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;
89 CDEBUG(D_CONFIG, "destroying %d lov targets\n",
91 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
92 tgt = lov->lov_tgts[i];
94 if (!tgt || !tgt->ltd_reap)
96 list_add(&tgt->ltd_kill, &kill);
97 /* XXX - right now there is a dependency on ld_tgt_count
98 * being the maximum tgt index for computing the
99 * mds_max_easize. So we can't shrink it. */
100 lov_ost_pool_remove(&lov->lov_packed, i);
101 lov->lov_tgts[i] = NULL;
102 lov->lov_death_row--;
104 mutex_unlock(&lov->lov_lock);
106 list_for_each_entry_safe(tgt, n, &kill, ltd_kill) {
107 list_del(&tgt->ltd_kill);
109 __lov_del_obd(obd, tgt);
112 if (lov->lov_tgts_kobj)
113 kobject_put(lov->lov_tgts_kobj);
116 mutex_unlock(&lov->lov_lock);
120 static int lov_set_osc_active(struct obd_device *obd, struct obd_uuid *uuid,
121 enum obd_notify_event ev);
122 static int lov_notify(struct obd_device *obd, struct obd_device *watched,
123 enum obd_notify_event ev, void *data);
125 #define MAX_STRING_SIZE 128
126 int lov_connect_obd(struct obd_device *obd, __u32 index, int activate,
127 struct obd_connect_data *data)
129 struct lov_obd *lov = &obd->u.lov;
130 struct obd_uuid *tgt_uuid;
131 struct obd_device *tgt_obd;
132 static struct obd_uuid lov_osc_uuid = { "LOV_OSC_UUID" };
133 struct obd_import *imp;
136 if (!lov->lov_tgts[index])
139 tgt_uuid = &lov->lov_tgts[index]->ltd_uuid;
140 tgt_obd = lov->lov_tgts[index]->ltd_obd;
142 if (!tgt_obd->obd_set_up) {
143 CERROR("Target %s not set up\n", obd_uuid2str(tgt_uuid));
147 /* override the sp_me from lov */
148 tgt_obd->u.cli.cl_sp_me = lov->lov_sp_me;
150 if (data && (data->ocd_connect_flags & OBD_CONNECT_INDEX))
151 data->ocd_index = index;
154 * Divine LOV knows that OBDs under it are OSCs.
156 imp = tgt_obd->u.cli.cl_import;
159 tgt_obd->obd_no_recov = 0;
160 /* FIXME this is probably supposed to be
161 ptlrpc_set_import_active. Horrible naming. */
162 ptlrpc_activate_import(imp);
165 rc = obd_register_observer(tgt_obd, obd);
167 CERROR("Target %s register_observer error %d\n",
168 obd_uuid2str(tgt_uuid), rc);
172 if (imp->imp_invalid) {
173 CDEBUG(D_CONFIG, "not connecting OSC %s; administratively disabled\n",
174 obd_uuid2str(tgt_uuid));
178 rc = obd_connect(NULL, &lov->lov_tgts[index]->ltd_exp, tgt_obd,
179 &lov_osc_uuid, data, NULL);
180 if (rc || !lov->lov_tgts[index]->ltd_exp) {
181 CERROR("Target %s connect error %d\n",
182 obd_uuid2str(tgt_uuid), rc);
186 lov->lov_tgts[index]->ltd_reap = 0;
188 CDEBUG(D_CONFIG, "Connected tgt idx %d %s (%s) %sactive\n", index,
189 obd_uuid2str(tgt_uuid), tgt_obd->obd_name, activate ? "":"in");
191 if (lov->lov_tgts_kobj)
192 /* Even if we failed, that's ok */
193 rc = sysfs_create_link(lov->lov_tgts_kobj, &tgt_obd->obd_kobj,
199 static int lov_connect(const struct lu_env *env,
200 struct obd_export **exp, struct obd_device *obd,
201 struct obd_uuid *cluuid, struct obd_connect_data *data,
204 struct lov_obd *lov = &obd->u.lov;
205 struct lov_tgt_desc *tgt;
206 struct lustre_handle conn;
209 CDEBUG(D_CONFIG, "connect #%d\n", lov->lov_connects);
211 rc = class_connect(&conn, obd, cluuid);
215 *exp = class_conn2export(&conn);
217 /* Why should there ever be more than 1 connect? */
219 LASSERT(lov->lov_connects == 1);
221 memset(&lov->lov_ocd, 0, sizeof(lov->lov_ocd));
223 lov->lov_ocd = *data;
227 lov->lov_tgts_kobj = kobject_create_and_add("target_obds",
230 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
231 tgt = lov->lov_tgts[i];
232 if (!tgt || obd_uuid_empty(&tgt->ltd_uuid))
234 /* Flags will be lowest common denominator */
235 rc = lov_connect_obd(obd, i, tgt->ltd_activate, &lov->lov_ocd);
237 CERROR("%s: lov connect tgt %d failed: %d\n",
238 obd->obd_name, i, rc);
241 /* connect to administrative disabled ost */
242 if (!lov->lov_tgts[i]->ltd_exp)
245 rc = lov_notify(obd, lov->lov_tgts[i]->ltd_exp->exp_obd,
246 OBD_NOTIFY_CONNECT, (void *)&i);
248 CERROR("%s error sending notify %d\n",
257 static int lov_disconnect_obd(struct obd_device *obd, struct lov_tgt_desc *tgt)
259 struct lov_obd *lov = &obd->u.lov;
260 struct obd_device *osc_obd;
263 osc_obd = class_exp2obd(tgt->ltd_exp);
264 CDEBUG(D_CONFIG, "%s: disconnecting target %s\n",
265 obd->obd_name, osc_obd ? osc_obd->obd_name : "NULL");
267 if (tgt->ltd_active) {
269 lov->desc.ld_active_tgt_count--;
270 tgt->ltd_exp->exp_obd->obd_inactive = 1;
274 if (lov->lov_tgts_kobj)
275 sysfs_remove_link(lov->lov_tgts_kobj,
278 /* Pass it on to our clients.
279 * XXX This should be an argument to disconnect,
280 * XXX not a back-door flag on the OBD. Ah well.
282 osc_obd->obd_force = obd->obd_force;
283 osc_obd->obd_fail = obd->obd_fail;
284 osc_obd->obd_no_recov = obd->obd_no_recov;
287 obd_register_observer(osc_obd, NULL);
289 rc = obd_disconnect(tgt->ltd_exp);
291 CERROR("Target %s disconnect error %d\n",
292 tgt->ltd_uuid.uuid, rc);
300 static int lov_disconnect(struct obd_export *exp)
302 struct obd_device *obd = class_exp2obd(exp);
303 struct lov_obd *lov = &obd->u.lov;
309 /* Only disconnect the underlying layers on the final disconnect. */
311 if (lov->lov_connects != 0) {
312 /* why should there be more than 1 connect? */
313 CERROR("disconnect #%d\n", lov->lov_connects);
317 /* Let's hold another reference so lov_del_obd doesn't spin through
321 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
322 if (lov->lov_tgts[i] && lov->lov_tgts[i]->ltd_exp) {
323 /* Disconnection is the last we know about an obd */
324 lov_del_target(obd, i, NULL, lov->lov_tgts[i]->ltd_gen);
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: {
836 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
837 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_active)
839 obd_precleanup(class_exp2obd(lov->lov_tgts[i]->ltd_exp),
851 static int lov_cleanup(struct obd_device *obd)
853 struct lov_obd *lov = &obd->u.lov;
854 struct list_head *pos, *tmp;
855 struct pool_desc *pool;
857 list_for_each_safe(pos, tmp, &lov->lov_pool_list) {
858 pool = list_entry(pos, struct pool_desc, pool_list);
859 /* free pool structs */
860 CDEBUG(D_INFO, "delete pool %p\n", pool);
861 /* In the function below, .hs_keycmp resolves to
862 * pool_hashkey_keycmp() */
863 /* coverity[overrun-buffer-val] */
864 lov_pool_del(obd, pool->pool_name);
866 cfs_hash_putref(lov->lov_pools_hash_body);
867 lov_ost_pool_free(&lov->lov_packed);
869 lprocfs_obd_cleanup(obd);
874 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
875 if (!lov->lov_tgts[i])
878 /* Inactive targets may never have connected */
879 if (lov->lov_tgts[i]->ltd_active ||
880 atomic_read(&lov->lov_refcount))
881 /* We should never get here - these
882 should have been removed in the
884 CERROR("lov tgt %d not cleaned! deathrow=%d, lovrc=%d\n",
885 i, lov->lov_death_row,
886 atomic_read(&lov->lov_refcount));
887 lov_del_target(obd, i, NULL, 0);
890 kfree(lov->lov_tgts);
891 lov->lov_tgt_size = 0;
896 int lov_process_config_base(struct obd_device *obd, struct lustre_cfg *lcfg,
897 __u32 *indexp, int *genp)
899 struct obd_uuid obd_uuid;
903 switch (cmd = lcfg->lcfg_command) {
904 case LCFG_LOV_ADD_OBD:
905 case LCFG_LOV_ADD_INA:
906 case LCFG_LOV_DEL_OBD: {
909 /* lov_modify_tgts add 0:lov_mdsA 1:ost1_UUID 2:0 3:1 */
910 if (LUSTRE_CFG_BUFLEN(lcfg, 1) > sizeof(obd_uuid.uuid)) {
915 obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1));
917 rc = kstrtoint(lustre_cfg_buf(lcfg, 2), 10, indexp);
920 rc = kstrtoint(lustre_cfg_buf(lcfg, 3), 10, genp);
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)
1065 struct lov_request_set *set;
1066 struct obd_info oinfo;
1067 struct lov_request *req;
1068 struct list_head *pos;
1069 struct lov_obd *lov;
1070 int rc = 0, err = 0;
1072 ASSERT_LSM_MAGIC(lsm);
1074 if (!exp || !exp->exp_obd)
1077 if (oa->o_valid & OBD_MD_FLCOOKIE) {
1079 LASSERT(oti->oti_logcookies);
1082 lov = &exp->exp_obd->u.lov;
1083 obd_getref(exp->exp_obd);
1084 rc = lov_prep_destroy_set(exp, &oinfo, oa, lsm, oti, &set);
1088 list_for_each(pos, &set->set_list) {
1089 req = list_entry(pos, struct lov_request, rq_link);
1091 if (oa->o_valid & OBD_MD_FLCOOKIE)
1092 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1094 err = obd_destroy(env, lov->lov_tgts[req->rq_idx]->ltd_exp,
1095 req->rq_oi.oi_oa, NULL, oti, NULL);
1096 err = lov_update_common_set(set, req, err);
1098 CERROR("%s: destroying objid "DOSTID" subobj "
1099 DOSTID" on OST idx %d: rc = %d\n",
1100 exp->exp_obd->obd_name, POSTID(&oa->o_oi),
1101 POSTID(&req->rq_oi.oi_oa->o_oi),
1109 LASSERT(lsm_op_find(lsm->lsm_magic) != NULL);
1110 rc = lsm_op_find(lsm->lsm_magic)->lsm_destroy(lsm, oa, md_exp);
1112 err = lov_fini_destroy_set(set);
1114 obd_putref(exp->exp_obd);
1115 return rc ? rc : err;
1118 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset,
1121 struct lov_request_set *lovset = (struct lov_request_set *)data;
1124 /* don't do attribute merge if this async op failed */
1126 atomic_set(&lovset->set_completes, 0);
1127 err = lov_fini_getattr_set(lovset);
1128 return rc ? rc : err;
1131 static int lov_getattr_async(struct obd_export *exp, struct obd_info *oinfo,
1132 struct ptlrpc_request_set *rqset)
1134 struct lov_request_set *lovset;
1135 struct lov_obd *lov;
1136 struct list_head *pos;
1137 struct lov_request *req;
1141 ASSERT_LSM_MAGIC(oinfo->oi_md);
1143 if (!exp || !exp->exp_obd)
1146 lov = &exp->exp_obd->u.lov;
1148 rc = lov_prep_getattr_set(exp, oinfo, &lovset);
1152 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1153 POSTID(&oinfo->oi_md->lsm_oi), oinfo->oi_md->lsm_stripe_count,
1154 oinfo->oi_md->lsm_stripe_size);
1156 list_for_each(pos, &lovset->set_list) {
1157 req = list_entry(pos, struct lov_request, rq_link);
1159 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1160 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1161 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1162 rc = obd_getattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1163 &req->rq_oi, rqset);
1165 CERROR("%s: getattr objid "DOSTID" subobj"
1166 DOSTID" on OST idx %d: rc = %d\n",
1167 exp->exp_obd->obd_name,
1168 POSTID(&oinfo->oi_oa->o_oi),
1169 POSTID(&req->rq_oi.oi_oa->o_oi),
1175 if (!list_empty(&rqset->set_requests)) {
1177 LASSERT(rqset->set_interpret == NULL);
1178 rqset->set_interpret = lov_getattr_interpret;
1179 rqset->set_arg = (void *)lovset;
1184 atomic_set(&lovset->set_completes, 0);
1185 err = lov_fini_getattr_set(lovset);
1186 return rc ? rc : err;
1189 static int lov_setattr_interpret(struct ptlrpc_request_set *rqset,
1192 struct lov_request_set *lovset = (struct lov_request_set *)data;
1196 atomic_set(&lovset->set_completes, 0);
1197 err = lov_fini_setattr_set(lovset);
1198 return rc ? rc : err;
1201 /* If @oti is given, the request goes from MDS and responses from OSTs are not
1202 needed. Otherwise, a client is waiting for responses. */
1203 static int lov_setattr_async(struct obd_export *exp, struct obd_info *oinfo,
1204 struct obd_trans_info *oti,
1205 struct ptlrpc_request_set *rqset)
1207 struct lov_request_set *set;
1208 struct lov_request *req;
1209 struct list_head *pos;
1210 struct lov_obd *lov;
1214 ASSERT_LSM_MAGIC(oinfo->oi_md);
1215 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE) {
1217 LASSERT(oti->oti_logcookies);
1220 if (!exp || !exp->exp_obd)
1223 lov = &exp->exp_obd->u.lov;
1224 rc = lov_prep_setattr_set(exp, oinfo, oti, &set);
1228 CDEBUG(D_INFO, "objid "DOSTID": %ux%u byte stripes\n",
1229 POSTID(&oinfo->oi_md->lsm_oi),
1230 oinfo->oi_md->lsm_stripe_count,
1231 oinfo->oi_md->lsm_stripe_size);
1233 list_for_each(pos, &set->set_list) {
1234 req = list_entry(pos, struct lov_request, rq_link);
1236 if (oinfo->oi_oa->o_valid & OBD_MD_FLCOOKIE)
1237 oti->oti_logcookies = set->set_cookies + req->rq_stripe;
1239 CDEBUG(D_INFO, "objid " DOSTID "[%d] has subobj " DOSTID " at idx%u\n",
1240 POSTID(&oinfo->oi_oa->o_oi), req->rq_stripe,
1241 POSTID(&req->rq_oi.oi_oa->o_oi), req->rq_idx);
1243 rc = obd_setattr_async(lov->lov_tgts[req->rq_idx]->ltd_exp,
1244 &req->rq_oi, oti, rqset);
1246 CERROR("error: setattr objid "DOSTID" subobj"
1247 DOSTID" on OST idx %d: rc = %d\n",
1248 POSTID(&set->set_oi->oi_oa->o_oi),
1249 POSTID(&req->rq_oi.oi_oa->o_oi),
1255 /* If we are not waiting for responses on async requests, return. */
1256 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)) {
1353 atomic_set(&set->set_completes, 0);
1354 err = lov_fini_statfs_set(set);
1355 return rc ? rc : err;
1358 LASSERT(rqset->set_interpret == NULL);
1359 rqset->set_interpret = lov_statfs_interpret;
1360 rqset->set_arg = (void *)set;
1364 static int lov_statfs(const struct lu_env *env, struct obd_export *exp,
1365 struct obd_statfs *osfs, __u64 max_age, __u32 flags)
1367 struct ptlrpc_request_set *set = NULL;
1368 struct obd_info oinfo = { };
1371 /* for obdclass we forbid using obd_statfs_rqset, but prefer using async
1372 * statfs requests */
1373 set = ptlrpc_prep_set();
1377 oinfo.oi_osfs = osfs;
1378 oinfo.oi_flags = flags;
1379 rc = lov_statfs_async(exp, &oinfo, max_age, set);
1381 rc = ptlrpc_set_wait(set);
1382 ptlrpc_set_destroy(set);
1387 static int lov_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
1388 void *karg, void *uarg)
1390 struct obd_device *obddev = class_exp2obd(exp);
1391 struct lov_obd *lov = &obddev->u.lov;
1392 int i = 0, rc = 0, count = lov->desc.ld_tgt_count;
1393 struct obd_uuid *uuidp;
1396 case IOC_OBD_STATFS: {
1397 struct obd_ioctl_data *data = karg;
1398 struct obd_device *osc_obd;
1399 struct obd_statfs stat_buf = {0};
1403 memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
1407 if (!lov->lov_tgts[index])
1408 /* Try again with the next index */
1410 if (!lov->lov_tgts[index]->ltd_active)
1413 osc_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
1418 if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(osc_obd),
1419 min((int) data->ioc_plen2,
1420 (int) sizeof(struct obd_uuid))))
1423 flags = uarg ? *(__u32 *)uarg : 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,
1504 if (tgt->ltd_exp == NULL)
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 */
1547 osc_obd = class_exp2obd(lov->lov_tgts[i]->ltd_exp);
1548 osc_obd->obd_force = obddev->obd_force;
1549 err = obd_iocontrol(cmd, lov->lov_tgts[i]->ltd_exp,
1551 if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) {
1554 if (lov->lov_tgts[i]->ltd_active) {
1555 CDEBUG(err == -ENOTTY ?
1556 D_IOCTL : D_WARNING,
1557 "iocontrol OSC %s on OST idx %d cmd %x: err = %d\n",
1558 lov_uuid2str(lov, i),
1575 #define FIEMAP_BUFFER_SIZE 4096
1578 * Non-zero fe_logical indicates that this is a continuation FIEMAP
1579 * call. The local end offset and the device are sent in the first
1580 * fm_extent. This function calculates the stripe number from the index.
1581 * This function returns a stripe_no on which mapping is to be restarted.
1583 * This function returns fm_end_offset which is the in-OST offset at which
1584 * mapping should be restarted. If fm_end_offset=0 is returned then caller
1585 * will re-calculate proper offset in next stripe.
1586 * Note that the first extent is passed to lov_get_info via the value field.
1588 * \param fiemap fiemap request header
1589 * \param lsm striping information for the file
1590 * \param fm_start logical start of mapping
1591 * \param fm_end logical end of mapping
1592 * \param start_stripe starting stripe will be returned in this
1594 static u64 fiemap_calc_fm_end_offset(struct ll_user_fiemap *fiemap,
1595 struct lov_stripe_md *lsm, u64 fm_start,
1596 u64 fm_end, int *start_stripe)
1598 u64 local_end = fiemap->fm_extents[0].fe_logical;
1599 u64 lun_start, lun_end;
1601 int stripe_no = -1, i;
1603 if (fiemap->fm_extent_count == 0 ||
1604 fiemap->fm_extents[0].fe_logical == 0)
1607 /* Find out stripe_no from ost_index saved in the fe_device */
1608 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1609 struct lov_oinfo *oinfo = lsm->lsm_oinfo[i];
1611 if (lov_oinfo_is_dummy(oinfo))
1614 if (oinfo->loi_ost_idx == fiemap->fm_extents[0].fe_device) {
1619 if (stripe_no == -1)
1622 /* If we have finished mapping on previous device, shift logical
1623 * offset to start of next device */
1624 if ((lov_stripe_intersects(lsm, stripe_no, fm_start, fm_end,
1625 &lun_start, &lun_end)) != 0 &&
1626 local_end < lun_end) {
1627 fm_end_offset = local_end;
1628 *start_stripe = stripe_no;
1630 /* This is a special value to indicate that caller should
1631 * calculate offset in next stripe. */
1633 *start_stripe = (stripe_no + 1) % lsm->lsm_stripe_count;
1636 return fm_end_offset;
1640 * We calculate on which OST the mapping will end. If the length of mapping
1641 * is greater than (stripe_size * stripe_count) then the last_stripe will
1642 * will be one just before start_stripe. Else we check if the mapping
1643 * intersects each OST and find last_stripe.
1644 * This function returns the last_stripe and also sets the stripe_count
1645 * over which the mapping is spread
1647 * \param lsm striping information for the file
1648 * \param fm_start logical start of mapping
1649 * \param fm_end logical end of mapping
1650 * \param start_stripe starting stripe of the mapping
1651 * \param stripe_count the number of stripes across which to map is returned
1653 * \retval last_stripe return the last stripe of the mapping
1655 static int fiemap_calc_last_stripe(struct lov_stripe_md *lsm, u64 fm_start,
1656 u64 fm_end, int start_stripe,
1660 u64 obd_start, obd_end;
1663 if (fm_end - fm_start > lsm->lsm_stripe_size * lsm->lsm_stripe_count) {
1664 last_stripe = start_stripe < 1 ? lsm->lsm_stripe_count - 1 :
1666 *stripe_count = lsm->lsm_stripe_count;
1668 for (j = 0, i = start_stripe; j < lsm->lsm_stripe_count;
1669 i = (i + 1) % lsm->lsm_stripe_count, j++) {
1670 if ((lov_stripe_intersects(lsm, i, fm_start, fm_end,
1671 &obd_start, &obd_end)) == 0)
1675 last_stripe = (start_stripe + j - 1) % lsm->lsm_stripe_count;
1682 * Set fe_device and copy extents from local buffer into main return buffer.
1684 * \param fiemap fiemap request header
1685 * \param lcl_fm_ext array of local fiemap extents to be copied
1686 * \param ost_index OST index to be written into the fm_device field for each
1688 * \param ext_count number of extents to be copied
1689 * \param current_extent where to start copying in main extent array
1691 static void fiemap_prepare_and_copy_exts(struct ll_user_fiemap *fiemap,
1692 struct ll_fiemap_extent *lcl_fm_ext,
1693 int ost_index, unsigned int ext_count,
1699 for (ext = 0; ext < ext_count; ext++) {
1700 lcl_fm_ext[ext].fe_device = ost_index;
1701 lcl_fm_ext[ext].fe_flags |= FIEMAP_EXTENT_NET;
1704 /* Copy fm_extent's from fm_local to return buffer */
1705 to = (char *)fiemap + fiemap_count_to_size(current_extent);
1706 memcpy(to, lcl_fm_ext, ext_count * sizeof(struct ll_fiemap_extent));
1710 * Break down the FIEMAP request and send appropriate calls to individual OSTs.
1711 * This also handles the restarting of FIEMAP calls in case mapping overflows
1712 * the available number of extents in single call.
1714 static int lov_fiemap(struct lov_obd *lov, __u32 keylen, void *key,
1715 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
1717 struct ll_fiemap_info_key *fm_key = key;
1718 struct ll_user_fiemap *fiemap = val;
1719 struct ll_user_fiemap *fm_local = NULL;
1720 struct ll_fiemap_extent *lcl_fm_ext;
1722 unsigned int get_num_extents = 0;
1723 int ost_index = 0, actual_start_stripe, start_stripe;
1724 u64 fm_start, fm_end, fm_length, fm_end_offset;
1726 int current_extent = 0, rc = 0, i;
1727 int ost_eof = 0; /* EOF for object */
1728 int ost_done = 0; /* done with required mapping for this OST? */
1730 int cur_stripe = 0, cur_stripe_wrap = 0, stripe_count;
1731 unsigned int buffer_size = FIEMAP_BUFFER_SIZE;
1733 if (!lsm_has_objects(lsm)) {
1738 if (fiemap_count_to_size(fm_key->fiemap.fm_extent_count) < buffer_size)
1739 buffer_size = fiemap_count_to_size(fm_key->fiemap.fm_extent_count);
1741 fm_local = libcfs_kvzalloc(buffer_size, GFP_NOFS);
1742 if (fm_local == NULL) {
1746 lcl_fm_ext = &fm_local->fm_extents[0];
1748 count_local = fiemap_size_to_count(buffer_size);
1750 memcpy(fiemap, &fm_key->fiemap, sizeof(*fiemap));
1751 fm_start = fiemap->fm_start;
1752 fm_length = fiemap->fm_length;
1753 /* Calculate start stripe, last stripe and length of mapping */
1754 actual_start_stripe = start_stripe = lov_stripe_number(lsm, fm_start);
1755 fm_end = (fm_length == ~0ULL ? fm_key->oa.o_size :
1756 fm_start + fm_length - 1);
1757 /* If fm_length != ~0ULL but fm_start+fm_length-1 exceeds file size */
1758 if (fm_end > fm_key->oa.o_size)
1759 fm_end = fm_key->oa.o_size;
1761 last_stripe = fiemap_calc_last_stripe(lsm, fm_start, fm_end,
1762 actual_start_stripe, &stripe_count);
1764 fm_end_offset = fiemap_calc_fm_end_offset(fiemap, lsm, fm_start,
1765 fm_end, &start_stripe);
1766 if (fm_end_offset == -EINVAL) {
1771 if (fiemap_count_to_size(fiemap->fm_extent_count) > *vallen)
1772 fiemap->fm_extent_count = fiemap_size_to_count(*vallen);
1773 if (fiemap->fm_extent_count == 0) {
1774 get_num_extents = 1;
1777 /* Check each stripe */
1778 for (cur_stripe = start_stripe, i = 0; i < stripe_count;
1779 i++, cur_stripe = (cur_stripe + 1) % lsm->lsm_stripe_count) {
1780 u64 req_fm_len; /* Stores length of required mapping */
1781 u64 len_mapped_single_call;
1782 u64 lun_start, lun_end, obd_object_end;
1783 unsigned int ext_count;
1785 cur_stripe_wrap = cur_stripe;
1787 /* Find out range of mapping on this stripe */
1788 if ((lov_stripe_intersects(lsm, cur_stripe, fm_start, fm_end,
1789 &lun_start, &obd_object_end)) == 0)
1792 if (lov_oinfo_is_dummy(lsm->lsm_oinfo[cur_stripe])) {
1797 /* If this is a continuation FIEMAP call and we are on
1798 * starting stripe then lun_start needs to be set to
1800 if (fm_end_offset != 0 && cur_stripe == start_stripe)
1801 lun_start = fm_end_offset;
1803 if (fm_length != ~0ULL) {
1804 /* Handle fm_start + fm_length overflow */
1805 if (fm_start + fm_length < fm_start)
1806 fm_length = ~0ULL - fm_start;
1807 lun_end = lov_size_to_stripe(lsm, fm_start + fm_length,
1813 if (lun_start == lun_end)
1816 req_fm_len = obd_object_end - lun_start;
1817 fm_local->fm_length = 0;
1818 len_mapped_single_call = 0;
1820 /* If the output buffer is very large and the objects have many
1821 * extents we may need to loop on a single OST repeatedly */
1825 if (get_num_extents == 0) {
1826 /* Don't get too many extents. */
1827 if (current_extent + count_local >
1828 fiemap->fm_extent_count)
1829 count_local = fiemap->fm_extent_count -
1833 lun_start += len_mapped_single_call;
1834 fm_local->fm_length = req_fm_len - len_mapped_single_call;
1835 req_fm_len = fm_local->fm_length;
1836 fm_local->fm_extent_count = count_local;
1837 fm_local->fm_mapped_extents = 0;
1838 fm_local->fm_flags = fiemap->fm_flags;
1840 fm_key->oa.o_oi = lsm->lsm_oinfo[cur_stripe]->loi_oi;
1841 ost_index = lsm->lsm_oinfo[cur_stripe]->loi_ost_idx;
1843 if (ost_index < 0 ||
1844 ost_index >= lov->desc.ld_tgt_count) {
1849 /* If OST is inactive, return extent with UNKNOWN flag */
1850 if (!lov->lov_tgts[ost_index]->ltd_active) {
1851 fm_local->fm_flags |= FIEMAP_EXTENT_LAST;
1852 fm_local->fm_mapped_extents = 1;
1854 lcl_fm_ext[0].fe_logical = lun_start;
1855 lcl_fm_ext[0].fe_length = obd_object_end -
1857 lcl_fm_ext[0].fe_flags |= FIEMAP_EXTENT_UNKNOWN;
1862 fm_local->fm_start = lun_start;
1863 fm_local->fm_flags &= ~FIEMAP_FLAG_DEVICE_ORDER;
1864 memcpy(&fm_key->fiemap, fm_local, sizeof(*fm_local));
1865 *vallen = fiemap_count_to_size(fm_local->fm_extent_count);
1866 rc = obd_get_info(NULL,
1867 lov->lov_tgts[ost_index]->ltd_exp,
1868 keylen, key, vallen, fm_local, lsm);
1873 ext_count = fm_local->fm_mapped_extents;
1874 if (ext_count == 0) {
1876 /* If last stripe has hole at the end,
1877 * then we need to return */
1878 if (cur_stripe_wrap == last_stripe) {
1879 fiemap->fm_mapped_extents = 0;
1885 /* If we just need num of extents then go to next device */
1886 if (get_num_extents) {
1887 current_extent += ext_count;
1891 len_mapped_single_call = lcl_fm_ext[ext_count-1].fe_logical -
1892 lun_start + lcl_fm_ext[ext_count - 1].fe_length;
1894 /* Have we finished mapping on this device? */
1895 if (req_fm_len <= len_mapped_single_call)
1898 /* Clear the EXTENT_LAST flag which can be present on
1900 if (lcl_fm_ext[ext_count-1].fe_flags & FIEMAP_EXTENT_LAST)
1901 lcl_fm_ext[ext_count - 1].fe_flags &=
1902 ~FIEMAP_EXTENT_LAST;
1904 curr_loc = lov_stripe_size(lsm,
1905 lcl_fm_ext[ext_count - 1].fe_logical+
1906 lcl_fm_ext[ext_count - 1].fe_length,
1908 if (curr_loc >= fm_key->oa.o_size)
1911 fiemap_prepare_and_copy_exts(fiemap, lcl_fm_ext,
1912 ost_index, ext_count,
1915 current_extent += ext_count;
1917 /* Ran out of available extents? */
1918 if (current_extent >= fiemap->fm_extent_count)
1920 } while (ost_done == 0 && ost_eof == 0);
1922 if (cur_stripe_wrap == last_stripe)
1927 /* Indicate that we are returning device offsets unless file just has
1929 if (lsm->lsm_stripe_count > 1)
1930 fiemap->fm_flags |= FIEMAP_FLAG_DEVICE_ORDER;
1932 if (get_num_extents)
1933 goto skip_last_device_calc;
1935 /* Check if we have reached the last stripe and whether mapping for that
1936 * stripe is done. */
1937 if (cur_stripe_wrap == last_stripe) {
1938 if (ost_done || ost_eof)
1939 fiemap->fm_extents[current_extent - 1].fe_flags |=
1943 skip_last_device_calc:
1944 fiemap->fm_mapped_extents = current_extent;
1951 static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
1952 __u32 keylen, void *key, __u32 *vallen, void *val,
1953 struct lov_stripe_md *lsm)
1955 struct obd_device *obddev = class_exp2obd(exp);
1956 struct lov_obd *lov = &obddev->u.lov;
1959 if (!vallen || !val)
1964 if (KEY_IS(KEY_LOCK_TO_STRIPE)) {
1967 struct ldlm_lock *lock;
1969 struct ldlm_res_id *res_id = &data->lock->l_resource->lr_name;
1970 struct lov_oinfo *loi;
1971 __u32 *stripe = val;
1973 if (*vallen < sizeof(*stripe)) {
1977 *vallen = sizeof(*stripe);
1979 /* XXX This is another one of those bits that will need to
1980 * change if we ever actually support nested LOVs. It uses
1981 * the lock's export to find out which stripe it is. */
1982 /* XXX - it's assumed all the locks for deleted OSTs have
1983 * been cancelled. Also, the export for deleted OSTs will
1984 * be NULL and won't match the lock's export. */
1985 for (i = 0; i < lsm->lsm_stripe_count; i++) {
1986 loi = lsm->lsm_oinfo[i];
1987 if (lov_oinfo_is_dummy(loi))
1990 if (!lov->lov_tgts[loi->loi_ost_idx])
1992 if (lov->lov_tgts[loi->loi_ost_idx]->ltd_exp ==
1993 data->lock->l_conn_export &&
1994 ostid_res_name_eq(&loi->loi_oi, res_id)) {
2000 LDLM_ERROR(data->lock, "lock on inode without such object");
2001 dump_lsm(D_ERROR, lsm);
2004 } else if (KEY_IS(KEY_LAST_ID)) {
2005 struct obd_id_info *info = val;
2006 __u32 size = sizeof(u64);
2007 struct lov_tgt_desc *tgt;
2009 LASSERT(*vallen == sizeof(struct obd_id_info));
2010 tgt = lov->lov_tgts[info->idx];
2012 if (!tgt || !tgt->ltd_active) {
2017 rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
2018 &size, info->data, NULL);
2021 } else if (KEY_IS(KEY_LOVDESC)) {
2022 struct lov_desc *desc_ret = val;
2023 *desc_ret = lov->desc;
2027 } else if (KEY_IS(KEY_FIEMAP)) {
2028 rc = lov_fiemap(lov, keylen, key, vallen, val, lsm);
2030 } else if (KEY_IS(KEY_CONNECT_FLAG)) {
2031 struct lov_tgt_desc *tgt;
2032 __u64 ost_idx = *((__u64 *)val);
2034 LASSERT(*vallen == sizeof(__u64));
2035 LASSERT(ost_idx < lov->desc.ld_tgt_count);
2036 tgt = lov->lov_tgts[ost_idx];
2038 if (!tgt || !tgt->ltd_exp) {
2043 *((__u64 *)val) = exp_connect_flags(tgt->ltd_exp);
2046 } else if (KEY_IS(KEY_TGT_COUNT)) {
2047 *((int *)val) = lov->desc.ld_tgt_count;
2059 static int lov_set_info_async(const struct lu_env *env, struct obd_export *exp,
2060 u32 keylen, void *key, u32 vallen,
2061 void *val, struct ptlrpc_request_set *set)
2063 struct obd_device *obddev = class_exp2obd(exp);
2064 struct lov_obd *lov = &obddev->u.lov;
2067 struct lov_tgt_desc *tgt;
2068 unsigned incr, check_uuid,
2069 do_inactive, no_set;
2070 unsigned next_id = 0, mds_con = 0;
2072 incr = check_uuid = do_inactive = no_set = 0;
2075 set = ptlrpc_prep_set();
2081 count = lov->desc.ld_tgt_count;
2083 if (KEY_IS(KEY_NEXT_ID)) {
2084 count = vallen / sizeof(struct obd_id_info);
2085 vallen = sizeof(u64);
2086 incr = sizeof(struct obd_id_info);
2089 } else if (KEY_IS(KEY_CHECKSUM)) {
2091 } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2092 /* use defaults: do_inactive = incr = 0; */
2093 } else if (KEY_IS(KEY_MDS_CONN)) {
2095 } else if (KEY_IS(KEY_CACHE_SET)) {
2096 LASSERT(lov->lov_cache == NULL);
2097 lov->lov_cache = val;
2101 for (i = 0; i < count; i++, val = (char *)val + incr) {
2103 tgt = lov->lov_tgts[((struct obd_id_info *)val)->idx];
2105 tgt = lov->lov_tgts[i];
2106 /* OST was disconnected */
2107 if (!tgt || !tgt->ltd_exp)
2110 /* OST is inactive and we don't want inactive OSCs */
2111 if (!tgt->ltd_active && !do_inactive)
2115 struct mds_group_info *mgi;
2117 LASSERT(vallen == sizeof(*mgi));
2118 mgi = (struct mds_group_info *)val;
2120 /* Only want a specific OSC */
2121 if (mgi->uuid && !obd_uuid_equals(mgi->uuid,
2125 err = obd_set_info_async(env, tgt->ltd_exp,
2126 keylen, key, sizeof(int),
2128 } else if (next_id) {
2129 err = obd_set_info_async(env, tgt->ltd_exp,
2130 keylen, key, vallen,
2131 ((struct obd_id_info *)val)->data, set);
2133 /* Only want a specific OSC */
2135 !obd_uuid_equals(val, &tgt->ltd_uuid))
2138 err = obd_set_info_async(env, tgt->ltd_exp,
2139 keylen, key, vallen, val, set);
2148 err = ptlrpc_set_wait(set);
2151 ptlrpc_set_destroy(set);
2156 void lov_stripe_lock(struct lov_stripe_md *md)
2157 __acquires(&md->lsm_lock)
2159 LASSERT(md->lsm_lock_owner != current_pid());
2160 spin_lock(&md->lsm_lock);
2161 LASSERT(md->lsm_lock_owner == 0);
2162 md->lsm_lock_owner = current_pid();
2164 EXPORT_SYMBOL(lov_stripe_lock);
2166 void lov_stripe_unlock(struct lov_stripe_md *md)
2167 __releases(&md->lsm_lock)
2169 LASSERT(md->lsm_lock_owner == current_pid());
2170 md->lsm_lock_owner = 0;
2171 spin_unlock(&md->lsm_lock);
2173 EXPORT_SYMBOL(lov_stripe_unlock);
2175 static int lov_quotactl(struct obd_device *obd, struct obd_export *exp,
2176 struct obd_quotactl *oqctl)
2178 struct lov_obd *lov = &obd->u.lov;
2179 struct lov_tgt_desc *tgt;
2181 __u64 bhardlimit = 0;
2184 if (oqctl->qc_cmd != LUSTRE_Q_QUOTAON &&
2185 oqctl->qc_cmd != LUSTRE_Q_QUOTAOFF &&
2186 oqctl->qc_cmd != Q_GETOQUOTA &&
2187 oqctl->qc_cmd != Q_INITQUOTA &&
2188 oqctl->qc_cmd != LUSTRE_Q_SETQUOTA &&
2189 oqctl->qc_cmd != Q_FINVALIDATE) {
2190 CERROR("bad quota opc %x for lov obd", oqctl->qc_cmd);
2196 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2199 tgt = lov->lov_tgts[i];
2204 if (!tgt->ltd_active || tgt->ltd_reap) {
2205 if (oqctl->qc_cmd == Q_GETOQUOTA &&
2206 lov->lov_tgts[i]->ltd_activate) {
2208 CERROR("ost %d is inactive\n", i);
2210 CDEBUG(D_HA, "ost %d is inactive\n", i);
2215 err = obd_quotactl(tgt->ltd_exp, oqctl);
2217 if (tgt->ltd_active && !rc)
2222 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2223 curspace += oqctl->qc_dqblk.dqb_curspace;
2224 bhardlimit += oqctl->qc_dqblk.dqb_bhardlimit;
2229 if (oqctl->qc_cmd == Q_GETOQUOTA) {
2230 oqctl->qc_dqblk.dqb_curspace = curspace;
2231 oqctl->qc_dqblk.dqb_bhardlimit = bhardlimit;
2236 static int lov_quotacheck(struct obd_device *obd, struct obd_export *exp,
2237 struct obd_quotactl *oqctl)
2239 struct lov_obd *lov = &obd->u.lov;
2244 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2245 if (!lov->lov_tgts[i])
2248 /* Skip quota check on the administratively disabled OSTs. */
2249 if (!lov->lov_tgts[i]->ltd_activate) {
2250 CWARN("lov idx %d was administratively disabled, skip quotacheck on it.\n",
2255 if (!lov->lov_tgts[i]->ltd_active) {
2256 CERROR("lov idx %d inactive\n", i);
2262 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
2265 if (!lov->lov_tgts[i] || !lov->lov_tgts[i]->ltd_activate)
2268 err = obd_quotacheck(lov->lov_tgts[i]->ltd_exp, oqctl);
2279 static struct obd_ops lov_obd_ops = {
2280 .o_owner = THIS_MODULE,
2281 .o_setup = lov_setup,
2282 .o_precleanup = lov_precleanup,
2283 .o_cleanup = lov_cleanup,
2284 /*.o_process_config = lov_process_config,*/
2285 .o_connect = lov_connect,
2286 .o_disconnect = lov_disconnect,
2287 .o_statfs = lov_statfs,
2288 .o_statfs_async = lov_statfs_async,
2289 .o_packmd = lov_packmd,
2290 .o_unpackmd = lov_unpackmd,
2291 .o_create = lov_create,
2292 .o_destroy = lov_destroy,
2293 .o_getattr_async = lov_getattr_async,
2294 .o_setattr_async = lov_setattr_async,
2295 .o_adjust_kms = lov_adjust_kms,
2296 .o_find_cbdata = lov_find_cbdata,
2297 .o_iocontrol = lov_iocontrol,
2298 .o_get_info = lov_get_info,
2299 .o_set_info_async = lov_set_info_async,
2300 .o_notify = lov_notify,
2301 .o_pool_new = lov_pool_new,
2302 .o_pool_rem = lov_pool_remove,
2303 .o_pool_add = lov_pool_add,
2304 .o_pool_del = lov_pool_del,
2305 .o_getref = lov_getref,
2306 .o_putref = lov_putref,
2307 .o_quotactl = lov_quotactl,
2308 .o_quotacheck = lov_quotacheck,
2311 struct kmem_cache *lov_oinfo_slab;
2313 static int __init lov_init(void)
2315 struct lprocfs_static_vars lvars = { NULL };
2318 /* print an address of _any_ initialized kernel symbol from this
2319 * module, to allow debugging with gdb that doesn't support data
2320 * symbols from modules.*/
2321 CDEBUG(D_INFO, "Lustre LOV module (%p).\n", &lov_caches);
2323 rc = lu_kmem_init(lov_caches);
2327 lov_oinfo_slab = kmem_cache_create("lov_oinfo",
2328 sizeof(struct lov_oinfo),
2329 0, SLAB_HWCACHE_ALIGN, NULL);
2330 if (lov_oinfo_slab == NULL) {
2331 lu_kmem_fini(lov_caches);
2334 lprocfs_lov_init_vars(&lvars);
2336 rc = class_register_type(&lov_obd_ops, NULL,
2337 LUSTRE_LOV_NAME, &lov_device_type);
2340 kmem_cache_destroy(lov_oinfo_slab);
2341 lu_kmem_fini(lov_caches);
2347 static void /*__exit*/ lov_exit(void)
2349 class_unregister_type(LUSTRE_LOV_NAME);
2350 kmem_cache_destroy(lov_oinfo_slab);
2352 lu_kmem_fini(lov_caches);
2355 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2356 MODULE_DESCRIPTION("Lustre Logical Object Volume OBD driver");
2357 MODULE_LICENSE("GPL");
2358 MODULE_VERSION(LUSTRE_VERSION_STRING);
2360 module_init(lov_init);
2361 module_exit(lov_exit);