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, 2015, 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 "../../include/linux/libcfs/libcfs.h"
43 #include "../include/obd_support.h"
44 #include "../include/lustre/lustre_ioctl.h"
45 #include "../include/lustre_lib.h"
46 #include "../include/lustre_net.h"
47 #include "../include/lustre/lustre_idl.h"
48 #include "../include/lustre_dlm.h"
49 #include "../include/lustre_mds.h"
50 #include "../include/obd_class.h"
51 #include "../include/lprocfs_status.h"
52 #include "../include/lustre_param.h"
53 #include "../include/cl_object.h"
54 #include "../include/lustre/ll_fiemap.h"
55 #include "../include/lustre_fid.h"
57 #include "lov_internal.h"
59 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
60 * Any function that expects lov_tgts to remain stationary must take a ref.
62 static void lov_getref(struct obd_device *obd)
64 struct lov_obd *lov = &obd->u.lov;
66 /* nobody gets through here until lov_putref is done */
67 mutex_lock(&lov->lov_lock);
68 atomic_inc(&lov->lov_refcount);
69 mutex_unlock(&lov->lov_lock);
73 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
75 static void lov_putref(struct obd_device *obd)
77 struct lov_obd *lov = &obd->u.lov;
79 mutex_lock(&lov->lov_lock);
80 /* ok to dec to 0 more than once -- ltd_exp's will be null */
81 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
84 struct lov_tgt_desc *tgt, *n;
86 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
88 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
89 tgt = lov->lov_tgts[i];
91 if (!tgt || !tgt->ltd_reap)
93 list_add(&tgt->ltd_kill, &kill);
94 /* XXX - right now there is a dependency on ld_tgt_count
95 * being the maximum tgt index for computing the
96 * mds_max_easize. So we can't shrink it.
98 lov_ost_pool_remove(&lov->lov_packed, i);
99 lov->lov_tgts[i] = NULL;
100 lov->lov_death_row--;
102 mutex_unlock(&lov->lov_lock);
104 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
105 list_del(&tgt->ltd_kill);
107 __lov_del_obd(obd, tgt);
110 if (lov->lov_tgts_kobj)
111 kobject_put(lov->lov_tgts_kobj);
114 mutex_unlock(&lov->lov_lock);
118 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
119 enum obd_notify_event ev);
120 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
121 enum obd_notify_event ev, void *data);
123 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
124 struct obd_connect_data *data)
126 struct lov_obd *lov = &obd->u.lov;
127 struct obd_uuid *tgt_uuid;
128 struct obd_device *tgt_obd;
129 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
130 struct obd_import *imp;
133 if (!lov->lov_tgts[index])
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.
160 ptlrpc_activate_import(imp);
163 rc = obd_register_observer(tgt_obd, obd);
165 CERROR("Target %s register_observer error %d\n",
166 obd_uuid2str(tgt_uuid), rc);
170 if (imp->imp_invalid) {
171 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n",
172 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->lov_tgts_kobj)
190 /* Even if we failed, that's ok */
191 rc = sysfs_create_link(lov->lov_tgts_kobj, &tgt_obd->obd_kobj,
197 static int lov_connect(const struct lu_env *env,
198 struct obd_export **exp, struct obd_device *obd,
199 struct obd_uuid *cluuid, struct obd_connect_data *data,
202 struct lov_obd *lov = &obd->u.lov;
203 struct lov_tgt_desc *tgt;
204 struct lustre_handle conn;
207 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
209 rc = class_connect(&conn, obd, cluuid);
213 *exp = class_conn2export(&conn);
215 /* Why should there ever be more than 1 connect? */
217 LASSERT(lov->lov_connects == 1);
219 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
221 lov->lov_ocd = *data;
225 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
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,
244 OBD_NOTIFY_CONNECT, (void *)&i);
246 CERROR("%s error sending notify %d\n",
255 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
257 struct lov_obd *lov = &obd->u.lov;
258 struct obd_device *osc_obd;
261 osc_obd = class_exp2obd(tgt->ltd_exp);
262 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
263 obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL");
265 if (tgt->ltd_active) {
267 lov->desc.ld_active_tgt_count--;
268 tgt->ltd_exp->exp_obd->obd_inactive = 1;
272 if (lov->lov_tgts_kobj)
273 sysfs_remove_link(lov->lov_tgts_kobj,
276 /* Pass it on to our clients.
277 * XXX This should be an argument to disconnect,
278 * XXX not a back-door flag on the OBD. Ah well.
280 osc_obd->obd_force = obd->obd_force;
281 osc_obd->obd_fail = obd->obd_fail;
282 osc_obd->obd_no_recov = obd->obd_no_recov;
285 obd_register_observer(osc_obd, NULL);
287 rc = obd_disconnect(tgt->ltd_exp);
289 CERROR("Target %s disconnect error %d\n",
290 tgt->ltd_uuid.uuid, rc);
298 static int lov_disconnect(struct obd_export *exp)
300 struct obd_device *obd = class_exp2obd(exp);
301 struct lov_obd *lov = &obd->u.lov;
307 /* Only disconnect the underlying layers on the final disconnect. */
309 if (lov->lov_connects != 0) {
310 /* why should there be more than 1 connect? */
311 CERROR("disconnect #%d\n", lov->lov_connects);
315 /* Let's hold another reference so lov_del_obd doesn't spin through
320 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
321 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
322 /* Disconnection is the last we know about an obd */
323 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
330 rc = class_disconnect(exp); /* bz 9811 */
336 * -EINVAL : UUID can't be found in the LOV's target list
337 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
338 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
339 * any >= 0 : is log target index
341 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
342 enum obd_notify_event ev)
344 struct lov_obd *lov = &obd->u.lov;
345 struct lov_tgt_desc *tgt;
346 int index, activate, active;
348 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
349 lov, uuid->uuid, ev);
352 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
353 tgt = lov->lov_tgts[index];
357 * LU-642, initially inactive OSC could miss the obd_connect,
358 * we make up for it here.
360 if (ev == OBD_NOTIFY_ACTIVATE && !tgt->ltd_exp &&
361 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
362 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
364 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
365 &lov_osc_uuid, &lov->lov_ocd, NULL);
370 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
371 index, obd_uuid2str(&tgt->ltd_uuid),
372 tgt->ltd_exp->exp_handle.h_cookie);
373 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
377 if (index == lov->desc.ld_tgt_count) {
382 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
383 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
385 if (lov->lov_tgts[index]->ltd_activate == activate) {
386 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
387 uuid->uuid, activate ? "" : "de");
389 lov->lov_tgts[index]->ltd_activate = activate;
390 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
391 activate ? "" : "de", obd_uuid2str(uuid));
394 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
395 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
397 if (lov->lov_tgts[index]->ltd_active == active) {
398 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
399 uuid->uuid, active ? "" : "in");
402 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
403 obd_uuid2str(uuid), active ? "" : "in");
405 lov->lov_tgts[index]->ltd_active = active;
407 lov->desc.ld_active_tgt_count++;
408 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
410 lov->desc.ld_active_tgt_count--;
411 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
414 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
422 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
423 enum obd_notify_event ev, void *data)
426 struct lov_obd *lov = &obd->u.lov;
428 down_read(&lov->lov_notify_lock);
429 if (!lov->lov_connects) {
430 up_read(&lov->lov_notify_lock);
434 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
435 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
436 struct obd_uuid *uuid;
440 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
441 up_read(&lov->lov_notify_lock);
442 CERROR("unexpected notification of %s %s!\n",
443 watched->obd_type->typ_name,
447 uuid = &watched->u.cli.cl_target_uuid;
449 /* Set OSC as active before notifying the observer, so the
450 * observer can use the OSC normally.
452 rc = lov_set_osc_active(obd, uuid, ev);
454 up_read(&lov->lov_notify_lock);
455 CERROR("event(%d) of %s failed: %d\n", ev,
456 obd_uuid2str(uuid), rc);
459 /* active event should be pass lov target index as data */
463 /* Pass the notification up the chain. */
465 rc = obd_notify_observer(obd, watched, ev, data);
467 /* NULL watched means all osc's in the lov (only for syncs) */
468 /* sync event should be send lov idx as data */
469 struct lov_obd *lov = &obd->u.lov;
473 is_sync = (ev == OBD_NOTIFY_SYNC) ||
474 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
477 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
478 if (!lov->lov_tgts[i])
481 /* don't send sync event if target not
482 * connected/activated
484 if (is_sync && !lov->lov_tgts[i]->ltd_active)
487 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
490 CERROR("%s: notify %s of %s failed %d\n",
492 obd->obd_observer->obd_name,
493 lov->lov_tgts[i]->ltd_obd->obd_name,
500 up_read(&lov->lov_notify_lock);
504 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
505 __u32 index, int gen, int active)
507 struct lov_obd *lov = &obd->u.lov;
508 struct lov_tgt_desc *tgt;
509 struct obd_device *tgt_obd;
512 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
513 uuidp->uuid, index, gen, active);
516 CERROR("request to add OBD %s with invalid generation: %d\n",
521 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
526 mutex_lock(&lov->lov_lock);
528 if ((index < lov->lov_tgt_size) && lov->lov_tgts[index]) {
529 tgt = lov->lov_tgts[index];
530 CERROR("UUID %s already assigned at LOV target index %d\n",
531 obd_uuid2str(&tgt->ltd_uuid), index);
532 mutex_unlock(&lov->lov_lock);
536 if (index >= lov->lov_tgt_size) {
537 /* We need to reallocate the lov target array. */
538 struct lov_tgt_desc **newtgts, **old = NULL;
539 __u32 newsize, oldsize = 0;
541 newsize = max_t(__u32, lov->lov_tgt_size, 2);
542 while (newsize < index + 1)
544 newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS);
546 mutex_unlock(&lov->lov_lock);
550 if (lov->lov_tgt_size) {
551 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
554 oldsize = lov->lov_tgt_size;
557 lov->lov_tgts = newtgts;
558 lov->lov_tgt_size = newsize;
562 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
563 lov->lov_tgts, lov->lov_tgt_size);
566 tgt = kzalloc(sizeof(*tgt), GFP_NOFS);
568 mutex_unlock(&lov->lov_lock);
572 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
574 mutex_unlock(&lov->lov_lock);
579 tgt->ltd_uuid = *uuidp;
580 tgt->ltd_obd = tgt_obd;
581 /* XXX - add a sanity check on the generation number. */
583 tgt->ltd_index = index;
584 tgt->ltd_activate = active;
585 lov->lov_tgts[index] = tgt;
586 if (index >= lov->desc.ld_tgt_count)
587 lov->desc.ld_tgt_count = index + 1;
589 mutex_unlock(&lov->lov_lock);
591 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
592 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
594 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
596 if (lov->lov_connects == 0) {
597 /* lov_connect hasn't been called yet. We'll do the
598 * lov_connect_obd on this target when that fn first runs,
599 * because we don't know the connect flags yet.
606 rc = lov_connect_obd(obd, index, active, &lov->lov_ocd);
610 /* connect to administrative disabled ost */
616 if (lov->lov_cache) {
617 rc = obd_set_info_async(NULL, tgt->ltd_exp,
618 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
619 sizeof(struct cl_client_cache),
620 lov->lov_cache, NULL);
625 rc = lov_notify(obd, tgt->ltd_exp->exp_obd,
626 active ? OBD_NOTIFY_CONNECT : OBD_NOTIFY_INACTIVE,
631 CERROR("add failed (%d), deleting %s\n", rc,
632 obd_uuid2str(&tgt->ltd_uuid));
633 lov_del_target(obd, index, NULL, 0);
639 /* Schedule a target for deletion */
640 int lov_del_target(struct obd_device *obd, __u32 index,
641 struct obd_uuid *uuidp, int gen)
643 struct lov_obd *lov = &obd->u.lov;
644 int count = lov->desc.ld_tgt_count;
647 if (index >= count) {
648 CERROR("LOV target index %d >= number of LOV OBDs %d.\n",
653 /* to make sure there's no ongoing lov_notify() now */
654 down_write(&lov->lov_notify_lock);
657 if (!lov->lov_tgts[index]) {
658 CERROR("LOV target at index %d is not setup.\n", index);
663 if (uuidp && !obd_uuid_equals(uuidp, &lov->lov_tgts[index]->ltd_uuid)) {
664 CERROR("LOV target UUID %s at index %d doesn't match %s.\n",
665 lov_uuid2str(lov, index), index,
666 obd_uuid2str(uuidp));
671 CDEBUG(D_CONFIG, "uuid: %s idx: %d gen: %d exp: %p active: %d\n",
672 lov_uuid2str(lov, index), index,
673 lov->lov_tgts[index]->ltd_gen, lov->lov_tgts[index]->ltd_exp,
674 lov->lov_tgts[index]->ltd_active);
676 lov->lov_tgts[index]->ltd_reap = 1;
677 lov->lov_death_row++;
678 /* we really delete it from obd_putref */
681 up_write(&lov->lov_notify_lock);
686 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
688 struct obd_device *osc_obd;
691 LASSERT(tgt->ltd_reap);
693 osc_obd = class_exp2obd(tgt->ltd_exp);
695 CDEBUG(D_CONFIG, "Removing tgt %s : %s\n",
697 osc_obd ? osc_obd->obd_name : "<no obd>");
700 lov_disconnect_obd(obd, tgt);
704 /* Manual cleanup - no cleanup logs to clean up the osc's. We must
705 * do it ourselves. And we can't do it from lov_cleanup,
706 * because we just lost our only reference to it.
709 class_manual_cleanup(osc_obd);
712 void lov_fix_desc_stripe_size(__u64 *val)
714 if (*val < LOV_MIN_STRIPE_SIZE) {
716 LCONSOLE_INFO("Increasing default stripe size to minimum %u\n",
717 LOV_DESC_STRIPE_SIZE_DEFAULT);
718 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
719 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
720 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
721 LCONSOLE_WARN("Changing default stripe size to %llu (a multiple of %u)\n",
722 *val, LOV_MIN_STRIPE_SIZE);
726 void lov_fix_desc_stripe_count(__u32 *val)
732 void lov_fix_desc_pattern(__u32 *val)
734 /* from lov_setstripe */
735 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
736 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
741 void lov_fix_desc_qos_maxage(__u32 *val)
744 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
747 void lov_fix_desc(struct lov_desc *desc)
749 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
750 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
751 lov_fix_desc_pattern(&desc->ld_pattern);
752 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
755 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
757 struct lprocfs_static_vars lvars = { NULL };
758 struct lov_desc *desc;
759 struct lov_obd *lov = &obd->u.lov;
762 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
763 CERROR("LOV setup requires a descriptor\n");
767 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
769 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
770 CERROR("descriptor size wrong: %d > %d\n",
771 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
775 if (desc->ld_magic != LOV_DESC_MAGIC) {
776 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
777 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
778 obd->obd_name, desc);
779 lustre_swab_lov_desc(desc);
781 CERROR("%s: Bad lov desc magic: %#x\n",
782 obd->obd_name, desc->ld_magic);
789 desc->ld_active_tgt_count = 0;
791 lov->lov_tgt_size = 0;
793 mutex_init(&lov->lov_lock);
794 atomic_set(&lov->lov_refcount, 0);
795 lov->lov_sp_me = LUSTRE_SP_CLI;
797 init_rwsem(&lov->lov_notify_lock);
799 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
801 HASH_POOLS_BKT_BITS, 0,
804 &pool_hash_operations,
806 INIT_LIST_HEAD(&lov->lov_pool_list);
807 lov->lov_pool_count = 0;
808 rc = lov_ost_pool_init(&lov->lov_packed, 0);
812 lprocfs_lov_init_vars(&lvars);
813 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
815 rc = ldebugfs_seq_create(obd->obd_debugfs_entry, "target_obd",
816 0444, &lov_proc_target_fops, obd);
818 CWARN("Error adding the target_obd file\n");
820 lov->lov_pool_debugfs_entry = ldebugfs_register("pools",
821 obd->obd_debugfs_entry,
829 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
831 struct lov_obd *lov = &obd->u.lov;
834 case OBD_CLEANUP_EARLY: {
837 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
838 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
840 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
852 static int lov_cleanup(struct obd_device *obd)
854 struct lov_obd *lov = &obd->u.lov;
855 struct list_head *pos, *tmp;
856 struct pool_desc *pool;
858 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
859 pool = list_entry(pos, struct pool_desc, pool_list);
860 /* free pool structs */
861 CDEBUG(D_INFO, "delete pool %p\n", pool);
862 /* In the function below, .hs_keycmp resolves to
863 * pool_hashkey_keycmp()
865 /* coverity[overrun-buffer-val] */
866 lov_pool_del(obd, pool->pool_name);
868 cfs_hash_putref(lov->lov_pools_hash_body);
869 lov_ost_pool_free(&lov->lov_packed);
871 lprocfs_obd_cleanup(obd);
876 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
877 if (!lov->lov_tgts[i])
880 /* Inactive targets may never have connected */
881 if (lov->lov_tgts[i]->ltd_active ||
882 atomic_read(&lov->lov_refcount))
883 /* We should never get here - these
884 * should have been removed in the
887 CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n",
888 i, lov->lov_death_row,
889 atomic_read(&lov->lov_refcount));
890 lov_del_target(obd, i, NULL, 0);
893 kfree(lov->lov_tgts);
894 lov->lov_tgt_size = 0;
897 if (lov->lov_cache) {
898 cl_cache_decref(lov->lov_cache);
899 lov->lov_cache = NULL;
905 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
906 __u32 *indexp, int *genp)
908 struct obd_uuid obd_uuid;
912 switch (cmd = lcfg->lcfg_command) {
913 case LCFG_LOV_ADD_OBD:
914 case LCFG_LOV_ADD_INA:
915 case LCFG_LOV_DEL_OBD: {
918 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
919 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
924 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
926 rc = kstrtoint(lustre_cfg_buf(lcfg, 2), 10, indexp);
929 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
934 if (cmd == LCFG_LOV_ADD_OBD)
935 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
936 else if (cmd == LCFG_LOV_ADD_INA)
937 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
939 rc = lov_del_target(obd, index, &obd_uuid, gen);
943 struct lprocfs_static_vars lvars = { NULL };
944 struct lov_desc *desc = &obd->u.lov.desc;
951 lprocfs_lov_init_vars(&lvars);
953 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
966 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
975 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
976 struct lov_stripe_md **ea, struct obd_trans_info *oti)
978 struct lov_stripe_md *obj_mdp, *lsm;
979 struct lov_obd *lov = &exp->exp_obd->u.lov;
983 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
984 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
986 obj_mdp = kzalloc(sizeof(*obj_mdp), GFP_NOFS);
990 ost_idx = src_oa->o_nlink;
996 if (ost_idx >= lov->desc.ld_tgt_count ||
997 !lov->lov_tgts[ost_idx]) {
1002 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1003 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1005 if (lov_oinfo_is_dummy(loi))
1008 if (loi->loi_ost_idx == ost_idx) {
1009 if (ostid_id(&loi->loi_oi) != ostid_id(&src_oa->o_oi)) {
1016 if (i == lsm->lsm_stripe_count) {
1021 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1022 src_oa, &obj_mdp, oti);
1028 /* the LOV expects oa->o_id to be set to the LOV object id */
1029 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1030 struct obdo *src_oa, struct lov_stripe_md **ea,
1031 struct obd_trans_info *oti)
1033 struct lov_obd *lov;
1040 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1041 src_oa->o_flags == OBD_FL_DELORPHAN) {
1042 /* should be used with LOV anymore */
1046 lov = &exp->exp_obd->u.lov;
1047 if (!lov->desc.ld_active_tgt_count)
1050 obd_getref(exp->exp_obd);
1051 /* Recreate a specific object id at the given OST index */
1052 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1053 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1054 rc = lov_recreate(exp, src_oa, ea, oti);
1057 obd_putref(exp->exp_obd);
1061 #define ASSERT_LSM_MAGIC(lsmp) \
1064 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1065 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1066 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1069 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1070 struct obdo *oa, struct lov_stripe_md *lsm,
1071 struct obd_trans_info *oti, struct obd_export *md_exp)
1073 struct lov_request_set *set;
1074 struct obd_info oinfo;
1075 struct lov_request *req;
1076 struct lov_obd *lov;
1077 int rc = 0, err = 0;
1079 ASSERT_LSM_MAGIC(lsm);
1081 if (!exp || !exp->exp_obd)
1084 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1086 LASSERT(oti->oti_logcookies);
1089 lov = &exp->exp_obd->u.lov;
1090 obd_getref(exp->exp_obd);
1091 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1095 list_for_each_entry(req, &set->set_list, rq_link) {
1096 if (oa->o_valid & OBD_MD_FLCOOKIE)
1097 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1099 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1100 req->rq_oi.oi_oa, NULL, oti, NULL);
1101 err = lov_update_common_set(set, req, err);
1103 CERROR("%s: destroying objid "DOSTID" subobj "
1104 DOSTID" on OST idx %d: rc = %d\n",
1105 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1106 POSTID(&req->rq_oi.oi_oa->o_oi),
1114 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1116 err = lov_fini_destroy_set(set);
1118 obd_putref(exp->exp_obd);
1119 return rc ? rc : err;
1122 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1125 struct lov_request_set *lovset = (struct lov_request_set *)data;
1128 /* don't do attribute merge if this async op failed */
1130 atomic_set(&lovset->set_completes, 0);
1131 err = lov_fini_getattr_set(lovset);
1132 return rc ? rc : err;
1135 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1136 struct ptlrpc_request_set *rqset)
1138 struct lov_request_set *lovset;
1139 struct lov_obd *lov;
1140 struct lov_request *req;
1144 ASSERT_LSM_MAGIC(oinfo->oi_md);
1146 if (!exp || !exp->exp_obd)
1149 lov = &exp->exp_obd->u.lov;
1151 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1155 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1156 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1157 oinfo->oi_md->lsm_stripe_size);
1159 list_for_each_entry(req, &lovset->set_list, rq_link) {
1160 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1161 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1162 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1163 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1164 &req->rq_oi, rqset);
1166 CERROR("%s: getattr objid "DOSTID" subobj"
1167 DOSTID" on OST idx %d: rc = %d\n",
1168 exp->exp_obd->obd_name,
1169 POSTID(&oinfo->oi_oa->o_oi),
1170 POSTID(&req->rq_oi.oi_oa->o_oi),
1176 if (!list_empty(&rqset->set_requests)) {
1178 LASSERT(!rqset->set_interpret);
1179 rqset->set_interpret = lov_getattr_interpret;
1180 rqset->set_arg = (void *)lovset;
1185 atomic_set(&lovset->set_completes, 0);
1186 err = lov_fini_getattr_set(lovset);
1187 return rc ? rc : err;
1190 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1193 struct lov_request_set *lovset = (struct lov_request_set *)data;
1197 atomic_set(&lovset->set_completes, 0);
1198 err = lov_fini_setattr_set(lovset);
1199 return rc ? rc : err;
1202 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1203 * needed. Otherwise, a client is waiting for responses.
1205 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1206 struct obd_trans_info *oti,
1207 struct ptlrpc_request_set *rqset)
1209 struct lov_request_set *set;
1210 struct lov_request *req;
1211 struct lov_obd *lov;
1215 ASSERT_LSM_MAGIC(oinfo->oi_md);
1216 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1218 LASSERT(oti->oti_logcookies);
1221 if (!exp || !exp->exp_obd)
1224 lov = &exp->exp_obd->u.lov;
1225 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1229 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1230 POSTID(&oinfo->oi_md->lsm_oi),
1231 oinfo->oi_md->lsm_stripe_count,
1232 oinfo->oi_md->lsm_stripe_size);
1234 list_for_each_entry(req, &set->set_list, rq_link) {
1235 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1236 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1238 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1239 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1240 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1242 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1243 &req->rq_oi, oti, rqset);
1245 CERROR("error: setattr objid "DOSTID" subobj"
1246 DOSTID" on OST idx %d: rc = %d\n",
1247 POSTID(&set->set_oi->oi_oa->o_oi),
1248 POSTID(&req->rq_oi.oi_oa->o_oi),
1254 /* If we are not waiting for responses on async requests, return. */
1255 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1259 atomic_set(&set->set_completes, 0);
1260 err = lov_fini_setattr_set(set);
1261 return rc ? rc : err;
1264 LASSERT(!rqset->set_interpret);
1265 rqset->set_interpret = lov_setattr_interpret;
1266 rqset->set_arg = (void *)set;
1271 /* find any ldlm lock of the inode in lov
1276 static int lov_find_cbdata(struct obd_export *exp,
1277 struct lov_stripe_md *lsm, ldlm_iterator_t it,
1280 struct lov_obd *lov;
1283 ASSERT_LSM_MAGIC(lsm);
1285 if (!exp || !exp->exp_obd)
1288 lov = &exp->exp_obd->u.lov;
1289 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1290 struct lov_stripe_md submd;
1291 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
1293 if (lov_oinfo_is_dummy(loi))
1296 if (!lov->lov_tgts[loi->loi_ost_idx]) {
1297 CDEBUG(D_HA, "lov idx %d NULL\n", loi->loi_ost_idx);
1301 submd.lsm_oi = loi->loi_oi;
1302 submd.lsm_stripe_count = 0;
1303 rc = obd_find_cbdata(lov->lov_tgts[loi->loi_ost_idx]->ltd_exp,
1311 int lov_statfs_interpret(struct ptlrpc_request_set *rqset, void *data, int rc)
1313 struct lov_request_set *lovset = (struct lov_request_set *)data;
1317 atomic_set(&lovset->set_completes, 0);
1319 err = lov_fini_statfs_set(lovset);
1320 return rc ? rc : err;
1323 static int lov_statfs_async(struct obd_export *exp, struct obd_info *oinfo,
1324 __u64 max_age, struct ptlrpc_request_set *rqset)
1326 struct obd_device *obd = class_exp2obd(exp);
1327 struct lov_request_set *set;
1328 struct lov_request *req;
1329 struct lov_obd *lov;
1332 LASSERT(oinfo->oi_osfs);
1335 rc = lov_prep_statfs_set(obd, oinfo, &set);
1339 list_for_each_entry(req, &set->set_list, rq_link) {
1340 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1341 &req->rq_oi, max_age, rqset);
1346 if (rc || list_empty(&rqset->set_requests)) {
1350 atomic_set(&set->set_completes, 0);
1351 err = lov_fini_statfs_set(set);
1352 return rc ? rc : err;
1355 LASSERT(!rqset->set_interpret);
1356 rqset->set_interpret = lov_statfs_interpret;
1357 rqset->set_arg = (void *)set;
1361 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1362 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1364 struct ptlrpc_request_set *set = NULL;
1365 struct obd_info oinfo = { };
1368 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1371 set = ptlrpc_prep_set();
1375 oinfo.oi_osfs = osfs;
1376 oinfo.oi_flags = flags;
1377 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1379 rc = ptlrpc_set_wait(set);
1380 ptlrpc_set_destroy(set);
1385 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1386 void *karg, void __user *uarg)
1388 struct obd_device *obddev = class_exp2obd(exp);
1389 struct lov_obd *lov = &obddev->u.lov;
1390 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1391 struct obd_uuid *uuidp;
1394 case IOC_OBD_STATFS: {
1395 struct obd_ioctl_data *data = karg;
1396 struct obd_device *osc_obd;
1397 struct obd_statfs stat_buf = {0};
1401 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1405 if (!lov->lov_tgts[index])
1406 /* Try again with the next index */
1408 if (!lov->lov_tgts[index]->ltd_active)
1411 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1416 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1417 min((int)data->ioc_plen2,
1418 (int)sizeof(struct obd_uuid))))
1421 memcpy(&flags, data->ioc_inlbuf1, sizeof(__u32));
1422 flags = flags & LL_STATFS_NODELAY ? OBD_STATFS_NODELAY : 0;
1424 /* got statfs data */
1425 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1426 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1430 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1431 min((int)data->ioc_plen1,
1432 (int)sizeof(stat_buf))))
1436 case OBD_IOC_LOV_GET_CONFIG: {
1437 struct obd_ioctl_data *data;
1438 struct lov_desc *desc;
1443 if (obd_ioctl_getdata(&buf, &len, uarg))
1446 data = (struct obd_ioctl_data *)buf;
1448 if (sizeof(*desc) > data->ioc_inllen1) {
1449 obd_ioctl_freedata(buf, len);
1453 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1454 obd_ioctl_freedata(buf, len);
1458 if (sizeof(__u32) * count > data->ioc_inllen3) {
1459 obd_ioctl_freedata(buf, len);
1463 desc = (struct lov_desc *)data->ioc_inlbuf1;
1464 memcpy(desc, &lov->desc, sizeof(*desc));
1466 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1467 genp = (__u32 *)data->ioc_inlbuf3;
1468 /* the uuid will be empty for deleted OSTs */
1469 for (i = 0; i < count; i++, uuidp++, genp++) {
1470 if (!lov->lov_tgts[i])
1472 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1473 *genp = lov->lov_tgts[i]->ltd_gen;
1476 if (copy_to_user(uarg, buf, len))
1478 obd_ioctl_freedata(buf, len);
1481 case LL_IOC_LOV_GETSTRIPE:
1482 rc = lov_getstripe(exp, karg, uarg);
1484 case OBD_IOC_QUOTACTL: {
1485 struct if_quotactl *qctl = karg;
1486 struct lov_tgt_desc *tgt = NULL;
1487 struct obd_quotactl *oqctl;
1489 if (qctl->qc_valid == QC_OSTIDX) {
1490 if (count <= qctl->qc_idx)
1493 tgt = lov->lov_tgts[qctl->qc_idx];
1494 if (!tgt || !tgt->ltd_exp)
1496 } else if (qctl->qc_valid == QC_UUID) {
1497 for (i = 0; i < count; i++) {
1498 tgt = lov->lov_tgts[i];
1500 !obd_uuid_equals(&tgt->ltd_uuid,
1516 LASSERT(tgt && tgt->ltd_exp);
1517 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1521 QCTL_COPY(oqctl, qctl);
1522 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1524 QCTL_COPY(qctl, oqctl);
1525 qctl->qc_valid = QC_OSTIDX;
1526 qctl->obd_uuid = tgt->ltd_uuid;
1537 for (i = 0; i < count; i++) {
1539 struct obd_device *osc_obd;
1541 /* OST was disconnected */
1542 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1545 /* ll_umount_begin() sets force flag but for lov, not
1546 * osc. Let's pass it through
1548 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1549 osc_obd->obd_force = obddev->obd_force;
1550 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1552 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK)
1555 if (lov->lov_tgts[i]->ltd_active) {
1556 CDEBUG(err == -ENOTTY ?
1557 D_IOCTL : D_WARNING,
1558 "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n",
1559 lov_uuid2str(lov, i),
1576 #define FIEMAP_BUFFER_SIZE 4096
1579 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1580 * call. The local end offset and the device are sent in the first
1581 * fm_extent. This function calculates the stripe number from the index.
1582 * This function returns a stripe_no on which mapping is to be restarted.
1584 * This function returns fm_end_offset which is the in-OST offset at which
1585 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1586 * will re-calculate proper offset in next stripe.
1587 * Note that the first extent is passed to lov_get_info via the value field.
1589 * \param fiemap fiemap request header
1590 * \param lsm striping information for the file
1591 * \param fm_start logical start of mapping
1592 * \param fm_end logical end of mapping
1593 * \param start_stripe starting stripe will be returned in this
1595 static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1596 struct lov_stripe_md *lsm, u64 fm_start,
1597 u64 fm_end, int *start_stripe)
1599 u64 local_end = fiemap->fm_extents[0].fe_logical;
1600 u64 lun_start, lun_end;
1602 int stripe_no = -1, i;
1604 if (fiemap->fm_extent_count == 0 ||
1605 fiemap->fm_extents[0].fe_logical == 0)
1608 /* Find out stripe_no from ost_index saved in the fe_device */
1609 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1610 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1612 if (lov_oinfo_is_dummy(oinfo))
1615 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1620 if (stripe_no == -1)
1623 /* If we have finished mapping on previous device, shift logical
1624 * offset to start of next device
1626 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1627 &lun_start, &lun_end)) != 0 &&
1628 local_end < lun_end) {
1629 fm_end_offset = local_end;
1630 *start_stripe = stripe_no;
1632 /* This is a special value to indicate that caller should
1633 * calculate offset in next stripe.
1636 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1639 return fm_end_offset;
1643 * We calculate on which OST the mapping will end. If the length of mapping
1644 * is greater than (stripe_size * stripe_count) then the last_stripe will
1645 * will be one just before start_stripe. Else we check if the mapping
1646 * intersects each OST and find last_stripe.
1647 * This function returns the last_stripe and also sets the stripe_count
1648 * over which the mapping is spread
1650 * \param lsm striping information for the file
1651 * \param fm_start logical start of mapping
1652 * \param fm_end logical end of mapping
1653 * \param start_stripe starting stripe of the mapping
1654 * \param stripe_count the number of stripes across which to map is returned
1656 * \retval last_stripe return the last stripe of the mapping
1658 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start,
1659 u64 fm_end, int start_stripe,
1663 u64 obd_start, obd_end;
1666 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1667 last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1669 *stripe_count = lsm->lsm_stripe_count;
1671 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1672 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1673 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1674 &obd_start, &obd_end)) == 0)
1678 last_stripe = (start_stripe + j - 1) % lsm->lsm_stripe_count;
1685 * Set fe_device and copy extents from local buffer into main return buffer.
1687 * \param fiemap fiemap request header
1688 * \param lcl_fm_ext array of local fiemap extents to be copied
1689 * \param ost_index OST index to be written into the fm_device field for each
1691 * \param ext_count number of extents to be copied
1692 * \param current_extent where to start copying in main extent array
1694 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1695 struct ll_fiemap_extent *lcl_fm_ext,
1696 int ost_index, unsigned int ext_count,
1702 for (ext = 0; ext < ext_count; ext++) {
1703 lcl_fm_ext[ext].fe_device = ost_index;
1704 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1707 /* Copy fm_extent's from fm_local to return buffer */
1708 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1709 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1713 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1714 * This also handles the restarting of FIEMAP calls in case mapping overflows
1715 * the available number of extents in single call.
1717 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1718 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1720 struct ll_fiemap_info_key *fm_key = key;
1721 struct ll_user_fiemap *fiemap = val;
1722 struct ll_user_fiemap *fm_local = NULL;
1723 struct ll_fiemap_extent *lcl_fm_ext;
1725 unsigned int get_num_extents = 0;
1726 int ost_index = 0, actual_start_stripe, start_stripe;
1727 u64 fm_start, fm_end, fm_length, fm_end_offset;
1729 int current_extent = 0, rc = 0, i;
1730 int ost_eof = 0; /* EOF for object */
1731 int ost_done = 0; /* done with required mapping for this OST? */
1733 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1734 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1736 if (!lsm_has_objects(lsm)) {
1737 if (lsm && lsm_is_released(lsm) && (fm_key->fiemap.fm_start <
1738 fm_key->oa.o_size)) {
1740 * released file, return a minimal FIEMAP if
1741 * request fits in file-size.
1743 fiemap->fm_mapped_extents = 1;
1744 fiemap->fm_extents[0].fe_logical =
1745 fm_key->fiemap.fm_start;
1746 if (fm_key->fiemap.fm_start + fm_key->fiemap.fm_length <
1747 fm_key->oa.o_size) {
1748 fiemap->fm_extents[0].fe_length =
1749 fm_key->fiemap.fm_length;
1751 fiemap->fm_extents[0].fe_length =
1752 fm_key->oa.o_size - fm_key->fiemap.fm_start;
1753 fiemap->fm_extents[0].fe_flags |=
1754 (FIEMAP_EXTENT_UNKNOWN |
1755 FIEMAP_EXTENT_LAST);
1762 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1763 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1765 fm_local = libcfs_kvzalloc(buffer_size, GFP_NOFS);
1770 lcl_fm_ext = &fm_local->fm_extents[0];
1772 count_local = fiemap_size_to_count(buffer_size);
1774 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1775 fm_start = fiemap->fm_start;
1776 fm_length = fiemap->fm_length;
1777 /* Calculate start stripe, last stripe and length of mapping */
1778 start_stripe = lov_stripe_number(lsm, fm_start);
1779 actual_start_stripe = start_stripe;
1780 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1781 fm_start + fm_length - 1);
1782 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1783 if (fm_end > fm_key->oa.o_size)
1784 fm_end = fm_key->oa.o_size;
1786 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1787 actual_start_stripe,
1790 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1791 fm_end, &start_stripe);
1792 if (fm_end_offset == -EINVAL) {
1797 if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1798 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1799 if (fiemap->fm_extent_count == 0) {
1800 get_num_extents = 1;
1803 /* Check each stripe */
1804 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1805 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1806 u64 req_fm_len; /* Stores length of required mapping */
1807 u64 len_mapped_single_call;
1808 u64 lun_start, lun_end, obd_object_end;
1809 unsigned int ext_count;
1811 cur_stripe_wrap = cur_stripe;
1813 /* Find out range of mapping on this stripe */
1814 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1815 &lun_start, &obd_object_end)) == 0)
1818 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) {
1823 /* If this is a continuation FIEMAP call and we are on
1824 * starting stripe then lun_start needs to be set to
1827 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1828 lun_start = fm_end_offset;
1830 if (fm_length != ~0ULL) {
1831 /* Handle fm_start + fm_length overflow */
1832 if (fm_start + fm_length < fm_start)
1833 fm_length = ~0ULL - fm_start;
1834 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1840 if (lun_start == lun_end)
1843 req_fm_len = obd_object_end - lun_start;
1844 fm_local->fm_length = 0;
1845 len_mapped_single_call = 0;
1847 /* If the output buffer is very large and the objects have many
1848 * extents we may need to loop on a single OST repeatedly
1853 if (get_num_extents == 0) {
1854 /* Don't get too many extents. */
1855 if (current_extent + count_local >
1856 fiemap->fm_extent_count)
1857 count_local = fiemap->fm_extent_count -
1861 lun_start += len_mapped_single_call;
1862 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1863 req_fm_len = fm_local->fm_length;
1864 fm_local->fm_extent_count = count_local;
1865 fm_local->fm_mapped_extents = 0;
1866 fm_local->fm_flags = fiemap->fm_flags;
1868 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1869 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1871 if (ost_index < 0 ||
1872 ost_index >= lov->desc.ld_tgt_count) {
1877 /* If OST is inactive, return extent with UNKNOWN flag */
1878 if (!lov->lov_tgts[ost_index]->ltd_active) {
1879 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1880 fm_local->fm_mapped_extents = 1;
1882 lcl_fm_ext[0].fe_logical = lun_start;
1883 lcl_fm_ext[0].fe_length = obd_object_end -
1885 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1890 fm_local->fm_start = lun_start;
1891 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1892 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1893 *vallen = fiemap_count_to_size(fm_local->fm_extent_count);
1894 rc = obd_get_info(NULL,
1895 lov->lov_tgts[ost_index]->ltd_exp,
1896 keylen, key, vallen, fm_local, lsm);
1901 ext_count = fm_local->fm_mapped_extents;
1902 if (ext_count == 0) {
1904 /* If last stripe has hole at the end,
1905 * then we need to return
1907 if (cur_stripe_wrap == last_stripe) {
1908 fiemap->fm_mapped_extents = 0;
1914 /* If we just need num of extents then go to next device */
1915 if (get_num_extents) {
1916 current_extent += ext_count;
1920 len_mapped_single_call =
1921 lcl_fm_ext[ext_count - 1].fe_logical -
1922 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1924 /* Have we finished mapping on this device? */
1925 if (req_fm_len <= len_mapped_single_call)
1928 /* Clear the EXTENT_LAST flag which can be present on
1931 if (lcl_fm_ext[ext_count - 1].fe_flags &
1933 lcl_fm_ext[ext_count - 1].fe_flags &=
1934 ~FIEMAP_EXTENT_LAST;
1936 curr_loc = lov_stripe_size(lsm,
1937 lcl_fm_ext[ext_count - 1].fe_logical +
1938 lcl_fm_ext[ext_count - 1].fe_length,
1940 if (curr_loc >= fm_key->oa.o_size)
1943 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1944 ost_index, ext_count,
1947 current_extent += ext_count;
1949 /* Ran out of available extents? */
1950 if (current_extent >= fiemap->fm_extent_count)
1952 } while (ost_done == 0 && ost_eof == 0);
1954 if (cur_stripe_wrap == last_stripe)
1959 /* Indicate that we are returning device offsets unless file just has
1962 if (lsm->lsm_stripe_count > 1)
1963 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1965 if (get_num_extents)
1966 goto skip_last_device_calc;
1968 /* Check if we have reached the last stripe and whether mapping for that
1971 if (cur_stripe_wrap == last_stripe) {
1972 if (ost_done || ost_eof)
1973 fiemap->fm_extents[current_extent - 1].fe_flags |=
1977 skip_last_device_calc:
1978 fiemap->fm_mapped_extents = current_extent;
1985 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1986 __u32 keylen, void *key, __u32 *vallen, void *val,
1987 struct lov_stripe_md *lsm)
1989 struct obd_device *obddev = class_exp2obd(exp);
1990 struct lov_obd *lov = &obddev->u.lov;
1993 if (!vallen || !val)
1998 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
2001 struct ldlm_lock *lock;
2003 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
2004 struct lov_oinfo *loi;
2005 __u32 *stripe = val;
2007 if (*vallen < sizeof(*stripe)) {
2011 *vallen = sizeof(*stripe);
2013 /* XXX This is another one of those bits that will need to
2014 * change if we ever actually support nested LOVs. It uses
2015 * the lock's export to find out which stripe it is.
2017 /* XXX - it's assumed all the locks for deleted OSTs have
2018 * been cancelled. Also, the export for deleted OSTs will
2019 * be NULL and won't match the lock's export.
2021 for (i = 0; i < lsm->lsm_stripe_count; i++) {
2022 loi = lsm->lsm_oinfo[i];
2023 if (lov_oinfo_is_dummy(loi))
2026 if (!lov->lov_tgts[loi->loi_ost_idx])
2028 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
2029 data->lock->l_conn_export &&
2030 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2036 LDLM_ERROR(data->lock, "lock on inode without such object");
2037 dump_lsm(D_ERROR, lsm);
2040 } else if (KEY_IS(KEY_LAST_ID)) {
2041 struct obd_id_info *info = val;
2042 __u32 size = sizeof(u64);
2043 struct lov_tgt_desc *tgt;
2045 LASSERT(*vallen == sizeof(struct obd_id_info));
2046 tgt = lov->lov_tgts[info->idx];
2048 if (!tgt || !tgt->ltd_active) {
2053 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2054 &size, info->data, NULL);
2057 } else if (KEY_IS(KEY_LOVDESC)) {
2058 struct lov_desc *desc_ret = val;
2059 *desc_ret = lov->desc;
2063 } else if (KEY_IS(KEY_FIEMAP)) {
2064 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2066 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2067 struct lov_tgt_desc *tgt;
2068 __u64 ost_idx = *((__u64 *)val);
2070 LASSERT(*vallen == sizeof(__u64));
2071 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2072 tgt = lov->lov_tgts[ost_idx];
2074 if (!tgt || !tgt->ltd_exp) {
2079 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2082 } else if (KEY_IS(KEY_TGT_COUNT)) {
2083 *((int *)val) = lov->desc.ld_tgt_count;
2095 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2096 u32 keylen, void *key, u32 vallen,
2097 void *val, struct ptlrpc_request_set *set)
2099 struct obd_device *obddev = class_exp2obd(exp);
2100 struct lov_obd *lov = &obddev->u.lov;
2103 struct lov_tgt_desc *tgt;
2104 unsigned int incr = 0, check_uuid = 0, do_inactive = 0, no_set = 0;
2105 unsigned int next_id = 0, mds_con = 0;
2109 set = ptlrpc_prep_set();
2115 count = lov->desc.ld_tgt_count;
2117 if (KEY_IS(KEY_NEXT_ID)) {
2118 count = vallen / sizeof(struct obd_id_info);
2119 vallen = sizeof(u64);
2120 incr = sizeof(struct obd_id_info);
2123 } else if (KEY_IS(KEY_CHECKSUM)) {
2125 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2126 /* use defaults: do_inactive = incr = 0; */
2127 } else if (KEY_IS(KEY_MDS_CONN)) {
2129 } else if (KEY_IS(KEY_CACHE_SET)) {
2130 LASSERT(!lov->lov_cache);
2131 lov->lov_cache = val;
2133 cl_cache_incref(lov->lov_cache);
2136 for (i = 0; i < count; i++, val = (char *)val + incr) {
2138 tgt = lov->lov_tgts[((struct obd_id_info *)val)->idx];
2140 tgt = lov->lov_tgts[i];
2141 /* OST was disconnected */
2142 if (!tgt || !tgt->ltd_exp)
2145 /* OST is inactive and we don't want inactive OSCs */
2146 if (!tgt->ltd_active && !do_inactive)
2150 struct mds_group_info *mgi;
2152 LASSERT(vallen == sizeof(*mgi));
2153 mgi = (struct mds_group_info *)val;
2155 /* Only want a specific OSC */
2156 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2160 err = obd_set_info_async(env, tgt->ltd_exp,
2161 keylen, key, sizeof(int),
2163 } else if (next_id) {
2164 err = obd_set_info_async(env, tgt->ltd_exp,
2165 keylen, key, vallen,
2166 ((struct obd_id_info *)val)->data, set);
2168 /* Only want a specific OSC */
2170 !obd_uuid_equals(val, &tgt->ltd_uuid))
2173 err = obd_set_info_async(env, tgt->ltd_exp,
2174 keylen, key, vallen, val, set);
2183 err = ptlrpc_set_wait(set);
2186 ptlrpc_set_destroy(set);
2191 void lov_stripe_lock(struct lov_stripe_md *md)
2192 __acquires(&md->lsm_lock)
2194 LASSERT(md->lsm_lock_owner != current_pid());
2195 spin_lock(&md->lsm_lock);
2196 LASSERT(md->lsm_lock_owner == 0);
2197 md->lsm_lock_owner = current_pid();
2200 void lov_stripe_unlock(struct lov_stripe_md *md)
2201 __releases(&md->lsm_lock)
2203 LASSERT(md->lsm_lock_owner == current_pid());
2204 md->lsm_lock_owner = 0;
2205 spin_unlock(&md->lsm_lock);
2208 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2209 struct obd_quotactl *oqctl)
2211 struct lov_obd *lov = &obd->u.lov;
2212 struct lov_tgt_desc *tgt;
2214 __u64 bhardlimit = 0;
2217 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2218 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2219 oqctl->qc_cmd != Q_GETOQUOTA &&
2220 oqctl->qc_cmd != Q_INITQUOTA &&
2221 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2222 oqctl->qc_cmd != Q_FINVALIDATE) {
2223 CERROR("bad quota opc %x for lov obd\n", oqctl->qc_cmd);
2229 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2232 tgt = lov->lov_tgts[i];
2237 if (!tgt->ltd_active || tgt->ltd_reap) {
2238 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2239 lov->lov_tgts[i]->ltd_activate) {
2241 CERROR("ost %d is inactive\n", i);
2243 CDEBUG(D_HA, "ost %d is inactive\n", i);
2248 err = obd_quotactl(tgt->ltd_exp, oqctl);
2250 if (tgt->ltd_active && !rc)
2255 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2256 curspace += oqctl->qc_dqblk.dqb_curspace;
2257 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2262 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2263 oqctl->qc_dqblk.dqb_curspace = curspace;
2264 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2269 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2270 struct obd_quotactl *oqctl)
2272 struct lov_obd *lov = &obd->u.lov;
2277 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2278 if (!lov->lov_tgts[i])
2281 /* Skip quota check on the administratively disabled OSTs. */
2282 if (!lov->lov_tgts[i]->ltd_activate) {
2283 CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n",
2288 if (!lov->lov_tgts[i]->ltd_active) {
2289 CERROR("lov idx %d inactive\n", i);
2295 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2298 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2301 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2312 static struct obd_ops lov_obd_ops = {
2313 .owner = THIS_MODULE,
2315 .precleanup = lov_precleanup,
2316 .cleanup = lov_cleanup,
2317 /*.process_config = lov_process_config,*/
2318 .connect = lov_connect,
2319 .disconnect = lov_disconnect,
2320 .statfs = lov_statfs,
2321 .statfs_async = lov_statfs_async,
2322 .packmd = lov_packmd,
2323 .unpackmd = lov_unpackmd,
2324 .create = lov_create,
2325 .destroy = lov_destroy,
2326 .getattr_async = lov_getattr_async,
2327 .setattr_async = lov_setattr_async,
2328 .adjust_kms = lov_adjust_kms,
2329 .find_cbdata = lov_find_cbdata,
2330 .iocontrol = lov_iocontrol,
2331 .get_info = lov_get_info,
2332 .set_info_async = lov_set_info_async,
2333 .notify = lov_notify,
2334 .pool_new = lov_pool_new,
2335 .pool_rem = lov_pool_remove,
2336 .pool_add = lov_pool_add,
2337 .pool_del = lov_pool_del,
2338 .getref = lov_getref,
2339 .putref = lov_putref,
2340 .quotactl = lov_quotactl,
2341 .quotacheck = lov_quotacheck,
2344 struct kmem_cache *lov_oinfo_slab;
2346 static int __init lov_init(void)
2348 struct lprocfs_static_vars lvars = { NULL };
2351 /* print an address of _any_ initialized kernel symbol from this
2352 * module, to allow debugging with gdb that doesn't support data
2353 * symbols from modules.
2355 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2357 rc = lu_kmem_init(lov_caches);
2361 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2362 sizeof(struct lov_oinfo),
2363 0, SLAB_HWCACHE_ALIGN, NULL);
2364 if (!lov_oinfo_slab) {
2365 lu_kmem_fini(lov_caches);
2368 lprocfs_lov_init_vars(&lvars);
2370 rc = class_register_type(&lov_obd_ops, NULL,
2371 LUSTRE_LOV_NAME, &lov_device_type);
2374 kmem_cache_destroy(lov_oinfo_slab);
2375 lu_kmem_fini(lov_caches);
2381 static void /*__exit*/ lov_exit(void)
2383 class_unregister_type(LUSTRE_LOV_NAME);
2384 kmem_cache_destroy(lov_oinfo_slab);
2386 lu_kmem_fini(lov_caches);
2389 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
2390 MODULE_DESCRIPTION("Lustre Logical Object Volume");
2391 MODULE_LICENSE("GPL");
2392 MODULE_VERSION(LUSTRE_VERSION_STRING);
2394 module_init(lov_init);
2395 module_exit(lov_exit);