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, 2012, 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 "../../include/linux/libcfs/libcfs.h"
47 #include "../include/obd_support.h"
48 #include "../include/lustre_lib.h"
49 #include "../include/lustre_net.h"
50 #include "../include/lustre/lustre_idl.h"
51 #include "../include/lustre_dlm.h"
52 #include "../include/lustre_mds.h"
53 #include "../include/obd_class.h"
54 #include "../include/lprocfs_status.h"
55 #include "../include/lustre_param.h"
56 #include "../include/cl_object.h"
57 #include "../include/lclient.h" /* for cl_client_lru */
58 #include "../include/lustre/ll_fiemap.h"
59 #include "../include/lustre_fid.h"
61 #include "lov_internal.h"
63 /* Keep a refcount of lov->tgt usage to prevent racing with addition/deletion.
64 Any function that expects lov_tgts to remain stationary must take a ref. */
65 static void lov_getref(struct obd_device *obd)
67 struct lov_obd *lov = &obd->u.lov;
69 /* nobody gets through here until lov_putref is done */
70 mutex_lock(&lov->lov_lock);
71 atomic_inc(&lov->lov_refcount);
72 mutex_unlock(&lov->lov_lock);
76 static void __lov_del_obd(struct obd_device *obd, struct lov_tgt_desc *tgt);
78 static void lov_putref(struct obd_device *obd)
80 struct lov_obd *lov = &obd->u.lov;
82 mutex_lock(&lov->lov_lock);
83 /* ok to dec to 0 more than once -- ltd_exp's will be null */
84 if (atomic_dec_and_test(&lov->lov_refcount) && lov->lov_death_row) {
87 struct lov_tgt_desc *tgt, *n;
88 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
90 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
91 tgt = lov->lov_tgts[i];
93 if (!tgt || !tgt->ltd_reap)
95 list_add(&tgt->ltd_kill, &kill);
96 /* XXX - right now there is a dependency on ld_tgt_count
97 * being the maximum tgt index for computing the
98 * mds_max_easize. So we can't shrink it. */
99 lov_ost_pool_remove(&lov->lov_packed, i);
100 lov->lov_tgts[i] = NULL;
101 lov->lov_death_row--;
103 mutex_unlock(&lov->lov_lock);
105 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
106 list_del(&tgt->ltd_kill);
108 __lov_del_obd(obd, tgt);
111 mutex_unlock(&lov->lov_lock);
115 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
116 enum obd_notify_event ev);
117 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
118 enum obd_notify_event ev, void *data);
121 #define MAX_STRING_SIZE 128
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;
132 if (!lov->lov_tgts[index])
135 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
136 tgt_obd = lov->lov_tgts[index]->ltd_obd;
138 if (!tgt_obd->obd_set_up) {
139 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
143 /* override the sp_me from lov */
144 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
146 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
147 data->ocd_index = index;
150 * Divine LOV knows that OBDs under it are OSCs.
152 imp = tgt_obd->u.cli.cl_import;
155 tgt_obd->obd_no_recov = 0;
156 /* FIXME this is probably supposed to be
157 ptlrpc_set_import_active. Horrible naming. */
158 ptlrpc_activate_import(imp);
161 rc = obd_register_observer(tgt_obd, obd);
163 CERROR("Target %s register_observer error %d\n",
164 obd_uuid2str(tgt_uuid), rc);
169 if (imp->imp_invalid) {
170 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n",
171 obd_uuid2str(tgt_uuid));
175 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
176 &lov_osc_uuid, data, NULL);
177 if (rc || !lov->lov_tgts[index]->ltd_exp) {
178 CERROR("Target %s connect error %d\n",
179 obd_uuid2str(tgt_uuid), rc);
183 lov->lov_tgts[index]->ltd_reap = 0;
185 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
186 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
188 if (lov->lov_tgts_kobj)
189 /* Even if we failed, that's ok */
190 rc = sysfs_create_link(lov->lov_tgts_kobj, &tgt_obd->obd_kobj,
196 static int lov_connect(const struct lu_env *env,
197 struct obd_export **exp, struct obd_device *obd,
198 struct obd_uuid *cluuid, struct obd_connect_data *data,
201 struct lov_obd *lov = &obd->u.lov;
202 struct lov_tgt_desc *tgt;
203 struct lustre_handle conn;
206 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
208 rc = class_connect(&conn, obd, cluuid);
212 *exp = class_conn2export(&conn);
214 /* Why should there ever be more than 1 connect? */
216 LASSERT(lov->lov_connects == 1);
218 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
220 lov->lov_ocd = *data;
224 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
227 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
228 tgt = lov->lov_tgts[i];
229 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
231 /* Flags will be lowest common denominator */
232 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
234 CERROR("%s: lov connect tgt %d failed: %d\n",
235 obd->obd_name, i, rc);
238 /* connect to administrative disabled ost */
239 if (!lov->lov_tgts[i]->ltd_exp)
242 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
243 OBD_NOTIFY_CONNECT, (void *)&i);
245 CERROR("%s error sending notify %d\n",
254 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
256 struct lov_obd *lov = &obd->u.lov;
257 struct obd_device *osc_obd;
260 osc_obd = class_exp2obd(tgt->ltd_exp);
261 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
262 obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL");
264 if (tgt->ltd_active) {
266 lov->desc.ld_active_tgt_count--;
267 tgt->ltd_exp->exp_obd->obd_inactive = 1;
271 if (lov->lov_tgts_kobj)
272 sysfs_remove_link(lov->lov_tgts_kobj,
275 /* Pass it on to our clients.
276 * XXX This should be an argument to disconnect,
277 * XXX not a back-door flag on the OBD. Ah well.
279 osc_obd->obd_force = obd->obd_force;
280 osc_obd->obd_fail = obd->obd_fail;
281 osc_obd->obd_no_recov = obd->obd_no_recov;
284 obd_register_observer(osc_obd, NULL);
286 rc = obd_disconnect(tgt->ltd_exp);
288 CERROR("Target %s disconnect error %d\n",
289 tgt->ltd_uuid.uuid, rc);
297 static int lov_disconnect(struct obd_export *exp)
299 struct obd_device *obd = class_exp2obd(exp);
300 struct lov_obd *lov = &obd->u.lov;
306 /* Only disconnect the underlying layers on the final disconnect. */
308 if (lov->lov_connects != 0) {
309 /* why should there be more than 1 connect? */
310 CERROR("disconnect #%d\n", lov->lov_connects);
314 /* Let's hold another reference so lov_del_obd doesn't spin through
318 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
319 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
320 /* Disconnection is the last we know about an obd */
321 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
325 if (lov->lov_tgts_kobj)
326 kobject_put(lov->lov_tgts_kobj);
331 rc = class_disconnect(exp); /* bz 9811 */
337 * -EINVAL : UUID can't be found in the LOV's target list
338 * -ENOTCONN: The UUID is found, but the target connection is bad (!)
339 * -EBADF : The UUID is found, but the OBD is the wrong type (!)
340 * any >= 0 : is log target index
342 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
343 enum obd_notify_event ev)
345 struct lov_obd *lov = &obd->u.lov;
346 struct lov_tgt_desc *tgt;
347 int index, activate, active;
349 CDEBUG(D_INFO, "Searching in lov %p for uuid %s event(%d)\n",
350 lov, uuid->uuid, ev);
353 for (index = 0; index < lov->desc.ld_tgt_count; index++) {
354 tgt = lov->lov_tgts[index];
358 * LU-642, initially inactive OSC could miss the obd_connect,
359 * we make up for it here.
361 if (ev == OBD_NOTIFY_ACTIVATE && tgt->ltd_exp == NULL &&
362 obd_uuid_equals(uuid, &tgt->ltd_uuid)) {
363 struct obd_uuid lov_osc_uuid = {"LOV_OSC_UUID"};
365 obd_connect(NULL, &tgt->ltd_exp, tgt->ltd_obd,
366 &lov_osc_uuid, &lov->lov_ocd, NULL);
371 CDEBUG(D_INFO, "lov idx %d is %s conn %#llx\n",
372 index, obd_uuid2str(&tgt->ltd_uuid),
373 tgt->ltd_exp->exp_handle.h_cookie);
374 if (obd_uuid_equals(uuid, &tgt->ltd_uuid))
378 if (index == lov->desc.ld_tgt_count) {
383 if (ev == OBD_NOTIFY_DEACTIVATE || ev == OBD_NOTIFY_ACTIVATE) {
384 activate = (ev == OBD_NOTIFY_ACTIVATE) ? 1 : 0;
386 if (lov->lov_tgts[index]->ltd_activate == activate) {
387 CDEBUG(D_INFO, "OSC %s already %sactivate!\n",
388 uuid->uuid, activate ? "" : "de");
390 lov->lov_tgts[index]->ltd_activate = activate;
391 CDEBUG(D_CONFIG, "%sactivate OSC %s\n",
392 activate ? "" : "de", obd_uuid2str(uuid));
395 } else if (ev == OBD_NOTIFY_INACTIVE || ev == OBD_NOTIFY_ACTIVE) {
396 active = (ev == OBD_NOTIFY_ACTIVE) ? 1 : 0;
398 if (lov->lov_tgts[index]->ltd_active == active) {
399 CDEBUG(D_INFO, "OSC %s already %sactive!\n",
400 uuid->uuid, active ? "" : "in");
403 CDEBUG(D_CONFIG, "Marking OSC %s %sactive\n",
404 obd_uuid2str(uuid), active ? "" : "in");
407 lov->lov_tgts[index]->ltd_active = active;
409 lov->desc.ld_active_tgt_count++;
410 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 0;
412 lov->desc.ld_active_tgt_count--;
413 lov->lov_tgts[index]->ltd_exp->exp_obd->obd_inactive = 1;
416 CERROR("Unknown event(%d) for uuid %s", ev, uuid->uuid);
424 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
425 enum obd_notify_event ev, void *data)
428 struct lov_obd *lov = &obd->u.lov;
430 down_read(&lov->lov_notify_lock);
431 if (!lov->lov_connects) {
432 up_read(&lov->lov_notify_lock);
436 if (ev == OBD_NOTIFY_ACTIVE || ev == OBD_NOTIFY_INACTIVE ||
437 ev == OBD_NOTIFY_ACTIVATE || ev == OBD_NOTIFY_DEACTIVATE) {
438 struct obd_uuid *uuid;
442 if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) {
443 up_read(&lov->lov_notify_lock);
444 CERROR("unexpected notification of %s %s!\n",
445 watched->obd_type->typ_name,
449 uuid = &watched->u.cli.cl_target_uuid;
451 /* Set OSC as active before notifying the observer, so the
452 * observer can use the OSC normally.
454 rc = lov_set_osc_active(obd, uuid, ev);
456 up_read(&lov->lov_notify_lock);
457 CERROR("event(%d) of %s failed: %d\n", ev,
458 obd_uuid2str(uuid), rc);
461 /* active event should be pass lov target index as data */
465 /* Pass the notification up the chain. */
467 rc = obd_notify_observer(obd, watched, ev, data);
469 /* NULL watched means all osc's in the lov (only for syncs) */
470 /* sync event should be send lov idx as data */
471 struct lov_obd *lov = &obd->u.lov;
475 is_sync = (ev == OBD_NOTIFY_SYNC) ||
476 (ev == OBD_NOTIFY_SYNC_NONBLOCK);
479 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
480 if (!lov->lov_tgts[i])
483 /* don't send sync event if target not
484 * connected/activated */
485 if (is_sync && !lov->lov_tgts[i]->ltd_active)
488 rc = obd_notify_observer(obd, lov->lov_tgts[i]->ltd_obd,
491 CERROR("%s: notify %s of %s failed %d\n",
493 obd->obd_observer->obd_name,
494 lov->lov_tgts[i]->ltd_obd->obd_name,
501 up_read(&lov->lov_notify_lock);
505 static int lov_add_target(struct obd_device *obd, struct obd_uuid *uuidp,
506 __u32 index, int gen, int active)
508 struct lov_obd *lov = &obd->u.lov;
509 struct lov_tgt_desc *tgt;
510 struct obd_device *tgt_obd;
513 CDEBUG(D_CONFIG, "uuid:%s idx:%d gen:%d active:%d\n",
514 uuidp->uuid, index, gen, active);
517 CERROR("request to add OBD %s with invalid generation: %d\n",
522 tgt_obd = class_find_client_obd(uuidp, LUSTRE_OSC_NAME,
527 mutex_lock(&lov->lov_lock);
529 if ((index < lov->lov_tgt_size) && (lov->lov_tgts[index] != NULL)) {
530 tgt = lov->lov_tgts[index];
531 CERROR("UUID %s already assigned at LOV target index %d\n",
532 obd_uuid2str(&tgt->ltd_uuid), index);
533 mutex_unlock(&lov->lov_lock);
537 if (index >= lov->lov_tgt_size) {
538 /* We need to reallocate the lov target array. */
539 struct lov_tgt_desc **newtgts, **old = NULL;
540 __u32 newsize, oldsize = 0;
542 newsize = max_t(__u32, lov->lov_tgt_size, 2);
543 while (newsize < index + 1)
545 newtgts = kcalloc(newsize, sizeof(*newtgts), GFP_NOFS);
546 if (newtgts == NULL) {
547 mutex_unlock(&lov->lov_lock);
551 if (lov->lov_tgt_size) {
552 memcpy(newtgts, lov->lov_tgts, sizeof(*newtgts) *
555 oldsize = lov->lov_tgt_size;
558 lov->lov_tgts = newtgts;
559 lov->lov_tgt_size = newsize;
563 CDEBUG(D_CONFIG, "tgts: %p size: %d\n",
564 lov->lov_tgts, lov->lov_tgt_size);
567 tgt = kzalloc(sizeof(*tgt), GFP_NOFS);
569 mutex_unlock(&lov->lov_lock);
573 rc = lov_ost_pool_add(&lov->lov_packed, index, lov->lov_tgt_size);
575 mutex_unlock(&lov->lov_lock);
580 tgt->ltd_uuid = *uuidp;
581 tgt->ltd_obd = tgt_obd;
582 /* XXX - add a sanity check on the generation number. */
584 tgt->ltd_index = index;
585 tgt->ltd_activate = active;
586 lov->lov_tgts[index] = tgt;
587 if (index >= lov->desc.ld_tgt_count)
588 lov->desc.ld_tgt_count = index + 1;
590 mutex_unlock(&lov->lov_lock);
592 CDEBUG(D_CONFIG, "idx=%d ltd_gen=%d ld_tgt_count=%d\n",
593 index, tgt->ltd_gen, lov->desc.ld_tgt_count);
595 rc = obd_notify(obd, tgt_obd, OBD_NOTIFY_CREATE, &index);
597 if (lov->lov_connects == 0) {
598 /* lov_connect hasn't been called yet. We'll do the
599 lov_connect_obd on this target when that fn first runs,
600 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 != NULL) {
617 rc = obd_set_info_async(NULL, tgt->ltd_exp,
618 sizeof(KEY_CACHE_SET), KEY_CACHE_SET,
619 sizeof(struct cl_client_cache), lov->lov_cache,
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. */
708 class_manual_cleanup(osc_obd);
711 void lov_fix_desc_stripe_size(__u64 *val)
713 if (*val < LOV_MIN_STRIPE_SIZE) {
715 LCONSOLE_INFO("Increasing default stripe size to minimum %u\n",
716 LOV_DESC_STRIPE_SIZE_DEFAULT);
717 *val = LOV_DESC_STRIPE_SIZE_DEFAULT;
718 } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
719 *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
720 LCONSOLE_WARN("Changing default stripe size to %llu (a multiple of %u)\n",
721 *val, LOV_MIN_STRIPE_SIZE);
725 void lov_fix_desc_stripe_count(__u32 *val)
731 void lov_fix_desc_pattern(__u32 *val)
733 /* from lov_setstripe */
734 if ((*val != 0) && (*val != LOV_PATTERN_RAID0)) {
735 LCONSOLE_WARN("Unknown stripe pattern: %#x\n", *val);
740 void lov_fix_desc_qos_maxage(__u32 *val)
743 *val = LOV_DESC_QOS_MAXAGE_DEFAULT;
746 void lov_fix_desc(struct lov_desc *desc)
748 lov_fix_desc_stripe_size(&desc->ld_default_stripe_size);
749 lov_fix_desc_stripe_count(&desc->ld_default_stripe_count);
750 lov_fix_desc_pattern(&desc->ld_pattern);
751 lov_fix_desc_qos_maxage(&desc->ld_qos_maxage);
754 int lov_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
756 struct lprocfs_static_vars lvars = { NULL };
757 struct lov_desc *desc;
758 struct lov_obd *lov = &obd->u.lov;
761 if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) {
762 CERROR("LOV setup requires a descriptor\n");
766 desc = (struct lov_desc *)lustre_cfg_buf(lcfg, 1);
768 if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) {
769 CERROR("descriptor size wrong: %d > %d\n",
770 (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1));
774 if (desc->ld_magic != LOV_DESC_MAGIC) {
775 if (desc->ld_magic == __swab32(LOV_DESC_MAGIC)) {
776 CDEBUG(D_OTHER, "%s: Swabbing lov desc %p\n",
777 obd->obd_name, desc);
778 lustre_swab_lov_desc(desc);
780 CERROR("%s: Bad lov desc magic: %#x\n",
781 obd->obd_name, desc->ld_magic);
788 desc->ld_active_tgt_count = 0;
790 lov->lov_tgt_size = 0;
792 mutex_init(&lov->lov_lock);
793 atomic_set(&lov->lov_refcount, 0);
794 lov->lov_sp_me = LUSTRE_SP_CLI;
796 init_rwsem(&lov->lov_notify_lock);
798 lov->lov_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
800 HASH_POOLS_BKT_BITS, 0,
803 &pool_hash_operations,
805 INIT_LIST_HEAD(&lov->lov_pool_list);
806 lov->lov_pool_count = 0;
807 rc = lov_ost_pool_init(&lov->lov_packed, 0);
811 lprocfs_lov_init_vars(&lvars);
812 lprocfs_obd_setup(obd, lvars.obd_vars, lvars.sysfs_vars);
814 rc = ldebugfs_seq_create(obd->obd_debugfs_entry, "target_obd",
815 0444, &lov_proc_target_fops, obd);
817 CWARN("Error adding the target_obd file\n");
819 lov->lov_pool_debugfs_entry = ldebugfs_register("pools",
820 obd->obd_debugfs_entry,
828 static int lov_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
830 struct lov_obd *lov = &obd->u.lov;
833 case OBD_CLEANUP_EARLY: {
835 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
836 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
838 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
850 static int lov_cleanup(struct obd_device *obd)
852 struct lov_obd *lov = &obd->u.lov;
853 struct list_head *pos, *tmp;
854 struct pool_desc *pool;
856 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
857 pool = list_entry(pos, struct pool_desc, pool_list);
858 /* free pool structs */
859 CDEBUG(D_INFO, "delete pool %p\n", pool);
860 /* In the function below, .hs_keycmp resolves to
861 * pool_hashkey_keycmp() */
862 /* coverity[overrun-buffer-val] */
863 lov_pool_del(obd, pool->pool_name);
865 cfs_hash_putref(lov->lov_pools_hash_body);
866 lov_ost_pool_free(&lov->lov_packed);
868 lprocfs_obd_cleanup(obd);
872 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
873 if (!lov->lov_tgts[i])
876 /* Inactive targets may never have connected */
877 if (lov->lov_tgts[i]->ltd_active ||
878 atomic_read(&lov->lov_refcount))
879 /* We should never get here - these
880 should have been removed in the
882 CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n",
883 i, lov->lov_death_row,
884 atomic_read(&lov->lov_refcount));
885 lov_del_target(obd, i, NULL, 0);
888 kfree(lov->lov_tgts);
889 lov->lov_tgt_size = 0;
894 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
895 __u32 *indexp, int *genp)
897 struct obd_uuid obd_uuid;
901 switch (cmd = lcfg->lcfg_command) {
902 case LCFG_LOV_ADD_OBD:
903 case LCFG_LOV_ADD_INA:
904 case LCFG_LOV_DEL_OBD: {
907 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
908 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
913 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
915 if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", indexp) != 1) {
919 if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", genp) != 1) {
925 if (cmd == LCFG_LOV_ADD_OBD)
926 rc = lov_add_target(obd, &obd_uuid, index, gen, 1);
927 else if (cmd == LCFG_LOV_ADD_INA)
928 rc = lov_add_target(obd, &obd_uuid, index, gen, 0);
930 rc = lov_del_target(obd, index, &obd_uuid, gen);
934 struct lprocfs_static_vars lvars = { NULL };
935 struct lov_desc *desc = &(obd->u.lov.desc);
942 lprocfs_lov_init_vars(&lvars);
944 rc = class_process_proc_param(PARAM_LOV, lvars.obd_vars,
957 CERROR("Unknown command: %d\n", lcfg->lcfg_command);
967 static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
968 struct lov_stripe_md **ea, struct obd_trans_info *oti)
970 struct lov_stripe_md *obj_mdp, *lsm;
971 struct lov_obd *lov = &exp->exp_obd->u.lov;
975 LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
976 src_oa->o_flags & OBD_FL_RECREATE_OBJS);
978 obj_mdp = kzalloc(sizeof(*obj_mdp), GFP_NOFS);
982 ost_idx = src_oa->o_nlink;
988 if (ost_idx >= lov->desc.ld_tgt_count ||
989 !lov->lov_tgts[ost_idx]) {
994 for (i = 0; i < lsm->lsm_stripe_count; i++) {
995 struct lov_oinfo *loi = lsm->lsm_oinfo[i];
997 if (lov_oinfo_is_dummy(loi))
1000 if (loi->loi_ost_idx == ost_idx) {
1001 if (ostid_id(&loi->loi_oi) != ostid_id(&src_oa->o_oi)) {
1008 if (i == lsm->lsm_stripe_count) {
1013 rc = obd_create(NULL, lov->lov_tgts[ost_idx]->ltd_exp,
1014 src_oa, &obj_mdp, oti);
1020 /* the LOV expects oa->o_id to be set to the LOV object id */
1021 static int lov_create(const struct lu_env *env, struct obd_export *exp,
1022 struct obdo *src_oa, struct lov_stripe_md **ea,
1023 struct obd_trans_info *oti)
1025 struct lov_obd *lov;
1028 LASSERT(ea != NULL);
1032 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1033 src_oa->o_flags == OBD_FL_DELORPHAN) {
1034 /* should be used with LOV anymore */
1038 lov = &exp->exp_obd->u.lov;
1039 if (!lov->desc.ld_active_tgt_count)
1042 obd_getref(exp->exp_obd);
1043 /* Recreate a specific object id at the given OST index */
1044 if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
1045 (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
1046 rc = lov_recreate(exp, src_oa, ea, oti);
1049 obd_putref(exp->exp_obd);
1053 #define ASSERT_LSM_MAGIC(lsmp) \
1055 LASSERT((lsmp) != NULL); \
1056 LASSERTF(((lsmp)->lsm_magic == LOV_MAGIC_V1 || \
1057 (lsmp)->lsm_magic == LOV_MAGIC_V3), \
1058 "%p->lsm_magic=%x\n", (lsmp), (lsmp)->lsm_magic); \
1061 static int lov_destroy(const struct lu_env *env, struct obd_export *exp,
1062 struct obdo *oa, struct lov_stripe_md *lsm,
1063 struct obd_trans_info *oti, struct obd_export *md_exp,
1066 struct lov_request_set *set;
1067 struct obd_info oinfo;
1068 struct lov_request *req;
1069 struct list_head *pos;
1070 struct lov_obd *lov;
1071 int rc = 0, err = 0;
1073 ASSERT_LSM_MAGIC(lsm);
1075 if (!exp || !exp->exp_obd)
1078 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1080 LASSERT(oti->oti_logcookies);
1083 lov = &exp->exp_obd->u.lov;
1084 obd_getref(exp->exp_obd);
1085 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1089 list_for_each(pos, &set->set_list) {
1090 req = list_entry(pos, struct lov_request, rq_link);
1092 if (oa->o_valid & OBD_MD_FLCOOKIE)
1093 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1095 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1096 req->rq_oi.oi_oa, NULL, oti, NULL, capa);
1097 err = lov_update_common_set(set, req, err);
1099 CERROR("%s: destroying objid "DOSTID" subobj "
1100 DOSTID" on OST idx %d: rc = %d\n",
1101 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1102 POSTID(&req->rq_oi.oi_oa->o_oi),
1110 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1111 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1113 err = lov_fini_destroy_set(set);
1115 obd_putref(exp->exp_obd);
1116 return rc ? rc : err;
1119 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1122 struct lov_request_set *lovset = (struct lov_request_set *)data;
1125 /* don't do attribute merge if this async op failed */
1127 atomic_set(&lovset->set_completes, 0);
1128 err = lov_fini_getattr_set(lovset);
1129 return rc ? rc : err;
1132 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1133 struct ptlrpc_request_set *rqset)
1135 struct lov_request_set *lovset;
1136 struct lov_obd *lov;
1137 struct list_head *pos;
1138 struct lov_request *req;
1142 ASSERT_LSM_MAGIC(oinfo->oi_md);
1144 if (!exp || !exp->exp_obd)
1147 lov = &exp->exp_obd->u.lov;
1149 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1153 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1154 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1155 oinfo->oi_md->lsm_stripe_size);
1157 list_for_each(pos, &lovset->set_list) {
1158 req = list_entry(pos, struct lov_request, 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 == NULL);
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. */
1204 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1205 struct obd_trans_info *oti,
1206 struct ptlrpc_request_set *rqset)
1208 struct lov_request_set *set;
1209 struct lov_request *req;
1210 struct list_head *pos;
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(pos, &set->set_list) {
1235 req = list_entry(pos, struct lov_request, rq_link);
1237 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1238 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1240 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1241 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1242 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1244 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1245 &req->rq_oi, oti, rqset);
1247 CERROR("error: setattr objid "DOSTID" subobj"
1248 DOSTID" on OST idx %d: rc = %d\n",
1249 POSTID(&set->set_oi->oi_oa->o_oi),
1250 POSTID(&req->rq_oi.oi_oa->o_oi),
1256 /* If we are not waiting for responses on async requests, return. */
1257 if (rc || !rqset || list_empty(&rqset->set_requests)) {
1260 atomic_set(&set->set_completes, 0);
1261 err = lov_fini_setattr_set(set);
1262 return rc ? rc : err;
1265 LASSERT(rqset->set_interpret == NULL);
1266 rqset->set_interpret = lov_setattr_interpret;
1267 rqset->set_arg = (void *)set;
1272 /* 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 list_head *pos;
1330 struct lov_obd *lov;
1333 LASSERT(oinfo != NULL);
1334 LASSERT(oinfo->oi_osfs != NULL);
1337 rc = lov_prep_statfs_set(obd, oinfo, &set);
1341 list_for_each(pos, &set->set_list) {
1342 req = list_entry(pos, struct lov_request, rq_link);
1343 rc = obd_statfs_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1344 &req->rq_oi, max_age, rqset);
1349 if (rc || list_empty(&rqset->set_requests)) {
1352 atomic_set(&set->set_completes, 0);
1353 err = lov_fini_statfs_set(set);
1354 return rc ? rc : err;
1357 LASSERT(rqset->set_interpret == NULL);
1358 rqset->set_interpret = lov_statfs_interpret;
1359 rqset->set_arg = (void *)set;
1363 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1364 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1366 struct ptlrpc_request_set *set = NULL;
1367 struct obd_info oinfo = { { { 0 } } };
1370 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1371 * statfs requests */
1372 set = ptlrpc_prep_set();
1376 oinfo.oi_osfs = osfs;
1377 oinfo.oi_flags = flags;
1378 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1380 rc = ptlrpc_set_wait(set);
1381 ptlrpc_set_destroy(set);
1386 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1387 void *karg, void *uarg)
1389 struct obd_device *obddev = class_exp2obd(exp);
1390 struct lov_obd *lov = &obddev->u.lov;
1391 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1392 struct obd_uuid *uuidp;
1395 case IOC_OBD_STATFS: {
1396 struct obd_ioctl_data *data = karg;
1397 struct obd_device *osc_obd;
1398 struct obd_statfs stat_buf = {0};
1402 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1403 if ((index >= count))
1406 if (!lov->lov_tgts[index])
1407 /* Try again with the next index */
1409 if (!lov->lov_tgts[index]->ltd_active)
1412 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1417 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1418 min((int) data->ioc_plen2,
1419 (int) sizeof(struct obd_uuid))))
1422 flags = uarg ? *(__u32 *)uarg : 0;
1423 /* got statfs data */
1424 rc = obd_statfs(NULL, lov->lov_tgts[index]->ltd_exp, &stat_buf,
1425 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1429 if (copy_to_user(data->ioc_pbuf1, &stat_buf,
1430 min((int) data->ioc_plen1,
1431 (int) sizeof(stat_buf))))
1435 case OBD_IOC_LOV_GET_CONFIG: {
1436 struct obd_ioctl_data *data;
1437 struct lov_desc *desc;
1442 if (obd_ioctl_getdata(&buf, &len, (void *)uarg))
1445 data = (struct obd_ioctl_data *)buf;
1447 if (sizeof(*desc) > data->ioc_inllen1) {
1448 obd_ioctl_freedata(buf, len);
1452 if (sizeof(uuidp->uuid) * count > data->ioc_inllen2) {
1453 obd_ioctl_freedata(buf, len);
1457 if (sizeof(__u32) * count > data->ioc_inllen3) {
1458 obd_ioctl_freedata(buf, len);
1462 desc = (struct lov_desc *)data->ioc_inlbuf1;
1463 memcpy(desc, &(lov->desc), sizeof(*desc));
1465 uuidp = (struct obd_uuid *)data->ioc_inlbuf2;
1466 genp = (__u32 *)data->ioc_inlbuf3;
1467 /* the uuid will be empty for deleted OSTs */
1468 for (i = 0; i < count; i++, uuidp++, genp++) {
1469 if (!lov->lov_tgts[i])
1471 *uuidp = lov->lov_tgts[i]->ltd_uuid;
1472 *genp = lov->lov_tgts[i]->ltd_gen;
1475 if (copy_to_user((void *)uarg, buf, len))
1477 obd_ioctl_freedata(buf, len);
1480 case LL_IOC_LOV_GETSTRIPE:
1481 rc = lov_getstripe(exp, karg, uarg);
1483 case OBD_IOC_QUOTACTL: {
1484 struct if_quotactl *qctl = karg;
1485 struct lov_tgt_desc *tgt = NULL;
1486 struct obd_quotactl *oqctl;
1488 if (qctl->qc_valid == QC_OSTIDX) {
1489 if (qctl->qc_idx < 0 || count <= qctl->qc_idx)
1492 tgt = lov->lov_tgts[qctl->qc_idx];
1493 if (!tgt || !tgt->ltd_exp)
1495 } else if (qctl->qc_valid == QC_UUID) {
1496 for (i = 0; i < count; i++) {
1497 tgt = lov->lov_tgts[i];
1499 !obd_uuid_equals(&tgt->ltd_uuid,
1503 if (tgt->ltd_exp == NULL)
1515 LASSERT(tgt && tgt->ltd_exp);
1516 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1520 QCTL_COPY(oqctl, qctl);
1521 rc = obd_quotactl(tgt->ltd_exp, oqctl);
1523 QCTL_COPY(qctl, oqctl);
1524 qctl->qc_valid = QC_OSTIDX;
1525 qctl->obd_uuid = tgt->ltd_uuid;
1536 for (i = 0; i < count; i++) {
1538 struct obd_device *osc_obd;
1540 /* OST was disconnected */
1541 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_exp)
1544 /* ll_umount_begin() sets force flag but for lov, not
1545 * osc. Let's pass it through */
1546 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1547 osc_obd->obd_force = obddev->obd_force;
1548 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1550 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1553 if (lov->lov_tgts[i]->ltd_active) {
1554 CDEBUG(err == -ENOTTY ?
1555 D_IOCTL : D_WARNING,
1556 "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n",
1557 lov_uuid2str(lov, i),
1574 #define FIEMAP_BUFFER_SIZE 4096
1577 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1578 * call. The local end offset and the device are sent in the first
1579 * fm_extent. This function calculates the stripe number from the index.
1580 * This function returns a stripe_no on which mapping is to be restarted.
1582 * This function returns fm_end_offset which is the in-OST offset at which
1583 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1584 * will re-calculate proper offset in next stripe.
1585 * Note that the first extent is passed to lov_get_info via the value field.
1587 * \param fiemap fiemap request header
1588 * \param lsm striping information for the file
1589 * \param fm_start logical start of mapping
1590 * \param fm_end logical end of mapping
1591 * \param start_stripe starting stripe will be returned in this
1593 static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1594 struct lov_stripe_md *lsm, u64 fm_start,
1595 u64 fm_end, int *start_stripe)
1597 u64 local_end = fiemap->fm_extents[0].fe_logical;
1598 u64 lun_start, lun_end;
1600 int stripe_no = -1, i;
1602 if (fiemap->fm_extent_count == 0 ||
1603 fiemap->fm_extents[0].fe_logical == 0)
1606 /* Find out stripe_no from ost_index saved in the fe_device */
1607 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1608 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1610 if (lov_oinfo_is_dummy(oinfo))
1613 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1618 if (stripe_no == -1)
1621 /* If we have finished mapping on previous device, shift logical
1622 * offset to start of next device */
1623 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1624 &lun_start, &lun_end)) != 0 &&
1625 local_end < lun_end) {
1626 fm_end_offset = local_end;
1627 *start_stripe = stripe_no;
1629 /* This is a special value to indicate that caller should
1630 * calculate offset in next stripe. */
1632 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1635 return fm_end_offset;
1639 * We calculate on which OST the mapping will end. If the length of mapping
1640 * is greater than (stripe_size * stripe_count) then the last_stripe will
1641 * will be one just before start_stripe. Else we check if the mapping
1642 * intersects each OST and find last_stripe.
1643 * This function returns the last_stripe and also sets the stripe_count
1644 * over which the mapping is spread
1646 * \param lsm striping information for the file
1647 * \param fm_start logical start of mapping
1648 * \param fm_end logical end of mapping
1649 * \param start_stripe starting stripe of the mapping
1650 * \param stripe_count the number of stripes across which to map is returned
1652 * \retval last_stripe return the last stripe of the mapping
1654 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start,
1655 u64 fm_end, int start_stripe,
1659 u64 obd_start, obd_end;
1662 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1663 last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1665 *stripe_count = lsm->lsm_stripe_count;
1667 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1668 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1669 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1670 &obd_start, &obd_end)) == 0)
1674 last_stripe = (start_stripe + j - 1) %lsm->lsm_stripe_count;
1681 * Set fe_device and copy extents from local buffer into main return buffer.
1683 * \param fiemap fiemap request header
1684 * \param lcl_fm_ext array of local fiemap extents to be copied
1685 * \param ost_index OST index to be written into the fm_device field for each
1687 * \param ext_count number of extents to be copied
1688 * \param current_extent where to start copying in main extent array
1690 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1691 struct ll_fiemap_extent *lcl_fm_ext,
1692 int ost_index, unsigned int ext_count,
1698 for (ext = 0; ext < ext_count; ext++) {
1699 lcl_fm_ext[ext].fe_device = ost_index;
1700 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1703 /* Copy fm_extent's from fm_local to return buffer */
1704 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1705 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1709 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1710 * This also handles the restarting of FIEMAP calls in case mapping overflows
1711 * the available number of extents in single call.
1713 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1714 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1716 struct ll_fiemap_info_key *fm_key = key;
1717 struct ll_user_fiemap *fiemap = val;
1718 struct ll_user_fiemap *fm_local = NULL;
1719 struct ll_fiemap_extent *lcl_fm_ext;
1721 unsigned int get_num_extents = 0;
1722 int ost_index = 0, actual_start_stripe, start_stripe;
1723 u64 fm_start, fm_end, fm_length, fm_end_offset;
1725 int current_extent = 0, rc = 0, i;
1726 int ost_eof = 0; /* EOF for object */
1727 int ost_done = 0; /* done with required mapping for this OST? */
1729 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1730 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1732 if (!lsm_has_objects(lsm)) {
1737 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1738 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1740 OBD_ALLOC_LARGE(fm_local, buffer_size);
1741 if (fm_local == NULL) {
1745 lcl_fm_ext = &fm_local->fm_extents[0];
1747 count_local = fiemap_size_to_count(buffer_size);
1749 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1750 fm_start = fiemap->fm_start;
1751 fm_length = fiemap->fm_length;
1752 /* Calculate start stripe, last stripe and length of mapping */
1753 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1754 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1755 fm_start + fm_length - 1);
1756 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1757 if (fm_end > fm_key->oa.o_size)
1758 fm_end = fm_key->oa.o_size;
1760 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1761 actual_start_stripe, &stripe_count);
1763 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1764 fm_end, &start_stripe);
1765 if (fm_end_offset == -EINVAL) {
1770 if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1771 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1772 if (fiemap->fm_extent_count == 0) {
1773 get_num_extents = 1;
1776 /* Check each stripe */
1777 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1778 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1779 u64 req_fm_len; /* Stores length of required mapping */
1780 u64 len_mapped_single_call;
1781 u64 lun_start, lun_end, obd_object_end;
1782 unsigned int ext_count;
1784 cur_stripe_wrap = cur_stripe;
1786 /* Find out range of mapping on this stripe */
1787 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1788 &lun_start, &obd_object_end)) == 0)
1791 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) {
1796 /* If this is a continuation FIEMAP call and we are on
1797 * starting stripe then lun_start needs to be set to
1799 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1800 lun_start = fm_end_offset;
1802 if (fm_length != ~0ULL) {
1803 /* Handle fm_start + fm_length overflow */
1804 if (fm_start + fm_length < fm_start)
1805 fm_length = ~0ULL - fm_start;
1806 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1812 if (lun_start == lun_end)
1815 req_fm_len = obd_object_end - lun_start;
1816 fm_local->fm_length = 0;
1817 len_mapped_single_call = 0;
1819 /* If the output buffer is very large and the objects have many
1820 * extents we may need to loop on a single OST repeatedly */
1824 if (get_num_extents == 0) {
1825 /* Don't get too many extents. */
1826 if (current_extent + count_local >
1827 fiemap->fm_extent_count)
1828 count_local = fiemap->fm_extent_count -
1832 lun_start += len_mapped_single_call;
1833 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1834 req_fm_len = fm_local->fm_length;
1835 fm_local->fm_extent_count = count_local;
1836 fm_local->fm_mapped_extents = 0;
1837 fm_local->fm_flags = fiemap->fm_flags;
1839 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1840 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1842 if (ost_index < 0 ||
1843 ost_index >= lov->desc.ld_tgt_count) {
1848 /* If OST is inactive, return extent with UNKNOWN flag */
1849 if (!lov->lov_tgts[ost_index]->ltd_active) {
1850 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1851 fm_local->fm_mapped_extents = 1;
1853 lcl_fm_ext[0].fe_logical = lun_start;
1854 lcl_fm_ext[0].fe_length = obd_object_end -
1856 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1861 fm_local->fm_start = lun_start;
1862 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1863 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1864 *vallen=fiemap_count_to_size(fm_local->fm_extent_count);
1865 rc = obd_get_info(NULL,
1866 lov->lov_tgts[ost_index]->ltd_exp,
1867 keylen, key, vallen, fm_local, lsm);
1872 ext_count = fm_local->fm_mapped_extents;
1873 if (ext_count == 0) {
1875 /* If last stripe has hole at the end,
1876 * then we need to return */
1877 if (cur_stripe_wrap == last_stripe) {
1878 fiemap->fm_mapped_extents = 0;
1884 /* If we just need num of extents then go to next device */
1885 if (get_num_extents) {
1886 current_extent += ext_count;
1890 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1891 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1893 /* Have we finished mapping on this device? */
1894 if (req_fm_len <= len_mapped_single_call)
1897 /* Clear the EXTENT_LAST flag which can be present on
1899 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1900 lcl_fm_ext[ext_count - 1].fe_flags &=
1901 ~FIEMAP_EXTENT_LAST;
1903 curr_loc = lov_stripe_size(lsm,
1904 lcl_fm_ext[ext_count - 1].fe_logical+
1905 lcl_fm_ext[ext_count - 1].fe_length,
1907 if (curr_loc >= fm_key->oa.o_size)
1910 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1911 ost_index, ext_count,
1914 current_extent += ext_count;
1916 /* Ran out of available extents? */
1917 if (current_extent >= fiemap->fm_extent_count)
1919 } while (ost_done == 0 && ost_eof == 0);
1921 if (cur_stripe_wrap == last_stripe)
1926 /* Indicate that we are returning device offsets unless file just has
1928 if (lsm->lsm_stripe_count > 1)
1929 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1931 if (get_num_extents)
1932 goto skip_last_device_calc;
1934 /* Check if we have reached the last stripe and whether mapping for that
1935 * stripe is done. */
1936 if (cur_stripe_wrap == last_stripe) {
1937 if (ost_done || ost_eof)
1938 fiemap->fm_extents[current_extent - 1].fe_flags |=
1942 skip_last_device_calc:
1943 fiemap->fm_mapped_extents = current_extent;
1946 OBD_FREE_LARGE(fm_local, buffer_size);
1950 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1951 __u32 keylen, void *key, __u32 *vallen, void *val,
1952 struct lov_stripe_md *lsm)
1954 struct obd_device *obddev = class_exp2obd(exp);
1955 struct lov_obd *lov = &obddev->u.lov;
1958 if (!vallen || !val)
1963 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
1966 struct ldlm_lock *lock;
1968 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
1969 struct lov_oinfo *loi;
1970 __u32 *stripe = val;
1972 if (*vallen < sizeof(*stripe)) {
1976 *vallen = sizeof(*stripe);
1978 /* XXX This is another one of those bits that will need to
1979 * change if we ever actually support nested LOVs. It uses
1980 * the lock's export to find out which stripe it is. */
1981 /* XXX - it's assumed all the locks for deleted OSTs have
1982 * been cancelled. Also, the export for deleted OSTs will
1983 * be NULL and won't match the lock's export. */
1984 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1985 loi = lsm->lsm_oinfo[i];
1986 if (lov_oinfo_is_dummy(loi))
1989 if (!lov->lov_tgts[loi->loi_ost_idx])
1991 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
1992 data->lock->l_conn_export &&
1993 ostid_res_name_eq(&loi->loi_oi, res_id)) {
1999 LDLM_ERROR(data->lock, "lock on inode without such object");
2000 dump_lsm(D_ERROR, lsm);
2003 } else if (KEY_IS(KEY_LAST_ID)) {
2004 struct obd_id_info *info = val;
2005 __u32 size = sizeof(u64);
2006 struct lov_tgt_desc *tgt;
2008 LASSERT(*vallen == sizeof(struct obd_id_info));
2009 tgt = lov->lov_tgts[info->idx];
2011 if (!tgt || !tgt->ltd_active) {
2016 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2017 &size, info->data, NULL);
2020 } else if (KEY_IS(KEY_LOVDESC)) {
2021 struct lov_desc *desc_ret = val;
2022 *desc_ret = lov->desc;
2026 } else if (KEY_IS(KEY_FIEMAP)) {
2027 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2029 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2030 struct lov_tgt_desc *tgt;
2031 __u64 ost_idx = *((__u64 *)val);
2033 LASSERT(*vallen == sizeof(__u64));
2034 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2035 tgt = lov->lov_tgts[ost_idx];
2037 if (!tgt || !tgt->ltd_exp) {
2042 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2045 } else if (KEY_IS(KEY_TGT_COUNT)) {
2046 *((int *)val) = lov->desc.ld_tgt_count;
2058 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2059 u32 keylen, void *key, u32 vallen,
2060 void *val, struct ptlrpc_request_set *set)
2062 struct obd_device *obddev = class_exp2obd(exp);
2063 struct lov_obd *lov = &obddev->u.lov;
2066 struct lov_tgt_desc *tgt;
2067 unsigned incr, check_uuid,
2068 do_inactive, no_set;
2069 unsigned next_id = 0, mds_con = 0, capa = 0;
2071 incr = check_uuid = do_inactive = no_set = 0;
2074 set = ptlrpc_prep_set();
2080 count = lov->desc.ld_tgt_count;
2082 if (KEY_IS(KEY_NEXT_ID)) {
2083 count = vallen / sizeof(struct obd_id_info);
2084 vallen = sizeof(u64);
2085 incr = sizeof(struct obd_id_info);
2088 } else if (KEY_IS(KEY_CHECKSUM)) {
2090 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2091 /* use defaults: do_inactive = incr = 0; */
2092 } else if (KEY_IS(KEY_MDS_CONN)) {
2094 } else if (KEY_IS(KEY_CAPA_KEY)) {
2096 } else if (KEY_IS(KEY_CACHE_SET)) {
2097 LASSERT(lov->lov_cache == NULL);
2098 lov->lov_cache = val;
2102 for (i = 0; i < count; i++, val = (char *)val + incr) {
2104 tgt = lov->lov_tgts[((struct obd_id_info *)val)->idx];
2106 tgt = lov->lov_tgts[i];
2108 /* OST was disconnected */
2109 if (!tgt || !tgt->ltd_exp)
2112 /* OST is inactive and we don't want inactive OSCs */
2113 if (!tgt->ltd_active && !do_inactive)
2117 struct mds_group_info *mgi;
2119 LASSERT(vallen == sizeof(*mgi));
2120 mgi = (struct mds_group_info *)val;
2122 /* Only want a specific OSC */
2123 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2127 err = obd_set_info_async(env, tgt->ltd_exp,
2128 keylen, key, sizeof(int),
2130 } else if (next_id) {
2131 err = obd_set_info_async(env, tgt->ltd_exp,
2132 keylen, key, vallen,
2133 ((struct obd_id_info *)val)->data, set);
2135 struct mds_capa_info *info = (struct mds_capa_info *)val;
2137 LASSERT(vallen == sizeof(*info));
2139 /* Only want a specific OSC */
2141 !obd_uuid_equals(info->uuid, &tgt->ltd_uuid))
2144 err = obd_set_info_async(env, tgt->ltd_exp, keylen,
2145 key, sizeof(*info->capa),
2148 /* Only want a specific OSC */
2150 !obd_uuid_equals(val, &tgt->ltd_uuid))
2153 err = obd_set_info_async(env, tgt->ltd_exp,
2154 keylen, key, vallen, val, set);
2163 err = ptlrpc_set_wait(set);
2166 ptlrpc_set_destroy(set);
2171 void lov_stripe_lock(struct lov_stripe_md *md)
2172 __acquires(&md->lsm_lock)
2174 LASSERT(md->lsm_lock_owner != current_pid());
2175 spin_lock(&md->lsm_lock);
2176 LASSERT(md->lsm_lock_owner == 0);
2177 md->lsm_lock_owner = current_pid();
2179 EXPORT_SYMBOL(lov_stripe_lock);
2181 void lov_stripe_unlock(struct lov_stripe_md *md)
2182 __releases(&md->lsm_lock)
2184 LASSERT(md->lsm_lock_owner == current_pid());
2185 md->lsm_lock_owner = 0;
2186 spin_unlock(&md->lsm_lock);
2188 EXPORT_SYMBOL(lov_stripe_unlock);
2190 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2191 struct obd_quotactl *oqctl)
2193 struct lov_obd *lov = &obd->u.lov;
2194 struct lov_tgt_desc *tgt;
2196 __u64 bhardlimit = 0;
2199 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2200 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2201 oqctl->qc_cmd != Q_GETOQUOTA &&
2202 oqctl->qc_cmd != Q_INITQUOTA &&
2203 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2204 oqctl->qc_cmd != Q_FINVALIDATE) {
2205 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2211 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2214 tgt = lov->lov_tgts[i];
2219 if (!tgt->ltd_active || tgt->ltd_reap) {
2220 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2221 lov->lov_tgts[i]->ltd_activate) {
2223 CERROR("ost %d is inactive\n", i);
2225 CDEBUG(D_HA, "ost %d is inactive\n", i);
2230 err = obd_quotactl(tgt->ltd_exp, oqctl);
2232 if (tgt->ltd_active && !rc)
2237 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2238 curspace += oqctl->qc_dqblk.dqb_curspace;
2239 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2244 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2245 oqctl->qc_dqblk.dqb_curspace = curspace;
2246 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2251 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2252 struct obd_quotactl *oqctl)
2254 struct lov_obd *lov = &obd->u.lov;
2259 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2260 if (!lov->lov_tgts[i])
2263 /* Skip quota check on the administratively disabled OSTs. */
2264 if (!lov->lov_tgts[i]->ltd_activate) {
2265 CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n",
2270 if (!lov->lov_tgts[i]->ltd_active) {
2271 CERROR("lov idx %d inactive\n", i);
2277 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2280 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2283 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2294 static struct obd_ops lov_obd_ops = {
2295 .o_owner = THIS_MODULE,
2296 .o_setup = lov_setup,
2297 .o_precleanup = lov_precleanup,
2298 .o_cleanup = lov_cleanup,
2299 /*.o_process_config = lov_process_config,*/
2300 .o_connect = lov_connect,
2301 .o_disconnect = lov_disconnect,
2302 .o_statfs = lov_statfs,
2303 .o_statfs_async = lov_statfs_async,
2304 .o_packmd = lov_packmd,
2305 .o_unpackmd = lov_unpackmd,
2306 .o_create = lov_create,
2307 .o_destroy = lov_destroy,
2308 .o_getattr_async = lov_getattr_async,
2309 .o_setattr_async = lov_setattr_async,
2310 .o_adjust_kms = lov_adjust_kms,
2311 .o_find_cbdata = lov_find_cbdata,
2312 .o_iocontrol = lov_iocontrol,
2313 .o_get_info = lov_get_info,
2314 .o_set_info_async = lov_set_info_async,
2315 .o_notify = lov_notify,
2316 .o_pool_new = lov_pool_new,
2317 .o_pool_rem = lov_pool_remove,
2318 .o_pool_add = lov_pool_add,
2319 .o_pool_del = lov_pool_del,
2320 .o_getref = lov_getref,
2321 .o_putref = lov_putref,
2322 .o_quotactl = lov_quotactl,
2323 .o_quotacheck = lov_quotacheck,
2326 struct kmem_cache *lov_oinfo_slab;
2328 static int __init lov_init(void)
2330 struct lprocfs_static_vars lvars = { NULL };
2333 /* print an address of _any_ initialized kernel symbol from this
2334 * module, to allow debugging with gdb that doesn't support data
2335 * symbols from modules.*/
2336 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2338 rc = lu_kmem_init(lov_caches);
2342 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2343 sizeof(struct lov_oinfo),
2344 0, SLAB_HWCACHE_ALIGN, NULL);
2345 if (lov_oinfo_slab == NULL) {
2346 lu_kmem_fini(lov_caches);
2349 lprocfs_lov_init_vars(&lvars);
2351 rc = class_register_type(&lov_obd_ops, NULL,
2352 LUSTRE_LOV_NAME, &lov_device_type);
2355 kmem_cache_destroy(lov_oinfo_slab);
2356 lu_kmem_fini(lov_caches);
2362 static void /*__exit*/ lov_exit(void)
2364 class_unregister_type(LUSTRE_LOV_NAME);
2365 kmem_cache_destroy(lov_oinfo_slab);
2367 lu_kmem_fini(lov_caches);
2370 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2371 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2372 MODULE_LICENSE("GPL");
2373 MODULE_VERSION(LUSTRE_VERSION_STRING);
2375 module_init(lov_init);
2376 module_exit(lov_exit);