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, 2016, 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
34 #include <linux/slab.h>
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/user_namespace.h>
38 #ifdef HAVE_UIDGID_HEADER
39 # include <linux/uidgid.h>
41 #include <linux/slab.h>
42 #include <linux/pagemap.h>
44 #include <linux/math64.h>
45 #include <linux/seq_file.h>
46 #include <linux/namei.h>
48 #include <lustre/lustre_idl.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 <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->cluuid = *cluuid;
215 lmv->conn_data = *data;
217 if (lmv->targets_proc_entry == NULL) {
218 lmv->targets_proc_entry = lprocfs_register("target_obds",
221 if (IS_ERR(lmv->targets_proc_entry)) {
222 CERROR("%s: cannot register "
223 "/proc/fs/lustre/%s/%s/target_obds\n",
224 obd->obd_name, obd->obd_type->typ_name,
226 lmv->targets_proc_entry = NULL;
230 rc = lmv_check_connect(obd);
239 if (lmv->targets_proc_entry != NULL)
240 lprocfs_remove(&lmv->targets_proc_entry);
242 class_disconnect(exp);
247 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
250 struct obd_device *obd = exp->exp_obd;
251 struct lmv_obd *lmv = &obd->u.lmv;
257 if (lmv->max_easize < easize) {
258 lmv->max_easize = easize;
261 if (lmv->max_def_easize < def_easize) {
262 lmv->max_def_easize = def_easize;
269 if (lmv->connected == 0)
272 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
273 struct lmv_tgt_desc *tgt = lmv->tgts[i];
275 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
276 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
280 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
282 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
283 " rc = %d\n", obd->obd_name, i, rc);
290 #define MAX_STRING_SIZE 128
292 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
294 struct lmv_obd *lmv = &obd->u.lmv;
295 struct obd_uuid *cluuid = &lmv->cluuid;
296 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
297 struct obd_device *mdc_obd;
298 struct obd_export *mdc_exp;
299 struct lu_fld_target target;
303 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
306 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
310 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
311 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
312 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
315 if (!mdc_obd->obd_set_up) {
316 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
320 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
321 &lmv->conn_data, NULL);
323 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
328 * Init fid sequence client for this mdc and add new fld target.
330 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
334 target.ft_srv = NULL;
335 target.ft_exp = mdc_exp;
336 target.ft_idx = tgt->ltd_idx;
338 fld_client_add_target(&lmv->lmv_fld, &target);
340 rc = obd_register_observer(mdc_obd, obd);
342 obd_disconnect(mdc_exp);
343 CERROR("target %s register_observer error %d\n",
344 tgt->ltd_uuid.uuid, rc);
348 if (obd->obd_observer) {
350 * Tell the observer about the new target.
352 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
355 obd_disconnect(mdc_exp);
361 tgt->ltd_exp = mdc_exp;
362 lmv->desc.ld_active_tgt_count++;
364 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
366 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
367 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
368 atomic_read(&obd->obd_refcount));
370 if (lmv->targets_proc_entry != NULL) {
371 struct proc_dir_entry *mdc_symlink;
373 LASSERT(mdc_obd->obd_type != NULL);
374 LASSERT(mdc_obd->obd_type->typ_name != NULL);
375 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
376 lmv->targets_proc_entry,
378 mdc_obd->obd_type->typ_name,
380 if (mdc_symlink == NULL) {
381 CERROR("cannot register LMV target "
382 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
383 obd->obd_type->typ_name, obd->obd_name,
390 static void lmv_del_target(struct lmv_obd *lmv, int index)
392 if (lmv->tgts[index] == NULL)
395 OBD_FREE_PTR(lmv->tgts[index]);
396 lmv->tgts[index] = NULL;
400 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
401 __u32 index, int gen)
403 struct obd_device *mdc_obd;
404 struct lmv_obd *lmv = &obd->u.lmv;
405 struct lmv_tgt_desc *tgt;
406 int orig_tgt_count = 0;
410 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
411 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
414 CERROR("%s: Target %s not attached: rc = %d\n",
415 obd->obd_name, uuidp->uuid, -EINVAL);
419 mutex_lock(&lmv->lmv_init_mutex);
420 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
421 tgt = lmv->tgts[index];
422 CERROR("%s: UUID %s already assigned at LOV target index %d:"
423 " rc = %d\n", obd->obd_name,
424 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
425 mutex_unlock(&lmv->lmv_init_mutex);
429 if (index >= lmv->tgts_size) {
430 /* We need to reallocate the lmv target array. */
431 struct lmv_tgt_desc **newtgts, **old = NULL;
435 while (newsize < index + 1)
436 newsize = newsize << 1;
437 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
438 if (newtgts == NULL) {
439 mutex_unlock(&lmv->lmv_init_mutex);
443 if (lmv->tgts_size) {
444 memcpy(newtgts, lmv->tgts,
445 sizeof(*newtgts) * lmv->tgts_size);
447 oldsize = lmv->tgts_size;
451 lmv->tgts_size = newsize;
454 OBD_FREE(old, sizeof(*old) * oldsize);
456 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
462 mutex_unlock(&lmv->lmv_init_mutex);
466 mutex_init(&tgt->ltd_fid_mutex);
467 tgt->ltd_idx = index;
468 tgt->ltd_uuid = *uuidp;
470 lmv->tgts[index] = tgt;
471 if (index >= lmv->desc.ld_tgt_count) {
472 orig_tgt_count = lmv->desc.ld_tgt_count;
473 lmv->desc.ld_tgt_count = index + 1;
476 if (lmv->connected == 0) {
477 /* lmv_check_connect() will connect this target. */
478 mutex_unlock(&lmv->lmv_init_mutex);
482 /* Otherwise let's connect it ourselves */
483 mutex_unlock(&lmv->lmv_init_mutex);
484 rc = lmv_connect_mdc(obd, tgt);
486 spin_lock(&lmv->lmv_lock);
487 if (lmv->desc.ld_tgt_count == index + 1)
488 lmv->desc.ld_tgt_count = orig_tgt_count;
489 memset(tgt, 0, sizeof(*tgt));
490 spin_unlock(&lmv->lmv_lock);
492 int easize = sizeof(struct lmv_stripe_md) +
493 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
494 lmv_init_ea_size(obd->obd_self_export, easize, 0);
500 static int lmv_check_connect(struct obd_device *obd)
502 struct lmv_obd *lmv = &obd->u.lmv;
503 struct lmv_tgt_desc *tgt;
512 mutex_lock(&lmv->lmv_init_mutex);
513 if (lmv->connected) {
514 mutex_unlock(&lmv->lmv_init_mutex);
518 if (lmv->desc.ld_tgt_count == 0) {
519 mutex_unlock(&lmv->lmv_init_mutex);
520 CERROR("%s: no targets configured.\n", obd->obd_name);
524 LASSERT(lmv->tgts != NULL);
526 if (lmv->tgts[0] == NULL) {
527 mutex_unlock(&lmv->lmv_init_mutex);
528 CERROR("%s: no target configured for index 0.\n",
533 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
534 lmv->cluuid.uuid, obd->obd_name);
536 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
540 rc = lmv_connect_mdc(obd, tgt);
546 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
547 lmv_init_ea_size(obd->obd_self_export, easize, 0);
548 mutex_unlock(&lmv->lmv_init_mutex);
559 --lmv->desc.ld_active_tgt_count;
560 rc2 = obd_disconnect(tgt->ltd_exp);
562 CERROR("LMV target %s disconnect on "
563 "MDC idx %d: error %d\n",
564 tgt->ltd_uuid.uuid, i, rc2);
569 mutex_unlock(&lmv->lmv_init_mutex);
574 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
576 struct lmv_obd *lmv = &obd->u.lmv;
577 struct obd_device *mdc_obd;
581 LASSERT(tgt != NULL);
582 LASSERT(obd != NULL);
584 mdc_obd = class_exp2obd(tgt->ltd_exp);
587 mdc_obd->obd_force = obd->obd_force;
588 mdc_obd->obd_fail = obd->obd_fail;
589 mdc_obd->obd_no_recov = obd->obd_no_recov;
591 if (lmv->targets_proc_entry != NULL)
592 lprocfs_remove_proc_entry(mdc_obd->obd_name,
593 lmv->targets_proc_entry);
596 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
598 CERROR("Can't finanize fids factory\n");
600 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
601 tgt->ltd_exp->exp_obd->obd_name,
602 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
604 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
605 rc = obd_disconnect(tgt->ltd_exp);
607 if (tgt->ltd_active) {
608 CERROR("Target %s disconnect error %d\n",
609 tgt->ltd_uuid.uuid, rc);
613 lmv_activate_target(lmv, tgt, 0);
618 static int lmv_disconnect(struct obd_export *exp)
620 struct obd_device *obd = class_exp2obd(exp);
621 struct lmv_obd *lmv = &obd->u.lmv;
629 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
630 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
633 lmv_disconnect_mdc(obd, lmv->tgts[i]);
636 if (lmv->targets_proc_entry != NULL)
637 lprocfs_remove(&lmv->targets_proc_entry);
639 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
640 obd->obd_type->typ_name, obd->obd_name);
644 * This is the case when no real connection is established by
645 * lmv_check_connect().
648 class_export_put(exp);
649 rc = class_disconnect(exp);
655 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
658 struct obd_device *obddev = class_exp2obd(exp);
659 struct lmv_obd *lmv = &obddev->u.lmv;
660 struct getinfo_fid2path *gf;
661 struct lmv_tgt_desc *tgt;
662 struct getinfo_fid2path *remote_gf = NULL;
663 struct lu_fid root_fid;
664 int remote_gf_size = 0;
668 tgt = lmv_find_target(lmv, &gf->gf_fid);
670 RETURN(PTR_ERR(tgt));
672 root_fid = *gf->gf_u.gf_root_fid;
673 LASSERT(fid_is_sane(&root_fid));
676 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
677 if (rc != 0 && rc != -EREMOTE)
678 GOTO(out_fid2path, rc);
680 /* If remote_gf != NULL, it means just building the
681 * path on the remote MDT, copy this path segement to gf */
682 if (remote_gf != NULL) {
683 struct getinfo_fid2path *ori_gf;
686 ori_gf = (struct getinfo_fid2path *)karg;
687 if (strlen(ori_gf->gf_u.gf_path) +
688 strlen(gf->gf_u.gf_path) > ori_gf->gf_pathlen)
689 GOTO(out_fid2path, rc = -EOVERFLOW);
691 ptr = ori_gf->gf_u.gf_path;
693 memmove(ptr + strlen(gf->gf_u.gf_path) + 1, ptr,
694 strlen(ori_gf->gf_u.gf_path));
696 strncpy(ptr, gf->gf_u.gf_path,
697 strlen(gf->gf_u.gf_path));
698 ptr += strlen(gf->gf_u.gf_path);
702 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
703 tgt->ltd_exp->exp_obd->obd_name,
704 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
708 GOTO(out_fid2path, rc);
710 /* sigh, has to go to another MDT to do path building further */
711 if (remote_gf == NULL) {
712 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
713 OBD_ALLOC(remote_gf, remote_gf_size);
714 if (remote_gf == NULL)
715 GOTO(out_fid2path, rc = -ENOMEM);
716 remote_gf->gf_pathlen = PATH_MAX;
719 if (!fid_is_sane(&gf->gf_fid)) {
720 CERROR("%s: invalid FID "DFID": rc = %d\n",
721 tgt->ltd_exp->exp_obd->obd_name,
722 PFID(&gf->gf_fid), -EINVAL);
723 GOTO(out_fid2path, rc = -EINVAL);
726 tgt = lmv_find_target(lmv, &gf->gf_fid);
728 GOTO(out_fid2path, rc = -EINVAL);
730 remote_gf->gf_fid = gf->gf_fid;
731 remote_gf->gf_recno = -1;
732 remote_gf->gf_linkno = -1;
733 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
734 *remote_gf->gf_u.gf_root_fid = root_fid;
736 goto repeat_fid2path;
739 if (remote_gf != NULL)
740 OBD_FREE(remote_gf, remote_gf_size);
744 static int lmv_hsm_req_count(struct lmv_obd *lmv,
745 const struct hsm_user_request *hur,
746 const struct lmv_tgt_desc *tgt_mds)
750 struct lmv_tgt_desc *curr_tgt;
752 /* count how many requests must be sent to the given target */
753 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
754 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
755 if (IS_ERR(curr_tgt))
756 RETURN(PTR_ERR(curr_tgt));
757 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
763 static int lmv_hsm_req_build(struct lmv_obd *lmv,
764 struct hsm_user_request *hur_in,
765 const struct lmv_tgt_desc *tgt_mds,
766 struct hsm_user_request *hur_out)
769 struct lmv_tgt_desc *curr_tgt;
771 /* build the hsm_user_request for the given target */
772 hur_out->hur_request = hur_in->hur_request;
774 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
775 curr_tgt = lmv_find_target(lmv,
776 &hur_in->hur_user_item[i].hui_fid);
777 if (IS_ERR(curr_tgt))
778 RETURN(PTR_ERR(curr_tgt));
779 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
780 hur_out->hur_user_item[nr_out] =
781 hur_in->hur_user_item[i];
785 hur_out->hur_request.hr_itemcount = nr_out;
786 memcpy(hur_data(hur_out), hur_data(hur_in),
787 hur_in->hur_request.hr_data_len);
792 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
793 struct lustre_kernelcomm *lk,
800 /* unregister request (call from llapi_hsm_copytool_fini) */
801 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
802 struct lmv_tgt_desc *tgt = lmv->tgts[i];
804 if (tgt == NULL || tgt->ltd_exp == NULL)
806 /* best effort: try to clean as much as possible
807 * (continue on error) */
808 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
811 /* Whatever the result, remove copytool from kuc groups.
812 * Unreached coordinators will get EPIPE on next requests
813 * and will unregister automatically.
815 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
820 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
821 struct lustre_kernelcomm *lk, __user void *uarg)
826 bool any_set = false;
827 struct kkuc_ct_data kcd = { 0 };
830 /* All or nothing: try to register to all MDS.
831 * In case of failure, unregister from previous MDS,
832 * except if it because of inactive target. */
833 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
834 struct lmv_tgt_desc *tgt = lmv->tgts[i];
836 if (tgt == NULL || tgt->ltd_exp == NULL)
838 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
840 if (tgt->ltd_active) {
841 /* permanent error */
842 CERROR("%s: iocontrol MDC %s on MDT"
843 " idx %d cmd %x: err = %d\n",
844 lmv2obd_dev(lmv)->obd_name,
845 tgt->ltd_uuid.uuid, i, cmd, err);
847 lk->lk_flags |= LK_FLG_STOP;
848 /* unregister from previous MDS */
849 for (j = 0; j < i; j++) {
851 if (tgt == NULL || tgt->ltd_exp == NULL)
853 obd_iocontrol(cmd, tgt->ltd_exp, len,
858 /* else: transient error.
859 * kuc will register to the missing MDT
867 /* no registration done: return error */
870 /* at least one registration done, with no failure */
871 filp = fget(lk->lk_wfd);
875 kcd.kcd_magic = KKUC_CT_DATA_MAGIC;
876 kcd.kcd_uuid = lmv->cluuid;
877 kcd.kcd_archive = lk->lk_data;
879 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group,
890 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
891 int len, void *karg, void __user *uarg)
893 struct obd_device *obddev = class_exp2obd(exp);
894 struct lmv_obd *lmv = &obddev->u.lmv;
895 struct lmv_tgt_desc *tgt = NULL;
899 __u32 count = lmv->desc.ld_tgt_count;
906 case IOC_OBD_STATFS: {
907 struct obd_ioctl_data *data = karg;
908 struct obd_device *mdc_obd;
909 struct obd_statfs stat_buf = {0};
912 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
913 if ((index >= count))
916 tgt = lmv->tgts[index];
917 if (tgt == NULL || !tgt->ltd_active)
920 mdc_obd = class_exp2obd(tgt->ltd_exp);
925 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
926 min((int) data->ioc_plen2,
927 (int) sizeof(struct obd_uuid))))
930 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
931 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
935 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
936 min((int) data->ioc_plen1,
937 (int) sizeof(stat_buf))))
941 case OBD_IOC_QUOTACTL: {
942 struct if_quotactl *qctl = karg;
943 struct obd_quotactl *oqctl;
945 if (qctl->qc_valid == QC_MDTIDX) {
946 if (count <= qctl->qc_idx)
949 tgt = lmv->tgts[qctl->qc_idx];
950 if (tgt == NULL || tgt->ltd_exp == NULL)
952 } else if (qctl->qc_valid == QC_UUID) {
953 for (i = 0; i < count; i++) {
957 if (!obd_uuid_equals(&tgt->ltd_uuid,
961 if (tgt->ltd_exp == NULL)
973 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
974 OBD_ALLOC_PTR(oqctl);
978 QCTL_COPY(oqctl, qctl);
979 rc = obd_quotactl(tgt->ltd_exp, oqctl);
981 QCTL_COPY(qctl, oqctl);
982 qctl->qc_valid = QC_MDTIDX;
983 qctl->obd_uuid = tgt->ltd_uuid;
988 case OBD_IOC_CHANGELOG_SEND:
989 case OBD_IOC_CHANGELOG_CLEAR: {
990 struct ioc_changelog *icc = karg;
992 if (icc->icc_mdtindex >= count)
995 tgt = lmv->tgts[icc->icc_mdtindex];
996 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
998 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1001 case LL_IOC_GET_CONNECT_FLAGS: {
1003 if (tgt == NULL || tgt->ltd_exp == NULL)
1005 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1008 case LL_IOC_FID2MDTIDX: {
1009 struct lu_fid *fid = karg;
1012 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1016 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1017 * point to user space memory for FID2MDTIDX. */
1018 *(__u32 *)uarg = mdt_index;
1021 case OBD_IOC_FID2PATH: {
1022 rc = lmv_fid2path(exp, len, karg, uarg);
1025 case LL_IOC_HSM_STATE_GET:
1026 case LL_IOC_HSM_STATE_SET:
1027 case LL_IOC_HSM_ACTION: {
1028 struct md_op_data *op_data = karg;
1030 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1032 RETURN(PTR_ERR(tgt));
1034 if (tgt->ltd_exp == NULL)
1037 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1040 case LL_IOC_HSM_PROGRESS: {
1041 const struct hsm_progress_kernel *hpk = karg;
1043 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1045 RETURN(PTR_ERR(tgt));
1046 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1049 case LL_IOC_HSM_REQUEST: {
1050 struct hsm_user_request *hur = karg;
1051 unsigned int reqcount = hur->hur_request.hr_itemcount;
1056 /* if the request is about a single fid
1057 * or if there is a single MDS, no need to split
1059 if (reqcount == 1 || count == 1) {
1060 tgt = lmv_find_target(lmv,
1061 &hur->hur_user_item[0].hui_fid);
1063 RETURN(PTR_ERR(tgt));
1064 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1066 /* split fid list to their respective MDS */
1067 for (i = 0; i < count; i++) {
1070 struct hsm_user_request *req;
1073 if (tgt == NULL || tgt->ltd_exp == NULL)
1076 nr = lmv_hsm_req_count(lmv, hur, tgt);
1079 if (nr == 0) /* nothing for this MDS */
1082 /* build a request with fids for this MDS */
1083 reqlen = offsetof(typeof(*hur),
1085 + hur->hur_request.hr_data_len;
1086 OBD_ALLOC_LARGE(req, reqlen);
1089 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1091 GOTO(hsm_req_err, rc1);
1092 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1095 if (rc1 != 0 && rc == 0)
1097 OBD_FREE_LARGE(req, reqlen);
1102 case LL_IOC_LOV_SWAP_LAYOUTS: {
1103 struct md_op_data *op_data = karg;
1104 struct lmv_tgt_desc *tgt1, *tgt2;
1106 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1108 RETURN(PTR_ERR(tgt1));
1110 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1112 RETURN(PTR_ERR(tgt2));
1114 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1117 /* only files on same MDT can have their layouts swapped */
1118 if (tgt1->ltd_idx != tgt2->ltd_idx)
1121 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1124 case LL_IOC_HSM_CT_START: {
1125 struct lustre_kernelcomm *lk = karg;
1126 if (lk->lk_flags & LK_FLG_STOP)
1127 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1129 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1133 for (i = 0; i < count; i++) {
1134 struct obd_device *mdc_obd;
1138 if (tgt == NULL || tgt->ltd_exp == NULL)
1140 /* ll_umount_begin() sets force flag but for lmv, not
1141 * mdc. Let's pass it through */
1142 mdc_obd = class_exp2obd(tgt->ltd_exp);
1143 mdc_obd->obd_force = obddev->obd_force;
1144 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1146 if (tgt->ltd_active) {
1147 CERROR("error: iocontrol MDC %s on MDT"
1148 " idx %d cmd %x: err = %d\n",
1149 tgt->ltd_uuid.uuid, i, cmd, err);
1163 * This is _inode_ placement policy function (not name).
1165 static int lmv_placement_policy(struct obd_device *obd,
1166 struct md_op_data *op_data, u32 *mds)
1168 struct lmv_obd *lmv = &obd->u.lmv;
1169 struct lmv_user_md *lum;
1173 LASSERT(mds != NULL);
1175 if (lmv->desc.ld_tgt_count == 1) {
1180 lum = op_data->op_data;
1182 * 1. See if the stripe offset is specified by lum.
1183 * 2. Then check if there is default stripe offset.
1184 * 3. Finally choose MDS by name hash if the parent
1185 * is striped directory. (see lmv_locate_mds()). */
1186 if (op_data->op_cli_flags & CLI_SET_MEA && lum != NULL &&
1187 le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1188 *mds = le32_to_cpu(lum->lum_stripe_offset);
1189 } else if (op_data->op_default_stripe_offset != (__u32)-1) {
1190 *mds = op_data->op_default_stripe_offset;
1191 op_data->op_mds = *mds;
1192 /* Correct the stripe offset in lum */
1194 lum->lum_stripe_offset = cpu_to_le32(*mds);
1196 *mds = op_data->op_mds;
1202 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1204 struct lmv_tgt_desc *tgt;
1208 tgt = lmv_get_target(lmv, mds, NULL);
1210 RETURN(PTR_ERR(tgt));
1213 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1214 * on server that seq in new allocated fid is not yet known.
1216 mutex_lock(&tgt->ltd_fid_mutex);
1218 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1219 GOTO(out, rc = -ENODEV);
1222 * Asking underlying tgt layer to allocate new fid.
1224 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1226 LASSERT(fid_is_sane(fid));
1232 mutex_unlock(&tgt->ltd_fid_mutex);
1236 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1237 struct lu_fid *fid, struct md_op_data *op_data)
1239 struct obd_device *obd = class_exp2obd(exp);
1240 struct lmv_obd *lmv = &obd->u.lmv;
1245 LASSERT(op_data != NULL);
1246 LASSERT(fid != NULL);
1248 rc = lmv_placement_policy(obd, op_data, &mds);
1250 CERROR("Can't get target for allocating fid, "
1255 rc = __lmv_fid_alloc(lmv, fid, mds);
1257 CERROR("Can't alloc new fid, rc %d\n", rc);
1264 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1266 struct lmv_obd *lmv = &obd->u.lmv;
1267 struct lmv_desc *desc;
1271 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1272 CERROR("LMV setup requires a descriptor\n");
1276 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1277 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1278 CERROR("Lmv descriptor size wrong: %d > %d\n",
1279 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1283 lmv->tgts_size = 32U;
1284 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1285 if (lmv->tgts == NULL)
1288 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1289 lmv->desc.ld_tgt_count = 0;
1290 lmv->desc.ld_active_tgt_count = 0;
1291 lmv->max_def_easize = 0;
1292 lmv->max_easize = 0;
1294 spin_lock_init(&lmv->lmv_lock);
1295 mutex_init(&lmv->lmv_init_mutex);
1297 #ifdef CONFIG_PROC_FS
1298 obd->obd_vars = lprocfs_lmv_obd_vars;
1299 lprocfs_obd_setup(obd);
1300 lprocfs_alloc_md_stats(obd, 0);
1301 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1302 0444, &lmv_proc_target_fops, obd);
1304 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1307 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1308 LUSTRE_CLI_FLD_HASH_DHT);
1310 CERROR("Can't init FLD, err %d\n", rc);
1320 static int lmv_cleanup(struct obd_device *obd)
1322 struct lmv_obd *lmv = &obd->u.lmv;
1325 fld_client_fini(&lmv->lmv_fld);
1326 if (lmv->tgts != NULL) {
1328 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1329 if (lmv->tgts[i] == NULL)
1331 lmv_del_target(lmv, i);
1333 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1339 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1341 struct lustre_cfg *lcfg = buf;
1342 struct obd_uuid obd_uuid;
1348 switch (lcfg->lcfg_command) {
1350 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1351 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1352 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1353 GOTO(out, rc = -EINVAL);
1355 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1357 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1358 GOTO(out, rc = -EINVAL);
1359 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1360 GOTO(out, rc = -EINVAL);
1361 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1364 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1365 GOTO(out, rc = -EINVAL);
1371 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1372 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1374 struct obd_device *obd = class_exp2obd(exp);
1375 struct lmv_obd *lmv = &obd->u.lmv;
1376 struct obd_statfs *temp;
1381 OBD_ALLOC(temp, sizeof(*temp));
1385 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1386 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1389 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1392 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1393 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1395 GOTO(out_free_temp, rc);
1400 /* If the statfs is from mount, it will needs
1401 * retrieve necessary information from MDT0.
1402 * i.e. mount does not need the merged osfs
1404 * And also clients can be mounted as long as
1405 * MDT0 is in service*/
1406 if (flags & OBD_STATFS_FOR_MDT0)
1407 GOTO(out_free_temp, rc);
1409 osfs->os_bavail += temp->os_bavail;
1410 osfs->os_blocks += temp->os_blocks;
1411 osfs->os_ffree += temp->os_ffree;
1412 osfs->os_files += temp->os_files;
1418 OBD_FREE(temp, sizeof(*temp));
1422 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1425 struct obd_device *obd = exp->exp_obd;
1426 struct lmv_obd *lmv = &obd->u.lmv;
1430 rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1434 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1435 u64 valid, const char *name,
1436 const char *input, int input_size, int output_size,
1437 int flags, struct ptlrpc_request **request)
1439 struct obd_device *obd = exp->exp_obd;
1440 struct lmv_obd *lmv = &obd->u.lmv;
1441 struct lmv_tgt_desc *tgt;
1445 tgt = lmv_find_target(lmv, fid);
1447 RETURN(PTR_ERR(tgt));
1449 rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1450 input_size, output_size, flags, request);
1455 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1456 u64 valid, const char *name,
1457 const char *input, int input_size, int output_size,
1458 int flags, __u32 suppgid,
1459 struct ptlrpc_request **request)
1461 struct obd_device *obd = exp->exp_obd;
1462 struct lmv_obd *lmv = &obd->u.lmv;
1463 struct lmv_tgt_desc *tgt;
1467 tgt = lmv_find_target(lmv, fid);
1469 RETURN(PTR_ERR(tgt));
1471 rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1472 input_size, output_size, flags, suppgid,
1478 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1479 struct ptlrpc_request **request)
1481 struct obd_device *obd = exp->exp_obd;
1482 struct lmv_obd *lmv = &obd->u.lmv;
1483 struct lmv_tgt_desc *tgt;
1487 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1489 RETURN(PTR_ERR(tgt));
1491 if (op_data->op_flags & MF_GET_MDT_IDX) {
1492 op_data->op_mds = tgt->ltd_idx;
1496 rc = md_getattr(tgt->ltd_exp, op_data, request);
1501 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1503 struct obd_device *obd = exp->exp_obd;
1504 struct lmv_obd *lmv = &obd->u.lmv;
1508 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1511 * With DNE every object can have two locks in different namespaces:
1512 * lookup lock in space of MDT storing direntry and update/open lock in
1513 * space of MDT storing inode.
1515 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1516 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1518 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1524 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1525 struct md_open_data *mod, struct ptlrpc_request **request)
1527 struct obd_device *obd = exp->exp_obd;
1528 struct lmv_obd *lmv = &obd->u.lmv;
1529 struct lmv_tgt_desc *tgt;
1533 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1535 RETURN(PTR_ERR(tgt));
1537 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1538 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1543 * Choosing the MDT by name or FID in @op_data.
1544 * For non-striped directory, it will locate MDT by fid.
1545 * For striped-directory, it will locate MDT by name. And also
1546 * it will reset op_fid1 with the FID of the choosen stripe.
1548 struct lmv_tgt_desc *
1549 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1550 const char *name, int namelen, struct lu_fid *fid,
1553 struct lmv_tgt_desc *tgt;
1554 const struct lmv_oinfo *oinfo;
1556 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1557 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1558 RETURN(ERR_PTR(-EBADF));
1559 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1561 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1563 RETURN(ERR_CAST(oinfo));
1567 *fid = oinfo->lmo_fid;
1569 *mds = oinfo->lmo_mds;
1571 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1573 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1574 PFID(&oinfo->lmo_fid));
1579 * Locate mds by fid or name
1581 * For striped directory (lsm != NULL), it will locate the stripe
1582 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1583 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1584 * walk through all of stripes to locate the entry.
1586 * For normal direcotry, it will locate MDS by FID directly.
1587 * \param[in] lmv LMV device
1588 * \param[in] op_data client MD stack parameters, name, namelen
1590 * \param[in] fid object FID used to locate MDS.
1592 * retval pointer to the lmv_tgt_desc if succeed.
1593 * ERR_PTR(errno) if failed.
1595 struct lmv_tgt_desc*
1596 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1599 struct lmv_stripe_md *lsm = op_data->op_mea1;
1600 struct lmv_tgt_desc *tgt;
1602 /* During creating VOLATILE file, it should honor the mdt
1603 * index if the file under striped dir is being restored, see
1605 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1606 (int)op_data->op_mds != -1) {
1608 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1613 /* refill the right parent fid */
1614 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1615 struct lmv_oinfo *oinfo;
1617 oinfo = &lsm->lsm_md_oinfo[i];
1618 if (oinfo->lmo_mds == op_data->op_mds) {
1619 *fid = oinfo->lmo_fid;
1624 if (i == lsm->lsm_md_stripe_count)
1625 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1631 if (lsm == NULL || op_data->op_namelen == 0) {
1632 tgt = lmv_find_target(lmv, fid);
1636 op_data->op_mds = tgt->ltd_idx;
1640 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1641 op_data->op_namelen, fid,
1645 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1646 const void *data, size_t datalen, umode_t mode, uid_t uid,
1647 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1648 struct ptlrpc_request **request)
1650 struct obd_device *obd = exp->exp_obd;
1651 struct lmv_obd *lmv = &obd->u.lmv;
1652 struct lmv_tgt_desc *tgt;
1656 if (!lmv->desc.ld_active_tgt_count)
1659 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1661 RETURN(PTR_ERR(tgt));
1663 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1664 (int)op_data->op_namelen, op_data->op_name,
1665 PFID(&op_data->op_fid1), op_data->op_mds);
1667 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1670 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1671 /* Send the create request to the MDT where the object
1672 * will be located */
1673 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1675 RETURN(PTR_ERR(tgt));
1677 op_data->op_mds = tgt->ltd_idx;
1679 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1682 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1683 PFID(&op_data->op_fid2), op_data->op_mds);
1685 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1686 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1687 cap_effective, rdev, request);
1689 if (*request == NULL)
1691 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1697 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1698 const union ldlm_policy_data *policy, struct md_op_data *op_data,
1699 struct lustre_handle *lockh, __u64 extra_lock_flags)
1701 struct obd_device *obd = exp->exp_obd;
1702 struct lmv_obd *lmv = &obd->u.lmv;
1703 struct lmv_tgt_desc *tgt;
1707 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1709 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1711 RETURN(PTR_ERR(tgt));
1713 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1714 PFID(&op_data->op_fid1), tgt->ltd_idx);
1716 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1723 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1724 struct ptlrpc_request **preq)
1726 struct ptlrpc_request *req = NULL;
1727 struct obd_device *obd = exp->exp_obd;
1728 struct lmv_obd *lmv = &obd->u.lmv;
1729 struct lmv_tgt_desc *tgt;
1730 struct mdt_body *body;
1734 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1736 RETURN(PTR_ERR(tgt));
1738 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1739 (int)op_data->op_namelen, op_data->op_name,
1740 PFID(&op_data->op_fid1), tgt->ltd_idx);
1742 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1746 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1747 LASSERT(body != NULL);
1749 if (body->mbo_valid & OBD_MD_MDS) {
1750 struct lu_fid rid = body->mbo_fid1;
1751 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1754 tgt = lmv_find_target(lmv, &rid);
1756 ptlrpc_req_finished(*preq);
1758 RETURN(PTR_ERR(tgt));
1761 op_data->op_fid1 = rid;
1762 op_data->op_valid |= OBD_MD_FLCROSSREF;
1763 op_data->op_namelen = 0;
1764 op_data->op_name = NULL;
1765 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1766 ptlrpc_req_finished(*preq);
1773 #define md_op_data_fid(op_data, fl) \
1774 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1775 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1776 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1777 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1780 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1781 struct md_op_data *op_data, __u32 op_tgt,
1782 enum ldlm_mode mode, int bits, int flag)
1784 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1785 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1786 union ldlm_policy_data policy = { { 0 } };
1790 if (!fid_is_sane(fid))
1794 tgt = lmv_find_target(lmv, fid);
1796 RETURN(PTR_ERR(tgt));
1799 if (tgt->ltd_idx != op_tgt) {
1800 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1801 policy.l_inodebits.bits = bits;
1802 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1803 mode, LCF_ASYNC, NULL);
1806 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1808 op_data->op_flags |= flag;
1816 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1819 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1820 struct ptlrpc_request **request)
1822 struct obd_device *obd = exp->exp_obd;
1823 struct lmv_obd *lmv = &obd->u.lmv;
1824 struct lmv_tgt_desc *tgt;
1828 LASSERT(op_data->op_namelen != 0);
1830 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1831 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1832 op_data->op_name, PFID(&op_data->op_fid1));
1834 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1835 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1836 op_data->op_cap = cfs_curproc_cap_pack();
1837 if (op_data->op_mea2 != NULL) {
1838 struct lmv_stripe_md *lsm = op_data->op_mea2;
1839 const struct lmv_oinfo *oinfo;
1841 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
1842 op_data->op_namelen);
1844 RETURN(PTR_ERR(oinfo));
1846 op_data->op_fid2 = oinfo->lmo_fid;
1849 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1851 RETURN(PTR_ERR(tgt));
1854 * Cancel UPDATE lock on child (fid1).
1856 op_data->op_flags |= MF_MDC_CANCEL_FID2;
1857 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
1858 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1862 rc = md_link(tgt->ltd_exp, op_data, request);
1867 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1868 const char *old, size_t oldlen,
1869 const char *new, size_t newlen,
1870 struct ptlrpc_request **request)
1872 struct obd_device *obd = exp->exp_obd;
1873 struct lmv_obd *lmv = &obd->u.lmv;
1874 struct lmv_tgt_desc *src_tgt;
1875 struct lmv_tgt_desc *tgt_tgt;
1876 struct obd_export *target_exp;
1877 struct mdt_body *body;
1881 LASSERT(oldlen != 0);
1883 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
1884 (int)oldlen, old, PFID(&op_data->op_fid1),
1885 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
1886 (int)newlen, new, PFID(&op_data->op_fid2),
1887 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
1889 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1890 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1891 op_data->op_cap = cfs_curproc_cap_pack();
1892 if (op_data->op_cli_flags & CLI_MIGRATE) {
1893 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
1894 PFID(&op_data->op_fid3));
1896 if (op_data->op_mea1 != NULL) {
1897 struct lmv_stripe_md *lsm = op_data->op_mea1;
1898 struct lmv_tgt_desc *tmp;
1900 /* Fix the parent fid for striped dir */
1901 tmp = lmv_locate_target_for_name(lmv, lsm, old,
1906 RETURN(PTR_ERR(tmp));
1909 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1913 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
1914 if (IS_ERR(src_tgt))
1915 RETURN(PTR_ERR(src_tgt));
1917 target_exp = src_tgt->ltd_exp;
1919 if (op_data->op_mea1 != NULL) {
1920 struct lmv_stripe_md *lsm = op_data->op_mea1;
1922 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
1927 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
1929 if (IS_ERR(src_tgt))
1930 RETURN(PTR_ERR(src_tgt));
1933 if (op_data->op_mea2 != NULL) {
1934 struct lmv_stripe_md *lsm = op_data->op_mea2;
1936 tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
1941 tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
1944 if (IS_ERR(tgt_tgt))
1945 RETURN(PTR_ERR(tgt_tgt));
1947 target_exp = tgt_tgt->ltd_exp;
1951 * LOOKUP lock on src child (fid3) should also be cancelled for
1952 * src_tgt in mdc_rename.
1954 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1957 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1960 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1961 LCK_EX, MDS_INODELOCK_UPDATE,
1962 MF_MDC_CANCEL_FID2);
1967 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
1969 if (fid_is_sane(&op_data->op_fid3)) {
1970 struct lmv_tgt_desc *tgt;
1972 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1974 RETURN(PTR_ERR(tgt));
1976 /* Cancel LOOKUP lock on its parent */
1977 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
1978 LCK_EX, MDS_INODELOCK_LOOKUP,
1979 MF_MDC_CANCEL_FID3);
1983 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1984 LCK_EX, MDS_INODELOCK_FULL,
1985 MF_MDC_CANCEL_FID3);
1992 * Cancel all the locks on tgt child (fid4).
1994 if (fid_is_sane(&op_data->op_fid4)) {
1995 struct lmv_tgt_desc *tgt;
1997 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1998 LCK_EX, MDS_INODELOCK_FULL,
1999 MF_MDC_CANCEL_FID4);
2003 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2005 RETURN(PTR_ERR(tgt));
2007 /* Since the target child might be destroyed, and it might
2008 * become orphan, and we can only check orphan on the local
2009 * MDT right now, so we send rename request to the MDT where
2010 * target child is located. If target child does not exist,
2011 * then it will send the request to the target parent */
2012 target_exp = tgt->ltd_exp;
2015 rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2018 if (rc != 0 && rc != -EXDEV)
2021 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2025 /* Not cross-ref case, just get out of here. */
2026 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2029 CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2030 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2032 op_data->op_fid4 = body->mbo_fid1;
2033 ptlrpc_req_finished(*request);
2038 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2039 void *ea, size_t ealen, struct ptlrpc_request **request)
2041 struct obd_device *obd = exp->exp_obd;
2042 struct lmv_obd *lmv = &obd->u.lmv;
2043 struct lmv_tgt_desc *tgt;
2047 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2048 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2050 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2051 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2053 RETURN(PTR_ERR(tgt));
2055 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2060 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2061 struct ptlrpc_request **request)
2063 struct obd_device *obd = exp->exp_obd;
2064 struct lmv_obd *lmv = &obd->u.lmv;
2065 struct lmv_tgt_desc *tgt;
2069 tgt = lmv_find_target(lmv, fid);
2071 RETURN(PTR_ERR(tgt));
2073 rc = md_fsync(tgt->ltd_exp, fid, request);
2078 * Get current minimum entry from striped directory
2080 * This function will search the dir entry, whose hash value is the
2081 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2082 * only being called for striped directory.
2084 * \param[in] exp export of LMV
2085 * \param[in] op_data parameters transferred beween client MD stack
2086 * stripe_information will be included in this
2088 * \param[in] cb_op ldlm callback being used in enqueue in
2090 * \param[in] hash_offset the hash value, which is used to locate
2091 * minum(closet) dir entry
2092 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2093 * index of last entry, so to avoid hash conflict
2094 * between stripes. It will also be used to
2095 * return the stripe index of current dir entry.
2096 * \param[in|out] entp the minum entry and it also is being used
2097 * to input the last dir entry to resolve the
2100 * \param[out] ppage the page which holds the minum entry
2102 * \retval = 0 get the entry successfully
2103 * negative errno (< 0) does not get the entry
2105 static int lmv_get_min_striped_entry(struct obd_export *exp,
2106 struct md_op_data *op_data,
2107 struct md_callback *cb_op,
2108 __u64 hash_offset, int *stripe_offset,
2109 struct lu_dirent **entp,
2110 struct page **ppage)
2112 struct obd_device *obd = exp->exp_obd;
2113 struct lmv_obd *lmv = &obd->u.lmv;
2114 struct lmv_stripe_md *lsm = op_data->op_mea1;
2115 struct lmv_tgt_desc *tgt;
2117 struct lu_dirent *min_ent = NULL;
2118 struct page *min_page = NULL;
2124 stripe_count = lsm->lsm_md_stripe_count;
2125 for (i = 0; i < stripe_count; i++) {
2126 struct lu_dirent *ent = NULL;
2127 struct page *page = NULL;
2128 struct lu_dirpage *dp;
2129 __u64 stripe_hash = hash_offset;
2131 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2133 GOTO(out, rc = PTR_ERR(tgt));
2135 /* op_data will be shared by each stripe, so we need
2136 * reset these value for each stripe */
2137 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2138 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2139 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2141 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2146 dp = page_address(page);
2147 for (ent = lu_dirent_start(dp); ent != NULL;
2148 ent = lu_dirent_next(ent)) {
2149 /* Skip dummy entry */
2150 if (le16_to_cpu(ent->lde_namelen) == 0)
2153 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2156 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2157 (*entp == ent || i < *stripe_offset))
2160 /* skip . and .. for other stripes */
2162 (strncmp(ent->lde_name, ".",
2163 le16_to_cpu(ent->lde_namelen)) == 0 ||
2164 strncmp(ent->lde_name, "..",
2165 le16_to_cpu(ent->lde_namelen)) == 0))
2171 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2177 /* reach the end of current stripe, go to next stripe */
2178 if (stripe_hash == MDS_DIR_END_OFF)
2184 if (min_ent != NULL) {
2185 if (le64_to_cpu(min_ent->lde_hash) >
2186 le64_to_cpu(ent->lde_hash)) {
2205 if (*ppage != NULL) {
2209 *stripe_offset = min_idx;
2216 * Build dir entry page from a striped directory
2218 * This function gets one entry by @offset from a striped directory. It will
2219 * read entries from all of stripes, and choose one closest to the required
2220 * offset(&offset). A few notes
2221 * 1. skip . and .. for non-zero stripes, because there can only have one .
2222 * and .. in a directory.
2223 * 2. op_data will be shared by all of stripes, instead of allocating new
2224 * one, so need to restore before reusing.
2225 * 3. release the entry page if that is not being chosen.
2227 * \param[in] exp obd export refer to LMV
2228 * \param[in] op_data hold those MD parameters of read_entry
2229 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2230 * \param[out] ldp the entry being read
2231 * \param[out] ppage the page holding the entry. Note: because the entry
2232 * will be accessed in upper layer, so we need hold the
2233 * page until the usages of entry is finished, see
2234 * ll_dir_entry_next.
2236 * retval =0 if get entry successfully
2237 * <0 cannot get entry
2239 static int lmv_read_striped_page(struct obd_export *exp,
2240 struct md_op_data *op_data,
2241 struct md_callback *cb_op,
2242 __u64 offset, struct page **ppage)
2244 struct lu_fid master_fid = op_data->op_fid1;
2245 struct inode *master_inode = op_data->op_data;
2246 __u64 hash_offset = offset;
2247 struct lu_dirpage *dp;
2248 struct page *min_ent_page = NULL;
2249 struct page *ent_page = NULL;
2250 struct lu_dirent *ent;
2253 struct lu_dirent *min_ent = NULL;
2254 struct lu_dirent *last_ent;
2259 /* Allocate a page and read entries from all of stripes and fill
2260 * the page by hash order */
2261 ent_page = alloc_page(GFP_KERNEL);
2262 if (ent_page == NULL)
2265 /* Initialize the entry page */
2266 dp = kmap(ent_page);
2267 memset(dp, 0, sizeof(*dp));
2268 dp->ldp_hash_start = cpu_to_le64(offset);
2269 dp->ldp_flags |= LDF_COLLIDE;
2272 left_bytes = PAGE_SIZE - sizeof(*dp);
2278 /* Find the minum entry from all sub-stripes */
2279 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2285 /* If it can not get minum entry, it means it already reaches
2286 * the end of this directory */
2287 if (min_ent == NULL) {
2288 last_ent->lde_reclen = 0;
2289 hash_offset = MDS_DIR_END_OFF;
2293 ent_size = le16_to_cpu(min_ent->lde_reclen);
2295 /* the last entry lde_reclen is 0, but it might not
2296 * the end of this entry of this temporay entry */
2298 ent_size = lu_dirent_calc_size(
2299 le16_to_cpu(min_ent->lde_namelen),
2300 le32_to_cpu(min_ent->lde_attrs));
2301 if (ent_size > left_bytes) {
2302 last_ent->lde_reclen = cpu_to_le16(0);
2303 hash_offset = le64_to_cpu(min_ent->lde_hash);
2307 memcpy(ent, min_ent, ent_size);
2309 /* Replace . with master FID and Replace .. with the parent FID
2310 * of master object */
2311 if (strncmp(ent->lde_name, ".",
2312 le16_to_cpu(ent->lde_namelen)) == 0 &&
2313 le16_to_cpu(ent->lde_namelen) == 1)
2314 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2315 else if (strncmp(ent->lde_name, "..",
2316 le16_to_cpu(ent->lde_namelen)) == 0 &&
2317 le16_to_cpu(ent->lde_namelen) == 2)
2318 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2320 left_bytes -= ent_size;
2321 ent->lde_reclen = cpu_to_le16(ent_size);
2323 ent = (void *)ent + ent_size;
2324 hash_offset = le64_to_cpu(min_ent->lde_hash);
2325 if (hash_offset == MDS_DIR_END_OFF) {
2326 last_ent->lde_reclen = 0;
2331 if (min_ent_page != NULL) {
2332 kunmap(min_ent_page);
2333 put_page(min_ent_page);
2336 if (unlikely(rc != 0)) {
2337 __free_page(ent_page);
2341 dp->ldp_flags |= LDF_EMPTY;
2342 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2343 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2346 /* We do not want to allocate md_op_data during each
2347 * dir entry reading, so op_data will be shared by every stripe,
2348 * then we need to restore it back to original value before
2349 * return to the upper layer */
2350 op_data->op_fid1 = master_fid;
2351 op_data->op_fid2 = master_fid;
2352 op_data->op_data = master_inode;
2359 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2360 struct md_callback *cb_op, __u64 offset,
2361 struct page **ppage)
2363 struct obd_device *obd = exp->exp_obd;
2364 struct lmv_obd *lmv = &obd->u.lmv;
2365 struct lmv_stripe_md *lsm = op_data->op_mea1;
2366 struct lmv_tgt_desc *tgt;
2370 if (unlikely(lsm != NULL)) {
2371 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2375 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2377 RETURN(PTR_ERR(tgt));
2379 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2385 * Unlink a file/directory
2387 * Unlink a file or directory under the parent dir. The unlink request
2388 * usually will be sent to the MDT where the child is located, but if
2389 * the client does not have the child FID then request will be sent to the
2390 * MDT where the parent is located.
2392 * If the parent is a striped directory then it also needs to locate which
2393 * stripe the name of the child is located, and replace the parent FID
2394 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2395 * it will walk through all of sub-stripes until the child is being
2398 * \param[in] exp export refer to LMV
2399 * \param[in] op_data different parameters transferred beween client
2400 * MD stacks, name, namelen, FIDs etc.
2401 * op_fid1 is the parent FID, op_fid2 is the child
2403 * \param[out] request point to the request of unlink.
2405 * retval 0 if succeed
2406 * negative errno if failed.
2408 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2409 struct ptlrpc_request **request)
2411 struct obd_device *obd = exp->exp_obd;
2412 struct lmv_obd *lmv = &obd->u.lmv;
2413 struct lmv_tgt_desc *tgt = NULL;
2414 struct lmv_tgt_desc *parent_tgt = NULL;
2415 struct mdt_body *body;
2417 int stripe_index = 0;
2418 struct lmv_stripe_md *lsm = op_data->op_mea1;
2422 /* For striped dir, we need to locate the parent as well */
2424 struct lmv_tgt_desc *tmp;
2426 LASSERT(op_data->op_name != NULL &&
2427 op_data->op_namelen != 0);
2429 tmp = lmv_locate_target_for_name(lmv, lsm,
2431 op_data->op_namelen,
2435 /* return -EBADFD means unknown hash type, might
2436 * need try all sub-stripe here */
2437 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2438 RETURN(PTR_ERR(tmp));
2440 /* Note: both migrating dir and unknown hash dir need to
2441 * try all of sub-stripes, so we need start search the
2442 * name from stripe 0, but migrating dir is already handled
2443 * inside lmv_locate_target_for_name(), so we only check
2444 * unknown hash type directory here */
2445 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2446 struct lmv_oinfo *oinfo;
2448 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2450 op_data->op_fid1 = oinfo->lmo_fid;
2451 op_data->op_mds = oinfo->lmo_mds;
2456 /* Send unlink requests to the MDT where the child is located */
2457 if (likely(!fid_is_zero(&op_data->op_fid2)))
2458 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2459 else if (lsm != NULL)
2460 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2462 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2465 RETURN(PTR_ERR(tgt));
2467 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2468 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2469 op_data->op_cap = cfs_curproc_cap_pack();
2472 * If child's fid is given, cancel unused locks for it if it is from
2473 * another export than parent.
2475 * LOOKUP lock for child (fid3) should also be cancelled on parent
2476 * tgt_tgt in mdc_unlink().
2478 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2481 * Cancel FULL locks on child (fid3).
2483 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2484 if (IS_ERR(parent_tgt))
2485 RETURN(PTR_ERR(parent_tgt));
2487 if (parent_tgt != tgt) {
2488 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2489 LCK_EX, MDS_INODELOCK_LOOKUP,
2490 MF_MDC_CANCEL_FID3);
2493 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2494 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2498 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2499 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2501 rc = md_unlink(tgt->ltd_exp, op_data, request);
2502 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2505 /* Try next stripe if it is needed. */
2506 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2507 struct lmv_oinfo *oinfo;
2510 if (stripe_index >= lsm->lsm_md_stripe_count)
2513 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2515 op_data->op_fid1 = oinfo->lmo_fid;
2516 op_data->op_mds = oinfo->lmo_mds;
2518 ptlrpc_req_finished(*request);
2521 goto try_next_stripe;
2524 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2528 /* Not cross-ref case, just get out of here. */
2529 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2532 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2533 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2535 /* This is a remote object, try remote MDT, Note: it may
2536 * try more than 1 time here, Considering following case
2537 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2538 * 1. Initially A does not know where remote1 is, it send
2539 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2540 * resend unlink RPC to MDT1 (retry 1st time).
2542 * 2. During the unlink RPC in flight,
2543 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2544 * and create new remote1, but on MDT0
2546 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2547 * /mnt/lustre, then lookup get fid of remote1, and find
2548 * it is remote dir again, and replay -EREMOTE again.
2550 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2552 * In theory, it might try unlimited time here, but it should
2553 * be very rare case. */
2554 op_data->op_fid2 = body->mbo_fid1;
2555 ptlrpc_req_finished(*request);
2561 static int lmv_precleanup(struct obd_device *obd)
2564 fld_client_proc_fini(&obd->u.lmv.lmv_fld);
2565 lprocfs_obd_cleanup(obd);
2566 lprocfs_free_md_stats(obd);
2571 * Get by key a value associated with a LMV device.
2573 * Dispatch request to lower-layer devices as needed.
2575 * \param[in] env execution environment for this thread
2576 * \param[in] exp export for the LMV device
2577 * \param[in] keylen length of key identifier
2578 * \param[in] key identifier of key to get value for
2579 * \param[in] vallen size of \a val
2580 * \param[out] val pointer to storage location for value
2581 * \param[in] lsm optional striping metadata of object
2583 * \retval 0 on success
2584 * \retval negative negated errno on failure
2586 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2587 __u32 keylen, void *key, __u32 *vallen, void *val)
2589 struct obd_device *obd;
2590 struct lmv_obd *lmv;
2594 obd = class_exp2obd(exp);
2596 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2597 exp->exp_handle.h_cookie);
2602 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2605 LASSERT(*vallen == sizeof(__u32));
2606 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2607 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2609 * All tgts should be connected when this gets called.
2611 if (tgt == NULL || tgt->ltd_exp == NULL)
2614 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2619 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2620 KEY_IS(KEY_DEFAULT_EASIZE) ||
2621 KEY_IS(KEY_CONN_DATA)) {
2623 * Forwarding this request to first MDS, it should know LOV
2626 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2628 if (!rc && KEY_IS(KEY_CONN_DATA))
2629 exp->exp_connect_data = *(struct obd_connect_data *)val;
2631 } else if (KEY_IS(KEY_TGT_COUNT)) {
2632 *((int *)val) = lmv->desc.ld_tgt_count;
2636 CDEBUG(D_IOCTL, "Invalid key\n");
2641 * Asynchronously set by key a value associated with a LMV device.
2643 * Dispatch request to lower-layer devices as needed.
2645 * \param[in] env execution environment for this thread
2646 * \param[in] exp export for the LMV device
2647 * \param[in] keylen length of key identifier
2648 * \param[in] key identifier of key to store value for
2649 * \param[in] vallen size of value to store
2650 * \param[in] val pointer to data to be stored
2651 * \param[in] set optional list of related ptlrpc requests
2653 * \retval 0 on success
2654 * \retval negative negated errno on failure
2656 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2657 __u32 keylen, void *key, __u32 vallen, void *val,
2658 struct ptlrpc_request_set *set)
2660 struct lmv_tgt_desc *tgt = NULL;
2661 struct obd_device *obd;
2662 struct lmv_obd *lmv;
2666 obd = class_exp2obd(exp);
2668 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2669 exp->exp_handle.h_cookie);
2674 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2675 KEY_IS(KEY_DEFAULT_EASIZE)) {
2678 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2681 if (tgt == NULL || tgt->ltd_exp == NULL)
2684 err = obd_set_info_async(env, tgt->ltd_exp,
2685 keylen, key, vallen, val, set);
2696 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2697 const struct lmv_mds_md_v1 *lmm1)
2699 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2706 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2707 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2708 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2709 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2710 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2712 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2713 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2714 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2715 sizeof(lsm->lsm_md_pool_name));
2717 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2720 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2721 "layout_version %d\n", lsm->lsm_md_stripe_count,
2722 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2723 lsm->lsm_md_layout_version);
2725 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2726 for (i = 0; i < stripe_count; i++) {
2727 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2728 &lmm1->lmv_stripe_fids[i]);
2729 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2730 &lsm->lsm_md_oinfo[i].lmo_mds);
2733 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2734 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2740 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2741 const union lmv_mds_md *lmm, size_t lmm_size)
2743 struct lmv_stripe_md *lsm;
2746 bool allocated = false;
2749 LASSERT(lsmp != NULL);
2753 if (lsm != NULL && lmm == NULL) {
2755 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2756 /* For migrating inode, the master stripe and master
2757 * object will be the same, so do not need iput, see
2758 * ll_update_lsm_md */
2759 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2760 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2761 iput(lsm->lsm_md_oinfo[i].lmo_root);
2763 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2764 OBD_FREE(lsm, lsm_size);
2769 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2773 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2774 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2775 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2776 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2781 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2782 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2785 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2786 * stripecount should be 0 then.
2788 lsm_size = lmv_stripe_md_size(0);
2790 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2792 OBD_ALLOC(lsm, lsm_size);
2799 switch (le32_to_cpu(lmm->lmv_magic)) {
2801 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2804 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2805 le32_to_cpu(lmm->lmv_magic));
2810 if (rc != 0 && allocated) {
2811 OBD_FREE(lsm, lsm_size);
2818 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2820 lmv_unpackmd(NULL, &lsm, NULL, 0);
2822 EXPORT_SYMBOL(lmv_free_memmd);
2824 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2825 union ldlm_policy_data *policy,
2826 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
2829 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2834 LASSERT(fid != NULL);
2836 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2837 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2840 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2843 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2851 static int lmv_set_lock_data(struct obd_export *exp,
2852 const struct lustre_handle *lockh,
2853 void *data, __u64 *bits)
2855 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2856 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2860 if (tgt == NULL || tgt->ltd_exp == NULL)
2862 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
2866 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
2867 const struct lu_fid *fid, enum ldlm_type type,
2868 union ldlm_policy_data *policy,
2869 enum ldlm_mode mode, struct lustre_handle *lockh)
2871 struct obd_device *obd = exp->exp_obd;
2872 struct lmv_obd *lmv = &obd->u.lmv;
2878 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2881 * With DNE every object can have two locks in different namespaces:
2882 * lookup lock in space of MDT storing direntry and update/open lock in
2883 * space of MDT storing inode. Try the MDT that the FID maps to first,
2884 * since this can be easily found, and only try others if that fails.
2886 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
2887 i < lmv->desc.ld_tgt_count;
2888 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
2890 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
2891 obd->obd_name, PFID(fid), tgt);
2895 if (lmv->tgts[tgt] == NULL ||
2896 lmv->tgts[tgt]->ltd_exp == NULL ||
2897 lmv->tgts[tgt]->ltd_active == 0)
2900 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
2901 type, policy, mode, lockh);
2909 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2910 struct obd_export *dt_exp, struct obd_export *md_exp,
2911 struct lustre_md *md)
2913 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2914 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2916 if (tgt == NULL || tgt->ltd_exp == NULL)
2919 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2922 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2924 struct obd_device *obd = exp->exp_obd;
2925 struct lmv_obd *lmv = &obd->u.lmv;
2926 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2929 if (md->lmv != NULL) {
2930 lmv_free_memmd(md->lmv);
2933 if (tgt == NULL || tgt->ltd_exp == NULL)
2935 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2938 int lmv_set_open_replay_data(struct obd_export *exp,
2939 struct obd_client_handle *och,
2940 struct lookup_intent *it)
2942 struct obd_device *obd = exp->exp_obd;
2943 struct lmv_obd *lmv = &obd->u.lmv;
2944 struct lmv_tgt_desc *tgt;
2947 tgt = lmv_find_target(lmv, &och->och_fid);
2949 RETURN(PTR_ERR(tgt));
2951 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
2954 int lmv_clear_open_replay_data(struct obd_export *exp,
2955 struct obd_client_handle *och)
2957 struct obd_device *obd = exp->exp_obd;
2958 struct lmv_obd *lmv = &obd->u.lmv;
2959 struct lmv_tgt_desc *tgt;
2962 tgt = lmv_find_target(lmv, &och->och_fid);
2964 RETURN(PTR_ERR(tgt));
2966 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2969 int lmv_intent_getattr_async(struct obd_export *exp,
2970 struct md_enqueue_info *minfo)
2972 struct md_op_data *op_data = &minfo->mi_data;
2973 struct obd_device *obd = exp->exp_obd;
2974 struct lmv_obd *lmv = &obd->u.lmv;
2975 struct lmv_tgt_desc *ptgt = NULL;
2976 struct lmv_tgt_desc *ctgt = NULL;
2980 if (!fid_is_sane(&op_data->op_fid2))
2983 ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2985 RETURN(PTR_ERR(ptgt));
2987 ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2989 RETURN(PTR_ERR(ctgt));
2992 * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
2993 * lock on parent, and UPDATE lock on child MDT, which makes all
2994 * complicated. Considering remote dir is rare case, and not supporting
2995 * it in statahead won't cause any issue, drop its support for now.
3000 rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3004 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3005 struct lu_fid *fid, __u64 *bits)
3007 struct obd_device *obd = exp->exp_obd;
3008 struct lmv_obd *lmv = &obd->u.lmv;
3009 struct lmv_tgt_desc *tgt;
3013 tgt = lmv_find_target(lmv, fid);
3015 RETURN(PTR_ERR(tgt));
3017 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3021 int lmv_get_fid_from_lsm(struct obd_export *exp,
3022 const struct lmv_stripe_md *lsm,
3023 const char *name, int namelen, struct lu_fid *fid)
3025 const struct lmv_oinfo *oinfo;
3027 LASSERT(lsm != NULL);
3028 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3030 return PTR_ERR(oinfo);
3032 *fid = oinfo->lmo_fid;
3038 * For lmv, only need to send request to master MDT, and the master MDT will
3039 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3040 * we directly fetch data from the slave MDTs.
3042 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3043 struct obd_quotactl *oqctl)
3045 struct obd_device *obd = class_exp2obd(exp);
3046 struct lmv_obd *lmv = &obd->u.lmv;
3047 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3050 __u64 curspace, curinodes;
3054 tgt->ltd_exp == NULL ||
3056 lmv->desc.ld_tgt_count == 0) {
3057 CERROR("master lmv inactive\n");
3061 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3062 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3066 curspace = curinodes = 0;
3067 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3071 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3074 err = obd_quotactl(tgt->ltd_exp, oqctl);
3076 CERROR("getquota on mdt %d failed. %d\n", i, err);
3080 curspace += oqctl->qc_dqblk.dqb_curspace;
3081 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3084 oqctl->qc_dqblk.dqb_curspace = curspace;
3085 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3090 static int lmv_merge_attr(struct obd_export *exp,
3091 const struct lmv_stripe_md *lsm,
3092 struct cl_attr *attr,
3093 ldlm_blocking_callback cb_blocking)
3098 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3102 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3103 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3105 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3106 " atime %lu ctime %lu, mtime %lu.\n",
3107 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3108 i_size_read(inode), (unsigned long long)inode->i_blocks,
3109 inode->i_nlink, LTIME_S(inode->i_atime),
3110 LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3112 /* for slave stripe, it needs to subtract nlink for . and .. */
3114 attr->cat_nlink += inode->i_nlink - 2;
3116 attr->cat_nlink = inode->i_nlink;
3118 attr->cat_size += i_size_read(inode);
3119 attr->cat_blocks += inode->i_blocks;
3121 if (attr->cat_atime < LTIME_S(inode->i_atime))
3122 attr->cat_atime = LTIME_S(inode->i_atime);
3124 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3125 attr->cat_ctime = LTIME_S(inode->i_ctime);
3127 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3128 attr->cat_mtime = LTIME_S(inode->i_mtime);
3133 struct obd_ops lmv_obd_ops = {
3134 .o_owner = THIS_MODULE,
3135 .o_setup = lmv_setup,
3136 .o_cleanup = lmv_cleanup,
3137 .o_precleanup = lmv_precleanup,
3138 .o_process_config = lmv_process_config,
3139 .o_connect = lmv_connect,
3140 .o_disconnect = lmv_disconnect,
3141 .o_statfs = lmv_statfs,
3142 .o_get_info = lmv_get_info,
3143 .o_set_info_async = lmv_set_info_async,
3144 .o_notify = lmv_notify,
3145 .o_get_uuid = lmv_get_uuid,
3146 .o_iocontrol = lmv_iocontrol,
3147 .o_quotactl = lmv_quotactl
3150 struct md_ops lmv_md_ops = {
3151 .m_get_root = lmv_get_root,
3152 .m_null_inode = lmv_null_inode,
3153 .m_close = lmv_close,
3154 .m_create = lmv_create,
3155 .m_enqueue = lmv_enqueue,
3156 .m_getattr = lmv_getattr,
3157 .m_getxattr = lmv_getxattr,
3158 .m_getattr_name = lmv_getattr_name,
3159 .m_intent_lock = lmv_intent_lock,
3161 .m_rename = lmv_rename,
3162 .m_setattr = lmv_setattr,
3163 .m_setxattr = lmv_setxattr,
3164 .m_fsync = lmv_fsync,
3165 .m_read_page = lmv_read_page,
3166 .m_unlink = lmv_unlink,
3167 .m_init_ea_size = lmv_init_ea_size,
3168 .m_cancel_unused = lmv_cancel_unused,
3169 .m_set_lock_data = lmv_set_lock_data,
3170 .m_lock_match = lmv_lock_match,
3171 .m_get_lustre_md = lmv_get_lustre_md,
3172 .m_free_lustre_md = lmv_free_lustre_md,
3173 .m_merge_attr = lmv_merge_attr,
3174 .m_set_open_replay_data = lmv_set_open_replay_data,
3175 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3176 .m_intent_getattr_async = lmv_intent_getattr_async,
3177 .m_revalidate_lock = lmv_revalidate_lock,
3178 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3179 .m_unpackmd = lmv_unpackmd,
3182 static int __init lmv_init(void)
3184 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3185 LUSTRE_LMV_NAME, NULL);
3188 static void __exit lmv_exit(void)
3190 class_unregister_type(LUSTRE_LMV_NAME);
3193 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3194 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3195 MODULE_VERSION(LUSTRE_VERSION_STRING);
3196 MODULE_LICENSE("GPL");
3198 module_init(lmv_init);
3199 module_exit(lmv_exit);