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;
328 /* Only disconnect the underlying layers on the final disconnect. */
330 if (lov->lov_connects != 0) {
331 /* why should there be more than 1 connect? */
332 CWARN("%s: unexpected disconnect #%d\n",
333 obd->obd_name, lov->lov_connects);
337 /* hold another ref so lov_del_obd() doesn't spin in putref each time */
340 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
341 if (lov->lov_tgts[index] && lov->lov_tgts[index]->ltd_exp) {
342 /* Disconnection is the last we know about an OBD */
343 lov_del_target(obd, index, NULL,
344 lov->lov_tgts[index]->ltd_gen);
349 if (lov->targets_proc_entry != NULL)
350 lprocfs_remove(&lov->targets_proc_entry);
353 rc = class_disconnect(exp); /* bz 9811 */
359 * -EINVAL : UUID can't be found in the LOV's target list
360 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
361 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
362 * any >= 0 : is log target index
364 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
365 enum obd_notify_event ev)
367 struct lov_obd *lov = &obd->u.lov;
368 struct lov_tgt_desc *tgt;
369 int index, activate, active;
372 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
373 lov, uuid->uuid, ev);
376 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
377 tgt = lov->lov_tgts[index];
381 * LU-642, initially inactive OSC could miss the obd_connect,
382 * we make up for it here.
384 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
385 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
386 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
388 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
389 &lov_osc_uuid, &lov->lov_ocd, NULL);
394 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
395 index, obd_uuid2str(&tgt->ltd_uuid),
396 tgt->ltd_exp->exp_handle.h_cookie);
397 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
401 if (index == lov->desc.ld_tgt_count)
402 GOTO(out, index = -EINVAL);
404 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
405 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
407 if (lov->lov_tgts[index]->ltd_activate == activate) {
408 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
409 uuid->uuid, activate ? "" : "de");
411 lov->lov_tgts[index]->ltd_activate = activate;
412 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
413 activate ? "" : "de", obd_uuid2str(uuid));
416 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
417 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
419 if (lov->lov_tgts[index]->ltd_active == active) {
420 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
421 uuid->uuid, active ? "" : "in");
424 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
425 obd_uuid2str(uuid), active ? "" : "in");
428 lov->lov_tgts[index]->ltd_active = active;
430 lov->desc.ld_active_tgt_count++;
431 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
433 lov->desc.ld_active_tgt_count--;
434 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
437 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
446 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
447 enum obd_notify_event ev)
450 struct lov_obd *lov = &obd->u.lov;
453 down_read(&lov->lov_notify_lock);
454 if (!lov->lov_connects)
455 GOTO(out_notify_lock, rc = 0);
457 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
458 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
459 struct obd_uuid *uuid;
463 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
464 CERROR("unexpected notification of %s %s\n",
465 watched->obd_type->typ_name, watched->obd_name);
466 GOTO(out_notify_lock, rc = -EINVAL);
469 uuid = &watched->u.cli.cl_target_uuid;
471 /* Set OSC as active before notifying the observer, so the
472 * observer can use the OSC normally.
474 rc = lov_set_osc_active(obd, uuid, ev);
476 CERROR("%s: event %d failed: rc = %d\n", obd->obd_name,
478 GOTO(out_notify_lock, rc);
482 /* Pass the notification up the chain. */
483 rc = obd_notify_observer(obd, watched, ev);
486 up_read(&lov->lov_notify_lock);
491 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
492 u32 index, int gen, int active)
494 struct lov_obd *lov = &obd->u.lov;
495 struct lov_tgt_desc *tgt;
496 struct obd_device *tgt_obd;
500 CDEBUG(D_CONFIG, "uuid:%s idx:%u gen:%d active:%d\n",
501 uuidp->uuid, index, gen, active);
504 CERROR("%s: request to add '%s' with invalid generation: %d\n",
505 obd->obd_name, uuidp->uuid, gen);
509 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, &obd->obd_uuid);
513 mutex_lock(&lov->lov_lock);
515 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
516 tgt = lov->lov_tgts[index];
518 CERROR("%s: UUID %s already assigned at index %d: rc = %d\n",
519 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), index, rc);
520 mutex_unlock(&lov->lov_lock);
524 if (index >= lov->lov_tgt_size) {
525 /* We need to reallocate the lov target array. */
526 struct lov_tgt_desc **newtgts, **old = NULL;
527 __u32 newsize, oldsize = 0;
529 newsize = max(lov->lov_tgt_size, (__u32)2);
530 while (newsize < index + 1)
531 newsize = newsize << 1;
532 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
533 if (newtgts == NULL) {
534 mutex_unlock(&lov->lov_lock);
538 if (lov->lov_tgt_size) {
539 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
542 oldsize = lov->lov_tgt_size;
545 lov->lov_tgts = newtgts;
546 lov->lov_tgt_size = newsize;
549 OBD_FREE(old, sizeof(*old) * oldsize);
551 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
552 lov->lov_tgts, lov->lov_tgt_size);
557 mutex_unlock(&lov->lov_lock);
561 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
563 mutex_unlock(&lov->lov_lock);
568 tgt->ltd_uuid = *uuidp;
569 tgt->ltd_obd = tgt_obd;
570 /* XXX - add a sanity check on the generation number. */
572 tgt->ltd_index = index;
573 tgt->ltd_activate = active;
574 lov->lov_tgts[index] = tgt;
575 if (index >= lov->desc.ld_tgt_count)
576 lov->desc.ld_tgt_count = index + 1;
578 mutex_unlock(&lov->lov_lock);
580 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
581 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
583 if (lov->lov_connects == 0) {
584 /* lov_connect hasn't been called yet. We'll do the
585 lov_connect_obd on this target when that fn first runs,
586 because we don't know the connect flags yet. */
592 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
596 /* connect to administrative disabled ost */
600 if (lov->lov_cache != NULL) {
601 rc = obd_set_info_async(NULL, tgt->ltd_exp,
602 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
603 sizeof(struct cl_client_cache), lov->lov_cache,
609 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
610 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
614 CERROR("%s: add failed, deleting %s: rc = %d\n",
615 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), rc);
616 lov_del_target(obd, index, NULL, 0);
622 /* Schedule a target for deletion */
623 int lov_del_target(struct obd_device *obd, u32 index,
624 struct obd_uuid *uuidp, int gen)
626 struct lov_obd *lov = &obd->u.lov;
627 int count = lov->desc.ld_tgt_count;
631 if (index >= count) {
632 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
637 /* to make sure there's no ongoing lov_notify() now */
638 down_write(&lov->lov_notify_lock);
641 if (!lov->lov_tgts[index]) {
642 CERROR("LOV target at index %d is not setup.\n", index);
643 GOTO(out, rc = -EINVAL);
646 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
647 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
648 lov_uuid2str(lov, index), index,
649 obd_uuid2str(uuidp));
650 GOTO(out, rc = -EINVAL);
653 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
654 lov_uuid2str(lov, index), index,
655 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
656 lov->lov_tgts[index]->ltd_active);
658 lov->lov_tgts[index]->ltd_reap = 1;
659 lov->lov_death_row++;
660 /* we really delete it from obd_putref */
663 up_write(&lov->lov_notify_lock);
668 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
670 struct obd_device *osc_obd;
673 LASSERT(tgt->ltd_reap);
675 osc_obd = class_exp2obd(tgt->ltd_exp);
677 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
679 osc_obd ? osc_obd->obd_name : "<no obd>");
682 lov_disconnect_obd(obd, tgt);
686 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
687 do it ourselves. And we can't do it from lov_cleanup,
688 because we just lost our only reference to it. */
690 class_manual_cleanup(osc_obd);
693 void lov_fix_desc_stripe_size(__u64 *val)
695 if (*val < LOV_MIN_STRIPE_SIZE) {
697 LCONSOLE_INFO("Increasing default stripe size to "
699 LOV_DESC_STRIPE_SIZE_DEFAULT);
700 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
701 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
702 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
703 LCONSOLE_WARN("Changing default stripe size to %llu (a "
705 *val, LOV_MIN_STRIPE_SIZE);
709 void lov_fix_desc_stripe_count(__u32 *val)
715 void lov_fix_desc_pattern(__u32 *val)
717 /* from lov_setstripe */
718 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
719 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
724 void lov_fix_desc_qos_maxage(__u32 *val)
727 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
730 void lov_fix_desc(struct lov_desc *desc)
732 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
733 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
734 lov_fix_desc_pattern(&desc->ld_pattern);
735 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
738 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
740 struct lov_desc *desc;
741 struct lov_obd *lov = &obd->u.lov;
742 #ifdef CONFIG_PROC_FS
743 struct obd_type *type;
748 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
749 CERROR("LOV setup requires a descriptor\n");
753 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
755 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
756 CERROR("descriptor size wrong: %d > %d\n",
757 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
761 if (desc->ld_magic != LOV_DESC_MAGIC) {
762 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
763 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
764 obd->obd_name, desc);
765 lustre_swab_lov_desc(desc);
767 CERROR("%s: Bad lov desc magic: %#x\n",
768 obd->obd_name, desc->ld_magic);
775 desc->ld_active_tgt_count = 0;
777 lov->lov_tgt_size = 0;
779 mutex_init(&lov->lov_lock);
780 atomic_set(&lov->lov_refcount, 0);
781 lov->lov_sp_me = LUSTRE_SP_CLI;
783 init_rwsem(&lov->lov_notify_lock);
785 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
787 HASH_POOLS_BKT_BITS, 0,
790 &pool_hash_operations,
792 INIT_LIST_HEAD(&lov->lov_pool_list);
793 lov->lov_pool_count = 0;
794 rc = lov_ost_pool_init(&lov->lov_packed, 0);
798 obd->obd_vars = lprocfs_lov_obd_vars;
799 #ifdef CONFIG_PROC_FS
800 /* If this is true then both client (lov) and server
801 * (lod) are on the same node. The lod layer if loaded
802 * first will register the lov proc directory. In that
803 * case obd->obd_type->typ_procroot will be not set.
804 * Instead we use type->typ_procsym as the parent.
806 type = class_search_type(LUSTRE_LOD_NAME);
807 if (type && type->typ_procsym) {
808 obd->obd_proc_entry = lprocfs_register(obd->obd_name,
811 if (IS_ERR(obd->obd_proc_entry)) {
812 rc = PTR_ERR(obd->obd_proc_entry);
813 CERROR("error %d setting up lprocfs for %s\n", rc,
815 obd->obd_proc_entry = NULL;
820 rc = lprocfs_obd_setup(obd, false);
824 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", 0444,
825 &lov_proc_target_fops, obd);
827 CWARN("%s: Error adding the target_obd file : rc %d\n",
830 lov->lov_pool_proc_entry = lprocfs_register("pools",
833 if (IS_ERR(lov->lov_pool_proc_entry)) {
834 rc = PTR_ERR(lov->lov_pool_proc_entry);
835 CERROR("%s: error setting up ldebugfs for pools : rc %d\n",
837 lov->lov_pool_proc_entry = NULL;
843 static int lov_cleanup(struct obd_device *obd)
845 struct lov_obd *lov = &obd->u.lov;
846 struct list_head *pos, *tmp;
847 struct pool_desc *pool;
850 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
851 pool = list_entry(pos, struct pool_desc, pool_list);
852 /* free pool structs */
853 CDEBUG(D_INFO, "delete pool %p\n", pool);
854 /* In the function below, .hs_keycmp resolves to
855 * pool_hashkey_keycmp() */
856 /* coverity[overrun-buffer-val] */
857 lov_pool_del(obd, pool->pool_name);
859 cfs_hash_putref(lov->lov_pools_hash_body);
860 lov_ost_pool_free(&lov->lov_packed);
862 lprocfs_obd_cleanup(obd);
866 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
867 if (!lov->lov_tgts[i])
870 /* Inactive targets may never have connected */
871 if (lov->lov_tgts[i]->ltd_active ||
872 atomic_read(&lov->lov_refcount))
873 /* We should never get here - these
874 * should have been removed in the
876 CERROR("%s: lov tgt %d not cleaned! "
877 "deathrow=%d, lovrc=%d\n",
878 obd->obd_name, i, lov->lov_death_row,
879 atomic_read(&lov->lov_refcount));
880 lov_del_target(obd, i, NULL, 0);
883 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
885 lov->lov_tgt_size = 0;
888 if (lov->lov_cache != NULL) {
889 cl_cache_decref(lov->lov_cache);
890 lov->lov_cache = NULL;
896 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
897 u32 *indexp, int *genp)
899 struct obd_uuid obd_uuid;
904 switch (cmd = lcfg->lcfg_command) {
908 case LCFG_LOV_ADD_OBD:
909 case LCFG_LOV_ADD_INA:
910 case LCFG_LOV_DEL_OBD: {
914 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
915 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
916 GOTO(out, rc = -EINVAL);
918 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
920 rc = kstrtou32(lustre_cfg_buf(lcfg, 2), 10, indexp);
923 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
928 if (cmd == LCFG_LOV_ADD_OBD)
929 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
930 else if (cmd == LCFG_LOV_ADD_INA)
931 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
933 rc = lov_del_target(obd, index, &obd_uuid, gen);
938 struct lov_desc *desc = &(obd->u.lov.desc);
941 GOTO(out, rc = -EINVAL);
943 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
956 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
957 GOTO(out, rc = -EINVAL);
965 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
966 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
968 struct obd_device *obd = class_exp2obd(exp);
969 struct lov_obd *lov = &obd->u.lov;
970 struct obd_info oinfo = {
974 struct ptlrpc_request_set *rqset;
975 struct lov_request_set *set = NULL;
976 struct lov_request *req;
982 rqset = ptlrpc_prep_set();
986 rc = lov_prep_statfs_set(obd, &oinfo, &set);
990 list_for_each_entry(req, &set->set_list, rq_link) {
991 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
992 &req->rq_oi, max_age, rqset);
997 rc = ptlrpc_set_wait(rqset);
1001 atomic_set(&set->set_completes, 0);
1003 rc2 = lov_fini_statfs_set(set);
1008 ptlrpc_set_destroy(rqset);
1013 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1014 void *karg, void __user *uarg)
1016 struct obd_device *obddev = class_exp2obd(exp);
1017 struct lov_obd *lov = &obddev->u.lov;
1018 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1019 struct obd_uuid *uuidp;
1023 case IOC_OBD_STATFS: {
1024 struct obd_ioctl_data *data = karg;
1025 struct obd_device *osc_obd;
1026 struct obd_statfs stat_buf = {0};
1030 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
1031 if ((index >= count))
1034 if (!lov->lov_tgts[index])
1035 /* Try again with the next index */
1037 if (!lov->lov_tgts[index]->ltd_active)
1040 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1045 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1046 min_t(unsigned long, data->ioc_plen2,
1047 sizeof(struct obd_uuid))))
1050 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1051 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1053 /* got statfs data */
1054 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1055 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1059 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1060 min_t(unsigned long, data->ioc_plen1,
1061 sizeof(struct obd_statfs))))
1065 case OBD_IOC_LOV_GET_CONFIG: {
1066 struct obd_ioctl_data *data;
1067 struct lov_desc *desc;
1072 if (obd_ioctl_getdata(&buf, &len, uarg))
1075 data = (struct obd_ioctl_data *)buf;
1077 if (sizeof(*desc) > data->ioc_inllen1) {
1078 OBD_FREE_LARGE(buf, len);
1082 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1083 OBD_FREE_LARGE(buf, len);
1087 if (sizeof(__u32) * count > data->ioc_inllen3) {
1088 OBD_FREE_LARGE(buf, len);
1092 desc = (struct lov_desc *)data->ioc_inlbuf1;
1093 memcpy(desc, &(lov->desc), sizeof(*desc));
1095 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1096 genp = (__u32 *)data->ioc_inlbuf3;
1097 /* the uuid will be empty for deleted OSTs */
1098 for (i = 0; i < count; i++, uuidp++, genp++) {
1099 if (!lov->lov_tgts[i])
1101 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1102 *genp = lov->lov_tgts[i]->ltd_gen;
1105 if (copy_to_user(uarg, buf, len))
1107 OBD_FREE_LARGE(buf, len);
1110 case OBD_IOC_QUOTACTL: {
1111 struct if_quotactl *qctl = karg;
1112 struct lov_tgt_desc *tgt = NULL;
1113 struct obd_quotactl *oqctl;
1115 if (qctl->qc_valid == QC_OSTIDX) {
1116 if (count <= qctl->qc_idx)
1119 tgt = lov->lov_tgts[qctl->qc_idx];
1120 if (!tgt || !tgt->ltd_exp)
1122 } else if (qctl->qc_valid == QC_UUID) {
1123 for (i = 0; i < count; i++) {
1124 tgt = lov->lov_tgts[i];
1126 !obd_uuid_equals(&tgt->ltd_uuid,
1130 if (tgt->ltd_exp == NULL)
1142 LASSERT(tgt && tgt->ltd_exp);
1143 OBD_ALLOC_PTR(oqctl);
1147 QCTL_COPY(oqctl, qctl);
1148 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1150 QCTL_COPY(qctl, oqctl);
1151 qctl->qc_valid = QC_OSTIDX;
1152 qctl->obd_uuid = tgt->ltd_uuid;
1154 OBD_FREE_PTR(oqctl);
1163 for (i = 0; i < count; i++) {
1165 struct obd_device *osc_obd;
1167 /* OST was disconnected */
1168 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1171 /* ll_umount_begin() sets force flag but for lov, not
1172 * osc. Let's pass it through */
1173 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1174 osc_obd->obd_force = obddev->obd_force;
1175 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1178 if (lov->lov_tgts[i]->ltd_active) {
1179 CDEBUG(err == -ENOTTY ?
1180 D_IOCTL : D_WARNING,
1181 "iocontrol OSC %s on OST "
1182 "idx %d cmd %x: err = %d\n",
1183 lov_uuid2str(lov, i),
1200 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1201 __u32 keylen, void *key, __u32 *vallen, void *val)
1203 struct obd_device *obddev = class_exp2obd(exp);
1204 struct lov_obd *lov = &obddev->u.lov;
1205 struct lov_desc *ld = &lov->desc;
1209 if (vallen == NULL || val == NULL)
1214 if (KEY_IS(KEY_MAX_EASIZE)) {
1215 u32 max_stripe_count = min_t(u32, ld->ld_active_tgt_count,
1216 LOV_MAX_STRIPE_COUNT);
1218 *((u32 *)val) = lov_mds_md_size(max_stripe_count, LOV_MAGIC_V3);
1219 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1220 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1221 LOV_MAX_STRIPE_COUNT);
1223 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1224 } else if (KEY_IS(KEY_TGT_COUNT)) {
1225 *((int *)val) = lov->desc.ld_tgt_count;
1235 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1236 __u32 keylen, void *key,
1237 __u32 vallen, void *val,
1238 struct ptlrpc_request_set *set)
1240 struct obd_device *obddev = class_exp2obd(exp);
1241 struct lov_obd *lov = &obddev->u.lov;
1242 struct lov_tgt_desc *tgt;
1243 bool do_inactive = false, no_set = false;
1252 set = ptlrpc_prep_set();
1259 if (KEY_IS(KEY_CHECKSUM)) {
1261 } else if (KEY_IS(KEY_CACHE_SET)) {
1262 LASSERT(lov->lov_cache == NULL);
1263 lov->lov_cache = val;
1265 cl_cache_incref(lov->lov_cache);
1268 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1269 tgt = lov->lov_tgts[i];
1271 /* OST was disconnected */
1272 if (tgt == NULL || tgt->ltd_exp == NULL)
1275 /* OST is inactive and we don't want inactive OSCs */
1276 if (!tgt->ltd_active && !do_inactive)
1279 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1286 /* cycle through MDC target for Data-on-MDT */
1287 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1288 struct obd_device *mdc;
1290 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1294 err = obd_set_info_async(env, mdc->obd_self_export,
1295 keylen, key, vallen, val, set);
1302 err = ptlrpc_set_wait(set);
1305 ptlrpc_set_destroy(set);
1310 void lov_stripe_lock(struct lov_stripe_md *md)
1311 __acquires(&md->lsm_lock)
1313 LASSERT(md->lsm_lock_owner != current_pid());
1314 spin_lock(&md->lsm_lock);
1315 LASSERT(md->lsm_lock_owner == 0);
1316 md->lsm_lock_owner = current_pid();
1319 void lov_stripe_unlock(struct lov_stripe_md *md)
1320 __releases(&md->lsm_lock)
1322 LASSERT(md->lsm_lock_owner == current_pid());
1323 md->lsm_lock_owner = 0;
1324 spin_unlock(&md->lsm_lock);
1327 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1328 struct obd_quotactl *oqctl)
1330 struct lov_obd *lov = &obd->u.lov;
1331 struct lov_tgt_desc *tgt;
1333 __u64 bhardlimit = 0;
1337 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1338 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA) {
1339 CERROR("%s: bad quota opc %x for lov obd\n",
1340 obd->obd_name, oqctl->qc_cmd);
1346 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1349 tgt = lov->lov_tgts[i];
1354 if (!tgt->ltd_active || tgt->ltd_reap) {
1355 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1356 lov->lov_tgts[i]->ltd_activate) {
1358 CERROR("ost %d is inactive\n", i);
1360 CDEBUG(D_HA, "ost %d is inactive\n", i);
1365 err = obd_quotactl(tgt->ltd_exp, oqctl);
1367 if (tgt->ltd_active && !rc)
1372 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1373 curspace += oqctl->qc_dqblk.dqb_curspace;
1374 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1379 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1380 oqctl->qc_dqblk.dqb_curspace = curspace;
1381 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1386 static struct obd_ops lov_obd_ops = {
1387 .o_owner = THIS_MODULE,
1388 .o_setup = lov_setup,
1389 .o_cleanup = lov_cleanup,
1390 .o_connect = lov_connect,
1391 .o_disconnect = lov_disconnect,
1392 .o_statfs = lov_statfs,
1393 .o_iocontrol = lov_iocontrol,
1394 .o_get_info = lov_get_info,
1395 .o_set_info_async = lov_set_info_async,
1396 .o_notify = lov_notify,
1397 .o_pool_new = lov_pool_new,
1398 .o_pool_rem = lov_pool_remove,
1399 .o_pool_add = lov_pool_add,
1400 .o_pool_del = lov_pool_del,
1401 .o_getref = lov_getref,
1402 .o_putref = lov_putref,
1403 .o_quotactl = lov_quotactl,
1406 struct kmem_cache *lov_oinfo_slab;
1408 static int __init lov_init(void)
1410 bool enable_proc = true;
1411 struct obd_type *type;
1415 /* print an address of _any_ initialized kernel symbol from this
1416 * module, to allow debugging with gdb that doesn't support data
1417 * symbols from modules.*/
1418 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1420 rc = lu_kmem_init(lov_caches);
1424 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1425 sizeof(struct lov_oinfo), 0,
1426 SLAB_HWCACHE_ALIGN, NULL);
1427 if (lov_oinfo_slab == NULL) {
1428 lu_kmem_fini(lov_caches);
1432 type = class_search_type(LUSTRE_LOD_NAME);
1433 if (type != NULL && type->typ_procsym != NULL)
1434 enable_proc = false;
1436 rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
1437 LUSTRE_LOV_NAME, &lov_device_type);
1440 kmem_cache_destroy(lov_oinfo_slab);
1441 lu_kmem_fini(lov_caches);
1447 static void __exit lov_exit(void)
1449 class_unregister_type(LUSTRE_LOV_NAME);
1450 kmem_cache_destroy(lov_oinfo_slab);
1451 lu_kmem_fini(lov_caches);
1454 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1455 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1456 MODULE_VERSION(LUSTRE_VERSION_STRING);
1457 MODULE_LICENSE("GPL");
1459 module_init(lov_init);
1460 module_exit(lov_exit);