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, void *data)
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, data);
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,
354 (void *)(tgt - lmv->tgts[0]));
356 obd_disconnect(mdc_exp);
362 tgt->ltd_exp = mdc_exp;
363 lmv->desc.ld_active_tgt_count++;
365 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
367 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
368 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
369 atomic_read(&obd->obd_refcount));
371 if (lmv->targets_proc_entry != NULL) {
372 struct proc_dir_entry *mdc_symlink;
374 LASSERT(mdc_obd->obd_type != NULL);
375 LASSERT(mdc_obd->obd_type->typ_name != NULL);
376 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
377 lmv->targets_proc_entry,
379 mdc_obd->obd_type->typ_name,
381 if (mdc_symlink == NULL) {
382 CERROR("cannot register LMV target "
383 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
384 obd->obd_type->typ_name, obd->obd_name,
391 static void lmv_del_target(struct lmv_obd *lmv, int index)
393 if (lmv->tgts[index] == NULL)
396 OBD_FREE_PTR(lmv->tgts[index]);
397 lmv->tgts[index] = NULL;
401 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
402 __u32 index, int gen)
404 struct obd_device *mdc_obd;
405 struct lmv_obd *lmv = &obd->u.lmv;
406 struct lmv_tgt_desc *tgt;
407 int orig_tgt_count = 0;
411 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
412 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
415 CERROR("%s: Target %s not attached: rc = %d\n",
416 obd->obd_name, uuidp->uuid, -EINVAL);
420 mutex_lock(&lmv->lmv_init_mutex);
421 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
422 tgt = lmv->tgts[index];
423 CERROR("%s: UUID %s already assigned at LOV target index %d:"
424 " rc = %d\n", obd->obd_name,
425 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
426 mutex_unlock(&lmv->lmv_init_mutex);
430 if (index >= lmv->tgts_size) {
431 /* We need to reallocate the lmv target array. */
432 struct lmv_tgt_desc **newtgts, **old = NULL;
436 while (newsize < index + 1)
437 newsize = newsize << 1;
438 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
439 if (newtgts == NULL) {
440 mutex_unlock(&lmv->lmv_init_mutex);
444 if (lmv->tgts_size) {
445 memcpy(newtgts, lmv->tgts,
446 sizeof(*newtgts) * lmv->tgts_size);
448 oldsize = lmv->tgts_size;
452 lmv->tgts_size = newsize;
455 OBD_FREE(old, sizeof(*old) * oldsize);
457 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
463 mutex_unlock(&lmv->lmv_init_mutex);
467 mutex_init(&tgt->ltd_fid_mutex);
468 tgt->ltd_idx = index;
469 tgt->ltd_uuid = *uuidp;
471 lmv->tgts[index] = tgt;
472 if (index >= lmv->desc.ld_tgt_count) {
473 orig_tgt_count = lmv->desc.ld_tgt_count;
474 lmv->desc.ld_tgt_count = index + 1;
477 if (lmv->connected == 0) {
478 /* lmv_check_connect() will connect this target. */
479 mutex_unlock(&lmv->lmv_init_mutex);
483 /* Otherwise let's connect it ourselves */
484 mutex_unlock(&lmv->lmv_init_mutex);
485 rc = lmv_connect_mdc(obd, tgt);
487 spin_lock(&lmv->lmv_lock);
488 if (lmv->desc.ld_tgt_count == index + 1)
489 lmv->desc.ld_tgt_count = orig_tgt_count;
490 memset(tgt, 0, sizeof(*tgt));
491 spin_unlock(&lmv->lmv_lock);
493 int easize = sizeof(struct lmv_stripe_md) +
494 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
495 lmv_init_ea_size(obd->obd_self_export, easize, 0);
501 static int lmv_check_connect(struct obd_device *obd)
503 struct lmv_obd *lmv = &obd->u.lmv;
504 struct lmv_tgt_desc *tgt;
513 mutex_lock(&lmv->lmv_init_mutex);
514 if (lmv->connected) {
515 mutex_unlock(&lmv->lmv_init_mutex);
519 if (lmv->desc.ld_tgt_count == 0) {
520 mutex_unlock(&lmv->lmv_init_mutex);
521 CERROR("%s: no targets configured.\n", obd->obd_name);
525 LASSERT(lmv->tgts != NULL);
527 if (lmv->tgts[0] == NULL) {
528 mutex_unlock(&lmv->lmv_init_mutex);
529 CERROR("%s: no target configured for index 0.\n",
534 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
535 lmv->cluuid.uuid, obd->obd_name);
537 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
541 rc = lmv_connect_mdc(obd, tgt);
547 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
548 lmv_init_ea_size(obd->obd_self_export, easize, 0);
549 mutex_unlock(&lmv->lmv_init_mutex);
560 --lmv->desc.ld_active_tgt_count;
561 rc2 = obd_disconnect(tgt->ltd_exp);
563 CERROR("LMV target %s disconnect on "
564 "MDC idx %d: error %d\n",
565 tgt->ltd_uuid.uuid, i, rc2);
570 mutex_unlock(&lmv->lmv_init_mutex);
575 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
577 struct lmv_obd *lmv = &obd->u.lmv;
578 struct obd_device *mdc_obd;
582 LASSERT(tgt != NULL);
583 LASSERT(obd != NULL);
585 mdc_obd = class_exp2obd(tgt->ltd_exp);
588 mdc_obd->obd_force = obd->obd_force;
589 mdc_obd->obd_fail = obd->obd_fail;
590 mdc_obd->obd_no_recov = obd->obd_no_recov;
592 if (lmv->targets_proc_entry != NULL)
593 lprocfs_remove_proc_entry(mdc_obd->obd_name,
594 lmv->targets_proc_entry);
597 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
599 CERROR("Can't finanize fids factory\n");
601 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
602 tgt->ltd_exp->exp_obd->obd_name,
603 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
605 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
606 rc = obd_disconnect(tgt->ltd_exp);
608 if (tgt->ltd_active) {
609 CERROR("Target %s disconnect error %d\n",
610 tgt->ltd_uuid.uuid, rc);
614 lmv_activate_target(lmv, tgt, 0);
619 static int lmv_disconnect(struct obd_export *exp)
621 struct obd_device *obd = class_exp2obd(exp);
622 struct lmv_obd *lmv = &obd->u.lmv;
630 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
631 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
634 lmv_disconnect_mdc(obd, lmv->tgts[i]);
637 if (lmv->targets_proc_entry != NULL)
638 lprocfs_remove(&lmv->targets_proc_entry);
640 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
641 obd->obd_type->typ_name, obd->obd_name);
645 * This is the case when no real connection is established by
646 * lmv_check_connect().
649 class_export_put(exp);
650 rc = class_disconnect(exp);
656 static int lmv_fid2path(struct obd_export *exp, int len, void *karg,
659 struct obd_device *obddev = class_exp2obd(exp);
660 struct lmv_obd *lmv = &obddev->u.lmv;
661 struct getinfo_fid2path *gf;
662 struct lmv_tgt_desc *tgt;
663 struct getinfo_fid2path *remote_gf = NULL;
664 struct lu_fid root_fid;
665 int remote_gf_size = 0;
669 tgt = lmv_find_target(lmv, &gf->gf_fid);
671 RETURN(PTR_ERR(tgt));
673 root_fid = *gf->gf_u.gf_root_fid;
674 LASSERT(fid_is_sane(&root_fid));
677 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
678 if (rc != 0 && rc != -EREMOTE)
679 GOTO(out_fid2path, rc);
681 /* If remote_gf != NULL, it means just building the
682 * path on the remote MDT, copy this path segement to gf */
683 if (remote_gf != NULL) {
684 struct getinfo_fid2path *ori_gf;
687 ori_gf = (struct getinfo_fid2path *)karg;
688 if (strlen(ori_gf->gf_u.gf_path) +
689 strlen(gf->gf_u.gf_path) > ori_gf->gf_pathlen)
690 GOTO(out_fid2path, rc = -EOVERFLOW);
692 ptr = ori_gf->gf_u.gf_path;
694 memmove(ptr + strlen(gf->gf_u.gf_path) + 1, ptr,
695 strlen(ori_gf->gf_u.gf_path));
697 strncpy(ptr, gf->gf_u.gf_path,
698 strlen(gf->gf_u.gf_path));
699 ptr += strlen(gf->gf_u.gf_path);
703 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: %llu ln: %u\n",
704 tgt->ltd_exp->exp_obd->obd_name,
705 gf->gf_u.gf_path, PFID(&gf->gf_fid), gf->gf_recno,
709 GOTO(out_fid2path, rc);
711 /* sigh, has to go to another MDT to do path building further */
712 if (remote_gf == NULL) {
713 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
714 OBD_ALLOC(remote_gf, remote_gf_size);
715 if (remote_gf == NULL)
716 GOTO(out_fid2path, rc = -ENOMEM);
717 remote_gf->gf_pathlen = PATH_MAX;
720 if (!fid_is_sane(&gf->gf_fid)) {
721 CERROR("%s: invalid FID "DFID": rc = %d\n",
722 tgt->ltd_exp->exp_obd->obd_name,
723 PFID(&gf->gf_fid), -EINVAL);
724 GOTO(out_fid2path, rc = -EINVAL);
727 tgt = lmv_find_target(lmv, &gf->gf_fid);
729 GOTO(out_fid2path, rc = -EINVAL);
731 remote_gf->gf_fid = gf->gf_fid;
732 remote_gf->gf_recno = -1;
733 remote_gf->gf_linkno = -1;
734 memset(remote_gf->gf_u.gf_path, 0, remote_gf->gf_pathlen);
735 *remote_gf->gf_u.gf_root_fid = root_fid;
737 goto repeat_fid2path;
740 if (remote_gf != NULL)
741 OBD_FREE(remote_gf, remote_gf_size);
745 static int lmv_hsm_req_count(struct lmv_obd *lmv,
746 const struct hsm_user_request *hur,
747 const struct lmv_tgt_desc *tgt_mds)
751 struct lmv_tgt_desc *curr_tgt;
753 /* count how many requests must be sent to the given target */
754 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
755 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
756 if (IS_ERR(curr_tgt))
757 RETURN(PTR_ERR(curr_tgt));
758 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
764 static int lmv_hsm_req_build(struct lmv_obd *lmv,
765 struct hsm_user_request *hur_in,
766 const struct lmv_tgt_desc *tgt_mds,
767 struct hsm_user_request *hur_out)
770 struct lmv_tgt_desc *curr_tgt;
772 /* build the hsm_user_request for the given target */
773 hur_out->hur_request = hur_in->hur_request;
775 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
776 curr_tgt = lmv_find_target(lmv,
777 &hur_in->hur_user_item[i].hui_fid);
778 if (IS_ERR(curr_tgt))
779 RETURN(PTR_ERR(curr_tgt));
780 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
781 hur_out->hur_user_item[nr_out] =
782 hur_in->hur_user_item[i];
786 hur_out->hur_request.hr_itemcount = nr_out;
787 memcpy(hur_data(hur_out), hur_data(hur_in),
788 hur_in->hur_request.hr_data_len);
793 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
794 struct lustre_kernelcomm *lk,
801 /* unregister request (call from llapi_hsm_copytool_fini) */
802 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
803 struct lmv_tgt_desc *tgt = lmv->tgts[i];
805 if (tgt == NULL || tgt->ltd_exp == NULL)
807 /* best effort: try to clean as much as possible
808 * (continue on error) */
809 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
812 /* Whatever the result, remove copytool from kuc groups.
813 * Unreached coordinators will get EPIPE on next requests
814 * and will unregister automatically.
816 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group);
821 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
822 struct lustre_kernelcomm *lk, __user void *uarg)
827 bool any_set = false;
828 struct kkuc_ct_data kcd = { 0 };
831 /* All or nothing: try to register to all MDS.
832 * In case of failure, unregister from previous MDS,
833 * except if it because of inactive target. */
834 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
835 struct lmv_tgt_desc *tgt = lmv->tgts[i];
837 if (tgt == NULL || tgt->ltd_exp == NULL)
839 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
841 if (tgt->ltd_active) {
842 /* permanent error */
843 CERROR("%s: iocontrol MDC %s on MDT"
844 " idx %d cmd %x: err = %d\n",
845 lmv2obd_dev(lmv)->obd_name,
846 tgt->ltd_uuid.uuid, i, cmd, err);
848 lk->lk_flags |= LK_FLG_STOP;
849 /* unregister from previous MDS */
850 for (j = 0; j < i; j++) {
852 if (tgt == NULL || tgt->ltd_exp == NULL)
854 obd_iocontrol(cmd, tgt->ltd_exp, len,
859 /* else: transient error.
860 * kuc will register to the missing MDT
868 /* no registration done: return error */
871 /* at least one registration done, with no failure */
872 filp = fget(lk->lk_wfd);
876 kcd.kcd_magic = KKUC_CT_DATA_MAGIC;
877 kcd.kcd_uuid = lmv->cluuid;
878 kcd.kcd_archive = lk->lk_data;
880 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group,
891 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
892 int len, void *karg, void __user *uarg)
894 struct obd_device *obddev = class_exp2obd(exp);
895 struct lmv_obd *lmv = &obddev->u.lmv;
896 struct lmv_tgt_desc *tgt = NULL;
900 __u32 count = lmv->desc.ld_tgt_count;
907 case IOC_OBD_STATFS: {
908 struct obd_ioctl_data *data = karg;
909 struct obd_device *mdc_obd;
910 struct obd_statfs stat_buf = {0};
913 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
914 if ((index >= count))
917 tgt = lmv->tgts[index];
918 if (tgt == NULL || !tgt->ltd_active)
921 mdc_obd = class_exp2obd(tgt->ltd_exp);
926 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
927 min((int) data->ioc_plen2,
928 (int) sizeof(struct obd_uuid))))
931 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
932 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
936 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
937 min((int) data->ioc_plen1,
938 (int) sizeof(stat_buf))))
942 case OBD_IOC_QUOTACTL: {
943 struct if_quotactl *qctl = karg;
944 struct obd_quotactl *oqctl;
946 if (qctl->qc_valid == QC_MDTIDX) {
947 if (count <= qctl->qc_idx)
950 tgt = lmv->tgts[qctl->qc_idx];
951 if (tgt == NULL || tgt->ltd_exp == NULL)
953 } else if (qctl->qc_valid == QC_UUID) {
954 for (i = 0; i < count; i++) {
958 if (!obd_uuid_equals(&tgt->ltd_uuid,
962 if (tgt->ltd_exp == NULL)
974 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
975 OBD_ALLOC_PTR(oqctl);
979 QCTL_COPY(oqctl, qctl);
980 rc = obd_quotactl(tgt->ltd_exp, oqctl);
982 QCTL_COPY(qctl, oqctl);
983 qctl->qc_valid = QC_MDTIDX;
984 qctl->obd_uuid = tgt->ltd_uuid;
989 case OBD_IOC_CHANGELOG_SEND:
990 case OBD_IOC_CHANGELOG_CLEAR: {
991 struct ioc_changelog *icc = karg;
993 if (icc->icc_mdtindex >= count)
996 tgt = lmv->tgts[icc->icc_mdtindex];
997 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
999 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1002 case LL_IOC_GET_CONNECT_FLAGS: {
1004 if (tgt == NULL || tgt->ltd_exp == NULL)
1006 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1009 case LL_IOC_FID2MDTIDX: {
1010 struct lu_fid *fid = karg;
1013 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1017 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1018 * point to user space memory for FID2MDTIDX. */
1019 *(__u32 *)uarg = mdt_index;
1022 case OBD_IOC_FID2PATH: {
1023 rc = lmv_fid2path(exp, len, karg, uarg);
1026 case LL_IOC_HSM_STATE_GET:
1027 case LL_IOC_HSM_STATE_SET:
1028 case LL_IOC_HSM_ACTION: {
1029 struct md_op_data *op_data = karg;
1031 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1033 RETURN(PTR_ERR(tgt));
1035 if (tgt->ltd_exp == NULL)
1038 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1041 case LL_IOC_HSM_PROGRESS: {
1042 const struct hsm_progress_kernel *hpk = karg;
1044 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1046 RETURN(PTR_ERR(tgt));
1047 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1050 case LL_IOC_HSM_REQUEST: {
1051 struct hsm_user_request *hur = karg;
1052 unsigned int reqcount = hur->hur_request.hr_itemcount;
1057 /* if the request is about a single fid
1058 * or if there is a single MDS, no need to split
1060 if (reqcount == 1 || count == 1) {
1061 tgt = lmv_find_target(lmv,
1062 &hur->hur_user_item[0].hui_fid);
1064 RETURN(PTR_ERR(tgt));
1065 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1067 /* split fid list to their respective MDS */
1068 for (i = 0; i < count; i++) {
1071 struct hsm_user_request *req;
1074 if (tgt == NULL || tgt->ltd_exp == NULL)
1077 nr = lmv_hsm_req_count(lmv, hur, tgt);
1080 if (nr == 0) /* nothing for this MDS */
1083 /* build a request with fids for this MDS */
1084 reqlen = offsetof(typeof(*hur),
1086 + hur->hur_request.hr_data_len;
1087 OBD_ALLOC_LARGE(req, reqlen);
1090 rc1 = lmv_hsm_req_build(lmv, hur, tgt, req);
1092 GOTO(hsm_req_err, rc1);
1093 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1096 if (rc1 != 0 && rc == 0)
1098 OBD_FREE_LARGE(req, reqlen);
1103 case LL_IOC_LOV_SWAP_LAYOUTS: {
1104 struct md_op_data *op_data = karg;
1105 struct lmv_tgt_desc *tgt1, *tgt2;
1107 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1109 RETURN(PTR_ERR(tgt1));
1111 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1113 RETURN(PTR_ERR(tgt2));
1115 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1118 /* only files on same MDT can have their layouts swapped */
1119 if (tgt1->ltd_idx != tgt2->ltd_idx)
1122 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1125 case LL_IOC_HSM_CT_START: {
1126 struct lustre_kernelcomm *lk = karg;
1127 if (lk->lk_flags & LK_FLG_STOP)
1128 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1130 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1134 for (i = 0; i < count; i++) {
1135 struct obd_device *mdc_obd;
1139 if (tgt == NULL || tgt->ltd_exp == NULL)
1141 /* ll_umount_begin() sets force flag but for lmv, not
1142 * mdc. Let's pass it through */
1143 mdc_obd = class_exp2obd(tgt->ltd_exp);
1144 mdc_obd->obd_force = obddev->obd_force;
1145 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1147 if (tgt->ltd_active) {
1148 CERROR("error: iocontrol MDC %s on MDT"
1149 " idx %d cmd %x: err = %d\n",
1150 tgt->ltd_uuid.uuid, i, cmd, err);
1164 * This is _inode_ placement policy function (not name).
1166 static int lmv_placement_policy(struct obd_device *obd,
1167 struct md_op_data *op_data, u32 *mds)
1169 struct lmv_obd *lmv = &obd->u.lmv;
1172 LASSERT(mds != NULL);
1174 if (lmv->desc.ld_tgt_count == 1) {
1179 if (op_data->op_default_stripe_offset != -1) {
1180 *mds = op_data->op_default_stripe_offset;
1185 * If stripe_offset is provided during setdirstripe
1186 * (setdirstripe -i xx), xx MDS will be choosen.
1188 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1189 struct lmv_user_md *lum;
1191 lum = op_data->op_data;
1193 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1194 *mds = le32_to_cpu(lum->lum_stripe_offset);
1196 /* -1 means default, which will be in the same MDT with
1198 *mds = op_data->op_mds;
1199 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1202 /* Allocate new fid on target according to operation type and
1203 * parent home mds. */
1204 *mds = op_data->op_mds;
1210 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1212 struct lmv_tgt_desc *tgt;
1216 tgt = lmv_get_target(lmv, mds, NULL);
1218 RETURN(PTR_ERR(tgt));
1221 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1222 * on server that seq in new allocated fid is not yet known.
1224 mutex_lock(&tgt->ltd_fid_mutex);
1226 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1227 GOTO(out, rc = -ENODEV);
1230 * Asking underlying tgt layer to allocate new fid.
1232 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1234 LASSERT(fid_is_sane(fid));
1240 mutex_unlock(&tgt->ltd_fid_mutex);
1244 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1245 struct lu_fid *fid, struct md_op_data *op_data)
1247 struct obd_device *obd = class_exp2obd(exp);
1248 struct lmv_obd *lmv = &obd->u.lmv;
1253 LASSERT(op_data != NULL);
1254 LASSERT(fid != NULL);
1256 rc = lmv_placement_policy(obd, op_data, &mds);
1258 CERROR("Can't get target for allocating fid, "
1263 rc = __lmv_fid_alloc(lmv, fid, mds);
1265 CERROR("Can't alloc new fid, rc %d\n", rc);
1272 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1274 struct lmv_obd *lmv = &obd->u.lmv;
1275 struct lmv_desc *desc;
1279 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1280 CERROR("LMV setup requires a descriptor\n");
1284 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1285 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1286 CERROR("Lmv descriptor size wrong: %d > %d\n",
1287 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1291 lmv->tgts_size = 32U;
1292 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1293 if (lmv->tgts == NULL)
1296 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1297 lmv->desc.ld_tgt_count = 0;
1298 lmv->desc.ld_active_tgt_count = 0;
1299 lmv->max_def_easize = 0;
1300 lmv->max_easize = 0;
1302 spin_lock_init(&lmv->lmv_lock);
1303 mutex_init(&lmv->lmv_init_mutex);
1305 #ifdef CONFIG_PROC_FS
1306 obd->obd_vars = lprocfs_lmv_obd_vars;
1307 lprocfs_obd_setup(obd);
1308 lprocfs_alloc_md_stats(obd, 0);
1309 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1310 0444, &lmv_proc_target_fops, obd);
1312 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1315 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1316 LUSTRE_CLI_FLD_HASH_DHT);
1318 CERROR("Can't init FLD, err %d\n", rc);
1328 static int lmv_cleanup(struct obd_device *obd)
1330 struct lmv_obd *lmv = &obd->u.lmv;
1333 fld_client_fini(&lmv->lmv_fld);
1334 if (lmv->tgts != NULL) {
1336 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1337 if (lmv->tgts[i] == NULL)
1339 lmv_del_target(lmv, i);
1341 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1347 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1349 struct lustre_cfg *lcfg = buf;
1350 struct obd_uuid obd_uuid;
1356 switch (lcfg->lcfg_command) {
1358 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1359 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1360 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1361 GOTO(out, rc = -EINVAL);
1363 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1365 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1366 GOTO(out, rc = -EINVAL);
1367 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1368 GOTO(out, rc = -EINVAL);
1369 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1372 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1373 GOTO(out, rc = -EINVAL);
1379 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1380 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1382 struct obd_device *obd = class_exp2obd(exp);
1383 struct lmv_obd *lmv = &obd->u.lmv;
1384 struct obd_statfs *temp;
1389 OBD_ALLOC(temp, sizeof(*temp));
1393 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1394 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1397 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1400 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1401 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1403 GOTO(out_free_temp, rc);
1408 /* If the statfs is from mount, it will needs
1409 * retrieve necessary information from MDT0.
1410 * i.e. mount does not need the merged osfs
1412 * And also clients can be mounted as long as
1413 * MDT0 is in service*/
1414 if (flags & OBD_STATFS_FOR_MDT0)
1415 GOTO(out_free_temp, rc);
1417 osfs->os_bavail += temp->os_bavail;
1418 osfs->os_blocks += temp->os_blocks;
1419 osfs->os_ffree += temp->os_ffree;
1420 osfs->os_files += temp->os_files;
1426 OBD_FREE(temp, sizeof(*temp));
1430 static int lmv_get_root(struct obd_export *exp, const char *fileset,
1433 struct obd_device *obd = exp->exp_obd;
1434 struct lmv_obd *lmv = &obd->u.lmv;
1438 rc = md_get_root(lmv->tgts[0]->ltd_exp, fileset, fid);
1442 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1443 u64 valid, const char *name,
1444 const char *input, int input_size, int output_size,
1445 int flags, struct ptlrpc_request **request)
1447 struct obd_device *obd = exp->exp_obd;
1448 struct lmv_obd *lmv = &obd->u.lmv;
1449 struct lmv_tgt_desc *tgt;
1453 tgt = lmv_find_target(lmv, fid);
1455 RETURN(PTR_ERR(tgt));
1457 rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1458 input_size, output_size, flags, request);
1463 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1464 u64 valid, const char *name,
1465 const char *input, int input_size, int output_size,
1466 int flags, __u32 suppgid,
1467 struct ptlrpc_request **request)
1469 struct obd_device *obd = exp->exp_obd;
1470 struct lmv_obd *lmv = &obd->u.lmv;
1471 struct lmv_tgt_desc *tgt;
1475 tgt = lmv_find_target(lmv, fid);
1477 RETURN(PTR_ERR(tgt));
1479 rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1480 input_size, output_size, flags, suppgid,
1486 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1487 struct ptlrpc_request **request)
1489 struct obd_device *obd = exp->exp_obd;
1490 struct lmv_obd *lmv = &obd->u.lmv;
1491 struct lmv_tgt_desc *tgt;
1495 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1497 RETURN(PTR_ERR(tgt));
1499 if (op_data->op_flags & MF_GET_MDT_IDX) {
1500 op_data->op_mds = tgt->ltd_idx;
1504 rc = md_getattr(tgt->ltd_exp, op_data, request);
1509 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1511 struct obd_device *obd = exp->exp_obd;
1512 struct lmv_obd *lmv = &obd->u.lmv;
1516 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1519 * With DNE every object can have two locks in different namespaces:
1520 * lookup lock in space of MDT storing direntry and update/open lock in
1521 * space of MDT storing inode.
1523 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1524 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1526 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1532 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1533 struct md_open_data *mod, struct ptlrpc_request **request)
1535 struct obd_device *obd = exp->exp_obd;
1536 struct lmv_obd *lmv = &obd->u.lmv;
1537 struct lmv_tgt_desc *tgt;
1541 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1543 RETURN(PTR_ERR(tgt));
1545 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1546 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1551 * Choosing the MDT by name or FID in @op_data.
1552 * For non-striped directory, it will locate MDT by fid.
1553 * For striped-directory, it will locate MDT by name. And also
1554 * it will reset op_fid1 with the FID of the choosen stripe.
1556 struct lmv_tgt_desc *
1557 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1558 const char *name, int namelen, struct lu_fid *fid,
1561 struct lmv_tgt_desc *tgt;
1562 const struct lmv_oinfo *oinfo;
1564 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1565 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1566 RETURN(ERR_PTR(-EBADF));
1567 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1569 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1571 RETURN(ERR_CAST(oinfo));
1575 *fid = oinfo->lmo_fid;
1577 *mds = oinfo->lmo_mds;
1579 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1581 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1582 PFID(&oinfo->lmo_fid));
1587 * Locate mds by fid or name
1589 * For striped directory (lsm != NULL), it will locate the stripe
1590 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1591 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1592 * walk through all of stripes to locate the entry.
1594 * For normal direcotry, it will locate MDS by FID directly.
1595 * \param[in] lmv LMV device
1596 * \param[in] op_data client MD stack parameters, name, namelen
1598 * \param[in] fid object FID used to locate MDS.
1600 * retval pointer to the lmv_tgt_desc if succeed.
1601 * ERR_PTR(errno) if failed.
1603 struct lmv_tgt_desc*
1604 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1607 struct lmv_stripe_md *lsm = op_data->op_mea1;
1608 struct lmv_tgt_desc *tgt;
1610 /* During creating VOLATILE file, it should honor the mdt
1611 * index if the file under striped dir is being restored, see
1613 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1614 (int)op_data->op_mds != -1) {
1616 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1621 /* refill the right parent fid */
1622 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1623 struct lmv_oinfo *oinfo;
1625 oinfo = &lsm->lsm_md_oinfo[i];
1626 if (oinfo->lmo_mds == op_data->op_mds) {
1627 *fid = oinfo->lmo_fid;
1632 if (i == lsm->lsm_md_stripe_count)
1633 *fid = lsm->lsm_md_oinfo[0].lmo_fid;
1639 if (lsm == NULL || op_data->op_namelen == 0) {
1640 tgt = lmv_find_target(lmv, fid);
1644 op_data->op_mds = tgt->ltd_idx;
1648 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1649 op_data->op_namelen, fid,
1653 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1654 const void *data, size_t datalen, umode_t mode, uid_t uid,
1655 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1656 struct ptlrpc_request **request)
1658 struct obd_device *obd = exp->exp_obd;
1659 struct lmv_obd *lmv = &obd->u.lmv;
1660 struct lmv_tgt_desc *tgt;
1664 if (!lmv->desc.ld_active_tgt_count)
1667 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1669 RETURN(PTR_ERR(tgt));
1671 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1672 (int)op_data->op_namelen, op_data->op_name,
1673 PFID(&op_data->op_fid1), op_data->op_mds);
1675 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1678 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1679 /* Send the create request to the MDT where the object
1680 * will be located */
1681 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1683 RETURN(PTR_ERR(tgt));
1685 op_data->op_mds = tgt->ltd_idx;
1687 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1690 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1691 PFID(&op_data->op_fid2), op_data->op_mds);
1693 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1694 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1695 cap_effective, rdev, request);
1697 if (*request == NULL)
1699 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1705 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1706 const union ldlm_policy_data *policy, struct md_op_data *op_data,
1707 struct lustre_handle *lockh, __u64 extra_lock_flags)
1709 struct obd_device *obd = exp->exp_obd;
1710 struct lmv_obd *lmv = &obd->u.lmv;
1711 struct lmv_tgt_desc *tgt;
1715 CDEBUG(D_INODE, "ENQUEUE on "DFID"\n", PFID(&op_data->op_fid1));
1717 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1719 RETURN(PTR_ERR(tgt));
1721 CDEBUG(D_INODE, "ENQUEUE on "DFID" -> mds #%u\n",
1722 PFID(&op_data->op_fid1), tgt->ltd_idx);
1724 rc = md_enqueue(tgt->ltd_exp, einfo, policy, op_data, lockh,
1731 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1732 struct ptlrpc_request **preq)
1734 struct ptlrpc_request *req = NULL;
1735 struct obd_device *obd = exp->exp_obd;
1736 struct lmv_obd *lmv = &obd->u.lmv;
1737 struct lmv_tgt_desc *tgt;
1738 struct mdt_body *body;
1742 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1744 RETURN(PTR_ERR(tgt));
1746 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1747 (int)op_data->op_namelen, op_data->op_name,
1748 PFID(&op_data->op_fid1), tgt->ltd_idx);
1750 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1754 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1755 LASSERT(body != NULL);
1757 if (body->mbo_valid & OBD_MD_MDS) {
1758 struct lu_fid rid = body->mbo_fid1;
1759 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1762 tgt = lmv_find_target(lmv, &rid);
1764 ptlrpc_req_finished(*preq);
1766 RETURN(PTR_ERR(tgt));
1769 op_data->op_fid1 = rid;
1770 op_data->op_valid |= OBD_MD_FLCROSSREF;
1771 op_data->op_namelen = 0;
1772 op_data->op_name = NULL;
1773 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1774 ptlrpc_req_finished(*preq);
1781 #define md_op_data_fid(op_data, fl) \
1782 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1783 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1784 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1785 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1788 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1789 struct md_op_data *op_data, __u32 op_tgt,
1790 enum ldlm_mode mode, int bits, int flag)
1792 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1793 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
1794 union ldlm_policy_data policy = { { 0 } };
1798 if (!fid_is_sane(fid))
1802 tgt = lmv_find_target(lmv, fid);
1804 RETURN(PTR_ERR(tgt));
1807 if (tgt->ltd_idx != op_tgt) {
1808 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1809 policy.l_inodebits.bits = bits;
1810 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1811 mode, LCF_ASYNC, NULL);
1814 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1816 op_data->op_flags |= flag;
1824 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1827 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1828 struct ptlrpc_request **request)
1830 struct obd_device *obd = exp->exp_obd;
1831 struct lmv_obd *lmv = &obd->u.lmv;
1832 struct lmv_tgt_desc *tgt;
1836 LASSERT(op_data->op_namelen != 0);
1838 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1839 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1840 op_data->op_name, PFID(&op_data->op_fid1));
1842 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1843 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1844 op_data->op_cap = cfs_curproc_cap_pack();
1845 if (op_data->op_mea2 != NULL) {
1846 struct lmv_stripe_md *lsm = op_data->op_mea2;
1847 const struct lmv_oinfo *oinfo;
1849 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
1850 op_data->op_namelen);
1852 RETURN(PTR_ERR(oinfo));
1854 op_data->op_fid2 = oinfo->lmo_fid;
1857 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
1859 RETURN(PTR_ERR(tgt));
1862 * Cancel UPDATE lock on child (fid1).
1864 op_data->op_flags |= MF_MDC_CANCEL_FID2;
1865 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
1866 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
1870 rc = md_link(tgt->ltd_exp, op_data, request);
1875 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
1876 const char *old, size_t oldlen,
1877 const char *new, size_t newlen,
1878 struct ptlrpc_request **request)
1880 struct obd_device *obd = exp->exp_obd;
1881 struct lmv_obd *lmv = &obd->u.lmv;
1882 struct lmv_tgt_desc *src_tgt;
1883 struct lmv_tgt_desc *tgt_tgt;
1884 struct obd_export *target_exp;
1885 struct mdt_body *body;
1889 LASSERT(oldlen != 0);
1891 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
1892 (int)oldlen, old, PFID(&op_data->op_fid1),
1893 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
1894 (int)newlen, new, PFID(&op_data->op_fid2),
1895 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
1897 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
1898 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
1899 op_data->op_cap = cfs_curproc_cap_pack();
1900 if (op_data->op_cli_flags & CLI_MIGRATE) {
1901 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
1902 PFID(&op_data->op_fid3));
1904 if (op_data->op_mea1 != NULL) {
1905 struct lmv_stripe_md *lsm = op_data->op_mea1;
1906 struct lmv_tgt_desc *tmp;
1908 /* Fix the parent fid for striped dir */
1909 tmp = lmv_locate_target_for_name(lmv, lsm, old,
1914 RETURN(PTR_ERR(tmp));
1917 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1921 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
1922 if (IS_ERR(src_tgt))
1923 RETURN(PTR_ERR(src_tgt));
1925 target_exp = src_tgt->ltd_exp;
1927 if (op_data->op_mea1 != NULL) {
1928 struct lmv_stripe_md *lsm = op_data->op_mea1;
1930 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
1935 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
1937 if (IS_ERR(src_tgt))
1938 RETURN(PTR_ERR(src_tgt));
1941 if (op_data->op_mea2 != NULL) {
1942 struct lmv_stripe_md *lsm = op_data->op_mea2;
1944 tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
1949 tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
1952 if (IS_ERR(tgt_tgt))
1953 RETURN(PTR_ERR(tgt_tgt));
1955 target_exp = tgt_tgt->ltd_exp;
1959 * LOOKUP lock on src child (fid3) should also be cancelled for
1960 * src_tgt in mdc_rename.
1962 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
1965 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
1968 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1969 LCK_EX, MDS_INODELOCK_UPDATE,
1970 MF_MDC_CANCEL_FID2);
1975 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
1977 if (fid_is_sane(&op_data->op_fid3)) {
1978 struct lmv_tgt_desc *tgt;
1980 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1982 RETURN(PTR_ERR(tgt));
1984 /* Cancel LOOKUP lock on its parent */
1985 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
1986 LCK_EX, MDS_INODELOCK_LOOKUP,
1987 MF_MDC_CANCEL_FID3);
1991 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
1992 LCK_EX, MDS_INODELOCK_FULL,
1993 MF_MDC_CANCEL_FID3);
2000 * Cancel all the locks on tgt child (fid4).
2002 if (fid_is_sane(&op_data->op_fid4)) {
2003 struct lmv_tgt_desc *tgt;
2005 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2006 LCK_EX, MDS_INODELOCK_FULL,
2007 MF_MDC_CANCEL_FID4);
2011 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2013 RETURN(PTR_ERR(tgt));
2015 /* Since the target child might be destroyed, and it might
2016 * become orphan, and we can only check orphan on the local
2017 * MDT right now, so we send rename request to the MDT where
2018 * target child is located. If target child does not exist,
2019 * then it will send the request to the target parent */
2020 target_exp = tgt->ltd_exp;
2023 rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2026 if (rc != 0 && rc != -EXDEV)
2029 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2033 /* Not cross-ref case, just get out of here. */
2034 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2037 CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2038 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2040 op_data->op_fid4 = body->mbo_fid1;
2041 ptlrpc_req_finished(*request);
2046 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2047 void *ea, size_t ealen, struct ptlrpc_request **request)
2049 struct obd_device *obd = exp->exp_obd;
2050 struct lmv_obd *lmv = &obd->u.lmv;
2051 struct lmv_tgt_desc *tgt;
2055 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2056 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2058 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2059 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2061 RETURN(PTR_ERR(tgt));
2063 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2068 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2069 struct ptlrpc_request **request)
2071 struct obd_device *obd = exp->exp_obd;
2072 struct lmv_obd *lmv = &obd->u.lmv;
2073 struct lmv_tgt_desc *tgt;
2077 tgt = lmv_find_target(lmv, fid);
2079 RETURN(PTR_ERR(tgt));
2081 rc = md_fsync(tgt->ltd_exp, fid, request);
2086 * Get current minimum entry from striped directory
2088 * This function will search the dir entry, whose hash value is the
2089 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2090 * only being called for striped directory.
2092 * \param[in] exp export of LMV
2093 * \param[in] op_data parameters transferred beween client MD stack
2094 * stripe_information will be included in this
2096 * \param[in] cb_op ldlm callback being used in enqueue in
2098 * \param[in] hash_offset the hash value, which is used to locate
2099 * minum(closet) dir entry
2100 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2101 * index of last entry, so to avoid hash conflict
2102 * between stripes. It will also be used to
2103 * return the stripe index of current dir entry.
2104 * \param[in|out] entp the minum entry and it also is being used
2105 * to input the last dir entry to resolve the
2108 * \param[out] ppage the page which holds the minum entry
2110 * \retval = 0 get the entry successfully
2111 * negative errno (< 0) does not get the entry
2113 static int lmv_get_min_striped_entry(struct obd_export *exp,
2114 struct md_op_data *op_data,
2115 struct md_callback *cb_op,
2116 __u64 hash_offset, int *stripe_offset,
2117 struct lu_dirent **entp,
2118 struct page **ppage)
2120 struct obd_device *obd = exp->exp_obd;
2121 struct lmv_obd *lmv = &obd->u.lmv;
2122 struct lmv_stripe_md *lsm = op_data->op_mea1;
2123 struct lmv_tgt_desc *tgt;
2125 struct lu_dirent *min_ent = NULL;
2126 struct page *min_page = NULL;
2132 stripe_count = lsm->lsm_md_stripe_count;
2133 for (i = 0; i < stripe_count; i++) {
2134 struct lu_dirent *ent = NULL;
2135 struct page *page = NULL;
2136 struct lu_dirpage *dp;
2137 __u64 stripe_hash = hash_offset;
2139 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2141 GOTO(out, rc = PTR_ERR(tgt));
2143 /* op_data will be shared by each stripe, so we need
2144 * reset these value for each stripe */
2145 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2146 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2147 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2149 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2154 dp = page_address(page);
2155 for (ent = lu_dirent_start(dp); ent != NULL;
2156 ent = lu_dirent_next(ent)) {
2157 /* Skip dummy entry */
2158 if (le16_to_cpu(ent->lde_namelen) == 0)
2161 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2164 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2165 (*entp == ent || i < *stripe_offset))
2168 /* skip . and .. for other stripes */
2170 (strncmp(ent->lde_name, ".",
2171 le16_to_cpu(ent->lde_namelen)) == 0 ||
2172 strncmp(ent->lde_name, "..",
2173 le16_to_cpu(ent->lde_namelen)) == 0))
2179 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2185 /* reach the end of current stripe, go to next stripe */
2186 if (stripe_hash == MDS_DIR_END_OFF)
2192 if (min_ent != NULL) {
2193 if (le64_to_cpu(min_ent->lde_hash) >
2194 le64_to_cpu(ent->lde_hash)) {
2213 if (*ppage != NULL) {
2217 *stripe_offset = min_idx;
2224 * Build dir entry page from a striped directory
2226 * This function gets one entry by @offset from a striped directory. It will
2227 * read entries from all of stripes, and choose one closest to the required
2228 * offset(&offset). A few notes
2229 * 1. skip . and .. for non-zero stripes, because there can only have one .
2230 * and .. in a directory.
2231 * 2. op_data will be shared by all of stripes, instead of allocating new
2232 * one, so need to restore before reusing.
2233 * 3. release the entry page if that is not being chosen.
2235 * \param[in] exp obd export refer to LMV
2236 * \param[in] op_data hold those MD parameters of read_entry
2237 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2238 * \param[out] ldp the entry being read
2239 * \param[out] ppage the page holding the entry. Note: because the entry
2240 * will be accessed in upper layer, so we need hold the
2241 * page until the usages of entry is finished, see
2242 * ll_dir_entry_next.
2244 * retval =0 if get entry successfully
2245 * <0 cannot get entry
2247 static int lmv_read_striped_page(struct obd_export *exp,
2248 struct md_op_data *op_data,
2249 struct md_callback *cb_op,
2250 __u64 offset, struct page **ppage)
2252 struct lu_fid master_fid = op_data->op_fid1;
2253 struct inode *master_inode = op_data->op_data;
2254 __u64 hash_offset = offset;
2255 struct lu_dirpage *dp;
2256 struct page *min_ent_page = NULL;
2257 struct page *ent_page = NULL;
2258 struct lu_dirent *ent;
2261 struct lu_dirent *min_ent = NULL;
2262 struct lu_dirent *last_ent;
2267 /* Allocate a page and read entries from all of stripes and fill
2268 * the page by hash order */
2269 ent_page = alloc_page(GFP_KERNEL);
2270 if (ent_page == NULL)
2273 /* Initialize the entry page */
2274 dp = kmap(ent_page);
2275 memset(dp, 0, sizeof(*dp));
2276 dp->ldp_hash_start = cpu_to_le64(offset);
2277 dp->ldp_flags |= LDF_COLLIDE;
2280 left_bytes = PAGE_SIZE - sizeof(*dp);
2286 /* Find the minum entry from all sub-stripes */
2287 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2293 /* If it can not get minum entry, it means it already reaches
2294 * the end of this directory */
2295 if (min_ent == NULL) {
2296 last_ent->lde_reclen = 0;
2297 hash_offset = MDS_DIR_END_OFF;
2301 ent_size = le16_to_cpu(min_ent->lde_reclen);
2303 /* the last entry lde_reclen is 0, but it might not
2304 * the end of this entry of this temporay entry */
2306 ent_size = lu_dirent_calc_size(
2307 le16_to_cpu(min_ent->lde_namelen),
2308 le32_to_cpu(min_ent->lde_attrs));
2309 if (ent_size > left_bytes) {
2310 last_ent->lde_reclen = cpu_to_le16(0);
2311 hash_offset = le64_to_cpu(min_ent->lde_hash);
2315 memcpy(ent, min_ent, ent_size);
2317 /* Replace . with master FID and Replace .. with the parent FID
2318 * of master object */
2319 if (strncmp(ent->lde_name, ".",
2320 le16_to_cpu(ent->lde_namelen)) == 0 &&
2321 le16_to_cpu(ent->lde_namelen) == 1)
2322 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2323 else if (strncmp(ent->lde_name, "..",
2324 le16_to_cpu(ent->lde_namelen)) == 0 &&
2325 le16_to_cpu(ent->lde_namelen) == 2)
2326 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2328 left_bytes -= ent_size;
2329 ent->lde_reclen = cpu_to_le16(ent_size);
2331 ent = (void *)ent + ent_size;
2332 hash_offset = le64_to_cpu(min_ent->lde_hash);
2333 if (hash_offset == MDS_DIR_END_OFF) {
2334 last_ent->lde_reclen = 0;
2339 if (min_ent_page != NULL) {
2340 kunmap(min_ent_page);
2341 put_page(min_ent_page);
2344 if (unlikely(rc != 0)) {
2345 __free_page(ent_page);
2349 dp->ldp_flags |= LDF_EMPTY;
2350 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2351 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2354 /* We do not want to allocate md_op_data during each
2355 * dir entry reading, so op_data will be shared by every stripe,
2356 * then we need to restore it back to original value before
2357 * return to the upper layer */
2358 op_data->op_fid1 = master_fid;
2359 op_data->op_fid2 = master_fid;
2360 op_data->op_data = master_inode;
2367 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2368 struct md_callback *cb_op, __u64 offset,
2369 struct page **ppage)
2371 struct obd_device *obd = exp->exp_obd;
2372 struct lmv_obd *lmv = &obd->u.lmv;
2373 struct lmv_stripe_md *lsm = op_data->op_mea1;
2374 struct lmv_tgt_desc *tgt;
2378 if (unlikely(lsm != NULL)) {
2379 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2383 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2385 RETURN(PTR_ERR(tgt));
2387 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2393 * Unlink a file/directory
2395 * Unlink a file or directory under the parent dir. The unlink request
2396 * usually will be sent to the MDT where the child is located, but if
2397 * the client does not have the child FID then request will be sent to the
2398 * MDT where the parent is located.
2400 * If the parent is a striped directory then it also needs to locate which
2401 * stripe the name of the child is located, and replace the parent FID
2402 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2403 * it will walk through all of sub-stripes until the child is being
2406 * \param[in] exp export refer to LMV
2407 * \param[in] op_data different parameters transferred beween client
2408 * MD stacks, name, namelen, FIDs etc.
2409 * op_fid1 is the parent FID, op_fid2 is the child
2411 * \param[out] request point to the request of unlink.
2413 * retval 0 if succeed
2414 * negative errno if failed.
2416 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2417 struct ptlrpc_request **request)
2419 struct obd_device *obd = exp->exp_obd;
2420 struct lmv_obd *lmv = &obd->u.lmv;
2421 struct lmv_tgt_desc *tgt = NULL;
2422 struct lmv_tgt_desc *parent_tgt = NULL;
2423 struct mdt_body *body;
2425 int stripe_index = 0;
2426 struct lmv_stripe_md *lsm = op_data->op_mea1;
2430 /* For striped dir, we need to locate the parent as well */
2432 struct lmv_tgt_desc *tmp;
2434 LASSERT(op_data->op_name != NULL &&
2435 op_data->op_namelen != 0);
2437 tmp = lmv_locate_target_for_name(lmv, lsm,
2439 op_data->op_namelen,
2443 /* return -EBADFD means unknown hash type, might
2444 * need try all sub-stripe here */
2445 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2446 RETURN(PTR_ERR(tmp));
2448 /* Note: both migrating dir and unknown hash dir need to
2449 * try all of sub-stripes, so we need start search the
2450 * name from stripe 0, but migrating dir is already handled
2451 * inside lmv_locate_target_for_name(), so we only check
2452 * unknown hash type directory here */
2453 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2454 struct lmv_oinfo *oinfo;
2456 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2458 op_data->op_fid1 = oinfo->lmo_fid;
2459 op_data->op_mds = oinfo->lmo_mds;
2464 /* Send unlink requests to the MDT where the child is located */
2465 if (likely(!fid_is_zero(&op_data->op_fid2)))
2466 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2467 else if (lsm != NULL)
2468 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2470 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2473 RETURN(PTR_ERR(tgt));
2475 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2476 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2477 op_data->op_cap = cfs_curproc_cap_pack();
2480 * If child's fid is given, cancel unused locks for it if it is from
2481 * another export than parent.
2483 * LOOKUP lock for child (fid3) should also be cancelled on parent
2484 * tgt_tgt in mdc_unlink().
2486 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2489 * Cancel FULL locks on child (fid3).
2491 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2492 if (IS_ERR(parent_tgt))
2493 RETURN(PTR_ERR(parent_tgt));
2495 if (parent_tgt != tgt) {
2496 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2497 LCK_EX, MDS_INODELOCK_LOOKUP,
2498 MF_MDC_CANCEL_FID3);
2501 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2502 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2506 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2507 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2509 rc = md_unlink(tgt->ltd_exp, op_data, request);
2510 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2513 /* Try next stripe if it is needed. */
2514 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2515 struct lmv_oinfo *oinfo;
2518 if (stripe_index >= lsm->lsm_md_stripe_count)
2521 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2523 op_data->op_fid1 = oinfo->lmo_fid;
2524 op_data->op_mds = oinfo->lmo_mds;
2526 ptlrpc_req_finished(*request);
2529 goto try_next_stripe;
2532 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2536 /* Not cross-ref case, just get out of here. */
2537 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2540 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2541 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2543 /* This is a remote object, try remote MDT, Note: it may
2544 * try more than 1 time here, Considering following case
2545 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2546 * 1. Initially A does not know where remote1 is, it send
2547 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2548 * resend unlink RPC to MDT1 (retry 1st time).
2550 * 2. During the unlink RPC in flight,
2551 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2552 * and create new remote1, but on MDT0
2554 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2555 * /mnt/lustre, then lookup get fid of remote1, and find
2556 * it is remote dir again, and replay -EREMOTE again.
2558 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2560 * In theory, it might try unlimited time here, but it should
2561 * be very rare case. */
2562 op_data->op_fid2 = body->mbo_fid1;
2563 ptlrpc_req_finished(*request);
2569 static int lmv_precleanup(struct obd_device *obd)
2572 fld_client_proc_fini(&obd->u.lmv.lmv_fld);
2573 lprocfs_obd_cleanup(obd);
2574 lprocfs_free_md_stats(obd);
2579 * Get by key a value associated with a LMV device.
2581 * Dispatch request to lower-layer devices as needed.
2583 * \param[in] env execution environment for this thread
2584 * \param[in] exp export for the LMV device
2585 * \param[in] keylen length of key identifier
2586 * \param[in] key identifier of key to get value for
2587 * \param[in] vallen size of \a val
2588 * \param[out] val pointer to storage location for value
2589 * \param[in] lsm optional striping metadata of object
2591 * \retval 0 on success
2592 * \retval negative negated errno on failure
2594 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2595 __u32 keylen, void *key, __u32 *vallen, void *val)
2597 struct obd_device *obd;
2598 struct lmv_obd *lmv;
2602 obd = class_exp2obd(exp);
2604 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2605 exp->exp_handle.h_cookie);
2610 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2613 LASSERT(*vallen == sizeof(__u32));
2614 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2615 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2617 * All tgts should be connected when this gets called.
2619 if (tgt == NULL || tgt->ltd_exp == NULL)
2622 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2627 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2628 KEY_IS(KEY_DEFAULT_EASIZE) ||
2629 KEY_IS(KEY_CONN_DATA)) {
2631 * Forwarding this request to first MDS, it should know LOV
2634 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2636 if (!rc && KEY_IS(KEY_CONN_DATA))
2637 exp->exp_connect_data = *(struct obd_connect_data *)val;
2639 } else if (KEY_IS(KEY_TGT_COUNT)) {
2640 *((int *)val) = lmv->desc.ld_tgt_count;
2644 CDEBUG(D_IOCTL, "Invalid key\n");
2649 * Asynchronously set by key a value associated with a LMV device.
2651 * Dispatch request to lower-layer devices as needed.
2653 * \param[in] env execution environment for this thread
2654 * \param[in] exp export for the LMV device
2655 * \param[in] keylen length of key identifier
2656 * \param[in] key identifier of key to store value for
2657 * \param[in] vallen size of value to store
2658 * \param[in] val pointer to data to be stored
2659 * \param[in] set optional list of related ptlrpc requests
2661 * \retval 0 on success
2662 * \retval negative negated errno on failure
2664 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2665 __u32 keylen, void *key, __u32 vallen, void *val,
2666 struct ptlrpc_request_set *set)
2668 struct lmv_tgt_desc *tgt = NULL;
2669 struct obd_device *obd;
2670 struct lmv_obd *lmv;
2674 obd = class_exp2obd(exp);
2676 CDEBUG(D_IOCTL, "Invalid client cookie %#llx\n",
2677 exp->exp_handle.h_cookie);
2682 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2683 KEY_IS(KEY_DEFAULT_EASIZE)) {
2686 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2689 if (tgt == NULL || tgt->ltd_exp == NULL)
2692 err = obd_set_info_async(env, tgt->ltd_exp,
2693 keylen, key, vallen, val, set);
2704 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2705 const struct lmv_mds_md_v1 *lmm1)
2707 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2714 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2715 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2716 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2717 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2718 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2720 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2721 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2722 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2723 sizeof(lsm->lsm_md_pool_name));
2725 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2728 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2729 "layout_version %d\n", lsm->lsm_md_stripe_count,
2730 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2731 lsm->lsm_md_layout_version);
2733 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2734 for (i = 0; i < stripe_count; i++) {
2735 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2736 &lmm1->lmv_stripe_fids[i]);
2737 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2738 &lsm->lsm_md_oinfo[i].lmo_mds);
2741 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2742 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2748 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2749 const union lmv_mds_md *lmm, size_t lmm_size)
2751 struct lmv_stripe_md *lsm;
2754 bool allocated = false;
2757 LASSERT(lsmp != NULL);
2761 if (lsm != NULL && lmm == NULL) {
2763 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2764 /* For migrating inode, the master stripe and master
2765 * object will be the same, so do not need iput, see
2766 * ll_update_lsm_md */
2767 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2768 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2769 iput(lsm->lsm_md_oinfo[i].lmo_root);
2771 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2772 OBD_FREE(lsm, lsm_size);
2777 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2781 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2782 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2783 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2784 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2789 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2790 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2793 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2794 * stripecount should be 0 then.
2796 lsm_size = lmv_stripe_md_size(0);
2798 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2800 OBD_ALLOC(lsm, lsm_size);
2807 switch (le32_to_cpu(lmm->lmv_magic)) {
2809 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
2812 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
2813 le32_to_cpu(lmm->lmv_magic));
2818 if (rc != 0 && allocated) {
2819 OBD_FREE(lsm, lsm_size);
2826 void lmv_free_memmd(struct lmv_stripe_md *lsm)
2828 lmv_unpackmd(NULL, &lsm, NULL, 0);
2830 EXPORT_SYMBOL(lmv_free_memmd);
2832 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
2833 union ldlm_policy_data *policy,
2834 enum ldlm_mode mode, enum ldlm_cancel_flags flags,
2837 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2842 LASSERT(fid != NULL);
2844 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2845 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2848 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
2851 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
2859 static int lmv_set_lock_data(struct obd_export *exp,
2860 const struct lustre_handle *lockh,
2861 void *data, __u64 *bits)
2863 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2864 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2868 if (tgt == NULL || tgt->ltd_exp == NULL)
2870 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
2874 enum ldlm_mode lmv_lock_match(struct obd_export *exp, __u64 flags,
2875 const struct lu_fid *fid, enum ldlm_type type,
2876 union ldlm_policy_data *policy,
2877 enum ldlm_mode mode, struct lustre_handle *lockh)
2879 struct obd_device *obd = exp->exp_obd;
2880 struct lmv_obd *lmv = &obd->u.lmv;
2886 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
2889 * With DNE every object can have two locks in different namespaces:
2890 * lookup lock in space of MDT storing direntry and update/open lock in
2891 * space of MDT storing inode. Try the MDT that the FID maps to first,
2892 * since this can be easily found, and only try others if that fails.
2894 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
2895 i < lmv->desc.ld_tgt_count;
2896 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
2898 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
2899 obd->obd_name, PFID(fid), tgt);
2903 if (lmv->tgts[tgt] == NULL ||
2904 lmv->tgts[tgt]->ltd_exp == NULL ||
2905 lmv->tgts[tgt]->ltd_active == 0)
2908 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
2909 type, policy, mode, lockh);
2917 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
2918 struct obd_export *dt_exp, struct obd_export *md_exp,
2919 struct lustre_md *md)
2921 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2922 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2924 if (tgt == NULL || tgt->ltd_exp == NULL)
2927 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
2930 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
2932 struct obd_device *obd = exp->exp_obd;
2933 struct lmv_obd *lmv = &obd->u.lmv;
2934 struct lmv_tgt_desc *tgt = lmv->tgts[0];
2937 if (md->lmv != NULL) {
2938 lmv_free_memmd(md->lmv);
2941 if (tgt == NULL || tgt->ltd_exp == NULL)
2943 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
2946 int lmv_set_open_replay_data(struct obd_export *exp,
2947 struct obd_client_handle *och,
2948 struct lookup_intent *it)
2950 struct obd_device *obd = exp->exp_obd;
2951 struct lmv_obd *lmv = &obd->u.lmv;
2952 struct lmv_tgt_desc *tgt;
2955 tgt = lmv_find_target(lmv, &och->och_fid);
2957 RETURN(PTR_ERR(tgt));
2959 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
2962 int lmv_clear_open_replay_data(struct obd_export *exp,
2963 struct obd_client_handle *och)
2965 struct obd_device *obd = exp->exp_obd;
2966 struct lmv_obd *lmv = &obd->u.lmv;
2967 struct lmv_tgt_desc *tgt;
2970 tgt = lmv_find_target(lmv, &och->och_fid);
2972 RETURN(PTR_ERR(tgt));
2974 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
2977 int lmv_intent_getattr_async(struct obd_export *exp,
2978 struct md_enqueue_info *minfo)
2980 struct md_op_data *op_data = &minfo->mi_data;
2981 struct obd_device *obd = exp->exp_obd;
2982 struct lmv_obd *lmv = &obd->u.lmv;
2983 struct lmv_tgt_desc *ptgt = NULL;
2984 struct lmv_tgt_desc *ctgt = NULL;
2988 if (!fid_is_sane(&op_data->op_fid2))
2991 ptgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2993 RETURN(PTR_ERR(ptgt));
2995 ctgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2997 RETURN(PTR_ERR(ctgt));
3000 * if child is on remote MDT, we need 2 async RPCs to fetch both LOOKUP
3001 * lock on parent, and UPDATE lock on child MDT, which makes all
3002 * complicated. Considering remote dir is rare case, and not supporting
3003 * it in statahead won't cause any issue, drop its support for now.
3008 rc = md_intent_getattr_async(ptgt->ltd_exp, minfo);
3012 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3013 struct lu_fid *fid, __u64 *bits)
3015 struct obd_device *obd = exp->exp_obd;
3016 struct lmv_obd *lmv = &obd->u.lmv;
3017 struct lmv_tgt_desc *tgt;
3021 tgt = lmv_find_target(lmv, fid);
3023 RETURN(PTR_ERR(tgt));
3025 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3029 int lmv_get_fid_from_lsm(struct obd_export *exp,
3030 const struct lmv_stripe_md *lsm,
3031 const char *name, int namelen, struct lu_fid *fid)
3033 const struct lmv_oinfo *oinfo;
3035 LASSERT(lsm != NULL);
3036 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3038 return PTR_ERR(oinfo);
3040 *fid = oinfo->lmo_fid;
3046 * For lmv, only need to send request to master MDT, and the master MDT will
3047 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3048 * we directly fetch data from the slave MDTs.
3050 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3051 struct obd_quotactl *oqctl)
3053 struct obd_device *obd = class_exp2obd(exp);
3054 struct lmv_obd *lmv = &obd->u.lmv;
3055 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3058 __u64 curspace, curinodes;
3062 tgt->ltd_exp == NULL ||
3064 lmv->desc.ld_tgt_count == 0) {
3065 CERROR("master lmv inactive\n");
3069 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3070 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3074 curspace = curinodes = 0;
3075 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3079 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3082 err = obd_quotactl(tgt->ltd_exp, oqctl);
3084 CERROR("getquota on mdt %d failed. %d\n", i, err);
3088 curspace += oqctl->qc_dqblk.dqb_curspace;
3089 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3092 oqctl->qc_dqblk.dqb_curspace = curspace;
3093 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3098 static int lmv_merge_attr(struct obd_export *exp,
3099 const struct lmv_stripe_md *lsm,
3100 struct cl_attr *attr,
3101 ldlm_blocking_callback cb_blocking)
3106 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3110 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3111 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3113 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3114 " atime %lu ctime %lu, mtime %lu.\n",
3115 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3116 i_size_read(inode), (unsigned long long)inode->i_blocks,
3117 inode->i_nlink, LTIME_S(inode->i_atime),
3118 LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3120 /* for slave stripe, it needs to subtract nlink for . and .. */
3122 attr->cat_nlink += inode->i_nlink - 2;
3124 attr->cat_nlink = inode->i_nlink;
3126 attr->cat_size += i_size_read(inode);
3127 attr->cat_blocks += inode->i_blocks;
3129 if (attr->cat_atime < LTIME_S(inode->i_atime))
3130 attr->cat_atime = LTIME_S(inode->i_atime);
3132 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3133 attr->cat_ctime = LTIME_S(inode->i_ctime);
3135 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3136 attr->cat_mtime = LTIME_S(inode->i_mtime);
3141 struct obd_ops lmv_obd_ops = {
3142 .o_owner = THIS_MODULE,
3143 .o_setup = lmv_setup,
3144 .o_cleanup = lmv_cleanup,
3145 .o_precleanup = lmv_precleanup,
3146 .o_process_config = lmv_process_config,
3147 .o_connect = lmv_connect,
3148 .o_disconnect = lmv_disconnect,
3149 .o_statfs = lmv_statfs,
3150 .o_get_info = lmv_get_info,
3151 .o_set_info_async = lmv_set_info_async,
3152 .o_notify = lmv_notify,
3153 .o_get_uuid = lmv_get_uuid,
3154 .o_iocontrol = lmv_iocontrol,
3155 .o_quotactl = lmv_quotactl
3158 struct md_ops lmv_md_ops = {
3159 .m_get_root = lmv_get_root,
3160 .m_null_inode = lmv_null_inode,
3161 .m_close = lmv_close,
3162 .m_create = lmv_create,
3163 .m_enqueue = lmv_enqueue,
3164 .m_getattr = lmv_getattr,
3165 .m_getxattr = lmv_getxattr,
3166 .m_getattr_name = lmv_getattr_name,
3167 .m_intent_lock = lmv_intent_lock,
3169 .m_rename = lmv_rename,
3170 .m_setattr = lmv_setattr,
3171 .m_setxattr = lmv_setxattr,
3172 .m_fsync = lmv_fsync,
3173 .m_read_page = lmv_read_page,
3174 .m_unlink = lmv_unlink,
3175 .m_init_ea_size = lmv_init_ea_size,
3176 .m_cancel_unused = lmv_cancel_unused,
3177 .m_set_lock_data = lmv_set_lock_data,
3178 .m_lock_match = lmv_lock_match,
3179 .m_get_lustre_md = lmv_get_lustre_md,
3180 .m_free_lustre_md = lmv_free_lustre_md,
3181 .m_merge_attr = lmv_merge_attr,
3182 .m_set_open_replay_data = lmv_set_open_replay_data,
3183 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3184 .m_intent_getattr_async = lmv_intent_getattr_async,
3185 .m_revalidate_lock = lmv_revalidate_lock,
3186 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3187 .m_unpackmd = lmv_unpackmd,
3190 static int __init lmv_init(void)
3192 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3193 LUSTRE_LMV_NAME, NULL);
3196 static void __exit lmv_exit(void)
3198 class_unregister_type(LUSTRE_LMV_NAME);
3201 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3202 MODULE_DESCRIPTION("Lustre Logical Metadata Volume");
3203 MODULE_VERSION(LUSTRE_VERSION_STRING);
3204 MODULE_LICENSE("GPL");
3206 module_init(lmv_init);
3207 module_exit(lmv_exit);