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 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_obd(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);
156 rc = obd_register_observer(tgt_obd, obd);
158 CERROR("Target %s register_observer error %d\n",
159 obd_uuid2str(tgt_uuid), rc);
164 if (imp->imp_invalid) {
165 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
166 "disabled\n", obd_uuid2str(tgt_uuid));
170 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
171 &lov_osc_uuid, data, NULL);
172 if (rc || !lov->lov_tgts[index]->ltd_exp) {
173 CERROR("Target %s connect error %d\n",
174 obd_uuid2str(tgt_uuid), rc);
178 lov->lov_tgts[index]->ltd_reap = 0;
180 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
181 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
183 if (lov->targets_proc_entry != NULL) {
184 struct proc_dir_entry *osc_symlink;
185 struct obd_device *osc_obd;
187 osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
189 LASSERT(osc_obd != NULL);
190 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
191 LASSERT(osc_obd->obd_type->typ_name != NULL);
193 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
194 lov->targets_proc_entry,
196 osc_obd->obd_type->typ_name,
198 if (osc_symlink == NULL) {
199 CERROR("cannot register LOV target "
200 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
201 obd->obd_type->typ_name, obd->obd_name,
208 static int lov_connect(const struct lu_env *env,
209 struct obd_export **exp, struct obd_device *obd,
210 struct obd_uuid *cluuid, struct obd_connect_data *data,
213 struct lov_obd *lov = &obd->u.lov;
214 struct lov_tgt_desc *tgt;
215 struct lustre_handle conn;
219 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
221 rc = class_connect(&conn, obd, cluuid);
225 *exp = class_conn2export(&conn);
227 /* Why should there ever be more than 1 connect? */
229 LASSERT(lov->lov_connects == 1);
231 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
233 lov->lov_ocd = *data;
235 lov->targets_proc_entry = lprocfs_register("target_obds",
238 if (IS_ERR(lov->targets_proc_entry)) {
239 CERROR("%s: cannot register "
240 "/proc/fs/lustre/%s/%s/target_obds\n",
241 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
242 lov->targets_proc_entry = NULL;
246 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
247 tgt = lov->lov_tgts[i];
248 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
250 /* Flags will be lowest common denominator */
251 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
253 CERROR("%s: lov connect tgt %d failed: %d\n",
254 obd->obd_name, i, rc);
257 /* connect to administrative disabled ost */
258 if (!lov->lov_tgts[i]->ltd_exp)
261 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
264 CERROR("%s error sending notify %d\n",
273 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
275 struct lov_obd *lov = &obd->u.lov;
276 struct obd_device *osc_obd;
280 osc_obd = class_exp2obd(tgt->ltd_exp);
281 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
282 obd->obd_name, osc_obd->obd_name);
284 if (tgt->ltd_active) {
286 lov->desc.ld_active_tgt_count--;
287 tgt->ltd_exp->exp_obd->obd_inactive = 1;
291 /* Pass it on to our clients.
292 * XXX This should be an argument to disconnect,
293 * XXX not a back-door flag on the OBD. Ah well.
295 osc_obd->obd_force = obd->obd_force;
296 osc_obd->obd_fail = obd->obd_fail;
297 osc_obd->obd_no_recov = obd->obd_no_recov;
299 if (lov->targets_proc_entry != NULL)
300 lprocfs_remove_proc_entry(osc_obd->obd_name,
301 lov->targets_proc_entry);
304 obd_register_observer(osc_obd, NULL);
306 rc = obd_disconnect(tgt->ltd_exp);
308 CERROR("Target %s disconnect error %d\n",
309 tgt->ltd_uuid.uuid, rc);
317 static int lov_disconnect(struct obd_export *exp)
319 struct obd_device *obd = class_exp2obd(exp);
320 struct lov_obd *lov = &obd->u.lov;
327 /* Only disconnect the underlying layers on the final disconnect. */
329 if (lov->lov_connects != 0) {
330 /* why should there be more than 1 connect? */
331 CERROR("disconnect #%d\n", lov->lov_connects);
335 /* Let's hold another reference so lov_del_obd doesn't spin through
339 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
340 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
341 /* Disconnection is the last we know about an obd */
342 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
347 if (lov->targets_proc_entry != NULL)
348 lprocfs_remove(&lov->targets_proc_entry);
351 rc = class_disconnect(exp); /* bz 9811 */
357 * -EINVAL : UUID can't be found in the LOV's target list
358 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
359 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
360 * any >= 0 : is log target index
362 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
363 enum obd_notify_event ev)
365 struct lov_obd *lov = &obd->u.lov;
366 struct lov_tgt_desc *tgt;
367 int index, activate, active;
370 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
371 lov, uuid->uuid, ev);
374 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
375 tgt = lov->lov_tgts[index];
379 * LU-642, initially inactive OSC could miss the obd_connect,
380 * we make up for it here.
382 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
383 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
384 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
386 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
387 &lov_osc_uuid, &lov->lov_ocd, NULL);
392 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
393 index, obd_uuid2str(&tgt->ltd_uuid),
394 tgt->ltd_exp->exp_handle.h_cookie);
395 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
399 if (index == lov->desc.ld_tgt_count)
400 GOTO(out, index = -EINVAL);
402 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
403 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
405 if (lov->lov_tgts[index]->ltd_activate == activate) {
406 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
407 uuid->uuid, activate ? "" : "de");
409 lov->lov_tgts[index]->ltd_activate = activate;
410 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
411 activate ? "" : "de", obd_uuid2str(uuid));
414 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
415 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
417 if (lov->lov_tgts[index]->ltd_active == active) {
418 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
419 uuid->uuid, active ? "" : "in");
422 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
423 obd_uuid2str(uuid), active ? "" : "in");
426 lov->lov_tgts[index]->ltd_active = active;
428 lov->desc.ld_active_tgt_count++;
429 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
431 lov->desc.ld_active_tgt_count--;
432 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
435 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
444 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
445 enum obd_notify_event ev)
448 struct lov_obd *lov = &obd->u.lov;
451 down_read(&lov->lov_notify_lock);
452 if (!lov->lov_connects)
453 GOTO(out_notify_lock, rc = 0);
455 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
456 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
457 struct obd_uuid *uuid;
461 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
462 CERROR("unexpected notification of %s %s\n",
463 watched->obd_type->typ_name, watched->obd_name);
464 GOTO(out_notify_lock, rc = -EINVAL);
467 uuid = &watched->u.cli.cl_target_uuid;
469 /* Set OSC as active before notifying the observer, so the
470 * observer can use the OSC normally.
472 rc = lov_set_osc_active(obd, uuid, ev);
474 CERROR("%s: event %d failed: rc = %d\n", obd->obd_name,
476 GOTO(out_notify_lock, rc);
480 /* Pass the notification up the chain. */
481 rc = obd_notify_observer(obd, watched, ev);
484 up_read(&lov->lov_notify_lock);
489 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
490 __u32 index, int gen, int active)
492 struct lov_obd *lov = &obd->u.lov;
493 struct lov_tgt_desc *tgt;
494 struct obd_device *tgt_obd;
498 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
499 uuidp->uuid, index, gen, active);
502 CERROR("request to add OBD %s with invalid generation: %d\n",
507 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
512 mutex_lock(&lov->lov_lock);
514 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
515 tgt = lov->lov_tgts[index];
516 CERROR("UUID %s already assigned at LOV target index %d\n",
517 obd_uuid2str(&tgt->ltd_uuid), index);
518 mutex_unlock(&lov->lov_lock);
522 if (index >= lov->lov_tgt_size) {
523 /* We need to reallocate the lov target array. */
524 struct lov_tgt_desc **newtgts, **old = NULL;
525 __u32 newsize, oldsize = 0;
527 newsize = max(lov->lov_tgt_size, (__u32)2);
528 while (newsize < index + 1)
529 newsize = newsize << 1;
530 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
531 if (newtgts == NULL) {
532 mutex_unlock(&lov->lov_lock);
536 if (lov->lov_tgt_size) {
537 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
540 oldsize = lov->lov_tgt_size;
543 lov->lov_tgts = newtgts;
544 lov->lov_tgt_size = newsize;
547 OBD_FREE(old, sizeof(*old) * oldsize);
549 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
550 lov->lov_tgts, lov->lov_tgt_size);
555 mutex_unlock(&lov->lov_lock);
559 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
561 mutex_unlock(&lov->lov_lock);
566 tgt->ltd_uuid = *uuidp;
567 tgt->ltd_obd = tgt_obd;
568 /* XXX - add a sanity check on the generation number. */
570 tgt->ltd_index = index;
571 tgt->ltd_activate = active;
572 lov->lov_tgts[index] = tgt;
573 if (index >= lov->desc.ld_tgt_count)
574 lov->desc.ld_tgt_count = index + 1;
576 mutex_unlock(&lov->lov_lock);
578 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
579 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
581 if (lov->lov_connects == 0) {
582 /* lov_connect hasn't been called yet. We'll do the
583 lov_connect_obd on this target when that fn first runs,
584 because we don't know the connect flags yet. */
590 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
594 /* connect to administrative disabled ost */
598 if (lov->lov_cache != NULL) {
599 rc = obd_set_info_async(NULL, tgt->ltd_exp,
600 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
601 sizeof(struct cl_client_cache), lov->lov_cache,
607 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
608 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
612 CERROR("add failed (%d), deleting %s\n", rc,
613 obd_uuid2str(&tgt->ltd_uuid));
614 lov_del_target(obd, index, NULL, 0);
620 /* Schedule a target for deletion */
621 int lov_del_target(struct obd_device *obd, __u32 index,
622 struct obd_uuid *uuidp, int gen)
624 struct lov_obd *lov = &obd->u.lov;
625 int count = lov->desc.ld_tgt_count;
629 if (index >= count) {
630 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
635 /* to make sure there's no ongoing lov_notify() now */
636 down_write(&lov->lov_notify_lock);
639 if (!lov->lov_tgts[index]) {
640 CERROR("LOV target at index %d is not setup.\n", index);
641 GOTO(out, rc = -EINVAL);
644 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
645 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
646 lov_uuid2str(lov, index), index,
647 obd_uuid2str(uuidp));
648 GOTO(out, rc = -EINVAL);
651 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
652 lov_uuid2str(lov, index), index,
653 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
654 lov->lov_tgts[index]->ltd_active);
656 lov->lov_tgts[index]->ltd_reap = 1;
657 lov->lov_death_row++;
658 /* we really delete it from obd_putref */
661 up_write(&lov->lov_notify_lock);
666 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
668 struct obd_device *osc_obd;
671 LASSERT(tgt->ltd_reap);
673 osc_obd = class_exp2obd(tgt->ltd_exp);
675 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
677 osc_obd ? osc_obd->obd_name : "<no obd>");
680 lov_disconnect_obd(obd, tgt);
684 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
685 do it ourselves. And we can't do it from lov_cleanup,
686 because we just lost our only reference to it. */
688 class_manual_cleanup(osc_obd);
691 void lov_fix_desc_stripe_size(__u64 *val)
693 if (*val < LOV_MIN_STRIPE_SIZE) {
695 LCONSOLE_INFO("Increasing default stripe size to "
697 LOV_DESC_STRIPE_SIZE_DEFAULT);
698 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
699 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
700 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
701 LCONSOLE_WARN("Changing default stripe size to %llu (a "
703 *val, LOV_MIN_STRIPE_SIZE);
707 void lov_fix_desc_stripe_count(__u32 *val)
713 void lov_fix_desc_pattern(__u32 *val)
715 /* from lov_setstripe */
716 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
717 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
722 void lov_fix_desc_qos_maxage(__u32 *val)
725 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
728 void lov_fix_desc(struct lov_desc *desc)
730 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
731 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
732 lov_fix_desc_pattern(&desc->ld_pattern);
733 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
736 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
738 struct lov_desc *desc;
739 struct lov_obd *lov = &obd->u.lov;
740 #ifdef CONFIG_PROC_FS
741 struct obd_type *type;
746 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
747 CERROR("LOV setup requires a descriptor\n");
751 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
753 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
754 CERROR("descriptor size wrong: %d > %d\n",
755 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
759 if (desc->ld_magic != LOV_DESC_MAGIC) {
760 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
761 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
762 obd->obd_name, desc);
763 lustre_swab_lov_desc(desc);
765 CERROR("%s: Bad lov desc magic: %#x\n",
766 obd->obd_name, desc->ld_magic);
773 desc->ld_active_tgt_count = 0;
775 lov->lov_tgt_size = 0;
777 mutex_init(&lov->lov_lock);
778 atomic_set(&lov->lov_refcount, 0);
779 lov->lov_sp_me = LUSTRE_SP_CLI;
781 init_rwsem(&lov->lov_notify_lock);
783 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
785 HASH_POOLS_BKT_BITS, 0,
788 &pool_hash_operations,
790 INIT_LIST_HEAD(&lov->lov_pool_list);
791 lov->lov_pool_count = 0;
792 rc = lov_ost_pool_init(&lov->lov_packed, 0);
796 obd->obd_vars = lprocfs_lov_obd_vars;
797 #ifdef CONFIG_PROC_FS
798 /* If this is true then both client (lov) and server
799 * (lod) are on the same node. The lod layer if loaded
800 * first will register the lov proc directory. In that
801 * case obd->obd_type->typ_procroot will be not set.
802 * Instead we use type->typ_procsym as the parent.
804 type = class_search_type(LUSTRE_LOD_NAME);
805 if (type && type->typ_procsym) {
806 obd->obd_proc_entry = lprocfs_register(obd->obd_name,
809 if (IS_ERR(obd->obd_proc_entry)) {
810 rc = PTR_ERR(obd->obd_proc_entry);
811 CERROR("error %d setting up lprocfs for %s\n", rc,
813 obd->obd_proc_entry = NULL;
818 rc = lprocfs_obd_setup(obd, false);
822 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", 0444,
823 &lov_proc_target_fops, obd);
825 CWARN("%s: Error adding the target_obd file : rc %d\n",
828 lov->lov_pool_proc_entry = lprocfs_register("pools",
831 if (IS_ERR(lov->lov_pool_proc_entry)) {
832 rc = PTR_ERR(lov->lov_pool_proc_entry);
833 CERROR("%s: error setting up ldebugfs for pools : rc %d\n",
835 lov->lov_pool_proc_entry = NULL;
841 static int lov_cleanup(struct obd_device *obd)
843 struct lov_obd *lov = &obd->u.lov;
844 struct list_head *pos, *tmp;
845 struct pool_desc *pool;
848 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
849 pool = list_entry(pos, struct pool_desc, pool_list);
850 /* free pool structs */
851 CDEBUG(D_INFO, "delete pool %p\n", pool);
852 /* In the function below, .hs_keycmp resolves to
853 * pool_hashkey_keycmp() */
854 /* coverity[overrun-buffer-val] */
855 lov_pool_del(obd, pool->pool_name);
857 cfs_hash_putref(lov->lov_pools_hash_body);
858 lov_ost_pool_free(&lov->lov_packed);
860 lprocfs_obd_cleanup(obd);
864 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
865 if (!lov->lov_tgts[i])
868 /* Inactive targets may never have connected */
869 if (lov->lov_tgts[i]->ltd_active ||
870 atomic_read(&lov->lov_refcount))
871 /* We should never get here - these
872 * should have been removed in the
874 CERROR("%s: lov tgt %d not cleaned! "
875 "deathrow=%d, lovrc=%d\n",
876 obd->obd_name, i, lov->lov_death_row,
877 atomic_read(&lov->lov_refcount));
878 lov_del_target(obd, i, NULL, 0);
881 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
883 lov->lov_tgt_size = 0;
886 if (lov->lov_cache != NULL) {
887 cl_cache_decref(lov->lov_cache);
888 lov->lov_cache = NULL;
894 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
895 __u32 *indexp, int *genp)
897 struct obd_uuid obd_uuid;
902 switch(cmd = lcfg->lcfg_command) {
906 case LCFG_LOV_ADD_OBD:
907 case LCFG_LOV_ADD_INA:
908 case LCFG_LOV_DEL_OBD: {
911 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
912 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
913 GOTO(out, rc = -EINVAL);
915 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
917 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", indexp) != 1)
918 GOTO(out, rc = -EINVAL);
919 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
920 GOTO(out, rc = -EINVAL);
923 if (cmd == LCFG_LOV_ADD_OBD)
924 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
925 else if (cmd == LCFG_LOV_ADD_INA)
926 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
928 rc = lov_del_target(obd, index, &obd_uuid, gen);
932 struct lov_desc *desc = &(obd->u.lov.desc);
935 GOTO(out, rc = -EINVAL);
937 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
950 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
951 GOTO(out, rc = -EINVAL);
959 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
960 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
962 struct obd_device *obd = class_exp2obd(exp);
963 struct lov_obd *lov = &obd->u.lov;
964 struct obd_info oinfo = {
968 struct ptlrpc_request_set *rqset;
969 struct lov_request_set *set = NULL;
970 struct lov_request *req;
976 rqset = ptlrpc_prep_set();
980 rc = lov_prep_statfs_set(obd, &oinfo, &set);
984 list_for_each_entry(req, &set->set_list, rq_link) {
985 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
986 &req->rq_oi, max_age, rqset);
991 rc = ptlrpc_set_wait(rqset);
995 atomic_set(&set->set_completes, 0);
997 rc2 = lov_fini_statfs_set(set);
1002 ptlrpc_set_destroy(rqset);
1007 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1008 void *karg, void __user *uarg)
1010 struct obd_device *obddev = class_exp2obd(exp);
1011 struct lov_obd *lov = &obddev->u.lov;
1012 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1013 struct obd_uuid *uuidp;
1017 case IOC_OBD_STATFS: {
1018 struct obd_ioctl_data *data = karg;
1019 struct obd_device *osc_obd;
1020 struct obd_statfs stat_buf = {0};
1024 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
1025 if ((index >= count))
1028 if (!lov->lov_tgts[index])
1029 /* Try again with the next index */
1031 if (!lov->lov_tgts[index]->ltd_active)
1034 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1039 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1040 min_t(unsigned long, data->ioc_plen2,
1041 sizeof(struct obd_uuid))))
1044 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1045 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1047 /* got statfs data */
1048 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1049 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1053 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1054 min_t(unsigned long, data->ioc_plen1,
1055 sizeof(struct obd_statfs))))
1059 case OBD_IOC_LOV_GET_CONFIG: {
1060 struct obd_ioctl_data *data;
1061 struct lov_desc *desc;
1066 if (obd_ioctl_getdata(&buf, &len, uarg))
1069 data = (struct obd_ioctl_data *)buf;
1071 if (sizeof(*desc) > data->ioc_inllen1) {
1072 OBD_FREE_LARGE(buf, len);
1076 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1077 OBD_FREE_LARGE(buf, len);
1081 if (sizeof(__u32) * count > data->ioc_inllen3) {
1082 OBD_FREE_LARGE(buf, len);
1086 desc = (struct lov_desc *)data->ioc_inlbuf1;
1087 memcpy(desc, &(lov->desc), sizeof(*desc));
1089 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1090 genp = (__u32 *)data->ioc_inlbuf3;
1091 /* the uuid will be empty for deleted OSTs */
1092 for (i = 0; i < count; i++, uuidp++, genp++) {
1093 if (!lov->lov_tgts[i])
1095 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1096 *genp = lov->lov_tgts[i]->ltd_gen;
1099 if (copy_to_user(uarg, buf, len))
1101 OBD_FREE_LARGE(buf, len);
1104 case OBD_IOC_QUOTACTL: {
1105 struct if_quotactl *qctl = karg;
1106 struct lov_tgt_desc *tgt = NULL;
1107 struct obd_quotactl *oqctl;
1109 if (qctl->qc_valid == QC_OSTIDX) {
1110 if (count <= qctl->qc_idx)
1113 tgt = lov->lov_tgts[qctl->qc_idx];
1114 if (!tgt || !tgt->ltd_exp)
1116 } else if (qctl->qc_valid == QC_UUID) {
1117 for (i = 0; i < count; i++) {
1118 tgt = lov->lov_tgts[i];
1120 !obd_uuid_equals(&tgt->ltd_uuid,
1124 if (tgt->ltd_exp == NULL)
1136 LASSERT(tgt && tgt->ltd_exp);
1137 OBD_ALLOC_PTR(oqctl);
1141 QCTL_COPY(oqctl, qctl);
1142 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1144 QCTL_COPY(qctl, oqctl);
1145 qctl->qc_valid = QC_OSTIDX;
1146 qctl->obd_uuid = tgt->ltd_uuid;
1148 OBD_FREE_PTR(oqctl);
1157 for (i = 0; i < count; i++) {
1159 struct obd_device *osc_obd;
1161 /* OST was disconnected */
1162 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1165 /* ll_umount_begin() sets force flag but for lov, not
1166 * osc. Let's pass it through */
1167 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1168 osc_obd->obd_force = obddev->obd_force;
1169 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1172 if (lov->lov_tgts[i]->ltd_active) {
1173 CDEBUG(err == -ENOTTY ?
1174 D_IOCTL : D_WARNING,
1175 "iocontrol OSC %s on OST "
1176 "idx %d cmd %x: err = %d\n",
1177 lov_uuid2str(lov, i),
1194 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1195 __u32 keylen, void *key, __u32 *vallen, void *val)
1197 struct obd_device *obddev = class_exp2obd(exp);
1198 struct lov_obd *lov = &obddev->u.lov;
1199 struct lov_desc *ld = &lov->desc;
1203 if (vallen == NULL || val == NULL)
1208 if (KEY_IS(KEY_MAX_EASIZE)) {
1209 u32 max_stripe_count = min_t(u32, ld->ld_active_tgt_count,
1210 LOV_MAX_STRIPE_COUNT);
1212 *((u32 *)val) = lov_mds_md_size(max_stripe_count, LOV_MAGIC_V3);
1213 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1214 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1215 LOV_MAX_STRIPE_COUNT);
1217 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1218 } else if (KEY_IS(KEY_TGT_COUNT)) {
1219 *((int *)val) = lov->desc.ld_tgt_count;
1229 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1230 __u32 keylen, void *key,
1231 __u32 vallen, void *val,
1232 struct ptlrpc_request_set *set)
1234 struct obd_device *obddev = class_exp2obd(exp);
1235 struct lov_obd *lov = &obddev->u.lov;
1236 struct lov_tgt_desc *tgt;
1237 bool do_inactive = false, no_set = false;
1246 set = ptlrpc_prep_set();
1253 if (KEY_IS(KEY_CHECKSUM)) {
1255 } else if (KEY_IS(KEY_CACHE_SET)) {
1256 LASSERT(lov->lov_cache == NULL);
1257 lov->lov_cache = val;
1259 cl_cache_incref(lov->lov_cache);
1262 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1263 tgt = lov->lov_tgts[i];
1265 /* OST was disconnected */
1266 if (tgt == NULL || tgt->ltd_exp == NULL)
1269 /* OST is inactive and we don't want inactive OSCs */
1270 if (!tgt->ltd_active && !do_inactive)
1273 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1280 /* cycle through MDC target for Data-on-MDT */
1281 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1282 struct obd_device *mdc;
1284 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1288 err = obd_set_info_async(env, mdc->obd_self_export,
1289 keylen, key, vallen, val, set);
1296 err = ptlrpc_set_wait(set);
1299 ptlrpc_set_destroy(set);
1304 void lov_stripe_lock(struct lov_stripe_md *md)
1305 __acquires(&md->lsm_lock)
1307 LASSERT(md->lsm_lock_owner != current_pid());
1308 spin_lock(&md->lsm_lock);
1309 LASSERT(md->lsm_lock_owner == 0);
1310 md->lsm_lock_owner = current_pid();
1313 void lov_stripe_unlock(struct lov_stripe_md *md)
1314 __releases(&md->lsm_lock)
1316 LASSERT(md->lsm_lock_owner == current_pid());
1317 md->lsm_lock_owner = 0;
1318 spin_unlock(&md->lsm_lock);
1321 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1322 struct obd_quotactl *oqctl)
1324 struct lov_obd *lov = &obd->u.lov;
1325 struct lov_tgt_desc *tgt;
1327 __u64 bhardlimit = 0;
1331 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1332 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA) {
1333 CERROR("%s: bad quota opc %x for lov obd\n",
1334 obd->obd_name, oqctl->qc_cmd);
1340 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1343 tgt = lov->lov_tgts[i];
1348 if (!tgt->ltd_active || tgt->ltd_reap) {
1349 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1350 lov->lov_tgts[i]->ltd_activate) {
1352 CERROR("ost %d is inactive\n", i);
1354 CDEBUG(D_HA, "ost %d is inactive\n", i);
1359 err = obd_quotactl(tgt->ltd_exp, oqctl);
1361 if (tgt->ltd_active && !rc)
1366 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1367 curspace += oqctl->qc_dqblk.dqb_curspace;
1368 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1373 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1374 oqctl->qc_dqblk.dqb_curspace = curspace;
1375 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1380 static struct obd_ops lov_obd_ops = {
1381 .o_owner = THIS_MODULE,
1382 .o_setup = lov_setup,
1383 .o_cleanup = lov_cleanup,
1384 .o_connect = lov_connect,
1385 .o_disconnect = lov_disconnect,
1386 .o_statfs = lov_statfs,
1387 .o_iocontrol = lov_iocontrol,
1388 .o_get_info = lov_get_info,
1389 .o_set_info_async = lov_set_info_async,
1390 .o_notify = lov_notify,
1391 .o_pool_new = lov_pool_new,
1392 .o_pool_rem = lov_pool_remove,
1393 .o_pool_add = lov_pool_add,
1394 .o_pool_del = lov_pool_del,
1395 .o_getref = lov_getref,
1396 .o_putref = lov_putref,
1397 .o_quotactl = lov_quotactl,
1400 struct kmem_cache *lov_oinfo_slab;
1402 static int __init lov_init(void)
1404 bool enable_proc = true;
1405 struct obd_type *type;
1409 /* print an address of _any_ initialized kernel symbol from this
1410 * module, to allow debugging with gdb that doesn't support data
1411 * symbols from modules.*/
1412 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1414 rc = lu_kmem_init(lov_caches);
1418 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1419 sizeof(struct lov_oinfo), 0,
1420 SLAB_HWCACHE_ALIGN, NULL);
1421 if (lov_oinfo_slab == NULL) {
1422 lu_kmem_fini(lov_caches);
1426 type = class_search_type(LUSTRE_LOD_NAME);
1427 if (type != NULL && type->typ_procsym != NULL)
1428 enable_proc = false;
1430 rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
1431 LUSTRE_LOV_NAME, &lov_device_type);
1434 kmem_cache_destroy(lov_oinfo_slab);
1435 lu_kmem_fini(lov_caches);
1441 static void __exit lov_exit(void)
1443 class_unregister_type(LUSTRE_LOV_NAME);
1444 kmem_cache_destroy(lov_oinfo_slab);
1445 lu_kmem_fini(lov_caches);
1448 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1449 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1450 MODULE_VERSION(LUSTRE_VERSION_STRING);
1451 MODULE_LICENSE("GPL");
1453 module_init(lov_init);
1454 module_exit(lov_exit);