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 static void lov_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 static void lov_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 if (lov->lov_tgts_kobj)
108 kobject_put(lov->lov_tgts_kobj);
111 mutex_unlock(&lov->lov_lock);
115 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
116 enum obd_notify_event ev);
117 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
118 enum obd_notify_event ev);
120 int lov_connect_obd(struct obd_device *obd, u32 index, int activate,
121 struct obd_connect_data *data)
123 struct lov_obd *lov = &obd->u.lov;
124 struct obd_uuid *tgt_uuid;
125 struct obd_device *tgt_obd;
126 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
127 struct obd_import *imp;
131 if (lov->lov_tgts[index] == NULL)
134 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
135 tgt_obd = lov->lov_tgts[index]->ltd_obd;
137 if (!tgt_obd->obd_set_up) {
138 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
142 /* override the sp_me from lov */
143 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
145 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
146 data->ocd_index = index;
149 * Divine LOV knows that OBDs under it are OSCs.
151 imp = tgt_obd->u.cli.cl_import;
154 tgt_obd->obd_no_recov = 0;
155 /* FIXME this is probably supposed to be
156 ptlrpc_set_import_active. Horrible naming. */
157 ptlrpc_activate_import(imp);
160 rc = obd_register_observer(tgt_obd, obd);
162 CERROR("Target %s register_observer error %d\n",
163 obd_uuid2str(tgt_uuid), rc);
168 if (imp->imp_invalid) {
169 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
170 "disabled\n", obd_uuid2str(tgt_uuid));
174 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
175 &lov_osc_uuid, data, NULL);
176 if (rc || !lov->lov_tgts[index]->ltd_exp) {
177 CERROR("Target %s connect error %d\n",
178 obd_uuid2str(tgt_uuid), rc);
182 lov->lov_tgts[index]->ltd_reap = 0;
184 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
185 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
187 if (lov->lov_tgts_kobj) {
188 /* Even if we failed, that's ok */
189 rc = sysfs_create_link(lov->lov_tgts_kobj,
190 &tgt_obd->obd_kset.kobj,
193 CERROR("%s: can't register LOV target /sys/fs/lustre/%s/%s/target_obds/%s : rc = %d\n",
194 obd->obd_name, obd->obd_type->typ_name,
196 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_name,
203 static int lov_connect(const struct lu_env *env,
204 struct obd_export **exp, struct obd_device *obd,
205 struct obd_uuid *cluuid, struct obd_connect_data *data,
208 struct lov_obd *lov = &obd->u.lov;
209 struct lov_tgt_desc *tgt;
210 struct lustre_handle conn;
214 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
216 rc = class_connect(&conn, obd, cluuid);
220 *exp = class_conn2export(&conn);
222 /* Why should there ever be more than 1 connect? */
224 LASSERT(lov->lov_connects == 1);
226 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
228 lov->lov_ocd = *data;
232 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
233 &obd->obd_kset.kobj);
235 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
236 tgt = lov->lov_tgts[i];
237 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
239 /* Flags will be lowest common denominator */
240 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
242 CERROR("%s: lov connect tgt %d failed: %d\n",
243 obd->obd_name, i, rc);
246 /* connect to administrative disabled ost */
247 if (!lov->lov_tgts[i]->ltd_exp)
250 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
253 CERROR("%s error sending notify %d\n",
262 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
264 struct lov_obd *lov = &obd->u.lov;
265 struct obd_device *osc_obd;
269 osc_obd = class_exp2obd(tgt->ltd_exp);
270 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
271 obd->obd_name, osc_obd->obd_name);
273 if (tgt->ltd_active) {
275 lov->desc.ld_active_tgt_count--;
276 tgt->ltd_exp->exp_obd->obd_inactive = 1;
280 if (lov->lov_tgts_kobj)
281 sysfs_remove_link(lov->lov_tgts_kobj,
284 /* Pass it on to our clients.
285 * XXX This should be an argument to disconnect,
286 * XXX not a back-door flag on the OBD. Ah well.
288 osc_obd->obd_force = obd->obd_force;
289 osc_obd->obd_fail = obd->obd_fail;
290 osc_obd->obd_no_recov = obd->obd_no_recov;
292 if (lov->targets_proc_entry != NULL)
293 lprocfs_remove_proc_entry(osc_obd->obd_name,
294 lov->targets_proc_entry);
297 obd_register_observer(osc_obd, NULL);
299 rc = obd_disconnect(tgt->ltd_exp);
301 CERROR("Target %s disconnect error %d\n",
302 tgt->ltd_uuid.uuid, rc);
310 static int lov_disconnect(struct obd_export *exp)
312 struct obd_device *obd = class_exp2obd(exp);
313 struct lov_obd *lov = &obd->u.lov;
321 /* Only disconnect the underlying layers on the final disconnect. */
323 if (lov->lov_connects != 0) {
324 /* why should there be more than 1 connect? */
325 CWARN("%s: unexpected disconnect #%d\n",
326 obd->obd_name, lov->lov_connects);
330 /* hold another ref so lov_del_obd() doesn't spin in putref each time */
333 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
334 if (lov->lov_tgts[index] && lov->lov_tgts[index]->ltd_exp) {
335 /* Disconnection is the last we know about an OBD */
336 lov_del_target(obd, index, NULL,
337 lov->lov_tgts[index]->ltd_gen);
343 rc = class_disconnect(exp); /* bz 9811 */
349 * -EINVAL : UUID can't be found in the LOV's target list
350 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
351 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
352 * any >= 0 : is log target index
354 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
355 enum obd_notify_event ev)
357 struct lov_obd *lov = &obd->u.lov;
358 struct lov_tgt_desc *tgt;
359 int index, activate, active;
362 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
363 lov, uuid->uuid, ev);
366 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
367 tgt = lov->lov_tgts[index];
371 * LU-642, initially inactive OSC could miss the obd_connect,
372 * we make up for it here.
374 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
375 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
376 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
378 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
379 &lov_osc_uuid, &lov->lov_ocd, NULL);
384 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
385 index, obd_uuid2str(&tgt->ltd_uuid),
386 tgt->ltd_exp->exp_handle.h_cookie);
387 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
391 if (index == lov->desc.ld_tgt_count)
392 GOTO(out, index = -EINVAL);
394 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
395 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
397 if (lov->lov_tgts[index]->ltd_activate == activate) {
398 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
399 uuid->uuid, activate ? "" : "de");
401 lov->lov_tgts[index]->ltd_activate = activate;
402 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
403 activate ? "" : "de", obd_uuid2str(uuid));
406 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
407 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
409 if (lov->lov_tgts[index]->ltd_active == active) {
410 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
411 uuid->uuid, active ? "" : "in");
414 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
415 obd_uuid2str(uuid), active ? "" : "in");
418 lov->lov_tgts[index]->ltd_active = active;
420 lov->desc.ld_active_tgt_count++;
421 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
423 lov->desc.ld_active_tgt_count--;
424 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
427 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
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_obd on this target when that fn first runs,
576 because we don't know the connect flags yet. */
582 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
586 /* connect to administrative disabled ost */
590 if (lov->lov_cache != NULL) {
591 rc = obd_set_info_async(NULL, tgt->ltd_exp,
592 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
593 sizeof(struct cl_client_cache), lov->lov_cache,
599 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
600 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
604 CERROR("%s: add failed, deleting %s: rc = %d\n",
605 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), rc);
606 lov_del_target(obd, index, NULL, 0);
612 /* Schedule a target for deletion */
613 int lov_del_target(struct obd_device *obd, u32 index,
614 struct obd_uuid *uuidp, int gen)
616 struct lov_obd *lov = &obd->u.lov;
617 int count = lov->desc.ld_tgt_count;
621 if (index >= count) {
622 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
627 /* to make sure there's no ongoing lov_notify() now */
628 down_write(&lov->lov_notify_lock);
631 if (!lov->lov_tgts[index]) {
632 CERROR("LOV target at index %d is not setup.\n", index);
633 GOTO(out, rc = -EINVAL);
636 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
637 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
638 lov_uuid2str(lov, index), index,
639 obd_uuid2str(uuidp));
640 GOTO(out, rc = -EINVAL);
643 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
644 lov_uuid2str(lov, index), index,
645 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
646 lov->lov_tgts[index]->ltd_active);
648 lov->lov_tgts[index]->ltd_reap = 1;
649 lov->lov_death_row++;
650 /* we really delete it from obd_putref */
653 up_write(&lov->lov_notify_lock);
658 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
660 struct obd_device *osc_obd;
663 LASSERT(tgt->ltd_reap);
665 osc_obd = class_exp2obd(tgt->ltd_exp);
667 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
669 osc_obd ? osc_obd->obd_name : "<no obd>");
672 lov_disconnect_obd(obd, tgt);
676 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
677 do it ourselves. And we can't do it from lov_cleanup,
678 because we just lost our only reference to it. */
680 class_manual_cleanup(osc_obd);
683 void lov_fix_desc_stripe_size(__u64 *val)
685 if (*val < LOV_MIN_STRIPE_SIZE) {
687 LCONSOLE_INFO("Increasing default stripe size to "
689 LOV_DESC_STRIPE_SIZE_DEFAULT);
690 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
691 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
692 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
693 LCONSOLE_WARN("Changing default stripe size to %llu (a "
695 *val, LOV_MIN_STRIPE_SIZE);
699 void lov_fix_desc_stripe_count(__u32 *val)
705 void lov_fix_desc_pattern(__u32 *val)
707 /* from lov_setstripe */
708 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
709 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
714 void lov_fix_desc_qos_maxage(__u32 *val)
717 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
720 void lov_fix_desc(struct lov_desc *desc)
722 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
723 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
724 lov_fix_desc_pattern(&desc->ld_pattern);
725 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
728 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
730 struct lov_desc *desc;
731 struct lov_obd *lov = &obd->u.lov;
735 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
736 CERROR("LOV setup requires a descriptor\n");
740 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
742 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
743 CERROR("descriptor size wrong: %d > %d\n",
744 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
748 if (desc->ld_magic != LOV_DESC_MAGIC) {
749 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
750 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
751 obd->obd_name, desc);
752 lustre_swab_lov_desc(desc);
754 CERROR("%s: Bad lov desc magic: %#x\n",
755 obd->obd_name, desc->ld_magic);
762 desc->ld_active_tgt_count = 0;
764 lov->lov_tgt_size = 0;
766 mutex_init(&lov->lov_lock);
767 atomic_set(&lov->lov_refcount, 0);
768 lov->lov_sp_me = LUSTRE_SP_CLI;
770 init_rwsem(&lov->lov_notify_lock);
772 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
774 HASH_POOLS_BKT_BITS, 0,
777 &pool_hash_operations,
779 INIT_LIST_HEAD(&lov->lov_pool_list);
780 lov->lov_pool_count = 0;
781 rc = lov_ost_pool_init(&lov->lov_packed, 0);
785 rc = lov_tunables_init(obd);
790 static int lov_cleanup(struct obd_device *obd)
792 struct lov_obd *lov = &obd->u.lov;
793 struct list_head *pos, *tmp;
794 struct pool_desc *pool;
797 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
798 pool = list_entry(pos, struct pool_desc, pool_list);
799 /* free pool structs */
800 CDEBUG(D_INFO, "delete pool %p\n", pool);
801 /* In the function below, .hs_keycmp resolves to
802 * pool_hashkey_keycmp() */
803 /* coverity[overrun-buffer-val] */
804 lov_pool_del(obd, pool->pool_name);
806 cfs_hash_putref(lov->lov_pools_hash_body);
807 lov_ost_pool_free(&lov->lov_packed);
809 lprocfs_obd_cleanup(obd);
813 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
814 if (!lov->lov_tgts[i])
817 /* Inactive targets may never have connected */
818 if (lov->lov_tgts[i]->ltd_active ||
819 atomic_read(&lov->lov_refcount))
820 /* We should never get here - these
821 * should have been removed in the
823 CERROR("%s: lov tgt %d not cleaned! "
824 "deathrow=%d, lovrc=%d\n",
825 obd->obd_name, i, lov->lov_death_row,
826 atomic_read(&lov->lov_refcount));
827 lov_del_target(obd, i, NULL, 0);
830 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
832 lov->lov_tgt_size = 0;
835 if (lov->lov_cache != NULL) {
836 cl_cache_decref(lov->lov_cache);
837 lov->lov_cache = NULL;
843 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
844 u32 *indexp, int *genp)
846 struct obd_uuid obd_uuid;
851 switch (cmd = lcfg->lcfg_command) {
855 case LCFG_LOV_ADD_OBD:
856 case LCFG_LOV_ADD_INA:
857 case LCFG_LOV_DEL_OBD: {
861 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
862 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
863 GOTO(out, rc = -EINVAL);
865 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
867 rc = kstrtou32(lustre_cfg_buf(lcfg, 2), 10, indexp);
870 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
875 if (cmd == LCFG_LOV_ADD_OBD)
876 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
877 else if (cmd == LCFG_LOV_ADD_INA)
878 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
880 rc = lov_del_target(obd, index, &obd_uuid, gen);
885 struct lov_desc *desc = &(obd->u.lov.desc);
889 GOTO(out, rc = -EINVAL);
891 count = class_modify_config(lcfg, PARAM_LOV,
892 &obd->obd_kset.kobj);
893 GOTO(out, rc = count < 0 ? count : 0);
902 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
903 GOTO(out, rc = -EINVAL);
911 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
912 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
914 struct obd_device *obd = class_exp2obd(exp);
915 struct lov_obd *lov = &obd->u.lov;
916 struct obd_info oinfo = {
920 struct ptlrpc_request_set *rqset;
921 struct lov_request_set *set = NULL;
922 struct lov_request *req;
928 rqset = ptlrpc_prep_set();
932 rc = lov_prep_statfs_set(obd, &oinfo, &set);
936 list_for_each_entry(req, &set->set_list, rq_link) {
937 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
938 &req->rq_oi, max_age, rqset);
943 rc = ptlrpc_set_wait(rqset);
947 atomic_set(&set->set_completes, 0);
949 rc2 = lov_fini_statfs_set(set);
954 ptlrpc_set_destroy(rqset);
959 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
960 void *karg, void __user *uarg)
962 struct obd_device *obddev = class_exp2obd(exp);
963 struct lov_obd *lov = &obddev->u.lov;
964 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
965 struct obd_uuid *uuidp;
969 case IOC_OBD_STATFS: {
970 struct obd_ioctl_data *data = karg;
971 struct obd_device *osc_obd;
972 struct obd_statfs stat_buf = {0};
976 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
977 if ((index >= count))
980 if (!lov->lov_tgts[index])
981 /* Try again with the next index */
983 if (!lov->lov_tgts[index]->ltd_active)
986 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
991 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
992 min_t(unsigned long, data->ioc_plen2,
993 sizeof(struct obd_uuid))))
996 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
997 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
999 /* got statfs data */
1000 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1001 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1005 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1006 min_t(unsigned long, data->ioc_plen1,
1007 sizeof(struct obd_statfs))))
1011 case OBD_IOC_LOV_GET_CONFIG: {
1012 struct obd_ioctl_data *data;
1013 struct lov_desc *desc;
1018 if (obd_ioctl_getdata(&buf, &len, uarg))
1021 data = (struct obd_ioctl_data *)buf;
1023 if (sizeof(*desc) > data->ioc_inllen1) {
1024 OBD_FREE_LARGE(buf, len);
1028 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1029 OBD_FREE_LARGE(buf, len);
1033 if (sizeof(__u32) * count > data->ioc_inllen3) {
1034 OBD_FREE_LARGE(buf, len);
1038 desc = (struct lov_desc *)data->ioc_inlbuf1;
1039 memcpy(desc, &(lov->desc), sizeof(*desc));
1041 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1042 genp = (__u32 *)data->ioc_inlbuf3;
1043 /* the uuid will be empty for deleted OSTs */
1044 for (i = 0; i < count; i++, uuidp++, genp++) {
1045 if (!lov->lov_tgts[i])
1047 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1048 *genp = lov->lov_tgts[i]->ltd_gen;
1051 if (copy_to_user(uarg, buf, len))
1053 OBD_FREE_LARGE(buf, len);
1056 case OBD_IOC_QUOTACTL: {
1057 struct if_quotactl *qctl = karg;
1058 struct lov_tgt_desc *tgt = NULL;
1059 struct obd_quotactl *oqctl;
1061 if (qctl->qc_valid == QC_OSTIDX) {
1062 if (count <= qctl->qc_idx)
1065 tgt = lov->lov_tgts[qctl->qc_idx];
1066 if (!tgt || !tgt->ltd_exp)
1068 } else if (qctl->qc_valid == QC_UUID) {
1069 for (i = 0; i < count; i++) {
1070 tgt = lov->lov_tgts[i];
1072 !obd_uuid_equals(&tgt->ltd_uuid,
1076 if (tgt->ltd_exp == NULL)
1088 LASSERT(tgt && tgt->ltd_exp);
1089 OBD_ALLOC_PTR(oqctl);
1093 QCTL_COPY(oqctl, qctl);
1094 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1096 QCTL_COPY(qctl, oqctl);
1097 qctl->qc_valid = QC_OSTIDX;
1098 qctl->obd_uuid = tgt->ltd_uuid;
1100 OBD_FREE_PTR(oqctl);
1109 for (i = 0; i < count; i++) {
1111 struct obd_device *osc_obd;
1113 /* OST was disconnected */
1114 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1117 /* ll_umount_begin() sets force flag but for lov, not
1118 * osc. Let's pass it through */
1119 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1120 osc_obd->obd_force = obddev->obd_force;
1121 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1124 if (lov->lov_tgts[i]->ltd_active) {
1125 CDEBUG(err == -ENOTTY ?
1126 D_IOCTL : D_WARNING,
1127 "iocontrol OSC %s on OST "
1128 "idx %d cmd %x: err = %d\n",
1129 lov_uuid2str(lov, i),
1146 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1147 __u32 keylen, void *key, __u32 *vallen, void *val)
1149 struct obd_device *obddev = class_exp2obd(exp);
1150 struct lov_obd *lov = &obddev->u.lov;
1151 struct lov_desc *ld = &lov->desc;
1155 if (vallen == NULL || val == NULL)
1160 if (KEY_IS(KEY_MAX_EASIZE)) {
1161 u32 max_stripe_count = min_t(u32, ld->ld_active_tgt_count,
1162 LOV_MAX_STRIPE_COUNT);
1164 *((u32 *)val) = lov_mds_md_size(max_stripe_count, LOV_MAGIC_V3);
1165 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1166 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1167 LOV_MAX_STRIPE_COUNT);
1169 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1170 } else if (KEY_IS(KEY_TGT_COUNT)) {
1171 *((int *)val) = lov->desc.ld_tgt_count;
1181 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1182 __u32 keylen, void *key,
1183 __u32 vallen, void *val,
1184 struct ptlrpc_request_set *set)
1186 struct obd_device *obddev = class_exp2obd(exp);
1187 struct lov_obd *lov = &obddev->u.lov;
1188 struct lov_tgt_desc *tgt;
1189 bool do_inactive = false, no_set = false;
1198 set = ptlrpc_prep_set();
1205 if (KEY_IS(KEY_CHECKSUM)) {
1207 } else if (KEY_IS(KEY_CACHE_SET)) {
1208 LASSERT(lov->lov_cache == NULL);
1209 lov->lov_cache = val;
1211 cl_cache_incref(lov->lov_cache);
1214 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1215 tgt = lov->lov_tgts[i];
1217 /* OST was disconnected */
1218 if (tgt == NULL || tgt->ltd_exp == NULL)
1221 /* OST is inactive and we don't want inactive OSCs */
1222 if (!tgt->ltd_active && !do_inactive)
1225 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1232 /* cycle through MDC target for Data-on-MDT */
1233 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1234 struct obd_device *mdc;
1236 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1240 err = obd_set_info_async(env, mdc->obd_self_export,
1241 keylen, key, vallen, val, set);
1248 err = ptlrpc_set_wait(set);
1251 ptlrpc_set_destroy(set);
1256 void lov_stripe_lock(struct lov_stripe_md *md)
1257 __acquires(&md->lsm_lock)
1259 LASSERT(md->lsm_lock_owner != current_pid());
1260 spin_lock(&md->lsm_lock);
1261 LASSERT(md->lsm_lock_owner == 0);
1262 md->lsm_lock_owner = current_pid();
1265 void lov_stripe_unlock(struct lov_stripe_md *md)
1266 __releases(&md->lsm_lock)
1268 LASSERT(md->lsm_lock_owner == current_pid());
1269 md->lsm_lock_owner = 0;
1270 spin_unlock(&md->lsm_lock);
1273 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1274 struct obd_quotactl *oqctl)
1276 struct lov_obd *lov = &obd->u.lov;
1277 struct lov_tgt_desc *tgt;
1279 __u64 bhardlimit = 0;
1283 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1284 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA) {
1285 CERROR("%s: bad quota opc %x for lov obd\n",
1286 obd->obd_name, oqctl->qc_cmd);
1292 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1295 tgt = lov->lov_tgts[i];
1300 if (!tgt->ltd_active || tgt->ltd_reap) {
1301 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1302 lov->lov_tgts[i]->ltd_activate) {
1304 CERROR("ost %d is inactive\n", i);
1306 CDEBUG(D_HA, "ost %d is inactive\n", i);
1311 err = obd_quotactl(tgt->ltd_exp, oqctl);
1313 if (tgt->ltd_active && !rc)
1318 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1319 curspace += oqctl->qc_dqblk.dqb_curspace;
1320 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1325 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1326 oqctl->qc_dqblk.dqb_curspace = curspace;
1327 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1332 static struct obd_ops lov_obd_ops = {
1333 .o_owner = THIS_MODULE,
1334 .o_setup = lov_setup,
1335 .o_cleanup = lov_cleanup,
1336 .o_connect = lov_connect,
1337 .o_disconnect = lov_disconnect,
1338 .o_statfs = lov_statfs,
1339 .o_iocontrol = lov_iocontrol,
1340 .o_get_info = lov_get_info,
1341 .o_set_info_async = lov_set_info_async,
1342 .o_notify = lov_notify,
1343 .o_pool_new = lov_pool_new,
1344 .o_pool_rem = lov_pool_remove,
1345 .o_pool_add = lov_pool_add,
1346 .o_pool_del = lov_pool_del,
1347 .o_getref = lov_getref,
1348 .o_putref = lov_putref,
1349 .o_quotactl = lov_quotactl,
1352 struct kmem_cache *lov_oinfo_slab;
1354 static int __init lov_init(void)
1356 bool enable_proc = true;
1357 struct obd_type *type;
1361 /* print an address of _any_ initialized kernel symbol from this
1362 * module, to allow debugging with gdb that doesn't support data
1363 * symbols from modules.*/
1364 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1366 rc = lu_kmem_init(lov_caches);
1370 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1371 sizeof(struct lov_oinfo), 0,
1372 SLAB_HWCACHE_ALIGN, NULL);
1373 if (lov_oinfo_slab == NULL) {
1374 lu_kmem_fini(lov_caches);
1378 type = class_search_type(LUSTRE_LOD_NAME);
1379 if (type != NULL && type->typ_procsym != NULL)
1380 enable_proc = false;
1382 rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
1383 LUSTRE_LOV_NAME, &lov_device_type);
1386 kmem_cache_destroy(lov_oinfo_slab);
1387 lu_kmem_fini(lov_caches);
1393 static void __exit lov_exit(void)
1395 class_unregister_type(LUSTRE_LOV_NAME);
1396 kmem_cache_destroy(lov_oinfo_slab);
1397 lu_kmem_fini(lov_caches);
1400 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1401 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1402 MODULE_VERSION(LUSTRE_VERSION_STRING);
1403 MODULE_LICENSE("GPL");
1405 module_init(lov_init);
1406 module_exit(lov_exit);