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_lib.h>
47 #include <lustre_mds.h>
48 #include <lustre_net.h>
49 #include <uapi/linux/lustre/lustre_param.h>
50 #include <lustre_swab.h>
51 #include <lprocfs_status.h>
52 #include <obd_class.h>
53 #include <obd_support.h>
55 #include "lov_internal.h"
57 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
58 Any function that expects lov_tgts to remain stationary must take a ref. */
59 void lov_tgts_getref(struct obd_device *obd)
61 struct lov_obd *lov = &obd->u.lov;
63 /* nobody gets through here until lov_putref is done */
64 mutex_lock(&lov->lov_lock);
65 atomic_inc(&lov->lov_refcount);
66 mutex_unlock(&lov->lov_lock);
69 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
71 void lov_tgts_putref(struct obd_device *obd)
73 struct lov_obd *lov = &obd->u.lov;
75 mutex_lock(&lov->lov_lock);
76 /* ok to dec to 0 more than once -- ltd_exp's will be null */
77 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
79 struct lov_tgt_desc *tgt, *n;
82 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
84 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
85 tgt = lov->lov_tgts[i];
87 if (!tgt || !tgt->ltd_reap)
89 list_add(&tgt->ltd_kill, &kill);
90 /* XXX - right now there is a dependency on ld_tgt_count
91 * being the maximum tgt index for computing the
92 * mds_max_easize. So we can't shrink it. */
93 lu_tgt_pool_remove(&lov->lov_packed, i);
94 lov->lov_tgts[i] = NULL;
97 mutex_unlock(&lov->lov_lock);
99 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
100 list_del(&tgt->ltd_kill);
102 __lov_del_obd(obd, tgt);
105 mutex_unlock(&lov->lov_lock);
109 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
110 enum obd_notify_event ev);
112 int lov_connect_osc(struct obd_device *obd, u32 index, int activate,
113 struct obd_connect_data *data)
115 struct lov_obd *lov = &obd->u.lov;
116 struct obd_uuid *tgt_uuid;
117 struct obd_device *tgt_obd;
118 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
119 struct obd_import *imp;
123 if (lov->lov_tgts[index] == NULL)
126 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
127 tgt_obd = lov->lov_tgts[index]->ltd_obd;
129 if (!tgt_obd->obd_set_up) {
130 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
134 /* override the sp_me from lov */
135 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
137 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
138 data->ocd_index = index;
141 * Divine LOV knows that OBDs under it are OSCs.
143 imp = tgt_obd->u.cli.cl_import;
146 tgt_obd->obd_no_recov = 0;
147 /* FIXME this is probably supposed to be
148 ptlrpc_set_import_active. Horrible naming. */
149 ptlrpc_activate_import(imp, false);
152 rc = obd_register_observer(tgt_obd, obd);
154 CERROR("Target %s register_observer error %d\n",
155 obd_uuid2str(tgt_uuid), rc);
159 if (imp->imp_invalid) {
160 CDEBUG(D_CONFIG, "%s: not connecting - administratively disabled\n",
161 obd_uuid2str(tgt_uuid));
165 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
166 &lov_osc_uuid, data, lov->lov_cache);
167 if (rc || !lov->lov_tgts[index]->ltd_exp) {
168 CERROR("Target %s connect error %d\n",
169 obd_uuid2str(tgt_uuid), rc);
173 lov->lov_tgts[index]->ltd_reap = 0;
175 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
176 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
178 if (lov->lov_tgts_kobj) {
179 /* Even if we failed, that's ok */
180 rc = sysfs_create_link(lov->lov_tgts_kobj,
181 &tgt_obd->obd_kset.kobj,
184 CERROR("%s: can't register LOV target /sys/fs/lustre/%s/%s/target_obds/%s : rc = %d\n",
185 obd->obd_name, obd->obd_type->typ_name,
187 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_name,
194 static int lov_connect(const struct lu_env *env,
195 struct obd_export **exp, struct obd_device *obd,
196 struct obd_uuid *cluuid, struct obd_connect_data *data,
199 struct lov_obd *lov = &obd->u.lov;
200 struct lov_tgt_desc *tgt;
201 struct lustre_handle conn;
205 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
207 rc = class_connect(&conn, obd, cluuid);
211 *exp = class_conn2export(&conn);
213 /* Why should there ever be more than 1 connect? */
215 LASSERT(lov->lov_connects == 1);
217 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
219 lov->lov_ocd = *data;
221 lov_tgts_getref(obd);
224 lov->lov_cache = localdata;
225 cl_cache_incref(lov->lov_cache);
228 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
229 tgt = lov->lov_tgts[i];
230 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
232 /* Flags will be lowest common denominator */
233 rc = lov_connect_osc(obd, i, tgt->ltd_activate, &lov->lov_ocd);
235 CERROR("%s: lov connect tgt %d failed: %d\n",
236 obd->obd_name, i, rc);
239 /* connect to administrative disabled ost */
240 if (!lov->lov_tgts[i]->ltd_exp)
243 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
246 CERROR("%s error sending notify %d\n",
251 lov_tgts_putref(obd);
256 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
258 struct lov_obd *lov = &obd->u.lov;
259 struct obd_device *osc_obd;
263 osc_obd = class_exp2obd(tgt->ltd_exp);
264 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n", obd->obd_name,
265 osc_obd ? osc_obd->obd_name : "<no obd>");
267 if (tgt->ltd_active) {
269 lov->desc.ld_active_tgt_count--;
270 tgt->ltd_exp->exp_obd->obd_inactive = 1;
274 if (lov->lov_tgts_kobj)
275 sysfs_remove_link(lov->lov_tgts_kobj,
278 /* Pass it on to our clients.
279 * XXX This should be an argument to disconnect,
280 * XXX not a back-door flag on the OBD. Ah well.
282 osc_obd->obd_force = obd->obd_force;
283 osc_obd->obd_fail = obd->obd_fail;
284 osc_obd->obd_no_recov = obd->obd_no_recov;
287 obd_register_observer(osc_obd, NULL);
289 rc = obd_disconnect(tgt->ltd_exp);
291 CERROR("Target %s disconnect error %d\n",
292 tgt->ltd_uuid.uuid, rc);
300 static int lov_disconnect(struct obd_export *exp)
302 struct obd_device *obd = class_exp2obd(exp);
303 struct lov_obd *lov = &obd->u.lov;
311 /* Only disconnect the underlying layers on the final disconnect. */
312 if (lov->lov_connects == 0) {
313 CWARN("%s: was disconnected already #%d\n",
314 obd->obd_name, lov->lov_connects);
319 if (lov->lov_connects > 0) {
320 /* why should there be more than 1 connect? */
321 CWARN("%s: unexpected disconnect #%d\n",
322 obd->obd_name, lov->lov_connects);
326 /* hold another ref so lov_del_obd() doesn't spin in putref each time */
327 lov_tgts_getref(obd);
329 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
330 if (lov->lov_tgts[index] && lov->lov_tgts[index]->ltd_exp) {
331 /* Disconnection is the last we know about an OBD */
332 lov_del_target(obd, index, NULL,
333 lov->lov_tgts[index]->ltd_gen);
336 lov_tgts_putref(obd);
339 rc = class_disconnect(exp); /* bz 9811 */
345 * -EINVAL : UUID can't be found in the LOV's target list
346 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
347 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
348 * any >= 0 : is log target index
350 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
351 enum obd_notify_event ev)
353 struct lov_obd *lov = &obd->u.lov;
354 struct lov_tgt_desc *tgt;
356 bool activate, active;
359 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
360 lov, uuid->uuid, ev);
362 lov_tgts_getref(obd);
363 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
364 tgt = lov->lov_tgts[index];
365 if (tgt && obd_uuid_equals(uuid, &tgt->ltd_uuid))
369 if (index == lov->desc.ld_tgt_count)
370 GOTO(out, index = -EINVAL);
372 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
373 activate = (ev == OBD_NOTIFY_ACTIVATE);
376 * LU-642, initially inactive OSC could miss the obd_connect,
377 * we make up for it here.
379 if (activate && !tgt->ltd_exp) {
381 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
383 rc = obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
384 &lov_osc_uuid, &lov->lov_ocd,
386 if (rc || !tgt->ltd_exp)
387 GOTO(out, index = rc);
390 if (lov->lov_tgts[index]->ltd_activate == activate) {
391 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
392 uuid->uuid, activate ? "" : "de");
394 lov->lov_tgts[index]->ltd_activate = activate;
395 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
396 activate ? "" : "de", obd_uuid2str(uuid));
398 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
399 active = (ev == OBD_NOTIFY_ACTIVE);
401 if (lov->lov_tgts[index]->ltd_active == active) {
402 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
403 uuid->uuid, active ? "" : "in");
406 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
407 obd_uuid2str(uuid), active ? "" : "in");
409 lov->lov_tgts[index]->ltd_active = active;
411 lov->desc.ld_active_tgt_count++;
412 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
414 lov->desc.ld_active_tgt_count--;
415 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
418 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
423 CDEBUG(D_INFO, "%s: lov idx %d conn %llx\n", obd_uuid2str(uuid),
424 index, tgt->ltd_exp->exp_handle.h_cookie);
427 lov_tgts_putref(obd);
431 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
432 enum obd_notify_event ev)
435 struct lov_obd *lov = &obd->u.lov;
438 down_read(&lov->lov_notify_lock);
439 if (!lov->lov_connects)
440 GOTO(out_notify_lock, rc = 0);
442 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
443 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
444 struct obd_uuid *uuid;
448 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
449 CERROR("unexpected notification of %s %s\n",
450 watched->obd_type->typ_name, watched->obd_name);
451 GOTO(out_notify_lock, rc = -EINVAL);
454 uuid = &watched->u.cli.cl_target_uuid;
456 /* Set OSC as active before notifying the observer, so the
457 * observer can use the OSC normally.
459 rc = lov_set_osc_active(obd, uuid, ev);
461 CERROR("%s: event %d failed: rc = %d\n", obd->obd_name,
463 GOTO(out_notify_lock, rc);
467 /* Pass the notification up the chain. */
468 rc = obd_notify_observer(obd, watched, ev);
471 up_read(&lov->lov_notify_lock);
476 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
477 u32 index, int gen, int active)
479 struct lov_obd *lov = &obd->u.lov;
480 struct lov_tgt_desc *tgt;
481 struct obd_device *tgt_obd;
485 CDEBUG(D_CONFIG, "uuid:%s idx:%u gen:%d active:%d\n",
486 uuidp->uuid, index, gen, active);
489 CERROR("%s: request to add '%s' with invalid generation: %d\n",
490 obd->obd_name, uuidp->uuid, gen);
494 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, &obd->obd_uuid);
498 mutex_lock(&lov->lov_lock);
500 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
501 tgt = lov->lov_tgts[index];
503 CERROR("%s: UUID %s already assigned at index %d: rc = %d\n",
504 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), index, rc);
505 mutex_unlock(&lov->lov_lock);
509 if (index >= lov->lov_tgt_size) {
510 /* We need to reallocate the lov target array. */
511 struct lov_tgt_desc **newtgts, **old = NULL;
512 __u32 newsize, oldsize = 0;
514 newsize = max(lov->lov_tgt_size, 2U);
515 while (newsize < index + 1)
516 newsize = newsize << 1;
517 OBD_ALLOC_PTR_ARRAY(newtgts, newsize);
518 if (newtgts == NULL) {
519 mutex_unlock(&lov->lov_lock);
523 if (lov->lov_tgt_size) {
524 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
527 oldsize = lov->lov_tgt_size;
530 lov->lov_tgts = newtgts;
531 lov->lov_tgt_size = newsize;
534 OBD_FREE_PTR_ARRAY(old, oldsize);
536 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
537 lov->lov_tgts, lov->lov_tgt_size);
542 mutex_unlock(&lov->lov_lock);
546 rc = lu_tgt_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
548 mutex_unlock(&lov->lov_lock);
553 tgt->ltd_uuid = *uuidp;
554 tgt->ltd_obd = tgt_obd;
555 /* XXX - add a sanity check on the generation number. */
557 tgt->ltd_index = index;
558 tgt->ltd_activate = active;
559 lov->lov_tgts[index] = tgt;
560 if (index >= lov->desc.ld_tgt_count)
561 lov->desc.ld_tgt_count = index + 1;
563 mutex_unlock(&lov->lov_lock);
565 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
566 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
568 if (lov->lov_connects == 0) {
569 /* lov_connect hasn't been called yet. We'll do the
570 lov_connect_osc on this target when that fn first runs,
571 because we don't know the connect flags yet. */
575 lov_tgts_getref(obd);
577 rc = lov_connect_osc(obd, index, active, &lov->lov_ocd);
581 /* connect to administrative disabled ost */
585 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
586 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
590 CERROR("%s: add failed, deleting %s: rc = %d\n",
591 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), rc);
592 lov_del_target(obd, index, NULL, 0);
594 lov_tgts_putref(obd);
598 /* Schedule a target for deletion */
599 int lov_del_target(struct obd_device *obd, u32 index,
600 struct obd_uuid *uuidp, int gen)
602 struct lov_obd *lov = &obd->u.lov;
603 int count = lov->desc.ld_tgt_count;
607 if (index >= count) {
608 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
613 /* to make sure there's no ongoing lov_notify() now */
614 down_write(&lov->lov_notify_lock);
615 lov_tgts_getref(obd);
617 if (!lov->lov_tgts[index]) {
618 CERROR("LOV target at index %d is not setup.\n", index);
619 GOTO(out, rc = -EINVAL);
622 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
623 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
624 lov_uuid2str(lov, index), index,
625 obd_uuid2str(uuidp));
626 GOTO(out, rc = -EINVAL);
629 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
630 lov_uuid2str(lov, index), index,
631 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
632 lov->lov_tgts[index]->ltd_active);
634 lov->lov_tgts[index]->ltd_reap = 1;
635 lov->lov_death_row++;
636 /* we really delete it from lov_tgts_putref() */
638 lov_tgts_putref(obd);
639 up_write(&lov->lov_notify_lock);
644 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
646 struct obd_device *osc_obd;
649 LASSERT(tgt->ltd_reap);
651 osc_obd = class_exp2obd(tgt->ltd_exp);
653 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
655 osc_obd ? osc_obd->obd_name : "<no obd>");
658 lov_disconnect_obd(obd, tgt);
662 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
663 do it ourselves. And we can't do it from lov_cleanup,
664 because we just lost our only reference to it. */
666 class_manual_cleanup(osc_obd);
669 void lov_fix_desc_stripe_size(__u64 *val)
671 if (*val < LOV_MIN_STRIPE_SIZE) {
673 LCONSOLE_INFO("Increasing default stripe size to "
675 LOV_DESC_STRIPE_SIZE_DEFAULT);
676 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
677 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
678 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
679 LCONSOLE_WARN("Changing default stripe size to %llu (a "
681 *val, LOV_MIN_STRIPE_SIZE);
685 void lov_fix_desc_stripe_count(__u32 *val)
691 void lov_fix_desc_pattern(__u32 *val)
693 /* from lov_setstripe */
694 if ((*val != 0) && !lov_pattern_supported_normal_comp(*val)) {
695 LCONSOLE_WARN("lov: Unknown stripe pattern: %#x\n", *val);
700 void lov_fix_desc_qos_maxage(__u32 *val)
703 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
706 void lov_fix_desc(struct lov_desc *desc)
708 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
709 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
710 lov_fix_desc_pattern(&desc->ld_pattern);
711 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
714 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
716 struct lov_desc *desc;
717 struct lov_obd *lov = &obd->u.lov;
721 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
722 CERROR("LOV setup requires a descriptor\n");
726 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
728 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
729 CERROR("descriptor size wrong: %d > %d\n",
730 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
734 if (desc->ld_magic != LOV_DESC_MAGIC) {
735 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
736 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
737 obd->obd_name, desc);
738 lustre_swab_lov_desc(desc);
740 CERROR("%s: Bad lov desc magic: %#x\n",
741 obd->obd_name, desc->ld_magic);
748 desc->ld_active_tgt_count = 0;
750 lov->lov_tgt_size = 0;
752 mutex_init(&lov->lov_lock);
753 atomic_set(&lov->lov_refcount, 0);
754 lov->lov_sp_me = LUSTRE_SP_CLI;
756 init_rwsem(&lov->lov_notify_lock);
758 INIT_LIST_HEAD(&lov->lov_pool_list);
759 lov->lov_pool_count = 0;
760 rc = lov_pool_hash_init(&lov->lov_pools_hash_body);
764 rc = lu_tgt_pool_init(&lov->lov_packed, 0);
768 rc = lov_tunables_init(obd);
772 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
773 &obd->obd_kset.kobj);
779 static int lov_cleanup(struct obd_device *obd)
781 struct lov_obd *lov = &obd->u.lov;
782 struct pool_desc *pool, *tmp;
785 if (lov->lov_tgts_kobj) {
786 kobject_put(lov->lov_tgts_kobj);
787 lov->lov_tgts_kobj = NULL;
790 list_for_each_entry_safe(pool, tmp, &lov->lov_pool_list, pool_list) {
791 /* free pool structs */
792 CDEBUG(D_INFO, "delete pool %p\n", pool);
793 /* In the function below, .hs_keycmp resolves to
794 * pool_hashkey_keycmp() */
795 /* coverity[overrun-buffer-val] */
796 lov_pool_del(obd, pool->pool_name);
798 lov_pool_hash_destroy(&lov->lov_pools_hash_body);
799 lu_tgt_pool_free(&lov->lov_packed);
801 lprocfs_obd_cleanup(obd);
804 lov_tgts_getref(obd);
805 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
806 if (!lov->lov_tgts[i])
809 /* Inactive targets may never have connected */
810 if (lov->lov_tgts[i]->ltd_active)
811 /* We should never get here - these
812 * should have been removed in the
814 CERROR("%s: lov tgt %d not cleaned! "
815 "deathrow=%d, lovrc=%d\n",
816 obd->obd_name, i, lov->lov_death_row,
817 atomic_read(&lov->lov_refcount));
818 lov_del_target(obd, i, NULL, 0);
820 lov_tgts_putref(obd);
821 OBD_FREE_PTR_ARRAY(lov->lov_tgts, lov->lov_tgt_size);
822 lov->lov_tgt_size = 0;
825 if (lov->lov_cache != NULL) {
826 cl_cache_decref(lov->lov_cache);
827 lov->lov_cache = NULL;
833 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
834 u32 *indexp, int *genp)
836 struct obd_uuid obd_uuid;
841 switch (cmd = lcfg->lcfg_command) {
845 case LCFG_LOV_ADD_OBD:
846 case LCFG_LOV_ADD_INA:
847 case LCFG_LOV_DEL_OBD: {
851 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
852 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
853 GOTO(out, rc = -EINVAL);
855 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
857 rc = kstrtou32(lustre_cfg_buf(lcfg, 2), 10, indexp);
860 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
865 if (cmd == LCFG_LOV_ADD_OBD)
866 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
867 else if (cmd == LCFG_LOV_ADD_INA)
868 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
870 rc = lov_del_target(obd, index, &obd_uuid, gen);
875 struct lov_desc *desc = &(obd->u.lov.desc);
879 GOTO(out, rc = -EINVAL);
881 count = class_modify_config(lcfg, PARAM_LOV,
882 &obd->obd_kset.kobj);
883 GOTO(out, rc = count < 0 ? count : 0);
892 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
893 GOTO(out, rc = -EINVAL);
901 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
902 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
904 struct obd_device *obd = class_exp2obd(exp);
905 struct lov_obd *lov = &obd->u.lov;
906 struct obd_info oinfo = {
910 struct ptlrpc_request_set *rqset;
911 struct lov_request_set *set = NULL;
912 struct lov_request *req;
918 rqset = ptlrpc_prep_set();
922 rc = lov_prep_statfs_set(obd, &oinfo, &set);
926 list_for_each_entry(req, &set->set_list, rq_link) {
927 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
928 &req->rq_oi, max_age, rqset);
933 rc = ptlrpc_set_wait(env, rqset);
937 atomic_set(&set->set_completes, 0);
939 rc2 = lov_fini_statfs_set(set);
944 ptlrpc_set_destroy(rqset);
949 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
950 void *karg, void __user *uarg)
952 struct obd_device *obd = class_exp2obd(exp);
953 struct lov_obd *lov = &obd->u.lov;
954 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
957 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
958 exp->exp_obd->obd_name, cmd, len, karg, uarg);
960 /* exit early for unknown ioctl types */
961 if (unlikely(_IOC_TYPE(cmd) != 'f' && cmd != IOC_OSC_SET_ACTIVE))
962 RETURN(OBD_IOC_DEBUG(D_IOCTL, obd->obd_name, cmd, "unknown",
965 /* can't do a generic karg == NULL check here, since it is too noisy and
966 * we need to return -ENOTTY for unsupported ioctls instead of -EINVAL.
969 case IOC_OBD_STATFS: {
970 struct obd_ioctl_data *data;
971 struct obd_device *osc_obd;
972 struct obd_statfs stat_buf = {0};
973 struct obd_import *imp;
977 if (unlikely(karg == NULL))
978 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=null",
982 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
986 if (!lov->lov_tgts[index])
987 /* Try again with the next index */
990 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
994 imp = osc_obd->u.cli.cl_import;
995 if (!lov->lov_tgts[index]->ltd_active &&
996 imp->imp_state != LUSTRE_IMP_IDLE)
1000 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1001 min_t(unsigned long, data->ioc_plen2,
1002 sizeof(struct obd_uuid))))
1005 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1006 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1008 /* got statfs data */
1009 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1010 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1014 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1015 min_t(unsigned long, data->ioc_plen1,
1016 sizeof(struct obd_statfs))))
1020 case OBD_IOC_QUOTACTL: {
1021 struct if_quotactl *qctl;
1022 struct lov_tgt_desc *tgt = NULL;
1023 struct obd_quotactl *oqctl;
1024 struct obd_import *imp;
1026 if (unlikely(karg == NULL))
1027 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=null",
1031 if (qctl->qc_valid == QC_OSTIDX) {
1032 if (count <= qctl->qc_idx)
1035 tgt = lov->lov_tgts[qctl->qc_idx];
1041 } else if (qctl->qc_valid == QC_UUID) {
1042 for (i = 0; i < count; i++) {
1043 tgt = lov->lov_tgts[i];
1045 !obd_uuid_equals(&tgt->ltd_uuid,
1049 if (tgt->ltd_exp == NULL)
1061 LASSERT(tgt && tgt->ltd_exp);
1062 imp = class_exp2cliimp(tgt->ltd_exp);
1063 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
1064 qctl->qc_valid = QC_OSTIDX;
1065 qctl->obd_uuid = tgt->ltd_uuid;
1069 OBD_ALLOC_PTR(oqctl);
1073 QCTL_COPY(oqctl, qctl);
1074 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1076 QCTL_COPY(qctl, oqctl);
1077 qctl->qc_valid = QC_OSTIDX;
1078 qctl->obd_uuid = tgt->ltd_uuid;
1080 OBD_FREE_PTR(oqctl);
1089 for (i = 0; i < count; i++) {
1091 struct obd_device *osc_obd;
1093 /* OST was disconnected */
1094 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1097 /* ll_umount_begin() sets force on lov, pass to osc */
1098 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1100 osc_obd->obd_force = obd->obd_force;
1101 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1104 if (lov->lov_tgts[i]->ltd_active) {
1105 OBD_IOC_DEBUG(err == -ENOTTY ?
1106 D_IOCTL : D_WARNING,
1108 lov_uuid2str(lov, i),
1129 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1130 __u32 keylen, void *key, __u32 *vallen, void *val)
1132 struct obd_device *obd = class_exp2obd(exp);
1133 struct lov_obd *lov = &obd->u.lov;
1134 struct lov_desc *ld = &lov->desc;
1138 if (vallen == NULL || val == NULL)
1141 lov_tgts_getref(obd);
1143 if (KEY_IS(KEY_MAX_EASIZE)) {
1144 *((u32 *)val) = exp->exp_connect_data.ocd_max_easize;
1145 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1146 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1147 LOV_MAX_STRIPE_COUNT);
1149 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1150 } else if (KEY_IS(KEY_TGT_COUNT)) {
1151 *((int *)val) = lov->desc.ld_tgt_count;
1156 lov_tgts_putref(obd);
1161 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1162 __u32 keylen, void *key,
1163 __u32 vallen, void *val,
1164 struct ptlrpc_request_set *set)
1166 struct obd_device *obd = class_exp2obd(exp);
1167 struct lov_obd *lov = &obd->u.lov;
1168 struct lov_tgt_desc *tgt;
1169 bool do_inactive = false, no_set = false;
1178 set = ptlrpc_prep_set();
1183 lov_tgts_getref(obd);
1185 if (KEY_IS(KEY_CHECKSUM))
1188 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1189 tgt = lov->lov_tgts[i];
1191 /* OST was disconnected */
1192 if (tgt == NULL || tgt->ltd_exp == NULL)
1195 /* OST is inactive and we don't want inactive OSCs */
1196 if (!tgt->ltd_active && !do_inactive)
1199 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1206 /* cycle through MDC target for Data-on-MDT */
1207 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1208 struct obd_device *mdc;
1210 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1214 err = obd_set_info_async(env, mdc->obd_self_export,
1215 keylen, key, vallen, val, set);
1220 lov_tgts_putref(obd);
1222 err = ptlrpc_set_wait(env, set);
1225 ptlrpc_set_destroy(set);
1230 void lov_stripe_lock(struct lov_stripe_md *md)
1231 __acquires(&md->lsm_lock)
1233 LASSERT(md->lsm_lock_owner != current->pid);
1234 spin_lock(&md->lsm_lock);
1235 LASSERT(md->lsm_lock_owner == 0);
1236 md->lsm_lock_owner = current->pid;
1239 void lov_stripe_unlock(struct lov_stripe_md *md)
1240 __releases(&md->lsm_lock)
1242 LASSERT(md->lsm_lock_owner == current->pid);
1243 md->lsm_lock_owner = 0;
1244 spin_unlock(&md->lsm_lock);
1247 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1248 struct obd_quotactl *oqctl)
1250 struct lov_obd *lov = &obd->u.lov;
1251 struct lov_tgt_desc *tgt;
1252 struct pool_desc *pool = NULL;
1254 __u64 bhardlimit = 0;
1258 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1259 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
1260 oqctl->qc_cmd != LUSTRE_Q_GETQUOTAPOOL) {
1262 CERROR("%s: bad quota opc %x for lov obd: rc = %d\n",
1263 obd->obd_name, oqctl->qc_cmd, rc);
1267 if (oqctl->qc_cmd == LUSTRE_Q_GETQUOTAPOOL) {
1268 pool = lov_pool_find(obd, oqctl->qc_poolname);
1271 /* Set Q_GETOQUOTA back as targets report it's own
1272 * usage and doesn't care about pools */
1273 oqctl->qc_cmd = Q_GETOQUOTA;
1277 lov_tgts_getref(obd);
1278 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1281 tgt = lov->lov_tgts[i];
1287 lu_tgt_check_index(tgt->ltd_index, &pool->pool_obds))
1290 if (!tgt->ltd_active || tgt->ltd_reap) {
1291 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1292 lov->lov_tgts[i]->ltd_activate) {
1294 CERROR("%s: ost %d is inactive: rc = %d\n",
1295 obd->obd_name, i, rc);
1297 CDEBUG(D_HA, "ost %d is inactive\n", i);
1302 err = obd_quotactl(tgt->ltd_exp, oqctl);
1304 if (tgt->ltd_active && !rc)
1309 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1310 curspace += oqctl->qc_dqblk.dqb_curspace;
1311 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1314 lov_tgts_putref(obd);
1316 lov_pool_putref(pool);
1318 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1319 oqctl->qc_dqblk.dqb_curspace = curspace;
1320 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1325 static const struct obd_ops lov_obd_ops = {
1326 .o_owner = THIS_MODULE,
1327 .o_setup = lov_setup,
1328 .o_cleanup = lov_cleanup,
1329 .o_connect = lov_connect,
1330 .o_disconnect = lov_disconnect,
1331 .o_statfs = lov_statfs,
1332 .o_iocontrol = lov_iocontrol,
1333 .o_get_info = lov_get_info,
1334 .o_set_info_async = lov_set_info_async,
1335 .o_notify = lov_notify,
1336 .o_pool_new = lov_pool_new,
1337 .o_pool_rem = lov_pool_remove,
1338 .o_pool_add = lov_pool_add,
1339 .o_pool_del = lov_pool_del,
1340 .o_quotactl = lov_quotactl,
1343 struct kmem_cache *lov_oinfo_slab;
1345 static int __init lov_init(void)
1350 /* print an address of _any_ initialized kernel symbol from this
1351 * module, to allow debugging with gdb that doesn't support data
1352 * symbols from modules.*/
1353 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1355 rc = lu_kmem_init(lov_caches);
1359 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1360 sizeof(struct lov_oinfo), 0,
1361 SLAB_HWCACHE_ALIGN, NULL);
1362 if (lov_oinfo_slab == NULL) {
1363 lu_kmem_fini(lov_caches);
1367 rc = class_register_type(&lov_obd_ops, NULL, true,
1368 LUSTRE_LOV_NAME, &lov_device_type);
1370 kmem_cache_destroy(lov_oinfo_slab);
1371 lu_kmem_fini(lov_caches);
1377 static void __exit lov_exit(void)
1379 class_unregister_type(LUSTRE_LOV_NAME);
1380 kmem_cache_destroy(lov_oinfo_slab);
1381 lu_kmem_fini(lov_caches);
1384 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1385 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1386 MODULE_VERSION(LUSTRE_VERSION_STRING);
1387 MODULE_LICENSE("GPL");
1389 module_init(lov_init);
1390 module_exit(lov_exit);