4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/lov/lov_obd.c
34 * Author: Phil Schwan <phil@clusterfs.com>
35 * Author: Peter Braam <braam@clusterfs.com>
36 * Author: Mike Shaver <shaver@clusterfs.com>
37 * Author: Nathan Rutman <nathan@clusterfs.com>
40 #define DEBUG_SUBSYSTEM S_LOV
41 #include <libcfs/libcfs.h>
43 #include <cl_object.h>
44 #include <lustre_dlm.h>
45 #include <lustre_fid.h>
46 #include <uapi/linux/lustre/lustre_ioctl.h>
47 #include <lustre_lib.h>
48 #include <lustre_mds.h>
49 #include <lustre_net.h>
50 #include <uapi/linux/lustre/lustre_param.h>
51 #include <lustre_swab.h>
52 #include <lprocfs_status.h>
53 #include <obd_class.h>
54 #include <obd_support.h>
56 #include "lov_internal.h"
58 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
59 Any function that expects lov_tgts to remain stationary must take a ref. */
60 void lov_tgts_getref(struct obd_device *obd)
62 struct lov_obd *lov = &obd->u.lov;
64 /* nobody gets through here until lov_putref is done */
65 mutex_lock(&lov->lov_lock);
66 atomic_inc(&lov->lov_refcount);
67 mutex_unlock(&lov->lov_lock);
71 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
73 void lov_tgts_putref(struct obd_device *obd)
75 struct lov_obd *lov = &obd->u.lov;
77 mutex_lock(&lov->lov_lock);
78 /* ok to dec to 0 more than once -- ltd_exp's will be null */
79 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
80 struct list_head kill = LIST_HEAD_INIT(kill);
81 struct lov_tgt_desc *tgt, *n;
84 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
86 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
87 tgt = lov->lov_tgts[i];
89 if (!tgt || !tgt->ltd_reap)
91 list_add(&tgt->ltd_kill, &kill);
92 /* XXX - right now there is a dependency on ld_tgt_count
93 * being the maximum tgt index for computing the
94 * mds_max_easize. So we can't shrink it. */
95 lov_ost_pool_remove(&lov->lov_packed, i);
96 lov->lov_tgts[i] = NULL;
99 mutex_unlock(&lov->lov_lock);
101 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
102 list_del(&tgt->ltd_kill);
104 __lov_del_obd(obd, tgt);
107 mutex_unlock(&lov->lov_lock);
111 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
112 enum obd_notify_event ev);
113 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
114 enum obd_notify_event ev);
116 int lov_connect_osc(struct obd_device *obd, u32 index, int activate,
117 struct obd_connect_data *data)
119 struct lov_obd *lov = &obd->u.lov;
120 struct obd_uuid *tgt_uuid;
121 struct obd_device *tgt_obd;
122 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
123 struct obd_import *imp;
127 if (lov->lov_tgts[index] == NULL)
130 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
131 tgt_obd = lov->lov_tgts[index]->ltd_obd;
133 if (!tgt_obd->obd_set_up) {
134 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
138 /* override the sp_me from lov */
139 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
141 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
142 data->ocd_index = index;
145 * Divine LOV knows that OBDs under it are OSCs.
147 imp = tgt_obd->u.cli.cl_import;
150 tgt_obd->obd_no_recov = 0;
151 /* FIXME this is probably supposed to be
152 ptlrpc_set_import_active. Horrible naming. */
153 ptlrpc_activate_import(imp, false);
156 rc = obd_register_observer(tgt_obd, obd);
158 CERROR("Target %s register_observer error %d\n",
159 obd_uuid2str(tgt_uuid), rc);
163 if (imp->imp_invalid) {
164 CDEBUG(D_CONFIG, "%s: not connecting - administratively disabled\n",
165 obd_uuid2str(tgt_uuid));
169 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
170 &lov_osc_uuid, data, lov->lov_cache);
171 if (rc || !lov->lov_tgts[index]->ltd_exp) {
172 CERROR("Target %s connect error %d\n",
173 obd_uuid2str(tgt_uuid), rc);
177 lov->lov_tgts[index]->ltd_reap = 0;
179 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
180 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
182 if (lov->lov_tgts_kobj) {
183 /* Even if we failed, that's ok */
184 rc = sysfs_create_link(lov->lov_tgts_kobj,
185 &tgt_obd->obd_kset.kobj,
188 CERROR("%s: can't register LOV target /sys/fs/lustre/%s/%s/target_obds/%s : rc = %d\n",
189 obd->obd_name, obd->obd_type->typ_name,
191 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_name,
198 static int lov_connect(const struct lu_env *env,
199 struct obd_export **exp, struct obd_device *obd,
200 struct obd_uuid *cluuid, struct obd_connect_data *data,
203 struct lov_obd *lov = &obd->u.lov;
204 struct lov_tgt_desc *tgt;
205 struct lustre_handle conn;
209 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
211 rc = class_connect(&conn, obd, cluuid);
215 *exp = class_conn2export(&conn);
217 /* Why should there ever be more than 1 connect? */
219 LASSERT(lov->lov_connects == 1);
221 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
223 lov->lov_ocd = *data;
225 lov_tgts_getref(obd);
228 lov->lov_cache = localdata;
229 cl_cache_incref(lov->lov_cache);
232 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
233 tgt = lov->lov_tgts[i];
234 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
236 /* Flags will be lowest common denominator */
237 rc = lov_connect_osc(obd, i, tgt->ltd_activate, &lov->lov_ocd);
239 CERROR("%s: lov connect tgt %d failed: %d\n",
240 obd->obd_name, i, rc);
243 /* connect to administrative disabled ost */
244 if (!lov->lov_tgts[i]->ltd_exp)
247 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
250 CERROR("%s error sending notify %d\n",
255 lov_tgts_putref(obd);
260 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
262 struct lov_obd *lov = &obd->u.lov;
263 struct obd_device *osc_obd;
267 osc_obd = class_exp2obd(tgt->ltd_exp);
268 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
269 obd->obd_name, osc_obd->obd_name);
271 if (tgt->ltd_active) {
273 lov->desc.ld_active_tgt_count--;
274 tgt->ltd_exp->exp_obd->obd_inactive = 1;
278 if (lov->lov_tgts_kobj)
279 sysfs_remove_link(lov->lov_tgts_kobj,
282 /* Pass it on to our clients.
283 * XXX This should be an argument to disconnect,
284 * XXX not a back-door flag on the OBD. Ah well.
286 osc_obd->obd_force = obd->obd_force;
287 osc_obd->obd_fail = obd->obd_fail;
288 osc_obd->obd_no_recov = obd->obd_no_recov;
290 if (lov->targets_proc_entry != NULL)
291 lprocfs_remove_proc_entry(osc_obd->obd_name,
292 lov->targets_proc_entry);
295 obd_register_observer(osc_obd, NULL);
297 rc = obd_disconnect(tgt->ltd_exp);
299 CERROR("Target %s disconnect error %d\n",
300 tgt->ltd_uuid.uuid, rc);
308 static int lov_disconnect(struct obd_export *exp)
310 struct obd_device *obd = class_exp2obd(exp);
311 struct lov_obd *lov = &obd->u.lov;
319 /* Only disconnect the underlying layers on the final disconnect. */
321 if (lov->lov_connects != 0) {
322 /* why should there be more than 1 connect? */
323 CWARN("%s: unexpected disconnect #%d\n",
324 obd->obd_name, lov->lov_connects);
328 /* hold another ref so lov_del_obd() doesn't spin in putref each time */
329 lov_tgts_getref(obd);
331 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
332 if (lov->lov_tgts[index] && lov->lov_tgts[index]->ltd_exp) {
333 /* Disconnection is the last we know about an OBD */
334 lov_del_target(obd, index, NULL,
335 lov->lov_tgts[index]->ltd_gen);
338 lov_tgts_putref(obd);
341 rc = class_disconnect(exp); /* bz 9811 */
347 * -EINVAL : UUID can't be found in the LOV's target list
348 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
349 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
350 * any >= 0 : is log target index
352 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
353 enum obd_notify_event ev)
355 struct lov_obd *lov = &obd->u.lov;
356 struct lov_tgt_desc *tgt;
357 int index, activate, active;
360 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
361 lov, uuid->uuid, ev);
363 lov_tgts_getref(obd);
364 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
365 tgt = lov->lov_tgts[index];
368 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
372 if (index == lov->desc.ld_tgt_count)
373 GOTO(out, index = -EINVAL);
375 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
376 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
379 * LU-642, initially inactive OSC could miss the obd_connect,
380 * we make up for it here.
382 if (activate && !tgt->ltd_exp) {
384 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
386 rc = obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
387 &lov_osc_uuid, &lov->lov_ocd,
389 if (rc || tgt->ltd_exp == NULL)
390 GOTO(out, index = rc);
393 if (lov->lov_tgts[index]->ltd_activate == activate) {
394 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
395 uuid->uuid, activate ? "" : "de");
397 lov->lov_tgts[index]->ltd_activate = activate;
398 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
399 activate ? "" : "de", obd_uuid2str(uuid));
402 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
403 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
405 if (lov->lov_tgts[index]->ltd_active == active) {
406 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
407 uuid->uuid, active ? "" : "in");
410 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
411 obd_uuid2str(uuid), active ? "" : "in");
414 lov->lov_tgts[index]->ltd_active = active;
416 lov->desc.ld_active_tgt_count++;
417 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
419 lov->desc.ld_active_tgt_count--;
420 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
423 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
428 CDEBUG(D_INFO, "%s: lov idx %d conn %llx\n", obd_uuid2str(uuid),
429 index, tgt->ltd_exp->exp_handle.h_cookie);
432 lov_tgts_putref(obd);
436 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
437 enum obd_notify_event ev)
440 struct lov_obd *lov = &obd->u.lov;
443 down_read(&lov->lov_notify_lock);
444 if (!lov->lov_connects)
445 GOTO(out_notify_lock, rc = 0);
447 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
448 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
449 struct obd_uuid *uuid;
453 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
454 CERROR("unexpected notification of %s %s\n",
455 watched->obd_type->typ_name, watched->obd_name);
456 GOTO(out_notify_lock, rc = -EINVAL);
459 uuid = &watched->u.cli.cl_target_uuid;
461 /* Set OSC as active before notifying the observer, so the
462 * observer can use the OSC normally.
464 rc = lov_set_osc_active(obd, uuid, ev);
466 CERROR("%s: event %d failed: rc = %d\n", obd->obd_name,
468 GOTO(out_notify_lock, rc);
472 /* Pass the notification up the chain. */
473 rc = obd_notify_observer(obd, watched, ev);
476 up_read(&lov->lov_notify_lock);
481 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
482 u32 index, int gen, int active)
484 struct lov_obd *lov = &obd->u.lov;
485 struct lov_tgt_desc *tgt;
486 struct obd_device *tgt_obd;
490 CDEBUG(D_CONFIG, "uuid:%s idx:%u gen:%d active:%d\n",
491 uuidp->uuid, index, gen, active);
494 CERROR("%s: request to add '%s' with invalid generation: %d\n",
495 obd->obd_name, uuidp->uuid, gen);
499 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, &obd->obd_uuid);
503 mutex_lock(&lov->lov_lock);
505 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
506 tgt = lov->lov_tgts[index];
508 CERROR("%s: UUID %s already assigned at index %d: rc = %d\n",
509 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), index, rc);
510 mutex_unlock(&lov->lov_lock);
514 if (index >= lov->lov_tgt_size) {
515 /* We need to reallocate the lov target array. */
516 struct lov_tgt_desc **newtgts, **old = NULL;
517 __u32 newsize, oldsize = 0;
519 newsize = max(lov->lov_tgt_size, (__u32)2);
520 while (newsize < index + 1)
521 newsize = newsize << 1;
522 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
523 if (newtgts == NULL) {
524 mutex_unlock(&lov->lov_lock);
528 if (lov->lov_tgt_size) {
529 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
532 oldsize = lov->lov_tgt_size;
535 lov->lov_tgts = newtgts;
536 lov->lov_tgt_size = newsize;
539 OBD_FREE(old, sizeof(*old) * oldsize);
541 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
542 lov->lov_tgts, lov->lov_tgt_size);
547 mutex_unlock(&lov->lov_lock);
551 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
553 mutex_unlock(&lov->lov_lock);
558 tgt->ltd_uuid = *uuidp;
559 tgt->ltd_obd = tgt_obd;
560 /* XXX - add a sanity check on the generation number. */
562 tgt->ltd_index = index;
563 tgt->ltd_activate = active;
564 lov->lov_tgts[index] = tgt;
565 if (index >= lov->desc.ld_tgt_count)
566 lov->desc.ld_tgt_count = index + 1;
568 mutex_unlock(&lov->lov_lock);
570 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
571 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
573 if (lov->lov_connects == 0) {
574 /* lov_connect hasn't been called yet. We'll do the
575 lov_connect_osc on this target when that fn first runs,
576 because we don't know the connect flags yet. */
580 lov_tgts_getref(obd);
582 rc = lov_connect_osc(obd, index, active, &lov->lov_ocd);
586 /* connect to administrative disabled ost */
590 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
591 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
595 CERROR("%s: add failed, deleting %s: rc = %d\n",
596 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), rc);
597 lov_del_target(obd, index, NULL, 0);
599 lov_tgts_putref(obd);
603 /* Schedule a target for deletion */
604 int lov_del_target(struct obd_device *obd, u32 index,
605 struct obd_uuid *uuidp, int gen)
607 struct lov_obd *lov = &obd->u.lov;
608 int count = lov->desc.ld_tgt_count;
612 if (index >= count) {
613 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
618 /* to make sure there's no ongoing lov_notify() now */
619 down_write(&lov->lov_notify_lock);
620 lov_tgts_getref(obd);
622 if (!lov->lov_tgts[index]) {
623 CERROR("LOV target at index %d is not setup.\n", index);
624 GOTO(out, rc = -EINVAL);
627 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
628 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
629 lov_uuid2str(lov, index), index,
630 obd_uuid2str(uuidp));
631 GOTO(out, rc = -EINVAL);
634 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
635 lov_uuid2str(lov, index), index,
636 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
637 lov->lov_tgts[index]->ltd_active);
639 lov->lov_tgts[index]->ltd_reap = 1;
640 lov->lov_death_row++;
641 /* we really delete it from lov_tgts_putref() */
643 lov_tgts_putref(obd);
644 up_write(&lov->lov_notify_lock);
649 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
651 struct obd_device *osc_obd;
654 LASSERT(tgt->ltd_reap);
656 osc_obd = class_exp2obd(tgt->ltd_exp);
658 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
660 osc_obd ? osc_obd->obd_name : "<no obd>");
663 lov_disconnect_obd(obd, tgt);
667 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
668 do it ourselves. And we can't do it from lov_cleanup,
669 because we just lost our only reference to it. */
671 class_manual_cleanup(osc_obd);
674 void lov_fix_desc_stripe_size(__u64 *val)
676 if (*val < LOV_MIN_STRIPE_SIZE) {
678 LCONSOLE_INFO("Increasing default stripe size to "
680 LOV_DESC_STRIPE_SIZE_DEFAULT);
681 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
682 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
683 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
684 LCONSOLE_WARN("Changing default stripe size to %llu (a "
686 *val, LOV_MIN_STRIPE_SIZE);
690 void lov_fix_desc_stripe_count(__u32 *val)
696 void lov_fix_desc_pattern(__u32 *val)
698 /* from lov_setstripe */
699 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
700 LCONSOLE_WARN("lov: Unknown stripe pattern: %#x\n", *val);
705 void lov_fix_desc_qos_maxage(__u32 *val)
708 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
711 void lov_fix_desc(struct lov_desc *desc)
713 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
714 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
715 lov_fix_desc_pattern(&desc->ld_pattern);
716 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
719 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
721 struct lov_desc *desc;
722 struct lov_obd *lov = &obd->u.lov;
726 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
727 CERROR("LOV setup requires a descriptor\n");
731 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
733 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
734 CERROR("descriptor size wrong: %d > %d\n",
735 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
739 if (desc->ld_magic != LOV_DESC_MAGIC) {
740 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
741 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
742 obd->obd_name, desc);
743 lustre_swab_lov_desc(desc);
745 CERROR("%s: Bad lov desc magic: %#x\n",
746 obd->obd_name, desc->ld_magic);
753 desc->ld_active_tgt_count = 0;
755 lov->lov_tgt_size = 0;
757 mutex_init(&lov->lov_lock);
758 atomic_set(&lov->lov_refcount, 0);
759 lov->lov_sp_me = LUSTRE_SP_CLI;
761 init_rwsem(&lov->lov_notify_lock);
763 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
765 HASH_POOLS_BKT_BITS, 0,
768 &pool_hash_operations,
770 INIT_LIST_HEAD(&lov->lov_pool_list);
771 lov->lov_pool_count = 0;
772 rc = lov_ost_pool_init(&lov->lov_packed, 0);
776 rc = lov_tunables_init(obd);
780 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
781 &obd->obd_kset.kobj);
787 static int lov_cleanup(struct obd_device *obd)
789 struct lov_obd *lov = &obd->u.lov;
790 struct list_head *pos, *tmp;
791 struct pool_desc *pool;
794 if (lov->lov_tgts_kobj) {
795 kobject_put(lov->lov_tgts_kobj);
796 lov->lov_tgts_kobj = NULL;
799 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
800 pool = list_entry(pos, struct pool_desc, pool_list);
801 /* free pool structs */
802 CDEBUG(D_INFO, "delete pool %p\n", pool);
803 /* In the function below, .hs_keycmp resolves to
804 * pool_hashkey_keycmp() */
805 /* coverity[overrun-buffer-val] */
806 lov_pool_del(obd, pool->pool_name);
808 cfs_hash_putref(lov->lov_pools_hash_body);
809 lov_ost_pool_free(&lov->lov_packed);
811 lprocfs_obd_cleanup(obd);
814 lov_tgts_getref(obd);
815 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
816 if (!lov->lov_tgts[i])
819 /* Inactive targets may never have connected */
820 if (lov->lov_tgts[i]->ltd_active ||
821 atomic_read(&lov->lov_refcount))
822 /* We should never get here - these
823 * should have been removed in the
825 CERROR("%s: lov tgt %d not cleaned! "
826 "deathrow=%d, lovrc=%d\n",
827 obd->obd_name, i, lov->lov_death_row,
828 atomic_read(&lov->lov_refcount));
829 lov_del_target(obd, i, NULL, 0);
831 lov_tgts_putref(obd);
832 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
834 lov->lov_tgt_size = 0;
837 if (lov->lov_cache != NULL) {
838 cl_cache_decref(lov->lov_cache);
839 lov->lov_cache = NULL;
845 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
846 u32 *indexp, int *genp)
848 struct obd_uuid obd_uuid;
853 switch (cmd = lcfg->lcfg_command) {
857 case LCFG_LOV_ADD_OBD:
858 case LCFG_LOV_ADD_INA:
859 case LCFG_LOV_DEL_OBD: {
863 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
864 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
865 GOTO(out, rc = -EINVAL);
867 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
869 rc = kstrtou32(lustre_cfg_buf(lcfg, 2), 10, indexp);
872 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
877 if (cmd == LCFG_LOV_ADD_OBD)
878 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
879 else if (cmd == LCFG_LOV_ADD_INA)
880 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
882 rc = lov_del_target(obd, index, &obd_uuid, gen);
887 struct lov_desc *desc = &(obd->u.lov.desc);
891 GOTO(out, rc = -EINVAL);
893 count = class_modify_config(lcfg, PARAM_LOV,
894 &obd->obd_kset.kobj);
895 GOTO(out, rc = count < 0 ? count : 0);
904 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
905 GOTO(out, rc = -EINVAL);
913 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
914 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
916 struct obd_device *obd = class_exp2obd(exp);
917 struct lov_obd *lov = &obd->u.lov;
918 struct obd_info oinfo = {
922 struct ptlrpc_request_set *rqset;
923 struct lov_request_set *set = NULL;
924 struct lov_request *req;
930 rqset = ptlrpc_prep_set();
934 rc = lov_prep_statfs_set(obd, &oinfo, &set);
938 list_for_each_entry(req, &set->set_list, rq_link) {
939 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
940 &req->rq_oi, max_age, rqset);
945 rc = ptlrpc_set_wait(env, rqset);
949 atomic_set(&set->set_completes, 0);
951 rc2 = lov_fini_statfs_set(set);
956 ptlrpc_set_destroy(rqset);
961 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
962 void *karg, void __user *uarg)
964 struct obd_device *obd = class_exp2obd(exp);
965 struct lov_obd *lov = &obd->u.lov;
966 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
967 struct obd_uuid *uuidp;
971 case IOC_OBD_STATFS: {
972 struct obd_ioctl_data *data = karg;
973 struct obd_device *osc_obd;
974 struct obd_statfs stat_buf = {0};
975 struct obd_import *imp;
979 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
983 if (!lov->lov_tgts[index])
984 /* Try again with the next index */
987 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
991 imp = osc_obd->u.cli.cl_import;
992 if (!lov->lov_tgts[index]->ltd_active &&
993 imp->imp_state != LUSTRE_IMP_IDLE)
997 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
998 min_t(unsigned long, data->ioc_plen2,
999 sizeof(struct obd_uuid))))
1002 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1003 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1005 /* got statfs data */
1006 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1007 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1011 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1012 min_t(unsigned long, data->ioc_plen1,
1013 sizeof(struct obd_statfs))))
1017 case OBD_IOC_LOV_GET_CONFIG: {
1018 struct obd_ioctl_data *data;
1019 struct lov_desc *desc;
1024 if (obd_ioctl_getdata(&buf, &len, uarg))
1027 data = (struct obd_ioctl_data *)buf;
1029 if (sizeof(*desc) > data->ioc_inllen1) {
1030 OBD_FREE_LARGE(buf, len);
1034 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1035 OBD_FREE_LARGE(buf, len);
1039 if (sizeof(__u32) * count > data->ioc_inllen3) {
1040 OBD_FREE_LARGE(buf, len);
1044 desc = (struct lov_desc *)data->ioc_inlbuf1;
1045 memcpy(desc, &(lov->desc), sizeof(*desc));
1047 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1048 genp = (__u32 *)data->ioc_inlbuf3;
1049 /* the uuid will be empty for deleted OSTs */
1050 for (i = 0; i < count; i++, uuidp++, genp++) {
1051 if (!lov->lov_tgts[i])
1053 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1054 *genp = lov->lov_tgts[i]->ltd_gen;
1057 if (copy_to_user(uarg, buf, len))
1059 OBD_FREE_LARGE(buf, len);
1062 case OBD_IOC_QUOTACTL: {
1063 struct if_quotactl *qctl = karg;
1064 struct lov_tgt_desc *tgt = NULL;
1065 struct obd_quotactl *oqctl;
1067 if (qctl->qc_valid == QC_OSTIDX) {
1068 if (count <= qctl->qc_idx)
1071 tgt = lov->lov_tgts[qctl->qc_idx];
1072 if (!tgt || !tgt->ltd_exp)
1074 } else if (qctl->qc_valid == QC_UUID) {
1075 for (i = 0; i < count; i++) {
1076 tgt = lov->lov_tgts[i];
1078 !obd_uuid_equals(&tgt->ltd_uuid,
1082 if (tgt->ltd_exp == NULL)
1094 LASSERT(tgt && tgt->ltd_exp);
1095 OBD_ALLOC_PTR(oqctl);
1099 QCTL_COPY(oqctl, qctl);
1100 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1102 QCTL_COPY(qctl, oqctl);
1103 qctl->qc_valid = QC_OSTIDX;
1104 qctl->obd_uuid = tgt->ltd_uuid;
1106 OBD_FREE_PTR(oqctl);
1115 for (i = 0; i < count; i++) {
1117 struct obd_device *osc_obd;
1119 /* OST was disconnected */
1120 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1123 /* ll_umount_begin() sets force on lov, pass to osc */
1124 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1125 osc_obd->obd_force = obd->obd_force;
1126 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1129 if (lov->lov_tgts[i]->ltd_active) {
1130 CDEBUG(err == -ENOTTY ?
1131 D_IOCTL : D_WARNING,
1132 "iocontrol OSC %s on OST "
1133 "idx %d cmd %x: err = %d\n",
1134 lov_uuid2str(lov, i),
1151 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1152 __u32 keylen, void *key, __u32 *vallen, void *val)
1154 struct obd_device *obddev = class_exp2obd(exp);
1155 struct lov_obd *lov = &obddev->u.lov;
1156 struct lov_desc *ld = &lov->desc;
1160 if (vallen == NULL || val == NULL)
1163 lov_tgts_getref(obddev);
1165 if (KEY_IS(KEY_MAX_EASIZE)) {
1166 *((u32 *)val) = exp->exp_connect_data.ocd_max_easize;
1167 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1168 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1169 LOV_MAX_STRIPE_COUNT);
1171 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1172 } else if (KEY_IS(KEY_TGT_COUNT)) {
1173 *((int *)val) = lov->desc.ld_tgt_count;
1178 lov_tgts_putref(obddev);
1183 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1184 __u32 keylen, void *key,
1185 __u32 vallen, void *val,
1186 struct ptlrpc_request_set *set)
1188 struct obd_device *obddev = class_exp2obd(exp);
1189 struct lov_obd *lov = &obddev->u.lov;
1190 struct lov_tgt_desc *tgt;
1191 bool do_inactive = false, no_set = false;
1200 set = ptlrpc_prep_set();
1205 lov_tgts_getref(obddev);
1207 if (KEY_IS(KEY_CHECKSUM))
1210 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1211 tgt = lov->lov_tgts[i];
1213 /* OST was disconnected */
1214 if (tgt == NULL || tgt->ltd_exp == NULL)
1217 /* OST is inactive and we don't want inactive OSCs */
1218 if (!tgt->ltd_active && !do_inactive)
1221 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1228 /* cycle through MDC target for Data-on-MDT */
1229 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1230 struct obd_device *mdc;
1232 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1236 err = obd_set_info_async(env, mdc->obd_self_export,
1237 keylen, key, vallen, val, set);
1242 lov_tgts_putref(obddev);
1244 err = ptlrpc_set_wait(env, set);
1247 ptlrpc_set_destroy(set);
1252 void lov_stripe_lock(struct lov_stripe_md *md)
1253 __acquires(&md->lsm_lock)
1255 LASSERT(md->lsm_lock_owner != current_pid());
1256 spin_lock(&md->lsm_lock);
1257 LASSERT(md->lsm_lock_owner == 0);
1258 md->lsm_lock_owner = current_pid();
1261 void lov_stripe_unlock(struct lov_stripe_md *md)
1262 __releases(&md->lsm_lock)
1264 LASSERT(md->lsm_lock_owner == current_pid());
1265 md->lsm_lock_owner = 0;
1266 spin_unlock(&md->lsm_lock);
1269 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1270 struct obd_quotactl *oqctl)
1272 struct lov_obd *lov = &obd->u.lov;
1273 struct lov_tgt_desc *tgt;
1275 __u64 bhardlimit = 0;
1279 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1280 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA) {
1281 CERROR("%s: bad quota opc %x for lov obd\n",
1282 obd->obd_name, oqctl->qc_cmd);
1287 lov_tgts_getref(obd);
1288 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1291 tgt = lov->lov_tgts[i];
1296 if (!tgt->ltd_active || tgt->ltd_reap) {
1297 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1298 lov->lov_tgts[i]->ltd_activate) {
1300 CERROR("ost %d is inactive\n", i);
1302 CDEBUG(D_HA, "ost %d is inactive\n", i);
1307 err = obd_quotactl(tgt->ltd_exp, oqctl);
1309 if (tgt->ltd_active && !rc)
1314 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1315 curspace += oqctl->qc_dqblk.dqb_curspace;
1316 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1319 lov_tgts_putref(obd);
1321 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1322 oqctl->qc_dqblk.dqb_curspace = curspace;
1323 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1328 static struct obd_ops lov_obd_ops = {
1329 .o_owner = THIS_MODULE,
1330 .o_setup = lov_setup,
1331 .o_cleanup = lov_cleanup,
1332 .o_connect = lov_connect,
1333 .o_disconnect = lov_disconnect,
1334 .o_statfs = lov_statfs,
1335 .o_iocontrol = lov_iocontrol,
1336 .o_get_info = lov_get_info,
1337 .o_set_info_async = lov_set_info_async,
1338 .o_notify = lov_notify,
1339 .o_pool_new = lov_pool_new,
1340 .o_pool_rem = lov_pool_remove,
1341 .o_pool_add = lov_pool_add,
1342 .o_pool_del = lov_pool_del,
1343 .o_quotactl = lov_quotactl,
1346 struct kmem_cache *lov_oinfo_slab;
1348 static int __init lov_init(void)
1353 /* print an address of _any_ initialized kernel symbol from this
1354 * module, to allow debugging with gdb that doesn't support data
1355 * symbols from modules.*/
1356 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1358 rc = lu_kmem_init(lov_caches);
1362 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1363 sizeof(struct lov_oinfo), 0,
1364 SLAB_HWCACHE_ALIGN, NULL);
1365 if (lov_oinfo_slab == NULL) {
1366 lu_kmem_fini(lov_caches);
1370 rc = class_register_type(&lov_obd_ops, NULL, true, NULL,
1371 LUSTRE_LOV_NAME, &lov_device_type);
1373 kmem_cache_destroy(lov_oinfo_slab);
1374 lu_kmem_fini(lov_caches);
1380 static void __exit lov_exit(void)
1382 class_unregister_type(LUSTRE_LOV_NAME);
1383 kmem_cache_destroy(lov_oinfo_slab);
1384 lu_kmem_fini(lov_caches);
1387 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1388 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1389 MODULE_VERSION(LUSTRE_VERSION_STRING);
1390 MODULE_LICENSE("GPL");
1392 module_init(lov_init);
1393 module_exit(lov_exit);