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 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);
71 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
73 void lov_tgts_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->lov_tgts_kobj) {
184 /* Even if we failed, that's ok */
185 rc = sysfs_create_link(lov->lov_tgts_kobj,
186 &tgt_obd->obd_kset.kobj,
189 CERROR("%s: can't register LOV target /sys/fs/lustre/%s/%s/target_obds/%s : rc = %d\n",
190 obd->obd_name, obd->obd_type->typ_name,
192 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_name,
199 static int lov_connect(const struct lu_env *env,
200 struct obd_export **exp, struct obd_device *obd,
201 struct obd_uuid *cluuid, struct obd_connect_data *data,
204 struct lov_obd *lov = &obd->u.lov;
205 struct lov_tgt_desc *tgt;
206 struct lustre_handle conn;
210 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
212 rc = class_connect(&conn, obd, cluuid);
216 *exp = class_conn2export(&conn);
218 /* Why should there ever be more than 1 connect? */
220 LASSERT(lov->lov_connects == 1);
222 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
224 lov->lov_ocd = *data;
226 lov_tgts_getref(obd);
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_obd(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",
265 obd->obd_name, osc_obd->obd_name);
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;
286 if (lov->targets_proc_entry != NULL)
287 lprocfs_remove_proc_entry(osc_obd->obd_name,
288 lov->targets_proc_entry);
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. */
317 if (lov->lov_connects != 0) {
318 /* why should there be more than 1 connect? */
319 CWARN("%s: unexpected disconnect #%d\n",
320 obd->obd_name, lov->lov_connects);
324 /* hold another ref so lov_del_obd() doesn't spin in putref each time */
325 lov_tgts_getref(obd);
327 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
328 if (lov->lov_tgts[index] && lov->lov_tgts[index]->ltd_exp) {
329 /* Disconnection is the last we know about an OBD */
330 lov_del_target(obd, index, NULL,
331 lov->lov_tgts[index]->ltd_gen);
334 lov_tgts_putref(obd);
337 rc = class_disconnect(exp); /* bz 9811 */
343 * -EINVAL : UUID can't be found in the LOV's target list
344 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
345 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
346 * any >= 0 : is log target index
348 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
349 enum obd_notify_event ev)
351 struct lov_obd *lov = &obd->u.lov;
352 struct lov_tgt_desc *tgt;
353 int index, activate, active;
356 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
357 lov, uuid->uuid, ev);
359 lov_tgts_getref(obd);
360 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
361 tgt = lov->lov_tgts[index];
365 * LU-642, initially inactive OSC could miss the obd_connect,
366 * we make up for it here.
368 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
369 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
370 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
372 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
373 &lov_osc_uuid, &lov->lov_ocd, NULL);
378 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
379 index, obd_uuid2str(&tgt->ltd_uuid),
380 tgt->ltd_exp->exp_handle.h_cookie);
381 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
385 if (index == lov->desc.ld_tgt_count)
386 GOTO(out, index = -EINVAL);
388 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
389 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
391 if (lov->lov_tgts[index]->ltd_activate == activate) {
392 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
393 uuid->uuid, activate ? "" : "de");
395 lov->lov_tgts[index]->ltd_activate = activate;
396 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
397 activate ? "" : "de", obd_uuid2str(uuid));
400 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
401 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
403 if (lov->lov_tgts[index]->ltd_active == active) {
404 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
405 uuid->uuid, active ? "" : "in");
408 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
409 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 lov_tgts_putref(obd);
430 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
431 enum obd_notify_event ev)
434 struct lov_obd *lov = &obd->u.lov;
437 down_read(&lov->lov_notify_lock);
438 if (!lov->lov_connects)
439 GOTO(out_notify_lock, rc = 0);
441 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
442 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
443 struct obd_uuid *uuid;
447 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
448 CERROR("unexpected notification of %s %s\n",
449 watched->obd_type->typ_name, watched->obd_name);
450 GOTO(out_notify_lock, rc = -EINVAL);
453 uuid = &watched->u.cli.cl_target_uuid;
455 /* Set OSC as active before notifying the observer, so the
456 * observer can use the OSC normally.
458 rc = lov_set_osc_active(obd, uuid, ev);
460 CERROR("%s: event %d failed: rc = %d\n", obd->obd_name,
462 GOTO(out_notify_lock, rc);
466 /* Pass the notification up the chain. */
467 rc = obd_notify_observer(obd, watched, ev);
470 up_read(&lov->lov_notify_lock);
475 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
476 u32 index, int gen, int active)
478 struct lov_obd *lov = &obd->u.lov;
479 struct lov_tgt_desc *tgt;
480 struct obd_device *tgt_obd;
484 CDEBUG(D_CONFIG, "uuid:%s idx:%u gen:%d active:%d\n",
485 uuidp->uuid, index, gen, active);
488 CERROR("%s: request to add '%s' with invalid generation: %d\n",
489 obd->obd_name, uuidp->uuid, gen);
493 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME, &obd->obd_uuid);
497 mutex_lock(&lov->lov_lock);
499 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
500 tgt = lov->lov_tgts[index];
502 CERROR("%s: UUID %s already assigned at index %d: rc = %d\n",
503 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), index, rc);
504 mutex_unlock(&lov->lov_lock);
508 if (index >= lov->lov_tgt_size) {
509 /* We need to reallocate the lov target array. */
510 struct lov_tgt_desc **newtgts, **old = NULL;
511 __u32 newsize, oldsize = 0;
513 newsize = max(lov->lov_tgt_size, (__u32)2);
514 while (newsize < index + 1)
515 newsize = newsize << 1;
516 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
517 if (newtgts == NULL) {
518 mutex_unlock(&lov->lov_lock);
522 if (lov->lov_tgt_size) {
523 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
526 oldsize = lov->lov_tgt_size;
529 lov->lov_tgts = newtgts;
530 lov->lov_tgt_size = newsize;
533 OBD_FREE(old, sizeof(*old) * oldsize);
535 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
536 lov->lov_tgts, lov->lov_tgt_size);
541 mutex_unlock(&lov->lov_lock);
545 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
547 mutex_unlock(&lov->lov_lock);
552 tgt->ltd_uuid = *uuidp;
553 tgt->ltd_obd = tgt_obd;
554 /* XXX - add a sanity check on the generation number. */
556 tgt->ltd_index = index;
557 tgt->ltd_activate = active;
558 lov->lov_tgts[index] = tgt;
559 if (index >= lov->desc.ld_tgt_count)
560 lov->desc.ld_tgt_count = index + 1;
562 mutex_unlock(&lov->lov_lock);
564 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
565 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
567 if (lov->lov_connects == 0) {
568 /* lov_connect hasn't been called yet. We'll do the
569 lov_connect_obd on this target when that fn first runs,
570 because we don't know the connect flags yet. */
574 lov_tgts_getref(obd);
576 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
580 /* connect to administrative disabled ost */
584 if (lov->lov_cache != NULL) {
585 rc = obd_set_info_async(NULL, tgt->ltd_exp,
586 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
587 sizeof(struct cl_client_cache), lov->lov_cache,
593 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
594 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE);
598 CERROR("%s: add failed, deleting %s: rc = %d\n",
599 obd->obd_name, obd_uuid2str(&tgt->ltd_uuid), rc);
600 lov_del_target(obd, index, NULL, 0);
602 lov_tgts_putref(obd);
606 /* Schedule a target for deletion */
607 int lov_del_target(struct obd_device *obd, u32 index,
608 struct obd_uuid *uuidp, int gen)
610 struct lov_obd *lov = &obd->u.lov;
611 int count = lov->desc.ld_tgt_count;
615 if (index >= count) {
616 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
621 /* to make sure there's no ongoing lov_notify() now */
622 down_write(&lov->lov_notify_lock);
623 lov_tgts_getref(obd);
625 if (!lov->lov_tgts[index]) {
626 CERROR("LOV target at index %d is not setup.\n", index);
627 GOTO(out, rc = -EINVAL);
630 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
631 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
632 lov_uuid2str(lov, index), index,
633 obd_uuid2str(uuidp));
634 GOTO(out, rc = -EINVAL);
637 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
638 lov_uuid2str(lov, index), index,
639 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
640 lov->lov_tgts[index]->ltd_active);
642 lov->lov_tgts[index]->ltd_reap = 1;
643 lov->lov_death_row++;
644 /* we really delete it from lov_tgts_putref() */
646 lov_tgts_putref(obd);
647 up_write(&lov->lov_notify_lock);
652 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
654 struct obd_device *osc_obd;
657 LASSERT(tgt->ltd_reap);
659 osc_obd = class_exp2obd(tgt->ltd_exp);
661 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
663 osc_obd ? osc_obd->obd_name : "<no obd>");
666 lov_disconnect_obd(obd, tgt);
670 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
671 do it ourselves. And we can't do it from lov_cleanup,
672 because we just lost our only reference to it. */
674 class_manual_cleanup(osc_obd);
677 void lov_fix_desc_stripe_size(__u64 *val)
679 if (*val < LOV_MIN_STRIPE_SIZE) {
681 LCONSOLE_INFO("Increasing default stripe size to "
683 LOV_DESC_STRIPE_SIZE_DEFAULT);
684 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
685 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
686 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
687 LCONSOLE_WARN("Changing default stripe size to %llu (a "
689 *val, LOV_MIN_STRIPE_SIZE);
693 void lov_fix_desc_stripe_count(__u32 *val)
699 void lov_fix_desc_pattern(__u32 *val)
701 /* from lov_setstripe */
702 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
703 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
708 void lov_fix_desc_qos_maxage(__u32 *val)
711 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
714 void lov_fix_desc(struct lov_desc *desc)
716 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
717 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
718 lov_fix_desc_pattern(&desc->ld_pattern);
719 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
722 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
724 struct lov_desc *desc;
725 struct lov_obd *lov = &obd->u.lov;
729 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
730 CERROR("LOV setup requires a descriptor\n");
734 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
736 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
737 CERROR("descriptor size wrong: %d > %d\n",
738 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
742 if (desc->ld_magic != LOV_DESC_MAGIC) {
743 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
744 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
745 obd->obd_name, desc);
746 lustre_swab_lov_desc(desc);
748 CERROR("%s: Bad lov desc magic: %#x\n",
749 obd->obd_name, desc->ld_magic);
756 desc->ld_active_tgt_count = 0;
758 lov->lov_tgt_size = 0;
760 mutex_init(&lov->lov_lock);
761 atomic_set(&lov->lov_refcount, 0);
762 lov->lov_sp_me = LUSTRE_SP_CLI;
764 init_rwsem(&lov->lov_notify_lock);
766 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
768 HASH_POOLS_BKT_BITS, 0,
771 &pool_hash_operations,
773 INIT_LIST_HEAD(&lov->lov_pool_list);
774 lov->lov_pool_count = 0;
775 rc = lov_ost_pool_init(&lov->lov_packed, 0);
779 rc = lov_tunables_init(obd);
783 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
784 &obd->obd_kset.kobj);
790 static int lov_cleanup(struct obd_device *obd)
792 struct lov_obd *lov = &obd->u.lov;
793 struct list_head *pos, *tmp;
794 struct pool_desc *pool;
797 if (lov->lov_tgts_kobj) {
798 kobject_put(lov->lov_tgts_kobj);
799 lov->lov_tgts_kobj = NULL;
802 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
803 pool = list_entry(pos, struct pool_desc, pool_list);
804 /* free pool structs */
805 CDEBUG(D_INFO, "delete pool %p\n", pool);
806 /* In the function below, .hs_keycmp resolves to
807 * pool_hashkey_keycmp() */
808 /* coverity[overrun-buffer-val] */
809 lov_pool_del(obd, pool->pool_name);
811 cfs_hash_putref(lov->lov_pools_hash_body);
812 lov_ost_pool_free(&lov->lov_packed);
814 lprocfs_obd_cleanup(obd);
817 lov_tgts_getref(obd);
818 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
819 if (!lov->lov_tgts[i])
822 /* Inactive targets may never have connected */
823 if (lov->lov_tgts[i]->ltd_active ||
824 atomic_read(&lov->lov_refcount))
825 /* We should never get here - these
826 * should have been removed in the
828 CERROR("%s: lov tgt %d not cleaned! "
829 "deathrow=%d, lovrc=%d\n",
830 obd->obd_name, i, lov->lov_death_row,
831 atomic_read(&lov->lov_refcount));
832 lov_del_target(obd, i, NULL, 0);
834 lov_tgts_putref(obd);
835 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
837 lov->lov_tgt_size = 0;
840 if (lov->lov_cache != NULL) {
841 cl_cache_decref(lov->lov_cache);
842 lov->lov_cache = NULL;
848 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
849 u32 *indexp, int *genp)
851 struct obd_uuid obd_uuid;
856 switch (cmd = lcfg->lcfg_command) {
860 case LCFG_LOV_ADD_OBD:
861 case LCFG_LOV_ADD_INA:
862 case LCFG_LOV_DEL_OBD: {
866 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
867 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
868 GOTO(out, rc = -EINVAL);
870 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
872 rc = kstrtou32(lustre_cfg_buf(lcfg, 2), 10, indexp);
875 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
880 if (cmd == LCFG_LOV_ADD_OBD)
881 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
882 else if (cmd == LCFG_LOV_ADD_INA)
883 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
885 rc = lov_del_target(obd, index, &obd_uuid, gen);
890 struct lov_desc *desc = &(obd->u.lov.desc);
894 GOTO(out, rc = -EINVAL);
896 count = class_modify_config(lcfg, PARAM_LOV,
897 &obd->obd_kset.kobj);
898 GOTO(out, rc = count < 0 ? count : 0);
907 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
908 GOTO(out, rc = -EINVAL);
916 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
917 struct obd_statfs *osfs, time64_t max_age, __u32 flags)
919 struct obd_device *obd = class_exp2obd(exp);
920 struct lov_obd *lov = &obd->u.lov;
921 struct obd_info oinfo = {
925 struct ptlrpc_request_set *rqset;
926 struct lov_request_set *set = NULL;
927 struct lov_request *req;
933 rqset = ptlrpc_prep_set();
937 rc = lov_prep_statfs_set(obd, &oinfo, &set);
941 list_for_each_entry(req, &set->set_list, rq_link) {
942 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
943 &req->rq_oi, max_age, rqset);
948 rc = ptlrpc_set_wait(env, rqset);
952 atomic_set(&set->set_completes, 0);
954 rc2 = lov_fini_statfs_set(set);
959 ptlrpc_set_destroy(rqset);
964 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
965 void *karg, void __user *uarg)
967 struct obd_device *obd = class_exp2obd(exp);
968 struct lov_obd *lov = &obd->u.lov;
969 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
970 struct obd_uuid *uuidp;
974 case IOC_OBD_STATFS: {
975 struct obd_ioctl_data *data = karg;
976 struct obd_device *osc_obd;
977 struct obd_statfs stat_buf = {0};
978 struct obd_import *imp;
982 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
986 if (!lov->lov_tgts[index])
987 /* Try again with the next index */
989 imp = lov->lov_tgts[index]->ltd_exp->exp_obd->u.cli.cl_import;
990 if (!lov->lov_tgts[index]->ltd_active &&
991 imp->imp_state != LUSTRE_IMP_IDLE)
994 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
999 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1000 min_t(unsigned long, data->ioc_plen2,
1001 sizeof(struct obd_uuid))))
1004 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1005 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1007 /* got statfs data */
1008 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1009 ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
1013 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1014 min_t(unsigned long, data->ioc_plen1,
1015 sizeof(struct obd_statfs))))
1019 case OBD_IOC_LOV_GET_CONFIG: {
1020 struct obd_ioctl_data *data;
1021 struct lov_desc *desc;
1026 if (obd_ioctl_getdata(&buf, &len, uarg))
1029 data = (struct obd_ioctl_data *)buf;
1031 if (sizeof(*desc) > data->ioc_inllen1) {
1032 OBD_FREE_LARGE(buf, len);
1036 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1037 OBD_FREE_LARGE(buf, len);
1041 if (sizeof(__u32) * count > data->ioc_inllen3) {
1042 OBD_FREE_LARGE(buf, len);
1046 desc = (struct lov_desc *)data->ioc_inlbuf1;
1047 memcpy(desc, &(lov->desc), sizeof(*desc));
1049 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1050 genp = (__u32 *)data->ioc_inlbuf3;
1051 /* the uuid will be empty for deleted OSTs */
1052 for (i = 0; i < count; i++, uuidp++, genp++) {
1053 if (!lov->lov_tgts[i])
1055 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1056 *genp = lov->lov_tgts[i]->ltd_gen;
1059 if (copy_to_user(uarg, buf, len))
1061 OBD_FREE_LARGE(buf, len);
1064 case OBD_IOC_QUOTACTL: {
1065 struct if_quotactl *qctl = karg;
1066 struct lov_tgt_desc *tgt = NULL;
1067 struct obd_quotactl *oqctl;
1069 if (qctl->qc_valid == QC_OSTIDX) {
1070 if (count <= qctl->qc_idx)
1073 tgt = lov->lov_tgts[qctl->qc_idx];
1074 if (!tgt || !tgt->ltd_exp)
1076 } else if (qctl->qc_valid == QC_UUID) {
1077 for (i = 0; i < count; i++) {
1078 tgt = lov->lov_tgts[i];
1080 !obd_uuid_equals(&tgt->ltd_uuid,
1084 if (tgt->ltd_exp == NULL)
1096 LASSERT(tgt && tgt->ltd_exp);
1097 OBD_ALLOC_PTR(oqctl);
1101 QCTL_COPY(oqctl, qctl);
1102 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1104 QCTL_COPY(qctl, oqctl);
1105 qctl->qc_valid = QC_OSTIDX;
1106 qctl->obd_uuid = tgt->ltd_uuid;
1108 OBD_FREE_PTR(oqctl);
1117 for (i = 0; i < count; i++) {
1119 struct obd_device *osc_obd;
1121 /* OST was disconnected */
1122 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1125 /* ll_umount_begin() sets force on lov, pass to osc */
1126 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1127 osc_obd->obd_force = obd->obd_force;
1128 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1131 if (lov->lov_tgts[i]->ltd_active) {
1132 CDEBUG(err == -ENOTTY ?
1133 D_IOCTL : D_WARNING,
1134 "iocontrol OSC %s on OST "
1135 "idx %d cmd %x: err = %d\n",
1136 lov_uuid2str(lov, i),
1153 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1154 __u32 keylen, void *key, __u32 *vallen, void *val)
1156 struct obd_device *obddev = class_exp2obd(exp);
1157 struct lov_obd *lov = &obddev->u.lov;
1158 struct lov_desc *ld = &lov->desc;
1162 if (vallen == NULL || val == NULL)
1165 lov_tgts_getref(obddev);
1167 if (KEY_IS(KEY_MAX_EASIZE)) {
1168 u32 max_stripe_count = min_t(u32, ld->ld_active_tgt_count,
1169 LOV_MAX_STRIPE_COUNT);
1171 *((u32 *)val) = lov_mds_md_size(max_stripe_count, LOV_MAGIC_V3);
1172 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1173 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1174 LOV_MAX_STRIPE_COUNT);
1176 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1177 } else if (KEY_IS(KEY_TGT_COUNT)) {
1178 *((int *)val) = lov->desc.ld_tgt_count;
1183 lov_tgts_putref(obddev);
1188 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1189 __u32 keylen, void *key,
1190 __u32 vallen, void *val,
1191 struct ptlrpc_request_set *set)
1193 struct obd_device *obddev = class_exp2obd(exp);
1194 struct lov_obd *lov = &obddev->u.lov;
1195 struct lov_tgt_desc *tgt;
1196 bool do_inactive = false, no_set = false;
1205 set = ptlrpc_prep_set();
1210 lov_tgts_getref(obddev);
1212 if (KEY_IS(KEY_CHECKSUM)) {
1214 } else if (KEY_IS(KEY_CACHE_SET)) {
1215 LASSERT(lov->lov_cache == NULL);
1216 lov->lov_cache = val;
1218 cl_cache_incref(lov->lov_cache);
1221 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1222 tgt = lov->lov_tgts[i];
1224 /* OST was disconnected */
1225 if (tgt == NULL || tgt->ltd_exp == NULL)
1228 /* OST is inactive and we don't want inactive OSCs */
1229 if (!tgt->ltd_active && !do_inactive)
1232 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1239 /* cycle through MDC target for Data-on-MDT */
1240 for (i = 0; i < LOV_MDC_TGT_MAX; i++) {
1241 struct obd_device *mdc;
1243 mdc = lov->lov_mdc_tgts[i].lmtd_mdc;
1247 err = obd_set_info_async(env, mdc->obd_self_export,
1248 keylen, key, vallen, val, set);
1253 lov_tgts_putref(obddev);
1255 err = ptlrpc_set_wait(env, set);
1258 ptlrpc_set_destroy(set);
1263 void lov_stripe_lock(struct lov_stripe_md *md)
1264 __acquires(&md->lsm_lock)
1266 LASSERT(md->lsm_lock_owner != current_pid());
1267 spin_lock(&md->lsm_lock);
1268 LASSERT(md->lsm_lock_owner == 0);
1269 md->lsm_lock_owner = current_pid();
1272 void lov_stripe_unlock(struct lov_stripe_md *md)
1273 __releases(&md->lsm_lock)
1275 LASSERT(md->lsm_lock_owner == current_pid());
1276 md->lsm_lock_owner = 0;
1277 spin_unlock(&md->lsm_lock);
1280 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1281 struct obd_quotactl *oqctl)
1283 struct lov_obd *lov = &obd->u.lov;
1284 struct lov_tgt_desc *tgt;
1286 __u64 bhardlimit = 0;
1290 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1291 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA) {
1292 CERROR("%s: bad quota opc %x for lov obd\n",
1293 obd->obd_name, oqctl->qc_cmd);
1298 lov_tgts_getref(obd);
1299 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1302 tgt = lov->lov_tgts[i];
1307 if (!tgt->ltd_active || tgt->ltd_reap) {
1308 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1309 lov->lov_tgts[i]->ltd_activate) {
1311 CERROR("ost %d is inactive\n", i);
1313 CDEBUG(D_HA, "ost %d is inactive\n", i);
1318 err = obd_quotactl(tgt->ltd_exp, oqctl);
1320 if (tgt->ltd_active && !rc)
1325 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1326 curspace += oqctl->qc_dqblk.dqb_curspace;
1327 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1330 lov_tgts_putref(obd);
1332 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1333 oqctl->qc_dqblk.dqb_curspace = curspace;
1334 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1339 static struct obd_ops lov_obd_ops = {
1340 .o_owner = THIS_MODULE,
1341 .o_setup = lov_setup,
1342 .o_cleanup = lov_cleanup,
1343 .o_connect = lov_connect,
1344 .o_disconnect = lov_disconnect,
1345 .o_statfs = lov_statfs,
1346 .o_iocontrol = lov_iocontrol,
1347 .o_get_info = lov_get_info,
1348 .o_set_info_async = lov_set_info_async,
1349 .o_notify = lov_notify,
1350 .o_pool_new = lov_pool_new,
1351 .o_pool_rem = lov_pool_remove,
1352 .o_pool_add = lov_pool_add,
1353 .o_pool_del = lov_pool_del,
1354 .o_quotactl = lov_quotactl,
1357 struct kmem_cache *lov_oinfo_slab;
1359 static int __init lov_init(void)
1361 bool enable_proc = true;
1362 struct obd_type *type;
1366 /* print an address of _any_ initialized kernel symbol from this
1367 * module, to allow debugging with gdb that doesn't support data
1368 * symbols from modules.*/
1369 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1371 rc = lu_kmem_init(lov_caches);
1375 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1376 sizeof(struct lov_oinfo), 0,
1377 SLAB_HWCACHE_ALIGN, NULL);
1378 if (lov_oinfo_slab == NULL) {
1379 lu_kmem_fini(lov_caches);
1383 type = class_search_type(LUSTRE_LOD_NAME);
1384 if (type != NULL && type->typ_procsym != NULL)
1385 enable_proc = false;
1387 rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
1388 LUSTRE_LOV_NAME, &lov_device_type);
1391 kmem_cache_destroy(lov_oinfo_slab);
1392 lu_kmem_fini(lov_caches);
1398 static void __exit lov_exit(void)
1400 class_unregister_type(LUSTRE_LOV_NAME);
1401 kmem_cache_destroy(lov_oinfo_slab);
1402 lu_kmem_fini(lov_caches);
1405 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1406 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1407 MODULE_VERSION(LUSTRE_VERSION_STRING);
1408 MODULE_LICENSE("GPL");
1410 module_init(lov_init);
1411 module_exit(lov_exit);