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 <obd_class.h>
58 #include <lprocfs_status.h>
59 #include <lustre_param.h>
60 #include <cl_object.h>
62 #include <lustre/ll_fiemap.h>
63 #include <lustre_fid.h>
65 #include "lov_internal.h"
67 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
68 Any function that expects lov_tgts to remain stationary must take a ref. */
69 static void lov_getref(struct obd_device *obd)
71 struct lov_obd *lov = &obd->u.lov;
73 /* nobody gets through here until lov_putref is done */
74 mutex_lock(&lov->lov_lock);
75 atomic_inc(&lov->lov_refcount);
76 mutex_unlock(&lov->lov_lock);
80 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
82 static void lov_putref(struct obd_device *obd)
84 struct lov_obd *lov = &obd->u.lov;
86 mutex_lock(&lov->lov_lock);
87 /* ok to dec to 0 more than once -- ltd_exp's will be null */
88 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
89 struct list_head kill = LIST_HEAD_INIT(kill);
90 struct lov_tgt_desc *tgt, *n;
93 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
95 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
96 tgt = lov->lov_tgts[i];
98 if (!tgt || !tgt->ltd_reap)
100 cfs_list_add(&tgt->ltd_kill, &kill);
101 /* XXX - right now there is a dependency on ld_tgt_count
102 * being the maximum tgt index for computing the
103 * mds_max_easize. So we can't shrink it. */
104 lov_ost_pool_remove(&lov->lov_packed, i);
105 lov->lov_tgts[i] = NULL;
106 lov->lov_death_row--;
108 mutex_unlock(&lov->lov_lock);
110 cfs_list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
111 cfs_list_del(&tgt->ltd_kill);
113 __lov_del_obd(obd, tgt);
116 mutex_unlock(&lov->lov_lock);
120 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
121 enum obd_notify_event ev);
122 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
123 enum obd_notify_event ev, void *data);
125 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
126 struct obd_connect_data *data)
128 struct lov_obd *lov = &obd->u.lov;
129 struct obd_uuid *tgt_uuid;
130 struct obd_device *tgt_obd;
131 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
132 struct obd_import *imp;
134 struct proc_dir_entry *lov_proc_dir;
139 if (!lov->lov_tgts[index])
142 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
143 tgt_obd = lov->lov_tgts[index]->ltd_obd;
145 if (!tgt_obd->obd_set_up) {
146 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
150 /* override the sp_me from lov */
151 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
153 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
154 data->ocd_index = index;
157 * Divine LOV knows that OBDs under it are OSCs.
159 imp = tgt_obd->u.cli.cl_import;
162 tgt_obd->obd_no_recov = 0;
163 /* FIXME this is probably supposed to be
164 ptlrpc_set_import_active. Horrible naming. */
165 ptlrpc_activate_import(imp);
168 rc = obd_register_observer(tgt_obd, obd);
170 CERROR("Target %s register_observer error %d\n",
171 obd_uuid2str(tgt_uuid), rc);
176 if (imp->imp_invalid) {
177 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
178 "disabled\n", obd_uuid2str(tgt_uuid));
182 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
183 &lov_osc_uuid, data, NULL);
184 if (rc || !lov->lov_tgts[index]->ltd_exp) {
185 CERROR("Target %s connect error %d\n",
186 obd_uuid2str(tgt_uuid), rc);
190 lov->lov_tgts[index]->ltd_reap = 0;
192 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
193 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
196 lov_proc_dir = obd->obd_proc_private;
198 struct obd_device *osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
199 struct proc_dir_entry *osc_symlink;
201 LASSERT(osc_obd != NULL);
202 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
203 LASSERT(osc_obd->obd_type->typ_name != NULL);
205 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
208 osc_obd->obd_type->typ_name,
210 if (osc_symlink == NULL) {
211 CERROR("could not register LOV target "
212 "/proc/fs/lustre/%s/%s/target_obds/%s.",
213 obd->obd_type->typ_name, obd->obd_name,
215 lprocfs_remove(&lov_proc_dir);
216 obd->obd_proc_private = NULL;
224 static int lov_connect(const struct lu_env *env,
225 struct obd_export **exp, struct obd_device *obd,
226 struct obd_uuid *cluuid, struct obd_connect_data *data,
229 struct lov_obd *lov = &obd->u.lov;
230 struct lov_tgt_desc *tgt;
231 struct lustre_handle conn;
235 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
237 rc = class_connect(&conn, obd, cluuid);
241 *exp = class_conn2export(&conn);
243 /* Why should there ever be more than 1 connect? */
245 LASSERT(lov->lov_connects == 1);
247 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
249 lov->lov_ocd = *data;
252 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
253 tgt = lov->lov_tgts[i];
254 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
256 /* Flags will be lowest common denominator */
257 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
259 CERROR("%s: lov connect tgt %d failed: %d\n",
260 obd->obd_name, i, rc);
263 /* connect to administrative disabled ost */
264 if (!lov->lov_tgts[i]->ltd_exp)
267 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
268 OBD_NOTIFY_CONNECT, (void *)&i);
270 CERROR("%s error sending notify %d\n",
279 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
281 struct proc_dir_entry *lov_proc_dir;
282 struct lov_obd *lov = &obd->u.lov;
283 struct obd_device *osc_obd;
287 osc_obd = class_exp2obd(tgt->ltd_exp);
288 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
289 obd->obd_name, osc_obd->obd_name);
291 if (tgt->ltd_active) {
293 lov->desc.ld_active_tgt_count--;
294 tgt->ltd_exp->exp_obd->obd_inactive = 1;
297 lov_proc_dir = obd->obd_proc_private;
299 lprocfs_remove_proc_entry(osc_obd->obd_name, lov_proc_dir);
302 /* Pass it on to our clients.
303 * XXX This should be an argument to disconnect,
304 * XXX not a back-door flag on the OBD. Ah well.
306 osc_obd->obd_force = obd->obd_force;
307 osc_obd->obd_fail = obd->obd_fail;
308 osc_obd->obd_no_recov = obd->obd_no_recov;
311 obd_register_observer(osc_obd, NULL);
313 rc = obd_disconnect(tgt->ltd_exp);
315 CERROR("Target %s disconnect error %d\n",
316 tgt->ltd_uuid.uuid, rc);
324 static int lov_disconnect(struct obd_export *exp)
326 struct obd_device *obd = class_exp2obd(exp);
327 struct lov_obd *lov = &obd->u.lov;
334 /* Only disconnect the underlying layers on the final disconnect. */
336 if (lov->lov_connects != 0) {
337 /* why should there be more than 1 connect? */
338 CERROR("disconnect #%d\n", lov->lov_connects);
342 /* Let's hold another reference so lov_del_obd doesn't spin through
346 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
347 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
348 /* Disconnection is the last we know about an obd */
349 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
355 rc = class_disconnect(exp); /* bz 9811 */
361 * -EINVAL : UUID can't be found in the LOV's target list
362 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
363 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
364 * any >= 0 : is log target index
366 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
367 enum obd_notify_event ev)
369 struct lov_obd *lov = &obd->u.lov;
370 struct lov_tgt_desc *tgt;
371 int index, activate, active;
374 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
375 lov, uuid->uuid, ev);
378 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
379 tgt = lov->lov_tgts[index];
383 * LU-642, initially inactive OSC could miss the obd_connect,
384 * we make up for it here.
386 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
387 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
388 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
390 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
391 &lov_osc_uuid, &lov->lov_ocd, NULL);
396 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
397 index, obd_uuid2str(&tgt->ltd_uuid),
398 tgt->ltd_exp->exp_handle.h_cookie);
399 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
403 if (index == lov->desc.ld_tgt_count)
404 GOTO(out, index = -EINVAL);
406 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
407 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
409 if (lov->lov_tgts[index]->ltd_activate == activate) {
410 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
411 uuid->uuid, activate ? "" : "de");
413 lov->lov_tgts[index]->ltd_activate = activate;
414 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
415 activate ? "" : "de", obd_uuid2str(uuid));
418 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
419 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
421 if (lov->lov_tgts[index]->ltd_active == active) {
422 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
423 uuid->uuid, active ? "" : "in");
426 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
427 obd_uuid2str(uuid), active ? "" : "in");
430 lov->lov_tgts[index]->ltd_active = active;
432 lov->desc.ld_active_tgt_count++;
433 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
435 lov->desc.ld_active_tgt_count--;
436 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
439 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
447 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
448 enum obd_notify_event ev, void *data)
451 struct lov_obd *lov = &obd->u.lov;
454 down_read(&lov->lov_notify_lock);
455 if (!lov->lov_connects) {
456 up_read(&lov->lov_notify_lock);
460 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
461 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
462 struct obd_uuid *uuid;
466 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
467 up_read(&lov->lov_notify_lock);
468 CERROR("unexpected notification of %s %s!\n",
469 watched->obd_type->typ_name,
473 uuid = &watched->u.cli.cl_target_uuid;
475 /* Set OSC as active before notifying the observer, so the
476 * observer can use the OSC normally.
478 rc = lov_set_osc_active(obd, uuid, ev);
480 up_read(&lov->lov_notify_lock);
481 CERROR("event(%d) of %s failed: %d\n", ev,
482 obd_uuid2str(uuid), rc);
485 /* active event should be pass lov target index as data */
489 /* Pass the notification up the chain. */
491 rc = obd_notify_observer(obd, watched, ev, data);
493 /* NULL watched means all osc's in the lov (only for syncs) */
494 /* sync event should be send lov idx as data */
495 struct lov_obd *lov = &obd->u.lov;
499 is_sync = (ev == OBD_NOTIFY_SYNC) ||
500 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
503 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
504 if (!lov->lov_tgts[i])
507 /* don't send sync event if target not
508 * connected/activated */
509 if (is_sync && !lov->lov_tgts[i]->ltd_active)
512 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
515 CERROR("%s: notify %s of %s failed %d\n",
517 obd->obd_observer->obd_name,
518 lov->lov_tgts[i]->ltd_obd->obd_name,
525 up_read(&lov->lov_notify_lock);
529 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
530 __u32 index, int gen, int active)
532 struct lov_obd *lov = &obd->u.lov;
533 struct lov_tgt_desc *tgt;
534 struct obd_device *tgt_obd;
538 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
539 uuidp->uuid, index, gen, active);
542 CERROR("request to add OBD %s with invalid generation: %d\n",
547 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
552 mutex_lock(&lov->lov_lock);
554 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
555 tgt = lov->lov_tgts[index];
556 CERROR("UUID %s already assigned at LOV target index %d\n",
557 obd_uuid2str(&tgt->ltd_uuid), index);
558 mutex_unlock(&lov->lov_lock);
562 if (index >= lov->lov_tgt_size) {
563 /* We need to reallocate the lov target array. */
564 struct lov_tgt_desc **newtgts, **old = NULL;
565 __u32 newsize, oldsize = 0;
567 newsize = max(lov->lov_tgt_size, (__u32)2);
568 while (newsize < index + 1)
569 newsize = newsize << 1;
570 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
571 if (newtgts == NULL) {
572 mutex_unlock(&lov->lov_lock);
576 if (lov->lov_tgt_size) {
577 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
580 oldsize = lov->lov_tgt_size;
583 lov->lov_tgts = newtgts;
584 lov->lov_tgt_size = newsize;
587 OBD_FREE(old, sizeof(*old) * oldsize);
589 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
590 lov->lov_tgts, lov->lov_tgt_size);
595 mutex_unlock(&lov->lov_lock);
599 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
601 mutex_unlock(&lov->lov_lock);
606 tgt->ltd_uuid = *uuidp;
607 tgt->ltd_obd = tgt_obd;
608 /* XXX - add a sanity check on the generation number. */
610 tgt->ltd_index = index;
611 tgt->ltd_activate = active;
612 lov->lov_tgts[index] = tgt;
613 if (index >= lov->desc.ld_tgt_count)
614 lov->desc.ld_tgt_count = index + 1;
616 mutex_unlock(&lov->lov_lock);
618 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
619 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
621 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
623 if (lov->lov_connects == 0) {
624 /* lov_connect hasn't been called yet. We'll do the
625 lov_connect_obd on this target when that fn first runs,
626 because we don't know the connect flags yet. */
632 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
636 /* connect to administrative disabled ost */
640 if (lov->lov_cache != NULL) {
641 rc = obd_set_info_async(NULL, tgt->ltd_exp,
642 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
643 sizeof(struct cl_client_cache), lov->lov_cache,
649 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
650 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
655 CERROR("add failed (%d), deleting %s\n", rc,
656 obd_uuid2str(&tgt->ltd_uuid));
657 lov_del_target(obd, index, 0, 0);
663 /* Schedule a target for deletion */
664 int lov_del_target(struct obd_device *obd, __u32 index,
665 struct obd_uuid *uuidp, int gen)
667 struct lov_obd *lov = &obd->u.lov;
668 int count = lov->desc.ld_tgt_count;
672 if (index >= count) {
673 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
678 /* to make sure there's no ongoing lov_notify() now */
679 down_write(&lov->lov_notify_lock);
682 if (!lov->lov_tgts[index]) {
683 CERROR("LOV target at index %d is not setup.\n", index);
684 GOTO(out, rc = -EINVAL);
687 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
688 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
689 lov_uuid2str(lov, index), index,
690 obd_uuid2str(uuidp));
691 GOTO(out, rc = -EINVAL);
694 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
695 lov_uuid2str(lov, index), index,
696 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
697 lov->lov_tgts[index]->ltd_active);
699 lov->lov_tgts[index]->ltd_reap = 1;
700 lov->lov_death_row++;
701 /* we really delete it from obd_putref */
704 up_write(&lov->lov_notify_lock);
709 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
711 struct obd_device *osc_obd;
714 LASSERT(tgt->ltd_reap);
716 osc_obd = class_exp2obd(tgt->ltd_exp);
718 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
720 osc_obd ? osc_obd->obd_name : "<no obd>");
723 lov_disconnect_obd(obd, tgt);
727 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
728 do it ourselves. And we can't do it from lov_cleanup,
729 because we just lost our only reference to it. */
731 class_manual_cleanup(osc_obd);
734 void lov_fix_desc_stripe_size(__u64 *val)
736 if (*val < LOV_MIN_STRIPE_SIZE) {
738 LCONSOLE_INFO("Increasing default stripe size to "
740 LOV_DESC_STRIPE_SIZE_DEFAULT);
741 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
742 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
743 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
744 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
746 *val, LOV_MIN_STRIPE_SIZE);
750 void lov_fix_desc_stripe_count(__u32 *val)
756 void lov_fix_desc_pattern(__u32 *val)
758 /* from lov_setstripe */
759 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
760 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
765 void lov_fix_desc_qos_maxage(__u32 *val)
768 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
771 void lov_fix_desc(struct lov_desc *desc)
773 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
774 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
775 lov_fix_desc_pattern(&desc->ld_pattern);
776 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
779 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
781 struct lov_desc *desc;
782 struct lov_obd *lov = &obd->u.lov;
786 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
787 CERROR("LOV setup requires a descriptor\n");
791 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
793 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
794 CERROR("descriptor size wrong: %d > %d\n",
795 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
799 if (desc->ld_magic != LOV_DESC_MAGIC) {
800 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
801 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
802 obd->obd_name, desc);
803 lustre_swab_lov_desc(desc);
805 CERROR("%s: Bad lov desc magic: %#x\n",
806 obd->obd_name, desc->ld_magic);
813 desc->ld_active_tgt_count = 0;
815 lov->lov_tgt_size = 0;
817 mutex_init(&lov->lov_lock);
818 atomic_set(&lov->lov_refcount, 0);
819 lov->lov_sp_me = LUSTRE_SP_CLI;
821 init_rwsem(&lov->lov_notify_lock);
823 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
825 HASH_POOLS_BKT_BITS, 0,
828 &pool_hash_operations,
830 CFS_INIT_LIST_HEAD(&lov->lov_pool_list);
831 lov->lov_pool_count = 0;
832 rc = lov_ost_pool_init(&lov->lov_packed, 0);
837 obd->obd_vars = lprocfs_lov_obd_vars;
838 lprocfs_seq_obd_setup(obd);
839 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", 0444,
840 &lov_proc_target_fops, obd);
842 CWARN("Error adding the target_obd file\n");
844 lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
854 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
857 struct lov_obd *lov = &obd->u.lov;
862 case OBD_CLEANUP_EARLY: {
864 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
865 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
867 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
879 static int lov_cleanup(struct obd_device *obd)
881 struct lov_obd *lov = &obd->u.lov;
882 cfs_list_t *pos, *tmp;
883 struct pool_desc *pool;
886 cfs_list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
887 pool = cfs_list_entry(pos, struct pool_desc, pool_list);
888 /* free pool structs */
889 CDEBUG(D_INFO, "delete pool %p\n", pool);
890 /* In the function below, .hs_keycmp resolves to
891 * pool_hashkey_keycmp() */
892 /* coverity[overrun-buffer-val] */
893 lov_pool_del(obd, pool->pool_name);
895 cfs_hash_putref(lov->lov_pools_hash_body);
896 lov_ost_pool_free(&lov->lov_packed);
898 lprocfs_obd_cleanup(obd);
902 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
903 if (!lov->lov_tgts[i])
906 /* Inactive targets may never have connected */
907 if (lov->lov_tgts[i]->ltd_active ||
908 atomic_read(&lov->lov_refcount))
909 /* We should never get here - these
910 * should have been removed in the
912 CERROR("%s: lov tgt %d not cleaned! "
913 "deathrow=%d, lovrc=%d\n",
914 obd->obd_name, i, lov->lov_death_row,
915 atomic_read(&lov->lov_refcount));
916 lov_del_target(obd, i, 0, 0);
919 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
921 lov->lov_tgt_size = 0;
926 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
927 __u32 *indexp, int *genp)
929 struct obd_uuid obd_uuid;
934 switch(cmd = lcfg->lcfg_command) {
935 case LCFG_LOV_ADD_OBD:
936 case LCFG_LOV_ADD_INA:
937 case LCFG_LOV_DEL_OBD: {
940 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
941 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
942 GOTO(out, rc = -EINVAL);
944 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
946 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1)
947 GOTO(out, rc = -EINVAL);
948 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
949 GOTO(out, rc = -EINVAL);
952 if (cmd == LCFG_LOV_ADD_OBD)
953 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
954 else if (cmd == LCFG_LOV_ADD_INA)
955 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
957 rc = lov_del_target(obd, index, &obd_uuid, gen);
961 struct lov_desc *desc = &(obd->u.lov.desc);
964 GOTO(out, rc = -EINVAL);
966 rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
979 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
980 GOTO(out, rc = -EINVAL);
988 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
989 struct lov_stripe_md **ea, struct obd_trans_info *oti)
991 struct lov_stripe_md *obj_mdp, *lsm;
992 struct lov_obd *lov = &exp->exp_obd->u.lov;
997 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
998 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1000 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1001 if (obj_mdp == NULL)
1004 ost_idx = src_oa->o_nlink;
1007 GOTO(out, rc = -EINVAL);
1008 if (ost_idx >= lov->desc.ld_tgt_count ||
1009 !lov->lov_tgts[ost_idx])
1010 GOTO(out, rc = -EINVAL);
1012 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1013 if (lsm->lsm_oinfo[i]->loi_ost_idx == ost_idx) {
1014 if (ostid_id(&lsm->lsm_oinfo[i]->loi_oi) !=
1015 ostid_id(&src_oa->o_oi))
1016 GOTO(out, rc = -EINVAL);
1020 if (i == lsm->lsm_stripe_count)
1021 GOTO(out, rc = -EINVAL);
1023 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1024 src_oa, &obj_mdp, oti);
1026 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1030 /* the LOV expects oa->o_id to be set to the LOV object id */
1031 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1032 struct obdo *src_oa, struct lov_stripe_md **ea,
1033 struct obd_trans_info *oti)
1035 struct lov_obd *lov;
1039 LASSERT(ea != NULL);
1043 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1044 src_oa->o_flags == OBD_FL_DELORPHAN) {
1045 /* should be used with LOV anymore */
1049 lov = &exp->exp_obd->u.lov;
1050 if (!lov->desc.ld_active_tgt_count)
1053 obd_getref(exp->exp_obd);
1054 /* Recreate a specific object id at the given OST index */
1055 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1056 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1057 rc = lov_recreate(exp, src_oa, ea, oti);
1060 obd_putref(exp->exp_obd);
1064 #define ASSERT_LSM_MAGIC(lsmp) \
1066 LASSERT((lsmp) != NULL); \
1067 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1068 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1069 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1072 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1073 struct obdo *oa, struct lov_stripe_md *lsm,
1074 struct obd_trans_info *oti, struct obd_export *md_exp,
1077 struct lov_request_set *set;
1078 struct obd_info oinfo;
1079 struct lov_request *req;
1081 struct lov_obd *lov;
1082 int rc = 0, err = 0;
1085 ASSERT_LSM_MAGIC(lsm);
1087 if (!exp || !exp->exp_obd)
1090 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1092 LASSERT(oti->oti_logcookies);
1095 lov = &exp->exp_obd->u.lov;
1096 obd_getref(exp->exp_obd);
1097 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1101 cfs_list_for_each (pos, &set->set_list) {
1102 req = cfs_list_entry(pos, struct lov_request, rq_link);
1104 if (oa->o_valid & OBD_MD_FLCOOKIE)
1105 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1107 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1108 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1109 err = lov_update_common_set(set, req, err);
1111 CERROR("%s: destroying objid "DOSTID" subobj "
1112 DOSTID" on OST idx %d: rc = %d\n",
1113 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1114 POSTID(&req->rq_oi.oi_oa->o_oi),
1122 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1123 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1125 err = lov_fini_destroy_set(set);
1127 obd_putref(exp->exp_obd);
1128 RETURN(rc ? rc : err);
1131 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1134 struct lov_request_set *lovset = (struct lov_request_set *)data;
1138 /* don't do attribute merge if this aysnc op failed */
1140 atomic_set(&lovset->set_completes, 0);
1141 err = lov_fini_getattr_set(lovset);
1142 RETURN(rc ? rc : err);
1145 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1146 struct ptlrpc_request_set *rqset)
1148 struct lov_request_set *lovset;
1149 struct lov_obd *lov;
1151 struct lov_request *req;
1156 ASSERT_LSM_MAGIC(oinfo->oi_md);
1158 if (!exp || !exp->exp_obd)
1161 lov = &exp->exp_obd->u.lov;
1163 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1167 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1168 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1169 oinfo->oi_md->lsm_stripe_size);
1171 cfs_list_for_each(pos, &lovset->set_list) {
1172 req = cfs_list_entry(pos, struct lov_request, rq_link);
1174 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1175 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1176 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1177 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1178 &req->rq_oi, rqset);
1180 CERROR("%s: getattr objid "DOSTID" subobj"
1181 DOSTID" on OST idx %d: rc = %d\n",
1182 exp->exp_obd->obd_name,
1183 POSTID(&oinfo->oi_oa->o_oi),
1184 POSTID(&req->rq_oi.oi_oa->o_oi),
1190 if (!cfs_list_empty(&rqset->set_requests)) {
1192 LASSERT (rqset->set_interpret == NULL);
1193 rqset->set_interpret = lov_getattr_interpret;
1194 rqset->set_arg = (void *)lovset;
1199 atomic_set(&lovset->set_completes, 0);
1200 err = lov_fini_getattr_set(lovset);
1201 RETURN(rc ? rc : err);
1204 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1207 struct lov_request_set *lovset = (struct lov_request_set *)data;
1212 atomic_set(&lovset->set_completes, 0);
1213 err = lov_fini_setattr_set(lovset);
1214 RETURN(rc ? rc : err);
1217 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1218 needed. Otherwise, a client is waiting for responses. */
1219 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1220 struct obd_trans_info *oti,
1221 struct ptlrpc_request_set *rqset)
1223 struct lov_request_set *set;
1224 struct lov_request *req;
1226 struct lov_obd *lov;
1231 ASSERT_LSM_MAGIC(oinfo->oi_md);
1232 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1234 LASSERT(oti->oti_logcookies);
1237 if (!exp || !exp->exp_obd)
1240 lov = &exp->exp_obd->u.lov;
1241 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1245 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1246 POSTID(&oinfo->oi_md->lsm_oi),
1247 oinfo->oi_md->lsm_stripe_count,
1248 oinfo->oi_md->lsm_stripe_size);
1250 cfs_list_for_each(pos, &set->set_list) {
1251 req = cfs_list_entry(pos, struct lov_request, rq_link);
1253 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1254 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1256 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1257 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1258 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1260 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1261 &req->rq_oi, oti, rqset);
1263 CERROR("error: setattr objid "DOSTID" subobj"
1264 DOSTID" on OST idx %d: rc = %d\n",
1265 POSTID(&set->set_oi->oi_oa->o_oi),
1266 POSTID(&req->rq_oi.oi_oa->o_oi),
1272 /* If we are not waiting for responses on async requests, return. */
1273 if (rc || !rqset || cfs_list_empty(&rqset->set_requests)) {
1276 atomic_set(&set->set_completes, 0);
1277 err = lov_fini_setattr_set(set);
1278 RETURN(rc ? rc : err);
1281 LASSERT(rqset->set_interpret == NULL);
1282 rqset->set_interpret = lov_setattr_interpret;
1283 rqset->set_arg = (void *)set;
1288 static int lov_change_cbdata(struct obd_export *exp,
1289 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1292 struct lov_obd *lov;
1296 ASSERT_LSM_MAGIC(lsm);
1298 if (!exp || !exp->exp_obd)
1301 lov = &exp->exp_obd->u.lov;
1302 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1303 struct lov_stripe_md submd;
1304 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1306 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1307 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1311 submd.lsm_oi = loi->loi_oi;
1312 submd.lsm_stripe_count = 0;
1313 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1319 /* find any ldlm lock of the inode in lov
1323 static int lov_find_cbdata(struct obd_export *exp,
1324 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1327 struct lov_obd *lov;
1331 ASSERT_LSM_MAGIC(lsm);
1333 if (!exp || !exp->exp_obd)
1336 lov = &exp->exp_obd->u.lov;
1337 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1338 struct lov_stripe_md submd;
1339 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1341 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1342 CDEBUG(D_HA, "lov idx %d NULL \n", loi->loi_ost_idx);
1345 submd.lsm_oi = loi->loi_oi;
1346 submd.lsm_stripe_count = 0;
1347 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1355 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1357 struct lov_request_set *lovset = (struct lov_request_set *)data;
1362 atomic_set(&lovset->set_completes, 0);
1364 err = lov_fini_statfs_set(lovset);
1365 RETURN(rc ? rc : err);
1368 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1369 __u64 max_age, struct ptlrpc_request_set *rqset)
1371 struct obd_device *obd = class_exp2obd(exp);
1372 struct lov_request_set *set;
1373 struct lov_request *req;
1375 struct lov_obd *lov;
1379 LASSERT(oinfo != NULL);
1380 LASSERT(oinfo->oi_osfs != NULL);
1383 rc = lov_prep_statfs_set(obd, oinfo, &set);
1387 cfs_list_for_each (pos, &set->set_list) {
1388 req = cfs_list_entry(pos, struct lov_request, rq_link);
1389 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1390 &req->rq_oi, max_age, rqset);
1395 if (rc || cfs_list_empty(&rqset->set_requests)) {
1398 atomic_set(&set->set_completes, 0);
1399 err = lov_fini_statfs_set(set);
1400 RETURN(rc ? rc : err);
1403 LASSERT(rqset->set_interpret == NULL);
1404 rqset->set_interpret = lov_statfs_interpret;
1405 rqset->set_arg = (void *)set;
1409 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1410 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1412 struct ptlrpc_request_set *set = NULL;
1413 struct obd_info oinfo = { { { 0 } } };
1418 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1419 * statfs requests */
1420 set = ptlrpc_prep_set();
1424 oinfo.oi_osfs = osfs;
1425 oinfo.oi_flags = flags;
1426 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1428 rc = ptlrpc_set_wait(set);
1429 ptlrpc_set_destroy(set);
1434 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1435 void *karg, void *uarg)
1437 struct obd_device *obddev = class_exp2obd(exp);
1438 struct lov_obd *lov = &obddev->u.lov;
1439 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1440 struct obd_uuid *uuidp;
1444 case IOC_OBD_STATFS: {
1445 struct obd_ioctl_data *data = karg;
1446 struct obd_device *osc_obd;
1447 struct obd_statfs stat_buf = {0};
1451 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1452 if ((index >= count))
1455 if (!lov->lov_tgts[index])
1456 /* Try again with the next index */
1458 if (!lov->lov_tgts[index]->ltd_active)
1461 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1466 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1467 min((int)data->ioc_plen2,
1468 (int)sizeof(struct obd_uuid))))
1471 flags = uarg ? *(__u32*)uarg : 0;
1472 /* got statfs data */
1473 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1474 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1478 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1479 min((int) data->ioc_plen1,
1480 (int) sizeof(stat_buf))))
1484 case OBD_IOC_LOV_GET_CONFIG: {
1485 struct obd_ioctl_data *data;
1486 struct lov_desc *desc;
1491 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1494 data = (struct obd_ioctl_data *)buf;
1496 if (sizeof(*desc) > data->ioc_inllen1) {
1497 obd_ioctl_freedata(buf, len);
1501 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1502 obd_ioctl_freedata(buf, len);
1506 if (sizeof(__u32) * count > data->ioc_inllen3) {
1507 obd_ioctl_freedata(buf, len);
1511 desc = (struct lov_desc *)data->ioc_inlbuf1;
1512 memcpy(desc, &(lov->desc), sizeof(*desc));
1514 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1515 genp = (__u32 *)data->ioc_inlbuf3;
1516 /* the uuid will be empty for deleted OSTs */
1517 for (i = 0; i < count; i++, uuidp++, genp++) {
1518 if (!lov->lov_tgts[i])
1520 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1521 *genp = lov->lov_tgts[i]->ltd_gen;
1524 if (copy_to_user((void *)uarg, buf, len))
1526 obd_ioctl_freedata(buf, len);
1529 case LL_IOC_LOV_GETSTRIPE:
1530 rc = lov_getstripe(exp, karg, uarg);
1532 case OBD_IOC_QUOTACTL: {
1533 struct if_quotactl *qctl = karg;
1534 struct lov_tgt_desc *tgt = NULL;
1535 struct obd_quotactl *oqctl;
1537 if (qctl->qc_valid == QC_OSTIDX) {
1538 if (count <= qctl->qc_idx)
1541 tgt = lov->lov_tgts[qctl->qc_idx];
1542 if (!tgt || !tgt->ltd_exp)
1544 } else if (qctl->qc_valid == QC_UUID) {
1545 for (i = 0; i < count; i++) {
1546 tgt = lov->lov_tgts[i];
1548 !obd_uuid_equals(&tgt->ltd_uuid,
1552 if (tgt->ltd_exp == NULL)
1564 LASSERT(tgt && tgt->ltd_exp);
1565 OBD_ALLOC_PTR(oqctl);
1569 QCTL_COPY(oqctl, qctl);
1570 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1572 QCTL_COPY(qctl, oqctl);
1573 qctl->qc_valid = QC_OSTIDX;
1574 qctl->obd_uuid = tgt->ltd_uuid;
1576 OBD_FREE_PTR(oqctl);
1585 for (i = 0; i < count; i++) {
1587 struct obd_device *osc_obd;
1589 /* OST was disconnected */
1590 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1593 /* ll_umount_begin() sets force flag but for lov, not
1594 * osc. Let's pass it through */
1595 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1596 osc_obd->obd_force = obddev->obd_force;
1597 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1599 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1602 if (lov->lov_tgts[i]->ltd_active) {
1603 CDEBUG(err == -ENOTTY ?
1604 D_IOCTL : D_WARNING,
1605 "iocontrol OSC %s on OST "
1606 "idx %d cmd %x: err = %d\n",
1607 lov_uuid2str(lov, i),
1624 #define FIEMAP_BUFFER_SIZE 4096
1627 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1628 * call. The local end offset and the device are sent in the first
1629 * fm_extent. This function calculates the stripe number from the index.
1630 * This function returns a stripe_no on which mapping is to be restarted.
1632 * This function returns fm_end_offset which is the in-OST offset at which
1633 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1634 * will re-calculate proper offset in next stripe.
1635 * Note that the first extent is passed to lov_get_info via the value field.
1637 * \param fiemap fiemap request header
1638 * \param lsm striping information for the file
1639 * \param fm_start logical start of mapping
1640 * \param fm_end logical end of mapping
1641 * \param start_stripe starting stripe will be returned in this
1643 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1644 struct lov_stripe_md *lsm, obd_size fm_start,
1645 obd_size fm_end, int *start_stripe)
1647 obd_size local_end = fiemap->fm_extents[0].fe_logical;
1648 obd_off lun_start, lun_end;
1649 obd_size fm_end_offset;
1650 int stripe_no = -1, i;
1652 if (fiemap->fm_extent_count == 0 ||
1653 fiemap->fm_extents[0].fe_logical == 0)
1656 /* Find out stripe_no from ost_index saved in the fe_device */
1657 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1658 if (lsm->lsm_oinfo[i]->loi_ost_idx ==
1659 fiemap->fm_extents[0].fe_device) {
1664 if (stripe_no == -1)
1667 /* If we have finished mapping on previous device, shift logical
1668 * offset to start of next device */
1669 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1670 &lun_start, &lun_end)) != 0 &&
1671 local_end < lun_end) {
1672 fm_end_offset = local_end;
1673 *start_stripe = stripe_no;
1675 /* This is a special value to indicate that caller should
1676 * calculate offset in next stripe. */
1678 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1681 return fm_end_offset;
1685 * We calculate on which OST the mapping will end. If the length of mapping
1686 * is greater than (stripe_size * stripe_count) then the last_stripe will
1687 * will be one just before start_stripe. Else we check if the mapping
1688 * intersects each OST and find last_stripe.
1689 * This function returns the last_stripe and also sets the stripe_count
1690 * over which the mapping is spread
1692 * \param lsm striping information for the file
1693 * \param fm_start logical start of mapping
1694 * \param fm_end logical end of mapping
1695 * \param start_stripe starting stripe of the mapping
1696 * \param stripe_count the number of stripes across which to map is returned
1698 * \retval last_stripe return the last stripe of the mapping
1700 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
1701 obd_size fm_end, int start_stripe,
1705 obd_off obd_start, obd_end;
1708 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1709 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1711 *stripe_count = lsm->lsm_stripe_count;
1713 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1714 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1715 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1716 &obd_start, &obd_end)) == 0)
1720 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1727 * Set fe_device and copy extents from local buffer into main return buffer.
1729 * \param fiemap fiemap request header
1730 * \param lcl_fm_ext array of local fiemap extents to be copied
1731 * \param ost_index OST index to be written into the fm_device field for each
1733 * \param ext_count number of extents to be copied
1734 * \param current_extent where to start copying in main extent array
1736 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1737 struct ll_fiemap_extent *lcl_fm_ext,
1738 int ost_index, unsigned int ext_count,
1744 for (ext = 0; ext < ext_count; ext++) {
1745 lcl_fm_ext[ext].fe_device = ost_index;
1746 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1749 /* Copy fm_extent's from fm_local to return buffer */
1750 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1751 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1755 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1756 * This also handles the restarting of FIEMAP calls in case mapping overflows
1757 * the available number of extents in single call.
1759 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1760 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1762 struct ll_fiemap_info_key *fm_key = key;
1763 struct ll_user_fiemap *fiemap = val;
1764 struct ll_user_fiemap *fm_local = NULL;
1765 struct ll_fiemap_extent *lcl_fm_ext;
1767 unsigned int get_num_extents = 0;
1768 int ost_index = 0, actual_start_stripe, start_stripe;
1769 obd_size fm_start, fm_end, fm_length, fm_end_offset;
1771 int current_extent = 0, rc = 0, i;
1772 int ost_eof = 0; /* EOF for object */
1773 int ost_done = 0; /* done with required mapping for this OST? */
1775 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1776 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1778 if (!lsm_has_objects(lsm)) {
1779 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1780 fm_key->oa.o_size)) {
1781 /* released file, return a minimal FIEMAP if
1782 * request fits in file-size.
1784 fiemap->fm_mapped_extents = 1;
1785 fiemap->fm_extents[0].fe_logical =
1786 fm_key->fiemap.fm_start;
1787 if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1789 fiemap->fm_extents[0].fe_length =
1790 fm_key->fiemap.fm_length;
1792 fiemap->fm_extents[0].fe_length =
1794 fm_key->fiemap.fm_start;
1795 fiemap->fm_extents[0].fe_flags |=
1796 (FIEMAP_EXTENT_UNKNOWN |
1797 FIEMAP_EXTENT_LAST);
1802 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1803 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1805 OBD_ALLOC_LARGE(fm_local, buffer_size);
1806 if (fm_local == NULL)
1807 GOTO(out, rc = -ENOMEM);
1808 lcl_fm_ext = &fm_local->fm_extents[0];
1810 count_local = fiemap_size_to_count(buffer_size);
1812 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1813 fm_start = fiemap->fm_start;
1814 fm_length = fiemap->fm_length;
1815 /* Calculate start stripe, last stripe and length of mapping */
1816 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1817 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1818 fm_start + fm_length - 1);
1819 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1820 if (fm_end > fm_key->oa.o_size)
1821 fm_end = fm_key->oa.o_size;
1823 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1824 actual_start_stripe, &stripe_count);
1826 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1827 fm_end, &start_stripe);
1828 if (fm_end_offset == -EINVAL)
1829 GOTO(out, rc = -EINVAL);
1831 if (fiemap->fm_extent_count == 0) {
1832 get_num_extents = 1;
1836 /* Check each stripe */
1837 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1838 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1839 obd_size req_fm_len; /* Stores length of required mapping */
1840 obd_size len_mapped_single_call;
1841 obd_off lun_start, lun_end, obd_object_end;
1842 unsigned int ext_count;
1844 cur_stripe_wrap = cur_stripe;
1846 /* Find out range of mapping on this stripe */
1847 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1848 &lun_start, &obd_object_end)) == 0)
1851 /* If this is a continuation FIEMAP call and we are on
1852 * starting stripe then lun_start needs to be set to
1854 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1855 lun_start = fm_end_offset;
1857 if (fm_length != ~0ULL) {
1858 /* Handle fm_start + fm_length overflow */
1859 if (fm_start + fm_length < fm_start)
1860 fm_length = ~0ULL - fm_start;
1861 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1867 if (lun_start == lun_end)
1870 req_fm_len = obd_object_end - lun_start;
1871 fm_local->fm_length = 0;
1872 len_mapped_single_call = 0;
1874 /* If the output buffer is very large and the objects have many
1875 * extents we may need to loop on a single OST repeatedly */
1879 if (get_num_extents == 0) {
1880 /* Don't get too many extents. */
1881 if (current_extent + count_local >
1882 fiemap->fm_extent_count)
1883 count_local = fiemap->fm_extent_count -
1887 lun_start += len_mapped_single_call;
1888 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1889 req_fm_len = fm_local->fm_length;
1890 fm_local->fm_extent_count = count_local;
1891 fm_local->fm_mapped_extents = 0;
1892 fm_local->fm_flags = fiemap->fm_flags;
1894 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1895 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1897 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1898 GOTO(out, rc = -EINVAL);
1900 /* If OST is inactive, return extent with UNKNOWN flag */
1901 if (!lov->lov_tgts[ost_index]->ltd_active) {
1902 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1903 fm_local->fm_mapped_extents = 1;
1905 lcl_fm_ext[0].fe_logical = lun_start;
1906 lcl_fm_ext[0].fe_length = obd_object_end -
1908 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1913 fm_local->fm_start = lun_start;
1914 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1915 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1916 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1917 rc = obd_get_info(NULL,
1918 lov->lov_tgts[ost_index]->ltd_exp,
1919 keylen, key, vallen, fm_local, lsm);
1924 ext_count = fm_local->fm_mapped_extents;
1925 if (ext_count == 0) {
1927 /* If last stripe has hole at the end,
1928 * then we need to return */
1929 if (cur_stripe_wrap == last_stripe) {
1930 fiemap->fm_mapped_extents = 0;
1936 /* If we just need num of extents then go to next device */
1937 if (get_num_extents) {
1938 current_extent += ext_count;
1942 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1943 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1945 /* Have we finished mapping on this device? */
1946 if (req_fm_len <= len_mapped_single_call)
1949 /* Clear the EXTENT_LAST flag which can be present on
1951 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1952 lcl_fm_ext[ext_count - 1].fe_flags &=
1953 ~FIEMAP_EXTENT_LAST;
1955 curr_loc = lov_stripe_size(lsm,
1956 lcl_fm_ext[ext_count - 1].fe_logical+
1957 lcl_fm_ext[ext_count - 1].fe_length,
1959 if (curr_loc >= fm_key->oa.o_size)
1962 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1963 ost_index, ext_count,
1966 current_extent += ext_count;
1968 /* Ran out of available extents? */
1969 if (current_extent >= fiemap->fm_extent_count)
1971 } while (ost_done == 0 && ost_eof == 0);
1973 if (cur_stripe_wrap == last_stripe)
1978 /* Indicate that we are returning device offsets unless file just has
1980 if (lsm->lsm_stripe_count > 1)
1981 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1983 if (get_num_extents)
1984 goto skip_last_device_calc;
1986 /* Check if we have reached the last stripe and whether mapping for that
1987 * stripe is done. */
1988 if (cur_stripe_wrap == last_stripe) {
1989 if (ost_done || ost_eof)
1990 fiemap->fm_extents[current_extent - 1].fe_flags |=
1994 skip_last_device_calc:
1995 fiemap->fm_mapped_extents = current_extent;
1999 OBD_FREE_LARGE(fm_local, buffer_size);
2003 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2004 __u32 keylen, void *key, __u32 *vallen, void *val,
2005 struct lov_stripe_md *lsm)
2007 struct obd_device *obddev = class_exp2obd(exp);
2008 struct lov_obd *lov = &obddev->u.lov;
2012 if (!vallen || !val)
2017 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2020 struct ldlm_lock *lock;
2022 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2023 struct lov_oinfo *loi;
2024 __u32 *stripe = val;
2026 if (*vallen < sizeof(*stripe))
2027 GOTO(out, rc = -EFAULT);
2028 *vallen = sizeof(*stripe);
2030 /* XXX This is another one of those bits that will need to
2031 * change if we ever actually support nested LOVs. It uses
2032 * the lock's export to find out which stripe it is. */
2033 /* XXX - it's assumed all the locks for deleted OSTs have
2034 * been cancelled. Also, the export for deleted OSTs will
2035 * be NULL and won't match the lock's export. */
2036 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2037 loi = lsm->lsm_oinfo[i];
2038 if (!lov->lov_tgts[loi->loi_ost_idx])
2040 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2041 data->lock->l_conn_export &&
2042 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2047 LDLM_ERROR(data->lock, "lock on inode without such object");
2048 dump_lsm(D_ERROR, lsm);
2049 GOTO(out, rc = -ENXIO);
2050 } else if (KEY_IS(KEY_LAST_ID)) {
2051 struct obd_id_info *info = val;
2052 __u32 size = sizeof(obd_id);
2053 struct lov_tgt_desc *tgt;
2055 LASSERT(*vallen == sizeof(struct obd_id_info));
2056 tgt = lov->lov_tgts[info->idx];
2058 if (!tgt || !tgt->ltd_active)
2059 GOTO(out, rc = -ESRCH);
2061 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2062 &size, info->data, NULL);
2064 } else if (KEY_IS(KEY_LOVDESC)) {
2065 struct lov_desc *desc_ret = val;
2066 *desc_ret = lov->desc;
2069 } else if (KEY_IS(KEY_FIEMAP)) {
2070 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2072 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2073 struct lov_tgt_desc *tgt;
2074 __u64 ost_idx = *((__u64*)val);
2076 LASSERT(*vallen == sizeof(__u64));
2077 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2078 tgt = lov->lov_tgts[ost_idx];
2080 if (!tgt || !tgt->ltd_exp)
2081 GOTO(out, rc = -ESRCH);
2083 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2085 } else if (KEY_IS(KEY_TGT_COUNT)) {
2086 *((int *)val) = lov->desc.ld_tgt_count;
2097 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2098 obd_count keylen, void *key, obd_count vallen,
2099 void *val, struct ptlrpc_request_set *set)
2101 struct obd_device *obddev = class_exp2obd(exp);
2102 struct lov_obd *lov = &obddev->u.lov;
2105 struct lov_tgt_desc *tgt;
2106 unsigned incr, check_uuid,
2107 do_inactive, no_set;
2108 unsigned next_id = 0, mds_con = 0, capa = 0;
2111 incr = check_uuid = do_inactive = no_set = 0;
2114 set = ptlrpc_prep_set();
2120 count = lov->desc.ld_tgt_count;
2122 if (KEY_IS(KEY_NEXT_ID)) {
2123 count = vallen / sizeof(struct obd_id_info);
2124 vallen = sizeof(obd_id);
2125 incr = sizeof(struct obd_id_info);
2128 } else if (KEY_IS(KEY_CHECKSUM)) {
2130 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2131 /* use defaults: do_inactive = incr = 0; */
2132 } else if (KEY_IS(KEY_MDS_CONN)) {
2134 } else if (KEY_IS(KEY_CAPA_KEY)) {
2136 } else if (KEY_IS(KEY_CACHE_SET)) {
2137 LASSERT(lov->lov_cache == NULL);
2138 lov->lov_cache = val;
2142 for (i = 0; i < count; i++, val = (char *)val + incr) {
2144 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2146 tgt = lov->lov_tgts[i];
2148 /* OST was disconnected */
2149 if (!tgt || !tgt->ltd_exp)
2152 /* OST is inactive and we don't want inactive OSCs */
2153 if (!tgt->ltd_active && !do_inactive)
2157 struct mds_group_info *mgi;
2159 LASSERT(vallen == sizeof(*mgi));
2160 mgi = (struct mds_group_info *)val;
2162 /* Only want a specific OSC */
2163 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2167 err = obd_set_info_async(env, tgt->ltd_exp,
2168 keylen, key, sizeof(int),
2170 } else if (next_id) {
2171 err = obd_set_info_async(env, tgt->ltd_exp,
2172 keylen, key, vallen,
2173 ((struct obd_id_info*)val)->data, set);
2175 struct mds_capa_info *info = (struct mds_capa_info*)val;
2177 LASSERT(vallen == sizeof(*info));
2179 /* Only want a specific OSC */
2181 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2184 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2185 key, sizeof(*info->capa),
2188 /* Only want a specific OSC */
2190 !obd_uuid_equals(val, &tgt->ltd_uuid))
2193 err = obd_set_info_async(env, tgt->ltd_exp,
2194 keylen, key, vallen, val, set);
2203 err = ptlrpc_set_wait(set);
2206 ptlrpc_set_destroy(set);
2211 void lov_stripe_lock(struct lov_stripe_md *md)
2213 LASSERT(md->lsm_lock_owner != current_pid());
2214 spin_lock(&md->lsm_lock);
2215 LASSERT(md->lsm_lock_owner == 0);
2216 md->lsm_lock_owner = current_pid();
2219 void lov_stripe_unlock(struct lov_stripe_md *md)
2221 LASSERT(md->lsm_lock_owner == current_pid());
2222 md->lsm_lock_owner = 0;
2223 spin_unlock(&md->lsm_lock);
2226 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2227 struct obd_quotactl *oqctl)
2229 struct lov_obd *lov = &obd->u.lov;
2230 struct lov_tgt_desc *tgt;
2232 __u64 bhardlimit = 0;
2236 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2237 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2238 oqctl->qc_cmd != Q_GETOQUOTA &&
2239 oqctl->qc_cmd != Q_INITQUOTA &&
2240 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2241 oqctl->qc_cmd != Q_FINVALIDATE) {
2242 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2248 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2251 tgt = lov->lov_tgts[i];
2256 if (!tgt->ltd_active || tgt->ltd_reap) {
2257 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2258 lov->lov_tgts[i]->ltd_activate) {
2260 CERROR("ost %d is inactive\n", i);
2262 CDEBUG(D_HA, "ost %d is inactive\n", i);
2267 err = obd_quotactl(tgt->ltd_exp, oqctl);
2269 if (tgt->ltd_active && !rc)
2274 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2275 curspace += oqctl->qc_dqblk.dqb_curspace;
2276 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2281 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2282 oqctl->qc_dqblk.dqb_curspace = curspace;
2283 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2288 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2289 struct obd_quotactl *oqctl)
2291 struct lov_obd *lov = &obd->u.lov;
2297 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2298 if (!lov->lov_tgts[i])
2301 /* Skip quota check on the administratively disabled OSTs. */
2302 if (!lov->lov_tgts[i]->ltd_activate) {
2303 CWARN("lov idx %d was administratively disabled, "
2304 "skip quotacheck on it.\n", i);
2308 if (!lov->lov_tgts[i]->ltd_active) {
2309 CERROR("lov idx %d inactive\n", i);
2315 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2318 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2321 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2332 static struct obd_ops lov_obd_ops = {
2333 .o_owner = THIS_MODULE,
2334 .o_setup = lov_setup,
2335 .o_precleanup = lov_precleanup,
2336 .o_cleanup = lov_cleanup,
2337 .o_connect = lov_connect,
2338 .o_disconnect = lov_disconnect,
2339 .o_statfs = lov_statfs,
2340 .o_statfs_async = lov_statfs_async,
2341 .o_packmd = lov_packmd,
2342 .o_unpackmd = lov_unpackmd,
2343 .o_create = lov_create,
2344 .o_destroy = lov_destroy,
2345 .o_getattr_async = lov_getattr_async,
2346 .o_setattr_async = lov_setattr_async,
2347 .o_change_cbdata = lov_change_cbdata,
2348 .o_find_cbdata = lov_find_cbdata,
2349 .o_iocontrol = lov_iocontrol,
2350 .o_get_info = lov_get_info,
2351 .o_set_info_async = lov_set_info_async,
2352 .o_notify = lov_notify,
2353 .o_pool_new = lov_pool_new,
2354 .o_pool_rem = lov_pool_remove,
2355 .o_pool_add = lov_pool_add,
2356 .o_pool_del = lov_pool_del,
2357 .o_getref = lov_getref,
2358 .o_putref = lov_putref,
2359 .o_quotactl = lov_quotactl,
2360 .o_quotacheck = lov_quotacheck,
2363 struct kmem_cache *lov_oinfo_slab;
2365 extern struct lu_kmem_descr lov_caches[];
2367 int __init lov_init(void)
2372 /* print an address of _any_ initialized kernel symbol from this
2373 * module, to allow debugging with gdb that doesn't support data
2374 * symbols from modules.*/
2375 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2377 rc = lu_kmem_init(lov_caches);
2381 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2382 sizeof(struct lov_oinfo), 0,
2383 SLAB_HWCACHE_ALIGN, NULL);
2384 if (lov_oinfo_slab == NULL) {
2385 lu_kmem_fini(lov_caches);
2389 rc = class_register_type(&lov_obd_ops, NULL, NULL,
2390 #ifndef HAVE_ONLY_PROCFS_SEQ
2393 LUSTRE_LOV_NAME, &lov_device_type);
2396 kmem_cache_destroy(lov_oinfo_slab);
2397 lu_kmem_fini(lov_caches);
2404 static void /*__exit*/ lov_exit(void)
2406 class_unregister_type(LUSTRE_LOV_NAME);
2407 kmem_cache_destroy(lov_oinfo_slab);
2408 lu_kmem_fini(lov_caches);
2411 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2412 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2413 MODULE_LICENSE("GPL");
2415 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);