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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lov/lov_obd.c
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Peter Braam <braam@clusterfs.com>
40 * Author: Mike Shaver <shaver@clusterfs.com>
41 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_LOV
45 #include <libcfs/libcfs.h>
47 #include <lustre/lustre_idl.h>
49 #include <cl_object.h>
50 #include <lustre_dlm.h>
51 #include <lustre_fid.h>
52 #include <lustre_ioctl.h>
53 #include <lustre_lib.h>
54 #include <lustre_mds.h>
55 #include <lustre_net.h>
56 #include <lustre_param.h>
57 #include <lustre_swab.h>
58 #include <lprocfs_status.h>
59 #include <obd_class.h>
60 #include <obd_support.h>
62 #include "lov_internal.h"
64 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
65 Any function that expects lov_tgts to remain stationary must take a ref. */
66 static void lov_getref(struct obd_device *obd)
68 struct lov_obd *lov = &obd->u.lov;
70 /* nobody gets through here until lov_putref is done */
71 mutex_lock(&lov->lov_lock);
72 atomic_inc(&lov->lov_refcount);
73 mutex_unlock(&lov->lov_lock);
77 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
79 static void lov_putref(struct obd_device *obd)
81 struct lov_obd *lov = &obd->u.lov;
83 mutex_lock(&lov->lov_lock);
84 /* ok to dec to 0 more than once -- ltd_exp's will be null */
85 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
86 struct list_head kill = LIST_HEAD_INIT(kill);
87 struct lov_tgt_desc *tgt, *n;
90 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
92 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
93 tgt = lov->lov_tgts[i];
95 if (!tgt || !tgt->ltd_reap)
97 list_add(&tgt->ltd_kill, &kill);
98 /* XXX - right now there is a dependency on ld_tgt_count
99 * being the maximum tgt index for computing the
100 * mds_max_easize. So we can't shrink it. */
101 lov_ost_pool_remove(&lov->lov_packed, i);
102 lov->lov_tgts[i] = NULL;
103 lov->lov_death_row--;
105 mutex_unlock(&lov->lov_lock);
107 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
108 list_del(&tgt->ltd_kill);
110 __lov_del_obd(obd, tgt);
113 mutex_unlock(&lov->lov_lock);
117 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
118 enum obd_notify_event ev);
119 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
120 enum obd_notify_event ev, void *data);
122 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
123 struct obd_connect_data *data)
125 struct lov_obd *lov = &obd->u.lov;
126 struct obd_uuid *tgt_uuid;
127 struct obd_device *tgt_obd;
128 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
129 struct obd_import *imp;
133 if (lov->lov_tgts[index] == NULL)
136 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
137 tgt_obd = lov->lov_tgts[index]->ltd_obd;
139 if (!tgt_obd->obd_set_up) {
140 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
144 /* override the sp_me from lov */
145 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
147 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
148 data->ocd_index = index;
151 * Divine LOV knows that OBDs under it are OSCs.
153 imp = tgt_obd->u.cli.cl_import;
156 tgt_obd->obd_no_recov = 0;
157 /* FIXME this is probably supposed to be
158 ptlrpc_set_import_active. Horrible naming. */
159 ptlrpc_activate_import(imp);
162 rc = obd_register_observer(tgt_obd, obd);
164 CERROR("Target %s register_observer error %d\n",
165 obd_uuid2str(tgt_uuid), rc);
170 if (imp->imp_invalid) {
171 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively "
172 "disabled\n", obd_uuid2str(tgt_uuid));
176 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
177 &lov_osc_uuid, data, NULL);
178 if (rc || !lov->lov_tgts[index]->ltd_exp) {
179 CERROR("Target %s connect error %d\n",
180 obd_uuid2str(tgt_uuid), rc);
184 lov->lov_tgts[index]->ltd_reap = 0;
186 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
187 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
189 if (lov->targets_proc_entry != NULL) {
190 struct proc_dir_entry *osc_symlink;
191 struct obd_device *osc_obd;
193 osc_obd = lov->lov_tgts[index]->ltd_exp->exp_obd;
195 LASSERT(osc_obd != NULL);
196 LASSERT(osc_obd->obd_magic == OBD_DEVICE_MAGIC);
197 LASSERT(osc_obd->obd_type->typ_name != NULL);
199 osc_symlink = lprocfs_add_symlink(osc_obd->obd_name,
200 lov->targets_proc_entry,
202 osc_obd->obd_type->typ_name,
204 if (osc_symlink == NULL) {
205 CERROR("cannot register LOV target "
206 "/proc/fs/lustre/%s/%s/target_obds/%s\n",
207 obd->obd_type->typ_name, obd->obd_name,
214 static int lov_connect(const struct lu_env *env,
215 struct obd_export **exp, struct obd_device *obd,
216 struct obd_uuid *cluuid, struct obd_connect_data *data,
219 struct lov_obd *lov = &obd->u.lov;
220 struct lov_tgt_desc *tgt;
221 struct lustre_handle conn;
225 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
227 rc = class_connect(&conn, obd, cluuid);
231 *exp = class_conn2export(&conn);
233 /* Why should there ever be more than 1 connect? */
235 LASSERT(lov->lov_connects == 1);
237 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
239 lov->lov_ocd = *data;
241 lov->targets_proc_entry = lprocfs_register("target_obds",
244 if (IS_ERR(lov->targets_proc_entry)) {
245 CERROR("%s: cannot register "
246 "/proc/fs/lustre/%s/%s/target_obds\n",
247 obd->obd_name, obd->obd_type->typ_name, obd->obd_name);
248 lov->targets_proc_entry = NULL;
252 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
253 tgt = lov->lov_tgts[i];
254 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
256 /* Flags will be lowest common denominator */
257 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
259 CERROR("%s: lov connect tgt %d failed: %d\n",
260 obd->obd_name, i, rc);
263 /* connect to administrative disabled ost */
264 if (!lov->lov_tgts[i]->ltd_exp)
267 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
268 OBD_NOTIFY_CONNECT, (void *)&i);
270 CERROR("%s error sending notify %d\n",
279 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
281 struct lov_obd *lov = &obd->u.lov;
282 struct obd_device *osc_obd;
286 osc_obd = class_exp2obd(tgt->ltd_exp);
287 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
288 obd->obd_name, osc_obd->obd_name);
290 if (tgt->ltd_active) {
292 lov->desc.ld_active_tgt_count--;
293 tgt->ltd_exp->exp_obd->obd_inactive = 1;
297 /* Pass it on to our clients.
298 * XXX This should be an argument to disconnect,
299 * XXX not a back-door flag on the OBD. Ah well.
301 osc_obd->obd_force = obd->obd_force;
302 osc_obd->obd_fail = obd->obd_fail;
303 osc_obd->obd_no_recov = obd->obd_no_recov;
305 if (lov->targets_proc_entry != NULL)
306 lprocfs_remove_proc_entry(osc_obd->obd_name,
307 lov->targets_proc_entry);
310 obd_register_observer(osc_obd, NULL);
312 rc = obd_disconnect(tgt->ltd_exp);
314 CERROR("Target %s disconnect error %d\n",
315 tgt->ltd_uuid.uuid, rc);
323 static int lov_disconnect(struct obd_export *exp)
325 struct obd_device *obd = class_exp2obd(exp);
326 struct lov_obd *lov = &obd->u.lov;
333 /* Only disconnect the underlying layers on the final disconnect. */
335 if (lov->lov_connects != 0) {
336 /* why should there be more than 1 connect? */
337 CERROR("disconnect #%d\n", lov->lov_connects);
341 /* Let's hold another reference so lov_del_obd doesn't spin through
345 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
346 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
347 /* Disconnection is the last we know about an obd */
348 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
353 if (lov->targets_proc_entry != NULL)
354 lprocfs_remove(&lov->targets_proc_entry);
357 rc = class_disconnect(exp); /* bz 9811 */
363 * -EINVAL : UUID can't be found in the LOV's target list
364 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
365 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
366 * any >= 0 : is log target index
368 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
369 enum obd_notify_event ev)
371 struct lov_obd *lov = &obd->u.lov;
372 struct lov_tgt_desc *tgt;
373 int index, activate, active;
376 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
377 lov, uuid->uuid, ev);
380 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
381 tgt = lov->lov_tgts[index];
385 * LU-642, initially inactive OSC could miss the obd_connect,
386 * we make up for it here.
388 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
389 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
390 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
392 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
393 &lov_osc_uuid, &lov->lov_ocd, NULL);
398 CDEBUG(D_INFO, "lov idx %d is %s conn "LPX64"\n",
399 index, obd_uuid2str(&tgt->ltd_uuid),
400 tgt->ltd_exp->exp_handle.h_cookie);
401 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
405 if (index == lov->desc.ld_tgt_count)
406 GOTO(out, index = -EINVAL);
408 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
409 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
411 if (lov->lov_tgts[index]->ltd_activate == activate) {
412 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
413 uuid->uuid, activate ? "" : "de");
415 lov->lov_tgts[index]->ltd_activate = activate;
416 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
417 activate ? "" : "de", obd_uuid2str(uuid));
420 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
421 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
423 if (lov->lov_tgts[index]->ltd_active == active) {
424 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
425 uuid->uuid, active ? "" : "in");
428 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
429 obd_uuid2str(uuid), active ? "" : "in");
432 lov->lov_tgts[index]->ltd_active = active;
434 lov->desc.ld_active_tgt_count++;
435 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
437 lov->desc.ld_active_tgt_count--;
438 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
441 CERROR("%s: unknown event %d for uuid %s\n", obd->obd_name,
450 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
451 enum obd_notify_event ev, void *data)
454 struct lov_obd *lov = &obd->u.lov;
457 down_read(&lov->lov_notify_lock);
458 if (!lov->lov_connects) {
459 up_read(&lov->lov_notify_lock);
463 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
464 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
465 struct obd_uuid *uuid;
469 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
470 up_read(&lov->lov_notify_lock);
471 CERROR("unexpected notification of %s %s!\n",
472 watched->obd_type->typ_name,
476 uuid = &watched->u.cli.cl_target_uuid;
478 /* Set OSC as active before notifying the observer, so the
479 * observer can use the OSC normally.
481 rc = lov_set_osc_active(obd, uuid, ev);
483 up_read(&lov->lov_notify_lock);
484 CERROR("event(%d) of %s failed: %d\n", ev,
485 obd_uuid2str(uuid), rc);
488 /* active event should be pass lov target index as data */
492 /* Pass the notification up the chain. */
494 rc = obd_notify_observer(obd, watched, ev, data);
496 /* NULL watched means all osc's in the lov (only for syncs) */
497 /* sync event should be send lov idx as data */
498 struct lov_obd *lov = &obd->u.lov;
502 is_sync = (ev == OBD_NOTIFY_SYNC) ||
503 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
506 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
507 if (!lov->lov_tgts[i])
510 /* don't send sync event if target not
511 * connected/activated */
512 if (is_sync && !lov->lov_tgts[i]->ltd_active)
515 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
518 CERROR("%s: notify %s of %s failed %d\n",
520 obd->obd_observer->obd_name,
521 lov->lov_tgts[i]->ltd_obd->obd_name,
528 up_read(&lov->lov_notify_lock);
532 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
533 __u32 index, int gen, int active)
535 struct lov_obd *lov = &obd->u.lov;
536 struct lov_tgt_desc *tgt;
537 struct obd_device *tgt_obd;
541 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
542 uuidp->uuid, index, gen, active);
545 CERROR("request to add OBD %s with invalid generation: %d\n",
550 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
555 mutex_lock(&lov->lov_lock);
557 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
558 tgt = lov->lov_tgts[index];
559 CERROR("UUID %s already assigned at LOV target index %d\n",
560 obd_uuid2str(&tgt->ltd_uuid), index);
561 mutex_unlock(&lov->lov_lock);
565 if (index >= lov->lov_tgt_size) {
566 /* We need to reallocate the lov target array. */
567 struct lov_tgt_desc **newtgts, **old = NULL;
568 __u32 newsize, oldsize = 0;
570 newsize = max(lov->lov_tgt_size, (__u32)2);
571 while (newsize < index + 1)
572 newsize = newsize << 1;
573 OBD_ALLOC(newtgts, sizeof(*newtgts) * newsize);
574 if (newtgts == NULL) {
575 mutex_unlock(&lov->lov_lock);
579 if (lov->lov_tgt_size) {
580 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
583 oldsize = lov->lov_tgt_size;
586 lov->lov_tgts = newtgts;
587 lov->lov_tgt_size = newsize;
590 OBD_FREE(old, sizeof(*old) * oldsize);
592 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
593 lov->lov_tgts, lov->lov_tgt_size);
598 mutex_unlock(&lov->lov_lock);
602 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
604 mutex_unlock(&lov->lov_lock);
609 tgt->ltd_uuid = *uuidp;
610 tgt->ltd_obd = tgt_obd;
611 /* XXX - add a sanity check on the generation number. */
613 tgt->ltd_index = index;
614 tgt->ltd_activate = active;
615 lov->lov_tgts[index] = tgt;
616 if (index >= lov->desc.ld_tgt_count)
617 lov->desc.ld_tgt_count = index + 1;
619 mutex_unlock(&lov->lov_lock);
621 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
622 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
624 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
626 if (lov->lov_connects == 0) {
627 /* lov_connect hasn't been called yet. We'll do the
628 lov_connect_obd on this target when that fn first runs,
629 because we don't know the connect flags yet. */
635 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
639 /* connect to administrative disabled ost */
643 if (lov->lov_cache != NULL) {
644 rc = obd_set_info_async(NULL, tgt->ltd_exp,
645 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
646 sizeof(struct cl_client_cache), lov->lov_cache,
652 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
653 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
658 CERROR("add failed (%d), deleting %s\n", rc,
659 obd_uuid2str(&tgt->ltd_uuid));
660 lov_del_target(obd, index, NULL, 0);
666 /* Schedule a target for deletion */
667 int lov_del_target(struct obd_device *obd, __u32 index,
668 struct obd_uuid *uuidp, int gen)
670 struct lov_obd *lov = &obd->u.lov;
671 int count = lov->desc.ld_tgt_count;
675 if (index >= count) {
676 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
681 /* to make sure there's no ongoing lov_notify() now */
682 down_write(&lov->lov_notify_lock);
685 if (!lov->lov_tgts[index]) {
686 CERROR("LOV target at index %d is not setup.\n", index);
687 GOTO(out, rc = -EINVAL);
690 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
691 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
692 lov_uuid2str(lov, index), index,
693 obd_uuid2str(uuidp));
694 GOTO(out, rc = -EINVAL);
697 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
698 lov_uuid2str(lov, index), index,
699 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
700 lov->lov_tgts[index]->ltd_active);
702 lov->lov_tgts[index]->ltd_reap = 1;
703 lov->lov_death_row++;
704 /* we really delete it from obd_putref */
707 up_write(&lov->lov_notify_lock);
712 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
714 struct obd_device *osc_obd;
717 LASSERT(tgt->ltd_reap);
719 osc_obd = class_exp2obd(tgt->ltd_exp);
721 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
723 osc_obd ? osc_obd->obd_name : "<no obd>");
726 lov_disconnect_obd(obd, tgt);
730 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
731 do it ourselves. And we can't do it from lov_cleanup,
732 because we just lost our only reference to it. */
734 class_manual_cleanup(osc_obd);
737 void lov_fix_desc_stripe_size(__u64 *val)
739 if (*val < LOV_MIN_STRIPE_SIZE) {
741 LCONSOLE_INFO("Increasing default stripe size to "
743 LOV_DESC_STRIPE_SIZE_DEFAULT);
744 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
745 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
746 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
747 LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
749 *val, LOV_MIN_STRIPE_SIZE);
753 void lov_fix_desc_stripe_count(__u32 *val)
759 void lov_fix_desc_pattern(__u32 *val)
761 /* from lov_setstripe */
762 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
763 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
768 void lov_fix_desc_qos_maxage(__u32 *val)
771 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
774 void lov_fix_desc(struct lov_desc *desc)
776 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
777 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
778 lov_fix_desc_pattern(&desc->ld_pattern);
779 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
782 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
784 struct lov_desc *desc;
785 struct lov_obd *lov = &obd->u.lov;
786 #ifdef CONFIG_PROC_FS
787 struct obd_type *type;
792 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
793 CERROR("LOV setup requires a descriptor\n");
797 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
799 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
800 CERROR("descriptor size wrong: %d > %d\n",
801 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
805 if (desc->ld_magic != LOV_DESC_MAGIC) {
806 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
807 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
808 obd->obd_name, desc);
809 lustre_swab_lov_desc(desc);
811 CERROR("%s: Bad lov desc magic: %#x\n",
812 obd->obd_name, desc->ld_magic);
819 desc->ld_active_tgt_count = 0;
821 lov->lov_tgt_size = 0;
823 mutex_init(&lov->lov_lock);
824 atomic_set(&lov->lov_refcount, 0);
825 lov->lov_sp_me = LUSTRE_SP_CLI;
827 init_rwsem(&lov->lov_notify_lock);
829 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
831 HASH_POOLS_BKT_BITS, 0,
834 &pool_hash_operations,
836 INIT_LIST_HEAD(&lov->lov_pool_list);
837 lov->lov_pool_count = 0;
838 rc = lov_ost_pool_init(&lov->lov_packed, 0);
842 #ifdef CONFIG_PROC_FS
843 obd->obd_vars = lprocfs_lov_obd_vars;
844 /* If this is true then both client (lov) and server
845 * (lod) are on the same node. The lod layer if loaded
846 * first will register the lov proc directory. In that
847 * case obd->obd_type->typ_procroot will be not set.
848 * Instead we use type->typ_procsym as the parent. */
849 type = class_search_type(LUSTRE_LOD_NAME);
850 if (type != NULL && type->typ_procsym != NULL) {
851 obd->obd_proc_entry = lprocfs_register(obd->obd_name,
854 if (IS_ERR(obd->obd_proc_entry)) {
855 rc = PTR_ERR(obd->obd_proc_entry);
856 CERROR("error %d setting up lprocfs for %s\n", rc,
858 obd->obd_proc_entry = NULL;
861 rc = lprocfs_obd_setup(obd);
865 rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
866 0444, &lov_proc_target_fops, obd);
868 CWARN("Error adding the target_obd file\n");
870 lov->lov_pool_proc_entry = lprocfs_register("pools",
873 if (IS_ERR(lov->lov_pool_proc_entry)) {
874 rc = PTR_ERR(lov->lov_pool_proc_entry);
875 CERROR("error %d setting up lprocfs for pools\n", rc);
876 lov->lov_pool_proc_entry = NULL;
886 static int lov_cleanup(struct obd_device *obd)
888 struct lov_obd *lov = &obd->u.lov;
889 struct list_head *pos, *tmp;
890 struct pool_desc *pool;
893 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
894 pool = list_entry(pos, struct pool_desc, pool_list);
895 /* free pool structs */
896 CDEBUG(D_INFO, "delete pool %p\n", pool);
897 /* In the function below, .hs_keycmp resolves to
898 * pool_hashkey_keycmp() */
899 /* coverity[overrun-buffer-val] */
900 lov_pool_del(obd, pool->pool_name);
902 cfs_hash_putref(lov->lov_pools_hash_body);
903 lov_ost_pool_free(&lov->lov_packed);
905 lprocfs_obd_cleanup(obd);
909 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
910 if (!lov->lov_tgts[i])
913 /* Inactive targets may never have connected */
914 if (lov->lov_tgts[i]->ltd_active ||
915 atomic_read(&lov->lov_refcount))
916 /* We should never get here - these
917 * should have been removed in the
919 CERROR("%s: lov tgt %d not cleaned! "
920 "deathrow=%d, lovrc=%d\n",
921 obd->obd_name, i, lov->lov_death_row,
922 atomic_read(&lov->lov_refcount));
923 lov_del_target(obd, i, NULL, 0);
926 OBD_FREE(lov->lov_tgts, sizeof(*lov->lov_tgts) *
928 lov->lov_tgt_size = 0;
931 if (lov->lov_cache != NULL) {
932 cl_cache_decref(lov->lov_cache);
933 lov->lov_cache = NULL;
939 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
940 __u32 *indexp, int *genp)
942 struct obd_uuid obd_uuid;
947 switch(cmd = lcfg->lcfg_command) {
948 case LCFG_LOV_ADD_OBD:
949 case LCFG_LOV_ADD_INA:
950 case LCFG_LOV_DEL_OBD: {
953 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
954 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid))
955 GOTO(out, rc = -EINVAL);
957 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
959 if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", indexp) != 1)
960 GOTO(out, rc = -EINVAL);
961 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1)
962 GOTO(out, rc = -EINVAL);
965 if (cmd == LCFG_LOV_ADD_OBD)
966 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
967 else if (cmd == LCFG_LOV_ADD_INA)
968 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
970 rc = lov_del_target(obd, index, &obd_uuid, gen);
974 struct lov_desc *desc = &(obd->u.lov.desc);
977 GOTO(out, rc = -EINVAL);
979 rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
992 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
993 GOTO(out, rc = -EINVAL);
1001 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1003 struct lov_request_set *lovset = (struct lov_request_set *)data;
1008 atomic_set(&lovset->set_completes, 0);
1010 err = lov_fini_statfs_set(lovset);
1011 RETURN(rc ? rc : err);
1014 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1015 __u64 max_age, struct ptlrpc_request_set *rqset)
1017 struct obd_device *obd = class_exp2obd(exp);
1018 struct lov_request_set *set;
1019 struct lov_request *req;
1020 struct list_head *pos;
1021 struct lov_obd *lov;
1025 LASSERT(oinfo != NULL);
1026 LASSERT(oinfo->oi_osfs != NULL);
1029 rc = lov_prep_statfs_set(obd, oinfo, &set);
1033 list_for_each(pos, &set->set_list) {
1034 req = list_entry(pos, struct lov_request, rq_link);
1035 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1036 &req->rq_oi, max_age, rqset);
1041 if (rc || list_empty(&rqset->set_requests)) {
1044 atomic_set(&set->set_completes, 0);
1045 err = lov_fini_statfs_set(set);
1046 RETURN(rc ? rc : err);
1049 LASSERT(rqset->set_interpret == NULL);
1050 rqset->set_interpret = lov_statfs_interpret;
1051 rqset->set_arg = (void *)set;
1055 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1056 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1058 struct ptlrpc_request_set *set = NULL;
1059 struct obd_info oinfo = {
1067 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1068 * statfs requests */
1069 set = ptlrpc_prep_set();
1073 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1075 rc = ptlrpc_set_wait(set);
1077 ptlrpc_set_destroy(set);
1082 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1083 void *karg, void __user *uarg)
1085 struct obd_device *obddev = class_exp2obd(exp);
1086 struct lov_obd *lov = &obddev->u.lov;
1087 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1088 struct obd_uuid *uuidp;
1092 case IOC_OBD_STATFS: {
1093 struct obd_ioctl_data *data = karg;
1094 struct obd_device *osc_obd;
1095 struct obd_statfs stat_buf = {0};
1099 memcpy(&index, data->ioc_inlbuf2, sizeof(index));
1100 if ((index >= count))
1103 if (!lov->lov_tgts[index])
1104 /* Try again with the next index */
1106 if (!lov->lov_tgts[index]->ltd_active)
1109 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1114 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1115 min_t(unsigned long, data->ioc_plen2,
1116 sizeof(struct obd_uuid))))
1119 memcpy(&flags, data->ioc_inlbuf1, sizeof(flags));
1120 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1122 /* got statfs data */
1123 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1124 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1128 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1129 min_t(unsigned long, data->ioc_plen1,
1130 sizeof(struct obd_statfs))))
1134 case OBD_IOC_LOV_GET_CONFIG: {
1135 struct obd_ioctl_data *data;
1136 struct lov_desc *desc;
1141 if (obd_ioctl_getdata(&buf, &len, uarg))
1144 data = (struct obd_ioctl_data *)buf;
1146 if (sizeof(*desc) > data->ioc_inllen1) {
1147 obd_ioctl_freedata(buf, len);
1151 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1152 obd_ioctl_freedata(buf, len);
1156 if (sizeof(__u32) * count > data->ioc_inllen3) {
1157 obd_ioctl_freedata(buf, len);
1161 desc = (struct lov_desc *)data->ioc_inlbuf1;
1162 memcpy(desc, &(lov->desc), sizeof(*desc));
1164 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1165 genp = (__u32 *)data->ioc_inlbuf3;
1166 /* the uuid will be empty for deleted OSTs */
1167 for (i = 0; i < count; i++, uuidp++, genp++) {
1168 if (!lov->lov_tgts[i])
1170 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1171 *genp = lov->lov_tgts[i]->ltd_gen;
1174 if (copy_to_user(uarg, buf, len))
1176 obd_ioctl_freedata(buf, len);
1179 case OBD_IOC_QUOTACTL: {
1180 struct if_quotactl *qctl = karg;
1181 struct lov_tgt_desc *tgt = NULL;
1182 struct obd_quotactl *oqctl;
1184 if (qctl->qc_valid == QC_OSTIDX) {
1185 if (count <= qctl->qc_idx)
1188 tgt = lov->lov_tgts[qctl->qc_idx];
1189 if (!tgt || !tgt->ltd_exp)
1191 } else if (qctl->qc_valid == QC_UUID) {
1192 for (i = 0; i < count; i++) {
1193 tgt = lov->lov_tgts[i];
1195 !obd_uuid_equals(&tgt->ltd_uuid,
1199 if (tgt->ltd_exp == NULL)
1211 LASSERT(tgt && tgt->ltd_exp);
1212 OBD_ALLOC_PTR(oqctl);
1216 QCTL_COPY(oqctl, qctl);
1217 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1219 QCTL_COPY(qctl, oqctl);
1220 qctl->qc_valid = QC_OSTIDX;
1221 qctl->obd_uuid = tgt->ltd_uuid;
1223 OBD_FREE_PTR(oqctl);
1232 for (i = 0; i < count; i++) {
1234 struct obd_device *osc_obd;
1236 /* OST was disconnected */
1237 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1240 /* ll_umount_begin() sets force flag but for lov, not
1241 * osc. Let's pass it through */
1242 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1243 osc_obd->obd_force = obddev->obd_force;
1244 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1247 if (lov->lov_tgts[i]->ltd_active) {
1248 CDEBUG(err == -ENOTTY ?
1249 D_IOCTL : D_WARNING,
1250 "iocontrol OSC %s on OST "
1251 "idx %d cmd %x: err = %d\n",
1252 lov_uuid2str(lov, i),
1269 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1270 __u32 keylen, void *key, __u32 *vallen, void *val)
1272 struct obd_device *obddev = class_exp2obd(exp);
1273 struct lov_obd *lov = &obddev->u.lov;
1274 struct lov_desc *ld = &lov->desc;
1278 if (vallen == NULL || val == NULL)
1283 if (KEY_IS(KEY_MAX_EASIZE)) {
1284 u32 max_stripe_count = min_t(u32, ld->ld_active_tgt_count,
1285 LOV_MAX_STRIPE_COUNT);
1287 *((u32 *)val) = lov_mds_md_size(max_stripe_count, LOV_MAGIC_V3);
1288 } else if (KEY_IS(KEY_DEFAULT_EASIZE)) {
1289 u32 def_stripe_count = min_t(u32, ld->ld_default_stripe_count,
1290 LOV_MAX_STRIPE_COUNT);
1292 *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
1293 } else if (KEY_IS(KEY_TGT_COUNT)) {
1294 *((int *)val) = lov->desc.ld_tgt_count;
1304 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
1305 __u32 keylen, void *key,
1306 __u32 vallen, void *val,
1307 struct ptlrpc_request_set *set)
1309 struct obd_device *obddev = class_exp2obd(exp);
1310 struct lov_obd *lov = &obddev->u.lov;
1311 struct lov_tgt_desc *tgt;
1312 int do_inactive = 0;
1322 set = ptlrpc_prep_set();
1328 count = lov->desc.ld_tgt_count;
1330 if (KEY_IS(KEY_CHECKSUM)) {
1332 } else if (KEY_IS(KEY_CACHE_SET)) {
1333 LASSERT(lov->lov_cache == NULL);
1334 lov->lov_cache = val;
1336 cl_cache_incref(lov->lov_cache);
1339 for (i = 0; i < count; i++) {
1340 tgt = lov->lov_tgts[i];
1342 /* OST was disconnected */
1343 if (!tgt || !tgt->ltd_exp)
1346 /* OST is inactive and we don't want inactive OSCs */
1347 if (!tgt->ltd_active && !do_inactive)
1350 err = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
1358 err = ptlrpc_set_wait(set);
1361 ptlrpc_set_destroy(set);
1366 void lov_stripe_lock(struct lov_stripe_md *md)
1367 __acquires(&md->lsm_lock)
1369 LASSERT(md->lsm_lock_owner != current_pid());
1370 spin_lock(&md->lsm_lock);
1371 LASSERT(md->lsm_lock_owner == 0);
1372 md->lsm_lock_owner = current_pid();
1375 void lov_stripe_unlock(struct lov_stripe_md *md)
1376 __releases(&md->lsm_lock)
1378 LASSERT(md->lsm_lock_owner == current_pid());
1379 md->lsm_lock_owner = 0;
1380 spin_unlock(&md->lsm_lock);
1383 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
1384 struct obd_quotactl *oqctl)
1386 struct lov_obd *lov = &obd->u.lov;
1387 struct lov_tgt_desc *tgt;
1389 __u64 bhardlimit = 0;
1393 if (oqctl->qc_cmd != Q_GETOQUOTA &&
1394 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA) {
1395 CERROR("%s: bad quota opc %x for lov obd\n",
1396 obd->obd_name, oqctl->qc_cmd);
1402 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
1405 tgt = lov->lov_tgts[i];
1410 if (!tgt->ltd_active || tgt->ltd_reap) {
1411 if (oqctl->qc_cmd == Q_GETOQUOTA &&
1412 lov->lov_tgts[i]->ltd_activate) {
1414 CERROR("ost %d is inactive\n", i);
1416 CDEBUG(D_HA, "ost %d is inactive\n", i);
1421 err = obd_quotactl(tgt->ltd_exp, oqctl);
1423 if (tgt->ltd_active && !rc)
1428 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1429 curspace += oqctl->qc_dqblk.dqb_curspace;
1430 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
1435 if (oqctl->qc_cmd == Q_GETOQUOTA) {
1436 oqctl->qc_dqblk.dqb_curspace = curspace;
1437 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
1442 static struct obd_ops lov_obd_ops = {
1443 .o_owner = THIS_MODULE,
1444 .o_setup = lov_setup,
1445 .o_cleanup = lov_cleanup,
1446 .o_connect = lov_connect,
1447 .o_disconnect = lov_disconnect,
1448 .o_statfs = lov_statfs,
1449 .o_statfs_async = lov_statfs_async,
1450 .o_iocontrol = lov_iocontrol,
1451 .o_get_info = lov_get_info,
1452 .o_set_info_async = lov_set_info_async,
1453 .o_notify = lov_notify,
1454 .o_pool_new = lov_pool_new,
1455 .o_pool_rem = lov_pool_remove,
1456 .o_pool_add = lov_pool_add,
1457 .o_pool_del = lov_pool_del,
1458 .o_getref = lov_getref,
1459 .o_putref = lov_putref,
1460 .o_quotactl = lov_quotactl,
1463 struct kmem_cache *lov_oinfo_slab;
1465 static int __init lov_init(void)
1467 bool enable_proc = true;
1468 struct obd_type *type;
1472 /* print an address of _any_ initialized kernel symbol from this
1473 * module, to allow debugging with gdb that doesn't support data
1474 * symbols from modules.*/
1475 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
1477 rc = lu_kmem_init(lov_caches);
1481 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
1482 sizeof(struct lov_oinfo), 0,
1483 SLAB_HWCACHE_ALIGN, NULL);
1484 if (lov_oinfo_slab == NULL) {
1485 lu_kmem_fini(lov_caches);
1489 type = class_search_type(LUSTRE_LOD_NAME);
1490 if (type != NULL && type->typ_procsym != NULL)
1491 enable_proc = false;
1493 rc = class_register_type(&lov_obd_ops, NULL, enable_proc, NULL,
1494 LUSTRE_LOV_NAME, &lov_device_type);
1497 kmem_cache_destroy(lov_oinfo_slab);
1498 lu_kmem_fini(lov_caches);
1504 static void __exit lov_exit(void)
1506 class_unregister_type(LUSTRE_LOV_NAME);
1507 kmem_cache_destroy(lov_oinfo_slab);
1508 lu_kmem_fini(lov_caches);
1511 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
1512 MODULE_DESCRIPTION("Lustre Logical Object Volume");
1513 MODULE_VERSION(LUSTRE_VERSION_STRING);
1514 MODULE_LICENSE("GPL");
1516 module_init(lov_init);
1517 module_exit(lov_exit);