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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #define DEBUG_SUBSYSTEM S_LMV
38 #include <linux/slab.h>
39 #include <linux/module.h>
40 #include <linux/init.h>
41 #include <linux/user_namespace.h>
42 #ifdef HAVE_UIDGID_HEADER
43 # include <linux/uidgid.h>
45 #include <linux/slab.h>
46 #include <linux/pagemap.h>
48 #include <linux/math64.h>
49 #include <linux/seq_file.h>
50 #include <linux/namei.h>
52 #include <lustre/lustre_idl.h>
53 #include <obd_support.h>
54 #include <lustre_lib.h>
55 #include <lustre_net.h>
56 #include <obd_class.h>
57 #include <lustre_lmv.h>
58 #include <lprocfs_status.h>
59 #include <cl_object.h>
60 #include <lustre_fid.h>
61 #include <lustre_ioctl.h>
62 #include <lustre_kernelcomm.h>
63 #include "lmv_internal.h"
65 static void lmv_activate_target(struct lmv_obd *lmv,
66 struct lmv_tgt_desc *tgt,
69 if (tgt->ltd_active == activate)
72 tgt->ltd_active = activate;
73 lmv->desc.ld_active_tgt_count += (activate ? 1 : -1);
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 "LPX64"\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;
185 else if (ev == OBD_NOTIFY_DISCON) {
187 * For disconnect event, flush fld cache for failout MDS case.
189 fld_client_flush(&lmv->lmv_fld);
193 * Pass the notification up the chain.
195 if (obd->obd_observer)
196 rc = obd_notify(obd->obd_observer, watched, ev, data);
202 * This is fake connect function. Its purpose is to initialize lmv and say
203 * caller that everything is okay. Real connection will be performed later.
205 static int lmv_connect(const struct lu_env *env,
206 struct obd_export **exp, struct obd_device *obd,
207 struct obd_uuid *cluuid, struct obd_connect_data *data,
210 struct lmv_obd *lmv = &obd->u.lmv;
211 struct lustre_handle conn = { 0 };
216 * We don't want to actually do the underlying connections more than
217 * once, so keep track.
220 if (lmv->refcount > 1) {
225 rc = class_connect(&conn, obd, cluuid);
227 CERROR("class_connection() returned %d\n", rc);
231 *exp = class_conn2export(&conn);
232 class_export_get(*exp);
236 lmv->cluuid = *cluuid;
239 lmv->conn_data = *data;
241 if (lmv->targets_proc_entry == NULL) {
242 lmv->targets_proc_entry = lprocfs_register("target_obds",
245 if (IS_ERR(lmv->targets_proc_entry)) {
246 CERROR("%s: cannot register "
247 "/proc/fs/lustre/%s/%s/target_obds\n",
248 obd->obd_name, obd->obd_type->typ_name,
250 lmv->targets_proc_entry = NULL;
255 * All real clients should perform actual connection right away, because
256 * it is possible, that LMV will not have opportunity to connect targets
257 * and MDC stuff will be called directly, for instance while reading
258 * ../mdc/../kbytesfree procfs file, etc.
260 if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL))
261 rc = lmv_check_connect(obd);
263 if (rc && lmv->targets_proc_entry != NULL)
264 lprocfs_remove(&lmv->targets_proc_entry);
268 static int lmv_init_ea_size(struct obd_export *exp, __u32 easize,
271 struct obd_device *obd = exp->exp_obd;
272 struct lmv_obd *lmv = &obd->u.lmv;
278 if (lmv->max_easize < easize) {
279 lmv->max_easize = easize;
282 if (lmv->max_def_easize < def_easize) {
283 lmv->max_def_easize = def_easize;
290 if (lmv->connected == 0)
293 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
294 struct lmv_tgt_desc *tgt = lmv->tgts[i];
296 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) {
297 CWARN("%s: NULL export for %d\n", obd->obd_name, i);
301 rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize);
303 CERROR("%s: obd_init_ea_size() failed on MDT target %d:"
304 " rc = %d\n", obd->obd_name, i, rc);
311 #define MAX_STRING_SIZE 128
313 int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
315 struct lmv_obd *lmv = &obd->u.lmv;
316 struct obd_uuid *cluuid = &lmv->cluuid;
317 struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" };
318 struct obd_device *mdc_obd;
319 struct obd_export *mdc_exp;
320 struct lu_fld_target target;
324 mdc_obd = class_find_client_obd(&tgt->ltd_uuid, LUSTRE_MDC_NAME,
327 CERROR("target %s not attached\n", tgt->ltd_uuid.uuid);
331 CDEBUG(D_CONFIG, "connect to %s(%s) - %s, %s FOR %s\n",
332 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
333 tgt->ltd_uuid.uuid, obd->obd_uuid.uuid,
336 if (!mdc_obd->obd_set_up) {
337 CERROR("target %s is not set up\n", tgt->ltd_uuid.uuid);
341 rc = obd_connect(NULL, &mdc_exp, mdc_obd, &lmv_mdc_uuid,
342 &lmv->conn_data, NULL);
344 CERROR("target %s connect error %d\n", tgt->ltd_uuid.uuid, rc);
349 * Init fid sequence client for this mdc and add new fld target.
351 rc = obd_fid_init(mdc_obd, mdc_exp, LUSTRE_SEQ_METADATA);
355 target.ft_srv = NULL;
356 target.ft_exp = mdc_exp;
357 target.ft_idx = tgt->ltd_idx;
359 fld_client_add_target(&lmv->lmv_fld, &target);
361 rc = obd_register_observer(mdc_obd, obd);
363 obd_disconnect(mdc_exp);
364 CERROR("target %s register_observer error %d\n",
365 tgt->ltd_uuid.uuid, rc);
369 if (obd->obd_observer) {
371 * Tell the observer about the new target.
373 rc = obd_notify(obd->obd_observer, mdc_exp->exp_obd,
375 (void *)(tgt - lmv->tgts[0]));
377 obd_disconnect(mdc_exp);
383 tgt->ltd_exp = mdc_exp;
384 lmv->desc.ld_active_tgt_count++;
386 md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize);
388 CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n",
389 mdc_obd->obd_name, mdc_obd->obd_uuid.uuid,
390 atomic_read(&obd->obd_refcount));
392 if (lmv->targets_proc_entry != NULL) {
393 struct proc_dir_entry *mdc_symlink;
395 LASSERT(mdc_obd->obd_type != NULL);
396 LASSERT(mdc_obd->obd_type->typ_name != NULL);
397 mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name,
398 lmv->targets_proc_entry,
400 mdc_obd->obd_type->typ_name,
402 if (mdc_symlink == NULL) {
403 CERROR("cannot register LMV target "
404 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
405 obd->obd_type->typ_name, obd->obd_name,
412 static void lmv_del_target(struct lmv_obd *lmv, int index)
414 if (lmv->tgts[index] == NULL)
417 OBD_FREE_PTR(lmv->tgts[index]);
418 lmv->tgts[index] = NULL;
422 static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
423 __u32 index, int gen)
425 struct obd_device *mdc_obd;
426 struct lmv_obd *lmv = &obd->u.lmv;
427 struct lmv_tgt_desc *tgt;
428 int orig_tgt_count = 0;
432 CDEBUG(D_CONFIG, "Target uuid: %s. index %d\n", uuidp->uuid, index);
433 mdc_obd = class_find_client_obd(uuidp, LUSTRE_MDC_NAME,
436 CERROR("%s: Target %s not attached: rc = %d\n",
437 obd->obd_name, uuidp->uuid, -EINVAL);
441 mutex_lock(&lmv->lmv_init_mutex);
442 if ((index < lmv->tgts_size) && (lmv->tgts[index] != NULL)) {
443 tgt = lmv->tgts[index];
444 CERROR("%s: UUID %s already assigned at LOV target index %d:"
445 " rc = %d\n", obd->obd_name,
446 obd_uuid2str(&tgt->ltd_uuid), index, -EEXIST);
447 mutex_unlock(&lmv->lmv_init_mutex);
451 if (index >= lmv->tgts_size) {
452 /* We need to reallocate the lmv target array. */
453 struct lmv_tgt_desc **newtgts, **old = NULL;
457 while (newsize < index + 1)
458 newsize = newsize << 1;
459 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
460 if (newtgts == NULL) {
461 mutex_unlock(&lmv->lmv_init_mutex);
465 if (lmv->tgts_size) {
466 memcpy(newtgts, lmv->tgts,
467 sizeof(*newtgts) * lmv->tgts_size);
469 oldsize = lmv->tgts_size;
473 lmv->tgts_size = newsize;
476 OBD_FREE(old, sizeof(*old) * oldsize);
478 CDEBUG(D_CONFIG, "tgts: %p size: %d\n", lmv->tgts,
484 mutex_unlock(&lmv->lmv_init_mutex);
488 mutex_init(&tgt->ltd_fid_mutex);
489 tgt->ltd_idx = index;
490 tgt->ltd_uuid = *uuidp;
492 lmv->tgts[index] = tgt;
493 if (index >= lmv->desc.ld_tgt_count) {
494 orig_tgt_count = lmv->desc.ld_tgt_count;
495 lmv->desc.ld_tgt_count = index + 1;
498 if (lmv->connected == 0) {
499 /* lmv_check_connect() will connect this target. */
500 mutex_unlock(&lmv->lmv_init_mutex);
504 /* Otherwise let's connect it ourselves */
505 mutex_unlock(&lmv->lmv_init_mutex);
506 rc = lmv_connect_mdc(obd, tgt);
508 spin_lock(&lmv->lmv_lock);
509 if (lmv->desc.ld_tgt_count == index + 1)
510 lmv->desc.ld_tgt_count = orig_tgt_count;
511 memset(tgt, 0, sizeof(*tgt));
512 spin_unlock(&lmv->lmv_lock);
514 int easize = sizeof(struct lmv_stripe_md) +
515 lmv->desc.ld_tgt_count * sizeof(struct lu_fid);
516 lmv_init_ea_size(obd->obd_self_export, easize, 0);
522 int lmv_check_connect(struct obd_device *obd)
524 struct lmv_obd *lmv = &obd->u.lmv;
525 struct lmv_tgt_desc *tgt;
534 mutex_lock(&lmv->lmv_init_mutex);
535 if (lmv->connected) {
536 mutex_unlock(&lmv->lmv_init_mutex);
540 if (lmv->desc.ld_tgt_count == 0) {
541 mutex_unlock(&lmv->lmv_init_mutex);
542 CERROR("%s: no targets configured.\n", obd->obd_name);
546 LASSERT(lmv->tgts != NULL);
548 if (lmv->tgts[0] == NULL) {
549 mutex_unlock(&lmv->lmv_init_mutex);
550 CERROR("%s: no target configured for index 0.\n",
555 CDEBUG(D_CONFIG, "Time to connect %s to %s\n",
556 lmv->cluuid.uuid, obd->obd_name);
558 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
562 rc = lmv_connect_mdc(obd, tgt);
567 class_export_put(lmv->exp);
569 easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC);
570 lmv_init_ea_size(obd->obd_self_export, easize, 0);
571 mutex_unlock(&lmv->lmv_init_mutex);
582 --lmv->desc.ld_active_tgt_count;
583 rc2 = obd_disconnect(tgt->ltd_exp);
585 CERROR("LMV target %s disconnect on "
586 "MDC idx %d: error %d\n",
587 tgt->ltd_uuid.uuid, i, rc2);
591 class_disconnect(lmv->exp);
592 mutex_unlock(&lmv->lmv_init_mutex);
596 static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt)
598 struct lmv_obd *lmv = &obd->u.lmv;
599 struct obd_device *mdc_obd;
603 LASSERT(tgt != NULL);
604 LASSERT(obd != NULL);
606 mdc_obd = class_exp2obd(tgt->ltd_exp);
609 mdc_obd->obd_force = obd->obd_force;
610 mdc_obd->obd_fail = obd->obd_fail;
611 mdc_obd->obd_no_recov = obd->obd_no_recov;
613 if (lmv->targets_proc_entry != NULL)
614 lprocfs_remove_proc_entry(mdc_obd->obd_name,
615 lmv->targets_proc_entry);
618 rc = obd_fid_fini(tgt->ltd_exp->exp_obd);
620 CERROR("Can't finanize fids factory\n");
622 CDEBUG(D_INFO, "Disconnected from %s(%s) successfully\n",
623 tgt->ltd_exp->exp_obd->obd_name,
624 tgt->ltd_exp->exp_obd->obd_uuid.uuid);
626 obd_register_observer(tgt->ltd_exp->exp_obd, NULL);
627 rc = obd_disconnect(tgt->ltd_exp);
629 if (tgt->ltd_active) {
630 CERROR("Target %s disconnect error %d\n",
631 tgt->ltd_uuid.uuid, rc);
635 lmv_activate_target(lmv, tgt, 0);
640 static int lmv_disconnect(struct obd_export *exp)
642 struct obd_device *obd = class_exp2obd(exp);
643 struct lmv_obd *lmv = &obd->u.lmv;
652 * Only disconnect the underlying layers on the final disconnect.
655 if (lmv->refcount != 0)
658 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
659 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
662 lmv_disconnect_mdc(obd, lmv->tgts[i]);
665 if (lmv->targets_proc_entry != NULL)
666 lprocfs_remove(&lmv->targets_proc_entry);
668 CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n",
669 obd->obd_type->typ_name, obd->obd_name);
673 * This is the case when no real connection is established by
674 * lmv_check_connect().
677 class_export_put(exp);
678 rc = class_disconnect(exp);
679 if (lmv->refcount == 0)
684 static int lmv_fid2path(struct obd_export *exp, int len, void *karg, void *uarg)
686 struct obd_device *obddev = class_exp2obd(exp);
687 struct lmv_obd *lmv = &obddev->u.lmv;
688 struct getinfo_fid2path *gf;
689 struct lmv_tgt_desc *tgt;
690 struct getinfo_fid2path *remote_gf = NULL;
691 int remote_gf_size = 0;
694 gf = (struct getinfo_fid2path *)karg;
695 tgt = lmv_find_target(lmv, &gf->gf_fid);
697 RETURN(PTR_ERR(tgt));
700 rc = obd_iocontrol(OBD_IOC_FID2PATH, tgt->ltd_exp, len, gf, uarg);
701 if (rc != 0 && rc != -EREMOTE)
702 GOTO(out_fid2path, rc);
704 /* If remote_gf != NULL, it means just building the
705 * path on the remote MDT, copy this path segement to gf */
706 if (remote_gf != NULL) {
707 struct getinfo_fid2path *ori_gf;
710 ori_gf = (struct getinfo_fid2path *)karg;
711 if (strlen(ori_gf->gf_path) +
712 strlen(gf->gf_path) > ori_gf->gf_pathlen)
713 GOTO(out_fid2path, rc = -EOVERFLOW);
715 ptr = ori_gf->gf_path;
717 memmove(ptr + strlen(gf->gf_path) + 1, ptr,
718 strlen(ori_gf->gf_path));
720 strncpy(ptr, gf->gf_path, strlen(gf->gf_path));
721 ptr += strlen(gf->gf_path);
725 CDEBUG(D_INFO, "%s: get path %s "DFID" rec: "LPU64" ln: %u\n",
726 tgt->ltd_exp->exp_obd->obd_name,
727 gf->gf_path, PFID(&gf->gf_fid), gf->gf_recno,
731 GOTO(out_fid2path, rc);
733 /* sigh, has to go to another MDT to do path building further */
734 if (remote_gf == NULL) {
735 remote_gf_size = sizeof(*remote_gf) + PATH_MAX;
736 OBD_ALLOC(remote_gf, remote_gf_size);
737 if (remote_gf == NULL)
738 GOTO(out_fid2path, rc = -ENOMEM);
739 remote_gf->gf_pathlen = PATH_MAX;
742 if (!fid_is_sane(&gf->gf_fid)) {
743 CERROR("%s: invalid FID "DFID": rc = %d\n",
744 tgt->ltd_exp->exp_obd->obd_name,
745 PFID(&gf->gf_fid), -EINVAL);
746 GOTO(out_fid2path, rc = -EINVAL);
749 tgt = lmv_find_target(lmv, &gf->gf_fid);
751 GOTO(out_fid2path, rc = -EINVAL);
753 remote_gf->gf_fid = gf->gf_fid;
754 remote_gf->gf_recno = -1;
755 remote_gf->gf_linkno = -1;
756 memset(remote_gf->gf_path, 0, remote_gf->gf_pathlen);
758 goto repeat_fid2path;
761 if (remote_gf != NULL)
762 OBD_FREE(remote_gf, remote_gf_size);
766 static int lmv_hsm_req_count(struct lmv_obd *lmv,
767 const struct hsm_user_request *hur,
768 const struct lmv_tgt_desc *tgt_mds)
772 struct lmv_tgt_desc *curr_tgt;
774 /* count how many requests must be sent to the given target */
775 for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
776 curr_tgt = lmv_find_target(lmv, &hur->hur_user_item[i].hui_fid);
777 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid))
783 static void lmv_hsm_req_build(struct lmv_obd *lmv,
784 struct hsm_user_request *hur_in,
785 const struct lmv_tgt_desc *tgt_mds,
786 struct hsm_user_request *hur_out)
789 struct lmv_tgt_desc *curr_tgt;
791 /* build the hsm_user_request for the given target */
792 hur_out->hur_request = hur_in->hur_request;
794 for (i = 0; i < hur_in->hur_request.hr_itemcount; i++) {
795 curr_tgt = lmv_find_target(lmv,
796 &hur_in->hur_user_item[i].hui_fid);
797 if (obd_uuid_equals(&curr_tgt->ltd_uuid, &tgt_mds->ltd_uuid)) {
798 hur_out->hur_user_item[nr_out] =
799 hur_in->hur_user_item[i];
803 hur_out->hur_request.hr_itemcount = nr_out;
804 memcpy(hur_data(hur_out), hur_data(hur_in),
805 hur_in->hur_request.hr_data_len);
808 static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len,
809 struct lustre_kernelcomm *lk, void *uarg)
813 struct kkuc_ct_data *kcd = NULL;
816 /* unregister request (call from llapi_hsm_copytool_fini) */
817 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
818 struct lmv_tgt_desc *tgt = lmv->tgts[i];
820 if (tgt == NULL || tgt->ltd_exp == NULL)
822 /* best effort: try to clean as much as possible
823 * (continue on error) */
824 obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
827 /* Whatever the result, remove copytool from kuc groups.
828 * Unreached coordinators will get EPIPE on next requests
829 * and will unregister automatically.
831 rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd);
838 static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len,
839 struct lustre_kernelcomm *lk, void *uarg)
844 bool any_set = false;
845 struct kkuc_ct_data *kcd;
848 /* All or nothing: try to register to all MDS.
849 * In case of failure, unregister from previous MDS,
850 * except if it because of inactive target. */
851 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
852 struct lmv_tgt_desc *tgt = lmv->tgts[i];
854 if (tgt == NULL || tgt->ltd_exp == NULL)
856 err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg);
858 if (tgt->ltd_active) {
859 /* permanent error */
860 CERROR("%s: iocontrol MDC %s on MDT"
861 " idx %d cmd %x: err = %d\n",
862 class_exp2obd(lmv->exp)->obd_name,
863 tgt->ltd_uuid.uuid, i, cmd, err);
865 lk->lk_flags |= LK_FLG_STOP;
866 /* unregister from previous MDS */
867 for (j = 0; j < i; j++) {
869 if (tgt == NULL || tgt->ltd_exp == NULL)
871 obd_iocontrol(cmd, tgt->ltd_exp, len,
876 /* else: transient error.
877 * kuc will register to the missing MDT
885 /* no registration done: return error */
888 /* at least one registration done, with no failure */
889 filp = fget(lk->lk_wfd);
898 kcd->kcd_magic = KKUC_CT_DATA_MAGIC;
899 kcd->kcd_uuid = lmv->cluuid;
900 kcd->kcd_archive = lk->lk_data;
902 rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd);
915 static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp,
916 int len, void *karg, void *uarg)
918 struct obd_device *obddev = class_exp2obd(exp);
919 struct lmv_obd *lmv = &obddev->u.lmv;
920 struct lmv_tgt_desc *tgt = NULL;
924 __u32 count = lmv->desc.ld_tgt_count;
931 case IOC_OBD_STATFS: {
932 struct obd_ioctl_data *data = karg;
933 struct obd_device *mdc_obd;
934 struct obd_statfs stat_buf = {0};
937 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
938 if ((index >= count))
941 tgt = lmv->tgts[index];
942 if (tgt == NULL || !tgt->ltd_active)
945 mdc_obd = class_exp2obd(tgt->ltd_exp);
950 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(mdc_obd),
951 min((int) data->ioc_plen2,
952 (int) sizeof(struct obd_uuid))))
955 rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf,
956 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
960 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
961 min((int) data->ioc_plen1,
962 (int) sizeof(stat_buf))))
966 case OBD_IOC_QUOTACTL: {
967 struct if_quotactl *qctl = karg;
968 struct obd_quotactl *oqctl;
970 if (qctl->qc_valid == QC_MDTIDX) {
971 if (count <= qctl->qc_idx)
974 tgt = lmv->tgts[qctl->qc_idx];
975 if (tgt == NULL || tgt->ltd_exp == NULL)
977 } else if (qctl->qc_valid == QC_UUID) {
978 for (i = 0; i < count; i++) {
982 if (!obd_uuid_equals(&tgt->ltd_uuid,
986 if (tgt->ltd_exp == NULL)
998 LASSERT(tgt != NULL && tgt->ltd_exp != NULL);
999 OBD_ALLOC_PTR(oqctl);
1003 QCTL_COPY(oqctl, qctl);
1004 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1006 QCTL_COPY(qctl, oqctl);
1007 qctl->qc_valid = QC_MDTIDX;
1008 qctl->obd_uuid = tgt->ltd_uuid;
1010 OBD_FREE_PTR(oqctl);
1013 case OBD_IOC_CHANGELOG_SEND:
1014 case OBD_IOC_CHANGELOG_CLEAR: {
1015 struct ioc_changelog *icc = karg;
1017 if (icc->icc_mdtindex >= count)
1020 tgt = lmv->tgts[icc->icc_mdtindex];
1021 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
1023 rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL);
1026 case LL_IOC_GET_CONNECT_FLAGS: {
1028 if (tgt == NULL || tgt->ltd_exp == NULL)
1030 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1033 case LL_IOC_FID2MDTIDX: {
1034 struct lu_fid *fid = karg;
1037 rc = lmv_fld_lookup(lmv, fid, &mdt_index);
1041 /* Note: this is from llite(see ll_dir_ioctl()), @uarg does not
1042 * point to user space memory for FID2MDTIDX. */
1043 *(__u32 *)uarg = mdt_index;
1046 case OBD_IOC_FID2PATH: {
1047 rc = lmv_fid2path(exp, len, karg, uarg);
1050 case LL_IOC_HSM_STATE_GET:
1051 case LL_IOC_HSM_STATE_SET:
1052 case LL_IOC_HSM_ACTION: {
1053 struct md_op_data *op_data = karg;
1055 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1057 RETURN(PTR_ERR(tgt));
1059 if (tgt->ltd_exp == NULL)
1062 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1065 case LL_IOC_HSM_PROGRESS: {
1066 const struct hsm_progress_kernel *hpk = karg;
1068 tgt = lmv_find_target(lmv, &hpk->hpk_fid);
1070 RETURN(PTR_ERR(tgt));
1071 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1074 case LL_IOC_HSM_REQUEST: {
1075 struct hsm_user_request *hur = karg;
1076 unsigned int reqcount = hur->hur_request.hr_itemcount;
1081 /* if the request is about a single fid
1082 * or if there is a single MDS, no need to split
1084 if (reqcount == 1 || count == 1) {
1085 tgt = lmv_find_target(lmv,
1086 &hur->hur_user_item[0].hui_fid);
1088 RETURN(PTR_ERR(tgt));
1089 rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1091 /* split fid list to their respective MDS */
1092 for (i = 0; i < count; i++) {
1093 unsigned int nr, reqlen;
1095 struct hsm_user_request *req;
1098 if (tgt == NULL || tgt->ltd_exp == NULL)
1101 nr = lmv_hsm_req_count(lmv, hur, tgt);
1102 if (nr == 0) /* nothing for this MDS */
1105 /* build a request with fids for this MDS */
1106 reqlen = offsetof(typeof(*hur),
1108 + hur->hur_request.hr_data_len;
1109 OBD_ALLOC_LARGE(req, reqlen);
1113 lmv_hsm_req_build(lmv, hur, tgt, req);
1115 rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen,
1117 if (rc1 != 0 && rc == 0)
1119 OBD_FREE_LARGE(req, reqlen);
1124 case LL_IOC_LOV_SWAP_LAYOUTS: {
1125 struct md_op_data *op_data = karg;
1126 struct lmv_tgt_desc *tgt1, *tgt2;
1128 tgt1 = lmv_find_target(lmv, &op_data->op_fid1);
1130 RETURN(PTR_ERR(tgt1));
1132 tgt2 = lmv_find_target(lmv, &op_data->op_fid2);
1134 RETURN(PTR_ERR(tgt2));
1136 if ((tgt1->ltd_exp == NULL) || (tgt2->ltd_exp == NULL))
1139 /* only files on same MDT can have their layouts swapped */
1140 if (tgt1->ltd_idx != tgt2->ltd_idx)
1143 rc = obd_iocontrol(cmd, tgt1->ltd_exp, len, karg, uarg);
1146 case LL_IOC_HSM_CT_START: {
1147 struct lustre_kernelcomm *lk = karg;
1148 if (lk->lk_flags & LK_FLG_STOP)
1149 rc = lmv_hsm_ct_unregister(lmv, cmd, len, lk, uarg);
1151 rc = lmv_hsm_ct_register(lmv, cmd, len, lk, uarg);
1155 for (i = 0; i < count; i++) {
1156 struct obd_device *mdc_obd;
1160 if (tgt == NULL || tgt->ltd_exp == NULL)
1162 /* ll_umount_begin() sets force flag but for lmv, not
1163 * mdc. Let's pass it through */
1164 mdc_obd = class_exp2obd(tgt->ltd_exp);
1165 mdc_obd->obd_force = obddev->obd_force;
1166 err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg);
1168 if (tgt->ltd_active) {
1169 CERROR("error: iocontrol MDC %s on MDT"
1170 " idx %d cmd %x: err = %d\n",
1171 tgt->ltd_uuid.uuid, i, cmd, err);
1185 static int lmv_all_chars_policy(int count, const char *name,
1196 static int lmv_nid_policy(struct lmv_obd *lmv)
1198 struct obd_import *imp;
1202 * XXX: To get nid we assume that underlying obd device is mdc.
1204 imp = class_exp2cliimp(lmv->tgts[0].ltd_exp);
1205 id = imp->imp_connection->c_self ^ (imp->imp_connection->c_self >> 32);
1206 return id % lmv->desc.ld_tgt_count;
1209 static int lmv_choose_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1210 placement_policy_t placement)
1212 switch (placement) {
1213 case PLACEMENT_CHAR_POLICY:
1214 return lmv_all_chars_policy(lmv->desc.ld_tgt_count,
1216 op_data->op_namelen);
1217 case PLACEMENT_NID_POLICY:
1218 return lmv_nid_policy(lmv);
1224 CERROR("Unsupported placement policy %x\n", placement);
1230 * This is _inode_ placement policy function (not name).
1232 static int lmv_placement_policy(struct obd_device *obd,
1233 struct md_op_data *op_data, u32 *mds)
1235 struct lmv_obd *lmv = &obd->u.lmv;
1238 LASSERT(mds != NULL);
1240 if (lmv->desc.ld_tgt_count == 1) {
1245 if (op_data->op_default_stripe_offset != -1) {
1246 *mds = op_data->op_default_stripe_offset;
1251 * If stripe_offset is provided during setdirstripe
1252 * (setdirstripe -i xx), xx MDS will be choosen.
1254 if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) {
1255 struct lmv_user_md *lum;
1257 lum = op_data->op_data;
1259 if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) {
1260 *mds = le32_to_cpu(lum->lum_stripe_offset);
1262 /* -1 means default, which will be in the same MDT with
1264 *mds = op_data->op_mds;
1265 lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds);
1268 /* Allocate new fid on target according to operation type and
1269 * parent home mds. */
1270 *mds = op_data->op_mds;
1276 int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, u32 mds)
1278 struct lmv_tgt_desc *tgt;
1282 tgt = lmv_get_target(lmv, mds, NULL);
1284 RETURN(PTR_ERR(tgt));
1287 * New seq alloc and FLD setup should be atomic. Otherwise we may find
1288 * on server that seq in new allocated fid is not yet known.
1290 mutex_lock(&tgt->ltd_fid_mutex);
1292 if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL)
1293 GOTO(out, rc = -ENODEV);
1296 * Asking underlying tgt layer to allocate new fid.
1298 rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL);
1300 LASSERT(fid_is_sane(fid));
1306 mutex_unlock(&tgt->ltd_fid_mutex);
1310 int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp,
1311 struct lu_fid *fid, struct md_op_data *op_data)
1313 struct obd_device *obd = class_exp2obd(exp);
1314 struct lmv_obd *lmv = &obd->u.lmv;
1319 LASSERT(op_data != NULL);
1320 LASSERT(fid != NULL);
1322 rc = lmv_placement_policy(obd, op_data, &mds);
1324 CERROR("Can't get target for allocating fid, "
1329 rc = __lmv_fid_alloc(lmv, fid, mds);
1331 CERROR("Can't alloc new fid, rc %d\n", rc);
1338 static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
1340 struct lmv_obd *lmv = &obd->u.lmv;
1341 struct lmv_desc *desc;
1345 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
1346 CERROR("LMV setup requires a descriptor\n");
1350 desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1);
1351 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
1352 CERROR("Lmv descriptor size wrong: %d > %d\n",
1353 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
1357 lmv->tgts_size = 32U;
1358 OBD_ALLOC(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1359 if (lmv->tgts == NULL)
1362 obd_str2uuid(&lmv->desc.ld_uuid, desc->ld_uuid.uuid);
1363 lmv->desc.ld_tgt_count = 0;
1364 lmv->desc.ld_active_tgt_count = 0;
1365 lmv->max_def_easize = 0;
1366 lmv->max_easize = 0;
1367 lmv->lmv_placement = PLACEMENT_CHAR_POLICY;
1369 spin_lock_init(&lmv->lmv_lock);
1370 mutex_init(&lmv->lmv_init_mutex);
1372 #ifdef CONFIG_PROC_FS
1373 obd->obd_vars = lprocfs_lmv_obd_vars;
1374 lprocfs_obd_setup(obd);
1375 lprocfs_alloc_md_stats(obd, 0);
1376 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
1377 0444, &lmv_proc_target_fops, obd);
1379 CWARN("%s: error adding LMV target_obd file: rc = %d\n",
1382 rc = fld_client_init(&lmv->lmv_fld, obd->obd_name,
1383 LUSTRE_CLI_FLD_HASH_DHT);
1385 CERROR("Can't init FLD, err %d\n", rc);
1395 static int lmv_cleanup(struct obd_device *obd)
1397 struct lmv_obd *lmv = &obd->u.lmv;
1400 fld_client_fini(&lmv->lmv_fld);
1401 if (lmv->tgts != NULL) {
1403 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1404 if (lmv->tgts[i] == NULL)
1406 lmv_del_target(lmv, i);
1408 OBD_FREE(lmv->tgts, sizeof(*lmv->tgts) * lmv->tgts_size);
1414 static int lmv_process_config(struct obd_device *obd, size_t len, void *buf)
1416 struct lustre_cfg *lcfg = buf;
1417 struct obd_uuid obd_uuid;
1423 switch (lcfg->lcfg_command) {
1425 /* modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID
1426 * 2:0 3:1 4:lustre-MDT0000-mdc_UUID */
1427 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
1428 GOTO(out, rc = -EINVAL);
1430 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
1432 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1)
1433 GOTO(out, rc = -EINVAL);
1434 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1)
1435 GOTO(out, rc = -EINVAL);
1436 rc = lmv_add_target(obd, &obd_uuid, index, gen);
1439 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1440 GOTO(out, rc = -EINVAL);
1446 static int lmv_statfs(const struct lu_env *env, struct obd_export *exp,
1447 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1449 struct obd_device *obd = class_exp2obd(exp);
1450 struct lmv_obd *lmv = &obd->u.lmv;
1451 struct obd_statfs *temp;
1456 rc = lmv_check_connect(obd);
1460 OBD_ALLOC(temp, sizeof(*temp));
1464 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1465 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1468 rc = obd_statfs(env, lmv->tgts[i]->ltd_exp, temp,
1471 CERROR("can't stat MDS #%d (%s), error %d\n", i,
1472 lmv->tgts[i]->ltd_exp->exp_obd->obd_name,
1474 GOTO(out_free_temp, rc);
1479 /* If the statfs is from mount, it will needs
1480 * retrieve necessary information from MDT0.
1481 * i.e. mount does not need the merged osfs
1483 * And also clients can be mounted as long as
1484 * MDT0 is in service*/
1485 if (flags & OBD_STATFS_FOR_MDT0)
1486 GOTO(out_free_temp, rc);
1488 osfs->os_bavail += temp->os_bavail;
1489 osfs->os_blocks += temp->os_blocks;
1490 osfs->os_ffree += temp->os_ffree;
1491 osfs->os_files += temp->os_files;
1497 OBD_FREE(temp, sizeof(*temp));
1501 static int lmv_getstatus(struct obd_export *exp, struct lu_fid *fid)
1503 struct obd_device *obd = exp->exp_obd;
1504 struct lmv_obd *lmv = &obd->u.lmv;
1508 rc = lmv_check_connect(obd);
1512 rc = md_getstatus(lmv->tgts[0]->ltd_exp, fid);
1516 static int lmv_getxattr(struct obd_export *exp, const struct lu_fid *fid,
1517 u64 valid, const char *name,
1518 const char *input, int input_size, int output_size,
1519 int flags, struct ptlrpc_request **request)
1521 struct obd_device *obd = exp->exp_obd;
1522 struct lmv_obd *lmv = &obd->u.lmv;
1523 struct lmv_tgt_desc *tgt;
1527 rc = lmv_check_connect(obd);
1531 tgt = lmv_find_target(lmv, fid);
1533 RETURN(PTR_ERR(tgt));
1535 rc = md_getxattr(tgt->ltd_exp, fid, valid, name, input,
1536 input_size, output_size, flags, request);
1541 static int lmv_setxattr(struct obd_export *exp, const struct lu_fid *fid,
1542 u64 valid, const char *name,
1543 const char *input, int input_size, int output_size,
1544 int flags, __u32 suppgid,
1545 struct ptlrpc_request **request)
1547 struct obd_device *obd = exp->exp_obd;
1548 struct lmv_obd *lmv = &obd->u.lmv;
1549 struct lmv_tgt_desc *tgt;
1553 rc = lmv_check_connect(obd);
1557 tgt = lmv_find_target(lmv, fid);
1559 RETURN(PTR_ERR(tgt));
1561 rc = md_setxattr(tgt->ltd_exp, fid, valid, name, input,
1562 input_size, output_size, flags, suppgid,
1568 static int lmv_getattr(struct obd_export *exp, struct md_op_data *op_data,
1569 struct ptlrpc_request **request)
1571 struct obd_device *obd = exp->exp_obd;
1572 struct lmv_obd *lmv = &obd->u.lmv;
1573 struct lmv_tgt_desc *tgt;
1577 rc = lmv_check_connect(obd);
1581 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1583 RETURN(PTR_ERR(tgt));
1585 if (op_data->op_flags & MF_GET_MDT_IDX) {
1586 op_data->op_mds = tgt->ltd_idx;
1590 rc = md_getattr(tgt->ltd_exp, op_data, request);
1595 static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid)
1597 struct obd_device *obd = exp->exp_obd;
1598 struct lmv_obd *lmv = &obd->u.lmv;
1603 rc = lmv_check_connect(obd);
1607 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1610 * With DNE every object can have two locks in different namespaces:
1611 * lookup lock in space of MDT storing direntry and update/open lock in
1612 * space of MDT storing inode.
1614 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
1615 if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL)
1617 md_null_inode(lmv->tgts[i]->ltd_exp, fid);
1623 static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid,
1624 ldlm_iterator_t it, void *data)
1626 struct obd_device *obd = exp->exp_obd;
1627 struct lmv_obd *lmv = &obd->u.lmv;
1633 rc = lmv_check_connect(obd);
1637 CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid));
1640 * With DNE every object can have two locks in different namespaces:
1641 * lookup lock in space of MDT storing direntry and update/open lock in
1642 * space of MDT storing inode. Try the MDT that the FID maps to first,
1643 * since this can be easily found, and only try others if that fails.
1645 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
1646 i < lmv->desc.ld_tgt_count;
1647 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
1649 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
1650 obd->obd_name, PFID(fid), tgt);
1654 if (lmv->tgts[tgt] == NULL ||
1655 lmv->tgts[tgt]->ltd_exp == NULL)
1658 rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data);
1667 static int lmv_close(struct obd_export *exp, struct md_op_data *op_data,
1668 struct md_open_data *mod, struct ptlrpc_request **request)
1670 struct obd_device *obd = exp->exp_obd;
1671 struct lmv_obd *lmv = &obd->u.lmv;
1672 struct lmv_tgt_desc *tgt;
1676 rc = lmv_check_connect(obd);
1680 tgt = lmv_find_target(lmv, &op_data->op_fid1);
1682 RETURN(PTR_ERR(tgt));
1684 CDEBUG(D_INODE, "CLOSE "DFID"\n", PFID(&op_data->op_fid1));
1685 rc = md_close(tgt->ltd_exp, op_data, mod, request);
1690 * Choosing the MDT by name or FID in @op_data.
1691 * For non-striped directory, it will locate MDT by fid.
1692 * For striped-directory, it will locate MDT by name. And also
1693 * it will reset op_fid1 with the FID of the choosen stripe.
1695 struct lmv_tgt_desc *
1696 lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm,
1697 const char *name, int namelen, struct lu_fid *fid,
1700 struct lmv_tgt_desc *tgt;
1701 const struct lmv_oinfo *oinfo;
1703 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_NAME_HASH)) {
1704 if (cfs_fail_val >= lsm->lsm_md_stripe_count)
1705 RETURN(ERR_PTR(-EBADF));
1706 oinfo = &lsm->lsm_md_oinfo[cfs_fail_val];
1708 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
1710 RETURN(ERR_CAST(oinfo));
1714 *fid = oinfo->lmo_fid;
1716 *mds = oinfo->lmo_mds;
1718 tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
1720 CDEBUG(D_INFO, "locate on mds %u "DFID"\n", oinfo->lmo_mds,
1721 PFID(&oinfo->lmo_fid));
1726 * Locate mds by fid or name
1728 * For striped directory (lsm != NULL), it will locate the stripe
1729 * by name hash (see lsm_name_to_stripe_info()). Note: if the hash_type
1730 * is unknown, it will return -EBADFD, and lmv_intent_lookup might need
1731 * walk through all of stripes to locate the entry.
1733 * For normal direcotry, it will locate MDS by FID directly.
1734 * \param[in] lmv LMV device
1735 * \param[in] op_data client MD stack parameters, name, namelen
1737 * \param[in] fid object FID used to locate MDS.
1739 * retval pointer to the lmv_tgt_desc if succeed.
1740 * ERR_PTR(errno) if failed.
1742 struct lmv_tgt_desc*
1743 lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data,
1746 struct lmv_stripe_md *lsm = op_data->op_mea1;
1747 struct lmv_tgt_desc *tgt;
1749 /* During creating VOLATILE file, it should honor the mdt
1750 * index if the file under striped dir is being restored, see
1752 if (op_data->op_bias & MDS_CREATE_VOLATILE &&
1753 (int)op_data->op_mds != -1 && lsm != NULL) {
1755 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
1759 /* refill the right parent fid */
1760 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1761 struct lmv_oinfo *oinfo;
1763 oinfo = &lsm->lsm_md_oinfo[i];
1764 if (oinfo->lmo_mds == op_data->op_mds) {
1765 *fid = oinfo->lmo_fid;
1770 /* Hmm, can not find the stripe by mdt_index(op_mds) */
1771 if (i == lsm->lsm_md_stripe_count)
1772 tgt = ERR_PTR(-EINVAL);
1777 if (lsm == NULL || op_data->op_namelen == 0) {
1778 tgt = lmv_find_target(lmv, fid);
1782 op_data->op_mds = tgt->ltd_idx;
1786 return lmv_locate_target_for_name(lmv, lsm, op_data->op_name,
1787 op_data->op_namelen, fid,
1791 int lmv_create(struct obd_export *exp, struct md_op_data *op_data,
1792 const void *data, size_t datalen, umode_t mode, uid_t uid,
1793 gid_t gid, cfs_cap_t cap_effective, __u64 rdev,
1794 struct ptlrpc_request **request)
1796 struct obd_device *obd = exp->exp_obd;
1797 struct lmv_obd *lmv = &obd->u.lmv;
1798 struct lmv_tgt_desc *tgt;
1802 rc = lmv_check_connect(obd);
1806 if (!lmv->desc.ld_active_tgt_count)
1809 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1811 RETURN(PTR_ERR(tgt));
1813 CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n",
1814 (int)op_data->op_namelen, op_data->op_name,
1815 PFID(&op_data->op_fid1), op_data->op_mds);
1817 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1820 if (exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE) {
1821 /* Send the create request to the MDT where the object
1822 * will be located */
1823 tgt = lmv_find_target(lmv, &op_data->op_fid2);
1825 RETURN(PTR_ERR(tgt));
1827 op_data->op_mds = tgt->ltd_idx;
1829 CDEBUG(D_CONFIG, "Server doesn't support striped dirs\n");
1832 CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n",
1833 PFID(&op_data->op_fid2), op_data->op_mds);
1835 op_data->op_flags |= MF_MDC_CANCEL_FID1;
1836 rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid,
1837 cap_effective, rdev, request);
1839 if (*request == NULL)
1841 CDEBUG(D_INODE, "Created - "DFID"\n", PFID(&op_data->op_fid2));
1847 lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1848 const union ldlm_policy_data *policy,
1849 struct lookup_intent *it, struct md_op_data *op_data,
1850 struct lustre_handle *lockh, __u64 extra_lock_flags)
1852 struct obd_device *obd = exp->exp_obd;
1853 struct lmv_obd *lmv = &obd->u.lmv;
1854 struct lmv_tgt_desc *tgt;
1858 rc = lmv_check_connect(obd);
1862 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID"\n",
1863 LL_IT2STR(it), PFID(&op_data->op_fid1));
1865 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1867 RETURN(PTR_ERR(tgt));
1869 CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%u\n",
1870 LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx);
1872 rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh,
1879 lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data,
1880 struct ptlrpc_request **preq)
1882 struct ptlrpc_request *req = NULL;
1883 struct obd_device *obd = exp->exp_obd;
1884 struct lmv_obd *lmv = &obd->u.lmv;
1885 struct lmv_tgt_desc *tgt;
1886 struct mdt_body *body;
1890 rc = lmv_check_connect(obd);
1894 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
1896 RETURN(PTR_ERR(tgt));
1898 CDEBUG(D_INODE, "GETATTR_NAME for %*s on "DFID" -> mds #%d\n",
1899 (int)op_data->op_namelen, op_data->op_name,
1900 PFID(&op_data->op_fid1), tgt->ltd_idx);
1902 rc = md_getattr_name(tgt->ltd_exp, op_data, preq);
1906 body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY);
1907 LASSERT(body != NULL);
1909 if (body->mbo_valid & OBD_MD_MDS) {
1910 struct lu_fid rid = body->mbo_fid1;
1911 CDEBUG(D_INODE, "Request attrs for "DFID"\n",
1914 tgt = lmv_find_target(lmv, &rid);
1916 ptlrpc_req_finished(*preq);
1918 RETURN(PTR_ERR(tgt));
1921 op_data->op_fid1 = rid;
1922 op_data->op_valid |= OBD_MD_FLCROSSREF;
1923 op_data->op_namelen = 0;
1924 op_data->op_name = NULL;
1925 rc = md_getattr_name(tgt->ltd_exp, op_data, &req);
1926 ptlrpc_req_finished(*preq);
1933 #define md_op_data_fid(op_data, fl) \
1934 (fl == MF_MDC_CANCEL_FID1 ? &op_data->op_fid1 : \
1935 fl == MF_MDC_CANCEL_FID2 ? &op_data->op_fid2 : \
1936 fl == MF_MDC_CANCEL_FID3 ? &op_data->op_fid3 : \
1937 fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \
1940 static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt,
1941 struct md_op_data *op_data,
1942 __u32 op_tgt, ldlm_mode_t mode, int bits, int flag)
1944 struct lu_fid *fid = md_op_data_fid(op_data, flag);
1945 struct obd_device *obd = exp->exp_obd;
1946 struct lmv_obd *lmv = &obd->u.lmv;
1947 ldlm_policy_data_t policy = {{ 0 }};
1951 if (!fid_is_sane(fid))
1955 tgt = lmv_find_target(lmv, fid);
1957 RETURN(PTR_ERR(tgt));
1960 if (tgt->ltd_idx != op_tgt) {
1961 CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid));
1962 policy.l_inodebits.bits = bits;
1963 rc = md_cancel_unused(tgt->ltd_exp, fid, &policy,
1964 mode, LCF_ASYNC, NULL);
1967 "EARLY_CANCEL skip operation target %d on "DFID"\n",
1969 op_data->op_flags |= flag;
1977 * llite passes fid of an target inode in op_data->op_fid1 and id of directory in
1980 static int lmv_link(struct obd_export *exp, struct md_op_data *op_data,
1981 struct ptlrpc_request **request)
1983 struct obd_device *obd = exp->exp_obd;
1984 struct lmv_obd *lmv = &obd->u.lmv;
1985 struct lmv_tgt_desc *tgt;
1989 rc = lmv_check_connect(obd);
1993 LASSERT(op_data->op_namelen != 0);
1995 CDEBUG(D_INODE, "LINK "DFID":%*s to "DFID"\n",
1996 PFID(&op_data->op_fid2), (int)op_data->op_namelen,
1997 op_data->op_name, PFID(&op_data->op_fid1));
1999 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2000 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2001 op_data->op_cap = cfs_curproc_cap_pack();
2002 if (op_data->op_mea2 != NULL) {
2003 struct lmv_stripe_md *lsm = op_data->op_mea2;
2004 const struct lmv_oinfo *oinfo;
2006 oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name,
2007 op_data->op_namelen);
2009 RETURN(PTR_ERR(oinfo));
2011 op_data->op_fid2 = oinfo->lmo_fid;
2014 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2);
2016 RETURN(PTR_ERR(tgt));
2019 * Cancel UPDATE lock on child (fid1).
2021 op_data->op_flags |= MF_MDC_CANCEL_FID2;
2022 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2023 MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1);
2027 rc = md_link(tgt->ltd_exp, op_data, request);
2032 static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data,
2033 const char *old, size_t oldlen,
2034 const char *new, size_t newlen,
2035 struct ptlrpc_request **request)
2037 struct obd_device *obd = exp->exp_obd;
2038 struct lmv_obd *lmv = &obd->u.lmv;
2039 struct lmv_tgt_desc *src_tgt;
2040 struct lmv_tgt_desc *tgt_tgt;
2041 struct obd_export *target_exp;
2042 struct mdt_body *body;
2046 LASSERT(oldlen != 0);
2048 CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n",
2049 (int)oldlen, old, PFID(&op_data->op_fid1),
2050 op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0,
2051 (int)newlen, new, PFID(&op_data->op_fid2),
2052 op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0);
2054 rc = lmv_check_connect(obd);
2058 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2059 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2060 op_data->op_cap = cfs_curproc_cap_pack();
2061 if (op_data->op_cli_flags & CLI_MIGRATE) {
2062 LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n",
2063 PFID(&op_data->op_fid3));
2065 if (op_data->op_mea1 != NULL) {
2066 struct lmv_stripe_md *lsm = op_data->op_mea1;
2067 struct lmv_tgt_desc *tmp;
2069 /* Fix the parent fid for striped dir */
2070 tmp = lmv_locate_target_for_name(lmv, lsm, old,
2075 RETURN(PTR_ERR(tmp));
2078 rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
2082 src_tgt = lmv_find_target(lmv, &op_data->op_fid3);
2083 if (IS_ERR(src_tgt))
2084 RETURN(PTR_ERR(src_tgt));
2086 target_exp = src_tgt->ltd_exp;
2088 if (op_data->op_mea1 != NULL) {
2089 struct lmv_stripe_md *lsm = op_data->op_mea1;
2091 src_tgt = lmv_locate_target_for_name(lmv, lsm, old,
2096 src_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2098 if (IS_ERR(src_tgt))
2099 RETURN(PTR_ERR(src_tgt));
2102 if (op_data->op_mea2 != NULL) {
2103 struct lmv_stripe_md *lsm = op_data->op_mea2;
2105 tgt_tgt = lmv_locate_target_for_name(lmv, lsm, new,
2110 tgt_tgt = lmv_find_target(lmv, &op_data->op_fid2);
2113 if (IS_ERR(tgt_tgt))
2114 RETURN(PTR_ERR(tgt_tgt));
2116 target_exp = tgt_tgt->ltd_exp;
2120 * LOOKUP lock on src child (fid3) should also be cancelled for
2121 * src_tgt in mdc_rename.
2123 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2126 * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its
2129 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2130 LCK_EX, MDS_INODELOCK_UPDATE,
2131 MF_MDC_CANCEL_FID2);
2136 * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt.
2138 if (fid_is_sane(&op_data->op_fid3)) {
2139 struct lmv_tgt_desc *tgt;
2141 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2143 RETURN(PTR_ERR(tgt));
2145 /* Cancel LOOKUP lock on its parent */
2146 rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx,
2147 LCK_EX, MDS_INODELOCK_LOOKUP,
2148 MF_MDC_CANCEL_FID3);
2152 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2153 LCK_EX, MDS_INODELOCK_FULL,
2154 MF_MDC_CANCEL_FID3);
2161 * Cancel all the locks on tgt child (fid4).
2163 if (fid_is_sane(&op_data->op_fid4)) {
2164 struct lmv_tgt_desc *tgt;
2166 rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx,
2167 LCK_EX, MDS_INODELOCK_FULL,
2168 MF_MDC_CANCEL_FID4);
2172 tgt = lmv_find_target(lmv, &op_data->op_fid4);
2174 RETURN(PTR_ERR(tgt));
2176 /* Since the target child might be destroyed, and it might
2177 * become orphan, and we can only check orphan on the local
2178 * MDT right now, so we send rename request to the MDT where
2179 * target child is located. If target child does not exist,
2180 * then it will send the request to the target parent */
2181 target_exp = tgt->ltd_exp;
2184 rc = md_rename(target_exp, op_data, old, oldlen, new, newlen,
2187 if (rc != 0 && rc != -EXDEV)
2190 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2194 /* Not cross-ref case, just get out of here. */
2195 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2198 CDEBUG(D_INODE, "%s: try rename to another MDT for "DFID"\n",
2199 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2201 op_data->op_fid4 = body->mbo_fid1;
2202 ptlrpc_req_finished(*request);
2207 static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data,
2208 void *ea, size_t ealen, struct ptlrpc_request **request)
2210 struct obd_device *obd = exp->exp_obd;
2211 struct lmv_obd *lmv = &obd->u.lmv;
2212 struct lmv_tgt_desc *tgt;
2216 rc = lmv_check_connect(obd);
2220 CDEBUG(D_INODE, "SETATTR for "DFID", valid 0x%x\n",
2221 PFID(&op_data->op_fid1), op_data->op_attr.ia_valid);
2223 op_data->op_flags |= MF_MDC_CANCEL_FID1;
2224 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2226 RETURN(PTR_ERR(tgt));
2228 rc = md_setattr(tgt->ltd_exp, op_data, ea, ealen, request);
2233 static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid,
2234 struct ptlrpc_request **request)
2236 struct obd_device *obd = exp->exp_obd;
2237 struct lmv_obd *lmv = &obd->u.lmv;
2238 struct lmv_tgt_desc *tgt;
2242 rc = lmv_check_connect(obd);
2246 tgt = lmv_find_target(lmv, fid);
2248 RETURN(PTR_ERR(tgt));
2250 rc = md_fsync(tgt->ltd_exp, fid, request);
2255 * Get current minimum entry from striped directory
2257 * This function will search the dir entry, whose hash value is the
2258 * closest(>=) to @hash_offset, from all of sub-stripes, and it is
2259 * only being called for striped directory.
2261 * \param[in] exp export of LMV
2262 * \param[in] op_data parameters transferred beween client MD stack
2263 * stripe_information will be included in this
2265 * \param[in] cb_op ldlm callback being used in enqueue in
2267 * \param[in] hash_offset the hash value, which is used to locate
2268 * minum(closet) dir entry
2269 * \param[in|out] stripe_offset the caller use this to indicate the stripe
2270 * index of last entry, so to avoid hash conflict
2271 * between stripes. It will also be used to
2272 * return the stripe index of current dir entry.
2273 * \param[in|out] entp the minum entry and it also is being used
2274 * to input the last dir entry to resolve the
2277 * \param[out] ppage the page which holds the minum entry
2279 * \retval = 0 get the entry successfully
2280 * negative errno (< 0) does not get the entry
2282 static int lmv_get_min_striped_entry(struct obd_export *exp,
2283 struct md_op_data *op_data,
2284 struct md_callback *cb_op,
2285 __u64 hash_offset, int *stripe_offset,
2286 struct lu_dirent **entp,
2287 struct page **ppage)
2289 struct obd_device *obd = exp->exp_obd;
2290 struct lmv_obd *lmv = &obd->u.lmv;
2291 struct lmv_stripe_md *lsm = op_data->op_mea1;
2292 struct lmv_tgt_desc *tgt;
2294 struct lu_dirent *min_ent = NULL;
2295 struct page *min_page = NULL;
2301 stripe_count = lsm->lsm_md_stripe_count;
2302 for (i = 0; i < stripe_count; i++) {
2303 struct lu_dirent *ent = NULL;
2304 struct page *page = NULL;
2305 struct lu_dirpage *dp;
2306 __u64 stripe_hash = hash_offset;
2308 tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL);
2310 GOTO(out, rc = PTR_ERR(tgt));
2312 /* op_data will be shared by each stripe, so we need
2313 * reset these value for each stripe */
2314 op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid;
2315 op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid;
2316 op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root;
2318 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, stripe_hash,
2323 dp = page_address(page);
2324 for (ent = lu_dirent_start(dp); ent != NULL;
2325 ent = lu_dirent_next(ent)) {
2326 /* Skip dummy entry */
2327 if (le16_to_cpu(ent->lde_namelen) == 0)
2330 if (le64_to_cpu(ent->lde_hash) < hash_offset)
2333 if (le64_to_cpu(ent->lde_hash) == hash_offset &&
2334 (*entp == ent || i < *stripe_offset))
2337 /* skip . and .. for other stripes */
2339 (strncmp(ent->lde_name, ".",
2340 le16_to_cpu(ent->lde_namelen)) == 0 ||
2341 strncmp(ent->lde_name, "..",
2342 le16_to_cpu(ent->lde_namelen)) == 0))
2348 stripe_hash = le64_to_cpu(dp->ldp_hash_end);
2351 page_cache_release(page);
2354 /* reach the end of current stripe, go to next stripe */
2355 if (stripe_hash == MDS_DIR_END_OFF)
2361 if (min_ent != NULL) {
2362 if (le64_to_cpu(min_ent->lde_hash) >
2363 le64_to_cpu(ent->lde_hash)) {
2366 page_cache_release(min_page);
2371 page_cache_release(page);
2382 if (*ppage != NULL) {
2384 page_cache_release(*ppage);
2386 *stripe_offset = min_idx;
2393 * Build dir entry page from a striped directory
2395 * This function gets one entry by @offset from a striped directory. It will
2396 * read entries from all of stripes, and choose one closest to the required
2397 * offset(&offset). A few notes
2398 * 1. skip . and .. for non-zero stripes, because there can only have one .
2399 * and .. in a directory.
2400 * 2. op_data will be shared by all of stripes, instead of allocating new
2401 * one, so need to restore before reusing.
2402 * 3. release the entry page if that is not being chosen.
2404 * \param[in] exp obd export refer to LMV
2405 * \param[in] op_data hold those MD parameters of read_entry
2406 * \param[in] cb_op ldlm callback being used in enqueue in mdc_read_entry
2407 * \param[out] ldp the entry being read
2408 * \param[out] ppage the page holding the entry. Note: because the entry
2409 * will be accessed in upper layer, so we need hold the
2410 * page until the usages of entry is finished, see
2411 * ll_dir_entry_next.
2413 * retval =0 if get entry successfully
2414 * <0 cannot get entry
2416 static int lmv_read_striped_page(struct obd_export *exp,
2417 struct md_op_data *op_data,
2418 struct md_callback *cb_op,
2419 __u64 offset, struct page **ppage)
2421 struct obd_device *obd = exp->exp_obd;
2422 struct lu_fid master_fid = op_data->op_fid1;
2423 struct inode *master_inode = op_data->op_data;
2424 __u64 hash_offset = offset;
2425 struct lu_dirpage *dp;
2426 struct page *min_ent_page = NULL;
2427 struct page *ent_page = NULL;
2428 struct lu_dirent *ent;
2431 struct lu_dirent *min_ent = NULL;
2432 struct lu_dirent *last_ent;
2437 rc = lmv_check_connect(obd);
2441 /* Allocate a page and read entries from all of stripes and fill
2442 * the page by hash order */
2443 ent_page = alloc_page(GFP_KERNEL);
2444 if (ent_page == NULL)
2447 /* Initialize the entry page */
2448 dp = kmap(ent_page);
2449 memset(dp, 0, sizeof(*dp));
2450 dp->ldp_hash_start = cpu_to_le64(offset);
2451 dp->ldp_flags |= LDF_COLLIDE;
2454 left_bytes = PAGE_CACHE_SIZE - sizeof(*dp);
2460 /* Find the minum entry from all sub-stripes */
2461 rc = lmv_get_min_striped_entry(exp, op_data, cb_op, hash_offset,
2467 /* If it can not get minum entry, it means it already reaches
2468 * the end of this directory */
2469 if (min_ent == NULL) {
2470 last_ent->lde_reclen = 0;
2471 hash_offset = MDS_DIR_END_OFF;
2475 ent_size = le16_to_cpu(min_ent->lde_reclen);
2477 /* the last entry lde_reclen is 0, but it might not
2478 * the end of this entry of this temporay entry */
2480 ent_size = lu_dirent_calc_size(
2481 le16_to_cpu(min_ent->lde_namelen),
2482 le32_to_cpu(min_ent->lde_attrs));
2483 if (ent_size > left_bytes) {
2484 last_ent->lde_reclen = cpu_to_le16(0);
2485 hash_offset = le64_to_cpu(min_ent->lde_hash);
2489 memcpy(ent, min_ent, ent_size);
2491 /* Replace . with master FID and Replace .. with the parent FID
2492 * of master object */
2493 if (strncmp(ent->lde_name, ".",
2494 le16_to_cpu(ent->lde_namelen)) == 0 &&
2495 le16_to_cpu(ent->lde_namelen) == 1)
2496 fid_cpu_to_le(&ent->lde_fid, &master_fid);
2497 else if (strncmp(ent->lde_name, "..",
2498 le16_to_cpu(ent->lde_namelen)) == 0 &&
2499 le16_to_cpu(ent->lde_namelen) == 2)
2500 fid_cpu_to_le(&ent->lde_fid, &op_data->op_fid3);
2502 left_bytes -= ent_size;
2503 ent->lde_reclen = cpu_to_le16(ent_size);
2505 ent = (void *)ent + ent_size;
2506 hash_offset = le64_to_cpu(min_ent->lde_hash);
2507 if (hash_offset == MDS_DIR_END_OFF) {
2508 last_ent->lde_reclen = 0;
2513 if (min_ent_page != NULL) {
2514 kunmap(min_ent_page);
2515 page_cache_release(min_ent_page);
2518 if (unlikely(rc != 0)) {
2519 __free_page(ent_page);
2523 dp->ldp_flags |= LDF_EMPTY;
2524 dp->ldp_flags = cpu_to_le32(dp->ldp_flags);
2525 dp->ldp_hash_end = cpu_to_le64(hash_offset);
2528 /* We do not want to allocate md_op_data during each
2529 * dir entry reading, so op_data will be shared by every stripe,
2530 * then we need to restore it back to original value before
2531 * return to the upper layer */
2532 op_data->op_fid1 = master_fid;
2533 op_data->op_fid2 = master_fid;
2534 op_data->op_data = master_inode;
2541 int lmv_read_page(struct obd_export *exp, struct md_op_data *op_data,
2542 struct md_callback *cb_op, __u64 offset,
2543 struct page **ppage)
2545 struct obd_device *obd = exp->exp_obd;
2546 struct lmv_obd *lmv = &obd->u.lmv;
2547 struct lmv_stripe_md *lsm = op_data->op_mea1;
2548 struct lmv_tgt_desc *tgt;
2552 rc = lmv_check_connect(obd);
2556 if (unlikely(lsm != NULL)) {
2557 rc = lmv_read_striped_page(exp, op_data, cb_op, offset, ppage);
2561 tgt = lmv_find_target(lmv, &op_data->op_fid1);
2563 RETURN(PTR_ERR(tgt));
2565 rc = md_read_page(tgt->ltd_exp, op_data, cb_op, offset, ppage);
2571 * Unlink a file/directory
2573 * Unlink a file or directory under the parent dir. The unlink request
2574 * usually will be sent to the MDT where the child is located, but if
2575 * the client does not have the child FID then request will be sent to the
2576 * MDT where the parent is located.
2578 * If the parent is a striped directory then it also needs to locate which
2579 * stripe the name of the child is located, and replace the parent FID
2580 * (@op->op_fid1) with the stripe FID. Note: if the stripe is unknown,
2581 * it will walk through all of sub-stripes until the child is being
2584 * \param[in] exp export refer to LMV
2585 * \param[in] op_data different parameters transferred beween client
2586 * MD stacks, name, namelen, FIDs etc.
2587 * op_fid1 is the parent FID, op_fid2 is the child
2589 * \param[out] request point to the request of unlink.
2591 * retval 0 if succeed
2592 * negative errno if failed.
2594 static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data,
2595 struct ptlrpc_request **request)
2597 struct obd_device *obd = exp->exp_obd;
2598 struct lmv_obd *lmv = &obd->u.lmv;
2599 struct lmv_tgt_desc *tgt = NULL;
2600 struct lmv_tgt_desc *parent_tgt = NULL;
2601 struct mdt_body *body;
2603 int stripe_index = 0;
2604 struct lmv_stripe_md *lsm = op_data->op_mea1;
2607 rc = lmv_check_connect(obd);
2611 /* For striped dir, we need to locate the parent as well */
2613 struct lmv_tgt_desc *tmp;
2615 LASSERT(op_data->op_name != NULL &&
2616 op_data->op_namelen != 0);
2618 tmp = lmv_locate_target_for_name(lmv, lsm,
2620 op_data->op_namelen,
2624 /* return -EBADFD means unknown hash type, might
2625 * need try all sub-stripe here */
2626 if (IS_ERR(tmp) && PTR_ERR(tmp) != -EBADFD)
2627 RETURN(PTR_ERR(tmp));
2629 /* Note: both migrating dir and unknown hash dir need to
2630 * try all of sub-stripes, so we need start search the
2631 * name from stripe 0, but migrating dir is already handled
2632 * inside lmv_locate_target_for_name(), so we only check
2633 * unknown hash type directory here */
2634 if (!lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
2635 struct lmv_oinfo *oinfo;
2637 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2639 op_data->op_fid1 = oinfo->lmo_fid;
2640 op_data->op_mds = oinfo->lmo_mds;
2645 /* Send unlink requests to the MDT where the child is located */
2646 if (likely(!fid_is_zero(&op_data->op_fid2)))
2647 tgt = lmv_find_target(lmv, &op_data->op_fid2);
2648 else if (lsm != NULL)
2649 tgt = lmv_get_target(lmv, op_data->op_mds, NULL);
2651 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
2654 RETURN(PTR_ERR(tgt));
2656 op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2657 op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2658 op_data->op_cap = cfs_curproc_cap_pack();
2661 * If child's fid is given, cancel unused locks for it if it is from
2662 * another export than parent.
2664 * LOOKUP lock for child (fid3) should also be cancelled on parent
2665 * tgt_tgt in mdc_unlink().
2667 op_data->op_flags |= MF_MDC_CANCEL_FID1 | MF_MDC_CANCEL_FID3;
2670 * Cancel FULL locks on child (fid3).
2672 parent_tgt = lmv_find_target(lmv, &op_data->op_fid1);
2673 if (IS_ERR(parent_tgt))
2674 RETURN(PTR_ERR(parent_tgt));
2676 if (parent_tgt != tgt) {
2677 rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx,
2678 LCK_EX, MDS_INODELOCK_LOOKUP,
2679 MF_MDC_CANCEL_FID3);
2682 rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX,
2683 MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3);
2687 CDEBUG(D_INODE, "unlink with fid="DFID"/"DFID" -> mds #%u\n",
2688 PFID(&op_data->op_fid1), PFID(&op_data->op_fid2), tgt->ltd_idx);
2690 rc = md_unlink(tgt->ltd_exp, op_data, request);
2691 if (rc != 0 && rc != -EREMOTE && rc != -ENOENT)
2694 /* Try next stripe if it is needed. */
2695 if (rc == -ENOENT && lsm != NULL && lmv_need_try_all_stripes(lsm)) {
2696 struct lmv_oinfo *oinfo;
2699 if (stripe_index >= lsm->lsm_md_stripe_count)
2702 oinfo = &lsm->lsm_md_oinfo[stripe_index];
2704 op_data->op_fid1 = oinfo->lmo_fid;
2705 op_data->op_mds = oinfo->lmo_mds;
2707 ptlrpc_req_finished(*request);
2710 goto try_next_stripe;
2713 body = req_capsule_server_get(&(*request)->rq_pill, &RMF_MDT_BODY);
2717 /* Not cross-ref case, just get out of here. */
2718 if (likely(!(body->mbo_valid & OBD_MD_MDS)))
2721 CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n",
2722 exp->exp_obd->obd_name, PFID(&body->mbo_fid1));
2724 /* This is a remote object, try remote MDT, Note: it may
2725 * try more than 1 time here, Considering following case
2726 * /mnt/lustre is root on MDT0, remote1 is on MDT1
2727 * 1. Initially A does not know where remote1 is, it send
2728 * unlink RPC to MDT0, MDT0 return -EREMOTE, it will
2729 * resend unlink RPC to MDT1 (retry 1st time).
2731 * 2. During the unlink RPC in flight,
2732 * client B mv /mnt/lustre/remote1 /mnt/lustre/remote2
2733 * and create new remote1, but on MDT0
2735 * 3. MDT1 get unlink RPC(from A), then do remote lock on
2736 * /mnt/lustre, then lookup get fid of remote1, and find
2737 * it is remote dir again, and replay -EREMOTE again.
2739 * 4. Then A will resend unlink RPC to MDT0. (retry 2nd times).
2741 * In theory, it might try unlimited time here, but it should
2742 * be very rare case. */
2743 op_data->op_fid2 = body->mbo_fid1;
2744 ptlrpc_req_finished(*request);
2750 static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
2752 struct lmv_obd *lmv = &obd->u.lmv;
2756 case OBD_CLEANUP_EARLY:
2757 /* XXX: here should be calling obd_precleanup() down to
2760 case OBD_CLEANUP_EXPORTS:
2761 fld_client_proc_fini(&lmv->lmv_fld);
2762 lprocfs_obd_cleanup(obd);
2763 lprocfs_free_md_stats(obd);
2772 * Get by key a value associated with a LMV device.
2774 * Dispatch request to lower-layer devices as needed.
2776 * \param[in] env execution environment for this thread
2777 * \param[in] exp export for the LMV device
2778 * \param[in] keylen length of key identifier
2779 * \param[in] key identifier of key to get value for
2780 * \param[in] vallen size of \a val
2781 * \param[out] val pointer to storage location for value
2782 * \param[in] lsm optional striping metadata of object
2784 * \retval 0 on success
2785 * \retval negative negated errno on failure
2787 static int lmv_get_info(const struct lu_env *env, struct obd_export *exp,
2788 __u32 keylen, void *key, __u32 *vallen, void *val)
2790 struct obd_device *obd;
2791 struct lmv_obd *lmv;
2795 obd = class_exp2obd(exp);
2797 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2798 exp->exp_handle.h_cookie);
2803 if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) {
2806 rc = lmv_check_connect(obd);
2810 LASSERT(*vallen == sizeof(__u32));
2811 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2812 struct lmv_tgt_desc *tgt = lmv->tgts[i];
2814 * All tgts should be connected when this gets called.
2816 if (tgt == NULL || tgt->ltd_exp == NULL)
2819 if (!obd_get_info(env, tgt->ltd_exp, keylen, key,
2824 } else if (KEY_IS(KEY_MAX_EASIZE) ||
2825 KEY_IS(KEY_DEFAULT_EASIZE) ||
2826 KEY_IS(KEY_CONN_DATA)) {
2827 rc = lmv_check_connect(obd);
2832 * Forwarding this request to first MDS, it should know LOV
2835 rc = obd_get_info(env, lmv->tgts[0]->ltd_exp, keylen, key,
2837 if (!rc && KEY_IS(KEY_CONN_DATA))
2838 exp->exp_connect_data = *(struct obd_connect_data *)val;
2840 } else if (KEY_IS(KEY_TGT_COUNT)) {
2841 *((int *)val) = lmv->desc.ld_tgt_count;
2845 CDEBUG(D_IOCTL, "Invalid key\n");
2850 * Asynchronously set by key a value associated with a LMV device.
2852 * Dispatch request to lower-layer devices as needed.
2854 * \param[in] env execution environment for this thread
2855 * \param[in] exp export for the LMV device
2856 * \param[in] keylen length of key identifier
2857 * \param[in] key identifier of key to store value for
2858 * \param[in] vallen size of value to store
2859 * \param[in] val pointer to data to be stored
2860 * \param[in] set optional list of related ptlrpc requests
2862 * \retval 0 on success
2863 * \retval negative negated errno on failure
2865 int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp,
2866 __u32 keylen, void *key, __u32 vallen, void *val,
2867 struct ptlrpc_request_set *set)
2869 struct lmv_tgt_desc *tgt = NULL;
2870 struct obd_device *obd;
2871 struct lmv_obd *lmv;
2875 obd = class_exp2obd(exp);
2877 CDEBUG(D_IOCTL, "Invalid client cookie "LPX64"\n",
2878 exp->exp_handle.h_cookie);
2883 if (KEY_IS(KEY_READ_ONLY) || KEY_IS(KEY_FLUSH_CTX) ||
2884 KEY_IS(KEY_DEFAULT_EASIZE)) {
2887 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
2890 if (tgt == NULL || tgt->ltd_exp == NULL)
2893 err = obd_set_info_async(env, tgt->ltd_exp,
2894 keylen, key, vallen, val, set);
2905 static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm,
2906 const struct lmv_mds_md_v1 *lmm1)
2908 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
2915 lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic);
2916 lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2917 lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index);
2918 if (OBD_FAIL_CHECK(OBD_FAIL_UNKNOWN_LMV_STRIPE))
2919 lsm->lsm_md_hash_type = LMV_HASH_TYPE_UNKNOWN;
2921 lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type);
2922 lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version);
2923 cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name,
2924 sizeof(lsm->lsm_md_pool_name));
2926 if (cplen >= sizeof(lsm->lsm_md_pool_name))
2929 CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d"
2930 "layout_version %d\n", lsm->lsm_md_stripe_count,
2931 lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type,
2932 lsm->lsm_md_layout_version);
2934 stripe_count = le32_to_cpu(lmm1->lmv_stripe_count);
2935 for (i = 0; i < stripe_count; i++) {
2936 fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid,
2937 &lmm1->lmv_stripe_fids[i]);
2938 rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid,
2939 &lsm->lsm_md_oinfo[i].lmo_mds);
2942 CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i,
2943 PFID(&lsm->lsm_md_oinfo[i].lmo_fid));
2949 static int lmv_unpackmd(struct obd_export *exp, struct lmv_stripe_md **lsmp,
2950 const union lmv_mds_md *lmm, size_t lmm_size)
2952 struct lmv_stripe_md *lsm;
2955 bool allocated = false;
2958 LASSERT(lsmp != NULL);
2962 if (lsm != NULL && lmm == NULL) {
2964 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
2965 /* For migrating inode, the master stripe and master
2966 * object will be the same, so do not need iput, see
2967 * ll_update_lsm_md */
2968 if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION &&
2969 i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL)
2970 iput(lsm->lsm_md_oinfo[i].lmo_root);
2972 lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count);
2973 OBD_FREE(lsm, lsm_size);
2978 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE)
2982 if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 &&
2983 le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) {
2984 CERROR("%s: invalid lmv magic %x: rc = %d\n",
2985 exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic),
2990 if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1)
2991 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
2994 * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md,
2995 * stripecount should be 0 then.
2997 lsm_size = lmv_stripe_md_size(0);
2999 lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm));
3001 OBD_ALLOC(lsm, lsm_size);
3008 switch (le32_to_cpu(lmm->lmv_magic)) {
3010 rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1);
3013 CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name,
3014 le32_to_cpu(lmm->lmv_magic));
3019 if (rc != 0 && allocated) {
3020 OBD_FREE(lsm, lsm_size);
3027 void lmv_free_memmd(struct lmv_stripe_md *lsm)
3029 lmv_unpackmd(NULL, &lsm, NULL, 0);
3031 EXPORT_SYMBOL(lmv_free_memmd);
3033 static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
3034 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3035 ldlm_cancel_flags_t flags, void *opaque)
3037 struct obd_device *obd = exp->exp_obd;
3038 struct lmv_obd *lmv = &obd->u.lmv;
3044 LASSERT(fid != NULL);
3046 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3047 struct lmv_tgt_desc *tgt = lmv->tgts[i];
3049 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3052 err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags,
3060 int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
3063 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3064 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3068 if (tgt == NULL || tgt->ltd_exp == NULL)
3070 rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits);
3074 ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags,
3075 const struct lu_fid *fid, ldlm_type_t type,
3076 ldlm_policy_data_t *policy, ldlm_mode_t mode,
3077 struct lustre_handle *lockh)
3079 struct obd_device *obd = exp->exp_obd;
3080 struct lmv_obd *lmv = &obd->u.lmv;
3086 CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid));
3089 * With DNE every object can have two locks in different namespaces:
3090 * lookup lock in space of MDT storing direntry and update/open lock in
3091 * space of MDT storing inode. Try the MDT that the FID maps to first,
3092 * since this can be easily found, and only try others if that fails.
3094 for (i = 0, tgt = lmv_find_target_index(lmv, fid);
3095 i < lmv->desc.ld_tgt_count;
3096 i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) {
3098 CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n",
3099 obd->obd_name, PFID(fid), tgt);
3103 if (lmv->tgts[tgt] == NULL ||
3104 lmv->tgts[tgt]->ltd_exp == NULL ||
3105 lmv->tgts[tgt]->ltd_active == 0)
3108 rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid,
3109 type, policy, mode, lockh);
3117 int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
3118 struct obd_export *dt_exp, struct obd_export *md_exp,
3119 struct lustre_md *md)
3121 struct lmv_obd *lmv = &exp->exp_obd->u.lmv;
3122 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3124 if (tgt == NULL || tgt->ltd_exp == NULL)
3127 return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md);
3130 int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
3132 struct obd_device *obd = exp->exp_obd;
3133 struct lmv_obd *lmv = &obd->u.lmv;
3134 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3137 if (md->lmv != NULL) {
3138 lmv_free_memmd(md->lmv);
3141 if (tgt == NULL || tgt->ltd_exp == NULL)
3143 RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md));
3146 int lmv_set_open_replay_data(struct obd_export *exp,
3147 struct obd_client_handle *och,
3148 struct lookup_intent *it)
3150 struct obd_device *obd = exp->exp_obd;
3151 struct lmv_obd *lmv = &obd->u.lmv;
3152 struct lmv_tgt_desc *tgt;
3155 tgt = lmv_find_target(lmv, &och->och_fid);
3157 RETURN(PTR_ERR(tgt));
3159 RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it));
3162 int lmv_clear_open_replay_data(struct obd_export *exp,
3163 struct obd_client_handle *och)
3165 struct obd_device *obd = exp->exp_obd;
3166 struct lmv_obd *lmv = &obd->u.lmv;
3167 struct lmv_tgt_desc *tgt;
3170 tgt = lmv_find_target(lmv, &och->och_fid);
3172 RETURN(PTR_ERR(tgt));
3174 RETURN(md_clear_open_replay_data(tgt->ltd_exp, och));
3177 static int lmv_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
3178 u32 suppgid, struct ptlrpc_request **request)
3180 struct obd_device *obd = exp->exp_obd;
3181 struct lmv_obd *lmv = &obd->u.lmv;
3182 struct lmv_tgt_desc *tgt;
3186 rc = lmv_check_connect(obd);
3190 tgt = lmv_find_target(lmv, fid);
3192 RETURN(PTR_ERR(tgt));
3194 rc = md_get_remote_perm(tgt->ltd_exp, fid, suppgid, request);
3198 int lmv_intent_getattr_async(struct obd_export *exp,
3199 struct md_enqueue_info *minfo,
3200 struct ldlm_enqueue_info *einfo)
3202 struct md_op_data *op_data = &minfo->mi_data;
3203 struct obd_device *obd = exp->exp_obd;
3204 struct lmv_obd *lmv = &obd->u.lmv;
3205 struct lmv_tgt_desc *tgt = NULL;
3209 rc = lmv_check_connect(obd);
3213 tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
3215 RETURN(PTR_ERR(tgt));
3217 rc = md_intent_getattr_async(tgt->ltd_exp, minfo, einfo);
3221 int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
3222 struct lu_fid *fid, __u64 *bits)
3224 struct obd_device *obd = exp->exp_obd;
3225 struct lmv_obd *lmv = &obd->u.lmv;
3226 struct lmv_tgt_desc *tgt;
3230 rc = lmv_check_connect(obd);
3234 tgt = lmv_find_target(lmv, fid);
3236 RETURN(PTR_ERR(tgt));
3238 rc = md_revalidate_lock(tgt->ltd_exp, it, fid, bits);
3242 int lmv_get_fid_from_lsm(struct obd_export *exp,
3243 const struct lmv_stripe_md *lsm,
3244 const char *name, int namelen, struct lu_fid *fid)
3246 const struct lmv_oinfo *oinfo;
3248 LASSERT(lsm != NULL);
3249 oinfo = lsm_name_to_stripe_info(lsm, name, namelen);
3251 return PTR_ERR(oinfo);
3253 *fid = oinfo->lmo_fid;
3259 * For lmv, only need to send request to master MDT, and the master MDT will
3260 * process with other slave MDTs. The only exception is Q_GETOQUOTA for which
3261 * we directly fetch data from the slave MDTs.
3263 int lmv_quotactl(struct obd_device *unused, struct obd_export *exp,
3264 struct obd_quotactl *oqctl)
3266 struct obd_device *obd = class_exp2obd(exp);
3267 struct lmv_obd *lmv = &obd->u.lmv;
3268 struct lmv_tgt_desc *tgt = lmv->tgts[0];
3271 __u64 curspace, curinodes;
3275 tgt->ltd_exp == NULL ||
3277 lmv->desc.ld_tgt_count == 0) {
3278 CERROR("master lmv inactive\n");
3282 if (oqctl->qc_cmd != Q_GETOQUOTA) {
3283 rc = obd_quotactl(tgt->ltd_exp, oqctl);
3287 curspace = curinodes = 0;
3288 for (i = 0; i < lmv->desc.ld_tgt_count; i++) {
3292 if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active)
3295 err = obd_quotactl(tgt->ltd_exp, oqctl);
3297 CERROR("getquota on mdt %d failed. %d\n", i, err);
3301 curspace += oqctl->qc_dqblk.dqb_curspace;
3302 curinodes += oqctl->qc_dqblk.dqb_curinodes;
3305 oqctl->qc_dqblk.dqb_curspace = curspace;
3306 oqctl->qc_dqblk.dqb_curinodes = curinodes;
3311 static int lmv_merge_attr(struct obd_export *exp,
3312 const struct lmv_stripe_md *lsm,
3313 struct cl_attr *attr,
3314 ldlm_blocking_callback cb_blocking)
3319 rc = lmv_revalidate_slaves(exp, lsm, cb_blocking, 0);
3323 for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
3324 struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root;
3326 CDEBUG(D_INFO, ""DFID" size %llu, blocks %llu nlink %u,"
3327 " atime %lu ctime %lu, mtime %lu.\n",
3328 PFID(&lsm->lsm_md_oinfo[i].lmo_fid),
3329 i_size_read(inode), (unsigned long long)inode->i_blocks,
3330 inode->i_nlink, LTIME_S(inode->i_atime),
3331 LTIME_S(inode->i_ctime), LTIME_S(inode->i_mtime));
3333 /* for slave stripe, it needs to subtract nlink for . and .. */
3335 attr->cat_nlink += inode->i_nlink - 2;
3337 attr->cat_nlink = inode->i_nlink;
3339 attr->cat_size += i_size_read(inode);
3340 attr->cat_blocks += inode->i_blocks;
3342 if (attr->cat_atime < LTIME_S(inode->i_atime))
3343 attr->cat_atime = LTIME_S(inode->i_atime);
3345 if (attr->cat_ctime < LTIME_S(inode->i_ctime))
3346 attr->cat_ctime = LTIME_S(inode->i_ctime);
3348 if (attr->cat_mtime < LTIME_S(inode->i_mtime))
3349 attr->cat_mtime = LTIME_S(inode->i_mtime);
3354 struct obd_ops lmv_obd_ops = {
3355 .o_owner = THIS_MODULE,
3356 .o_setup = lmv_setup,
3357 .o_cleanup = lmv_cleanup,
3358 .o_precleanup = lmv_precleanup,
3359 .o_process_config = lmv_process_config,
3360 .o_connect = lmv_connect,
3361 .o_disconnect = lmv_disconnect,
3362 .o_statfs = lmv_statfs,
3363 .o_get_info = lmv_get_info,
3364 .o_set_info_async = lmv_set_info_async,
3365 .o_notify = lmv_notify,
3366 .o_get_uuid = lmv_get_uuid,
3367 .o_iocontrol = lmv_iocontrol,
3368 .o_quotactl = lmv_quotactl
3371 struct md_ops lmv_md_ops = {
3372 .m_getstatus = lmv_getstatus,
3373 .m_null_inode = lmv_null_inode,
3374 .m_find_cbdata = lmv_find_cbdata,
3375 .m_close = lmv_close,
3376 .m_create = lmv_create,
3377 .m_enqueue = lmv_enqueue,
3378 .m_getattr = lmv_getattr,
3379 .m_getxattr = lmv_getxattr,
3380 .m_getattr_name = lmv_getattr_name,
3381 .m_intent_lock = lmv_intent_lock,
3383 .m_rename = lmv_rename,
3384 .m_setattr = lmv_setattr,
3385 .m_setxattr = lmv_setxattr,
3386 .m_fsync = lmv_fsync,
3387 .m_read_page = lmv_read_page,
3388 .m_unlink = lmv_unlink,
3389 .m_init_ea_size = lmv_init_ea_size,
3390 .m_cancel_unused = lmv_cancel_unused,
3391 .m_set_lock_data = lmv_set_lock_data,
3392 .m_lock_match = lmv_lock_match,
3393 .m_get_lustre_md = lmv_get_lustre_md,
3394 .m_free_lustre_md = lmv_free_lustre_md,
3395 .m_merge_attr = lmv_merge_attr,
3396 .m_set_open_replay_data = lmv_set_open_replay_data,
3397 .m_clear_open_replay_data = lmv_clear_open_replay_data,
3398 .m_get_remote_perm = lmv_get_remote_perm,
3399 .m_intent_getattr_async = lmv_intent_getattr_async,
3400 .m_revalidate_lock = lmv_revalidate_lock,
3401 .m_get_fid_from_lsm = lmv_get_fid_from_lsm,
3402 .m_unpackmd = lmv_unpackmd,
3405 int __init lmv_init(void)
3407 return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL,
3408 LUSTRE_LMV_NAME, NULL);
3411 static void lmv_exit(void)
3413 class_unregister_type(LUSTRE_LMV_NAME);
3416 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
3417 MODULE_DESCRIPTION("Lustre Logical Metadata Volume OBD driver");
3418 MODULE_LICENSE("GPL");
3420 module_init(lmv_init);
3421 module_exit(lmv_exit);