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_ioctl.h>
53 #include <lustre_lib.h>
54 #include <lustre_net.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_dlm.h>
57 #include <lustre_mds.h>
58 #include <obd_class.h>
59 #include <lprocfs_status.h>
60 #include <lustre_param.h>
61 #include <cl_object.h>
63 #include <lustre/ll_fiemap.h>
64 #include <lustre_fid.h>
66 #include "lov_internal.h"
68 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
69 Any function that expects lov_tgts to remain stationary must take a ref. */
70 static void lov_getref(struct obd_device *obd)
72 struct lov_obd *lov = &obd->u.lov;
74 /* nobody gets through here until lov_putref is done */
75 mutex_lock(&lov->lov_lock);
76 atomic_inc(&lov->lov_refcount);
77 mutex_unlock(&lov->lov_lock);
81 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
83 static void lov_putref(struct obd_device *obd)
85 struct lov_obd *lov = &obd->u.lov;
87 mutex_lock(&lov->lov_lock);
88 /* ok to dec to 0 more than once -- ltd_exp's will be null */
89 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
90 struct list_head kill = LIST_HEAD_INIT(kill);
91 struct lov_tgt_desc *tgt, *n;
94 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
96 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
97 tgt = lov->lov_tgts[i];
99 if (!tgt || !tgt->ltd_reap)
101 list_add(&tgt->ltd_kill, &kill);
102 /* XXX - right now there is a dependency on ld_tgt_count
103 * being the maximum tgt index for computing the
104 * mds_max_easize. So we can't shrink it. */
105 lov_ost_pool_remove(&lov->lov_packed, i);
106 lov->lov_tgts[i] = NULL;
107 lov->lov_death_row--;
109 mutex_unlock(&lov->lov_lock);
111 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
112 list_del(&tgt->ltd_kill);
114 __lov_del_obd(obd, tgt);
117 mutex_unlock(&lov->lov_lock);
121 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
122 enum obd_notify_event ev);
123 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
124 enum obd_notify_event ev, void *data);
126 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
127 struct obd_connect_data *data)
129 struct lov_obd *lov = &obd->u.lov;
130 struct obd_uuid *tgt_uuid;
131 struct obd_device *tgt_obd;
132 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
133 struct obd_import *imp;
137 if (lov->lov_tgts[index] == NULL)
140 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
141 tgt_obd = lov->lov_tgts[index]->ltd_obd;
143 if (!tgt_obd->obd_set_up) {
144 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
148 /* override the sp_me from lov */
149 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
151 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
152 data->ocd_index = index;
155 * Divine LOV knows that OBDs under it are OSCs.
157 imp = tgt_obd->u.cli.cl_import;
160 tgt_obd->obd_no_recov = 0;
161 /* FIXME this is probably supposed to be
162 ptlrpc_set_import_active. Horrible naming. */
163 ptlrpc_activate_import(imp);
166 rc = obd_register_observer(tgt_obd, obd);
168 CERROR("Target %s register_observer error %d\n",
169 obd_uuid2str(tgt_uuid), rc);
174 if (imp->imp_invalid) {
175 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
176 "disabled\n", obd_uuid2str(tgt_uuid));
180 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
181 &lov_osc_uuid, data, NULL);
182 if (rc || !lov->lov_tgts[index]->ltd_exp) {
183 CERROR("Target %s connect error %d\n",
184 obd_uuid2str(tgt_uuid), rc);
188 lov->lov_tgts[index]->ltd_reap = 0;
190 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
191 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
193 if (lov->targets_proc_entry != NULL) {
194 struct proc_dir_entry *osc_symlink;
195 struct obd_device *osc_obd;
197 osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
199 LASSERT(osc_obd != NULL);
200 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
201 LASSERT(osc_obd->obd_type->typ_name != NULL);
203 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
204 lov->targets_proc_entry,
206 osc_obd->obd_type->typ_name,
208 if (osc_symlink == NULL) {
209 CERROR("cannot register LOV target "
210 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
211 obd->obd_type->typ_name, obd->obd_name,
218 static int lov_connect(const struct lu_env *env,
219 struct obd_export **exp, struct obd_device *obd,
220 struct obd_uuid *cluuid, struct obd_connect_data *data,
223 struct lov_obd *lov = &obd->u.lov;
224 struct lov_tgt_desc *tgt;
225 struct lustre_handle conn;
229 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
231 rc = class_connect(&conn, obd, cluuid);
235 *exp = class_conn2export(&conn);
237 /* Why should there ever be more than 1 connect? */
239 LASSERT(lov->lov_connects == 1);
241 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
243 lov->lov_ocd = *data;
245 lov->targets_proc_entry = lprocfs_seq_register("target_obds",
248 if (IS_ERR(lov->targets_proc_entry)) {
249 CERROR("%s: cannot register "
250 "/proc/fs/lustre/%s/%s/target_obds\n",
251 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
252 lov->targets_proc_entry = NULL;
256 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
257 tgt = lov->lov_tgts[i];
258 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
260 /* Flags will be lowest common denominator */
261 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
263 CERROR("%s: lov connect tgt %d failed: %d\n",
264 obd->obd_name, i, rc);
267 /* connect to administrative disabled ost */
268 if (!lov->lov_tgts[i]->ltd_exp)
271 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
272 OBD_NOTIFY_CONNECT, (void *)&i);
274 CERROR("%s error sending notify %d\n",
283 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
285 struct lov_obd *lov = &obd->u.lov;
286 struct obd_device *osc_obd;
290 osc_obd = class_exp2obd(tgt->ltd_exp);
291 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
292 obd->obd_name, osc_obd->obd_name);
294 if (tgt->ltd_active) {
296 lov->desc.ld_active_tgt_count--;
297 tgt->ltd_exp->exp_obd->obd_inactive = 1;
301 /* Pass it on to our clients.
302 * XXX This should be an argument to disconnect,
303 * XXX not a back-door flag on the OBD. Ah well.
305 osc_obd->obd_force = obd->obd_force;
306 osc_obd->obd_fail = obd->obd_fail;
307 osc_obd->obd_no_recov = obd->obd_no_recov;
309 if (lov->targets_proc_entry != NULL)
310 lprocfs_remove_proc_entry(osc_obd->obd_name,
311 lov->targets_proc_entry);
314 obd_register_observer(osc_obd, NULL);
316 rc = obd_disconnect(tgt->ltd_exp);
318 CERROR("Target %s disconnect error %d\n",
319 tgt->ltd_uuid.uuid, rc);
327 static int lov_disconnect(struct obd_export *exp)
329 struct obd_device *obd = class_exp2obd(exp);
330 struct lov_obd *lov = &obd->u.lov;
337 /* Only disconnect the underlying layers on the final disconnect. */
339 if (lov->lov_connects != 0) {
340 /* why should there be more than 1 connect? */
341 CERROR("disconnect #%d\n", lov->lov_connects);
345 /* Let's hold another reference so lov_del_obd doesn't spin through
349 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
350 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
351 /* Disconnection is the last we know about an obd */
352 lov_del_target(obd, i, 0, lov->lov_tgts[i]->ltd_gen);
357 if (lov->targets_proc_entry != NULL)
358 lprocfs_remove(&lov->targets_proc_entry);
361 rc = class_disconnect(exp); /* bz 9811 */
367 * -EINVAL : UUID can't be found in the LOV's target list
368 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
369 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
370 * any >= 0 : is log target index
372 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
373 enum obd_notify_event ev)
375 struct lov_obd *lov = &obd->u.lov;
376 struct lov_tgt_desc *tgt;
377 int index, activate, active;
380 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
381 lov, uuid->uuid, ev);
384 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
385 tgt = lov->lov_tgts[index];
389 * LU-642, initially inactive OSC could miss the obd_connect,
390 * we make up for it here.
392 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
393 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
394 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
396 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
397 &lov_osc_uuid, &lov->lov_ocd, NULL);
402 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
403 index, obd_uuid2str(&tgt->ltd_uuid),
404 tgt->ltd_exp->exp_handle.h_cookie);
405 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
409 if (index == lov->desc.ld_tgt_count)
410 GOTO(out, index = -EINVAL);
412 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
413 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
415 if (lov->lov_tgts[index]->ltd_activate == activate) {
416 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
417 uuid->uuid, activate ? "" : "de");
419 lov->lov_tgts[index]->ltd_activate = activate;
420 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
421 activate ? "" : "de", obd_uuid2str(uuid));
424 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
425 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
427 if (lov->lov_tgts[index]->ltd_active == active) {
428 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
429 uuid->uuid, active ? "" : "in");
432 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
433 obd_uuid2str(uuid), active ? "" : "in");
436 lov->lov_tgts[index]->ltd_active = active;
438 lov->desc.ld_active_tgt_count++;
439 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
441 lov->desc.ld_active_tgt_count--;
442 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
445 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
454 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
455 enum obd_notify_event ev, void *data)
458 struct lov_obd *lov = &obd->u.lov;
461 down_read(&lov->lov_notify_lock);
462 if (!lov->lov_connects) {
463 up_read(&lov->lov_notify_lock);
467 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
468 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
469 struct obd_uuid *uuid;
473 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
474 up_read(&lov->lov_notify_lock);
475 CERROR("unexpected notification of %s %s!\n",
476 watched->obd_type->typ_name,
480 uuid = &watched->u.cli.cl_target_uuid;
482 /* Set OSC as active before notifying the observer, so the
483 * observer can use the OSC normally.
485 rc = lov_set_osc_active(obd, uuid, ev);
487 up_read(&lov->lov_notify_lock);
488 CERROR("event(%d) of %s failed: %d\n", ev,
489 obd_uuid2str(uuid), rc);
492 /* active event should be pass lov target index as data */
496 /* Pass the notification up the chain. */
498 rc = obd_notify_observer(obd, watched, ev, data);
500 /* NULL watched means all osc's in the lov (only for syncs) */
501 /* sync event should be send lov idx as data */
502 struct lov_obd *lov = &obd->u.lov;
506 is_sync = (ev == OBD_NOTIFY_SYNC) ||
507 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
510 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
511 if (!lov->lov_tgts[i])
514 /* don't send sync event if target not
515 * connected/activated */
516 if (is_sync && !lov->lov_tgts[i]->ltd_active)
519 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
522 CERROR("%s: notify %s of %s failed %d\n",
524 obd->obd_observer->obd_name,
525 lov->lov_tgts[i]->ltd_obd->obd_name,
532 up_read(&lov->lov_notify_lock);
536 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
537 __u32 index, int gen, int active)
539 struct lov_obd *lov = &obd->u.lov;
540 struct lov_tgt_desc *tgt;
541 struct obd_device *tgt_obd;
545 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
546 uuidp->uuid, index, gen, active);
549 CERROR("request to add OBD %s with invalid generation: %d\n",
554 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
559 mutex_lock(&lov->lov_lock);
561 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
562 tgt = lov->lov_tgts[index];
563 CERROR("UUID %s already assigned at LOV target index %d\n",
564 obd_uuid2str(&tgt->ltd_uuid), index);
565 mutex_unlock(&lov->lov_lock);
569 if (index >= lov->lov_tgt_size) {
570 /* We need to reallocate the lov target array. */
571 struct lov_tgt_desc **newtgts, **old = NULL;
572 __u32 newsize, oldsize = 0;
574 newsize = max(lov->lov_tgt_size, (__u32)2);
575 while (newsize < index + 1)
576 newsize = newsize << 1;
577 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
578 if (newtgts == NULL) {
579 mutex_unlock(&lov->lov_lock);
583 if (lov->lov_tgt_size) {
584 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
587 oldsize = lov->lov_tgt_size;
590 lov->lov_tgts = newtgts;
591 lov->lov_tgt_size = newsize;
594 OBD_FREE(old, sizeof(*old) * oldsize);
596 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
597 lov->lov_tgts, lov->lov_tgt_size);
602 mutex_unlock(&lov->lov_lock);
606 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
608 mutex_unlock(&lov->lov_lock);
613 tgt->ltd_uuid = *uuidp;
614 tgt->ltd_obd = tgt_obd;
615 /* XXX - add a sanity check on the generation number. */
617 tgt->ltd_index = index;
618 tgt->ltd_activate = active;
619 lov->lov_tgts[index] = tgt;
620 if (index >= lov->desc.ld_tgt_count)
621 lov->desc.ld_tgt_count = index + 1;
623 mutex_unlock(&lov->lov_lock);
625 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
626 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
628 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
630 if (lov->lov_connects == 0) {
631 /* lov_connect hasn't been called yet. We'll do the
632 lov_connect_obd on this target when that fn first runs,
633 because we don't know the connect flags yet. */
639 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
643 /* connect to administrative disabled ost */
647 if (lov->lov_cache != NULL) {
648 rc = obd_set_info_async(NULL, tgt->ltd_exp,
649 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
650 sizeof(struct cl_client_cache), lov->lov_cache,
656 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
657 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
662 CERROR("add failed (%d), deleting %s\n", rc,
663 obd_uuid2str(&tgt->ltd_uuid));
664 lov_del_target(obd, index, 0, 0);
670 /* Schedule a target for deletion */
671 int lov_del_target(struct obd_device *obd, __u32 index,
672 struct obd_uuid *uuidp, int gen)
674 struct lov_obd *lov = &obd->u.lov;
675 int count = lov->desc.ld_tgt_count;
679 if (index >= count) {
680 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
685 /* to make sure there's no ongoing lov_notify() now */
686 down_write(&lov->lov_notify_lock);
689 if (!lov->lov_tgts[index]) {
690 CERROR("LOV target at index %d is not setup.\n", index);
691 GOTO(out, rc = -EINVAL);
694 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
695 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
696 lov_uuid2str(lov, index), index,
697 obd_uuid2str(uuidp));
698 GOTO(out, rc = -EINVAL);
701 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
702 lov_uuid2str(lov, index), index,
703 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
704 lov->lov_tgts[index]->ltd_active);
706 lov->lov_tgts[index]->ltd_reap = 1;
707 lov->lov_death_row++;
708 /* we really delete it from obd_putref */
711 up_write(&lov->lov_notify_lock);
716 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
718 struct obd_device *osc_obd;
721 LASSERT(tgt->ltd_reap);
723 osc_obd = class_exp2obd(tgt->ltd_exp);
725 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
727 osc_obd ? osc_obd->obd_name : "<no obd>");
730 lov_disconnect_obd(obd, tgt);
734 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
735 do it ourselves. And we can't do it from lov_cleanup,
736 because we just lost our only reference to it. */
738 class_manual_cleanup(osc_obd);
741 void lov_fix_desc_stripe_size(__u64 *val)
743 if (*val < LOV_MIN_STRIPE_SIZE) {
745 LCONSOLE_INFO("Increasing default stripe size to "
747 LOV_DESC_STRIPE_SIZE_DEFAULT);
748 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
749 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
750 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
751 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
753 *val, LOV_MIN_STRIPE_SIZE);
757 void lov_fix_desc_stripe_count(__u32 *val)
763 void lov_fix_desc_pattern(__u32 *val)
765 /* from lov_setstripe */
766 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
767 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
772 void lov_fix_desc_qos_maxage(__u32 *val)
775 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
778 void lov_fix_desc(struct lov_desc *desc)
780 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
781 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
782 lov_fix_desc_pattern(&desc->ld_pattern);
783 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
786 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
788 struct lov_desc *desc;
789 struct lov_obd *lov = &obd->u.lov;
791 struct obd_type *type;
796 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
797 CERROR("LOV setup requires a descriptor\n");
801 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
803 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
804 CERROR("descriptor size wrong: %d > %d\n",
805 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
809 if (desc->ld_magic != LOV_DESC_MAGIC) {
810 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
811 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
812 obd->obd_name, desc);
813 lustre_swab_lov_desc(desc);
815 CERROR("%s: Bad lov desc magic: %#x\n",
816 obd->obd_name, desc->ld_magic);
823 desc->ld_active_tgt_count = 0;
825 lov->lov_tgt_size = 0;
827 mutex_init(&lov->lov_lock);
828 atomic_set(&lov->lov_refcount, 0);
829 lov->lov_sp_me = LUSTRE_SP_CLI;
831 init_rwsem(&lov->lov_notify_lock);
833 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
835 HASH_POOLS_BKT_BITS, 0,
838 &pool_hash_operations,
840 INIT_LIST_HEAD(&lov->lov_pool_list);
841 lov->lov_pool_count = 0;
842 rc = lov_ost_pool_init(&lov->lov_packed, 0);
847 obd->obd_vars = lprocfs_lov_obd_vars;
848 /* If this is true then both client (lov) and server
849 * (lod) are on the same node. The lod layer if loaded
850 * first will register the lov proc directory. In that
851 * case obd->obd_type->typ_procroot will be not set.
852 * Instead we use type->typ_procsym as the parent. */
853 type = class_search_type(LUSTRE_LOD_NAME);
854 if (type != NULL && type->typ_procsym != NULL) {
855 obd->obd_proc_entry = lprocfs_seq_register(obd->obd_name,
858 if (IS_ERR(obd->obd_proc_entry)) {
859 rc = PTR_ERR(obd->obd_proc_entry);
860 CERROR("error %d setting up lprocfs for %s\n", rc,
862 obd->obd_proc_entry = NULL;
865 rc = lprocfs_seq_obd_setup(obd);
869 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
870 0444, &lov_proc_target_fops, obd);
872 CWARN("Error adding the target_obd file\n");
874 lov->lov_pool_proc_entry = lprocfs_seq_register("pools",
877 if (IS_ERR(lov->lov_pool_proc_entry)) {
878 rc = PTR_ERR(lov->lov_pool_proc_entry);
879 CERROR("error %d setting up lprocfs for pools\n", rc);
880 lov->lov_pool_proc_entry = NULL;
890 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
893 struct lov_obd *lov = &obd->u.lov;
898 case OBD_CLEANUP_EARLY: {
900 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
901 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
903 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
915 static int lov_cleanup(struct obd_device *obd)
917 struct lov_obd *lov = &obd->u.lov;
918 struct list_head *pos, *tmp;
919 struct pool_desc *pool;
922 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
923 pool = list_entry(pos, struct pool_desc, pool_list);
924 /* free pool structs */
925 CDEBUG(D_INFO, "delete pool %p\n", pool);
926 /* In the function below, .hs_keycmp resolves to
927 * pool_hashkey_keycmp() */
928 /* coverity[overrun-buffer-val] */
929 lov_pool_del(obd, pool->pool_name);
931 cfs_hash_putref(lov->lov_pools_hash_body);
932 lov_ost_pool_free(&lov->lov_packed);
934 lprocfs_obd_cleanup(obd);
938 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
939 if (!lov->lov_tgts[i])
942 /* Inactive targets may never have connected */
943 if (lov->lov_tgts[i]->ltd_active ||
944 atomic_read(&lov->lov_refcount))
945 /* We should never get here - these
946 * should have been removed in the
948 CERROR("%s: lov tgt %d not cleaned! "
949 "deathrow=%d, lovrc=%d\n",
950 obd->obd_name, i, lov->lov_death_row,
951 atomic_read(&lov->lov_refcount));
952 lov_del_target(obd, i, 0, 0);
955 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
957 lov->lov_tgt_size = 0;
962 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
963 __u32 *indexp, int *genp)
965 struct obd_uuid obd_uuid;
970 switch(cmd = lcfg->lcfg_command) {
971 case LCFG_LOV_ADD_OBD:
972 case LCFG_LOV_ADD_INA:
973 case LCFG_LOV_DEL_OBD: {
976 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
977 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
978 GOTO(out, rc = -EINVAL);
980 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
982 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", indexp) != 1)
983 GOTO(out, rc = -EINVAL);
984 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
985 GOTO(out, rc = -EINVAL);
988 if (cmd == LCFG_LOV_ADD_OBD)
989 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
990 else if (cmd == LCFG_LOV_ADD_INA)
991 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
993 rc = lov_del_target(obd, index, &obd_uuid, gen);
997 struct lov_desc *desc = &(obd->u.lov.desc);
1000 GOTO(out, rc = -EINVAL);
1002 rc = class_process_proc_seq_param(PARAM_LOV, obd->obd_vars,
1015 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
1016 GOTO(out, rc = -EINVAL);
1024 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
1025 struct lov_stripe_md **ea, struct obd_trans_info *oti)
1027 struct lov_stripe_md *obj_mdp, *lsm;
1028 struct lov_obd *lov = &exp->exp_obd->u.lov;
1033 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
1034 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
1036 OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
1037 if (obj_mdp == NULL)
1040 ost_idx = src_oa->o_nlink;
1043 GOTO(out, rc = -EINVAL);
1044 if (ost_idx >= lov->desc.ld_tgt_count ||
1045 !lov->lov_tgts[ost_idx])
1046 GOTO(out, rc = -EINVAL);
1048 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1049 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1051 if (lov_oinfo_is_dummy(loi))
1054 if (loi->loi_ost_idx == ost_idx) {
1055 if (ostid_id(&loi->loi_oi) !=
1056 ostid_id(&src_oa->o_oi))
1057 GOTO(out, rc = -EINVAL);
1061 if (i == lsm->lsm_stripe_count)
1062 GOTO(out, rc = -EINVAL);
1064 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1065 src_oa, &obj_mdp, oti);
1067 OBD_FREE(obj_mdp, sizeof(*obj_mdp));
1071 /* the LOV expects oa->o_id to be set to the LOV object id */
1072 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1073 struct obdo *src_oa, struct lov_stripe_md **ea,
1074 struct obd_trans_info *oti)
1076 struct lov_obd *lov;
1080 LASSERT(ea != NULL);
1084 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1085 src_oa->o_flags == OBD_FL_DELORPHAN) {
1086 /* should be used with LOV anymore */
1090 lov = &exp->exp_obd->u.lov;
1091 if (!lov->desc.ld_active_tgt_count)
1094 obd_getref(exp->exp_obd);
1095 /* Recreate a specific object id at the given OST index */
1096 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1097 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1098 rc = lov_recreate(exp, src_oa, ea, oti);
1101 obd_putref(exp->exp_obd);
1105 #define ASSERT_LSM_MAGIC(lsmp) \
1107 LASSERT((lsmp) != NULL); \
1108 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1109 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1110 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1113 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1114 struct obdo *oa, struct lov_stripe_md *lsm,
1115 struct obd_trans_info *oti, struct obd_export *md_exp,
1118 struct lov_request_set *set;
1119 struct obd_info oinfo;
1120 struct lov_request *req;
1121 struct list_head *pos;
1122 struct lov_obd *lov;
1123 int rc = 0, err = 0;
1126 ASSERT_LSM_MAGIC(lsm);
1128 if (!exp || !exp->exp_obd)
1131 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1133 LASSERT(oti->oti_logcookies);
1136 lov = &exp->exp_obd->u.lov;
1137 obd_getref(exp->exp_obd);
1138 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1142 list_for_each(pos, &set->set_list) {
1143 req = list_entry(pos, struct lov_request, rq_link);
1145 if (oa->o_valid & OBD_MD_FLCOOKIE)
1146 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1148 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1149 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1150 err = lov_update_common_set(set, req, err);
1152 CERROR("%s: destroying objid "DOSTID" subobj "
1153 DOSTID" on OST idx %d: rc = %d\n",
1154 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1155 POSTID(&req->rq_oi.oi_oa->o_oi),
1163 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1164 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1166 err = lov_fini_destroy_set(set);
1168 obd_putref(exp->exp_obd);
1169 RETURN(rc ? rc : err);
1172 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1175 struct lov_request_set *lovset = (struct lov_request_set *)data;
1179 /* don't do attribute merge if this aysnc op failed */
1181 atomic_set(&lovset->set_completes, 0);
1182 err = lov_fini_getattr_set(lovset);
1183 RETURN(rc ? rc : err);
1186 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1187 struct ptlrpc_request_set *rqset)
1189 struct lov_request_set *lovset;
1190 struct lov_obd *lov;
1191 struct list_head *pos;
1192 struct lov_request *req;
1197 ASSERT_LSM_MAGIC(oinfo->oi_md);
1199 if (!exp || !exp->exp_obd)
1202 lov = &exp->exp_obd->u.lov;
1204 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1208 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1209 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1210 oinfo->oi_md->lsm_stripe_size);
1212 list_for_each(pos, &lovset->set_list) {
1213 req = list_entry(pos, struct lov_request, rq_link);
1215 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1216 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1217 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1218 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1219 &req->rq_oi, rqset);
1221 CERROR("%s: getattr objid "DOSTID" subobj"
1222 DOSTID" on OST idx %d: rc = %d\n",
1223 exp->exp_obd->obd_name,
1224 POSTID(&oinfo->oi_oa->o_oi),
1225 POSTID(&req->rq_oi.oi_oa->o_oi),
1231 if (!list_empty(&rqset->set_requests)) {
1233 LASSERT (rqset->set_interpret == NULL);
1234 rqset->set_interpret = lov_getattr_interpret;
1235 rqset->set_arg = (void *)lovset;
1240 atomic_set(&lovset->set_completes, 0);
1241 err = lov_fini_getattr_set(lovset);
1242 RETURN(rc ? rc : err);
1245 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1248 struct lov_request_set *lovset = (struct lov_request_set *)data;
1253 atomic_set(&lovset->set_completes, 0);
1254 err = lov_fini_setattr_set(lovset);
1255 RETURN(rc ? rc : err);
1258 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1259 needed. Otherwise, a client is waiting for responses. */
1260 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1261 struct obd_trans_info *oti,
1262 struct ptlrpc_request_set *rqset)
1264 struct lov_request_set *set;
1265 struct lov_request *req;
1266 struct list_head *pos;
1267 struct lov_obd *lov;
1272 ASSERT_LSM_MAGIC(oinfo->oi_md);
1273 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1275 LASSERT(oti->oti_logcookies);
1278 if (!exp || !exp->exp_obd)
1281 lov = &exp->exp_obd->u.lov;
1282 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1286 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1287 POSTID(&oinfo->oi_md->lsm_oi),
1288 oinfo->oi_md->lsm_stripe_count,
1289 oinfo->oi_md->lsm_stripe_size);
1291 list_for_each(pos, &set->set_list) {
1292 req = list_entry(pos, struct lov_request, rq_link);
1294 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1295 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1297 CDEBUG(D_INFO, "objid "DOSTID"[%d] has subobj "DOSTID" at idx"
1298 "%u\n", POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1299 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1301 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1302 &req->rq_oi, oti, rqset);
1304 CERROR("error: setattr objid "DOSTID" subobj"
1305 DOSTID" on OST idx %d: rc = %d\n",
1306 POSTID(&set->set_oi->oi_oa->o_oi),
1307 POSTID(&req->rq_oi.oi_oa->o_oi),
1313 /* If we are not waiting for responses on async requests, return. */
1314 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1317 atomic_set(&set->set_completes, 0);
1318 err = lov_fini_setattr_set(set);
1319 RETURN(rc ? rc : err);
1322 LASSERT(rqset->set_interpret == NULL);
1323 rqset->set_interpret = lov_setattr_interpret;
1324 rqset->set_arg = (void *)set;
1329 static int lov_change_cbdata(struct obd_export *exp,
1330 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1333 struct lov_obd *lov;
1337 ASSERT_LSM_MAGIC(lsm);
1339 if (!exp || !exp->exp_obd)
1342 lov = &exp->exp_obd->u.lov;
1343 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1344 struct lov_stripe_md submd;
1345 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1347 if (lov_oinfo_is_dummy(loi))
1350 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1351 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1355 submd.lsm_oi = loi->loi_oi;
1356 submd.lsm_stripe_count = 0;
1357 rc = obd_change_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1363 /* find any ldlm lock of the inode in lov
1367 static int lov_find_cbdata(struct obd_export *exp,
1368 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1371 struct lov_obd *lov;
1375 ASSERT_LSM_MAGIC(lsm);
1377 if (!exp || !exp->exp_obd)
1380 lov = &exp->exp_obd->u.lov;
1381 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1382 struct lov_stripe_md submd;
1383 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1385 if (lov_oinfo_is_dummy(loi))
1388 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1389 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1393 submd.lsm_oi = loi->loi_oi;
1394 submd.lsm_stripe_count = 0;
1395 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1403 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1405 struct lov_request_set *lovset = (struct lov_request_set *)data;
1410 atomic_set(&lovset->set_completes, 0);
1412 err = lov_fini_statfs_set(lovset);
1413 RETURN(rc ? rc : err);
1416 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1417 __u64 max_age, struct ptlrpc_request_set *rqset)
1419 struct obd_device *obd = class_exp2obd(exp);
1420 struct lov_request_set *set;
1421 struct lov_request *req;
1422 struct list_head *pos;
1423 struct lov_obd *lov;
1427 LASSERT(oinfo != NULL);
1428 LASSERT(oinfo->oi_osfs != NULL);
1431 rc = lov_prep_statfs_set(obd, oinfo, &set);
1435 list_for_each(pos, &set->set_list) {
1436 req = list_entry(pos, struct lov_request, rq_link);
1437 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1438 &req->rq_oi, max_age, rqset);
1443 if (rc || list_empty(&rqset->set_requests)) {
1446 atomic_set(&set->set_completes, 0);
1447 err = lov_fini_statfs_set(set);
1448 RETURN(rc ? rc : err);
1451 LASSERT(rqset->set_interpret == NULL);
1452 rqset->set_interpret = lov_statfs_interpret;
1453 rqset->set_arg = (void *)set;
1457 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1458 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1460 struct ptlrpc_request_set *set = NULL;
1461 struct obd_info oinfo = { { { 0 } } };
1466 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1467 * statfs requests */
1468 set = ptlrpc_prep_set();
1472 oinfo.oi_osfs = osfs;
1473 oinfo.oi_flags = flags;
1474 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1476 rc = ptlrpc_set_wait(set);
1477 ptlrpc_set_destroy(set);
1482 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1483 void *karg, void *uarg)
1485 struct obd_device *obddev = class_exp2obd(exp);
1486 struct lov_obd *lov = &obddev->u.lov;
1487 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1488 struct obd_uuid *uuidp;
1492 case IOC_OBD_STATFS: {
1493 struct obd_ioctl_data *data = karg;
1494 struct obd_device *osc_obd;
1495 struct obd_statfs stat_buf = {0};
1499 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1500 if ((index >= count))
1503 if (!lov->lov_tgts[index])
1504 /* Try again with the next index */
1506 if (!lov->lov_tgts[index]->ltd_active)
1509 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1514 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1515 min((int)data->ioc_plen2,
1516 (int)sizeof(struct obd_uuid))))
1519 flags = uarg ? *(__u32*)uarg : 0;
1520 /* got statfs data */
1521 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1522 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1526 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1527 min((int) data->ioc_plen1,
1528 (int) sizeof(stat_buf))))
1532 case OBD_IOC_LOV_GET_CONFIG: {
1533 struct obd_ioctl_data *data;
1534 struct lov_desc *desc;
1539 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1542 data = (struct obd_ioctl_data *)buf;
1544 if (sizeof(*desc) > data->ioc_inllen1) {
1545 obd_ioctl_freedata(buf, len);
1549 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1550 obd_ioctl_freedata(buf, len);
1554 if (sizeof(__u32) * count > data->ioc_inllen3) {
1555 obd_ioctl_freedata(buf, len);
1559 desc = (struct lov_desc *)data->ioc_inlbuf1;
1560 memcpy(desc, &(lov->desc), sizeof(*desc));
1562 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1563 genp = (__u32 *)data->ioc_inlbuf3;
1564 /* the uuid will be empty for deleted OSTs */
1565 for (i = 0; i < count; i++, uuidp++, genp++) {
1566 if (!lov->lov_tgts[i])
1568 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1569 *genp = lov->lov_tgts[i]->ltd_gen;
1572 if (copy_to_user((void *)uarg, buf, len))
1574 obd_ioctl_freedata(buf, len);
1577 case LL_IOC_LOV_GETSTRIPE:
1578 rc = lov_getstripe(exp, karg, uarg);
1580 case OBD_IOC_QUOTACTL: {
1581 struct if_quotactl *qctl = karg;
1582 struct lov_tgt_desc *tgt = NULL;
1583 struct obd_quotactl *oqctl;
1585 if (qctl->qc_valid == QC_OSTIDX) {
1586 if (count <= qctl->qc_idx)
1589 tgt = lov->lov_tgts[qctl->qc_idx];
1590 if (!tgt || !tgt->ltd_exp)
1592 } else if (qctl->qc_valid == QC_UUID) {
1593 for (i = 0; i < count; i++) {
1594 tgt = lov->lov_tgts[i];
1596 !obd_uuid_equals(&tgt->ltd_uuid,
1600 if (tgt->ltd_exp == NULL)
1612 LASSERT(tgt && tgt->ltd_exp);
1613 OBD_ALLOC_PTR(oqctl);
1617 QCTL_COPY(oqctl, qctl);
1618 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1620 QCTL_COPY(qctl, oqctl);
1621 qctl->qc_valid = QC_OSTIDX;
1622 qctl->obd_uuid = tgt->ltd_uuid;
1624 OBD_FREE_PTR(oqctl);
1633 for (i = 0; i < count; i++) {
1635 struct obd_device *osc_obd;
1637 /* OST was disconnected */
1638 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1641 /* ll_umount_begin() sets force flag but for lov, not
1642 * osc. Let's pass it through */
1643 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1644 osc_obd->obd_force = obddev->obd_force;
1645 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1647 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1650 if (lov->lov_tgts[i]->ltd_active) {
1651 CDEBUG(err == -ENOTTY ?
1652 D_IOCTL : D_WARNING,
1653 "iocontrol OSC %s on OST "
1654 "idx %d cmd %x: err = %d\n",
1655 lov_uuid2str(lov, i),
1672 #define FIEMAP_BUFFER_SIZE 4096
1675 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1676 * call. The local end offset and the device are sent in the first
1677 * fm_extent. This function calculates the stripe number from the index.
1678 * This function returns a stripe_no on which mapping is to be restarted.
1680 * This function returns fm_end_offset which is the in-OST offset at which
1681 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1682 * will re-calculate proper offset in next stripe.
1683 * Note that the first extent is passed to lov_get_info via the value field.
1685 * \param fiemap fiemap request header
1686 * \param lsm striping information for the file
1687 * \param fm_start logical start of mapping
1688 * \param fm_end logical end of mapping
1689 * \param start_stripe starting stripe will be returned in this
1691 obd_size fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1692 struct lov_stripe_md *lsm, obd_size fm_start,
1693 obd_size fm_end, int *start_stripe)
1695 obd_size local_end = fiemap->fm_extents[0].fe_logical;
1696 obd_off lun_start, lun_end;
1697 obd_size fm_end_offset;
1698 int stripe_no = -1, i;
1700 if (fiemap->fm_extent_count == 0 ||
1701 fiemap->fm_extents[0].fe_logical == 0)
1704 /* Find out stripe_no from ost_index saved in the fe_device */
1705 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1706 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1708 if (lov_oinfo_is_dummy(oinfo))
1711 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1717 if (stripe_no == -1)
1720 /* If we have finished mapping on previous device, shift logical
1721 * offset to start of next device */
1722 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1723 &lun_start, &lun_end)) != 0 &&
1724 local_end < lun_end) {
1725 fm_end_offset = local_end;
1726 *start_stripe = stripe_no;
1728 /* This is a special value to indicate that caller should
1729 * calculate offset in next stripe. */
1731 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1734 return fm_end_offset;
1738 * We calculate on which OST the mapping will end. If the length of mapping
1739 * is greater than (stripe_size * stripe_count) then the last_stripe will
1740 * will be one just before start_stripe. Else we check if the mapping
1741 * intersects each OST and find last_stripe.
1742 * This function returns the last_stripe and also sets the stripe_count
1743 * over which the mapping is spread
1745 * \param lsm striping information for the file
1746 * \param fm_start logical start of mapping
1747 * \param fm_end logical end of mapping
1748 * \param start_stripe starting stripe of the mapping
1749 * \param stripe_count the number of stripes across which to map is returned
1751 * \retval last_stripe return the last stripe of the mapping
1753 int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, obd_size fm_start,
1754 obd_size fm_end, int start_stripe,
1758 obd_off obd_start, obd_end;
1761 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1762 last_stripe = (start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1764 *stripe_count = lsm->lsm_stripe_count;
1766 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1767 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1768 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1769 &obd_start, &obd_end)) == 0)
1773 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1780 * Set fe_device and copy extents from local buffer into main return buffer.
1782 * \param fiemap fiemap request header
1783 * \param lcl_fm_ext array of local fiemap extents to be copied
1784 * \param ost_index OST index to be written into the fm_device field for each
1786 * \param ext_count number of extents to be copied
1787 * \param current_extent where to start copying in main extent array
1789 void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1790 struct ll_fiemap_extent *lcl_fm_ext,
1791 int ost_index, unsigned int ext_count,
1797 for (ext = 0; ext < ext_count; ext++) {
1798 lcl_fm_ext[ext].fe_device = ost_index;
1799 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1802 /* Copy fm_extent's from fm_local to return buffer */
1803 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1804 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1808 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1809 * This also handles the restarting of FIEMAP calls in case mapping overflows
1810 * the available number of extents in single call.
1812 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1813 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1815 struct ll_fiemap_info_key *fm_key = key;
1816 struct ll_user_fiemap *fiemap = val;
1817 struct ll_user_fiemap *fm_local = NULL;
1818 struct ll_fiemap_extent *lcl_fm_ext;
1820 unsigned int get_num_extents = 0;
1821 int ost_index = 0, actual_start_stripe, start_stripe;
1822 obd_size fm_start, fm_end, fm_length, fm_end_offset;
1824 int current_extent = 0, rc = 0, i;
1825 int ost_eof = 0; /* EOF for object */
1826 int ost_done = 0; /* done with required mapping for this OST? */
1828 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1829 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1831 if (!lsm_has_objects(lsm)) {
1832 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1833 fm_key->oa.o_size)) {
1834 /* released file, return a minimal FIEMAP if
1835 * request fits in file-size.
1837 fiemap->fm_mapped_extents = 1;
1838 fiemap->fm_extents[0].fe_logical =
1839 fm_key->fiemap.fm_start;
1840 if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1842 fiemap->fm_extents[0].fe_length =
1843 fm_key->fiemap.fm_length;
1845 fiemap->fm_extents[0].fe_length =
1847 fm_key->fiemap.fm_start;
1848 fiemap->fm_extents[0].fe_flags |=
1849 (FIEMAP_EXTENT_UNKNOWN |
1850 FIEMAP_EXTENT_LAST);
1855 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1856 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1858 OBD_ALLOC_LARGE(fm_local, buffer_size);
1859 if (fm_local == NULL)
1860 GOTO(out, rc = -ENOMEM);
1861 lcl_fm_ext = &fm_local->fm_extents[0];
1863 count_local = fiemap_size_to_count(buffer_size);
1865 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1866 fm_start = fiemap->fm_start;
1867 fm_length = fiemap->fm_length;
1868 /* Calculate start stripe, last stripe and length of mapping */
1869 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1870 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1871 fm_start + fm_length - 1);
1872 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1873 if (fm_end > fm_key->oa.o_size)
1874 fm_end = fm_key->oa.o_size;
1876 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1877 actual_start_stripe, &stripe_count);
1879 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1880 fm_end, &start_stripe);
1881 if (fm_end_offset == -EINVAL)
1882 GOTO(out, rc = -EINVAL);
1884 if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1885 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1886 if (fiemap->fm_extent_count == 0) {
1887 get_num_extents = 1;
1890 /* Check each stripe */
1891 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1892 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1893 obd_size req_fm_len; /* Stores length of required mapping */
1894 obd_size len_mapped_single_call;
1895 obd_off lun_start, lun_end, obd_object_end;
1896 unsigned int ext_count;
1898 cur_stripe_wrap = cur_stripe;
1900 /* Find out range of mapping on this stripe */
1901 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1902 &lun_start, &obd_object_end)) == 0)
1905 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe]))
1906 GOTO(out, rc = -EIO);
1908 /* If this is a continuation FIEMAP call and we are on
1909 * starting stripe then lun_start needs to be set to
1911 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1912 lun_start = fm_end_offset;
1914 if (fm_length != ~0ULL) {
1915 /* Handle fm_start + fm_length overflow */
1916 if (fm_start + fm_length < fm_start)
1917 fm_length = ~0ULL - fm_start;
1918 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1924 if (lun_start == lun_end)
1927 req_fm_len = obd_object_end - lun_start;
1928 fm_local->fm_length = 0;
1929 len_mapped_single_call = 0;
1931 /* If the output buffer is very large and the objects have many
1932 * extents we may need to loop on a single OST repeatedly */
1936 if (get_num_extents == 0) {
1937 /* Don't get too many extents. */
1938 if (current_extent + count_local >
1939 fiemap->fm_extent_count)
1940 count_local = fiemap->fm_extent_count -
1944 lun_start += len_mapped_single_call;
1945 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1946 req_fm_len = fm_local->fm_length;
1947 fm_local->fm_extent_count = count_local;
1948 fm_local->fm_mapped_extents = 0;
1949 fm_local->fm_flags = fiemap->fm_flags;
1951 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1952 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1954 if (ost_index < 0 || ost_index >=lov->desc.ld_tgt_count)
1955 GOTO(out, rc = -EINVAL);
1957 /* If OST is inactive, return extent with UNKNOWN flag */
1958 if (!lov->lov_tgts[ost_index]->ltd_active) {
1959 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1960 fm_local->fm_mapped_extents = 1;
1962 lcl_fm_ext[0].fe_logical = lun_start;
1963 lcl_fm_ext[0].fe_length = obd_object_end -
1965 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1970 fm_local->fm_start = lun_start;
1971 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1972 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1973 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1974 rc = obd_get_info(NULL,
1975 lov->lov_tgts[ost_index]->ltd_exp,
1976 keylen, key, vallen, fm_local, lsm);
1981 ext_count = fm_local->fm_mapped_extents;
1982 if (ext_count == 0) {
1984 /* If last stripe has hole at the end,
1985 * then we need to return */
1986 if (cur_stripe_wrap == last_stripe) {
1987 fiemap->fm_mapped_extents = 0;
1993 /* If we just need num of extents then go to next device */
1994 if (get_num_extents) {
1995 current_extent += ext_count;
1999 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
2000 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
2002 /* Have we finished mapping on this device? */
2003 if (req_fm_len <= len_mapped_single_call)
2006 /* Clear the EXTENT_LAST flag which can be present on
2008 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
2009 lcl_fm_ext[ext_count - 1].fe_flags &=
2010 ~FIEMAP_EXTENT_LAST;
2012 curr_loc = lov_stripe_size(lsm,
2013 lcl_fm_ext[ext_count - 1].fe_logical+
2014 lcl_fm_ext[ext_count - 1].fe_length,
2016 if (curr_loc >= fm_key->oa.o_size)
2019 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
2020 ost_index, ext_count,
2023 current_extent += ext_count;
2025 /* Ran out of available extents? */
2026 if (current_extent >= fiemap->fm_extent_count)
2028 } while (ost_done == 0 && ost_eof == 0);
2030 if (cur_stripe_wrap == last_stripe)
2035 /* Indicate that we are returning device offsets unless file just has
2037 if (lsm->lsm_stripe_count > 1)
2038 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
2040 if (get_num_extents)
2041 goto skip_last_device_calc;
2043 /* Check if we have reached the last stripe and whether mapping for that
2044 * stripe is done. */
2045 if (cur_stripe_wrap == last_stripe) {
2046 if (ost_done || ost_eof)
2047 fiemap->fm_extents[current_extent - 1].fe_flags |=
2051 skip_last_device_calc:
2052 fiemap->fm_mapped_extents = current_extent;
2056 OBD_FREE_LARGE(fm_local, buffer_size);
2060 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
2061 __u32 keylen, void *key, __u32 *vallen, void *val,
2062 struct lov_stripe_md *lsm)
2064 struct obd_device *obddev = class_exp2obd(exp);
2065 struct lov_obd *lov = &obddev->u.lov;
2069 if (!vallen || !val)
2074 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2077 struct ldlm_lock *lock;
2079 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2080 struct lov_oinfo *loi;
2081 __u32 *stripe = val;
2083 if (*vallen < sizeof(*stripe))
2084 GOTO(out, rc = -EFAULT);
2085 *vallen = sizeof(*stripe);
2087 /* XXX This is another one of those bits that will need to
2088 * change if we ever actually support nested LOVs. It uses
2089 * the lock's export to find out which stripe it is. */
2090 /* XXX - it's assumed all the locks for deleted OSTs have
2091 * been cancelled. Also, the export for deleted OSTs will
2092 * be NULL and won't match the lock's export. */
2093 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2094 loi = lsm->lsm_oinfo[i];
2095 if (lov_oinfo_is_dummy(loi))
2098 if (!lov->lov_tgts[loi->loi_ost_idx])
2100 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2101 data->lock->l_conn_export &&
2102 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2107 LDLM_ERROR(data->lock, "lock on inode without such object");
2108 dump_lsm(D_ERROR, lsm);
2109 GOTO(out, rc = -ENXIO);
2110 } else if (KEY_IS(KEY_LAST_ID)) {
2111 struct obd_id_info *info = val;
2112 __u32 size = sizeof(obd_id);
2113 struct lov_tgt_desc *tgt;
2115 LASSERT(*vallen == sizeof(struct obd_id_info));
2116 tgt = lov->lov_tgts[info->idx];
2118 if (!tgt || !tgt->ltd_active)
2119 GOTO(out, rc = -ESRCH);
2121 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2122 &size, info->data, NULL);
2124 } else if (KEY_IS(KEY_LOVDESC)) {
2125 struct lov_desc *desc_ret = val;
2126 *desc_ret = lov->desc;
2129 } else if (KEY_IS(KEY_FIEMAP)) {
2130 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2132 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2133 struct lov_tgt_desc *tgt;
2134 __u64 ost_idx = *((__u64*)val);
2136 LASSERT(*vallen == sizeof(__u64));
2137 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2138 tgt = lov->lov_tgts[ost_idx];
2140 if (!tgt || !tgt->ltd_exp)
2141 GOTO(out, rc = -ESRCH);
2143 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2145 } else if (KEY_IS(KEY_TGT_COUNT)) {
2146 *((int *)val) = lov->desc.ld_tgt_count;
2157 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2158 obd_count keylen, void *key, obd_count vallen,
2159 void *val, struct ptlrpc_request_set *set)
2161 struct obd_device *obddev = class_exp2obd(exp);
2162 struct lov_obd *lov = &obddev->u.lov;
2165 struct lov_tgt_desc *tgt;
2166 unsigned incr, check_uuid,
2167 do_inactive, no_set;
2168 unsigned next_id = 0, mds_con = 0, capa = 0;
2171 incr = check_uuid = do_inactive = no_set = 0;
2174 set = ptlrpc_prep_set();
2180 count = lov->desc.ld_tgt_count;
2182 if (KEY_IS(KEY_NEXT_ID)) {
2183 count = vallen / sizeof(struct obd_id_info);
2184 vallen = sizeof(obd_id);
2185 incr = sizeof(struct obd_id_info);
2188 } else if (KEY_IS(KEY_CHECKSUM)) {
2190 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2191 /* use defaults: do_inactive = incr = 0; */
2192 } else if (KEY_IS(KEY_MDS_CONN)) {
2194 } else if (KEY_IS(KEY_CAPA_KEY)) {
2196 } else if (KEY_IS(KEY_CACHE_SET)) {
2197 LASSERT(lov->lov_cache == NULL);
2198 lov->lov_cache = val;
2202 for (i = 0; i < count; i++, val = (char *)val + incr) {
2204 tgt = lov->lov_tgts[((struct obd_id_info*)val)->idx];
2206 tgt = lov->lov_tgts[i];
2208 /* OST was disconnected */
2209 if (!tgt || !tgt->ltd_exp)
2212 /* OST is inactive and we don't want inactive OSCs */
2213 if (!tgt->ltd_active && !do_inactive)
2217 struct mds_group_info *mgi;
2219 LASSERT(vallen == sizeof(*mgi));
2220 mgi = (struct mds_group_info *)val;
2222 /* Only want a specific OSC */
2223 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2227 err = obd_set_info_async(env, tgt->ltd_exp,
2228 keylen, key, sizeof(int),
2230 } else if (next_id) {
2231 err = obd_set_info_async(env, tgt->ltd_exp,
2232 keylen, key, vallen,
2233 ((struct obd_id_info*)val)->data, set);
2235 struct mds_capa_info *info = (struct mds_capa_info*)val;
2237 LASSERT(vallen == sizeof(*info));
2239 /* Only want a specific OSC */
2241 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2244 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2245 key, sizeof(*info->capa),
2248 /* Only want a specific OSC */
2250 !obd_uuid_equals(val, &tgt->ltd_uuid))
2253 err = obd_set_info_async(env, tgt->ltd_exp,
2254 keylen, key, vallen, val, set);
2263 err = ptlrpc_set_wait(set);
2266 ptlrpc_set_destroy(set);
2271 void lov_stripe_lock(struct lov_stripe_md *md)
2273 LASSERT(md->lsm_lock_owner != current_pid());
2274 spin_lock(&md->lsm_lock);
2275 LASSERT(md->lsm_lock_owner == 0);
2276 md->lsm_lock_owner = current_pid();
2279 void lov_stripe_unlock(struct lov_stripe_md *md)
2281 LASSERT(md->lsm_lock_owner == current_pid());
2282 md->lsm_lock_owner = 0;
2283 spin_unlock(&md->lsm_lock);
2286 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2287 struct obd_quotactl *oqctl)
2289 struct lov_obd *lov = &obd->u.lov;
2290 struct lov_tgt_desc *tgt;
2292 __u64 bhardlimit = 0;
2296 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2297 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2298 oqctl->qc_cmd != Q_GETOQUOTA &&
2299 oqctl->qc_cmd != Q_INITQUOTA &&
2300 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2301 oqctl->qc_cmd != Q_FINVALIDATE) {
2302 CERROR("%s: bad quota opc %x for lov obd\n",
2303 obd->obd_name, oqctl->qc_cmd);
2309 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2312 tgt = lov->lov_tgts[i];
2317 if (!tgt->ltd_active || tgt->ltd_reap) {
2318 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2319 lov->lov_tgts[i]->ltd_activate) {
2321 CERROR("ost %d is inactive\n", i);
2323 CDEBUG(D_HA, "ost %d is inactive\n", i);
2328 err = obd_quotactl(tgt->ltd_exp, oqctl);
2330 if (tgt->ltd_active && !rc)
2335 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2336 curspace += oqctl->qc_dqblk.dqb_curspace;
2337 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2342 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2343 oqctl->qc_dqblk.dqb_curspace = curspace;
2344 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2349 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2350 struct obd_quotactl *oqctl)
2352 struct lov_obd *lov = &obd->u.lov;
2358 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2359 if (!lov->lov_tgts[i])
2362 /* Skip quota check on the administratively disabled OSTs. */
2363 if (!lov->lov_tgts[i]->ltd_activate) {
2364 CWARN("lov idx %d was administratively disabled, "
2365 "skip quotacheck on it.\n", i);
2369 if (!lov->lov_tgts[i]->ltd_active) {
2370 CERROR("lov idx %d inactive\n", i);
2376 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2379 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2382 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2393 static struct obd_ops lov_obd_ops = {
2394 .o_owner = THIS_MODULE,
2395 .o_setup = lov_setup,
2396 .o_precleanup = lov_precleanup,
2397 .o_cleanup = lov_cleanup,
2398 .o_connect = lov_connect,
2399 .o_disconnect = lov_disconnect,
2400 .o_statfs = lov_statfs,
2401 .o_statfs_async = lov_statfs_async,
2402 .o_packmd = lov_packmd,
2403 .o_unpackmd = lov_unpackmd,
2404 .o_create = lov_create,
2405 .o_destroy = lov_destroy,
2406 .o_getattr_async = lov_getattr_async,
2407 .o_setattr_async = lov_setattr_async,
2408 .o_change_cbdata = lov_change_cbdata,
2409 .o_find_cbdata = lov_find_cbdata,
2410 .o_iocontrol = lov_iocontrol,
2411 .o_get_info = lov_get_info,
2412 .o_set_info_async = lov_set_info_async,
2413 .o_notify = lov_notify,
2414 .o_pool_new = lov_pool_new,
2415 .o_pool_rem = lov_pool_remove,
2416 .o_pool_add = lov_pool_add,
2417 .o_pool_del = lov_pool_del,
2418 .o_getref = lov_getref,
2419 .o_putref = lov_putref,
2420 .o_quotactl = lov_quotactl,
2421 .o_quotacheck = lov_quotacheck,
2424 struct kmem_cache *lov_oinfo_slab;
2426 extern struct lu_kmem_descr lov_caches[];
2428 int __init lov_init(void)
2430 bool enable_proc = true;
2431 struct obd_type *type;
2435 /* print an address of _any_ initialized kernel symbol from this
2436 * module, to allow debugging with gdb that doesn't support data
2437 * symbols from modules.*/
2438 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2440 rc = lu_kmem_init(lov_caches);
2444 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2445 sizeof(struct lov_oinfo), 0,
2446 SLAB_HWCACHE_ALIGN, NULL);
2447 if (lov_oinfo_slab == NULL) {
2448 lu_kmem_fini(lov_caches);
2452 type = class_search_type(LUSTRE_LOD_NAME);
2453 if (type != NULL && type->typ_procsym != NULL)
2454 enable_proc = false;
2456 rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
2457 #ifndef HAVE_ONLY_PROCFS_SEQ
2460 LUSTRE_LOV_NAME, &lov_device_type);
2463 kmem_cache_destroy(lov_oinfo_slab);
2464 lu_kmem_fini(lov_caches);
2471 static void /*__exit*/ lov_exit(void)
2473 class_unregister_type(LUSTRE_LOV_NAME);
2474 kmem_cache_destroy(lov_oinfo_slab);
2475 lu_kmem_fini(lov_caches);
2478 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2479 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2480 MODULE_LICENSE("GPL");
2482 cfs_module(lov, LUSTRE_VERSION_STRING, lov_init, lov_exit);