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) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lov/lov_obd.c
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_LOV
46 #include <libcfs/libcfs.h>
48 #include <liblustre.h>
51 #include <obd_support.h>
52 #include <lustre_lib.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <lustre_dlm.h>
56 #include <lustre_mds.h>
57 #include <lustre_debug.h>
58 #include <obd_class.h>
61 #include <lprocfs_status.h>
62 #include <lustre_param.h>
63 #include <cl_object.h>
65 #include <lustre/ll_fiemap.h>
66 #include <lustre_fid.h>
68 #include "lov_internal.h"
70 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
71 Any function that expects lov_tgts to remain stationary must take a ref. */
72 static void lov_getref(struct obd_device *obd)
74 struct lov_obd *lov = &obd->u.lov;
76 /* nobody gets through here until lov_putref is done */
77 mutex_lock(&lov->lov_lock);
78 cfs_atomic_inc(&lov->lov_refcount);
79 mutex_unlock(&lov->lov_lock);
83 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
85 static void lov_putref(struct obd_device *obd)
87 struct lov_obd *lov = &obd->u.lov;
89 mutex_lock(&lov->lov_lock);
90 /* ok to dec to 0 more than once -- ltd_exp's will be null */
91 if (cfs_atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
94 struct lov_tgt_desc *tgt, *n;
95 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
97 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
98 tgt = lov->lov_tgts[i];
100 if (!tgt || !tgt->ltd_reap)
102 cfs_list_add(&tgt->ltd_kill, &kill);
103 /* XXX - right now there is a dependency on ld_tgt_count
104 * being the maximum tgt index for computing the
105 * mds_max_easize. So we can't shrink it. */
106 lov_ost_pool_remove(&lov->lov_packed, i);
107 lov->lov_tgts[i] = NULL;
108 lov->lov_death_row--;
110 mutex_unlock(&lov->lov_lock);
112 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
113 cfs_list_del(&tgt->ltd_kill);
115 __lov_del_obd(obd, tgt);
118 mutex_unlock(&lov->lov_lock);
122 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
123 enum obd_notify_event ev);
124 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
125 enum obd_notify_event ev, void *data);
127 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
128 struct obd_connect_data *data)
130 struct lov_obd *lov = &obd->u.lov;
131 struct obd_uuid *tgt_uuid;
132 struct obd_device *tgt_obd;
133 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
134 struct obd_import *imp;
136 struct proc_dir_entry *lov_proc_dir;
141 if (!lov->lov_tgts[index])
144 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
145 tgt_obd = lov->lov_tgts[index]->ltd_obd;
147 if (!tgt_obd->obd_set_up) {
148 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
152 /* override the sp_me from lov */
153 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
155 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
156 data->ocd_index = index;
159 * Divine LOV knows that OBDs under it are OSCs.
161 imp = tgt_obd->u.cli.cl_import;
164 tgt_obd->obd_no_recov = 0;
165 /* FIXME this is probably supposed to be
166 ptlrpc_set_import_active. Horrible naming. */
167 ptlrpc_activate_import(imp);
170 rc = obd_register_observer(tgt_obd, obd);
172 CERROR("Target %s register_observer error %d\n",
173 obd_uuid2str(tgt_uuid), rc);
178 if (imp->imp_invalid) {
179 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
180 "disabled\n", obd_uuid2str(tgt_uuid));
184 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
185 &lov_osc_uuid, data, NULL);
186 if (rc || !lov->lov_tgts[index]->ltd_exp) {
187 CERROR("Target %s connect error %d\n",
188 obd_uuid2str(tgt_uuid), rc);
192 lov->lov_tgts[index]->ltd_reap = 0;
194 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
195 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
198 lov_proc_dir = obd->obd_proc_private;
200 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
201 struct proc_dir_entry *osc_symlink;
203 LASSERT(osc_obd != NULL);
204 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
205 LASSERT(osc_obd->obd_type->typ_name != NULL);
207 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
210 osc_obd->obd_type->typ_name,
212 if (osc_symlink == NULL) {
213 CERROR("could not register LOV target "
214 "/proc/fs/lustre/%s/%s/target_obds/%s.",
215 obd->obd_type->typ_name, obd->obd_name,
217 lprocfs_remove(&lov_proc_dir);
218 obd->obd_proc_private = NULL;
226 static int lov_connect(const struct lu_env *env,
227 struct obd_export **exp, struct obd_device *obd,
228 struct obd_uuid *cluuid, struct obd_connect_data *data,
231 struct lov_obd *lov = &obd->u.lov;
232 struct lov_tgt_desc *tgt;
233 struct lustre_handle conn;
237 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
239 rc = class_connect(&conn, obd, cluuid);
243 *exp = class_conn2export(&conn);
245 /* Why should there ever be more than 1 connect? */
247 LASSERT(lov->lov_connects == 1);
249 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
251 lov->lov_ocd = *data;
254 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
255 tgt = lov->lov_tgts[i];
256 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
258 /* Flags will be lowest common denominator */
259 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
261 CERROR("%s: lov connect tgt %d failed: %d\n",
262 obd->obd_name, i, rc);
265 /* connect to administrative disabled ost */
266 if (!lov->lov_tgts[i]->ltd_exp)
269 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
270 OBD_NOTIFY_CONNECT, (void *)&i);
272 CERROR("%s error sending notify %d\n",
281 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
283 struct proc_dir_entry *lov_proc_dir;
284 struct lov_obd *lov = &obd->u.lov;
285 struct obd_device *osc_obd;
289 osc_obd = class_exp2obd(tgt->ltd_exp);
290 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
291 obd->obd_name, osc_obd->obd_name);
293 if (tgt->ltd_active) {
295 lov->desc.ld_active_tgt_count--;
296 tgt->ltd_exp->exp_obd->obd_inactive = 1;
299 lov_proc_dir = obd->obd_proc_private;
301 lprocfs_remove_proc_entry(osc_obd->obd_name, lov_proc_dir);
304 /* Pass it on to our clients.
305 * XXX This should be an argument to disconnect,
306 * XXX not a back-door flag on the OBD. Ah well.
308 osc_obd->obd_force = obd->obd_force;
309 osc_obd->obd_fail = obd->obd_fail;
310 osc_obd->obd_no_recov = obd->obd_no_recov;
313 obd_register_observer(osc_obd, NULL);
315 rc = obd_disconnect(tgt->ltd_exp);
317 CERROR("Target %s disconnect error %d\n",
318 tgt->ltd_uuid.uuid, rc);
326 static int lov_disconnect(struct obd_export *exp)
328 struct obd_device *obd = class_exp2obd(exp);
329 struct lov_obd *lov = &obd->u.lov;
336 /* Only disconnect the underlying layers on the final disconnect. */
338 if (lov->lov_connects != 0) {
339 /* why should there be more than 1 connect? */
340 CERROR("disconnect #%d\n", lov->lov_connects);
344 /* Let's hold another reference so lov_del_obd doesn't spin through
348 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
349 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
350 /* Disconnection is the last we know about an obd */
351 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
357 rc = class_disconnect(exp); /* bz 9811 */
363 * -EINVAL : UUID can't be found in the LOV's target list
364 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
365 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
366 * any >= 0 : is log target index
368 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
369 enum obd_notify_event ev)
371 struct lov_obd *lov = &obd->u.lov;
372 struct lov_tgt_desc *tgt;
373 int index, activate, active;
376 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
377 lov, uuid->uuid, ev);
380 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
381 tgt = lov->lov_tgts[index];
385 * LU-642, initially inactive OSC could miss the obd_connect,
386 * we make up for it here.
388 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
389 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
390 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
392 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
393 &lov_osc_uuid, &lov->lov_ocd, NULL);
398 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
399 index, obd_uuid2str(&tgt->ltd_uuid),
400 tgt->ltd_exp->exp_handle.h_cookie);
401 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
405 if (index == lov->desc.ld_tgt_count)
406 GOTO(out, index = -EINVAL);
408 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
409 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
411 if (lov->lov_tgts[index]->ltd_activate == activate) {
412 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
413 uuid->uuid, activate ? "" : "de");
415 lov->lov_tgts[index]->ltd_activate = activate;
416 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
417 activate ? "" : "de", obd_uuid2str(uuid));
420 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
421 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
423 if (lov->lov_tgts[index]->ltd_active == active) {
424 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
425 uuid->uuid, active ? "" : "in");
428 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
429 obd_uuid2str(uuid), active ? "" : "in");
432 lov->lov_tgts[index]->ltd_active = active;
434 lov->desc.ld_active_tgt_count++;
435 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
437 lov->desc.ld_active_tgt_count--;
438 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
441 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
449 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
450 enum obd_notify_event ev, void *data)
453 struct lov_obd *lov = &obd->u.lov;
456 down_read(&lov->lov_notify_lock);
457 if (!lov->lov_connects) {
458 up_read(&lov->lov_notify_lock);
462 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
463 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
464 struct obd_uuid *uuid;
468 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
469 up_read(&lov->lov_notify_lock);
470 CERROR("unexpected notification of %s %s!\n",
471 watched->obd_type->typ_name,
475 uuid = &watched->u.cli.cl_target_uuid;
477 /* Set OSC as active before notifying the observer, so the
478 * observer can use the OSC normally.
480 rc = lov_set_osc_active(obd, uuid, ev);
482 up_read(&lov->lov_notify_lock);
483 CERROR("event(%d) of %s failed: %d\n", ev,
484 obd_uuid2str(uuid), rc);
487 /* active event should be pass lov target index as data */
491 /* Pass the notification up the chain. */
493 rc = obd_notify_observer(obd, watched, ev, data);
495 /* NULL watched means all osc's in the lov (only for syncs) */
496 /* sync event should be send lov idx as data */
497 struct lov_obd *lov = &obd->u.lov;
501 is_sync = (ev == OBD_NOTIFY_SYNC) ||
502 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
505 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
506 if (!lov->lov_tgts[i])
509 /* don't send sync event if target not
510 * connected/activated */
511 if (is_sync && !lov->lov_tgts[i]->ltd_active)
514 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
517 CERROR("%s: notify %s of %s failed %d\n",
519 obd->obd_observer->obd_name,
520 lov->lov_tgts[i]->ltd_obd->obd_name,
527 up_read(&lov->lov_notify_lock);
531 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
532 __u32 index, int gen, int active)
534 struct lov_obd *lov = &obd->u.lov;
535 struct lov_tgt_desc *tgt;
536 struct obd_device *tgt_obd;
540 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
541 uuidp->uuid, index, gen, active);
544 CERROR("request to add OBD %s with invalid generation: %d\n",
549 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
554 mutex_lock(&lov->lov_lock);
556 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
557 tgt = lov->lov_tgts[index];
558 CERROR("UUID %s already assigned at LOV target index %d\n",
559 obd_uuid2str(&tgt->ltd_uuid), index);
560 mutex_unlock(&lov->lov_lock);
564 if (index >= lov->lov_tgt_size) {
565 /* We need to reallocate the lov target array. */
566 struct lov_tgt_desc **newtgts, **old = NULL;
567 __u32 newsize, oldsize = 0;
569 newsize = max(lov->lov_tgt_size, (__u32)2);
570 while (newsize < index + 1)
571 newsize = newsize << 1;
572 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
573 if (newtgts == NULL) {
574 mutex_unlock(&lov->lov_lock);
578 if (lov->lov_tgt_size) {
579 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
582 oldsize = lov->lov_tgt_size;
585 lov->lov_tgts = newtgts;
586 lov->lov_tgt_size = newsize;
589 OBD_FREE(old, sizeof(*old) * oldsize);
591 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
592 lov->lov_tgts, lov->lov_tgt_size);
597 mutex_unlock(&lov->lov_lock);
601 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
603 mutex_unlock(&lov->lov_lock);
608 tgt->ltd_uuid = *uuidp;
609 tgt->ltd_obd = tgt_obd;
610 /* XXX - add a sanity check on the generation number. */
612 tgt->ltd_index = index;
613 tgt->ltd_activate = active;
614 lov->lov_tgts[index] = tgt;
615 if (index >= lov->desc.ld_tgt_count)
616 lov->desc.ld_tgt_count = index + 1;
618 mutex_unlock(&lov->lov_lock);
620 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
621 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
623 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
625 if (lov->lov_connects == 0) {
626 /* lov_connect hasn't been called yet. We'll do the
627 lov_connect_obd on this target when that fn first runs,
628 because we don't know the connect flags yet. */
634 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
638 /* connect to administrative disabled ost */
642 if (lov->lov_cache != NULL) {
643 rc = obd_set_info_async(NULL, tgt->ltd_exp,
644 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
645 sizeof(struct cl_client_cache), lov->lov_cache,
651 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
652 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
657 CERROR("add failed (%d), deleting %s\n", rc,
658 obd_uuid2str(&tgt->ltd_uuid));
659 lov_del_target(obd, index, 0, 0);
665 /* Schedule a target for deletion */
666 int lov_del_target(struct obd_device *obd, __u32 index,
667 struct obd_uuid *uuidp, int gen)
669 struct lov_obd *lov = &obd->u.lov;
670 int count = lov->desc.ld_tgt_count;
674 if (index >= count) {
675 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
680 /* to make sure there's no ongoing lov_notify() now */
681 down_write(&lov->lov_notify_lock);
684 if (!lov->lov_tgts[index]) {
685 CERROR("LOV target at index %d is not setup.\n", index);
686 GOTO(out, rc = -EINVAL);
689 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
690 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
691 lov_uuid2str(lov, index), index,
692 obd_uuid2str(uuidp));
693 GOTO(out, rc = -EINVAL);
696 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
697 lov_uuid2str(lov, index), index,
698 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
699 lov->lov_tgts[index]->ltd_active);
701 lov->lov_tgts[index]->ltd_reap = 1;
702 lov->lov_death_row++;
703 /* we really delete it from obd_putref */
706 up_write(&lov->lov_notify_lock);
711 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
713 struct obd_device *osc_obd;
716 LASSERT(tgt->ltd_reap);
718 osc_obd = class_exp2obd(tgt->ltd_exp);
720 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
722 osc_obd ? osc_obd->obd_name : "<no obd>");
725 lov_disconnect_obd(obd, tgt);
729 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
730 do it ourselves. And we can't do it from lov_cleanup,
731 because we just lost our only reference to it. */
733 class_manual_cleanup(osc_obd);
736 void lov_fix_desc_stripe_size(__u64 *val)
738 if (*val < LOV_MIN_STRIPE_SIZE) {
740 LCONSOLE_INFO("Increasing default stripe size to "
742 LOV_DEFAULT_STRIPE_SIZE);
743 *val = LOV_DEFAULT_STRIPE_SIZE;
744 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
745 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
746 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
748 *val, LOV_MIN_STRIPE_SIZE);
752 void lov_fix_desc_stripe_count(__u32 *val)
758 void lov_fix_desc_pattern(__u32 *val)
760 /* from lov_setstripe */
761 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
762 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
767 void lov_fix_desc_qos_maxage(__u32 *val)
771 *val = QOS_DEFAULT_MAXAGE;
774 void lov_fix_desc(struct lov_desc *desc)
776 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
777 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
778 lov_fix_desc_pattern(&desc->ld_pattern);
779 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
782 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
784 struct lov_desc *desc;
785 struct lov_obd *lov = &obd->u.lov;
789 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
790 CERROR("LOV setup requires a descriptor\n");
794 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
796 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
797 CERROR("descriptor size wrong: %d > %d\n",
798 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
802 if (desc->ld_magic != LOV_DESC_MAGIC) {
803 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
804 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
805 obd->obd_name, desc);
806 lustre_swab_lov_desc(desc);
808 CERROR("%s: Bad lov desc magic: %#x\n",
809 obd->obd_name, desc->ld_magic);
816 desc->ld_active_tgt_count = 0;
818 lov->lov_tgt_size = 0;
820 mutex_init(&lov->lov_lock);
821 cfs_atomic_set(&lov->lov_refcount, 0);
822 lov->lov_sp_me = LUSTRE_SP_CLI;
824 init_rwsem(&lov->lov_notify_lock);
826 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
828 HASH_POOLS_BKT_BITS, 0,
831 &pool_hash_operations,
833 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
834 lov->lov_pool_count = 0;
835 rc = lov_ost_pool_init(&lov->lov_packed, 0);
840 obd->obd_vars = lprocfs_lov_obd_vars;
841 lprocfs_seq_obd_setup(obd);
842 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", 0444,
843 &lov_proc_target_fops, obd);
845 CWARN("Error adding the target_obd file\n");
847 lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
857 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
860 struct lov_obd *lov = &obd->u.lov;
865 case OBD_CLEANUP_EARLY: {
867 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
868 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
870 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
882 static int lov_cleanup(struct obd_device *obd)
884 struct lov_obd *lov = &obd->u.lov;
885 cfs_list_t *pos, *tmp;
886 struct pool_desc *pool;
889 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
890 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
891 /* free pool structs */
892 CDEBUG(D_INFO, "delete pool %p\n", pool);
893 /* In the function below, .hs_keycmp resolves to
894 * pool_hashkey_keycmp() */
895 /* coverity[overrun-buffer-val] */
896 lov_pool_del(obd, pool->pool_name);
898 cfs_hash_putref(lov->lov_pools_hash_body);
899 lov_ost_pool_free(&lov->lov_packed);
901 lprocfs_obd_cleanup(obd);
905 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
906 if (!lov->lov_tgts[i])
909 /* Inactive targets may never have connected */
910 if (lov->lov_tgts[i]->ltd_active ||
911 cfs_atomic_read(&lov->lov_refcount))
912 /* We should never get here - these
913 should have been removed in the
915 CERROR("lov tgt %d not cleaned!"
916 " deathrow=%d, lovrc=%d\n",
917 i, lov->lov_death_row,
918 cfs_atomic_read(&lov->lov_refcount));
919 lov_del_target(obd, i, 0, 0);
922 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
924 lov->lov_tgt_size = 0;
929 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
930 __u32 *indexp, int *genp)
932 struct obd_uuid obd_uuid;
937 switch(cmd = lcfg->lcfg_command) {
938 case LCFG_LOV_ADD_OBD:
939 case LCFG_LOV_ADD_INA:
940 case LCFG_LOV_DEL_OBD: {
943 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
944 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
945 GOTO(out, rc = -EINVAL);
947 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
949 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
950 GOTO(out, rc = -EINVAL);
951 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
952 GOTO(out, rc = -EINVAL);
955 if (cmd == LCFG_LOV_ADD_OBD)
956 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
957 else if (cmd == LCFG_LOV_ADD_INA)
958 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
960 rc = lov_del_target(obd, index, &obd_uuid, gen);
964 struct lov_desc *desc = &(obd->u.lov.desc);
967 GOTO(out, rc = -EINVAL);
969 rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
982 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
983 GOTO(out, rc = -EINVAL);
991 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
992 struct lov_stripe_md **ea, struct obd_trans_info *oti)
994 struct lov_stripe_md *obj_mdp, *lsm;
995 struct lov_obd *lov = &exp->exp_obd->u.lov;
1000 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1001 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1003 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1004 if (obj_mdp == NULL)
1007 ost_idx = src_oa->o_nlink;
1010 GOTO(out, rc = -EINVAL);
1011 if (ost_idx >= lov->desc.ld_tgt_count ||
1012 !lov->lov_tgts[ost_idx])
1013 GOTO(out, rc = -EINVAL);
1015 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1016 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1017 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1018 ostid_id(&src_oa->o_oi))
1019 GOTO(out, rc = -EINVAL);
1023 if (i == lsm->lsm_stripe_count)
1024 GOTO(out, rc = -EINVAL);
1026 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1027 src_oa, &obj_mdp, oti);
1029 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1033 /* the LOV expects oa->o_id to be set to the LOV object id */
1034 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1035 struct obdo *src_oa, struct lov_stripe_md **ea,
1036 struct obd_trans_info *oti)
1038 struct lov_obd *lov;
1042 LASSERT(ea != NULL);
1046 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1047 src_oa->o_flags == OBD_FL_DELORPHAN) {
1048 /* should be used with LOV anymore */
1052 lov = &exp->exp_obd->u.lov;
1053 if (!lov->desc.ld_active_tgt_count)
1056 obd_getref(exp->exp_obd);
1057 /* Recreate a specific object id at the given OST index */
1058 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1059 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1060 rc = lov_recreate(exp, src_oa, ea, oti);
1063 obd_putref(exp->exp_obd);
1067 #define ASSERT_LSM_MAGIC(lsmp) \
1069 LASSERT((lsmp) != NULL); \
1070 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1071 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1072 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1075 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1076 struct obdo *oa, struct lov_stripe_md *lsm,
1077 struct obd_trans_info *oti, struct obd_export *md_exp,
1080 struct lov_request_set *set;
1081 struct obd_info oinfo;
1082 struct lov_request *req;
1084 struct lov_obd *lov;
1085 int rc = 0, err = 0;
1088 ASSERT_LSM_MAGIC(lsm);
1090 if (!exp || !exp->exp_obd)
1093 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1095 LASSERT(oti->oti_logcookies);
1098 lov = &exp->exp_obd->u.lov;
1099 obd_getref(exp->exp_obd);
1100 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1104 cfs_list_for_each (pos, &set->set_list) {
1105 req = cfs_list_entry(pos, struct lov_request, rq_link);
1107 if (oa->o_valid & OBD_MD_FLCOOKIE)
1108 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1110 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1111 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1112 err = lov_update_common_set(set, req, err);
1114 CERROR("%s: destroying objid "DOSTID" subobj "
1115 DOSTID" on OST idx %d: rc = %d\n",
1116 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1117 POSTID(&req->rq_oi.oi_oa->o_oi),
1125 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1126 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1128 err = lov_fini_destroy_set(set);
1130 obd_putref(exp->exp_obd);
1131 RETURN(rc ? rc : err);
1134 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1137 struct lov_request_set *lovset = (struct lov_request_set *)data;
1141 /* don't do attribute merge if this aysnc op failed */
1143 cfs_atomic_set(&lovset->set_completes, 0);
1144 err = lov_fini_getattr_set(lovset);
1145 RETURN(rc ? rc : err);
1148 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1149 struct ptlrpc_request_set *rqset)
1151 struct lov_request_set *lovset;
1152 struct lov_obd *lov;
1154 struct lov_request *req;
1159 ASSERT_LSM_MAGIC(oinfo->oi_md);
1161 if (!exp || !exp->exp_obd)
1164 lov = &exp->exp_obd->u.lov;
1166 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1170 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1171 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1172 oinfo->oi_md->lsm_stripe_size);
1174 cfs_list_for_each(pos, &lovset->set_list) {
1175 req = cfs_list_entry(pos, struct lov_request, rq_link);
1177 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1178 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1179 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1180 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1181 &req->rq_oi, rqset);
1183 CERROR("%s: getattr objid "DOSTID" subobj"
1184 DOSTID" on OST idx %d: rc = %d\n",
1185 exp->exp_obd->obd_name,
1186 POSTID(&oinfo->oi_oa->o_oi),
1187 POSTID(&req->rq_oi.oi_oa->o_oi),
1193 if (!cfs_list_empty(&rqset->set_requests)) {
1195 LASSERT (rqset->set_interpret == NULL);
1196 rqset->set_interpret = lov_getattr_interpret;
1197 rqset->set_arg = (void *)lovset;
1202 cfs_atomic_set(&lovset->set_completes, 0);
1203 err = lov_fini_getattr_set(lovset);
1204 RETURN(rc ? rc : err);
1207 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1210 struct lov_request_set *lovset = (struct lov_request_set *)data;
1215 cfs_atomic_set(&lovset->set_completes, 0);
1216 err = lov_fini_setattr_set(lovset);
1217 RETURN(rc ? rc : err);
1220 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1221 needed. Otherwise, a client is waiting for responses. */
1222 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1223 struct obd_trans_info *oti,
1224 struct ptlrpc_request_set *rqset)
1226 struct lov_request_set *set;
1227 struct lov_request *req;
1229 struct lov_obd *lov;
1234 ASSERT_LSM_MAGIC(oinfo->oi_md);
1235 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1237 LASSERT(oti->oti_logcookies);
1240 if (!exp || !exp->exp_obd)
1243 lov = &exp->exp_obd->u.lov;
1244 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1248 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1249 POSTID(&oinfo->oi_md->lsm_oi),
1250 oinfo->oi_md->lsm_stripe_count,
1251 oinfo->oi_md->lsm_stripe_size);
1253 cfs_list_for_each(pos, &set->set_list) {
1254 req = cfs_list_entry(pos, struct lov_request, rq_link);
1256 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1257 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1259 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1260 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1261 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1263 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1264 &req->rq_oi, oti, rqset);
1266 CERROR("error: setattr objid "DOSTID" subobj"
1267 DOSTID" on OST idx %d: rc = %d\n",
1268 POSTID(&set->set_oi->oi_oa->o_oi),
1269 POSTID(&req->rq_oi.oi_oa->o_oi),
1275 /* If we are not waiting for responses on async requests, return. */
1276 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1279 cfs_atomic_set(&set->set_completes, 0);
1280 err = lov_fini_setattr_set(set);
1281 RETURN(rc ? rc : err);
1284 LASSERT(rqset->set_interpret == NULL);
1285 rqset->set_interpret = lov_setattr_interpret;
1286 rqset->set_arg = (void *)set;
1291 static int lov_change_cbdata(struct obd_export *exp,
1292 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1295 struct lov_obd *lov;
1299 ASSERT_LSM_MAGIC(lsm);
1301 if (!exp || !exp->exp_obd)
1304 lov = &exp->exp_obd->u.lov;
1305 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1306 struct lov_stripe_md submd;
1307 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1309 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1310 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1314 submd.lsm_oi = loi->loi_oi;
1315 submd.lsm_stripe_count = 0;
1316 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1322 /* find any ldlm lock of the inode in lov
1326 static int lov_find_cbdata(struct obd_export *exp,
1327 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1330 struct lov_obd *lov;
1334 ASSERT_LSM_MAGIC(lsm);
1336 if (!exp || !exp->exp_obd)
1339 lov = &exp->exp_obd->u.lov;
1340 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1341 struct lov_stripe_md submd;
1342 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1344 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1345 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1348 submd.lsm_oi = loi->loi_oi;
1349 submd.lsm_stripe_count = 0;
1350 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1358 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1360 struct lov_request_set *lovset = (struct lov_request_set *)data;
1365 cfs_atomic_set(&lovset->set_completes, 0);
1367 err = lov_fini_statfs_set(lovset);
1368 RETURN(rc ? rc : err);
1371 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1372 __u64 max_age, struct ptlrpc_request_set *rqset)
1374 struct obd_device *obd = class_exp2obd(exp);
1375 struct lov_request_set *set;
1376 struct lov_request *req;
1378 struct lov_obd *lov;
1382 LASSERT(oinfo != NULL);
1383 LASSERT(oinfo->oi_osfs != NULL);
1386 rc = lov_prep_statfs_set(obd, oinfo, &set);
1390 cfs_list_for_each (pos, &set->set_list) {
1391 req = cfs_list_entry(pos, struct lov_request, rq_link);
1392 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1393 &req->rq_oi, max_age, rqset);
1398 if (rc || cfs_list_empty(&rqset->set_requests)) {
1401 cfs_atomic_set(&set->set_completes, 0);
1402 err = lov_fini_statfs_set(set);
1403 RETURN(rc ? rc : err);
1406 LASSERT(rqset->set_interpret == NULL);
1407 rqset->set_interpret = lov_statfs_interpret;
1408 rqset->set_arg = (void *)set;
1412 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1413 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1415 struct ptlrpc_request_set *set = NULL;
1416 struct obd_info oinfo = { { { 0 } } };
1421 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1422 * statfs requests */
1423 set = ptlrpc_prep_set();
1427 oinfo.oi_osfs = osfs;
1428 oinfo.oi_flags = flags;
1429 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1431 rc = ptlrpc_set_wait(set);
1432 ptlrpc_set_destroy(set);
1437 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1438 void *karg, void *uarg)
1440 struct obd_device *obddev = class_exp2obd(exp);
1441 struct lov_obd *lov = &obddev->u.lov;
1442 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1443 struct obd_uuid *uuidp;
1447 case IOC_OBD_STATFS: {
1448 struct obd_ioctl_data *data = karg;
1449 struct obd_device *osc_obd;
1450 struct obd_statfs stat_buf = {0};
1454 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1455 if ((index >= count))
1458 if (!lov->lov_tgts[index])
1459 /* Try again with the next index */
1461 if (!lov->lov_tgts[index]->ltd_active)
1464 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1469 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1470 min((int)data->ioc_plen2,
1471 (int)sizeof(struct obd_uuid))))
1474 flags = uarg ? *(__u32*)uarg : 0;
1475 /* got statfs data */
1476 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1477 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1481 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1482 min((int) data->ioc_plen1,
1483 (int) sizeof(stat_buf))))
1487 case OBD_IOC_LOV_GET_CONFIG: {
1488 struct obd_ioctl_data *data;
1489 struct lov_desc *desc;
1494 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1497 data = (struct obd_ioctl_data *)buf;
1499 if (sizeof(*desc) > data->ioc_inllen1) {
1500 obd_ioctl_freedata(buf, len);
1504 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1505 obd_ioctl_freedata(buf, len);
1509 if (sizeof(__u32) * count > data->ioc_inllen3) {
1510 obd_ioctl_freedata(buf, len);
1514 desc = (struct lov_desc *)data->ioc_inlbuf1;
1515 memcpy(desc, &(lov->desc), sizeof(*desc));
1517 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1518 genp = (__u32 *)data->ioc_inlbuf3;
1519 /* the uuid will be empty for deleted OSTs */
1520 for (i = 0; i < count; i++, uuidp++, genp++) {
1521 if (!lov->lov_tgts[i])
1523 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1524 *genp = lov->lov_tgts[i]->ltd_gen;
1527 if (copy_to_user((void *)uarg, buf, len))
1529 obd_ioctl_freedata(buf, len);
1532 case LL_IOC_LOV_GETSTRIPE:
1533 rc = lov_getstripe(exp, karg, uarg);
1535 case OBD_IOC_QUOTACTL: {
1536 struct if_quotactl *qctl = karg;
1537 struct lov_tgt_desc *tgt = NULL;
1538 struct obd_quotactl *oqctl;
1540 if (qctl->qc_valid == QC_OSTIDX) {
1541 if (count <= qctl->qc_idx)
1544 tgt = lov->lov_tgts[qctl->qc_idx];
1545 if (!tgt || !tgt->ltd_exp)
1547 } else if (qctl->qc_valid == QC_UUID) {
1548 for (i = 0; i < count; i++) {
1549 tgt = lov->lov_tgts[i];
1551 !obd_uuid_equals(&tgt->ltd_uuid,
1555 if (tgt->ltd_exp == NULL)
1567 LASSERT(tgt && tgt->ltd_exp);
1568 OBD_ALLOC_PTR(oqctl);
1572 QCTL_COPY(oqctl, qctl);
1573 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1575 QCTL_COPY(qctl, oqctl);
1576 qctl->qc_valid = QC_OSTIDX;
1577 qctl->obd_uuid = tgt->ltd_uuid;
1579 OBD_FREE_PTR(oqctl);
1588 for (i = 0; i < count; i++) {
1590 struct obd_device *osc_obd;
1592 /* OST was disconnected */
1593 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1596 /* ll_umount_begin() sets force flag but for lov, not
1597 * osc. Let's pass it through */
1598 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1599 osc_obd->obd_force = obddev->obd_force;
1600 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1602 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1605 if (lov->lov_tgts[i]->ltd_active) {
1606 CDEBUG(err == -ENOTTY ?
1607 D_IOCTL : D_WARNING,
1608 "iocontrol OSC %s on OST "
1609 "idx %d cmd %x: err = %d\n",
1610 lov_uuid2str(lov, i),
1627 #define FIEMAP_BUFFER_SIZE 4096
1630 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1631 * call. The local end offset and the device are sent in the first
1632 * fm_extent. This function calculates the stripe number from the index.
1633 * This function returns a stripe_no on which mapping is to be restarted.
1635 * This function returns fm_end_offset which is the in-OST offset at which
1636 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1637 * will re-calculate proper offset in next stripe.
1638 * Note that the first extent is passed to lov_get_info via the value field.
1640 * \param fiemap fiemap request header
1641 * \param lsm striping information for the file
1642 * \param fm_start logical start of mapping
1643 * \param fm_end logical end of mapping
1644 * \param start_stripe starting stripe will be returned in this
1646 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1647 struct lov_stripe_md *lsm, obd_size fm_start,
1648 obd_size fm_end, int *start_stripe)
1650 obd_size local_end = fiemap->fm_extents[0].fe_logical;
1651 obd_off lun_start, lun_end;
1652 obd_size fm_end_offset;
1653 int stripe_no = -1, i;
1655 if (fiemap->fm_extent_count == 0 ||
1656 fiemap->fm_extents[0].fe_logical == 0)
1659 /* Find out stripe_no from ost_index saved in the fe_device */
1660 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1661 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
1662 fiemap->fm_extents[0].fe_device) {
1667 if (stripe_no == -1)
1670 /* If we have finished mapping on previous device, shift logical
1671 * offset to start of next device */
1672 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1673 &lun_start, &lun_end)) != 0 &&
1674 local_end < lun_end) {
1675 fm_end_offset = local_end;
1676 *start_stripe = stripe_no;
1678 /* This is a special value to indicate that caller should
1679 * calculate offset in next stripe. */
1681 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1684 return fm_end_offset;
1688 * We calculate on which OST the mapping will end. If the length of mapping
1689 * is greater than (stripe_size * stripe_count) then the last_stripe will
1690 * will be one just before start_stripe. Else we check if the mapping
1691 * intersects each OST and find last_stripe.
1692 * This function returns the last_stripe and also sets the stripe_count
1693 * over which the mapping is spread
1695 * \param lsm striping information for the file
1696 * \param fm_start logical start of mapping
1697 * \param fm_end logical end of mapping
1698 * \param start_stripe starting stripe of the mapping
1699 * \param stripe_count the number of stripes across which to map is returned
1701 * \retval last_stripe return the last stripe of the mapping
1703 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
1704 obd_size fm_end, int start_stripe,
1708 obd_off obd_start, obd_end;
1711 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1712 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1714 *stripe_count = lsm->lsm_stripe_count;
1716 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1717 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1718 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1719 &obd_start, &obd_end)) == 0)
1723 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1730 * Set fe_device and copy extents from local buffer into main return buffer.
1732 * \param fiemap fiemap request header
1733 * \param lcl_fm_ext array of local fiemap extents to be copied
1734 * \param ost_index OST index to be written into the fm_device field for each
1736 * \param ext_count number of extents to be copied
1737 * \param current_extent where to start copying in main extent array
1739 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1740 struct ll_fiemap_extent *lcl_fm_ext,
1741 int ost_index, unsigned int ext_count,
1747 for (ext = 0; ext < ext_count; ext++) {
1748 lcl_fm_ext[ext].fe_device = ost_index;
1749 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1752 /* Copy fm_extent's from fm_local to return buffer */
1753 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1754 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1758 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1759 * This also handles the restarting of FIEMAP calls in case mapping overflows
1760 * the available number of extents in single call.
1762 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1763 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1765 struct ll_fiemap_info_key *fm_key = key;
1766 struct ll_user_fiemap *fiemap = val;
1767 struct ll_user_fiemap *fm_local = NULL;
1768 struct ll_fiemap_extent *lcl_fm_ext;
1770 unsigned int get_num_extents = 0;
1771 int ost_index = 0, actual_start_stripe, start_stripe;
1772 obd_size fm_start, fm_end, fm_length, fm_end_offset;
1774 int current_extent = 0, rc = 0, i;
1775 int ost_eof = 0; /* EOF for object */
1776 int ost_done = 0; /* done with required mapping for this OST? */
1778 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1779 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1781 if (!lsm_has_objects(lsm)) {
1782 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1783 fm_key->oa.o_size)) {
1784 /* released file, return a minimal FIEMAP if
1785 * request fits in file-size.
1787 fiemap->fm_mapped_extents = 1;
1788 fiemap->fm_extents[0].fe_logical =
1789 fm_key->fiemap.fm_start;
1790 if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1792 fiemap->fm_extents[0].fe_length =
1793 fm_key->fiemap.fm_length;
1795 fiemap->fm_extents[0].fe_length =
1797 fm_key->fiemap.fm_start;
1798 fiemap->fm_extents[0].fe_flags |=
1799 (FIEMAP_EXTENT_UNKNOWN |
1800 FIEMAP_EXTENT_LAST);
1805 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1806 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1808 OBD_ALLOC_LARGE(fm_local, buffer_size);
1809 if (fm_local == NULL)
1810 GOTO(out, rc = -ENOMEM);
1811 lcl_fm_ext = &fm_local->fm_extents[0];
1813 count_local = fiemap_size_to_count(buffer_size);
1815 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1816 fm_start = fiemap->fm_start;
1817 fm_length = fiemap->fm_length;
1818 /* Calculate start stripe, last stripe and length of mapping */
1819 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1820 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1821 fm_start + fm_length - 1);
1822 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1823 if (fm_end > fm_key->oa.o_size)
1824 fm_end = fm_key->oa.o_size;
1826 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1827 actual_start_stripe, &stripe_count);
1829 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1830 fm_end, &start_stripe);
1831 if (fm_end_offset == -EINVAL)
1832 GOTO(out, rc = -EINVAL);
1834 if (fiemap->fm_extent_count == 0) {
1835 get_num_extents = 1;
1839 /* Check each stripe */
1840 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1841 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1842 obd_size req_fm_len; /* Stores length of required mapping */
1843 obd_size len_mapped_single_call;
1844 obd_off lun_start, lun_end, obd_object_end;
1845 unsigned int ext_count;
1847 cur_stripe_wrap = cur_stripe;
1849 /* Find out range of mapping on this stripe */
1850 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1851 &lun_start, &obd_object_end)) == 0)
1854 /* If this is a continuation FIEMAP call and we are on
1855 * starting stripe then lun_start needs to be set to
1857 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1858 lun_start = fm_end_offset;
1860 if (fm_length != ~0ULL) {
1861 /* Handle fm_start + fm_length overflow */
1862 if (fm_start + fm_length < fm_start)
1863 fm_length = ~0ULL - fm_start;
1864 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1870 if (lun_start == lun_end)
1873 req_fm_len = obd_object_end - lun_start;
1874 fm_local->fm_length = 0;
1875 len_mapped_single_call = 0;
1877 /* If the output buffer is very large and the objects have many
1878 * extents we may need to loop on a single OST repeatedly */
1882 if (get_num_extents == 0) {
1883 /* Don't get too many extents. */
1884 if (current_extent + count_local >
1885 fiemap->fm_extent_count)
1886 count_local = fiemap->fm_extent_count -
1890 lun_start += len_mapped_single_call;
1891 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1892 req_fm_len = fm_local->fm_length;
1893 fm_local->fm_extent_count = count_local;
1894 fm_local->fm_mapped_extents = 0;
1895 fm_local->fm_flags = fiemap->fm_flags;
1897 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1898 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1900 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1901 GOTO(out, rc = -EINVAL);
1903 /* If OST is inactive, return extent with UNKNOWN flag */
1904 if (!lov->lov_tgts[ost_index]->ltd_active) {
1905 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1906 fm_local->fm_mapped_extents = 1;
1908 lcl_fm_ext[0].fe_logical = lun_start;
1909 lcl_fm_ext[0].fe_length = obd_object_end -
1911 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1916 fm_local->fm_start = lun_start;
1917 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1918 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1919 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1920 rc = obd_get_info(NULL,
1921 lov->lov_tgts[ost_index]->ltd_exp,
1922 keylen, key, vallen, fm_local, lsm);
1927 ext_count = fm_local->fm_mapped_extents;
1928 if (ext_count == 0) {
1930 /* If last stripe has hole at the end,
1931 * then we need to return */
1932 if (cur_stripe_wrap == last_stripe) {
1933 fiemap->fm_mapped_extents = 0;
1939 /* If we just need num of extents then go to next device */
1940 if (get_num_extents) {
1941 current_extent += ext_count;
1945 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1946 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1948 /* Have we finished mapping on this device? */
1949 if (req_fm_len <= len_mapped_single_call)
1952 /* Clear the EXTENT_LAST flag which can be present on
1954 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1955 lcl_fm_ext[ext_count - 1].fe_flags &=
1956 ~FIEMAP_EXTENT_LAST;
1958 curr_loc = lov_stripe_size(lsm,
1959 lcl_fm_ext[ext_count - 1].fe_logical+
1960 lcl_fm_ext[ext_count - 1].fe_length,
1962 if (curr_loc >= fm_key->oa.o_size)
1965 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1966 ost_index, ext_count,
1969 current_extent += ext_count;
1971 /* Ran out of available extents? */
1972 if (current_extent >= fiemap->fm_extent_count)
1974 } while (ost_done == 0 && ost_eof == 0);
1976 if (cur_stripe_wrap == last_stripe)
1981 /* Indicate that we are returning device offsets unless file just has
1983 if (lsm->lsm_stripe_count > 1)
1984 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1986 if (get_num_extents)
1987 goto skip_last_device_calc;
1989 /* Check if we have reached the last stripe and whether mapping for that
1990 * stripe is done. */
1991 if (cur_stripe_wrap == last_stripe) {
1992 if (ost_done || ost_eof)
1993 fiemap->fm_extents[current_extent - 1].fe_flags |=
1997 skip_last_device_calc:
1998 fiemap->fm_mapped_extents = current_extent;
2002 OBD_FREE_LARGE(fm_local, buffer_size);
2006 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2007 __u32 keylen, void *key, __u32 *vallen, void *val,
2008 struct lov_stripe_md *lsm)
2010 struct obd_device *obddev = class_exp2obd(exp);
2011 struct lov_obd *lov = &obddev->u.lov;
2015 if (!vallen || !val)
2020 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2023 struct ldlm_lock *lock;
2025 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2026 struct lov_oinfo *loi;
2027 __u32 *stripe = val;
2029 if (*vallen < sizeof(*stripe))
2030 GOTO(out, rc = -EFAULT);
2031 *vallen = sizeof(*stripe);
2033 /* XXX This is another one of those bits that will need to
2034 * change if we ever actually support nested LOVs. It uses
2035 * the lock's export to find out which stripe it is. */
2036 /* XXX - it's assumed all the locks for deleted OSTs have
2037 * been cancelled. Also, the export for deleted OSTs will
2038 * be NULL and won't match the lock's export. */
2039 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2040 loi = lsm->lsm_oinfo[i];
2041 if (!lov->lov_tgts[loi->loi_ost_idx])
2043 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2044 data->lock->l_conn_export &&
2045 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2050 LDLM_ERROR(data->lock, "lock on inode without such object");
2051 dump_lsm(D_ERROR, lsm);
2052 GOTO(out, rc = -ENXIO);
2053 } else if (KEY_IS(KEY_LAST_ID)) {
2054 struct obd_id_info *info = val;
2055 __u32 size = sizeof(obd_id);
2056 struct lov_tgt_desc *tgt;
2058 LASSERT(*vallen == sizeof(struct obd_id_info));
2059 tgt = lov->lov_tgts[info->idx];
2061 if (!tgt || !tgt->ltd_active)
2062 GOTO(out, rc = -ESRCH);
2064 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2065 &size, info->data, NULL);
2067 } else if (KEY_IS(KEY_LOVDESC)) {
2068 struct lov_desc *desc_ret = val;
2069 *desc_ret = lov->desc;
2072 } else if (KEY_IS(KEY_FIEMAP)) {
2073 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2075 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2076 struct lov_tgt_desc *tgt;
2077 __u64 ost_idx = *((__u64*)val);
2079 LASSERT(*vallen == sizeof(__u64));
2080 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2081 tgt = lov->lov_tgts[ost_idx];
2083 if (!tgt || !tgt->ltd_exp)
2084 GOTO(out, rc = -ESRCH);
2086 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2088 } else if (KEY_IS(KEY_TGT_COUNT)) {
2089 *((int *)val) = lov->desc.ld_tgt_count;
2100 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2101 obd_count keylen, void *key, obd_count vallen,
2102 void *val, struct ptlrpc_request_set *set)
2104 struct obd_device *obddev = class_exp2obd(exp);
2105 struct lov_obd *lov = &obddev->u.lov;
2108 struct lov_tgt_desc *tgt;
2109 unsigned incr, check_uuid,
2110 do_inactive, no_set;
2111 unsigned next_id = 0, mds_con = 0, capa = 0;
2114 incr = check_uuid = do_inactive = no_set = 0;
2117 set = ptlrpc_prep_set();
2123 count = lov->desc.ld_tgt_count;
2125 if (KEY_IS(KEY_NEXT_ID)) {
2126 count = vallen / sizeof(struct obd_id_info);
2127 vallen = sizeof(obd_id);
2128 incr = sizeof(struct obd_id_info);
2131 } else if (KEY_IS(KEY_CHECKSUM)) {
2133 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2134 /* use defaults: do_inactive = incr = 0; */
2135 } else if (KEY_IS(KEY_MDS_CONN)) {
2137 } else if (KEY_IS(KEY_CAPA_KEY)) {
2139 } else if (KEY_IS(KEY_CACHE_SET)) {
2140 LASSERT(lov->lov_cache == NULL);
2141 lov->lov_cache = val;
2145 for (i = 0; i < count; i++, val = (char *)val + incr) {
2147 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2149 tgt = lov->lov_tgts[i];
2151 /* OST was disconnected */
2152 if (!tgt || !tgt->ltd_exp)
2155 /* OST is inactive and we don't want inactive OSCs */
2156 if (!tgt->ltd_active && !do_inactive)
2160 struct mds_group_info *mgi;
2162 LASSERT(vallen == sizeof(*mgi));
2163 mgi = (struct mds_group_info *)val;
2165 /* Only want a specific OSC */
2166 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2170 err = obd_set_info_async(env, tgt->ltd_exp,
2171 keylen, key, sizeof(int),
2173 } else if (next_id) {
2174 err = obd_set_info_async(env, tgt->ltd_exp,
2175 keylen, key, vallen,
2176 ((struct obd_id_info*)val)->data, set);
2178 struct mds_capa_info *info = (struct mds_capa_info*)val;
2180 LASSERT(vallen == sizeof(*info));
2182 /* Only want a specific OSC */
2184 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2187 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2188 key, sizeof(*info->capa),
2191 /* Only want a specific OSC */
2193 !obd_uuid_equals(val, &tgt->ltd_uuid))
2196 err = obd_set_info_async(env, tgt->ltd_exp,
2197 keylen, key, vallen, val, set);
2206 err = ptlrpc_set_wait(set);
2209 ptlrpc_set_destroy(set);
2214 void lov_stripe_lock(struct lov_stripe_md *md)
2216 LASSERT(md->lsm_lock_owner != current_pid());
2217 spin_lock(&md->lsm_lock);
2218 LASSERT(md->lsm_lock_owner == 0);
2219 md->lsm_lock_owner = current_pid();
2222 void lov_stripe_unlock(struct lov_stripe_md *md)
2224 LASSERT(md->lsm_lock_owner == current_pid());
2225 md->lsm_lock_owner = 0;
2226 spin_unlock(&md->lsm_lock);
2229 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2230 struct obd_quotactl *oqctl)
2232 struct lov_obd *lov = &obd->u.lov;
2233 struct lov_tgt_desc *tgt;
2235 __u64 bhardlimit = 0;
2239 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2240 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2241 oqctl->qc_cmd != Q_GETOQUOTA &&
2242 oqctl->qc_cmd != Q_INITQUOTA &&
2243 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2244 oqctl->qc_cmd != Q_FINVALIDATE) {
2245 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2251 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2254 tgt = lov->lov_tgts[i];
2259 if (!tgt->ltd_active || tgt->ltd_reap) {
2260 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2261 lov->lov_tgts[i]->ltd_activate) {
2263 CERROR("ost %d is inactive\n", i);
2265 CDEBUG(D_HA, "ost %d is inactive\n", i);
2270 err = obd_quotactl(tgt->ltd_exp, oqctl);
2272 if (tgt->ltd_active && !rc)
2277 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2278 curspace += oqctl->qc_dqblk.dqb_curspace;
2279 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2284 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2285 oqctl->qc_dqblk.dqb_curspace = curspace;
2286 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2291 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2292 struct obd_quotactl *oqctl)
2294 struct lov_obd *lov = &obd->u.lov;
2300 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2301 if (!lov->lov_tgts[i])
2304 /* Skip quota check on the administratively disabled OSTs. */
2305 if (!lov->lov_tgts[i]->ltd_activate) {
2306 CWARN("lov idx %d was administratively disabled, "
2307 "skip quotacheck on it.\n", i);
2311 if (!lov->lov_tgts[i]->ltd_active) {
2312 CERROR("lov idx %d inactive\n", i);
2318 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2321 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2324 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2335 static struct obd_ops lov_obd_ops = {
2336 .o_owner = THIS_MODULE,
2337 .o_setup = lov_setup,
2338 .o_precleanup = lov_precleanup,
2339 .o_cleanup = lov_cleanup,
2340 .o_connect = lov_connect,
2341 .o_disconnect = lov_disconnect,
2342 .o_statfs = lov_statfs,
2343 .o_statfs_async = lov_statfs_async,
2344 .o_packmd = lov_packmd,
2345 .o_unpackmd = lov_unpackmd,
2346 .o_create = lov_create,
2347 .o_destroy = lov_destroy,
2348 .o_getattr_async = lov_getattr_async,
2349 .o_setattr_async = lov_setattr_async,
2350 .o_adjust_kms = lov_adjust_kms,
2351 .o_change_cbdata = lov_change_cbdata,
2352 .o_find_cbdata = lov_find_cbdata,
2353 .o_iocontrol = lov_iocontrol,
2354 .o_get_info = lov_get_info,
2355 .o_set_info_async = lov_set_info_async,
2356 .o_notify = lov_notify,
2357 .o_pool_new = lov_pool_new,
2358 .o_pool_rem = lov_pool_remove,
2359 .o_pool_add = lov_pool_add,
2360 .o_pool_del = lov_pool_del,
2361 .o_getref = lov_getref,
2362 .o_putref = lov_putref,
2363 .o_quotactl = lov_quotactl,
2364 .o_quotacheck = lov_quotacheck,
2367 struct kmem_cache *lov_oinfo_slab;
2369 extern struct lu_kmem_descr lov_caches[];
2371 int __init lov_init(void)
2376 /* print an address of _any_ initialized kernel symbol from this
2377 * module, to allow debugging with gdb that doesn't support data
2378 * symbols from modules.*/
2379 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2381 rc = lu_kmem_init(lov_caches);
2385 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2386 sizeof(struct lov_oinfo), 0,
2387 SLAB_HWCACHE_ALIGN, NULL);
2388 if (lov_oinfo_slab == NULL) {
2389 lu_kmem_fini(lov_caches);
2393 rc = class_register_type(&lov_obd_ops, NULL, NULL,
2394 #ifndef HAVE_ONLY_PROCFS_SEQ
2397 LUSTRE_LOV_NAME, &lov_device_type);
2400 kmem_cache_destroy(lov_oinfo_slab);
2401 lu_kmem_fini(lov_caches);
2408 static void /*__exit*/ lov_exit(void)
2410 class_unregister_type(LUSTRE_LOV_NAME);
2411 kmem_cache_destroy(lov_oinfo_slab);
2412 lu_kmem_fini(lov_caches);
2415 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2416 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2417 MODULE_LICENSE("GPL");
2419 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);