgt->ltd_idx, LCK_EX,
2334 MDS_INODELOCK_LOOKUP,
2335 MF_MDC_CANCEL_FID4);
2344 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2345 void *ea, size_t ealen, struct ptlrpc_request **request)
2347 struct obd_device *obd = exp->exp_obd;
2348 struct lmv_obd *lmv = &obd->u.lmv;
2349 struct lmv_tgt_desc *tgt;
2353 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x/0x%x\n",
2354 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid,
2355 op_data->op_xvalid);
2357 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2358 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2360 RETURN(PTR_ERR(tgt));
2362 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2367 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2368 struct ptlrpc_request **request)
2370 struct obd_device *obd = exp->exp_obd;
2371 struct lmv_obd *lmv = &obd->u.lmv;
2372 struct lmv_tgt_desc *tgt;
2376 tgt = lmv_find_target(lmv, fid);
2378 RETURN(PTR_ERR(tgt));
2380 rc = md_fsync(tgt->ltd_exp, fid, request);
2384 struct stripe_dirent {
2385 struct page *sd_page;
2386 struct lu_dirpage *sd_dp;
2387 struct lu_dirent *sd_ent;
2391 struct lmv_dir_ctxt {
2392 struct lmv_obd *ldc_lmv;
2393 struct md_op_data *ldc_op_data;
2394 struct md_callback *ldc_cb_op;
2397 struct stripe_dirent ldc_stripes[0];
2400 static inline void stripe_dirent_unload(struct stripe_dirent *stripe)
2402 if (stripe->sd_page) {
2403 kunmap(stripe->sd_page);
2404 put_page(stripe->sd_page);
2405 stripe->sd_page = NULL;
2406 stripe->sd_ent = NULL;
2410 static inline void put_lmv_dir_ctxt(struct lmv_dir_ctxt *ctxt)
2414 for (i = 0; i < ctxt->ldc_count; i++)
2415 stripe_dirent_unload(&ctxt->ldc_stripes[i]);
2418 /* if @ent is dummy, or . .., get next */
2419 static struct lu_dirent *stripe_dirent_get(struct lmv_dir_ctxt *ctxt,
2420 struct lu_dirent *ent,
2423 for (; ent; ent = lu_dirent_next(ent)) {
2424 /* Skip dummy entry */
2425 if (le16_to_cpu(ent->lde_namelen) == 0)
2428 /* skip . and .. for other stripes */
2430 (strncmp(ent->lde_name, ".",
2431 le16_to_cpu(ent->lde_namelen)) == 0 ||
2432 strncmp(ent->lde_name, "..",
2433 le16_to_cpu(ent->lde_namelen)) == 0))
2436 if (le64_to_cpu(ent->lde_hash) >= ctxt->ldc_hash)
2443 static struct lu_dirent *stripe_dirent_load(struct lmv_dir_ctxt *ctxt,
2444 struct stripe_dirent *stripe,
2447 struct md_op_data *op_data = ctxt->ldc_op_data;
2448 struct lmv_oinfo *oinfo;
2449 struct lu_fid fid = op_data->op_fid1;
2450 struct inode *inode = op_data->op_data;
2451 struct lmv_tgt_desc *tgt;
2452 struct lu_dirent *ent = stripe->sd_ent;
2453 __u64 hash = ctxt->ldc_hash;
2458 LASSERT(stripe == &ctxt->ldc_stripes[stripe_index]);
2462 if (stripe->sd_page) {
2463 __u64 end = le64_to_cpu(stripe->sd_dp->ldp_hash_end);
2465 /* @hash should be the last dirent hash */
2466 LASSERTF(hash <= end,
2467 "ctxt@%p stripe@%p hash %llx end %llx\n",
2468 ctxt, stripe, hash, end);
2469 /* unload last page */
2470 stripe_dirent_unload(stripe);
2472 if (end == MDS_DIR_END_OFF) {
2473 stripe->sd_eof = true;
2479 oinfo = &op_data->op_mea1->lsm_md_oinfo[stripe_index];
2480 if (!oinfo->lmo_root) {
2485 tgt = lmv_get_target(ctxt->ldc_lmv, oinfo->lmo_mds, NULL);
2491 /* op_data is shared by stripes, reset after use */
2492 op_data->op_fid1 = oinfo->lmo_fid;
2493 op_data->op_fid2 = oinfo->lmo_fid;
2494 op_data->op_data = oinfo->lmo_root;
2496 rc = md_read_page(tgt->ltd_exp, op_data, ctxt->ldc_cb_op, hash,
2499 op_data->op_fid1 = fid;
2500 op_data->op_fid2 = fid;
2501 op_data->op_data = inode;
2506 stripe->sd_dp = page_address(stripe->sd_page);
2507 ent = stripe_dirent_get(ctxt, lu_dirent_start(stripe->sd_dp),
2509 /* in case a page filled with ., .. and dummy, read next */
2512 stripe->sd_ent = ent;
2515 /* treat error as eof, so dir can be partially accessed */
2516 stripe->sd_eof = true;
2517 LCONSOLE_WARN("dir "DFID" stripe %d readdir failed: %d, "
2518 "directory is partially accessed!\n",
2519 PFID(&ctxt->ldc_op_data->op_fid1), stripe_index,
2526 static int lmv_file_resync(struct obd_export *exp, struct md_op_data *data)
2528 struct obd_device *obd = exp->exp_obd;
2529 struct lmv_obd *lmv = &obd->u.lmv;
2530 struct lmv_tgt_desc *tgt;
2534 rc = lmv_check_connect(obd);
2538 tgt = lmv_find_target(lmv, &data->op_fid1);
2540 RETURN(PTR_ERR(tgt));
2542 data->op_flags |= MF_MDC_CANCEL_FID1;
2543 rc = md_file_resync(tgt->ltd_exp, data);
2548 * Get dirent with the closest hash for striped directory
2550 * This function will search the dir entry, whose hash value is the
2551 * closest(>=) to hash from all of sub-stripes, and it is only being called
2552 * for striped directory.
2554 * \param[in] ctxt dir read context
2556 * \retval dirent get the entry successfully
2557 * NULL does not get the entry, normally it means
2558 * it reaches the end of the directory, while read
2559 * stripe dirent error is ignored to allow partial
2562 static struct lu_dirent *lmv_dirent_next(struct lmv_dir_ctxt *ctxt)
2564 struct stripe_dirent *stripe;
2565 struct lu_dirent *ent = NULL;
2569 /* TODO: optimize with k-way merge sort */
2570 for (i = 0; i < ctxt->ldc_count; i++) {
2571 stripe = &ctxt->ldc_stripes[i];
2575 if (!stripe->sd_ent) {
2576 stripe_dirent_load(ctxt, stripe, i);
2577 if (!stripe->sd_ent) {
2578 LASSERT(stripe->sd_eof);
2584 le64_to_cpu(ctxt->ldc_stripes[min].sd_ent->lde_hash) >
2585 le64_to_cpu(stripe->sd_ent->lde_hash)) {
2587 if (le64_to_cpu(stripe->sd_ent->lde_hash) ==
2594 stripe = &ctxt->ldc_stripes[min];
2595 ent = stripe->sd_ent;
2596 /* pop found dirent */
2597 stripe->sd_ent = stripe_dirent_get(ctxt, lu_dirent_next(ent),
2605 * Build dir entry page for striped directory
2607 * This function gets one entry by @offset from a striped directory. It will
2608 * read entries from all of stripes, and choose one closest to the required
2609 * offset(&offset). A few notes
2610 * 1. skip . and .. for non-zero stripes, because there can only have one .
2611 * and .. in a directory.
2612 * 2. op_data will be shared by all of stripes, instead of allocating new
2613 * one, so need to restore before reusing.
2615 * \param[in] exp obd export refer to LMV
2616 * \param[in] op_data hold those MD parameters of read_entry
2617 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2618 * \param[in] offset starting hash offset
2619 * \param[out] ppage the page holding the entry. Note: because the entry
2620 * will be accessed in upper layer, so we need hold the
2621 * page until the usages of entry is finished, see
2622 * ll_dir_entry_next.
2624 * retval =0 if get entry successfully
2625 * <0 cannot get entry
2627 static int lmv_striped_read_page(struct obd_export *exp,
2628 struct md_op_data *op_data,
2629 struct md_callback *cb_op,
2630 __u64 offset, struct page **ppage)
2632 struct page *page = NULL;
2633 struct lu_dirpage *dp;
2635 struct lu_dirent *ent;
2636 struct lu_dirent *last_ent;
2638 struct lmv_dir_ctxt *ctxt;
2639 struct lu_dirent *next = NULL;
2645 /* Allocate a page and read entries from all of stripes and fill
2646 * the page by hash order */
2647 page = alloc_page(GFP_KERNEL);
2651 /* Initialize the entry page */
2653 memset(dp, 0, sizeof(*dp));
2654 dp->ldp_hash_start = cpu_to_le64(offset);
2657 left_bytes = PAGE_SIZE - sizeof(*dp);
2661 /* initalize dir read context */
2662 stripe_count = op_data->op_mea1->lsm_md_stripe_count;
2663 OBD_ALLOC(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2665 GOTO(free_page, rc = -ENOMEM);
2666 ctxt->ldc_lmv = &exp->exp_obd->u.lmv;
2667 ctxt->ldc_op_data = op_data;
2668 ctxt->ldc_cb_op = cb_op;
2669 ctxt->ldc_hash = offset;
2670 ctxt->ldc_count = stripe_count;
2673 next = lmv_dirent_next(ctxt);
2675 /* end of directory */
2677 ctxt->ldc_hash = MDS_DIR_END_OFF;
2680 ctxt->ldc_hash = le64_to_cpu(next->lde_hash);
2682 ent_size = le16_to_cpu(next->lde_reclen);
2684 /* the last entry lde_reclen is 0, but it might not be the last
2685 * one of this temporay dir page */
2687 ent_size = lu_dirent_calc_size(
2688 le16_to_cpu(next->lde_namelen),
2689 le32_to_cpu(next->lde_attrs));
2691 if (ent_size > left_bytes)
2694 memcpy(ent, next, ent_size);
2696 /* Replace . with master FID and Replace .. with the parent FID
2697 * of master object */
2698 if (strncmp(ent->lde_name, ".",
2699 le16_to_cpu(ent->lde_namelen)) == 0 &&
2700 le16_to_cpu(ent->lde_namelen) == 1)
2701 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid1);
2702 else if (strncmp(ent->lde_name, "..",
2703 le16_to_cpu(ent->lde_namelen)) == 0 &&
2704 le16_to_cpu(ent->lde_namelen) == 2)
2705 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2707 CDEBUG(D_INODE, "entry %.*s hash %#llx\n",
2708 le16_to_cpu(ent->lde_namelen), ent->lde_name,
2709 le64_to_cpu(ent->lde_hash));
2711 left_bytes -= ent_size;
2712 ent->lde_reclen = cpu_to_le16(ent_size);
2714 ent = (void *)ent + ent_size;
2717 last_ent->lde_reclen = 0;
2720 dp->ldp_flags |= LDF_EMPTY;
2721 else if (ctxt->ldc_hash == le64_to_cpu(last_ent->lde_hash))
2722 dp->ldp_flags |= LDF_COLLIDE;
2723 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2724 dp->ldp_hash_end = cpu_to_le64(ctxt->ldc_hash);
2726 put_lmv_dir_ctxt(ctxt);
2727 OBD_FREE(ctxt, offsetof(typeof(*ctxt), ldc_stripes[stripe_count]));
2740 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2741 struct md_callback *cb_op, __u64 offset,
2742 struct page **ppage)
2744 struct obd_device *obd = exp->exp_obd;
2745 struct lmv_obd *lmv = &obd->u.lmv;
2746 struct lmv_tgt_desc *tgt;
2751 if (unlikely(lmv_dir_foreign(op_data->op_mea1)))
2754 if (unlikely(lmv_dir_striped(op_data->op_mea1))) {
2755 rc = lmv_striped_read_page(exp, op_data, cb_op, offset, ppage);
2759 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2761 RETURN(PTR_ERR(tgt));
2763 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2769 * Unlink a file/directory
2771 * Unlink a file or directory under the parent dir. The unlink request
2772 * usually will be sent to the MDT where the child is located, but if
2773 * the client does not have the child FID then request will be sent to the
2774 * MDT where the parent is located.
2776 * If the parent is a striped directory then it also needs to locate which
2777 * stripe the name of the child is located, and replace the parent FID
2778 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2779 * it will walk through all of sub-stripes until the child is being
2782 * \param[in] exp export refer to LMV
2783 * \param[in] op_data different parameters transferred beween client
2784 * MD stacks, name, namelen, FIDs etc.
2785 * op_fid1 is the parent FID, op_fid2 is the child
2787 * \param[out] request point to the request of unlink.
2789 * retval 0 if succeed
2790 * negative errno if failed.
2792 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2793 struct ptlrpc_request **request)
2795 struct obd_device *obd = exp->exp_obd;
2796 struct lmv_obd *lmv = &obd->u.lmv;
2797 struct lmv_tgt_desc *tgt;
2798 struct lmv_tgt_desc *parent_tgt;
2799 struct mdt_body *body;
2804 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2805 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2806 op_data->op_cap = cfs_curproc_cap_pack();
2809 parent_tgt = lmv_locate_tgt(lmv, op_data);
2810 if (IS_ERR(parent_tgt))
2811 RETURN(PTR_ERR(parent_tgt));
2813 if (likely(!fid_is_zero(&op_data->op_fid2))) {
2814 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2816 RETURN(PTR_ERR(tgt));
2822 * If child's fid is given, cancel unused locks for it if it is from
2823 * another export than parent.
2825 * LOOKUP lock for child (fid3) should also be cancelled on parent
2826 * tgt_tgt in mdc_unlink().
2828 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2830 if (parent_tgt != tgt)
2831 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2832 LCK_EX, MDS_INODELOCK_LOOKUP,
2833 MF_MDC_CANCEL_FID3);
2835 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2836 MDS_INODELOCK_ELC, MF_MDC_CANCEL_FID3);
2840 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2841 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2843 rc = md_unlink(tgt->ltd_exp, op_data, request);
2844 if (rc == -ENOENT && lmv_dir_retry_check_update(op_data)) {
2845 ptlrpc_req_finished(*request);
2853 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2857 /* Not cross-ref case, just get out of here. */
2858 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2861 /* This is a remote object, try remote MDT. */
2862 op_data->op_fid2 = body->mbo_fid1;
2863 ptlrpc_req_finished(*request);
2866 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2868 RETURN(PTR_ERR(tgt));
2873 static int lmv_precleanup(struct obd_device *obd)
2876 libcfs_kkuc_group_rem(&obd->obd_uuid, 0, KUC_GRP_HSM);
2877 fld_client_debugfs_fini(&obd->u.lmv.lmv_fld);
2878 lprocfs_obd_cleanup(obd);
2879 lprocfs_free_md_stats(obd);
2884 * Get by key a value associated with a LMV device.
2886 * Dispatch request to lower-layer devices as needed.
2888 * \param[in] env execution environment for this thread
2889 * \param[in] exp export for the LMV device
2890 * \param[in] keylen length of key identifier
2891 * \param[in] key identifier of key to get value for
2892 * \param[in] vallen size of \a val
2893 * \param[out] val pointer to storage location for value
2894 * \param[in] lsm optional striping metadata of object
2896 * \retval 0 on success
2897 * \retval negative negated errno on failure
2899 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2900 __u32 keylen, void *key, __u32 *vallen, void *val)
2902 struct obd_device *obd;
2903 struct lmv_obd *lmv;
2907 obd = class_exp2obd(exp);
2909 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2910 exp->exp_handle.h_cookie);
2915 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2918 LASSERT(*vallen == sizeof(__u32));
2919 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2920 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2922 * All tgts should be connected when this gets called.
2924 if (tgt == NULL || tgt->ltd_exp == NULL)
2927 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2932 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2933 KEY_IS(KEY_DEFAULT_EASIZE) ||
2934 KEY_IS(KEY_CONN_DATA)) {
2936 * Forwarding this request to first MDS, it should know LOV
2939 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2941 if (!rc && KEY_IS(KEY_CONN_DATA))
2942 exp->exp_connect_data = *(struct obd_connect_data *)val;
2944 } else if (KEY_IS(KEY_TGT_COUNT)) {
2945 *((int *)val) = lmv->desc.ld_tgt_count;
2949 CDEBUG(D_IOCTL, "Invalid key\n");
2954 * Asynchronously set by key a value associated with a LMV device.
2956 * Dispatch request to lower-layer devices as needed.
2958 * \param[in] env execution environment for this thread
2959 * \param[in] exp export for the LMV device
2960 * \param[in] keylen length of key identifier
2961 * \param[in] key identifier of key to store value for
2962 * \param[in] vallen size of value to store
2963 * \param[in] val pointer to data to be stored
2964 * \param[in] set optional list of related ptlrpc requests
2966 * \retval 0 on success
2967 * \retval negative negated errno on failure
2969 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2970 __u32 keylen, void *key, __u32 vallen, void *val,
2971 struct ptlrpc_request_set *set)
2973 struct lmv_tgt_desc *tgt = NULL;
2974 struct obd_device *obd;
2975 struct lmv_obd *lmv;
2979 obd = class_exp2obd(exp);
2981 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2982 exp->exp_handle.h_cookie);
2987 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2988 KEY_IS(KEY_DEFAULT_EASIZE)) {
2991 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2994 if (tgt == NULL || tgt->ltd_exp == NULL)
2997 err = obd_set_info_async(env, tgt->ltd_exp,
2998 keylen, key, vallen, val, set);
3009 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
3010 const struct lmv_mds_md_v1 *lmm1)
3012 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3019 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
3020 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3021 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
3022 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
3023 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
3025 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
3026 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
3027 lsm->lsm_md_migrate_offset = le32_to_cpu(lmm1->lmv_migrate_offset);
3028 lsm->lsm_md_migrate_hash = le32_to_cpu(lmm1->lmv_migrate_hash);
3029 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
3030 sizeof(lsm->lsm_md_pool_name));
3032 if (cplen >= sizeof(lsm->lsm_md_pool_name))
3035 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %#x "
3036 "layout_version %d\n", lsm->lsm_md_stripe_count,
3037 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
3038 lsm->lsm_md_layout_version);
3040 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
3041 for (i = 0; i < stripe_count; i++) {
3042 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
3043 &lmm1->lmv_stripe_fids[i]);
3045 * set default value -1, so lmv_locate_tgt() knows this stripe
3046 * target is not initialized.
3048 lsm->lsm_md_oinfo[i].lmo_mds = (u32)-1;
3049 if (!fid_is_sane(&lsm->lsm_md_oinfo[i].lmo_fid))
3052 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
3053 &lsm->lsm_md_oinfo[i].lmo_mds);
3060 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
3061 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
3067 static inline int lmv_unpack_user_md(struct obd_export *exp,
3068 struct lmv_stripe_md *lsm,
3069 const struct lmv_user_md *lmu)
3071 lsm->lsm_md_magic = le32_to_cpu(lmu->lum_magic);
3072 lsm->lsm_md_stripe_count = le32_to_cpu(lmu->lum_stripe_count);
3073 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmu->lum_stripe_offset);
3074 lsm->lsm_md_hash_type = le32_to_cpu(lmu->lum_hash_type);
3079 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
3080 const union lmv_mds_md *lmm, size_t lmm_size)
3082 struct lmv_stripe_md *lsm;
3085 bool allocated = false;
3088 LASSERT(lsmp != NULL);
3092 if (lsm != NULL && lmm == NULL) {
3094 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3096 if (lfm->lfm_magic == LMV_MAGIC_FOREIGN) {
3099 lfm_size = lfm->lfm_length + offsetof(typeof(*lfm),
3101 OBD_FREE_LARGE(lfm, lfm_size);
3105 if (lmv_dir_striped(lsm)) {
3106 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3107 if (lsm->lsm_md_oinfo[i].lmo_root)
3108 iput(lsm->lsm_md_oinfo[i].lmo_root);
3110 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
3112 lsm_size = lmv_stripe_md_size(0);
3114 OBD_FREE(lsm, lsm_size);
3119 /* foreign lmv case */
3120 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_FOREIGN) {
3121 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lsm;
3124 OBD_ALLOC_LARGE(lfm, lmm_size);
3127 *lsmp = (struct lmv_stripe_md *)lfm;
3129 lfm->lfm_magic = le32_to_cpu(lmm->lmv_foreign_md.lfm_magic);
3130 lfm->lfm_length = le32_to_cpu(lmm->lmv_foreign_md.lfm_length);
3131 lfm->lfm_type = le32_to_cpu(lmm->lmv_foreign_md.lfm_type);
3132 lfm->lfm_flags = le32_to_cpu(lmm->lmv_foreign_md.lfm_flags);
3133 memcpy(&lfm->lfm_value, &lmm->lmv_foreign_md.lfm_value,
3138 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
3142 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
3143 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
3144 CERROR("%s: invalid lmv magic %x: rc = %d\n",
3145 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
3150 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
3151 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3154 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
3155 * stripecount should be 0 then.
3157 lsm_size = lmv_stripe_md_size(0);
3160 OBD_ALLOC(lsm, lsm_size);
3167 switch (le32_to_cpu(lmm->lmv_magic)) {
3169 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3171 case LMV_USER_MAGIC:
3172 rc = lmv_unpack_user_md(exp, lsm, &lmm->lmv_user_md);
3175 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3176 le32_to_cpu(lmm->lmv_magic));
3181 if (rc != 0 && allocated) {
3182 OBD_FREE(lsm, lsm_size);
3189 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3191 lmv_unpackmd(NULL, &lsm, NULL, 0);
3193 EXPORT_SYMBOL(lmv_free_memmd);
3195 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3196 union ldlm_policy_data *policy,
3197 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
3200 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3205 LASSERT(fid != NULL);
3207 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3208 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3211 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3214 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3222 static int lmv_set_lock_data(struct obd_export *exp,
3223 const struct lustre_handle *lockh,
3224 void *data, __u64 *bits)
3226 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3227 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3231 if (tgt == NULL || tgt->ltd_exp == NULL)
3233 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3237 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
3238 const struct lu_fid *fid, enum ldlm_type type,
3239 union ldlm_policy_data *policy,
3240 enum ldlm_mode mode, struct lustre_handle *lockh)
3242 struct obd_device *obd = exp->exp_obd;
3243 struct lmv_obd *lmv = &obd->u.lmv;
3249 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3252 * With DNE every object can have two locks in different namespaces:
3253 * lookup lock in space of MDT storing direntry and update/open lock in
3254 * space of MDT storing inode. Try the MDT that the FID maps to first,
3255 * since this can be easily found, and only try others if that fails.
3257 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3258 i < lmv->desc.ld_tgt_count;
3259 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3261 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3262 obd->obd_name, PFID(fid), tgt);
3266 if (lmv->tgts[tgt] == NULL ||
3267 lmv->tgts[tgt]->ltd_exp == NULL ||
3268 lmv->tgts[tgt]->ltd_active == 0)
3271 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3272 type, policy, mode, lockh);
3280 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3281 struct obd_export *dt_exp, struct obd_export *md_exp,
3282 struct lustre_md *md)
3284 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3285 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3287 if (tgt == NULL || tgt->ltd_exp == NULL)
3290 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3293 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3295 struct obd_device *obd = exp->exp_obd;
3296 struct lmv_obd *lmv = &obd->u.lmv;
3297 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3300 if (md->default_lmv) {
3301 lmv_free_memmd(md->default_lmv);
3302 md->default_lmv = NULL;
3304 if (md->lmv != NULL) {
3305 lmv_free_memmd(md->lmv);
3308 if (tgt == NULL || tgt->ltd_exp == NULL)
3310 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3313 int lmv_set_open_replay_data(struct obd_export *exp,
3314 struct obd_client_handle *och,
3315 struct lookup_intent *it)
3317 struct obd_device *obd = exp->exp_obd;
3318 struct lmv_obd *lmv = &obd->u.lmv;
3319 struct lmv_tgt_desc *tgt;
3322 tgt = lmv_find_target(lmv, &och->och_fid);
3324 RETURN(PTR_ERR(tgt));
3326 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3329 int lmv_clear_open_replay_data(struct obd_export *exp,
3330 struct obd_client_handle *och)
3332 struct obd_device *obd = exp->exp_obd;
3333 struct lmv_obd *lmv = &obd->u.lmv;
3334 struct lmv_tgt_desc *tgt;
3337 tgt = lmv_find_target(lmv, &och->och_fid);
3339 RETURN(PTR_ERR(tgt));
3341 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3344 int lmv_intent_getattr_async(struct obd_export *exp,
3345 struct md_enqueue_info *minfo)
3347 struct md_op_data *op_data = &minfo->mi_data;
3348 struct obd_device *obd = exp->exp_obd;
3349 struct lmv_obd *lmv = &obd->u.lmv;
3350 struct lmv_tgt_desc *tgt = NULL;
3354 if (!fid_is_sane(&op_data->op_fid2))
3357 tgt = lmv_find_target(lmv, &op_data->op_fid1);
3359 RETURN(PTR_ERR(tgt));
3362 * no special handle for remote dir, which needs to fetch both LOOKUP
3363 * lock on parent, and then UPDATE lock on child MDT, which makes all
3364 * complicated because this is done async. So only LOOKUP lock is
3365 * fetched for remote dir, but considering remote dir is rare case,
3366 * and not supporting it in statahead won't cause any issue, just leave
3370 rc = md_intent_getattr_async(tgt->ltd_exp, minfo);
3374 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3375 struct lu_fid *fid, __u64 *bits)
3377 struct obd_device *obd = exp->exp_obd;
3378 struct lmv_obd *lmv = &obd->u.lmv;
3379 struct lmv_tgt_desc *tgt;
3383 tgt = lmv_find_target(lmv, fid);
3385 RETURN(PTR_ERR(tgt));
3387 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3391 int lmv_get_fid_from_lsm(struct obd_export *exp,
3392 const struct lmv_stripe_md *lsm,
3393 const char *name, int namelen, struct lu_fid *fid)
3395 const struct lmv_oinfo *oinfo;
3397 LASSERT(lmv_dir_striped(lsm));
3399 oinfo = lsm_name_to_stripe_info(lsm, name, namelen, false);
3401 return PTR_ERR(oinfo);
3403 *fid = oinfo->lmo_fid;
3409 * For lmv, only need to send request to master MDT, and the master MDT will
3410 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3411 * we directly fetch data from the slave MDTs.
3413 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3414 struct obd_quotactl *oqctl)
3416 struct obd_device *obd = class_exp2obd(exp);
3417 struct lmv_obd *lmv = &obd->u.lmv;
3418 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3421 __u64 curspace, curinodes;
3425 tgt->ltd_exp == NULL ||
3427 lmv->desc.ld_tgt_count == 0) {
3428 CERROR("master lmv inactive\n");
3432 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3433 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3437 curspace = curinodes = 0;
3438 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3442 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3445 err = obd_quotactl(tgt->ltd_exp, oqctl);
3447 CERROR("getquota on mdt %d failed. %d\n", i, err);
3451 curspace += oqctl->qc_dqblk.dqb_curspace;
3452 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3455 oqctl->qc_dqblk.dqb_curspace = curspace;
3456 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3461 static int lmv_merge_attr(struct obd_export *exp,
3462 const struct lmv_stripe_md *lsm,
3463 struct cl_attr *attr,
3464 ldlm_blocking_callback cb_blocking)
3469 if (!lmv_dir_striped(lsm))
3472 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3476 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3477 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3483 "" DFID " size %llu, blocks %llu nlink %u, atime %lld ctime %lld, mtime %lld.\n",
3484 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3485 i_size_read(inode), (unsigned long long)inode->i_blocks,
3486 inode->i_nlink, (s64)inode->i_atime.tv_sec,
3487 (s64)inode->i_ctime.tv_sec, (s64)inode->i_mtime.tv_sec);
3489 /* for slave stripe, it needs to subtract nlink for . and .. */
3491 attr->cat_nlink += inode->i_nlink - 2;
3493 attr->cat_nlink = inode->i_nlink;
3495 attr->cat_size += i_size_read(inode);
3496 attr->cat_blocks += inode->i_blocks;
3498 if (attr->cat_atime < inode->i_atime.tv_sec)
3499 attr->cat_atime = inode->i_atime.tv_sec;
3501 if (attr->cat_ctime < inode->i_ctime.tv_sec)
3502 attr->cat_ctime = inode->i_ctime.tv_sec;
3504 if (attr->cat_mtime < inode->i_mtime.tv_sec)
3505 attr->cat_mtime = inode->i_mtime.tv_sec;
3510 struct obd_ops lmv_obd_ops = {
3511 .o_owner = THIS_MODULE,
3512 .o_setup = lmv_setup,
3513 .o_cleanup = lmv_cleanup,
3514 .o_precleanup = lmv_precleanup,
3515 .o_process_config = lmv_process_config,
3516 .o_connect = lmv_connect,
3517 .o_disconnect = lmv_disconnect,
3518 .o_statfs = lmv_statfs,
3519 .o_get_info = lmv_get_info,
3520 .o_set_info_async = lmv_set_info_async,
3521 .o_notify = lmv_notify,
3522 .o_get_uuid = lmv_get_uuid,
3523 .o_iocontrol = lmv_iocontrol,
3524 .o_quotactl = lmv_quotactl
3527 struct md_ops lmv_md_ops = {
3528 .m_get_root = lmv_get_root,
3529 .m_null_inode = lmv_null_inode,
3530 .m_close = lmv_close,
3531 .m_create = lmv_create,
3532 .m_enqueue = lmv_enqueue,
3533 .m_getattr = lmv_getattr,
3534 .m_getxattr = lmv_getxattr,
3535 .m_getattr_name = lmv_getattr_name,
3536 .m_intent_lock = lmv_intent_lock,
3538 .m_rename = lmv_rename,
3539 .m_setattr = lmv_setattr,
3540 .m_setxattr = lmv_setxattr,
3541 .m_fsync = lmv_fsync,
3542 .m_file_resync = lmv_file_resync,
3543 .m_read_page = lmv_read_page,
3544 .m_unlink = lmv_unlink,
3545 .m_init_ea_size = lmv_init_ea_size,
3546 .m_cancel_unused = lmv_cancel_unused,
3547 .m_set_lock_data = lmv_set_lock_data,
3548 .m_lock_match = lmv_lock_match,
3549 .m_get_lustre_md = lmv_get_lustre_md,
3550 .m_free_lustre_md = lmv_free_lustre_md,
3551 .m_merge_attr = lmv_merge_attr,
3552 .m_set_open_replay_data = lmv_set_open_replay_data,
3553 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3554 .m_intent_getattr_async = lmv_intent_getattr_async,
3555 .m_revalidate_lock = lmv_revalidate_lock,
3556 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3557 .m_unpackmd = lmv_unpackmd,
3560 static int __init lmv_init(void)
3562 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3563 LUSTRE_LMV_NAME, NULL);
3566 static void __exit lmv_exit(void)
3568 class_unregister_type(LUSTRE_LMV_NAME);
3571 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3572 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3573 MODULE_VERSION(LUSTRE_VERSION_STRING);
3574 MODULE_LICENSE("GPL");
3576 module_init(lmv_init);
3577 module_exit(lmv_exit);