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/
31 * lustre/lov/lov_obd.c
33 * Author: Phil Schwan <phil@clusterfs.com>
34 * Author: Peter Braam <braam@clusterfs.com>
35 * Author: Mike Shaver <shaver@clusterfs.com>
36 * Author: Nathan Rutman <nathan@clusterfs.com>
39 #define DEBUG_SUBSYSTEM S_LOV
40 #include <libcfs/libcfs.h>
42 #include <cl_object.h>
43 #include <lustre_dlm.h>
44 #include <lustre_fid.h>
45 #include <uapi/linux/lustre/lustre_ioctl.h>
46 #include <lustre_ioctl_old.h>
47 #include <lustre_lib.h>
48 #include <lustre_mds.h>
49 #include <lustre_net.h>
50 #include <uapi/linux/lustre/lustre_param.h>
51 #include <lustre_swab.h>
52 #include <lprocfs_status.h>
53 #include <obd_class.h>
54 #include <obd_support.h>
56 #include "lov_internal.h"
58 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
59 Any function that expects lov_tgts to remain stationary must take a ref. */
60 void lov_tgts_getref(struct obd_device *obd)
62 struct lov_obd *lov = &obd->u.lov;
64 /* nobody gets through here until lov_putref is done */
65 mutex_lock(&lov->lov_lock);
66 atomic_inc(&lov->lov_refcount);
67 mutex_unlock(&lov->lov_lock);
70 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
72 void lov_tgts_putref(struct obd_device *obd)
74 struct lov_obd *lov = &obd->u.lov;
76 mutex_lock(&lov->lov_lock);
77 /* ok to dec to 0 more than once -- ltd_exp's will be null */
78 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
80 struct lov_tgt_desc *tgt, *n;
83 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
85 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
86 tgt = lov->lov_tgts[i];
88 if (!tgt || !tgt->ltd_reap)
90 list_add(&tgt->ltd_kill, &kill);
91 /* XXX - right now there is a dependency on ld_tgt_count
92 * being the maximum tgt index for computing the
93 * mds_max_easize. So we can't shrink it. */
94 lu_tgt_pool_remove(&lov->lov_packed, i);
95 lov->lov_tgts[i] = NULL;
98 mutex_unlock(&lov->lov_lock);
100 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
101 list_del(&tgt->ltd_kill);
103 __lov_del_obd(obd, tgt);
106 mutex_unlock(&lov->lov_lock);
110 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
111 enum obd_notify_event ev);
113 int lov_connect_osc(struct obd_device *obd, u32 index, int activate,
114 struct obd_connect_data *data)
116 struct lov_obd *lov = &obd->u.lov;
117 struct obd_uuid *tgt_uuid;
118 struct obd_device *tgt_obd;
119 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
120 struct obd_import *imp;
124 if (lov->lov_tgts[index] == NULL)
127 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
128 tgt_obd = lov->lov_tgts[index]->ltd_obd;
130 if (!tgt_obd->obd_set_up) {
132 CERROR("%s: target not set up: rc = %d\n",
133 obd_uuid2str(tgt_uuid), rc);
137 /* override the sp_me from lov */
138 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
140 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
141 data->ocd_index = index;
144 * Divine LOV knows that OBDs under it are OSCs.
146 imp = tgt_obd->u.cli.cl_import;
149 tgt_obd->obd_no_recov = 0;
150 /* FIXME this is probably supposed to be
151 ptlrpc_set_import_active. Horrible naming. */
152 ptlrpc_activate_import(imp, false);
155 rc = obd_register_observer(tgt_obd, obd);
157 CERROR("%s: target register_observer error: rc = %d\n",
158 obd_uuid2str(tgt_uuid), rc);
162 if (imp->imp_invalid) {
163 CDEBUG(D_CONFIG, "%s: not connecting - administratively disabled\n",
164 obd_uuid2str(tgt_uuid));
168 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
169 &lov_osc_uuid, data, lov->lov_cache);
170 if (rc || !lov->lov_tgts[index]->ltd_exp) {
171 CERROR("%s: target connect error: rc = %d\n",
172 obd_uuid2str(tgt_uuid), rc);
173 obd_register_observer(tgt_obd, NULL);
177 lov->lov_tgts[index]->ltd_reap = 0;
179 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
180 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
182 if (lov->lov_tgts_kobj) {
183 /* Even if we failed, that's ok */
184 rc = sysfs_create_link(lov->lov_tgts_kobj,
185 &tgt_obd->obd_kset.kobj,
188 CERROR("%s: can't register LOV target /sys/fs/lustre/%s/%s/target_obds/%s : rc = %d\n",
189 obd->obd_name, obd->obd_type->typ_name,
191 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_name,
198 static int lov_connect(const struct lu_env *env,
199 struct obd_export **exp, struct obd_device *obd,
200 struct obd_uuid *cluuid, struct obd_connect_data *data,
203 struct lov_obd *lov = &obd->u.lov;
204 struct lov_tgt_desc *tgt;
205 struct lustre_handle conn;
209 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
211 rc = class_connect(&conn, obd, cluuid);
215 *exp = class_conn2export(&conn);
217 /* Why should there ever be more than 1 connect? */
219 LASSERT(lov->lov_connects == 1);
221 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
223 lov->lov_ocd = *data;
225 lov_tgts_getref(obd);
228 lov->lov_cache = localdata;
229 cl_cache_incref(lov->lov_cache);
232 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
233 tgt = lov->lov_tgts[i];
234 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
236 /* Flags will be lowest common denominator */
237 rc = lov_connect_osc(obd, i, tgt->ltd_activate, &lov->lov_ocd);
239 CERROR("%s: lov connect tgt %d failed: %d\n",
240 obd->obd_name, i, rc);
243 /* connect to administrative disabled ost */
244 if (!lov->lov_tgts[i]->ltd_exp)
247 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
250 CERROR("%s error sending notify %d\n",
255 lov_tgts_putref(obd);
260 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
262 struct lov_obd *lov = &obd->u.lov;
263 struct obd_device *osc_obd;
267 osc_obd = class_exp2obd(tgt->ltd_exp);
268 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n", obd->obd_name,
269 osc_obd ? osc_obd->obd_name : "<no obd>");
271 if (tgt->ltd_active) {
273 lov->desc.ld_active_tgt_count--;
274 tgt->ltd_exp->exp_obd->obd_inactive = 1;
278 if (lov->lov_tgts_kobj)
279 sysfs_remove_link(lov->lov_tgts_kobj,
282 /* Pass it on to our clients.
283 * XXX This should be an argument to disconnect,
284 * XXX not a back-door flag on the OBD. Ah well.
286 osc_obd->obd_force = obd->obd_force;
287 osc_obd->obd_fail = obd->obd_fail;
288 osc_obd->obd_no_recov = obd->obd_no_recov;
291 obd_register_observer(osc_obd, NULL);
293 rc = obd_disconnect(tgt->ltd_exp);
295 CERROR("Target %s disconnect error %d\n",
296 tgt->ltd_uuid.uuid, rc);
304 static int lov_disconnect(struct obd_export *exp)
306 struct obd_device *obd = class_exp2obd(exp);
307 struct lov_obd *lov = &obd->u.lov;
315 /* Only disconnect the underlying layers on the final disconnect. */
316 if (lov->lov_connects == 0) {
317 CWARN("%s: was disconnected already #%d\n",
318 obd->obd_name, lov->lov_connects);
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 */
331 lov_tgts_getref(obd);
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);
340 lov_tgts_putref(obd);
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;
360 bool activate, active;
363 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
364 lov, uuid->uuid, ev);
366 lov_tgts_getref(obd);
367 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
368 tgt = lov->lov_tgts[index];
369 if (tgt && obd_uuid_equals(uuid, &tgt->ltd_uuid))
373 if (index == lov->desc.ld_tgt_count)
374 GOTO(out, index = -EINVAL);
376 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
377 activate = (ev == OBD_NOTIFY_ACTIVATE);
380 * LU-642, initially inactive OSC could miss the obd_connect,
381 * we make up for it here.
383 if (activate && !tgt->ltd_exp) {
385 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
387 rc = obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
388 &lov_osc_uuid, &lov->lov_ocd,
390 if (rc || !tgt->ltd_exp)
391 GOTO(out, index = rc);
394 if (lov->lov_tgts[index]->ltd_activate == activate) {
395 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
396 uuid->uuid, activate ? "" : "de");
398 lov->lov_tgts[index]->ltd_activate = activate;
399 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
400 activate ? "" : "de", obd_uuid2str(uuid));
402 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
403 active = (ev == OBD_NOTIFY_ACTIVE);
405 if (lov->lov_tgts[index]->ltd_active == active) {
406 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
407 uuid->uuid, active ? "" : "in");
410 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
411 obd_uuid2str(uuid), active ? "" : "in");
413 lov->lov_tgts[index]->ltd_active = active;
415 lov->desc.ld_active_tgt_count++;
416 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
418 lov->desc.ld_active_tgt_count--;
419 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
422 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
427 CDEBUG(D_INFO, "%s: lov idx %d conn %llx\n", obd_uuid2str(uuid),
428 index, tgt->ltd_exp->exp_handle.h_cookie);
431 lov_tgts_putref(obd);
435 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
436 enum obd_notify_event ev)
439 struct lov_obd *lov = &obd->u.lov;
442 down_read(&lov->lov_notify_lock);
443 if (!lov->lov_connects)
444 GOTO(out_notify_lock, rc = 0);
446 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
447 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
448 struct obd_uuid *uuid;
452 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
453 CERROR("unexpected notification of %s %s\n",
454 watched->obd_type->typ_name, watched->obd_name);
455 GOTO(out_notify_lock, rc = -EINVAL);
458 uuid = &watched->u.cli.cl_target_uuid;
460 /* Set OSC as active before notifying the observer, so the
461 * observer can use the OSC normally.
463 rc = lov_set_osc_active(obd, uuid, ev);
465 CERROR("%s: event %d failed: rc = %d\n", obd->obd_name,
467 GOTO(out_notify_lock, rc);
471 /* Pass the notification up the chain. */
472 rc = obd_notify_observer(obd, watched, ev);
475 up_read(&lov->lov_notify_lock);
480 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
481 u32 index, int gen, int active)
483 struct lov_obd *lov = &obd->u.lov;
484 struct lov_tgt_desc *tgt;
485 struct obd_device *tgt_obd;
489 CDEBUG(D_CONFIG, "uuid:%s idx:%u gen:%d active:%d\n",
490 uuidp->uuid, index, gen, active);
493 CERROR("%s: request to add '%s' with invalid generation: %d\n",
494 obd->obd_name, uuidp->uuid, gen);
498 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, &obd->obd_uuid);
502 mutex_lock(&lov->lov_lock);
504 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
505 tgt = lov->lov_tgts[index];
507 CERROR("%s: UUID %s already assigned at index %d: rc = %d\n",
508 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), index, rc);
509 mutex_unlock(&lov->lov_lock);
513 if (index >= lov->lov_tgt_size) {
514 /* We need to reallocate the lov target array. */
515 struct lov_tgt_desc **newtgts, **old = NULL;
516 __u32 newsize, oldsize = 0;
518 newsize = max(lov->lov_tgt_size, 2U);
519 while (newsize < index + 1)
520 newsize = newsize << 1;
521 OBD_ALLOC_PTR_ARRAY(newtgts, newsize);
522 if (newtgts == NULL) {
523 mutex_unlock(&lov->lov_lock);
527 if (lov->lov_tgt_size) {
528 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
531 oldsize = lov->lov_tgt_size;
534 lov->lov_tgts = newtgts;
535 lov->lov_tgt_size = newsize;
538 OBD_FREE_PTR_ARRAY(old, oldsize);
540 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
541 lov->lov_tgts, lov->lov_tgt_size);
546 mutex_unlock(&lov->lov_lock);
550 rc = lu_tgt_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
552 mutex_unlock(&lov->lov_lock);
557 tgt->ltd_uuid = *uuidp;
558 tgt->ltd_obd = tgt_obd;
559 /* XXX - add a sanity check on the generation number. */
561 tgt->ltd_index = index;
562 tgt->ltd_activate = active;
563 lov->lov_tgts[index] = tgt;
564 if (index >= lov->desc.ld_tgt_count)
565 lov->desc.ld_tgt_count = index + 1;
567 mutex_unlock(&lov->lov_lock);
569 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
570 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
572 if (lov->lov_connects == 0) {
573 /* lov_connect hasn't been called yet. We'll do the
574 lov_connect_osc on this target when that fn first runs,
575 because we don't know the connect flags yet. */
579 lov_tgts_getref(obd);
581 rc = lov_connect_osc(obd, index, active, &lov->lov_ocd);
585 /* connect to administrative disabled ost */
589 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
590 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
594 CERROR("%s: add failed, deleting %s: rc = %d\n",
595 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), rc);
596 lov_del_target(obd, index, NULL, 0);
598 lov_tgts_putref(obd);
602 /* Schedule a target for deletion */
603 int lov_del_target(struct obd_device *obd, u32 index,
604 struct obd_uuid *uuidp, int gen)
606 struct lov_obd *lov = &obd->u.lov;
607 int count = lov->desc.ld_tgt_count;
611 if (index >= count) {
612 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
617 /* to make sure there's no ongoing lov_notify() now */
618 down_write(&lov->lov_notify_lock);
619 lov_tgts_getref(obd);
621 if (!lov->lov_tgts[index]) {
622 CERROR("LOV target at index %d is not setup.\n", index);
623 GOTO(out, rc = -EINVAL);
626 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
627 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
628 lov_uuid2str(lov, index), index,
629 obd_uuid2str(uuidp));
630 GOTO(out, rc = -EINVAL);
633 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
634 lov_uuid2str(lov, index), index,
635 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
636 lov->lov_tgts[index]->ltd_active);
638 lov->lov_tgts[index]->ltd_reap = 1;
639 lov->lov_death_row++;
640 /* we really delete it from lov_tgts_putref() */
642 lov_tgts_putref(obd);
643 up_write(&lov->lov_notify_lock);
648 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
650 struct obd_device *osc_obd;
653 LASSERT(tgt->ltd_reap);
655 osc_obd = class_exp2obd(tgt->ltd_exp);
657 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
659 osc_obd ? osc_obd->obd_name : "<no obd>");
662 lov_disconnect_obd(obd, tgt);
666 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
667 do it ourselves. And we can't do it from lov_cleanup,
668 because we just lost our only reference to it. */
670 class_manual_cleanup(osc_obd);
673 void lov_fix_desc_stripe_size(__u64 *val)
675 if (*val < LOV_MIN_STRIPE_SIZE) {
677 LCONSOLE_INFO("Increasing default stripe size to "
679 LOV_DESC_STRIPE_SIZE_DEFAULT);
680 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
681 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
682 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
683 LCONSOLE_WARN("Changing default stripe size to %llu (a "
685 *val, LOV_MIN_STRIPE_SIZE);
689 void lov_fix_desc_stripe_count(__u32 *val)
695 void lov_fix_desc_pattern(__u32 *val)
697 /* from lov_setstripe */
698 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
699 LCONSOLE_WARN("lov: Unknown stripe pattern: %#x\n", *val);
704 void lov_fix_desc_qos_maxage(__u32 *val)
707 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
710 void lov_fix_desc(struct lov_desc *desc)
712 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
713 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
714 lov_fix_desc_pattern(&desc->ld_pattern);
715 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
718 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
720 struct lov_desc *desc;
721 struct lov_obd *lov = &obd->u.lov;
725 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
726 CERROR("LOV setup requires a descriptor\n");
730 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
732 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
733 CERROR("descriptor size wrong: %d > %d\n",
734 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
738 if (desc->ld_magic != LOV_DESC_MAGIC) {
739 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
740 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
741 obd->obd_name, desc);
742 lustre_swab_lov_desc(desc);
744 CERROR("%s: Bad lov desc magic: %#x\n",
745 obd->obd_name, desc->ld_magic);
752 desc->ld_active_tgt_count = 0;
754 lov->lov_tgt_size = 0;
756 mutex_init(&lov->lov_lock);
757 atomic_set(&lov->lov_refcount, 0);
758 lov->lov_sp_me = LUSTRE_SP_CLI;
760 init_rwsem(&lov->lov_notify_lock);
762 INIT_LIST_HEAD(&lov->lov_pool_list);
763 lov->lov_pool_count = 0;
764 rc = lov_pool_hash_init(&lov->lov_pools_hash_body);
768 rc = lu_tgt_pool_init(&lov->lov_packed, 0);
772 rc = lov_tunables_init(obd);
776 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
777 &obd->obd_kset.kobj);
783 static int lov_cleanup(struct obd_device *obd)
785 struct lov_obd *lov = &obd->u.lov;
786 struct pool_desc *pool, *tmp;
789 if (lov->lov_tgts_kobj) {
790 kobject_put(lov->lov_tgts_kobj);
791 lov->lov_tgts_kobj = NULL;
794 list_for_each_entry_safe(pool, tmp, &lov->lov_pool_list, pool_list) {
795 /* free pool structs */
796 CDEBUG(D_INFO, "delete pool %p\n", pool);
797 /* In the function below, .hs_keycmp resolves to
798 * pool_hashkey_keycmp() */
799 lov_pool_del(obd, pool->pool_name);
801 lov_pool_hash_destroy(&lov->lov_pools_hash_body);
802 lu_tgt_pool_free(&lov->lov_packed);
804 lprocfs_obd_cleanup(obd);
807 lov_tgts_getref(obd);
808 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
809 if (!lov->lov_tgts[i])
812 /* Inactive targets may never have connected */
813 if (lov->lov_tgts[i]->ltd_active)
814 /* We should never get here - these
815 * should have been removed in the
817 CERROR("%s: lov tgt %d not cleaned! "
818 "deathrow=%d, lovrc=%d\n",
819 obd->obd_name, i, lov->lov_death_row,
820 atomic_read(&lov->lov_refcount));
821 lov_del_target(obd, i, NULL, 0);
823 lov_tgts_putref(obd);
824 OBD_FREE_PTR_ARRAY(lov->lov_tgts, lov->lov_tgt_size);
825 lov->lov_tgt_size = 0;
828 if (lov->lov_cache != NULL) {
829 cl_cache_decref(lov->lov_cache);
830 lov->lov_cache = NULL;
836 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
837 u32 *indexp, int *genp)
839 struct obd_uuid obd_uuid;
844 switch (cmd = lcfg->lcfg_command) {
848 case LCFG_LOV_ADD_OBD:
849 case LCFG_LOV_ADD_INA:
850 case LCFG_LOV_DEL_OBD: {
854 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
855 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
856 GOTO(out, rc = -EINVAL);
858 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
860 rc = kstrtou32(lustre_cfg_buf(lcfg, 2), 10, indexp);
863 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
868 if (cmd == LCFG_LOV_ADD_OBD)
869 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
870 else if (cmd == LCFG_LOV_ADD_INA)
871 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
873 rc = lov_del_target(obd, index, &obd_uuid, gen);
878 struct lov_desc *desc = &(obd->u.lov.desc);
882 GOTO(out, rc = -EINVAL);
884 count = class_modify_config(lcfg, PARAM_LOV,
885 &obd->obd_kset.kobj);
886 GOTO(out, rc = count < 0 ? count : 0);
895 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
896 GOTO(out, rc = -EINVAL);
904 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
905 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
907 struct obd_device *obd = class_exp2obd(exp);
908 struct lov_obd *lov = &obd->u.lov;
909 struct obd_info oinfo = {
913 struct ptlrpc_request_set *rqset;
914 struct lov_request_set *set = NULL;
915 struct lov_request *req;
921 rqset = ptlrpc_prep_set();
925 rc = lov_prep_statfs_set(obd, &oinfo, &set);
929 list_for_each_entry(req, &set->set_list, rq_link) {
930 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
931 &req->rq_oi, max_age, rqset);
936 rc = ptlrpc_set_wait(env, rqset);
940 atomic_set(&set->set_completes, 0);
942 rc2 = lov_fini_statfs_set(set);
947 ptlrpc_set_destroy(rqset);
952 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
953 void *karg, void __user *uarg)
955 struct obd_device *obd = class_exp2obd(exp);
956 struct lov_obd *lov = &obd->u.lov;
957 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
960 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
961 exp->exp_obd->obd_name, cmd, len, karg, uarg);
963 /* exit early for unknown ioctl types. */
964 if (unlikely(_IOC_TYPE(cmd) != 'f' && !IOC_OSC_SET_ACTIVE_ALLOW(cmd)))
965 RETURN(OBD_IOC_DEBUG(D_IOCTL, obd->obd_name, cmd, "unknown",
968 /* can't do a generic karg == NULL check here, since it is too noisy and
969 * we need to return -ENOTTY for unsupported ioctls instead of -EINVAL.
972 case IOC_OBD_STATFS: {
973 struct obd_ioctl_data *data;
974 struct obd_device *osc_obd;
975 struct obd_statfs stat_buf = {0};
976 struct obd_import *imp;
980 if (unlikely(karg == NULL))
981 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=null",
985 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
989 if (!lov->lov_tgts[index])
990 /* Try again with the next index */
993 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
997 imp = osc_obd->u.cli.cl_import;
998 if (!lov->lov_tgts[index]->ltd_active &&
999 imp->imp_state != LUSTRE_IMP_IDLE)
1003 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1004 min_t(unsigned long, data->ioc_plen2,
1005 sizeof(struct obd_uuid))))
1008 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1009 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1011 /* got statfs data */
1012 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1013 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1017 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1018 min_t(unsigned long, data->ioc_plen1,
1019 sizeof(struct obd_statfs))))
1023 case OBD_IOC_QUOTACTL: {
1024 struct if_quotactl *qctl;
1025 struct lov_tgt_desc *tgt = NULL;
1026 struct obd_quotactl *oqctl;
1027 struct obd_import *imp;
1029 if (unlikely(karg == NULL))
1030 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=null",
1034 if (qctl->qc_valid == QC_OSTIDX) {
1035 if (count <= qctl->qc_idx)
1038 tgt = lov->lov_tgts[qctl->qc_idx];
1044 } else if (qctl->qc_valid == QC_UUID) {
1045 for (i = 0; i < count; i++) {
1046 tgt = lov->lov_tgts[i];
1048 !obd_uuid_equals(&tgt->ltd_uuid,
1052 if (tgt->ltd_exp == NULL)
1064 LASSERT(tgt && tgt->ltd_exp);
1065 imp = class_exp2cliimp(tgt->ltd_exp);
1066 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
1067 qctl->qc_valid = QC_OSTIDX;
1068 qctl->obd_uuid = tgt->ltd_uuid;
1072 OBD_ALLOC_PTR(oqctl);
1076 QCTL_COPY(oqctl, qctl);
1077 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1079 QCTL_COPY(qctl, oqctl);
1080 qctl->qc_valid = QC_OSTIDX;
1081 qctl->obd_uuid = tgt->ltd_uuid;
1083 OBD_FREE_PTR(oqctl);
1092 for (i = 0; i < count; i++) {
1094 struct obd_device *osc_obd;
1096 /* OST was disconnected */
1097 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1100 /* ll_umount_begin() sets force on lov, pass to osc */
1101 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1103 osc_obd->obd_force = obd->obd_force;
1104 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1107 if (lov->lov_tgts[i]->ltd_active) {
1108 OBD_IOC_DEBUG(err == -ENOTTY ?
1109 D_IOCTL : D_WARNING,
1111 lov_uuid2str(lov, i),
1132 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1133 __u32 keylen, void *key, __u32 *vallen, void *val)
1135 struct obd_device *obd = class_exp2obd(exp);
1136 struct lov_obd *lov = &obd->u.lov;
1137 struct lov_desc *ld = &lov->desc;
1141 if (vallen == NULL || val == NULL)
1144 lov_tgts_getref(obd);
1146 if (KEY_IS(KEY_MAX_EASIZE)) {
1147 *((u32 *)val) = exp->exp_connect_data.ocd_max_easize;
1148 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1149 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1150 LOV_MAX_STRIPE_COUNT);
1152 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1153 } else if (KEY_IS(KEY_TGT_COUNT)) {
1154 *((int *)val) = lov->desc.ld_tgt_count;
1159 lov_tgts_putref(obd);
1164 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1165 __u32 keylen, void *key,
1166 __u32 vallen, void *val,
1167 struct ptlrpc_request_set *set)
1169 struct obd_device *obd = class_exp2obd(exp);
1170 struct lov_obd *lov = &obd->u.lov;
1171 struct lov_tgt_desc *tgt;
1172 bool do_inactive = false, no_set = false;
1181 set = ptlrpc_prep_set();
1186 lov_tgts_getref(obd);
1188 if (KEY_IS(KEY_CHECKSUM))
1191 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1192 tgt = lov->lov_tgts[i];
1194 /* OST was disconnected */
1195 if (tgt == NULL || tgt->ltd_exp == NULL)
1198 /* OST is inactive and we don't want inactive OSCs */
1199 if (!tgt->ltd_active && !do_inactive)
1202 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1209 /* cycle through MDC target for Data-on-MDT */
1210 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1211 struct obd_device *mdc;
1213 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1217 err = obd_set_info_async(env, mdc->obd_self_export,
1218 keylen, key, vallen, val, set);
1223 lov_tgts_putref(obd);
1225 err = ptlrpc_set_wait(env, set);
1228 ptlrpc_set_destroy(set);
1233 void lov_stripe_lock(struct lov_stripe_md *md)
1234 __acquires(&md->lsm_lock)
1236 LASSERT(md->lsm_lock_owner != current->pid);
1237 spin_lock(&md->lsm_lock);
1238 LASSERT(md->lsm_lock_owner == 0);
1239 md->lsm_lock_owner = current->pid;
1242 void lov_stripe_unlock(struct lov_stripe_md *md)
1243 __releases(&md->lsm_lock)
1245 LASSERT(md->lsm_lock_owner == current->pid);
1246 md->lsm_lock_owner = 0;
1247 spin_unlock(&md->lsm_lock);
1250 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1251 struct obd_quotactl *oqctl)
1253 struct lov_obd *lov = &obd->u.lov;
1254 struct lov_tgt_desc *tgt;
1255 struct pool_desc *pool = NULL;
1257 __u64 bhardlimit = 0;
1261 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1262 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
1263 oqctl->qc_cmd != LUSTRE_Q_GETQUOTAPOOL) {
1265 CERROR("%s: bad quota opc %x for lov obd: rc = %d\n",
1266 obd->obd_name, oqctl->qc_cmd, rc);
1270 if (oqctl->qc_cmd == LUSTRE_Q_GETQUOTAPOOL) {
1271 pool = lov_pool_find(obd, oqctl->qc_poolname);
1274 /* Set Q_GETOQUOTA back as targets report it's own
1275 * usage and doesn't care about pools */
1276 oqctl->qc_cmd = Q_GETOQUOTA;
1280 lov_tgts_getref(obd);
1281 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1284 tgt = lov->lov_tgts[i];
1290 lu_tgt_check_index(tgt->ltd_index, &pool->pool_obds))
1293 if (!tgt->ltd_active || tgt->ltd_reap) {
1294 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1295 lov->lov_tgts[i]->ltd_activate) {
1297 CERROR("%s: ost %d is inactive: rc = %d\n",
1298 obd->obd_name, i, rc);
1300 CDEBUG(D_HA, "ost %d is inactive\n", i);
1305 err = obd_quotactl(tgt->ltd_exp, oqctl);
1307 if (tgt->ltd_active && !rc)
1312 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1313 curspace += oqctl->qc_dqblk.dqb_curspace;
1314 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1317 lov_tgts_putref(obd);
1319 lov_pool_putref(pool);
1321 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1322 oqctl->qc_dqblk.dqb_curspace = curspace;
1323 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1328 static const struct obd_ops lov_obd_ops = {
1329 .o_owner = THIS_MODULE,
1330 .o_setup = lov_setup,
1331 .o_cleanup = lov_cleanup,
1332 .o_connect = lov_connect,
1333 .o_disconnect = lov_disconnect,
1334 .o_statfs = lov_statfs,
1335 .o_iocontrol = lov_iocontrol,
1336 .o_get_info = lov_get_info,
1337 .o_set_info_async = lov_set_info_async,
1338 .o_notify = lov_notify,
1339 .o_pool_new = lov_pool_new,
1340 .o_pool_rem = lov_pool_remove,
1341 .o_pool_add = lov_pool_add,
1342 .o_pool_del = lov_pool_del,
1343 .o_quotactl = lov_quotactl,
1346 struct kmem_cache *lov_oinfo_slab;
1348 static int __init lov_init(void)
1353 /* print an address of _any_ initialized kernel symbol from this
1354 * module, to allow debugging with gdb that doesn't support data
1355 * symbols from modules.*/
1356 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1358 rc = lu_kmem_init(lov_caches);
1362 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1363 sizeof(struct lov_oinfo), 0,
1364 SLAB_HWCACHE_ALIGN, NULL);
1365 if (lov_oinfo_slab == NULL) {
1366 lu_kmem_fini(lov_caches);
1370 rc = class_register_type(&lov_obd_ops, NULL, true,
1371 LUSTRE_LOV_NAME, &lov_device_type);
1373 kmem_cache_destroy(lov_oinfo_slab);
1374 lu_kmem_fini(lov_caches);
1380 static void __exit lov_exit(void)
1382 class_unregister_type(LUSTRE_LOV_NAME);
1383 kmem_cache_destroy(lov_oinfo_slab);
1384 lu_kmem_fini(lov_caches);
1387 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1388 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1389 MODULE_VERSION(LUSTRE_VERSION_STRING);
1390 MODULE_LICENSE("GPL");
1392 module_init(lov_init);
1393 module_exit(lov_exit);