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 static 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, struct obd_export **exp,
199 struct obd_device *obd, struct obd_uuid *cluuid,
200 struct obd_connect_data *data, void *localdata)
202 struct lov_obd *lov = &obd->u.lov;
203 struct lov_tgt_desc *tgt;
204 struct lustre_handle conn;
208 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
210 rc = class_connect(&conn, obd, cluuid);
214 *exp = class_conn2export(&conn);
216 /* Why should there ever be more than 1 connect? */
218 LASSERT(lov->lov_connects == 1);
220 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
222 lov->lov_ocd = *data;
224 lov_tgts_getref(obd);
227 lov->lov_cache = localdata;
228 cl_cache_incref(lov->lov_cache);
231 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
232 tgt = lov->lov_tgts[i];
233 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
235 /* Flags will be lowest common denominator */
236 rc = lov_connect_osc(obd, i, tgt->ltd_activate, &lov->lov_ocd);
238 CERROR("%s: lov connect tgt %d failed: rc = %d\n",
239 obd->obd_name, i, rc);
242 /* connect to administrative disabled ost */
243 if (!lov->lov_tgts[i]->ltd_exp)
246 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
249 CERROR("%s error sending notify: rc = %d\n",
254 lov_tgts_putref(obd);
259 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
261 struct lov_obd *lov = &obd->u.lov;
262 struct obd_device *osc_obd;
266 osc_obd = class_exp2obd(tgt->ltd_exp);
267 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n", obd->obd_name,
268 osc_obd ? osc_obd->obd_name : "<no obd>");
270 if (tgt->ltd_active) {
272 lov->desc.ld_active_tgt_count--;
273 tgt->ltd_exp->exp_obd->obd_inactive = 1;
277 if (lov->lov_tgts_kobj)
278 sysfs_remove_link(lov->lov_tgts_kobj,
281 /* Pass it on to our clients.
282 * XXX This should be an argument to disconnect,
283 * XXX not a back-door flag on the OBD. Ah well.
285 osc_obd->obd_force = obd->obd_force;
286 osc_obd->obd_fail = obd->obd_fail;
287 osc_obd->obd_no_recov = obd->obd_no_recov;
290 obd_register_observer(osc_obd, NULL);
292 rc = obd_disconnect(tgt->ltd_exp);
294 CERROR("%s: Target %s disconnect error %d\n",
295 obd->obd_name, tgt->ltd_uuid.uuid, rc);
303 static int lov_disconnect(struct obd_export *exp)
305 struct obd_device *obd = class_exp2obd(exp);
306 struct lov_obd *lov = &obd->u.lov;
314 /* Only disconnect the underlying layers on the final disconnect. */
315 if (lov->lov_connects == 0) {
316 CWARN("%s: was disconnected already #%d\n",
317 obd->obd_name, lov->lov_connects);
322 if (lov->lov_connects > 0) {
323 /* why should there be more than 1 connect? */
324 CWARN("%s: unexpected disconnect #%d\n",
325 obd->obd_name, lov->lov_connects);
329 /* hold another ref so lov_del_obd() doesn't spin in putref each time */
330 lov_tgts_getref(obd);
332 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
333 if (lov->lov_tgts[index] && lov->lov_tgts[index]->ltd_exp) {
334 /* Disconnection is the last we know about an OBD */
335 lov_del_target(obd, index, NULL,
336 lov->lov_tgts[index]->ltd_gen);
339 lov_tgts_putref(obd);
342 rc = class_disconnect(exp); /* bz 9811 */
348 * -EINVAL : UUID can't be found in the LOV's target list
349 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
350 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
351 * any >= 0 : is log target index
353 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
354 enum obd_notify_event ev)
356 struct lov_obd *lov = &obd->u.lov;
357 struct lov_tgt_desc *tgt;
359 bool activate, active;
362 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
363 lov, uuid->uuid, ev);
365 lov_tgts_getref(obd);
366 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
367 tgt = lov->lov_tgts[index];
368 if (tgt && obd_uuid_equals(uuid, &tgt->ltd_uuid))
372 if (index == lov->desc.ld_tgt_count)
373 GOTO(out, index = -EINVAL);
375 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
376 activate = (ev == OBD_NOTIFY_ACTIVATE);
379 * LU-642, initially inactive OSC could miss the obd_connect,
380 * we make up for it here.
382 if (activate && !tgt->ltd_exp) {
384 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
386 rc = obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
387 &lov_osc_uuid, &lov->lov_ocd,
389 if (rc || !tgt->ltd_exp)
390 GOTO(out, index = rc);
393 if (lov->lov_tgts[index]->ltd_activate == activate) {
394 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
395 uuid->uuid, activate ? "" : "de");
397 lov->lov_tgts[index]->ltd_activate = activate;
398 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
399 activate ? "" : "de", obd_uuid2str(uuid));
401 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
402 active = (ev == OBD_NOTIFY_ACTIVE);
404 if (lov->lov_tgts[index]->ltd_active == active) {
405 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
406 uuid->uuid, active ? "" : "in");
409 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
410 obd_uuid2str(uuid), active ? "" : "in");
412 lov->lov_tgts[index]->ltd_active = active;
414 lov->desc.ld_active_tgt_count++;
415 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
417 lov->desc.ld_active_tgt_count--;
418 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
421 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
426 CDEBUG(D_INFO, "%s: lov idx %d conn %llx\n", obd_uuid2str(uuid),
427 index, tgt->ltd_exp->exp_handle.h_cookie);
430 lov_tgts_putref(obd);
434 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
435 enum obd_notify_event ev)
438 struct lov_obd *lov = &obd->u.lov;
441 down_read(&lov->lov_notify_lock);
442 if (!lov->lov_connects)
443 GOTO(out_notify_lock, rc = 0);
445 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
446 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
447 struct obd_uuid *uuid;
451 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
452 CERROR("unexpected notification of %s %s\n",
453 watched->obd_type->typ_name, watched->obd_name);
454 GOTO(out_notify_lock, rc = -EINVAL);
457 uuid = &watched->u.cli.cl_target_uuid;
459 /* Set OSC as active before notifying the observer, so the
460 * observer can use the OSC normally.
462 rc = lov_set_osc_active(obd, uuid, ev);
464 CERROR("%s: event %d failed: rc = %d\n", obd->obd_name,
466 GOTO(out_notify_lock, rc);
470 /* Pass the notification up the chain. */
471 rc = obd_notify_observer(obd, watched, ev);
474 up_read(&lov->lov_notify_lock);
479 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
480 u32 index, int gen, int active)
482 struct lov_obd *lov = &obd->u.lov;
483 struct lov_tgt_desc *tgt;
484 struct obd_device *tgt_obd;
488 CDEBUG(D_CONFIG, "uuid:%s idx:%u gen:%d active:%d\n",
489 uuidp->uuid, index, gen, active);
492 CERROR("%s: request to add '%s' with invalid generation: %d\n",
493 obd->obd_name, uuidp->uuid, gen);
497 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, &obd->obd_uuid);
501 mutex_lock(&lov->lov_lock);
503 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
504 tgt = lov->lov_tgts[index];
506 CERROR("%s: UUID %s already assigned at index %d: rc = %d\n",
507 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), index, rc);
508 mutex_unlock(&lov->lov_lock);
512 if (index >= lov->lov_tgt_size) {
513 /* We need to reallocate the lov target array. */
514 struct lov_tgt_desc **newtgts, **old = NULL;
515 __u32 newsize, oldsize = 0;
517 newsize = max(lov->lov_tgt_size, 2U);
518 while (newsize < index + 1)
519 newsize = newsize << 1;
520 OBD_ALLOC_PTR_ARRAY(newtgts, newsize);
521 if (newtgts == NULL) {
522 mutex_unlock(&lov->lov_lock);
526 if (lov->lov_tgt_size) {
527 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
530 oldsize = lov->lov_tgt_size;
533 lov->lov_tgts = newtgts;
534 lov->lov_tgt_size = newsize;
537 OBD_FREE_PTR_ARRAY(old, oldsize);
539 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
540 lov->lov_tgts, lov->lov_tgt_size);
545 mutex_unlock(&lov->lov_lock);
549 rc = lu_tgt_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
551 mutex_unlock(&lov->lov_lock);
556 tgt->ltd_uuid = *uuidp;
557 tgt->ltd_obd = tgt_obd;
558 /* XXX - add a sanity check on the generation number. */
560 tgt->ltd_index = index;
561 tgt->ltd_activate = active;
562 lov->lov_tgts[index] = tgt;
563 if (index >= lov->desc.ld_tgt_count)
564 lov->desc.ld_tgt_count = index + 1;
566 mutex_unlock(&lov->lov_lock);
568 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
569 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
571 if (lov->lov_connects == 0) {
572 /* lov_connect hasn't been called yet. We'll do the
573 * lov_connect_osc on this target when that fn first runs,
574 * because we don't know the connect flags yet. */
578 lov_tgts_getref(obd);
580 rc = lov_connect_osc(obd, index, active, &lov->lov_ocd);
584 /* connect to administrative disabled ost */
588 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
589 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
593 CERROR("%s: add failed, deleting %s: rc = %d\n",
594 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), rc);
595 lov_del_target(obd, index, NULL, 0);
597 lov_tgts_putref(obd);
601 /* Schedule a target for deletion */
602 int lov_del_target(struct obd_device *obd, u32 index, struct obd_uuid *uuidp,
605 struct lov_obd *lov = &obd->u.lov;
606 int count = lov->desc.ld_tgt_count;
610 if (index >= count) {
611 CERROR("%s: LOV target index %d >= number of LOV OBDs %d: rc = %d\n",
612 obd->obd_name, index, count, -EINVAL);
616 /* to make sure there's no ongoing lov_notify() now */
617 down_write(&lov->lov_notify_lock);
618 lov_tgts_getref(obd);
620 if (!lov->lov_tgts[index]) {
621 CERROR("%s: LOV target at index %d is not setup: rc = %d\n",
622 obd->obd_name, index, -EINVAL);
623 GOTO(out, rc = -EINVAL);
626 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
627 CERROR("%s: LOV target UUID %s at index %d doesn't match %s: rc = %d\n",
628 obd->obd_name, lov_uuid2str(lov, index), index,
629 obd_uuid2str(uuidp), -EINVAL);
631 GOTO(out, rc = -EINVAL);
634 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
635 lov_uuid2str(lov, index), index,
636 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
637 lov->lov_tgts[index]->ltd_active);
639 lov->lov_tgts[index]->ltd_reap = 1;
640 lov->lov_death_row++;
641 /* we really delete it from lov_tgts_putref() */
643 lov_tgts_putref(obd);
644 up_write(&lov->lov_notify_lock);
649 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
651 struct obd_device *osc_obd;
654 LASSERT(tgt->ltd_reap);
656 osc_obd = class_exp2obd(tgt->ltd_exp);
658 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n", tgt->ltd_uuid.uuid,
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("%s: LOV setup requires a descriptor: rc = %d\n",
727 obd->obd_name, -EINVAL);
731 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
733 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
734 CERROR("%s: descriptor size wrong: %d > %d: rc = %d\n",
735 obd->obd_name, (int)sizeof(*desc),
736 LUSTRE_CFG_BUFLEN(lcfg, 1), -EINVAL);
740 if (desc->ld_magic != LOV_DESC_MAGIC) {
741 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
742 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
743 obd->obd_name, desc);
744 lustre_swab_lov_desc(desc);
746 CERROR("%s: Bad lov desc magic: %#x: rc = %d\n",
747 obd->obd_name, desc->ld_magic, -EINVAL);
754 desc->ld_active_tgt_count = 0;
756 lov->lov_tgt_size = 0;
758 mutex_init(&lov->lov_lock);
759 atomic_set(&lov->lov_refcount, 0);
760 lov->lov_sp_me = LUSTRE_SP_CLI;
762 init_rwsem(&lov->lov_notify_lock);
764 INIT_LIST_HEAD(&lov->lov_pool_list);
765 lov->lov_pool_count = 0;
766 rc = lov_pool_hash_init(&lov->lov_pools_hash_body);
770 rc = lu_tgt_pool_init(&lov->lov_packed, 0);
774 rc = lov_tunables_init(obd);
778 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
779 &obd->obd_kset.kobj);
784 static int lov_cleanup(struct obd_device *obd)
786 struct lov_obd *lov = &obd->u.lov;
787 struct lov_pool_desc *pool, *tmp;
790 if (lov->lov_tgts_kobj) {
791 kobject_put(lov->lov_tgts_kobj);
792 lov->lov_tgts_kobj = NULL;
795 list_for_each_entry_safe(pool, tmp, &lov->lov_pool_list, pool_list) {
796 /* free pool structs */
797 CDEBUG(D_INFO, "delete pool %p\n", pool);
798 /* In the function below, .hs_keycmp resolves to
799 * pool_hashkey_keycmp() */
800 lov_pool_del(obd, pool->pool_name);
802 lov_pool_hash_destroy(&lov->lov_pools_hash_body);
803 lu_tgt_pool_free(&lov->lov_packed);
805 lprocfs_obd_cleanup(obd);
809 lov_tgts_getref(obd);
810 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
811 if (!lov->lov_tgts[i])
814 /* Inactive targets may never have connected */
815 if (lov->lov_tgts[i]->ltd_active)
816 /* We should never get here - these
817 * should have been removed in the
819 CERROR("%s: lov tgt %d not cleaned! "
820 "deathrow=%d, lovrc=%d\n",
821 obd->obd_name, i, lov->lov_death_row,
822 atomic_read(&lov->lov_refcount));
823 lov_del_target(obd, i, NULL, 0);
825 lov_tgts_putref(obd);
826 OBD_FREE_PTR_ARRAY(lov->lov_tgts, lov->lov_tgt_size);
827 lov->lov_tgt_size = 0;
830 if (lov->lov_cache != NULL) {
831 cl_cache_decref(lov->lov_cache);
832 lov->lov_cache = NULL;
838 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
839 u32 *indexp, int *genp)
841 struct obd_uuid obd_uuid;
846 switch (cmd = lcfg->lcfg_command) {
850 case LCFG_LOV_ADD_OBD:
851 case LCFG_LOV_ADD_INA:
852 case LCFG_LOV_DEL_OBD: {
856 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
857 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
858 GOTO(out, rc = -EINVAL);
860 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
862 rc = kstrtou32(lustre_cfg_buf(lcfg, 2), 10, indexp);
865 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
870 if (cmd == LCFG_LOV_ADD_OBD)
871 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
872 else if (cmd == LCFG_LOV_ADD_INA)
873 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
875 rc = lov_del_target(obd, index, &obd_uuid, gen);
880 struct lov_desc *desc = &(obd->u.lov.desc);
884 GOTO(out, rc = -EINVAL);
886 count = class_modify_config(lcfg, PARAM_LOV,
887 &obd->obd_kset.kobj);
888 GOTO(out, rc = count < 0 ? count : 0);
897 CERROR("%s: Unknown command: %d: rc = %d\n",
898 obd->obd_name, lcfg->lcfg_command, -EINVAL);
899 GOTO(out, rc = -EINVAL);
906 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
907 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
909 struct obd_device *obd = class_exp2obd(exp);
910 struct lov_obd *lov = &obd->u.lov;
911 struct obd_info oinfo = {
915 struct ptlrpc_request_set *rqset;
916 struct lov_request_set *set = NULL;
917 struct lov_request *req;
923 rqset = ptlrpc_prep_set();
927 rc = lov_prep_statfs_set(obd, &oinfo, &set);
931 list_for_each_entry(req, &set->set_list, rq_link) {
932 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
933 &req->rq_oi, max_age, rqset);
938 rc = ptlrpc_set_wait(env, rqset);
942 atomic_set(&set->set_completes, 0);
944 rc2 = lov_fini_statfs_set(set);
949 ptlrpc_set_destroy(rqset);
954 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
955 void *karg, void __user *uarg)
957 struct obd_device *obd = class_exp2obd(exp);
958 struct lov_obd *lov = &obd->u.lov;
959 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
962 CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
963 exp->exp_obd->obd_name, cmd, len, karg, uarg);
965 /* exit early for unknown ioctl types. */
966 if (unlikely(_IOC_TYPE(cmd) != 'f' && !IOC_OSC_SET_ACTIVE_ALLOW(cmd)))
967 RETURN(OBD_IOC_DEBUG(D_IOCTL, obd->obd_name, cmd, "unknown",
970 /* can't do a generic karg == NULL check here, since it is too noisy and
971 * we need to return -ENOTTY for unsupported ioctls instead of -EINVAL.
974 case IOC_OBD_STATFS: {
975 struct obd_ioctl_data *data;
976 struct obd_device *osc_obd;
977 struct obd_statfs stat_buf = {0};
978 struct obd_import *imp;
982 if (unlikely(karg == NULL))
983 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=null",
987 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
991 if (!lov->lov_tgts[index])
992 /* Try again with the next index */
995 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
999 imp = osc_obd->u.cli.cl_import;
1000 if (!lov->lov_tgts[index]->ltd_active &&
1001 imp->imp_state != LUSTRE_IMP_IDLE)
1005 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1006 min_t(unsigned long, data->ioc_plen2,
1007 sizeof(struct obd_uuid))))
1010 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1011 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1013 /* got statfs data */
1014 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1015 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1019 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1020 min_t(unsigned long, data->ioc_plen1,
1021 sizeof(struct obd_statfs))))
1025 case OBD_IOC_QUOTACTL: {
1026 struct if_quotactl *qctl;
1027 struct lov_tgt_desc *tgt = NULL;
1028 struct obd_quotactl *oqctl;
1029 struct obd_import *imp;
1031 if (unlikely(karg == NULL))
1032 RETURN(OBD_IOC_ERROR(obd->obd_name, cmd, "karg=null",
1036 if (qctl->qc_valid == QC_OSTIDX) {
1037 if (count <= qctl->qc_idx)
1040 tgt = lov->lov_tgts[qctl->qc_idx];
1046 } else if (qctl->qc_valid == QC_UUID) {
1047 for (i = 0; i < count; i++) {
1048 tgt = lov->lov_tgts[i];
1050 !obd_uuid_equals(&tgt->ltd_uuid,
1054 if (tgt->ltd_exp == NULL)
1066 LASSERT(tgt && tgt->ltd_exp);
1067 imp = class_exp2cliimp(tgt->ltd_exp);
1068 if (!tgt->ltd_active && imp->imp_state != LUSTRE_IMP_IDLE) {
1069 qctl->qc_valid = QC_OSTIDX;
1070 qctl->obd_uuid = tgt->ltd_uuid;
1074 OBD_ALLOC_PTR(oqctl);
1078 QCTL_COPY(oqctl, qctl);
1079 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1081 QCTL_COPY_NO_PNAME(qctl, oqctl);
1082 qctl->qc_valid = QC_OSTIDX;
1083 qctl->obd_uuid = tgt->ltd_uuid;
1085 OBD_FREE_PTR(oqctl);
1094 for (i = 0; i < count; i++) {
1096 struct obd_device *osc_obd;
1098 /* OST was disconnected */
1099 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1102 /* ll_umount_begin() sets force on lov, pass to osc */
1103 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1105 osc_obd->obd_force = obd->obd_force;
1106 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1109 if (lov->lov_tgts[i]->ltd_active) {
1110 OBD_IOC_DEBUG(err == -ENOTTY ?
1111 D_IOCTL : D_WARNING,
1113 lov_uuid2str(lov, i),
1134 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1135 __u32 keylen, void *key, __u32 *vallen, void *val)
1137 struct obd_device *obd = class_exp2obd(exp);
1138 struct lov_obd *lov = &obd->u.lov;
1139 struct lov_desc *ld = &lov->desc;
1143 if (vallen == NULL || val == NULL)
1146 lov_tgts_getref(obd);
1148 if (KEY_IS(KEY_MAX_EASIZE)) {
1149 *((u32 *)val) = exp->exp_connect_data.ocd_max_easize;
1150 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1151 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1152 LOV_MAX_STRIPE_COUNT);
1154 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1155 } else if (KEY_IS(KEY_TGT_COUNT)) {
1156 *((int *)val) = lov->desc.ld_tgt_count;
1161 lov_tgts_putref(obd);
1166 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1167 __u32 keylen, void *key,
1168 __u32 vallen, void *val,
1169 struct ptlrpc_request_set *set)
1171 struct obd_device *obd = class_exp2obd(exp);
1172 struct lov_obd *lov = &obd->u.lov;
1173 struct lov_tgt_desc *tgt;
1174 bool do_inactive = false, no_set = false;
1183 set = ptlrpc_prep_set();
1188 lov_tgts_getref(obd);
1190 if (KEY_IS(KEY_CHECKSUM))
1193 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1194 tgt = lov->lov_tgts[i];
1196 /* OST was disconnected */
1197 if (tgt == NULL || tgt->ltd_exp == NULL)
1200 /* OST is inactive and we don't want inactive OSCs */
1201 if (!tgt->ltd_active && !do_inactive)
1204 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1211 /* cycle through MDC target for Data-on-MDT */
1212 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1213 struct obd_device *mdc;
1215 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1219 err = obd_set_info_async(env, mdc->obd_self_export,
1220 keylen, key, vallen, val, set);
1225 lov_tgts_putref(obd);
1227 err = ptlrpc_set_wait(env, set);
1230 ptlrpc_set_destroy(set);
1235 void lov_stripe_lock(struct lov_stripe_md *md)
1236 __acquires(&md->lsm_lock)
1238 LASSERT(md->lsm_lock_owner != current->pid);
1239 spin_lock(&md->lsm_lock);
1240 LASSERT(md->lsm_lock_owner == 0);
1241 md->lsm_lock_owner = current->pid;
1244 void lov_stripe_unlock(struct lov_stripe_md *md)
1245 __releases(&md->lsm_lock)
1247 LASSERT(md->lsm_lock_owner == current->pid);
1248 md->lsm_lock_owner = 0;
1249 spin_unlock(&md->lsm_lock);
1252 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1253 struct obd_quotactl *oqctl)
1255 struct lov_obd *lov = &obd->u.lov;
1256 struct lov_tgt_desc *tgt;
1257 struct lov_pool_desc *pool = NULL;
1258 struct list_head *lst = NULL;
1260 __u64 bhardlimit = 0;
1264 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1265 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
1266 oqctl->qc_cmd != LUSTRE_Q_GETQUOTAPOOL &&
1267 oqctl->qc_cmd != LUSTRE_Q_ITEROQUOTA) {
1269 CERROR("%s: bad quota opc %x for lov obd: rc = %d\n",
1270 obd->obd_name, oqctl->qc_cmd, rc);
1274 if (oqctl->qc_cmd == LUSTRE_Q_GETQUOTAPOOL) {
1275 pool = lov_pool_find(obd, oqctl->qc_poolname);
1278 /* Set Q_GETOQUOTA back as targets report it's own
1279 * usage and doesn't care about pools */
1280 oqctl->qc_cmd = Q_GETOQUOTA;
1283 if (oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA)
1284 lst = (struct list_head *)oqctl->qc_iter_list;
1287 lov_tgts_getref(obd);
1288 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1291 tgt = lov->lov_tgts[i];
1297 lu_tgt_check_index(tgt->ltd_index, &pool->pool_obds))
1300 if (!tgt->ltd_active || tgt->ltd_reap) {
1301 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1302 lov->lov_tgts[i]->ltd_activate) {
1304 CERROR("%s: ost %d is inactive: rc = %d\n",
1305 obd->obd_name, i, rc);
1307 CDEBUG(D_HA, "ost %d is inactive\n", i);
1312 if (oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA)
1313 err = obd_quota_iter(tgt->ltd_exp, oqctl, lst);
1315 err = obd_quotactl(tgt->ltd_exp, oqctl);
1318 if (tgt->ltd_active && !rc)
1323 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1324 curspace += oqctl->qc_dqblk.dqb_curspace;
1325 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1328 lov_tgts_putref(obd);
1330 lov_pool_putref(pool);
1332 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1333 oqctl->qc_dqblk.dqb_curspace = curspace;
1334 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1340 static const struct obd_ops lov_obd_ops = {
1341 .o_owner = THIS_MODULE,
1342 .o_setup = lov_setup,
1343 .o_cleanup = lov_cleanup,
1344 .o_connect = lov_connect,
1345 .o_disconnect = lov_disconnect,
1346 .o_statfs = lov_statfs,
1347 .o_iocontrol = lov_iocontrol,
1348 .o_get_info = lov_get_info,
1349 .o_set_info_async = lov_set_info_async,
1350 .o_notify = lov_notify,
1351 .o_pool_new = lov_pool_new,
1352 .o_pool_rem = lov_pool_remove,
1353 .o_pool_add = lov_pool_add,
1354 .o_pool_del = lov_pool_del,
1355 .o_quotactl = lov_quotactl,
1358 struct kmem_cache *lov_oinfo_slab;
1360 static int __init lov_init(void)
1365 /* print an address of _any_ initialized kernel symbol from this
1366 * module, to allow debugging with gdb that doesn't support data
1367 * symbols from modules.*/
1368 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1370 rc = libcfs_setup();
1374 rc = lu_kmem_init(lov_caches);
1378 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1379 sizeof(struct lov_oinfo), 0,
1380 SLAB_HWCACHE_ALIGN, NULL);
1381 if (lov_oinfo_slab == NULL) {
1382 lu_kmem_fini(lov_caches);
1386 rc = class_register_type(&lov_obd_ops, NULL, true,
1387 LUSTRE_LOV_NAME, &lov_device_type);
1389 kmem_cache_destroy(lov_oinfo_slab);
1390 lu_kmem_fini(lov_caches);
1396 static void __exit lov_exit(void)
1398 class_unregister_type(LUSTRE_LOV_NAME);
1399 kmem_cache_destroy(lov_oinfo_slab);
1400 lu_kmem_fini(lov_caches);
1403 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1404 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1405 MODULE_VERSION(LUSTRE_VERSION_STRING);
1406 MODULE_LICENSE("GPL");
1408 module_init(lov_init);
1409 module_exit(lov_exit);